The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Now Quarterly FEDERAL RESERVE BANK OF ST. LOUIS REVIEW Federal Reserve Bank of St. Louis P.O. Box 442 St. Louis, MO 63166-0442 FIRST QUARTER 2014 VOLUME 96 | NUMBER 1 The Rise and Fall of Labor Force Participation in the United States James Bullard REVIEW The Great Trade Collapse and Rebound: A State-by-State View Cletus C. Coughlin A Guide to Tracking the U.S. Economy Kevin L. Kliesen QE: Is There a Portfolio Balance Effect? Daniel L. Thornton The Evolution of Federal Reserve Policy and the Impact of Monetary Policy Surprises on Asset Prices First Quarter 2014 • Volume 96, Number 1 Brett W. Fawley and Christopher J. Neely REVIEW Volume 96 • Number 1 President and CEO James Bullard Director of Research Christopher J. Waller 1 The Rise and Fall of Labor Force Participation in the United States James Bullard Policy Adviser Cletus C. Coughlin Deputy Director of Research David C. Wheelock Review Editor-in-Chief William T. Gavin Research Economists David Andolfatto Alejandro Badel Subhayu Bandyopadhyay Maria E. Canon YiLi Chien Silvio Contessi Riccardo DiCecio William Dupor Carlos Garriga Rubén Hernández-Murillo Kevin L. Kliesen Fernando M. Martin Michael W. McCracken Alexander Monge-Naranjo Christopher J. Neely Michael T. Owyang B. Ravikumar Juan M. Sánchez Daniel L. Thornton Yi Wen David Wiczer Christian M. Zimmermann 13 The Great Trade Collapse and Rebound: A State-by-State View Cletus C. Coughlin 35 A Guide to Tracking the U.S. Economy Kevin L. Kliesen 55 QE: Is There a Portfolio Balance Effect? Daniel L. Thornton 73 The Evolution of Federal Reserve Policy and the Impact of Monetary Policy Surprises on Asset Prices Brett W. Fawley and Christopher J. Neely Managing Editor George E. Fortier Editors Judith A. Ahlers Lydia H. Johnson Graphic Designer Donna M. Stiller Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 i Review Now Published Quarterly Review is published four times per year by the Research Division of the Federal Reserve Bank of St. Louis. Complimentary print subscriptions are available to U.S. addresses only. Full online access is available to all, free of charge. Online Access to Current and Past Issues The current issue and past issues dating back to 1967 may be accessed through our Research Division website: http://research.stlouisfed.org/publications/review. All nonproprietary and nonconfidential data and programs for the articles written by Federal Reserve Bank of St. Louis staff and published in Review also are available to our readers on this website. Review articles published before 1967 may be accessed through our digital archive, FRASER: http://fraser.stlouisfed.org/publication/?pid=820. Review is indexed in Fed in Print, the catalog of Federal Reserve publications (http://www.fedinprint.org/) and in IDEAS/RePEc, the free online bibliography hosted by the Research Division (http://ideas.repec.org/). Authorship and Disclaimer The majority of research published in Review is authored by economists on staff at the Federal Reserve Bank of St. Louis. Visiting scholars and others affiliated with the St. Louis Fed or the Federal Reserve System occasionally provide content as well. Review does not accept unsolicited manuscripts for publication. The views expressed in Review are those of the individual authors and do not necessarily reflect official positions of the Federal Reserve Bank of St. Louis, the Federal Reserve System, or the Board of Governors. Subscriptions and Alerts Single-copy subscriptions (U.S. addresses only) are available free of charge. Subscribe here: https://research.stlouisfed.org/publications/review/subscribe/. Our monthly email newsletter keeps you informed when new issues of Review, Economic Synopses, Regional Economist, and other publications become available; it also alerts you to new or enhanced data and information services provided by the St. Louis Fed. Subscribe to the newsletter here: http://research.stlouisfed.org/newsletter-subscribe.html. Copyright and Permissions Articles may be reprinted, reproduced, republished, distributed, displayed, and transmitted in their entirety if copyright notice, author name(s), and full citation are included. In these cases, there is no need to request written permission or approval. Please send a copy of any reprinted or republished materials to Review, Research Division of the Federal Reserve Bank of St. Louis, P.O. Box 442, St. Louis, MO 63166-0442; STLS.Research.Publications@stls.frb.org. Please note that any abstracts, synopses, translations, or other derivative work based on content published in Review may be made only with prior written permission of the Federal Reserve Bank of St. Louis. Please contact the Review editor at the above address to request this permission. Economic Data General economic data can be obtained through FRED (Federal Reserve Economic Data), our free database with more than 200,000 national, international, and regional data series, including data for our own Eighth Federal Reserve District. You may access FRED through our website: http://research.stlouisfed.org/fred2. © 2014, Federal Reserve Bank of St. Louis. ISSN 0014-9187 ii First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW In Memoriam: Michael J. Dueker ur friend and former colleague, Michael J. Dueker, passed away on Wednesday, January 29, 2014. Mike was a well-respected economist and policy advisor, a sought-after colleague and coauthor, and a close friend of many current and former employees of the St. Louis Fed. Mike joined the Research Division of the Federal Reserve Bank of St. Louis as an economist in 1991, following his graduation from the University of Washington, Seattle, where he earned a Ph.D. in economics. At the Bank, Mike was promoted three times—to senior economist, research officer, and assistant vice president—in recognition of his strong performance as a research economist and policy advisor. Mike left the Bank in 2008 to return to the Seattle area to work for Russell Investments. He then also became a regular contributor to the Blue Chip Economic Indicators panel of professional forecasters. Mike remained a good friend of the St. Louis Fed after leaving for Seattle, returning for several short visits to participate in Bank conferences and to work with coauthors. Mike was hired by Anatol B. (“Ted”) Balbach, who was the St. Louis Fed’s director of research at the time. Mike was a well-trained econometrician who had demonstrated strong technical skills, particularly in econometrics and statistics, in his Ph.D. dissertation. However, Ted was concerned whether Mike would make a good economist. “I know Mike is a good ‘metrician,’” Ted said, “but I don’t know about the ‘econ’ part—is he a good economist?” Mike clearly was the most talented of the new Ph.D.’s that the St. Louis Fed interviewed that year, however, so Ted was willing to take a chance on Mike. Ted was not disappointed. Mike’s research focused on developing new econometric methods and applying those methods to important economic policy questions. He was not an economic theorist, but he had a deep understanding of economics and knowledge of financial markets. Mike shared generously with his colleagues. In seminars, Mike asked penetrating questions. Frequently, he offered constructive comments and criticism that helped his colleagues build better economic models. Mike really was the economist that Ted hoped he would be, and then some. Mike’s understanding of economics, strong technical skills, and friendly manner made him a sought-after colleague and coauthor. He published over 50 articles on a broad range of topics. As of January 2014, Mike ranked among the top 5 percent of economists worldwide in terms of number of publications and measures of publication impact. Mike had coauthors all over the world. Early in his career, he spent a year at the Swiss National Bank, where he gained insights into the implementation of monetary policy in different countries and developed close working relationships with the Bank’s economists that led to several published articles. He also O Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 iii In Memoriam: Michael J. Dueker wrote with several of his St. Louis Fed colleagues on a wide range of topics, including the impact of price-level shocks on financial stability, Federal Reserve actions to smooth interest rates, and excess returns in foreign exchange market. Mike continued to do research and write with coauthors from the St. Louis Fed and elsewhere after he moved to Seattle in 2008. Many of Mike’s papers were highly technical, but he always found important applications. In one of his most highly cited works, Mike found that the difference between the yields on long- and short-term government securities is a good predictor of recessions.1 Mike built on this research to develop methods of classifying and forecasting the business cycle and other variables with discrete outcomes. Discrete outcome models parse the business cycle into phases—say, expansion and recession. One of Mike’s interests was in forecasting transitions from expansion to recession using currently available data. He recognized that output growth, employment, and other economic and financial market variables forecast business cycle transitions, while the behavior of those variables in turn reflected the current phase of the cycle. Mike developed a model he called the Qual-VAR that combines discrete variables such as the business cycle phase with the commonly used vector autoregression (VAR). The Qual-VAR led Mike to construct a business cycle index from which he assessed the probability that the economy would enter recession at some given date in the future.2 Mike also studied monetary policy and financial markets, first by modeling volatility in financial markets and then by identifying trends in foreign exchange rates. He also studied how Federal Reserve discount rate changes affected market rates, the setting of federal funds rate targets, and how explicit inflation targets might improve economic performance. Mike had a particular interest in modeling the behavior of variables, such as central bank policy rates and commercial bank prime lending rates, that change by discrete amounts. Mike proposed a methodology that both greatly simplified estimation of time-dependent classification models (such as the “dynamic probit” model) and incorporated features common to financial time-series models (autoregressive conditional heteroskedasticity). He applied the model to study the behavior of commercial bank prime lending rates and showed that periods of high volatility in the prime rate coincided with market uncertainty about monetary policy. Further, Mike showed that models of discrete variables that allow for changes in variance outperform constant variance specifications.3 This creative joining of disconnected models in the literature is something at which Mike excelled. Mike had a terrific career. He gave sound advice to policymakers and wrote important research papers. He was an outstanding, well-liked colleague who offered freely of his time and talents, and he will be missed. ■ NOTES iv 1 “Strengthening the Case for the Yield Curve as a Predictor of U.S. Recessions.” Federal Reserve Bank of St. Louis Review, March/April 1997, 79(2), pp. 41-51; http://research.stlouisfed.org/publications/review/article/3183. 2 “Dynamic Forecasts of Qualitative Variables: A Qual VAR Model of U.S. Recessions.” Journal of Business and Economic Statistics, January 2005, 23(1), pp. 96-104. 3 “Conditional Heteroscedasticity in Qualitative Response Models of Time Series: A Gibbs Sampling Approach to the Bank Prime Rate.” Journal of Business and Economic Statistics, October 1999, 17(4), pp. 466-72. First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW The Rise and Fall of Labor Force Participation in the United States James Bullard Monetary policy choices going forward are explicitly tied to labor market performance. Hence, the sharp decline in the labor force participation rate following the 2007-09 recession has become a salient topic. Presented here are a summary of labor force participation rate data and projections, a survey of the literature that studies the recent decline in the participation rate, and a view toward fruitful paths for future research. (JEL E32, E52, J11, J21, J22) Federal Reserve Bank of St. Louis Review, First Quarter 2014, 96(1), pp. 1-12. abor force participation in the United States has been a controversial subject in current macroeconomic discussions. In this article, I try to offer my own perspectives on the issue. The participation rate—a measure of the number of people actively involved in labor markets—has generally been a secondary concern in macroeconomics. However, with recent sharp declines following the financial crisis and recession of 2007-09, it has suddenly become a salient topic, and one that gets discussed even in non-economic settings. At its broadest level, the debate about the labor force participation rate is a debate about the nature of the U.S. economy over the 4½ years since the end of the recession, in the summer of 2009. Should we characterize the economy as substantially back to normal after a very severe recession? Or has little progress really been made, so that the economy remains far from its potential? There are clear lines of argument on both sides, sometimes blurring political boundaries. Some suggest that the extraordinary policy response since the end of the recession has been largely ineffectual, perhaps citing the very flat employment-to-population ratio since 2009,1 and that their own suggested policy responses would have produced better outcomes. Others emphasize the risk associated with the extraordinary policy response, perhaps citing the Fed’s now $4.1 trillion balance sheet and the nation’s relatively high debt-to-gross domestic product (GDP) ratio. Still others argue that the economy has recovered as well as can be expected in the wake of a major financial crisis, perhaps citing a recovery in real consumption expendi- L James Bullard is president and CEO of the Federal Reserve Bank of St. Louis. An earlier version of this article was delivered as a speech to the Exchequer Club, Washington DC, on February 19, 2014. The author thanks his staff for helpful comments. © 2014, The Federal Reserve Bank of St. Louis. The views expressed in this article are those of the author(s) and do not necessarily reflect the views of the Federal Reserve System, the Board of Governors, or the regional Federal Reserve Banks. Articles may be reprinted, reproduced, published, distributed, displayed, and transmitted in their entirety if copyright notice, author name(s), and full citation are included. Abstracts, synopses, and other derivative works may be made only with prior written permission of the Federal Reserve Bank of St. Louis. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 1 Bullard Figure 1 Unemployment Rate Before, During, and After the 2007-09 Recession Percent 12 10 8 6 4 2 0 Jan-03 Jan-04 Jan-05 Jan-06 Jan-07 Jan-08 Jan-09 Jan-10 Jan-11 Jan-12 Jan-13 Jan-14 NOTE: The shaded bar indicates the 2007-09 recession as determined by the National Bureau of Economic Research. SOURCE: Bureau of Labor Statistics and National Bureau of Economic Research. Last observation January 2014. tures, an improved housing market, a recovery in equity price valuations, and substantially lower unemployment. This last group might point to the euro area as an example of an economy that has suffered through a double-dip recession over the past several years, eventually leading to unemployment rates exceeding 12 percent, while the United States avoided this fate. Labor market performance is at the heart of the debate over how to characterize the state of the U.S. economy. While unemployment in the United States was at 10 percent in the fall of 2009 (Figure 1), it has now declined to 6.6 percent according to the latest reading and has generally declined much faster than many forecasters anticipated. In tandem with this rosy development, however, there has been a substantial decline in labor force participation. Some say that the decline in labor force participation is a bad omen for U.S. macroeconomic performance, with labor market dropouts reflecting frustration with the state of the economy. I call this the “bad omen” view. Under this interpretation, the decline in the unemployment rate does not really reflect an improving labor market, and policymakers should look elsewhere to measure labor market outcomes. Others, however, argue that the decline in labor force participation simply reflects changing demographics in the U.S. economy and that different demographic groups have different propensities to participate in market work. As we have different numbers of people in these different demographic groups, we should naturally expect the aggregate labor force participation rate to change. I call this the “demographics” view. Under this inter2 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Bullard pretation, the unemployment rate remains about as good an indicator of overall labor market health as ever, and recent sharp declines in the unemployment rate should indeed be taken as indicative of an improving economy and an improving labor market. In sum, the bad omen view interprets the recent declines in labor force participation as suggestive of a very weak labor market and discounts the signal coming from recent fasterthan-expected declines in unemployment. The demographics view interprets recent declines in labor force participation as more benign and takes the signal coming from recent faster-thanexpected declines in unemployment at face value. Since the Federal Open Market Committee (FOMC) has explicitly tied monetary policy choices to labor market performance, it is of considerable importance which view is more nearly correct. I offer three perspectives on these questions. First, I simply summarize the data on labor force participation and provide some background on why this variable has suffered in relative obscurity until now. Second, I summarize my views on some of the available literature concerning labor force participation as it exists today. In my opinion, this literature is generally supportive of the demographics view, although there are different strands and many issues are not satisfactorily resolved. Third, I discuss the future of research in this area, which is to move to more-sophisticated approaches to labor force participation. The more-sophisticated class of models might be based on the so-called home production literature. Without going into the details of this approach, I believe that future progress in this area must become more serious about the incentives of households to supply labor to market work versus nonmarket work.2 THE LABOR FORCE PARTICIPATION RATE The concept of labor force participation divides the population into three groups: employed, unemployed, and out of the labor force. This last group could also be called “nonparticipants” because they are neither working nor searching for market work. The labor force participation rate is a ratio. In the numerator is the labor force, the sum of all persons employed and persons unemployed. We think of this group as participants in the market workplace. Many have jobs, and the rest are looking for jobs. In the denominator is the civilian noninstitutional population 16 years of age or older—that is, a generally accepted collection of everyone who conceivably could be working. Here are some round numbers to keep in mind. The employed group is currently on the order of 145 million people. The unemployed group is on the order of 10 million people. And the nonparticipant group is on the order of 91 million people. The groups vary greatly in size; in particular, the nonparticipant group is large relative to the unemployed group.3 One quirk of organizing the data this way is that people routinely report moving from nonparticipation to market work without reporting themselves as unemployed. In other words—at least officially—they were not working and were not searching for a job but nevertheless ended up working at a job in the next reporting period. Evidently, they were not really properly categorized as “nonparticipants.” I have always found this to be an unsatisfactory aspect of this method of data organization. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 3 Bullard Figure 2 Labor Force Participation Rate Percent 68 66 64 62 60 58 56 Jan-48 Jan-54 Jan-60 Jan-66 Jan-72 Jan-78 Jan-84 Jan-90 Jan-96 Jan-02 Jan-08 Jan-14 NOTE: The shaded bars indicate recessions as determined by the National Bureau of Economic Research. SOURCE: Bureau of Labor Statistics and National Bureau of Economic Research. Last observation January 2014. Many discussions of contemporary unemployment forecasts focus on the extent to which nonparticipants will rejoin the labor market. During the late 1990s, for example, many workers seemed to come off the sidelines into the workplace because of an exceptionally strong economy. At the Federal Reserve Bank of St. Louis, we have constructed unemployment forecasts in recent years assuming that movements from nonparticipation to employment would be minimal while unemployment was at relatively high levels. This has served us well, as we have more accurately predicted declines in unemployment in the past year than many other forecasters.4 If you know only one aspect of the data on labor force participation, it should be this: Labor force participation used to be relatively low. It rose during the 1970s, 1980s, and 1990s; peaked in 2000; and has generally been declining since 2000 (Figure 2). From 1948 to 1966, the labor force participation rate was relatively low and relatively stable, averaging 59.1 percent—substantially lower than today’s value of 63 percent. It is important to note that we normally consider the U.S. economy to have performed relatively well during this period, especially during the long expansion of the 1960s. Evidently, low labor force participation does not equate with weak economic growth. Surely this is because the factors driving economic growth differ from those driving labor force participation. 4 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Bullard After about three decades of trending upward, the labor force participation rate peaked in the first half of 2000 at 67.3 percent. The rate of increase was slower in the 1990s than in the 1970s or 1980s. The peak was more than 8 percentage points higher than the average level during 1948-66. Many labor force participation studies during this period focused on the increasing participation rates of women. However, whatever effects came from that source, or any other source, the labor force participation rate could not continue to increase forever. Households make choices about how much labor to supply given current wages and work environments, and women newly joining the labor force would find the right level of participation and stop there. Since 2000, the labor force participation rate has generally been declining. The pace of decline was particularly sharp during the 2007-09 recession, but the participation rate also declined steadily in the early 2000s and since the end of the recession in mid-2009. The general picture, then, is one of a hump shape in U.S. aggregate labor force participation during the postwar era. A satisfactory theory must account for this hump shape. One way to build such a theory is to appeal to demographics. The nation’s workforce had a younger profile as the Baby Boom generation came of age, and it will have an older profile as the Baby Boom generation continues to retire. Since different age groups have different propensities to participate, this suggests a promising avenue to explain the labor force participation data. I daresay that the demographic explanation is the gut instinct of many macroeconomists. This is why labor force participation sits in the backseat of many macroeconomic models. Many, including me, might reason that a good demographic model combined with more women in the labor force during the 1970s, 1980s, and 1990s could explain a very large fraction of the hump-shaped movements in aggregate labor force participation over the postwar era. If such a model were fitted to the data, only a small amount of variation in the participation rate would remain to be explained. That small remaining amount of variation might be attributable to business cycle (cyclical) effects, or it might just be noise about the fundamental hump-shaped trend. Relatively minor cyclical effects on labor force participation would likely be too small to have major macroeconomic implications given everything else going on in a macroeconomic model. Consequently, it might seem that we do not need to worry too much about the labor force participation rate for business cycle purposes. But all this is just in the heads of macroeconomists. I now turn to some of the recent research on labor force participation to determine the extent to which such a theory has actually been devised. RECENT RESEARCH ON LABOR FORCE PARTICIPATION Let’s start with the Bureau of Labor Statistics (BLS). The BLS is, of course, very close to the data and it routinely projects labor force participation over the medium term. In general, its medium-term forecasts from the mid-2000s proved to be too high, meaning that its forecast labor force participation rate was considerably higher than the values actually observed. More recent medium-term BLS forecasts call for a declining rate of participation over the next decade or so, all the way down to 61.6 percent in 2022 (Figure 3).5 Recall that today’s participation rate Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 5 Bullard Figure 3 Labor Force Participation Rate: BLS Data and Projections Percent 68 Data Toossi (2013) projection 67 66 65 64 63 62 61 60 1990 1993 1996 1999 2002 2005 2008 2011 2014 2017 2020 SOURCE: Bureau of Labor Statistics and Toossi (2013). Last observation 2013. is 63.0 percent, so the rate is projected to continue to decline by around 15 basis points per year. According to BLS projections,6 more than 70 percent of this decline is due to purely demographic factors—that is, changes in population shares by age groups, assuming unchanged participation rates for each group.7 To the extent that this forecast pans out, the basic direction for the labor force participation rate is down, not up. Those waiting for an upward swing in labor force participation as the economy continues to expand will be disappointed, on average, if this forecast comes to pass. I read the BLS work as supportive of the demographics hypothesis I described previously. Fujita (2013) provides some additional insight concerning the decline in aggregate U.S. labor force participation since 2000. Fujita’s calculations suggest that about 65 percent of the decline in the participation rate was due to retirements and disability. Fujita points out that the empirical evidence suggests members of these groups have only a small probability of returning to the labor force. If we limit attention only to a period of relatively high economic stress, such as 2007:Q1 to 2011:Q2, we do see more of the decline in participation attributable to discouraged workers; but, even then, this is only about 25 percent according to Fujita’s calculations. Over a less stressful period, such as 2012:Q1 to 2013:Q2, the entire decline in the aggregate labor force participation rate is attributable to retirements, with no effect at all coming from an increase in discouraged workers. I read Fujita’s contribution as also supportive of the demographics hypothesis. 6 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Bullard Figure 4 Labor Force Participation Rate: BLS Data and Aaronson et al. (2006) Projections Percent 68 Data Aaronson et al. (2006) projection 67 66 65 64 63 62 61 60 1990 1993 1996 1999 2002 2005 2008 2011 2014 SOURCE: Bureau of Labor Statistics and Aaronson et al. (2006). Last observation 2013. Davig and Mustre-del-Río (2013) provide some analysis of the “shadow” labor supply to gain insight into whether this group is likely to return to the labor force. The shadow group is defined as those who want a job but are not actively seeking one. The authors document that this group is demographically similar to the unemployed. They suggest that any impact on aggregate labor force participation from this group is likely to be small, because flows from this group to unemployment are small and less likely to occur as the recovery continues. I read this as also supportive of the demographics view. In a somewhat older paper, Aaronson, Fallick, Figura, Pingle, and Wascher (2006) examined the decline in labor force participation following the 2001 recession and tried to ascertain how much of the decline at that time was cyclical. It is perhaps important to recall that there was an earlier debate on declining labor force participation, long before the deep recession of 2007-09. Their paper contains as part of the analysis an empirical model of the trend labor force participation rate that includes demographic factors. If that trend model is projected forward to today from 2006, it predicts nearly exactly the labor force participation rate observed in 2012 and 2013 (Figure 4). What a great piece of out-of-sample forecasting! I read this as supportive of the demographics view. This model also projects continued decline in the labor force participation rate in the years ahead. Kudlyak (2013) follows up on the empirical model proposed by Aaronson et al. (2006). Again, the model contains key demographic information such as age, gender, and birth-year Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 7 Bullard cohort effects. The model suggests that current aggregate labor force participation rates are not far off from the model’s predicted trend participation rate. Again, I read this as supportive of the demographic view.8 Erceg and Levin (2013) study the intersection between the labor force participation rate and monetary policy. Their paper is a “thinking outside the box” exercise. In what I have presented so far, there is a certain inevitable logic. I stated that the data on labor force participation cry out for an explanation based mostly on increasing labor force participation by women and slowly changing demographics. The existing literature more or less provides such an explanation. Erceg and Levin instead ask whether there are other ways to think about this issue. They present evidence from U.S. states on prime-age males and suggest that the declines in labor force participation after the 2007-09 recession for this particular group were mostly cyclical. The authors then ask how monetary policy might be conducted in a world where labor force participation has an important cyclical component. They suggest that the participation decision should receive more attention in monetary policy research, a point on which I will agree below. I do not find the evidence on cyclical versus structural changes in labor force participation in Erceg and Levin (2013) as persuasive as the other empirical work I have reviewed.9 Labor force participation for prime-age males, for instance, has also been on a secular decline for many years. Nevertheless, Erceg and Levin’s points about how to conduct monetary policy in a world with important cyclical components in labor force participation are well made.10 In fact, I will argue that incorporating home production in economic models—as Erceg and Levin do—is where the future lies. Some authors report somewhat higher estimates of the fraction of the decline in labor force participation since 2000 due to cyclical factors. For instance, Aaronson, Davis, and Hu (2012) use still another empirical model with demographic factors included and conclude that more than half of the decline in aggregate labor force participation from 2000 to 2011 is due to cyclical factors. Van Zandweghe (2012) tries an alternative method of decomposing the data from 2007 to 2011 and concludes that more than half of the decline is cyclical. The approach used by Hotchkiss and Rios-Avila (2013) emphasizes nonlinear factors following the severe 2007-09 recession; these authors conclude that nearly all of the decline in aggregate labor force participation following the recession was cyclical. Bengali, Daly, and Valletta (2013) study the correlation in the changes in employment and labor force participation in state-level data to gain insight; they conclude that a substantial cyclical component exists in the observed aggregate decline in labor force participation. I am not necessarily swayed by these alternative approaches or results. But they certainly do show that there are many ways to cut the data and interpret the findings. This leads me to my final remarks—namely, where should the literature on labor force participation go next? HOME PRODUCTION AS THE FUTURE So far, I have reviewed some interesting economic literature on a topic that has been hot, not just among economists, but also among politicians, the media, financial markets, and even others who are not normally close students of macroeconomic developments. Much of the literature I have reviewed uses the same basic idea: Certain demographic groups have a 8 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Bullard certain propensity to participate in market work, and one of the main things we need to do as economists is project the number of people in each of these groups to determine a reasonable estimate of the expected (or “normal” or “trend”) labor force participation rate in the U.S. economy. Much of the literature concludes that demographics have contributed substantially to the observed decline in U.S. labor force participation since 2000. Still, most of the literature I reviewed is a bit hollow. Simply saying that people in certain demographic groups tend to make the participation decision one way or another does not do enough to analyze the incentives of household labor supply decisions. The more we know about the details of the household labor supply choices, including choices to participate in market work, the better we can predict the impact of policy on labor force participation. Furthermore, we would like these decisions to be part of the macroeconomic model, as Erceg and Levin (2013) suggest.11 There is one strand of the literature that does provide a more complete picture of household incentives to supply labor and participate in labor markets: the literature on so-called home production. We need not go into the details here, but the idea is simple. Think of a household as the owner of capital and labor. The household members combine their home capital— refrigerators, ovens, dishwashers, cars, houses—with their labor time to produce home goods, such as a trip, a meal, or some child care.12 These goods are not acquired in the market and are not counted in GDP, but they matter to the household. The home labor provided does not count in the aggregate statistics on labor supply. The household then has to make decisions about how much time to supply to market work versus work at home, including how many members of the household should participate in market work.13 If we were to add to a household production model more explicit treatment of household retirement decisionmaking, in addition to decisions by younger households to acquire human capital, we would get to a more complete model of the labor force participation rate. This approach is much more detailed regarding household decisionmaking than the research described above. But the extra complexity comes with a benefit, as the approach also allows macroeconomists to better understand the factors driving household labor supply decisions in terms of actual options inside the home, as well as with respect to the informal labor market. More-detailed models in this direction will likely be necessary in the future if we want to move the debate on labor force participation forward.14 Some researchers have made progress in this direction. Tripier (2004) analyzes a real business cycle version of the Diamond-Mortensen-Pissarides search and matching model that incorporates a home production decision. Tripier concludes that such a model has counterfactual implications for the unemployment and participation rates, while it can account for the behavior of the employment rate over the business cycle. Veracierto (2008) confirms these counterfactual implications in a richer model with endogenous job-acceptance and job-separation decisions. Tüzemen (2012) extends Tripier’s analysis to allow for on-the-job search and, hence, job-to-job transitions. This model performs better in matching the business cycle features of the major labor market variables. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 9 Bullard CONCLUSION While the unemployment rate has declined in recent years, labor force participation has also been declining, perhaps suggesting that unemployment is not as reliable an indicator of macroeconomic performance as it may have been in the past. Here I have given three perspectives on labor force participation: First, I reviewed the data; second, I reviewed the literature; and third, I suggested directions for future research. The post-WWII data on aggregate U.S. labor force participation show a hump-shaped pattern. Participation rose in the 1970s, 1980s, and 1990s before peaking in 2000 and heading into decline up until the present day. Current BLS projections suggest that this decline will continue over the coming decade. The rise in labor force participation is often attributed in part to the maturing of the Baby Boom generation, as well as to the increase in the number of women in the workforce. The decline has often been attributed to the aging of the U.S. labor force. A satisfactory model has to account for the rise and fall over many decades. A demographically based model would seem to have a good chance of success in explaining these data. I reviewed some of the available literature on this topic. My view of the literature is that carefully constructed, demographically based empirical models of the hump-shaped trend in the U.S. labor force participation rate do a good job of explaining the data. These models suggest that the current participation rate is not far from the predicted trend. This means, in turn, that the cyclical component in labor force participation is likely to be relatively small. To the extent these models are correct, then, the observed unemployment rate remains as good an indicator of overall labor market health as it has been historically. In particular, the recent, relatively rapid declines in unemployment can be understood as representing an improving labor market. This is the judgment that should inform monetary policy going forward. The literature is not completely satisfactory, however. I discussed how researchers might gain additional insight by including more detailed household decisionmaking inside economic models. This would allow us to better understand what motivates or deters labor force participation. I look forward to seeing future research pushing in this direction. ■ 10 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Bullard NOTES 1 For a recent analysis of the employment-to-population ratio as a labor market indicator, see Kapon and Tracy (2014). 2 Tripier (2004), Veracierto (2008), and Tüzemen (2012) take important steps in this direction. 3 See Canon, Kudlyak, and Reed (2014) for an analysis of the relative magnitude of the flows to employment from unemployment and nonparticipation. 4 See Bullard (2014). 5 See Toossi (2013). 6 Author’s calculation based on figures from Toossi (2013). 7 For more on this topic, see Canon, Debbaut, and Kudlyak (2013). 8 A forthcoming working paper by Hornstein and Kudlyak (2013) includes a more elaborate version of this model. Their main finding remains that current labor force participation rates are close to what would be predicted by an empirical model with carefully constructed demographic factors. 9 For more on this issue, see Canon, Debbaut, and Kudlyak (2013). 10 I also largely agree with the points made by Orphanides (2013) in a comment on the paper, in effect that the new labor market slack indicators proposed by Erceg and Levin would be subject to considerable uncertainty and could lead policymakers badly astray. 11 For an example of a detailed macroeconomic model with an explicit participation decision that has an impact on policy recommendations, see Imrohoroğlu and Kitao (2012). 12 Time use surveys, such as the American Time Use Survey (ATUS) conducted by the BLS, provide a wealth of data to quantify labor supplied to home production. For example, see Aguiar and Hurst (2007). 13 For an example of the different perspective the home production literature provides on issues in monetary eco- nomics, see Aruoba, Davis, and Wright (2012). 14 For an example of the interplay between home production and labor force participation, see Greenwood, Seshadri, and Yorukoglu (2005). REFERENCES Aaronson, Daniel; Davis, Jonathan and Hu, Luojia. “Explaining the Decline in the U.S. Labor Force Participation Rate.” Federal Reserve Bank of Chicago Chicago Fed Letter, No. 296, March 2012; http://www.chicagofed.org/digital_assets/publications/chicago_fed_letter/2012/cflmarch2012_296.pdf. Aaronson, Stephanie; Fallick, Bruce; Figura, Andrew; Pingle, Jonathan and Wascher, William. “The Recent Decline in the Labor Force Participation Rate and Its Implications for Potential Labor Supply.” Brooking Papers on Economic Activity, Spring 2006, 37(1), pp. 69-134; http://www.brookings.edu/~/media/projects/bpea/spring%202006/2006a_bpea_aaronson.pdf. Aguiar, Mark and Hurst, Erik. “Measuring Trends in Leisure: The Allocation of Time over Five Decades.” Quarterly Journal of Economics, August 2007, 122(3), pp. 969-1006; http://www.jstor.org/stable/25098866. Aruoba, S. Boragan; Davis, Morris A. and Wright, Randall. “Homework in Monetary Economics: Inflation, Home Production, and the Production of Homes.” NBER Working Paper No. 18276, National Bureau of Economic Research, August 2012; http://www.nber.org/papers/w18276. Bengali, Leila; Daly, Mary and Valletta, Rob. “Will Labor Force Participation Bounce Back?” Federal Reserve Bank of San Francisco Economic Letter, 2013-14, May 13, 2013; http://www.frbsf.org/economic-research/publications/economic-letter/2013/may/will-labor-force-participationbounce-back/. Bullard, James. “Ghosts of Forecasts Past and Future.” Remarks delivered at the Indiana Bankers Association Economic Outlook Forum Luncheon, Indianapolis, IN, January 10, 2014; http://research.stlouisfed.org/econ/bullard/pdf/Bullard-IN-Bankers-Association-January-10-2014-Final.pdf. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 11 Bullard Canon, Maria; Debbaut, Peter and Kudlyak, Marianna. “A Closer Look at the Decline in the Labor Force Participation Rate.” Federal Reserve Bank of St. Louis Regional Economist, October 2013, 21(4), pp. 10-11; http://www.stlouisfed.org/publications/pub_assets/pdf/re/2013/d/labor_force.pdf. Canon, Maria; Kudlyak, Marianna and Reed, Marisa. “Not Everyone Who Joins the Ranks of the Employed Was ‘Unemployed.’” Federal Reserve Bank of St. Louis Regional Economist, January 2014, 22(1), pp. 14-16; http://www.stlouisfed.org/publications/pub_assets/pdf/re/2014/a/unemployed.pdf. Davig, Troy and Mustre-del-Río, José. “The Shadow Labor Supply and Its Implications for the Unemployment Rate.” Federal Reserve Bank of Kansas City Economic Review, Third Quarter 2013, pp. 5-29; http://www.kc.frb.org/publicat/econrev/pdf/13q3Davig-Mustre.pdf. Erceg, Christopher J. and Levin, Andrew T. “Labor Force Participation and Monetary Policy in the Wake of the Great Recession.” IMF Working Paper No. 13/245, International Monetary Fund, July 2013; http://www.imf.org/external/pubs/ft/wp/2013/wp13245.pdf. Fujita, Shigeru. “On the Causes of Declines in the Labor Force Participation Rate.” Federal Reserve Bank of Philadelphia Research Rap Special Report, November 2013; revised February 6, 2014; http://philadelphiafed.org/research-and-data/publications/research-rap/2013/on-the-causes-of-declines-in-thelabor-force-participation-rate.pdf. Greenwood, Jeremy; Seshadri, Ananth and Yorukoglu, Mehmet. “Engines of Liberation.” Review of Economic Studies, January 2005, 72(1), pp. 109-33; http://restud.oxfordjournals.org/content/72/1/109.full.pdf+html. Hornstein, Andreas and Kudlyak, Marianna. “Estimating the Trend in the LFP Rate.” Federal Reserve Bank of Richmond, forthcoming working paper, December 2013. Hotchkiss, Julie L. and Rios-Avila, Fernando. “Identifying Factors behind the Decline in the U.S. Labor Force Participation Rate.” Business and Economic Research, June 2013, 3(1), pp. 257-75; http://www.macrothink.org/journal/index.php/ber/article/view/3370/2921. Imrohoroğlu, Selahattin and Kitao, Sagiri. “Social Security Reforms: Benefit Claiming, Labor Force Participation, and Long-Run Sustainability.” American Economic Journal: Macroeconomics, July 2012, 4(3), pp. 96-127; http://www.aeaweb.org/articles.php?doi=10.1257/mac.4.3.96. Kapon, Samuel and Tracy, Joseph. “A Mis-Leading Labor Market Indicator.” Federal Reserve Bank of New York Liberty Street Economics (blog), February 3, 2014; http://libertystreeteconomics.newyorkfed.org/2014/02/a-mis-leading-labor-market-indicator.html. Kudlyak, Marianna. “A Cohort Model of Labor Force Participation.” Federal Reserve Bank of Richmond Economic Quarterly, First Quarter 2013, 99(1), pp. 25-43; http://www.richmondfed.org/publications/research/economic_quarterly/2013/q1/pdf/kudlyak.pdf. Orphanides, Athanasios. “Discussion of ‘Labor Force Participation and Monetary Policy in the Wake of the Great Recession’ by Erceg and Levin.” Delivered at the Federal Reserve Bank of Boston 57th Economic Conference Fulfilling the Full Employment Mandate, April 2013. Toossi, Mitra. “Labor Force Projections to 2022: The Labor Force Participation Rate Continues to Fall.” Monthly Labor Review, December 2013, pp. 1-28; http://www.bls.gov/opub/mlr/2013/article/pdf/labor-force-projections-to-2022-the-labor-force-participationrate-continues-to-fall.pdf. Tripier, Fabien. “Can the Labor Search Model Explain the Fluctuations of Allocations of Time?” Economic Modelling, January 2004, 21(1), pp. 131-46; http://www.sciencedirect.com/science/article/pii/S0264999302000871. Tüzemen, Didem. “Labor Market Dynamics with Endogenous Labor Force Participation and On-the-Job Search.” Working Paper No. 12-07, Federal Reserve Bank of Kansas City, October 2012; http://www.kansascityfed.org/publicat/reswkpap/pdf/rwp12-07.pdf. Van Zandweghe, Willem. “Interpreting the Recent Decline in Labor Force Participation.” Federal Reserve Bank of Kansas City Economic Review, First Quarter 2012, pp. 5-34; http://www.kc.frb.org/publicat/econrev/pdf/12q1VanZandweghe.pdf. Veracierto, Marcelo. “On the Cyclical Behavior of Employment, Unemployment and Labor Force Participation.” Journal of Monetary Economics, September 2008, 55(6), pp. 1143-57; http://www.sciencedirect.com/science/article/pii/S0304393208000998. 12 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW The Great Trade Collapse and Rebound: A State-by-State View Cletus C. Coughlin During the Great Trade Collapse in the United States, which began in late 2008, one concern was that such a large collapse would transform exporting firms into strictly domestic firms or, worse, drive them out of business. In either case, it was feared that U.S. exporting might, at best, revive slowly. However, this fear about long-lived export impacts did not materialize. Clearly there were large export effects, but the sharp decline was quickly followed by a sharp rebound that began in mid-2009. In contrast to previous research, this study examines this historic episode from the perspective of individual states. A comparison of intensive and extensive trade margins reveals that the adjustment occurred to a greater extent on the intensive than on the extensive trade margin. In other words, the adjustment process entailed relatively larger changes in average exports per firm than in the number of exporting firms. It is likely much easier to alter existing export levels than to, first, eliminate all exports by a firm and, second, either restart exports by this firm or become a new entrant into exporting. The bottom line is that the U.S. export sector weathered the challenges associated with the global recession and financial crisis quite well. The fact that relatively large firms dominate U.S. exports likely contributes to the resiliency of the U.S. export sector. (JEL F10, F14, N72, R12) Federal Reserve Bank of St. Louis Review, First Quarter 2014, 96(1), pp. 13-33. n 2008, U.S. exports of goods totaled $1,147.4 billion in chained 2005 dollars (Figure 1). As part of the sudden, steep, and synchronized decline in trade worldwide, termed the “Great Trade Collapse” (GTC) by Baldwin (2009), U.S. exports declined 13.4 percent to $993.9 billion in 2009. At the same time, U.S. gross domestic product (GDP) declined 2.8 percent. This collapse was short-lived, as export growth rebounded sharply and exports in 2010 totaled $1,142.2 billion, which is roughly the 2008 level.1 This 14.9 percent growth in exports substantially exceeded the 2.5 percent increase in GDP during 2010. This episode in trade history, unprecedented in U.S. post-World War II economic history, affected countries throughout the world and has attracted the attention of many researchers.2 Most analyses have relied on country-level data and focused on explanations for the GTC and, to a lesser extent, the subsequent rebound. I Cletus C. Coughlin is senior vice president and chief of staff at the Federal Reserve Bank of St. Louis. The author appreciates the research assistance of Li Li and Diana Cooke. © 2014, The Federal Reserve Bank of St. Louis. The views expressed in this article are those of the author(s) and do not necessarily reflect the views of the Federal Reserve System, the Board of Governors, or the regional Federal Reserve Banks. Articles may be reprinted, reproduced, published, distributed, displayed, and transmitted in their entirety if copyright notice, author name(s), and full citation are included. Abstracts, synopses, and other derivative works may be made only with prior written permission of the Federal Reserve Bank of St. Louis. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 13 Coughlin Figure 1 U.S. Exports of Goods Billion, SA Chained U.S. Dollars, 2005 1,400 1,300 1,200 1,100 1,000 900 800 2006 2007 2008 2009 2010 2011 2012 NOTE: The export data are Census-based rather than balance of payments-based. SA, seasonally adjusted. Crowley and Luo (2011) examine three primary hypotheses for the GTC.3 First, and likely the key explanation, is that trade collapsed as a result of a decline in aggregate demand. The fact that trade declined proportionately far more than GDP suggests a demand shock explanation that also accounts for different types of goods, vertical specialization, and inventory adjustments is likely to be important.4,5 For example, Eaton et al. (2011) find that more than 80 percent of the decline in the ratio of trade to GDP resulted from a spending shift away from manufactured goods, particularly durable goods. A second explanation stresses the disruption in the supply of traded goods because of the increased difficulties in securing trade finance during the financial crisis. Generally speaking, most would agree that trade finance conditions deteriorated but that their impact was much less important than the decline in demand. Chor and Manova (2012) find evidence that financing difficulties were a contributing factor, but pinning down their quantitative importance is difficult, partially because of a lack of data. Small- and medium-sized firms were probably affected more by credit constraints than larger firms, which do the bulk of exporting. A third explanation focuses on increased trade barriers. Despite a frequent finding that import restrictions tend to increase during periods of economic weakness, empirical evidence suggests this explanation is quantitatively unimportant in understanding the GTC.6 The rapid rebound in trade also casts doubt on the importance of the import restrictions argument. My focus is not on the causes of the GTC, but rather on the international trade experiences of U.S. states during the collapse (2008-09) and the rebound (2009-10). Most importantly, the goal is to provide information on the relative importance of the adjustments on the extensive (i.e., the number of exporting firms) and intensive (i.e., the average exports per firm) trade margins. A close look at these margins might provide insights into reasons for the sharp rebound in export growth. Such a sharp rebound was initially viewed as unlikely by many 14 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Coughlin because the very large declines in exports during the GTC raised the possibility of long-lived adverse effects on trading relationships. Examining the extensive margin shows how the number of exporting firms changed during the collapse and rebound. Examining the intensive margin shows how the average exports of firms changed. Not surprisingly, I find changes on both margins; however, changes on the intensive margin are found to be relatively more important than changes on the extensive margin during both the collapse and the rebound. Prior to presenting the analysis, the next section provides a description of the export data used and some basic facts about states and their exporting performance. This description is followed by an analysis of trade margins at the state level. Next, a closer look at state-level exports and their respective trade margins over time and across countries is undertaken. A summary of results and conclusion complete the article. A NATIONAL OVERVIEW USING STATE EXPORT DATA Data on exports by U.S. states to foreign destinations are from the Origin of Movement series.7 These data are compiled by the Foreign Trade Division of the U.S. Census Bureau. The data in this series identify the state from which an export begins its journey to a foreign country. Beginning in 1987, the Origin of Movement series provides the current-year export sales, or free-alongside-ship costs if not sold, for all 50 U.S. states to 242 foreign destinations (generally countries).8 These export sales are for merchandise sales only and do not include services exports. Since 1992, the Foreign Trade Division of the U.S. Census Bureau has issued an annual report providing a profile of exporting companies. These reports are based on economic census and survey data on file at the Census Bureau, administrative records from other government agencies, and documents filed for export clearances. Until recently, only exporting companies were profiled, but both exporters and importers are now profiled.9 My analysis uses data for 2008, 2009, and 2010 and requires the use of state data—most importantly, state-destination data. The profile contains information on identified companies and their export values. This linkage generates what are termed “known export values.” As a result, the data in Table 1 contain entries for total exports, some portion of which cannot be linked to individual exporters, as well as the portion that can be linked.10 In 2008, the year prior to the majority of the trade collapse, identified exporters accounted for $1,150.9 billion in current-dollar exports, or 89.4 percent of the total value of exported goods. During 2009, known exports declined 18.3 percent. From 2008 to 2009, the number of identified exporters declined from 289,711 to 276,643, a decrease of 4.5 percent. Turning to the rebound, in 2010 identified exporters accounted for $1,140.4 billion in exports or 89.2 percent of the total value of exported goods.11 Relative to 2009, known exports rose 21.3 percent in 2010. Meanwhile, the number of identified exporters rose from 276,643 to 293,988, an increase of 6.3 percent. Using these national data and subsequently adjusting the export values for price changes reveals that adjustments on the intensive margin exceeded adjustments on the extensive margin during both the trade collapse and trade rebound.12 For example, during the trade collapse, Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 15 Coughlin Table 1 Exports by Identified Companies U.S. Exports and Exporters 2008 2009 2010 Total exports ($ bil.) 1,287.4 1,056.0 1,278.3 Known value (Identified exporters) ($ bil.) 1,150.9 940.4 1,140.4 Known value/Total exports (%) No. of identified exporters 89.4 89.1 89.2 289,711 276,643 293,988 SOURCE: U.S. Census Bureau. A Profile of U.S. Importing and Exporting Companies, 2010-2011 and prior reports in this series. known exports declined 13.6 percent. A straightforward decomposition of this total change shows that the number of exporting firms declined 4.5 percent and average exports per firm declined 9.6 percent.13 Thus, the decline in the intensive margin was more than double that in the extensive margin.14 A similar, but not as stark, finding pertains to the trade rebound when known exports increased 15.1 percent. Exporting firms increased by 6.3 percent and average exports per firm increased by 8.3 percent. Adjustments clearly occurred on both margins, but these findings suggest that the intensive margin accounts for the majority of the adjustments during both the collapse and the rebound. This finding is similar to the lack of destruction of trade relationships during the GTC in other countries—France (see Fontagné and Gaulier, 2009), Japan (see Wakasugi, 2009), and Belgium (see Behrens, Corcos, and Mion, 2013). This evidence is also consistent with findings by Schott (2009) for other U.S. “trade shock” episodes. Schott used detailed, firm-level trade data to analyze U.S. firms’ responses to the 2001 recession and the 1997 Asian financial crisis and found that the collapse of trade nationally was driven primarily by changes in the intensive margin. In other words, trade fell because firms sold less of what they were already selling rather than eliminating trade altogether.15 By using state-level data, I hope to provide additional insights as to the regularity of this finding across states. I highlight some well-known facts about exporting firms and export markets to provide additional background for the analysis. Despite being outnumbered by smaller firms, relatively large firms dominate U.S. exports. In fact, 287,498 (97.8 percent) of the total 293,988 exporters in 2010 had fewer than 500 employees, while 6,490 exporters (2.2 percent) had 500 or more employees. Nonetheless, large firms (>500 employees) accounted for 66.2 percent of known exports, while relatively small firms (<500 employees) accounted for the remaining 33.8 percent of known exports. In addition, a relatively small number of firms account for the majority of exports. For 2010, the top 50 exporters (roughly 0.02 percent of all exporters) accounted for 29.0 percent of all known exports, and the top 2,000 exporters (roughly 0.68 percent of all exporters) accounted for 76.9 percent of all known exports. Not surprisingly, relatively large exporters tend to export to more countries than their relatively smaller counterparts. Slightly more than one in four large exporters (26.1 percent) shipped to only one country. These exports accounted for 0.4 percent of the exports of large 16 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Coughlin exporters. Meanwhile, smaller exporters were more likely to ship to only one country; as a result, the percentages of these firms and their export shares were larger than those of their larger counterparts. As the number of export destinations increases, the relative importance of large exporters increases. For 2010, 69.3 percent of the exports of large firms were shipped to 50 or more countries. Meanwhile, very few small exporters (roughly 430, or 0.15 percent of all small exporters) shipped to 50 or more countries. Finally, the top 25 export destinations account for most U.S. exports. For 2010, the top 25 markets accounted for 82.9 percent of U.S. exports.16 A similar percentage of U.S. exporters, 81.8 percent, ship to these 25 markets. Not surprisingly, large exporters handle the majority (67.4 percent) of these exports, despite representing a relatively small share (2.6 percent) of the total number of exporters supplying these 25 markets. Ideally, for the analyses in this article, individual firm data would be used to allow examination of the trading behavior of individual firms. Such data would allow observation of the beginning or ending of a firm’s trade involving a specific good or destination and the change in the value of a firm’s ongoing trade involving a specific good or destination. In the present case, I have state-level data, so I am limited to using the number of firms in a state that export to a specific country. The state-level data limitation highlights the distinction between plants and firms. Because a firm with plants in different states could export to a given country from plants in different states, the sum of the number of exporters over all states would exceed the number of exporters nationally. Also, a firm with plants in multiple states might export to different countries depending on the plant. If the exporters over all countries were summed, this sum would exceed the number of exporters at the national level. Thus, the analysis here combines single-plant firms with multiplant firms. Why might combining single- and multiplant firms matter? The response to a decline (or an increase) in foreign demand can differ between a single-plant firm and a multiplant firm. For example, assume a firm has plants in two states and that both plants produce exports for the same foreign country. In response to a decline in foreign demand, the firm might choose to serve the foreign country from one plant rather than two. As a result, from the firm’s perspective, the adjustment is completely on the intensive margin, while from a plant’s perspective, the adjustment differs across states. From the perspective of one state, there is a decline on the extensive margin and an uncertain change on the intensive margin, while from the perspective of the other state, there is no change on the extensive margin. Any change is on the intensive margin. The key point of this example is that the dynamics based on firms can differ from the dynamics based on plants. Any interpretation of the results in this article must keep this possibility in mind. Table 2 contains summary information, by state, for the number of foreign destinations with export activity.17 Virtually every state serves the top 30 export markets. Moreover, even during the GTC, more often than not the number of export destinations tended to increase. Relative to 2008, 27 states in 2009 experienced an increase in export destinations, 18 states a decrease, and 5 states no change. Especially noteworthy is that every state in New England added 30 or more export destinations. Only one state, North Dakota, suffered a double-digit Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 17 Coughlin Table 2 Number of Export Destinations by State State Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming Total level 2008 Change 2008-09 Total level 2009 Change 2009-10 Total level 2010 129 34 143 101 184 128 138 87 178 165 44 84 166 142 120 125 117 137 96 157 156 139 150 108 131 61 101 109 106 166 79 172 157 79 162 118 124 161 95 144 64 144 176 123 86 162 158 76 148 43 –9 2 –7 –7 1 –2 37 –2 –1 2 –2 –2 20 33 0 –1 9 1 53 –8 30 41 1 0 1 –3 6 2 46 24 3 19 –1 –11 27 –4 1 27 41 1 0 0 –1 5 43 –3 –6 –2 32 0 120 36 136 94 185 126 175 85 177 167 42 82 186 175 120 124 126 138 149 149 186 180 151 108 132 58 107 111 152 190 82 191 156 68 189 114 125 188 136 145 64 144 175 128 129 159 152 74 180 43 48 49 43 53 8 49 –2 49 15 22 43 55 2 –5 45 49 39 36 9 30 –2 –5 28 48 37 42 47 37 –2 –5 57 –2 27 48 –4 46 52 4 –1 28 44 38 15 42 1 26 34 36 –2 48 168 85 179 147 193 175 173 134 192 189 85 137 188 170 165 173 165 174 158 179 184 175 179 156 169 100 154 148 150 185 139 189 183 116 185 160 177 192 135 173 108 182 190 170 130 185 186 110 178 91 SOURCE: Author’s calculations using WISER data. 18 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Coughlin Figure 2 Export Destinations and State Size (2010) No. of Countries 210 190 170 150 130 110 90 70 50 9 10 11 12 13 14 15 ln(gross state product) SOURCE: WISER and Bureau of Economic Analysis. loss (11) in the number of export destinations. Not surprisingly, in light of the rebound, the vast majority of states (40) served more export destinations in 2010 than in 2009. No state suffered a loss of more than five export destinations. Figures 2 through 4 show some noteworthy differences across states. Figure 2, which uses 2010 data, shows the positive relationship between state size (gross state product in natural logarithms) and the number of destination countries.18 In other words, larger states tend to export to more countries. The three largest U.S. states—California, New York, and Texas— export to many more countries than the five smallest states—South Dakota, Wyoming, Montana, North Dakota, and Vermont. Figure 2 also shows that the rate of increase in the number of destination countries decreases as state size increases. For the largest states, the number of destination countries approaches 200. Figure 3 shows that larger foreign countries (higher GDPs in natural logarithms) tend to import from a larger number of U.S. states. Moreover, once a country’s GDP reaches a certain size, it generally imports from all 50 states.19 All countries with GDPs larger than $401 billion import from every state, while the 10 countries with the smallest GDPs imported from an average of slightly more than 21 states. Figure 4 shows the number of exporters in a state is related positively to state size (gross state product in natural logarithms).20 For example, the number of exporters in the largest states—California (72,092 exporters), New York (40,377), and Texas (38,276)—exceeds the number in the smallest states—South Dakota (965 exporters), Wyoming (421), Montana (1,539), North Dakota (1,870), and Vermont (1,244)—by a factor of more than 20. Figure 4 Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 19 Coughlin Figure 3 States and Export Destination Size (2010) No. of States 50 45 40 35 30 25 20 15 10 5 0 15 17 19 21 23 25 27 29 ln(GDP) SOURCE: WISER and the World Bank. Figure 4 Exporters and State Size (2010) No. of Exporting Firms in a Given State 80,000 70,000 60,000 50,000 40,000 30,000 20,000 10,000 0 10 11 12 13 14 15 ln(gross state product) SOURCE: U.S. Census Bureau and the World Bank. 20 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Coughlin also reveals that a given change in gross state product is associated with increases in exporters at an increasing rate. TRADE MARGINS DURING THE GREAT TRADE COLLAPSE AND REBOUND As I did for the decomposition of the percentage change in exports at the national level in the previous section, here I undertake the same calculation at the level of individual states. First, I examine changes during the 2008-09 GTC and, second, I examine changes during the 2009-10 trade rebound.21 During the GTC, known exports declined 13.6 percent nationally; all states but Alaska and Utah experienced a decline in known exports.22 Exports from Alaska increased 3.2 percent and exports from Utah increased 6.5 percent. Meanwhile, exports from New Mexico declined the most: –53.7 percent. As shown in Table 3, decreasing exports resulted from declines on both the extensive and intensive margins for most states. For example, the number of exporting firms declined in all but four states—Delaware, Louisiana, Maryland, and Rhode Island. The number of exporting firms declined even in the two states in which exports increased. Not surprisingly, large states, such as California and New York, suffered the largest absolute declines in exporters, with each state losing more than 2,000 exporting firms. Small states, such as Montana, South Dakota, and North Dakota, suffered the largest percentage declines in exporters. Each of these states experienced a decline in exporting firms of more than 14 percent. While the number of exporting firms generally declined across states during the GTC, exports per firm across states also tended to decline. Exports per firm declined in all but nine states—Alaska, Arkansas, Connecticut, Kentucky, Nebraska, Nevada, North Dakota, Utah, and Vermont. Alaska, with a 14.9 percent increase in exports per firm, had the largest percentage increase, while New Mexico, with a 51.6 percent decline, had the largest decrease. In comparing the percentage changes in the extensive and intensive margins, declines on the intensive margin exceed those on the extensive margin in 33 of the 50 states. Thus, not surprisingly in light of the national numbers, relatively more of the adjustment occurs in terms of exports per firm than in the numbers of exporters.23 Table 4 restates the results in Table 3 to highlight the ranking of states based on the relative importance of the percentage change in the intensive margin as a share of the percentage change in exports. During the GTC, seven states show percentage changes in the intensive margin relative to exports that are 1 or larger. The values for Alaska and Utah reflect the importance of the positive changes in the intensive margin in accounting for increased exports. The values for Delaware, Maryland, Rhode Island, Louisiana, and Florida reflect the importance of the negative changes in the intensive margin in accounting for the decrease in exports. Exports rose in nearly all states during the 2009-10 rebound. Export growth was most pronounced in Maine and New Hampshire, where growth exceeded 36 percent. The exceptions to positive export growth were Arkansas, Nevada, and Washington. Exports from Arkansas declined almost 11 percent. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 21 Coughlin Table 3 Decomposition of Trade Changes GTC (2008-09) State United States Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming Extensive margin –4.5 –5.0 –10.2 –3.7 –8.0 –3.0 –5.6 –6.1 10.2 0.0 –4.0 –4.3 –10.8 –5.2 –5.5 –8.6 –2.3 –5.3 0.3 –8.2 2.4 –4.4 –6.7 –5.2 –4.3 –6.6 –17.4 –7.1 –5.7 –7.3 –4.4 –4.3 –5.6 –2.7 –14.6 –5.5 –5.4 –7.0 –5.5 1.6 –6.5 –16.0 –4.5 –1.4 –2.2 –9.3 –7.2 –4.9 –3.8 –3.9 –8.3 Intensive margin –9.6 –14.4 14.9 –22.3 5.8 –8.8 –15.3 5.5 –14.0 –8.4 –4.9 –10.7 –3.4 –13.8 –2.9 –13.4 –27.6 3.1 –18.6 –16.2 –16.1 –7.8 –17.2 –9.8 –10.3 –16.0 –1.6 1.5 5.4 –5.3 –13.4 –51.6 –14.4 –5.9 1.7 –17.9 –2.4 –11.2 –7.7 –23.0 –4.7 –25.7 –1.3 –9.2 8.9 2.5 –7.3 –14.2 –10.2 –9.9 –1.8 Joint change 0.4 0.7 –1.5 0.8 –0.5 0.3 0.9 –0.3 –1.4 0.0 0.2 0.5 0.4 0.7 0.2 1.2 0.6 –0.2 –0.1 1.3 –0.4 0.3 1.2 0.5 0.4 1.0 0.3 –0.1 –0.3 0.4 0.6 2.2 0.8 0.2 –0.3 1.0 0.1 0.8 0.4 –0.4 0.3 4.1 0.1 0.1 –0.2 –0.2 0.5 0.7 0.4 0.4 0.2 Rebound (2009-10) Change in Extensive exports (%) margin –13.6 –18.6 3.2 –25.2 –2.6 –11.5 –20.0 –0.9 –5.3 –8.4 –8.7 –14.5 –13.8 –18.3 –8.2 –20.8 –29.2 –2.3 –18.3 –23.0 –14.1 –11.9 –22.8 –14.5 –14.2 –21.5 –18.7 –5.7 –0.7 –12.2 –17.2 –53.7 –19.2 –8.4 –13.1 –22.4 –7.7 –17.4 –12.8 –21.8 –10.9 –37.6 –5.7 –10.5 6.5 –7.1 –13.9 –18.4 –13.5 –13.4 –10.0 6.3 1.3 12.4 10.8 12.1 6.1 9.3 6.2 12.7 5.5 5.3 9.3 13.7 5.2 5.5 10.6 6.9 6.9 5.2 5.9 14.1 6.4 5.4 7.2 7.5 6.4 16.4 7.0 8.0 10.7 1.8 6.6 5.9 8.0 17.2 5.4 2.9 7.2 5.8 0.2 5.4 7.1 5.0 6.5 7.8 4.3 3.2 6.7 8.1 4.1 11.7 Intensive margin 8.3 23.8 5.8 –3.5 –20.4 7.1 0.5 1.7 –6.0 6.8 9.5 1.3 10.8 9.4 12.9 3.2 –1.6 –1.4 13.2 30.9 –7.6 –0.3 24.1 7.7 22.2 22.2 7.8 7.4 –10.0 23.6 9.0 5.1 7.9 1.1 –6.5 14.6 12.2 3.4 10.9 26.9 10.7 9.8 14.9 14.5 12.1 24.0 4.7 –7.8 16.9 7.8 –7.0 Joint change 0.5 0.3 0.7 –0.4 –2.5 0.4 0.0 0.1 –0.8 0.4 0.5 0.1 1.5 0.5 0.7 0.3 –0.1 –0.1 0.7 1.8 –1.1 0.0 1.3 0.6 1.7 1.4 1.3 0.5 –0.8 2.5 0.2 0.3 0.5 0.1 –1.1 0.8 0.4 0.2 0.6 0.0 0.6 0.7 0.7 0.9 0.9 1.0 0.1 –0.5 1.4 0.3 –0.8 Change in exports (%) 15.1 25.5 18.9 6.9 –10.8 13.7 9.9 7.9 6.0 12.6 15.3 10.7 26.0 15.2 19.1 14.1 5.2 5.4 19.1 38.7 5.4 6.1 30.8 15.5 31.4 30.0 25.4 15.0 –2.7 36.8 10.9 12.1 14.2 9.2 9.7 20.9 15.5 10.9 17.4 27.1 16.7 17.6 20.7 21.9 20.8 29.3 8.1 –1.6 26.4 12.2 3.8 SOURCE: Author’s calculations using U.S. Census data. 22 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Coughlin Table 4 Relative Importance of Intensive Margin Changes Intensive Margin Relative to Percent Change in Exports State Alaska Delaware Utah Maryland Rhode Island Louisiana Florida New Mexico Kansas Arizona Texas Ohio New Jersey Alabama Washington Colorado California Illinois Michigan West Virginia New York Missouri Wisconsin Hawaii Mississippi United States Maine North Carolina South Dakota Minnesota Massachusetts Iowa Oregon Pennsylvania Georgia Virginia New Hampshire South Carolina Indiana Oklahoma Idaho Tennessee Wyoming Montana North Dakota Nebraska Vermont Kentucky Arkansas Connecticut Nevada Federal Reserve Bank of St. Louis REVIEW GTC (2008-09) 4.70 2.66 1.37 1.14 1.06 1.01 1.00 0.96 0.94 0.89 0.88 0.80 0.78 0.77 0.77 0.76 0.76 0.76 0.76 0.75 0.75 0.74 0.74 0.73 0.73 0.71 0.70 0.70 0.68 0.68 0.66 0.64 0.64 0.60 0.57 0.52 0.43 0.43 0.35 0.32 0.25 0.22 0.18 0.09 –0.13 –0.26 –0.34 –1.35 –2.20 –5.89 –8.08 Origin Washington Nevada Arkansas Rhode Island Alabama New Jersey Vermont Maine Oklahoma Michigan Missouri Tennessee Mississippi Ohio Louisiana Indiana Texas South Carolina New Hampshire Wisconsin West Virginia Pennsylvania Illinois Georgia Virginia Utah South Dakota New York United States Florida California Minnesota Nebraska New Mexico Idaho Oregon Alaska Montana Iowa Connecticut Hawaii North Carolina Colorado Massachusetts Kentucky Kansas Arizona North Dakota Delaware Maryland Wyoming Rebound (2009-10) 4.97 3.64 1.89 0.99 0.93 0.82 0.82 0.80 0.79 0.78 0.74 0.72 0.71 0.70 0.69 0.68 0.66 0.64 0.64 0.64 0.64 0.63 0.62 0.62 0.59 0.58 0.56 0.55 0.55 0.53 0.52 0.50 0.50 0.42 0.41 0.31 0.31 0.30 0.23 0.21 0.12 0.12 0.05 –0.04 –0.26 –0.31 –0.51 –0.67 –1.00 –1.42 –1.83 First Quarter 2014 23 Coughlin As shown in Table 3, rising exports resulted from increases on the extensive margin for all states and increases on the intensive margin for most states. The growth in the number of exporting firms exceeded 10 percent in 11 states—Alaska, Arizona, Arkansas, Delaware, Idaho, Iowa, Maryland, Montana, New Hampshire, North Dakota, and Wyoming. Results for the intensive margin reveal that 39 of the 50 states experienced increases; in 8 states the increases exceeded 22 percent. Maine led the way with an increase of more than 30 percent, which accounts for its overall large percentage increase in exports. For the 11 states with declines in their intensive margins, the 20.4 percent decline in Arkansas was more than double the decline in Nevada, which experienced the second-largest decline. Obviously, the decline in Arkansas is the key factor accounting for the overall decline in its exports. When the percentage changes in the extensive and intensive margins are compared, increases on the intensive margin exceeded those on the extensive margin in 29 states during the rebound. Thus (not surprisingly) in light of the national numbers and similar to the finding during the GTC, relatively more of the adjustment occurred in terms of exports per firm than in the numbers of exporters.24 Table 4 ranks states based on the relative importance of the percentage change in the intensive margin as a share of the percentage change in exports. Washington, Nevada, and Arkansas, the exceptions to positive growth in known exports during the rebound, lead the way. Rhode Island and Alabama are the fourth- and fifth-ranked states. In both cases, positive changes in the intensive margin account for more than 90 percent of the change in exports. TRADE DURING THE COLLAPSE AND REBOUND: TIME, SIZE, GEOGRAPHIC, AND DEMAND PERSPECTIVES So far, I have explored the trade collapse and rebound separately without regard to possible connections over time or over space. This section presents an elementary-level exploration of some of the many possible relationships, beginning with a focus on time and then considering other perspectives. Time Perspective Are there any obvious connections between the two periods with respect to time? For example, do states with relatively larger trade collapses have relatively larger trade rebounds? At most, I find an association for such a relationship that is very weak and not statistically significant. The correlation between the percentage changes in known exports across states for the two periods is –0.16, which is not statistically significant at the 5 percent level. For total exports, the correlation coefficient is –0.20, which is also not statistically significant at the 5 percent level. Similarly, I find no statistically significant association for the percentage change in the intensive margin across states for the two periods. The correlation coefficient is –0.16. However, with a correlation coefficient of –0.28, I do find a statistically significant association for the percentage change in the extensive margin across states for the two periods. In other words, states with relatively larger declines in their extensive margins experienced relatively larger rebounds in this margin. 24 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Coughlin Size Perspective This section explores whether there are differences during either period based on state size. State size does not appear to be associated with the change in exports. For example, the correlation between state GDP in 2008 and the percentage change in total exports during the collapse is 0.04, and the correlation between state GDP in 2009 and the percentage change in total exports during the rebound is –0.14. Neither is statistically significant at the 5 percent level. No statistically significant relationship is found between the intensive margin and state size for either the collapse or the rebound. The correlation coefficient is –0.07 for the collapse and 0.02 for the rebound, neither of which is statistically significant at the 5 percent level. Meanwhile, state size is associated with the extensive margin. During the collapse, larger states tended to have relatively smaller declines in their extensive margins; similarly, during the rebound, larger states tended to have relatively smaller increases in their extensive margins. In the former case, the correlation coefficient is 0.24, while in the latter case, the correlation coefficient is –0.26. While neither case is statistically significant at the 5 percent level, both are statistically significant at the 10 percent level. Geographic Perspective One way to assess the similarity in the changes in trade margins and exports of nearby states from a geographic perspective is to calculate the Moran’s I spatial statistic. This statistic, with a range of +1 to –1, indicates clustering when its value is close to +1 and dispersion when its value is close to –1. If states with high values are located near other states with high values and states with low values are located near other states with low values, then the associated Moran’s I value will be close to 1. Meanwhile, if a state with a high value tends to be located near a state with a low value (and vice versa), then the associated Moran’s I value will be close to –1. If there is no pattern (i.e., random), then the value will be near zero.25 I use the trade margin measures in Table 3 and the percentage changes in total and known exports during both the collapse and the rebound to calculate the associated Moran’s I statistic for a number of cases. For the intensive margin, I find a random geographic distribution across the 48 contiguous U.S. states.26 In other words, the Moran’s I values are close to zero: –0.10 for 2008-09 and –0.06 for 2009-10, suggesting no statistical association. The maps in Figures 5 and 6 reflect this lack of geographic association for the intensive margin. Figure 5 shows the quintile distribution by state on the intensive margin for the trade collapse, while Figure 6 pertains to the rebound. The lightest color shows the states in the lowest (smallest values) quintile, while the darkest color shows the states in the highest (largest values) quintile. A positive association would be suggested by a clustering of states with the same color, while a negative association would be suggested by states with the lightest color that are contiguous to states with the darkest color. No association would be suggested by a random distribution of the colors of the states. Both maps reveal a random distribution. For example, in both Figures 5 and 6, states with the lowest values are scattered throughout the country; in no instance are more than two of these states contiguous. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 25 Coughlin For the extensive margin, during both the collapse and the rebound, I find a positive association at the 0.05 percent statistical significance level. For the collapse, the Moran’s I value is 0.33; for the rebound, the value is 0.18. The maps in Figures 7 and 8 show this positive association as states of a similar color tend to cluster. For example, in Figure 7, states with the lowest values—Maine, New Hampshire, and Vermont plus Idaho, Montana, Wyoming, North Dakota, and South Dakota—tend to cluster. The clustering of Illinois, Indiana, Ohio, Pennsylvania, Kentucky, Tennessee, and Alabama is also easily observable in the third quintile. In Figure 8, the clustering of Michigan, Indiana, Ohio, Pennsylvania, and New York is easily observable in the second quintile. In addition, states in the three highest quintiles tend to be located west of the Mississippi River, while states in the two lowest quintiles tend to be located east of the Mississippi River. Determining whether the percentage changes in exports cluster, regardless of using total or known exports for either of the periods, reveals values close to zero, indicating no statistically significant association. For example, for the collapse, the Moran’s I value is –0.12 for the percentage change in total exports and –0.11 for the percentage change in known exports. For the rebound, the corresponding values are –0.14 and –0.09. Recall that the percentage changes in the intensive margin were relatively more important than the extensive margin in accounting for the percentage change in known exports, so finding that the pattern for the extensive margin does not lead to a pattern for known exports is not surprising. Demand Perspective As a final topic, I explore the impact of foreign demand on state exports. Consistent with the explanation that a decline in aggregate demand played the key role in the GTC, state exports would be expected to be tied to changes in the economic performance of a state’s trading partners. A simple measure of the change in the economic performance of a state’s trading partners is the trade-weighted growth of GDP. In other words, weighting each trading partner’s growth in 2009 (2010) by its share of a state’s trade in 2008 (2009) provides a measure of the overall performance of these trading partners. A reasonable expectation is that a state’s export growth in a specific year would be related positively to this summary measure of economic performance. Using known exports, I find correlation coefficients of 0.11 for the trade collapse and 0.12 for the trade rebound; however, for neither period do I find statistically significant relationships. I also explore the possible connection between both the intensive and extensive margins and the trade-weighted growth of the state’s trading partners.27 I do not find statistically significant relationships between trade-weighted growth and percentage changes in the intensive margin for either period. The correlation is –0.08 for the collapse and 0.09 for the rebound. With respect to the extensive margin, a correlation coefficient of 0.48 suggests the larger trade-weighted declines in foreign growth were associated with larger percentage declines in the extensive margin during the collapse, but the correlation only slightly exceeds zero during the rebound. 26 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Coughlin Figure 5 Intensive Margin (Distribution of Percentage Change in Average Exports per Firm, 2008-09) Change (%) –51.6 to –16.1 –16.1 to –11.2 –11.2 to –7.3 –7.3 to –1.6 –1.6 to 8.9 Figure 6 Intensive Margin (Distribution of Percentage Change in Average Exports per Firm, 2009-10) Change (%) –20.4 to –1.4 –1.4 to 7.1 7.1 to 10.7 10.7 to 14.9 14.9 to 30.9 Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 27 Coughlin Figure 7 Extensive Margin (Distribution of Percentage Change in Number of Exporting Firms, 2008-09) Change (%) –17.4 to –7.3 –7.3 to –5.6 –5.6 to –4.5 –4.5 to –3.0 –3.0 to 10.2 Figure 8 Extensive Margin (Distribution of Percentage Change in Number of Exporting Firms, 2009-10) Change (%) 0.2 to 5.2 5.2 to 6.1 6.1 to 7.1 7.1 to 10.6 10.6 to 17.2 28 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Coughlin SUMMARY AND CONCLUSION The GTC and its aftermath provide an excellent opportunity to examine the resiliency of the U.S. export sector. In contrast to previous studies, I examine this historic episode from the perspective of individual states. One concern during the GTC was that such a large collapse would transform firms from exporters into strictly domestic firms or, worse, drive them out of business. In either case, it was feared that U.S. exporting might, at best, revive slowly. Obviously, this fear about long-lived export impacts did not materialize. Thus, the U.S. experience during the GTC was similar to prior trade declines. Clearly, large export effects occurred, but the sharp decline during the GTC was quickly followed by a sharp rebound. In both periods, the percentage change in exports far exceeded the contemporaneous percentage change in GDP. Examining the intensive and extensive trade margins shows that the adjustment occurred to a greater extent on the intensive than on the extensive trade margin. In other words, the adjustment process entailed relatively larger changes in average exports per firm than in the number of exporting firms. As indicated by the analysis of individual states, this finding at the national level is consistent with the results for the majority of states. It is likely much easier to alter existing export levels than to, first, eliminate all exports by a firm and, second, either restart exports by this firm or become a new entrant into exporting. The findings in this article, which rely on state data, are consistent with previous work based on national data for numerous countries, including the United States, France, Japan, and Belgium. Moreover, the findings are consistent with current international trade theory that emphasizes productivity differences across firms and the importance of the additional costs that must be incurred to engage in international trade.28 Obviously, the more productive the firm, the more likely it will find exporting to be profitable. The increased costs associated with exporting include transportation costs, import duties, legal fees, marketing fees, and the development of distribution networks. Some of the increased costs can be viewed as market entry costs. Large and sunk market entry costs deter firms from exiting a foreign market, especially if a decline in demand is viewed as temporary. Rather than exiting, firms scale back their operations and wait for better times. When better times arrive, the firms ramp up their exports. With respect to additional results, our examination of the time-related adjustment process revealed little association between the magnitudes of the trade collapses and the trade rebounds. In other words, I found states with the relatively larger collapses did not experience relatively larger (or smaller) rebounds. Similarly, I found no significant association between percentage changes in the intensive margins between the two periods. With respect to the extensive margin, I did find that relatively larger collapses were associated with relatively larger rebounds. I found little association between state size and either the trade collapse or rebound experienced by states. I also found state size was not associated with the extent of the changes in the intensive margin but was associated with the extensive margin. Relatively larger states tended to have relatively smaller declines in their extensive margin during the trade collapse and smaller increases in their extensive margin during the trade rebound. In comparing the similarity of a state’s experience with those of its neighbors, I find no clustering in terms of percentage changes in exports during either the trade collapse or rebound. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 29 Coughlin Similar to my previous results, I find no clustering of states with respect to the intensive margin, but I do find that the extensive margin experiences of nearby states are similar. Finally, the growth of a state’s trading partners does not systematically affect export growth during either the collapse or the rebound. This result might be caused by my reliance on annual data. It is possible that export changes lag the growth changes. Moreover, in addition to income, there may be other changes that obscure the impact of foreign growth. Similarly, an identified systematic relationship is lacking at the intensive margin during both periods. With respect to the extensive margin, I found that larger declines in foreign growth were associated with larger declines in the extensive margin during the collapse, but larger increases in foreign growth were not associated with larger increases in the extensive margin during the rebound. The bottom line is that the U.S. export sector weathered the challenges associated with the global recession and financial crisis quite well. The fact that relatively large firms dominate U.S. exports likely contributes to the resiliency of the U.S. export sector. ■ 30 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Coughlin NOTES 1 The export data are Census-based rather than balance of payments-based. Annual data are highlighted because data for the analyses of trade margins at the state level are available only annually. Quarterly data show that U.S. exports of goods peaked in the third quarter of 2008. Exports declined during each of the next three quarters and then began to recover during the second half of 2009. By the fourth quarter of 2010, exports were slightly higher than the previous peak. 2 As noted in Baldwin (2009), the GTC is the steepest decrease in recorded history. Between World War II and the GTC, global trade declined three times: in 1974-75, 1982-83, and 2001-02. Bussière et al. (2013) state that on a year-over-year basis global trade declined more than 10 percent in real terms in 2009, while global real GDP fell 0.6 percent. 3 For a more technical examination of these hypotheses, see Bems, Johnson, and Yi (2013). Their conclusions match those of Crowley and Luo. 4 See Levchenko, Lewis, and Tesar (2010), Eaton et al. (2011), and Bussière et al. (2013) for analysis of compositional effects (e.g., the decline in demand was likely skewed toward tradable goods) and vertical specialization effects (e.g., firms spread their production processes across countries, so that the value of trade, entailing mostly intermediate goods, will necessarily exceed value added or GDP). In the latter case, declining demand is multiplied because of the increasing role of international supply chains. See Allesandria, Kaboski, and Midrigan (2010) for an analysis of inventory adjustments. 5 See Contessi and de Nicola (2013) for a review of the theoretical and empirical relationships between finance and international trade. 6 See Crowley and Luo (2011) and Bown and Crowley (2013). Eaton et al. (2011) also conclude that increased trade barriers had only a minimal effect on global declines in the ratio of trade to GDP. 7 Other studies using the Origin of Movement series include those by Smith (1999), Coughlin and Wall (2003), Coughlin (2004), Cassey (2009, 2011), and Coughlin and Novy (2013). 8 “Free alongside ship” means that the goods are delivered to a port to the point of loading aboard a vessel for shipment. Thus, the cost of the goods does not include the costs of moving the goods from alongside the vessel to the foreign buyer. 9 The most recent annual report is titled A Profile of U.S. Importing and Exporting Companies, 2010-2011 (http://www.census.gov/foreign-trade/Press-Release/edb/2011/edbrel.pdf). Two years are reported each year; the older year contains revised data and the more recent year contains data subject to revision. 10 In other words, if the individual exporter can be identified, then the exports are said to be known. Total exports are determined by adding known exports and the exports for which no specific exporter can be identified. 11 Because the share of known to total exports is virtually constant over 2008-10, the percentage changes in total and known exports are very similar: –18.3 versus –18.0 for 2008-09 and 21.3 versus 21.1 for 2009-10. 12 Our definitions of trade margins follow those of Lawless (2010) and Eaton, Kortum, and Kramarz (2004). As dis- cussed by Lawless (2010), alternative definitions of these terms have appeared in the literature. Moreover, in the context of multiproduct firms, within-firm margins related to the number of products exported and average exports across products could be explored. 13 The decomposition of the percentage change in exports is straightforward: %∆X y1,y0 = [(∆Fy1,y0 * Xy0 /Fy0)/Xy0] * 100 + [(((Xy1/Fy1) – (Xy0 /Fy0)) * Fy0)/Xy0] * 100 + [(∆Fy1,y0 * ((Xy1/Fy1) – (Xy0 /Fy0))/Xy0 ] * 100, where X is the priceadjusted level of known exports, F is the number of identified exporters, and the subscripts identify the years for the calculation. The first of the three right-hand-side terms is the extensive margin, the second is the intensive margin, and the third is the joint effect. 14 The joint effect of the changes in the two margins is 0.4 percent. 15 Such a finding is consistent with what is termed “hysteresis in trade”; see Baldwin (1988). 16 Beginning with the destination for the most U.S. exports, the top 25 U.S exports destinations in 2010 were Canada, Mexico, China, Japan, the United Kingdom, Germany, Korea, Brazil, the Netherlands, Singapore, France, Hong Kong, Taiwan, Belgium, Australia, Switzerland, India, Italy, Malaysia, Colombia, the United Arab Emirates, Saudi Arabia, Israel, Chile, and Venezuela. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 31 Coughlin 17 See Cassey (2011) for a more thorough summary of state export patterns. 18 The simple correlation between state size in natural logarithms and the number of destination countries is 0.86. Using state size in levels, the simple correlation is 0.57. Both correlations are statistically significant at the 5 percent level. 19 The simple correlation between GDP in natural logarithms and the number of states that export to a given country is 0.70, which is statistically significant at the 5 percent level. Meanwhile, using GDP in levels, the simple correlation is 0.26, which is also statistically significant at the 5 percent level. 20 The simple correlation between state size as measured by gross state product in natural logarithms and the num- ber of exporting firms of plants in the state is 0.76. Using state size in levels, the simple correlation is 0.94. Both correlations are statistically significant at the 5 percent level. 21 Depending on the situation, I use state-level total and known exports in my analysis. Simple correlations of total and known exports across states are virtually 1 for 2008, 2009, and 2010. The simple correlations of percentage changes in total and known exports across states for 2008-09 and 2009-10 are 0.89 and 0.97, respectively. This suggests that total and known exports are interchangeable for purposes of this article. 22 Utah is the only state that experienced an increase in both total and known exports. Total exports declined in Alaska during the collapse. All other states experienced declines in both total and known exports. 23 Simple correlations between the percentage changes in trade margins and exports reinforce this conclusion. The correlation between the intensive margin and exports is 0.93, which is statistically significant at the 5 percent level, while the correlation between the extensive margin and exports is 0.13, which is not statistically significant at the 5 percent level. 24 Simple correlations between the percentage changes in trade margins and exports reinforce this conclusion. The correlation between the intensive margin and exports is 0.94, which is statistically significant at the 5 percent level, while the correlation between the extensive margin and exports is –0.20, which is not statistically significant at the 5 percent level. 25 A simple illustration is a checkerboard with white and black squares. Normally, the white and black squares are dispersed yielding a Moran’s I value close to –1. However, if all the white squares were placed together on one side of the board and the black squares on the other side, then the associated Moran’s I value would be close to 1. A random arrangement would yield a value close to zero. 26 Alaska and Hawaii are omitted because they have no contiguous states. 27 Increases in a foreign country’s income, holding other things constant, should tend to increase a state’s total exports to the country. However, as demonstrated by Lawless (2010), the effect in theory on the intensive margin is ambiguous. By inducing the entry of new exporting firms, average exports per firm may increase, decrease, or remain unchanged. Because of the induced entry of new exporters, the extensive margin in a state should increase as a foreign country’s income increases. 28 See Melitz (2003). REFERENCES Allesandria, George; Kaboski, Joseph P. and Midrigan, Virgiliu. “The Great Trade Collapse of 2008-09: An Inventory Adjustment?” IMF Economic Review, 2010, 58(2), pp. 254-94. Baldwin, Richard. “Hysteresis in Import Prices: The Beachhead Effect.” American Economic Review, September 1988, 78(4), pp. 773-85. Baldwin, Richard, ed. The Great Trade Collapse: Causes, Consequences, and Prospects. VoxEU.Org eBooks, 2009; http://www.voxeu.org/epubs/cepr-reports/great-trade-collapse-causes-consequences-and-prospects. Behrens, Kristian; Corcos, Gregory and Mion, Giordano. “Trade Crisis? What Trade Crisis?” Review of Economics and Statistics, May 2013, 95(2), pp. 702-09. Bems, Rudolfs; Johnson, Robert C. and Yi, Kei-Mu. “The Great Trade Collapse.” Annual Review of Economics, 2013, 5, pp. 375-400. 32 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Coughlin Bown, Chad P. and Crowley, Meredith A. “Import Protection, Business Cycles, and Exchange Rates: Evidence from the Great Recession.” Journal of International Economics, May 2013, 90(1), pp. 50-64. Bussière, Matthieu; Callegari, Giovanni; Ghironi, Fabio; Sestieri, Giulia and Yamano, Norihiko. “Estimating Trade Elasticities: Demand Composition and the Trade Collapse of 2008-2009.” American Economic Journal: Macroeconomics, July 2013, 5(3), pp. 118-51. Cassey, Andrew J. “State Export Data: Origin of Movement vs. Origin of Production.” Journal of Economic and Social Measurement, 2009, 34(4), pp. 241-68. Cassey, Andrew J. “State Foreign Export Patterns.” Southern Economic Journal, October 2011, 78(2), pp. 308-29. Chor, Davin and Manova, Kalina. “Off the Cliff and Back? Credit Conditions and International Trade During the Global Financial Crisis.” Journal of International Economics, May 2012, 87(1), pp. 117-33. Contessi, Silvio and de Nicola, Francesca. “What Do We Know About the Relationship Between Access to Finance and International Trade?” Working Paper No. 2012-054B, Federal Reserve Bank of St. Louis, March 2013; http://research.stlouisfed.org/wp/2012/2012-054.pdf. Coughlin, Cletus C. “The Increasing Importance of Proximity for Exports from U.S. States.” Federal Reserve Bank of St. Louis Review, November/December 2004, 86(6), pp. 1-18; http://research.stlouisfed.org/publications/review/04/11/Coughlin.pdf. Coughlin, Cletus C. and Novy, Dennis. “Is the International Border Effect Larger than the Domestic Effect? Evidence from U.S. Trade.” CESifo Economic Studies, June 2013, 59(2), pp. 249-76; doi:10.1093/cesifo/ifs002. Coughlin, Cletus C. and Wall, Howard J. “NAFTA and the Changing Pattern of State Exports.” Papers in Regional Science, October 2003, 82(4), pp. 427-50. Crowley, Meredith A. and Luo, Xi. “Understanding the Great Trade Collapse of 2008-09 and the Subsequent Trade Recovery.” Federal Reserve Bank of Chicago Economic Perspectives, Second Quarter 2011, 35(2), pp. 44-70; http://www.chicagofed.org/digital_assets/publications/economic_perspectives/2011/2qtr2011_part1_crowley_luo.pdf. Eaton, Jonathan; Kortum, Samuel and Kramarz, Francis. “Dissecting Trade: Firms, Industries, and Export Destinations.” American Economic Review, May 2004, 94(2), pp. 150-54. Eaton, Jonathan; Kortum, Samuel; Neiman, Brent and Romalis, John. “Trade and the Global Recession.” NBER Working Paper No. 16666, National Bureau of Economic Research, January 2011; http://www.nber.org/papers/w16666.pdf?new_window=1. Fontagné, Lionel and Gaulier, Guillaume. “French Exporters and the Global Crisis,” in Richard Baldwin, ed., The Great Trade Collapse: Causes, Consequences and Prospects. Chap. 16. VoxEU.Org eBooks, 2009; http://www.voxeu.org/article/french-exporters-and-global-crisis. Lawless, Martina. “Deconstructing Gravity: Trade Costs and Extensive and Intensive Margins.” Canadian Journal of Economics, November 2010, 43(4), pp. 1149-72. Levchenko, Andrei A.; Lewis, Logan T. and Tesar, Linda L. “The Collapse of International Trade during the 2008-2009 Crisis: In Search of the Smoking Gun.” IMF Economic Review, 2010, 58(2), pp. 214-53. Melitz, Marc J. “The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity.” Econometrica, November 2003, 71(6), pp. 1695-725. Schott, Peter K. “US Trade Margins during the 2008 Crisis,” in Richard Baldwin, ed., The Great Trade Collapse: Causes, Consequences and Prospects. Chap. 15. VoxEU.Org eBooks, 2009; http://www.voxeu.org/article/us-trade-margins-during-2008-crisis. Smith, Pamela J. “Are Weak Patent Rights a Barrier to U.S. Exports?” Journal of International Economics, June 1999, 48(1), pp. 151-77. Wakasugi, Ryuhei. “Why Was Japan’s Trade Hit So Much Harder?” in Richard Baldwin, ed., The Great Trade Collapse: Causes, Consequences and Prospects. Chap. 23. VoxEU.Org eBooks, 2009; http://www.voxeu.org/article/why-was-japan-s-trade-hit-so-much-harder. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 33 34 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW A Guide to Tracking the U.S. Economy Kevin L. Kliesen Analyzing and forecasting the performance and direction of a large, complex economy like that of the United States is a difficult task. The process involves parsing a great deal of data, understanding key economic relationships, and assessing which events or factors might cause monetary or fiscal policymakers to change policy. One purpose of this article is to reinforce several key principles that are useful for tracking the U.S. economy’s performance in real time. Two principles stand out: First, the economy is regularly hit by unexpected economic disturbances (shocks) that policymakers and forecasting models cannot predict. Second, most key data used to measure the economy and track its performance are often revised—and by substantial amounts. (JEL E32, E66) Federal Reserve Bank of St. Louis Review, First Quarter 2014, 96(1), pp. 35-54. nalyzing and forecasting the performance of the U.S. and global economies is a daunting challenge, even for trained, professional economists. This means the challenge facing the nonpractitioner is probably much more difficult. For example, suppose a furniture retailer would like to know the direction of interest rates and the unemployment rate over the next year or two. The direction of interest rates is important because sales of durable goods such as furniture tend to be interest rate sensitive. Likewise, if an increasing percentage of the labor force becomes unemployed, then sales will tend to suffer. But other economic variables are also important. Household wealth, home sales, and consumer sentiment are often used by forecasters and some monetary policymakers to help predict the future path of consumer spending on durable goods. If the retailer guesses wrong and orders too much or too little furniture from the factory, this may lead to either too much or too little inventory on hand. If too much furniture is ordered, the retailer’s costs of carrying the extra inventory would increase, whereas if too little is ordered, the retailer’s sales might suffer. In both instances, the retailer’s profits would probably be reduced relative to what was expected. In short, a furniture retailer has a powerful incentive to form some assessment of the economy’s future performance. A Kevin L. Kliesen is a research officer and economist at the Federal Reserve Bank of St. Louis. Douglas C. Smith and Lowell R. Ricketts provided research assistance. © 2014, The Federal Reserve Bank of St. Louis. The views expressed in this article are those of the author(s) and do not necessarily reflect the views of the Federal Reserve System, the Board of Governors, or the regional Federal Reserve Banks. Articles may be reprinted, reproduced, published, distributed, displayed, and transmitted in their entirety if copyright notice, author name(s), and full citation are included. Abstracts, synopses, and other derivative works may be made only with prior written permission of the Federal Reserve Bank of St. Louis. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 35 Kliesen This article is not a how-to exercise in building economic models. Rather, it is intended to assist the noneconomist (nonpractitioner) who wants to analyze and interpret patterns of economic activity at the macro level. A nonpractitioner can be a businessperson, an investor, or any individual interested in monitoring the U.S. economy and/or developing an expectation of its short-term direction. A key conclusion is that the economy’s performance can change rapidly. Accordingly, the nonpractitioner seeking some clues about the short-term direction of the economy is advised to monitor a handful of key data and then balance this information against freely available consensus forecasts of the economy over the next six months or so. Over time, consensus forecasts, which are simple averages of a group of professional forecasters, tend to be more accurate than any individual forecast. A BASIC MODEL OF ECONOMIC FLUCTUATIONS A practicing forecaster usually needs a model of how the macroeconomy works. For professional forecasters, the “model” is usually a sophisticated system of equations designed to explain key aspects of the economy—such as growth of real gross domestic product (GDP), inflation, interest rates, stock prices, and the unemployment rate. Nonpractitioners—those not actively managing a large econometric forecasting model—tend to be at a distinct disadvantage in this domain. To compensate, the nonpractitioner who needs to make some judgment about the future direction of the economy should adopt a less formal economic model. Such a model would convey a broad notion of how the economy evolves over the business cycle. One simplistic model the nonpractitioner can use to organize his or her thoughts would be the following: U.S. economic activity—or real GDP—revolves around a trend that grows roughly at a rate determined by the sum of labor productivity growth and population growth.1 This trend is sometimes called the growth rate of the economy’s potential output. Deviations around this trend—termed “economic fluctuations”—occur because of unexpected disturbances (termed “shocks”), new technologies, and the ever-evolving preferences of consumers, firms, and government policymakers to save, spend, and regulate. With this simplistic model, the nonpractitioner can make reasonably accurate assessments about the likely direction of the economy over the next several months or quarters. For example, if auto and home sales are strengthening, the unemployment rate is falling, job gains are picking up, and stock prices are rising, then these factors are usually reliable signals that the economy is on an upswing. The nonpractitioner should thus exploit the fact that many key variables move together, which is known as comovement (see the boxed insert on the next page). Comovement is important because the economy’s natural state is one of positive growth— where this growth is dependent on the economy’s fundamentals. At any point in time, then, the economy will be growing above or below this trend rate of growth, which will then affect important variables such as, inflation, interest rates, and the unemployment rate.2 A model of inflation also differentiates between short- and long-run movements. Inflation can vary over shorter periods of time with changes in energy prices or labor costs. However, over longer periods (several years), actions taken by monetary policymakers will have a significant influence on the economy’s inflation rate. Importantly, this transmission stems from 36 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Kliesen Exploit the Principle of Comovement Private sector professional forecasters and forecasters employed by central banks typically use sophisticated econometric models based on either economic theory or statistical relationships designed to exploit comovements among key variables. The table below depicts one way to measure comovement: by noting correlations among key economic variables. The economic variables listed in the table are real GDP, the unemployment rate, equity prices, the conventional mortgage rate, new orders for manufactured goods, single-family home sales, and consumer price index (CPI) inflation. Comovement: Cross-Correlation of Four-Quarter Changes or Growth Rates Indicator Real GDP Unemployment rate Equity prices Conventional mortgage rate New manufacturing orders Single-family home sales CPI inflation Unemployment Real GDP rate 1.00 –0.78 0.68 0.46 0.65 0.47 0.34 1.00 –0.54 –0.33 –0.73 –0.39 –0.50 Equity prices 1.00 0.18 0.62 0.35 0.16 New Single-family Conventional mortgage manufacturing home rate orders sales CPI inflation 1.00 0.40 –0.23 0.31 1.00 0.12 0.58 1.00 –0.03 1.00 NOTE: Correlations are based on the sample period 1992:Q3–2013:Q1. SOURCE: Author’s calculations. The correlations are based on changes over four quarters; the sample period is 1992 to the present.1 As an example, the growth of real GDP is highly correlated with the unemployment rate (–0.78), equity prices (0.68), and new manufacturing orders (0.65). These correlations suggest, therefore, that faster real GDP growth tends to be associated with falling unemployment rates, higher stock prices, and faster growth of factory orders. Economists then exploit this comovement over time, as well as economic theory, when building forecasting models or thinking about how the economy evolves in response to changes in some key data. Economists have also exploited comovement to construct economic indexes designed to measure economic activity using several separate economic series. There are several such indexes designed to mirror broader trends in the economy. Among the more notable are those published by the Federal Reserve Bank of Chicago (Chicago Fed National Activity Index) and the Federal Reserve Bank of Philadelphia (Aruoba-Diebold-Scotti Business Conditions Index).2 1 The data are four-quarter growth rates (percent changes), except for the unemployment rate and the conventional mortgage; these are simple changes over four quarters. For example, the unemployment rate in 2013:Q1 was 7.7 percent and in 2012:Q1 was 8.3 percent. Thus, the fourquarter change was 0.6 percentage points. 2 For the Chicago Fed National Activity Index, see http://www.chicagofed.org/webpages/research/data/cfnai/current_data.cfm. For the Aruoba- Diebold-Scotti Business Conditions Index, see http://www.phil.frb.org/research-and-data/real-time-center/business-conditions-index/. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 37 Kliesen the effects of these policy actions on the market’s expectations about future inflation. William Poole (2005, pp. 303-04), former president of the Federal Reserve Bank of St. Louis, provided a nice summary of how this might work in practice: My sense of what I do, which I think is not dissimilar to what most FOMC [Federal Open Market Committee] members do, is attempt to intuit future inflation pressures from current observed pressures as they show up in both price changes and resource pressures, or real gaps, in individual markets. The approach is not totally without theory; for example, wage changes are evaluated in light of expected productivity trends. I attempt to sort out temporary from more lasting wage and price changes and attempt informally to construct an appropriately weighted average of disparate experience in various sectors. I look closely at data on inflation expectations, but treat such data carefully because longer-run expectations are really a vote of confidence on the Fed and not an independent reading on inflation. I am extremely uncomfortable with this approach and believe that it is an invitation to future mistakes. I don’t know what better to do. The nonpractitioner faces another key disadvantage relative to professional forecasters or economic policymakers: resource constraints. Thus, returning to our earlier example, small firms tend to be at a disadvantage compared with large firms in trying to analyze the direction of the economy. Large firms have the resources to hire economists to sift through the data and construct their own sophisticated forecasting models, or they can benefit from professional forecasting services on a contract basis. To help offset this disadvantage, a small business owner will probably adopt some form of naive forecasting (“what happened last year will happen again this year”) by reading economic and financial market commentaries from trade associations or perusing “reputable” economic blogs. Some may also use common rules of thumb purported to gauge the strength and direction of the economy, such as the direction of the stock market. The challenge of economic forecasting extends beyond the technical expertise required to make accurate forecasts. Other factors contributing to this difficult task include the sheer volume of data, persistent data revisions, and correct interpretation of data that may send conflicting signals. Other complications are the responses of monetary and fiscal policymakers and foreign economic developments. Before expounding on how a nonpractitioner might try to overcome these challenges, the next section offers a brief discussion of the events leading up to 2008. The so-called Financial Panic of 2008 and the Great Recession offer several examples of the difficulties both nonpractitioners and professional forecasters face as they attempt to learn about the direction of the economy and forecast its short-term future path. THE PERILS OF FORECASTING: A LOOK BACK AT 2008 The late economist John Kenneth Galbraith reportedly once remarked that there are two types of forecasters: those who don’t know and those who don’t know they don’t know. Galbraith’s aphorism reveals an underappreciated aspect of forecasting: It is inherently difficult. Thus, it was not surprising that the onset of the recent recession was not foreseen by the majority of the professional forecasting community. According to the Business Cycle Dating 38 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Kliesen Figure 1 A Timeline of Blue Chip Forecasts for Real GDP Growth in 2008 Percent (Annual Rate) 4.0 2008 (Real GDP) December 2007 September 2008 3.0 2.0 1.0 –0.2 Initial Actual 0 –1.0 –2.0 –3.3 Current Estimate –3.0 –4.0 Jan. 2007 Jul. 2007 Jan. 2008 Jul. 2008 Jan. 2009 Jul. 2009 SOURCE: Blue Chip Economic Indicators, various issues. Committee of the National Bureau of Economic Research (NBER), the U.S. economic expansion that began in November 2001 ended sometime in December 2007.3 However, by the end of 2007, very few professional forecasters were predicting a recession in 2008. In fact, in the December 2007 Blue Chip Economic Indicators, the consensus of the Blue Chip forecasters was that real GDP would increase by 2.2 percent in 2008. The average of the 10 most pessimistic forecasters was 1.6 percent, while the average of the 10 most optimistic forecasters was 2.7 percent.4 The NBER Business Cycle Dating Committee, like many nonpractitioners, tends to look at real GDP as a key indicator (among other indicators) of the economy’s performance. For example, increases (decreases) in expenditures for real final goods and services—such as automobiles, refrigerators, or physician services—are regularly followed by increases (decreases) in employment and a lower (higher) unemployment rate. As Figure 1 shows, throughout most of 2007 the Blue Chip Consensus (BCC) of professional forecasters was that real GDP would increase by about 3 percent in 2008. This figure plots a timeline of BCC forecasts for real GDP growth in 2008. The first forecast was published in January 2007. Beginning in September 2007, though, forecasters began to steadily lower their projections for real GDP growth in 2008. In particular, as discussed below, the forecasts for real GDP growth for 2008 turned sharply lower after the widespread financial turmoil in September 2008. By the end of November 2008, when the NBER announced that the recession began sometime in December 2007, the BCC forecast for real GDP growth in 2008 had dipped slightly below zero. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 39 Kliesen Figure 2 A Timeline of Blue Chip Forecasts for CPI Inflation in 2008 Percent (Annual Rate) 5.0 2008 (CPI Inflation) December 2007 September 2008 4.5 4.0 3.5 3.0 2.5 2.0 1.6 Actual 1.5 1.0 0.5 0 Jan. 2007 Jul. 2007 Jan. 2008 Jul. 2008 Jan. 2009 Jul. 2009 SOURCE: Blue Chip Economic Indicators, various issues. The direction of inflation is another key indicator of economic performance. First, longterm interest rates such as mortgage rates and corporate bond yields have an inflation premium.5 Accordingly, if inflation or the perceived risk of higher inflation in the future increases, then interest rates also usually rise. A higher inflation rate may also spur the Fed to raise its short-term interest rate target, which could also cause long-term rates to rise.6 The direction of inflation was markedly different over a good portion of this period. As Figure 2 shows, from January 2007 until March 2008, the BCC forecast was that the CPI would increase by a bit less than 2.5 percent in 2008. The relative stability of inflation expectations was somewhat surprising given the behavior of oil prices and actual inflation over this period. From January 2007 to March 2008, crude oil prices rose from about $55 per barrel to about $106 per barrel. Over the same period, the year-to-year percent change in the CPI rose from 2.1 percent to 4 percent. As oil prices and actual inflation continued to rise over the first half of 2008, forecasters began to dramatically raise their forecasts for inflation in 2008—from about 2.75 percent in April to 4.5 percent in September.7 Interestingly, though, forecasts for CPI inflation in 2009 (not shown) rose only slightly, which suggests that most forecasters tended to believe that the upsurge in inflation in 2008 would be temporary. This forecast proved to be accurate. (See the boxed insert on p. 46.) A key takeaway message from Figures 1 and 2 is that significant, unexpected economic shocks can have important effects on the expectations of forecasters—and thus investors and economic policymakers. The remainder of the article discusses a methodology the nonprac40 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Kliesen Table 1 Free Economic Forecasts Name Source Frequency Survey of Professional Forecasters Federal Reserve Bank of Philadelphia Quarterly http://www.phil.frb.org/research-and-data/ real-time-center/survey-of-professionalforecasters/ FOMC Summary of Economic Projections Federal Open Market Committee Quarterly http://www.federalreserve.gov/monetarypolicy/ fomccalendars.htm IMF World Economic Outlook Reports International Monetary Fund NABE Outlook (partial) National Association for Business Economics Quarterly http://nabe.com/NABE_Outlook_Summary Budget and Economic Outlook Congressional Budget Office Annually http://www.cbo.gov/publication/43907 Economic Outlook OECD Semiannually Semiannually URL http://www.imf.org/external/ns/cs.aspx?id=29 http://www.oecd.org/eco/economicoutlook.htm NOTE: IMF, International Monetary Fund; NABE, National Association for Business Economics; OECD, Organisation for Economic Co-operation and Development. titioner can use to help analyze the current and short-term performance of the U.S. economy. This approach relies on publicly available data and macroeconomic forecasts. In this framework, “reading the tea leaves” requires an assessment of the following economic conditions: • the economy’s momentum (slowing or accelerating); • the headwinds or tailwinds affecting this momentum and how long they are expected to last; and • the risks to the outlook—that is, what could produce growth that is either faster or slower than expected for economic activity and prices. Endnotes are used for those seeking references or a more in-depth discussion about analyzing general business conditions and the macroeconomy. Although the U.S. economy is obviously affected by events in other countries, the discussion focuses primarily on U.S. data flows and the decisions adopted by U.S. economic policymakers and their potential economic consequences. KEY PRINCIPLES FOR TRACKING THE ECONOMY Principle #1: Use Freely Available Forecasts The nonpractitioner should adhere to a set of key economic principles. One such principle is comparative advantage. That is, the nonpractitioner should use forecasts developed by professional economists with advanced training and experience in modeling and forecasting. Another principle pertains to the law of demand, which relates the price of a good to its quantity demanded: Free is usually better. Fortunately, reputable forecasts are freely available Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 41 Kliesen Figure 3 Forecasts of Four Key Economic Variables What Are Forecasters Predicting for Real GDP Growth? What Are Forecasters Predicting for Inflation? Percent Percent 5 5 Actual 4 Actual Forecast Forecast 4 3 3 2 2 1 1 0 –1 2011:Q4 2012:Q3 2013:Q2 2014:Q1 2014:Q4 0 2011:Q4 2012:Q3 2013:Q2 2014:Q1 2014:Q4 What Are Forecasters Predicting for the Unemployment Rate? What Are Forecasters Predicting for the 10-Year Treasury Yield? Percent Percent 10 5 Actual 9 Actual Forecast Forecast 4 8 3 7 2 6 5 2011:Q4 2012:Q3 2013:Q2 2014:Q1 2014:Q4 1 2011:Q4 2012:Q3 2013:Q2 2014:Q1 2014:Q4 SOURCE: Survey of Professional Forecasters, November 2013. to the public (Table 1). The law of large numbers is also a related principle: An average, or consensus, of many forecasts is usually better than a single forecast by any one forecaster. Figure 3 shows forecasts of four key economic variables: real GDP growth, inflation, the unemployment rate, and the 10-year Treasury yield. The forecasts are based on a survey of professional forecasters and published four times per year by the Federal Reserve Bank of Philadelphia in its Survey of Professional Forecasters (SPF). In the November 2013 SPF, the consensus of professional forecasters was that the economy would continue to improve. This was evident by a modest acceleration in real GDP growth, a modest reduction in the unemployment rate, and a modest upswing in long-term interest rates. Forecasters also expected inflation to remain relatively low and stable. 42 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Kliesen Principle #2: Use Forecast Revisions to Gauge Changes in Economic Expectations Consensus forecasts are valuable because they provide a “best guess” approach to the economic outlook. This approach differs sharply from relying on a single forecaster whose model or theoretical biases may not be readily known. However, a drawback to consensus forecasts—and, for all practical purposes, all forecasts—is that forecast horizons less than a year or two (four to eight quarters) ahead can change dramatically because of unexpected economic events. Still, the nonpractitioner can use this knowledge to help assess whether the economy is experiencing faster or slower momentum. Just as a car speeds up or slows down, the economy goes through periods when growth of real GDP, inflation, or employment is faster or slower than expected. Before discussing how the nonpractitioner can assess changes in the economy’s momentum, it is crucial to acknowledge some key facts about the U.S. economy. First, the economy’s normal state of affairs is one of positive growth in real GDP and in prices (inflation). According to the NBER Business Cycle Dating Committee, from January 1948 to December 2013, the U.S. economy has spent 670 of 792 months (or 85 percent of the time) in expansion. Second, the growth rate of key indicators, such as employment, retail sales, real GDP, and inflation, can vary tremendously from month to month, quarter to quarter, or year to year. Third, actions by the Federal Open Market Committee (FOMC) can influence the economy in important respects, but generally not immediately. Fourth, unexpected disturbances regularly occur that cause forecasts to go awry. Figures 1 and 2 show how changes in forecasters’ expectations are reflected in the economy’s momentum. If the economy is exhibiting stable momentum, this generally suggests that the incoming data flows are in line with expectations. In this case, forecasts for real GDP growth (and other key indicators) will remain relatively unchanged, as they were over the first part of 2008. However, faster momentum suggests the incoming data are exceeding expectations (in a good way), and this will be translated into upward revisions in forecasts for real GDP growth. The opposite holds for slower momentum. An example of the latter situation is the downgrading of forecasts for real GDP growth that began in 2008 (see Figure 1). In terms of inflation momentum, Figure 2 shows forecasters continually raised their estimates for CPI inflation for 2008 over the first eight months of the year in response to rising energy prices. Changes in momentum, as reflected in data flows, are important because they help forecasters identify possible shocks to the economy, which can be either positive or negative. These changes thus feed back into revised forecasts. One drawback to this approach is that freely available forecasts tend to be published at a quarterly or annual frequency (see Table 1). However, identifying momentum changes from the monthly forecasts used in Figures 1 and 2 requires a paid subscription to the Blue Chip Economic Indicators. As an aside, many professional forecasters tend to update their modelbased forecasts on a daily or weekly basis using the latest available data. But a lot can happen in three months, so nonpractitioners who use quarterly forecasts need to augment this framework with something else to identify momentum shifts. One relatively easy method is to systematically track key economic data flows to infer future forecast revisions. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 43 Kliesen Principle #3: Follow the Data to Help Identify Momentum Swings Momentum swings during the course of the business cycle can be measured by tracking the evolution of the data relative to the expectations of forecasters and/or financial market participants. These data flows or other economic news may alter the expectations of investors, professional forecasters, and policymakers regarding the strength or weakness of the economy. A recent example of this principle was cited by former Federal Reserve Chairman Ben Bernanke (2013, p. 4) in his press conference following the June 19, 2013, FOMC meeting: Although the Committee left the pace of purchases unchanged at today’s meeting, it has stated that it may vary the pace of purchases as economic conditions evolve. Any such change would reflect the incoming data and their implications for the outlook. In the context of measuring economic momentum, if key data repeatedly surprise on the upside (downside), then this is a signal that forecasters have been underestimating (overestimating) the strength of the economy. To successfully use this framework, the nonpractitioner must first decide which economic data to focus on.8 This step is crucial for two reasons. First, some data are more important than others. And second, some data directly influence forecasts for real GDP and inflation, but most do not. In this section, the discussion focuses on key nonfinancial variables.9 The importance of financial market conditions is discussed later. Table 2 provides a list of key data that the noneconomist should monitor on a regular basis.10 In particular, key series released early in the monthly data cycle include • the manufacturing and nonmanufacturing purchasing managers indexes (PMIs), which provide a broad-based overview of economic activity; • the nonfarm payroll employment and unemployment rate series published by the Bureau of Labor Statistics in “The Employment Situation”; and • reports on manufacturing activity (durable goods orders and industrial production), consumer spending (retail sales and auto sales), and housing activity (housing starts and new and existing home sales). A weekly series—initial claims for unemployment insurance benefits—is also included. Initial claims is an important indicator because (i) it is released each week and (ii) the series tends to have some predictive power for the number of individuals moving into and out of jobs. For example, Kliesen, McCracken, and Zheng (2011) show that job growth tends to weaken or strengthen when the number of initial claims rises above or falls below 400,000.11 Table 2 also includes a market-based forecast for each of the indicators released on a recurring basis. For each series, economists and market analysts are surveyed and asked to provide their forecast, or best guess estimate, for the key economic data to be released that week. These market-based expectations for key upcoming data releases are found on many freely available economic calendars.12 Table 2 shows how a practitioner can use these expectations to construct a systematic, simple approach to gauge potential changes in economic momentum in real time based on data surprises. This method is depicted in the last three columns of the table. First, for each 44 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Kliesen Table 2 A Time Horizon of Key Data Flows (April 2013–May 2013) Date Indicator Market Period expectations 3/31/2013 4/1/2013 4/1/2013 4/2/2013 4/2/2013 4/3/2013 4/4/2013 4/5/2013 4/5/2013 4/5/2013 4/5/2013 4/9/2013 4/10/2013 4/11/2013 4/12/2013 4/12/2013 4/12/2013 4/12/2013 4/12/2013 4/16/2013 4/16/2013 4/16/2013 4/16/2013 4/16/2013 4/16/2013 4/18/2013 4/22/2013 4/23/2013 4/24/2013 4/24/2013 4/26/2013 4/29/2013 4/29/2013 4/29/2013 4/30/2013 5/1/2013 5/2/2013 5/2/2013 5/2/2013 5/7/2013 ISM Manufacturing PMI Construction spending Factory orders Total vehicle sales ISM Non-Manufacturing PMI Initial claims Total nonfarm payrolls Private payrolls Unemployment rate International trade balance Wholesale inventories Federal budget balance Import prices PPI Core PPI Retail sales Retail sales excluding autos Business inventories Housing starts Building permits CPI Core CPI Industrial production CU rate Index of Leading Economic Indicators Existing home sales (total) New home sales Durable goods Durable goods excluding transportation Real GDP Personal income PCE (expenditures) Core PCE (prices) Employment Cost Index Construction spending International trade Productivity Unit labor costs Consumer credit ($) March February February March March March 29 March March March February February March March March March March March February March March March March March March March March March March March Q1 Advance March March March Q1 Advance March March Q1 Advance Q1 Advance March 5/9/2013 Initial claims May 4 Actual 54.2 1.0 2.9 15.3 55.8 347 200 209 7.7 –44.6 0.5 –156.0 –0.5 –0.1 0.2 0.0 0.1 0.4 0.930 0.940 0.0 0.2 0.2 78.4 0.1 5.010 0.420 –2.8 0.5 3.0 0.4 0.0 0.1 0.5 0.7 –42.2 1.5 0.6 15.0 51.3 1.2 3.0 15.2 54.4 385 88 95 7.6 –43.0 –0.3 –106.5 –0.5 –0.6 0.2 –0.4 -0.4 0.1 1.036 0.902 –0.2 0.1 0.4 78.5 –0.1 4.920 0.417 –5.7 –1.4 2.5 0.2 0.2 0.0 0.3 –1.7 –38.8 0.7 0.5 8.0 335 323 Better than expected? Sign NTI No Yes Yes No No No No No Yes Yes No No Same Yes Same No No No Yes No Yes Yes Yes Yes No No No No No No No Yes Yes Yes No Yes No Yes No 0 –1 1 1 –1 –1 –1 –1 –1 1 1 –1 –1 0 1 0 –1 –1 –1 1 –1 1 1 1 1 –1 –1 –1 –1 –1 –1 –1 1 1 1 –1 1 –1 1 –1 0 –1 0 1 0 –1 –2 –2 –3 –2 –1 –2 –3 –3 –2 –2 –3 –4 –5 –4 –5 –4 –3 –2 –1 –2 –3 –4 –5 –6 –7 –8 –7 –6 –5 –6 –5 –6 –5 –6 Yes 1 –5 NOTE: CU, capacity utilization; ISM, Institute for Supply Management; PCE, personal consumption expenditures; PPI, producer price index. SOURCE: Thomson Reuters and author’s calculations. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 45 Kliesen Assessing Risks to the Economic Outlook When economic conditions are perceived as “normal,” the risk to the outlook is viewed as balanced. This means that forecasters generally see little reason to alter their assessment of the short-term outlook. However, if there are developments in the domestic or global economy that are judged as “abnormal,” then the risks are considered skewed to the upside or downside. The 2007-09 recession came on the heels of two shocks that have historically proved damaging to the economy’s short-term performance: an unexpectedly large increase in crude oil prices and an epic decline in house prices and housing activity.1 Oil prices are important because, historically, unexpected increases in oil prices have preceded nearly every post-World War II U.S. recession. Increases in oil prices tend to (i) raise headline inflation rates and (ii) slow the growth of economic activity. Thus, the analyst or investor who concludes that rising oil prices increase the probability of weaker growth and higher inflation is usually on safe terrain.2 In a similar vein, housing is important because construction ripples through the economy—for example, affecting sales of consumer durables such as furniture and appliances—and house prices can change household wealth and thus perhaps consumer expenditures (the largest component of GDP).3 As this episode demonstrates, nonpractitioners would be wise to pay attention to developments in energy and housing markets to gauge unfolding risks to the outlook.4 1 Stock and Watson (2012) argue that the sharp rise in oil prices and the financial turmoil were key factors that caused the 2007-09 recession. 2 Hamilton (2008) has documented that 9 of the 10 recessions between 1948 and 2001 were preceded by a rise in oil prices. Rising oil prices also appear to be an important factor explaining business cycles in other advanced economies. See Engemann, Kliesen, and Owyang (2011). 3 Boldrin et al. (2013) discuss and estimate these housing spillover effects; also see Leamer (2008). Some economists believe that changes in equity prices also have a significant effect on household consumption. 4 At the same time, many forecasts, including those by the staff members who advise the FOMC, consider numerous alternative scenarios to the so-called baseline forecast. Some forecasts, for example, attach a probability to a recession scenario, but the probability is generally much smaller than the baseline “no recession” scenario. release, determine whether the data were better than expected. Second, if so, arbitrarily assign an indicator value of +1; if not, assign a –1 (worse than expected). If the data met expectations, then assign a value of 0. Third, sum the indicator values (+1, –1, and 0) to obtain a net tracking index (NTI). Using the first indicator in Table 2 (the Institute for Supply Management [ISM] Manufacturing PMI), the market’s expectation for March 2013 was 54.2 but the actual estimate was 51.3, which was worse than expected, so we assign a value of –1. By the end of the list on May 9, the cumulative series—which is the NTI—has a value of –5. A negative value thus indicates that, on net, the data have come in worse than expected and, by assumption, this implies some weaker economic momentum over this period of data flows. Figure 4 plots the NTI for data flows that measure economic activity in the fourth quarter of 2012 and the first quarter of 2013. Interpreting the NTI is relatively straightforward since it is conditional on one’s assumption about the direction of the expected change in economic activity. And since much of the data feed directly into estimates of real GDP or are indicators of economic activity more broadly, the NTI is thus one proxy for the expected change in real GDP in a given quarter— actual or forecasted. Though not shown here, the nonpractitioner could also construct an NTI for inflation pressures. It should be noted that the NTI date listed in Figures 4A and 4B is not the same as the period of economic measurement. For example, total nonfarm payrolls for March 2013 were released on April 5, 2013. Figure 4A shows that beginning in the second week of December 46 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Kliesen Figure 4 Net Economic Tracking Index A. 2012:Q4 B. 2013:Q1 +1 = Better Than Expected –1 = Worse Than Expected +1 = Better Than Expected –1 = Worse Than Expected 30 30 25 25 20 20 15 15 10 10 5 5 0 0 –5 10/8/12 11/7/12 12/7/12 1/6/13 2/5/13 3/7/13 –5 1/9/13 2/8/13 3/10/13 4/9/13 5/9/13 SOURCE: Author’s calculations. 2012, the data flows began to be better than expected, on net. All else equal, this was a signal that in the fourth quarter the economy was strengthening by more than expected. However, when the advance estimate was released at the end of January 2013, real GDP for the fourth quarter of 2012 instead was shown to have declined at a 0.1 percent annual rate. This was far below the consensus forecast. Consistent with the NTI, though, subsequent revisions by the BEA slightly raised the advance estimate for real GDP growth in the fourth quarter from –0.1 percent to 0.1 percent. By contrast, Figure 4B shows that the NTI performed modestly better in the first quarter of 2013. Beginning in late February and early March 2013, the data began to come in consistently better than expected. In response, forecasters began raising their first-quarter estimate for real GDP growth. When the advance estimate was released in late April 2013, the economy was shown to have grown at a 2.5 percent annual rate. Although the NTI is a very simple metric for measuring momentum, there are a few drawbacks to consider. First, the NTI does not discriminate between the value added of expenditures (e.g., retail sales), employment, survey-based measures, or prices. A second criticism is that the NTI assigns each series the same weight (equal importance). Thus, key data such as payroll employment or housing starts should probably be assigned larger weights than series such as wholesale inventories. One problem confronting all practitioners and nonpractitioners is the inevitability of data revisions. Although the NTI, by design, cannot account for subsequent revisions, it does help minimize this problem because it includes survey data and other types of data that are not revised (e.g., consumer confidence or weekly initial claims). But if the goal is to use the latest data to get a reading on real GDP growth, then revisions are an issue that the nonpractitioner will need to confront. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 47 Kliesen Most Key Data Are Based on Sample Estimates At any point in time, those interested in reading the economy’s tea leaves have a morass of data to follow. The layperson needs to recognize, however, the limitations of many of these data series. For example, most key government statistics reported by the Census Bureau, the Bureau of Economic Analysis (BEA), and other agencies are based on sample-based estimates rather than universe-based counts. For example, to measure the monthly number of jobs created in the United States each month, the Bureau of Labor Statistics (BLS) does not count every new job at every single business and government entity each month. Instead, the BLS surveys approximately 145,000 private nonagricultural businesses and government agencies each month.1 These businesses and agencies represent approximately 557,000 individual worksites. This sample-based estimate is then used to construct an estimate of total nonfarm payroll employment for the nation.2 Despite what many people may believe, a separate survey is used to estimate the nation’s unemployment rate. Each month, the Census Bureau surveys about 60,000 households about their labor force status. These data are then analyzed and published by the BLS in its monthly employment report. In addition to providing an estimate of the unemployment rate, the BLS publishes an estimate of the nation’s population, labor force, and the number of people employed, among other metrics. Many other key data series, such as factory orders, housing starts, retail sales, and industrial production, are also based on survey data collected by government and private sector entities. An additional consideration is that most of the key data produced by the government statistical agencies and private sector firms are seasonally adjusted. This is another potential source of measurement error. 1 These entities employ approximately 9 million nonfarm workers. Technically, the establishment survey counts the number of jobs rather than workers because some people have more than one job. 2 More detailed information on the methodology used by the BLS to construct labor market measures from the Current Employment Statistics or the Current Population Survey can be found in the BLS Handbook of Methods; see http://www.bls.gov/opub/hom/. Principle #4: Beware of Data Revisions As noted in the description of Principle 3, revisions to data compound the difficulty of correctly identifying shocks and their significance in real time. These revisions occur largely because much of the source data collected by the U.S. government statistical agencies are based on surveys of a sample of economic entities (firms, households, and government offices and agencies), rather than a survey of the universe of all economic entities. (This process is discussed in the boxed insert above.) As an example of this process, consider the quarterly estimate for the growth of real GDP, which is subject to numerous revisions. These revisions generally reflect updates in the underlying source data or new data based on more complete surveys or income tax records. Sometimes, revisions are made to prices or the underlying statistical methodology used by the government agencies to construct the estimate. To see how revisions can dramatically change the portrait of the economy’s performance, consider the estimate of real GDP growth for the fourth quarter of 2007. According to the NBER, this quarter was the peak of the 2001-07 business expansion. As shown in Figure 5, the BEA released nine estimates of the annual rate of change for real GDP growth in the fourth quarter of 2007. In the advance estimate released in late January 2008, the BEA reported that real GDP rose at a 0.6 percent annual rate. However, when the annual National Income and Product Accounts (NIPA) revision was released in late July 2008, the estimate for real GDP growth in the fourth quarter of 2007 was changed to –0.2 percent. This estimate was subsequently changed to 2.9 percent per year in the 2010 annual NIPA revision but was subsequently marked back down by nearly 1.5 percentage points with the release of the July 2013 NIPA revision. 48 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Kliesen Figure 5 Real-Time Estimates of Real GDP Growth During 2007:Q4 Percent Change (Annual Rate) 3.5 2.9 3.0 2.5 2.1 2.0 1.7 1.5 1.7 1.5 1.0 0.6 0.6 0.6 0.5 0 –0.2 –0.5 Jan. 2008 Feb. 2008 Mar. 2008 Jul. 2008 Jul. 2009 Jul. 2010 Jul. 2011 Jul. 2012 Jul. 2013 NOTE: Dates reflect the month in which the estimates were published. SOURCE: Bureau of Economic Analysis and Haver Analytics. What should the nonpractitioner take away from this discussion? First, the data may not correctly portray the economy’s momentum. This possibility suggests that the nonpractitioner should take the monthly data flows and the consensus forecasts with a grain of salt. But what is the alternative? After all, policymakers, FOMC members, businesspersons, and investors have little choice but to react to the incoming data flows.13 One way a nonpractitioner can minimize the potential havoc caused by data revisions is to avoid point estimates. Thus, instead of becoming enamored with a forecast for real GDP growth of 3.5 percent (the point estimate), an interest rate of 3 percent, or an unemployment rate of 6.5 percent, the nonpractitioner would attempt to assess whether the economic momentum revealed by forecast revisions and the NTI suggest something more likely on either side of the point estimate. Another method of minimizing the impact of data revisions is to track financial market conditions. Although there is the possibility of a chicken-versus-egg problem since financial markets also react to incoming data flows that are subsequently revised, some financial market series have long been recognized for their leading indicator properties. Principle #5: Track Financial Market Conditions14 Economic historians have long known that disturbances in the financial sector can have significant effects on the economy.15 Moreover, stabilizing the real economy through its interventions in financial markets is one of the key reasons central banks exist.16 The Financial Panic of 2008 provides another example of the financial sector’s far-reaching effects on the macroeconomy when asset prices and other key financial market indicators are changing significantly.17 Thus, for a more-complete portrait of the economy and potential changes in shortFederal Reserve Bank of St. Louis REVIEW First Quarter 2014 49 Kliesen Figure 6 The St. Louis Fed Financial Stress Index Weekly Data 6 Week of August 5, 2011: S&P Downgrade and FOMC 5 European Turmoil, May/June 2010 4 3 September/October 2008 2 1 0 –1 –2 12/31/93 12/31/97 12/31/01 12/31/05 12/31/09 12/31/13 NOTE: The last observation is for the week ending January 31, 2014. term economic momentum, it is important for the nonpractitioner to understand and track financial market conditions. One of the key lessons policymakers learned from the 2007-08 experience is an old one: It is extraordinarily difficult to predict financial crises with any degree of confidence. But it can also be difficult to monitor financial conditions because there are literally thousands of different types of financial indicators—ranging from stock price indexes, to interest rates on government and corporate debt, to foreign exchange rates, to more elaborate indicators such as credit default swaps and mortgage-backed securities.18 Fortunately, the nonpractitioner can overcome much of this difficulty by focusing on a handful of key indicators. Three come to mind. The first key financial indicator is a financial stress index (FSI). FSIs are designed to measure changes in financial conditions. For example, when financial market conditions are viewed as stable, then financial stresses tend to be relatively normal. In this situation, lenders are no more risk averse than normal and the volatility of asset prices, such as stock and bond prices, exhibits no unusual movements. By and large, financial market participants have a rather sanguine view of the economy. By contrast, if lenders are becoming more risk averse, asset prices are falling, and volatility is increasing, then financial stresses are on the rise. In this instance, uncertainty about the health of the economy is increasing. Rising levels of financial stress tend to weaken the real economy through a variety of transmission mechanisms. These include reduced wealth, a reduction in bank lending, and balance-sheet effects that reduce the value of a firm’s collateral. The key innovation of FSIs is that they combine different types of financial market indicators into one index—much as the CPI is one measure of the economy’s price level constructed from tens of thousands of different prices on goods and services. For example, economic research has convincingly shown there is significant information content, and thus predictive power, in the U.S. Treasury yield 50 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Kliesen curve, commonly calculated as the difference between yields on 10-year Treasury securities and 3-month Treasury bills. The yield curve tends to be upward sloping during times of positive growth in economic activity and tends to narrow as the pace of economic activity slows and— importantly—to invert before recessions.19 Another measure of financial market stress is the spread between 10-year Treasury securities and Baa-rated corporate bonds. This interest rate spread, called the credit risk spread, accounts for default risks in private credit markets. Thus, if interest rates are rising and the pace of real GDP growth is slowing, firms tend to experience slowing sales, which adversely affects their revenues and thus their financial condition. In response, the risk associated with lending to firms increases. One well-known financial stress index that accounts for the information content in these two interest rate spreads is the St. Louis Fed Financial Stress Index (STLFSI). Figure 6 plots the STLFSI and shows that financial market stresses rose sharply prior to several recent economic upheavals that were transmitted to financial markets. These included recent developments in Europe, the downgrade of U.S. sovereign debt by Standard and Poor’s, and the unexpected downward revision in U.S. real GDP growth in July 2011 discussed previously. Empirical evidence suggests that rising levels of financial stress are associated with weak or negative growth of industrial production and other measures of economic activity going forward.20 CONCLUSION Analyzing and forecasting the performance and direction of a large, complex economy like that of the United States is exceptionally difficult. The process involves parsing a great deal of data, understanding key economic relationships, and assessing which events or factors might cause monetary or fiscal policymakers to change policy. One purpose of this article is to reinforce several key principles that a nonpractitioner should use to analyze U.S. economic and financial market conditions. The nonpractitioner can do a reasonably good job of tracking changes in the economy’s momentum by taking advantage of freely available macroeconomic forecasts and tracking key data. ■ Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 51 Kliesen NOTES 1 Labor productivity is output per hour. Broadly speaking, output is the total value of real GDP. Real GDP measures the inflation-adjusted dollar value of goods and services produced by labor and property located in the United States; these are known as factors of production. A “real” series has been adjusted to remove the effects of changes in prices over time. This adjustment better accounts for increases in the production and consumption of goods and services (i.e., volumes) in response to underlying factors that most affect supply and demand—such as changes in population, productivity, and technological innovations. These real factors are the ones that lead to increases (or decreases) in living standards, what economists term “economic growth.” See Gutierrez et al. (2007) for a primer on GDP and its construction. 2 In periods of economic recession, the level of real GDP is declining from one quarter to the next (at a negative growth rate). 3 A rough rule of thumb is that the NBER assumes that the inflection point occurred at the middle of the month. 4 The Blue Chip Consensus is a survey of roughly 50 private sector forecasters. For example, each forecaster submits his or her forecast for real GDP growth and other key macroeconomic and financial variables for the current and upcoming year. The consensus is the simple average (mean) of these forecasts. These forecasts are published in the Blue Chip Economic Indicators on or about the 10th of each month. 5 Economists use the Fisher equation to help analyze changes in interest rates. According to this equation, the nominal interest rate on, for example, a 30-year U.S. Treasury bond is the sum of (i) the real rate of interest earned over this period and (ii) the average inflation rate expected over this period (the inflation premium, a premium demanded by lenders to compensate them for the expected inflation rate over the maturity of the bond). This simple formulation ignores other risk premiums embedded within these nominal interest rates and the complication of selling a security before it matures, which can significantly affect the holding period rate of return on the bond. 6 Many sophisticated forecasting models, including those used by the staff economists at the Board of Governors of the Federal Reserve, assume that the yield on, say, the 10-year Treasury security, is simply an average of a series of 1-year expected future interest rates. Thus, by changing the level of the overnight federal funds rate, the model assumes that the Fed can affect the long-term interest rate. 7 Crude oil prices (measured by West Texas Intermediate) rose from a little more than $105 per barrel in March 2008 to about $134 per barrel in July 2008 (monthly averages). Over the same period, the year-to-year percent change in the CPI increased from 4 percent to 5.6 percent. 8 The St. Louis Fed’s FRED (Federal Reserve Economic Data) database contains one of the world’s largest collections of freely available economic and financial data. FRED can be accessed at http://research.stlouisfed.org/fred2/. 9 From a GDP-accounting perspective, data that flow directly into real GDP—such as housing starts, retail sales, and factory shipments—help economists estimate whether the growth of real GDP is likely to change from its previous-quarter estimate. However, other data flows such as employment, initial unemployment claims, inflation, and consumer confidence might also be signals of changes in aggregate demand or supply and thus elicit reactions from policymakers and financial market participants. 10 All these data, and more, are available on FRED, which currently contains more than 60,000 economic series. 11 Gavin and Kliesen (2002) provide a description and overview of the initial claims data. They also show that the ini- tial claims indicator has some statistically significant ability to predict monthly changes in payroll employment. 12 Many of these forecasts can be found on the “Calendar of Releases” (http://research.stlouisfed.org/publications/ usfd/cover.pdf ) published each week in the St. Louis Fed’s U.S. Financial Data publication (http://research.stlouisfed.org/publications/usfd/). Yahoo! provides an economic calendar with market forecasts of these and other variables (http://biz.yahoo.com/c/ec/201315.html). 13 Orphanides and Van Nordren (2002) have shown that the timeliness of the data and subsequent data revisions make it extremely difficult for monetary policymakers to identify the strength of real GDP relative to potential real GDP (the output gap) in real time. 14 This section draws from Kliesen, Owyang, and Vermann (2012). 15 See Kindleberger and Aliber (2005) or Reinhart and Rogoff (2009). 52 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Kliesen 16 This principle is known as the lender of last resort. Walter Bagehot’s Lombard Street (1873) is a classic text on the role of central banks during financial crises. 17 Several other events around this period contributed to the rise of financial market instability. For example, see “The Financial Crisis: A Timeline of Events and Policy Actions” on the St. Louis Fed’s website (http://timeline.stlouisfed.org/). 18 Zheng (2012) provides a good summary of key financial indicators (such as the yield curve) for the noneconomist. 19 See Wheelock and Wohar (2009) for a summary of this research. 20 See Kliesen, Owyang, and Vermann (2012). REFERENCES Bagehot, Walter. Lombard Street: A Description of the Money Market. London: H.S. King, 1873. Bernanke, Ben. “Transcript of Chairman Bernanke’s Press Conference.” June 19, 2013; http://www.federalreserve.gov/mediacenter/files/FOMCpresconf20130619.pdf. Boldrin, Michele; Garriga, Carlos; Peralta-Alva, Adrian and Sánchez, Juan M. “Reconstructing the Great Recession.” Federal Reserve Bank of St. Louis Working Paper 2013-006B, February 2013, revised June 2013; http://research.stlouisfed.org/wp/2013/2013-006.pdf. Engemann, Kristie M.; Kliesen, Kevin L. and Owyang, Michael T. “Do Oil Shocks Drive Business Cycles? Some U.S. and International Evidence.” Macroeconomic Dynamics, November 2011 (Suppl. S3), 15(3), pp. 498-517. Gavin, William T. and Kliesen, Kevin L. “Unemployment Insurance Claims and Economic Activity.” Federal Reserve Bank of St. Louis Review, May/June 2002, 84(3), pp. 15-28; http://research.stlouisfed.org/publications/review/02/05/15-28GavinKliesen.pdf. Gutierrez, Carlos M.; Glassman, Cynthia A.; Landefeld, J. Steven and Marcuss, Rosemary D. “Measuring the Economy: A Primer on GDP and the National Income and Product Accounts.” Bureau of Economic Analysis, September 2007. Hamilton, James D. “Oil and the Macroeconomy,” in Steven N. Durlauf and Lawrence E. Blume (eds.), The New Palgrave Dictionary of Economics. Second Edition. New York: Palgrave Macmillan, 2008. Kindleberger, Charles P. and Aliber, Robert. Manias, Panics, and Crashes: A History of Financial Crises. Fifth Edition. Hoboken, NJ: John Wiley & Sons, 2005. Kliesen, Kevin L.; McCracken, Michael W. and Zheng, Linpeng. “Initial Claims and Employment Growth: Are We at the Threshold?” Federal Reserve Bank of St. Louis Economic Synopses, 2011, No. 41, December 14, 2011; http://research.stlouisfed.org/publications/es/11/ES1141.pdf. Kliesen, Kevin L.; Owyang, Michael T. and Vermann, E. Katarina. “Disentangling Diverse Measures: A Survey of Financial Stress Indexes.” Federal Reserve Bank of St. Louis Review, September/October 2012, 94(5), pp. 369-97; http://research.stlouisfed.org/publications/review/12/09/369-398Kliesen.pdf. Leamer, Edward E. “Housing Is the Business Cycle,” in Housing, Housing Finance, and Monetary Policy. Proceedings of the 2008 Jackson Hole Economic Policy Symposium, Jackson Hole, Wyoming, August 30-September 1, 2007. Kansas City, MO: Federal Reserve Bank of Kansas City, 2008, pp. 149-233; http://www.kansascityfed.org/publicat/sympos/2007/PDF/Leamer_0415.pdf. Orphanides, Athanasios and Van Norden, Simon. “The Unreliability of Output-Gap Estimates in Real Time.” Review of Economics and Statistics, November 2002, 84(4), pp. 569-83. Poole, William. “Safeguarding Good Policy Practice.” Federal Reserve Bank of St. Louis Review, March/April 2005, 87(2, Part 2), pp. 303-06; http://research.stlouisfed.org/publications/review/05/03/part2/PanelDiscussion2.pdf. Reinhart, Carmen N. and Rogoff, Kenneth S. This Time Is Different: Eight Centuries of Financial Folly. Princeton, NJ: Princeton University Press, 2009. Stock, James H. and Watson, Mark W. “Disentangling the Channels of the 2007-2009 Recession.” Brookings Papers on Economic Activity, Spring 2012, pp. 81-135. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 53 Kliesen Wheelock, David C. and Wohar, Mark E. “Can the Term Spread Predict Output Growth and Recessions? A Survey of the Literature.” Federal Reserve Bank of St. Louis Review, September/October 2009, 91(5, Part 1), pp. 419-40; http://research.stlouisfed.org/publications/review/09/09/part1/Wheelock.pdf. Zheng, Linpeng. “What Do Financial Market Indicators Tell Us?” Federal Reserve Bank of St. Louis Liber8 Economic Information Newsletter. January 2012; http://research.stlouisfed.org/pageone-economics/uploads/newsletter/2012/Lib0112.pdf. 54 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW QE: Is There a Portfolio Balance Effect? Daniel L. Thornton The Federal Open Market Committee has recently attempted to stimulate economic growth using unconventional methods. Prominent among these is quantitative easing (QE)—the purchase of a large quantity of longer-term debt on the assumption that it will reduce long-term yields through the portfolio balance channel. Former Federal Reserve Chairman Ben Bernanke and others suggest that QE works through the portfolio balance channel, which implies a strong, statistically significant positive relationship between the public’s holding of long-term Treasury debt and long-term Treasury yields. The author uses the econometric approach of Gagnon et al. (2011) and others to investigate the relationship between a variety of measures of the public’s debt holding and various yield measures in the literature. The empirical results provide virtually no support for the portfolio balance channel. (JEL E52, E58, E43, E44) Federal Reserve Bank of St. Louis Review, First Quarter 2014, 96(1), pp. 55-72. All that quantitative easing (QE) does is to restructure the maturity of U.S. government debt in private hands. Now, of all the stories you’ve heard why unemployment is stubbornly high, how plausible is this: “The main problem is the maturity structure of debt. If only Treasury had issued $600 billion more bills and not all these 5 year notes, unemployment wouldn’t be so high. It’s a good thing the Fed can undo this mistake.” Of course that’s preposterous. —John H. Cochrane, December 7, 2010 he Federal Reserve aggressively increased the size of its balance sheet in the wake of Lehman Brothers’ bankruptcy announcement on September 15, 2008. Coincident with the massive increase in the supply of reserves, the federal funds rate fell to nearly zero.1 With the funds rate at effectively zero, the Federal Open Market Committee (FOMC) turned to unconventional monetary policies. Prominent among these is the policy of large-scale asset purchases (LSAPs), referred to as quantitative easing (QE). The goal of QE is to stimulate investment and consumption by reducing longer-term yields (see, e.g., Woodford, 2001, 2012). Considerable research has been devoted to determining the impact of the Fed’s QE T Daniel L. Thornton is vice president and presidential adviser at the Federal Reserve Bank of St. Louis. The author thanks Clemens Kool for helpful comments and Bryan Noeth, Sean Grover, and Li Li for valuable research assistance. © 2014, The Federal Reserve Bank of St. Louis. The views expressed in this article are those of the author(s) and do not necessarily reflect the views of the Federal Reserve System, the Board of Governors, or the regional Federal Reserve Banks. Articles may be reprinted, reproduced, published, distributed, displayed, and transmitted in their entirety if copyright notice, author name(s), and full citation are included. Abstracts, synopses, and other derivative works may be made only with prior written permission of the Federal Reserve Bank of St. Louis. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 55 Thornton operations on long-term yields. This effort can be divided into two broad strands of the literature. The first is event studies that analyze high-frequency changes in bond yields following QE announcements (e.g., Gagnon et al., 2011; Krishnamurthy and Vissing-Jorgensen, 2011; Joyce et al., 2010; Neely, 2013; Swanson, 2011; Bauer and Rudebusch, 2011; Wright, 2012).2 The second strand of the literature uses lower-frequency (monthly) data to test the implication of the portfolio balance effect—namely, that there is a positive relationship between bond term premiums (and, consequently, bond yields) and the maturity structure of the public’s holding of Treasury debt and long-term Treasury yields (e.g., Gagnon et al., 2011; Hamilton and Wu, 2012; Krishnamurthy and Vissing-Jorgensen, 2012; Greenwood and Vayanos, forthcoming). This strand of the literature investigates the relationship between bond yields or term premiums and various measures of the public’s holdings of Treasury debt prior to the FOMC’s QE activities. Evidence of a statistically significant and economically important positive relationship prior to the FOMC’s actions is used to infer how the FOMC’s actions to reduce these supply measures should have affected term premiums or bond yields. This article contributes to this literature by investigating the relationship between long-term yields and the public’s holding of long-term debt using a wide array of public debt and bond yield measures found in the literature. Following Gagnon et al. (2011) and Krishnamurthy and Vissing-Jorgensen (2012), I investigate the relationship within the context of a simple reduced-form framework that controls for a variety of macroeconomic and other variables. In implementing this methodology, I account for the trend in term premiums and bond yields over the sample period. To preview the empirical results, when the trend is accounted for, there is little evidence of a statistically significant effect of the Fed’s LSAPs on yields and no evidence of an economically meaningful effect. The article proceeds as follows. The next section briefly discusses the portfolio balance channel. I then review the previous empirical work in the literature and examine a variety of public debt, maturity/duration, and interest rate measures used in the literature. The next section presents the empirical results using these measures. THE PORTFOLIO BALANCE CHANNEL Many researchers and policymakers hypothesize that the Fed’s LSAPs affect long-term yields through a variety of channels (see, e.g., Krishnamurthy and Jorgensen, 2011). However, many analysts (e.g., Bernanke, 2010, and Gagnon et al., 2011) have suggested that QE works through the portfolio balance channel.3 For example, Gagnon et al. (2011, p. 7) say that QE reduces long-term yields because the Fed’s LSAPs remove “a considerable amount of assets with high duration from the markets. With less duration risk to hold in the aggregate, the market should require a lower premium to hold that risk.” For the portfolio balance channel to be operative, the market for long-term debt must be effectively segmented from the rest of the financial market. Until recently, the idea that markets are segmented had gained relatively little traction among financial economists and policymakers. Skepticism that financial markets are segmented stems in part from the fact that yield differentials create arbitrage opportunities that the market will exploit. 56 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Thornton Consequently, it is not surprising that several analysts have expressed skepticism about the empirical significance of the portfolio balance channel. For example, Cochrane (2011, p. 16) suggests that the portfolio balance channel should be relatively weak because (i) “much of the Treasury bond supply is locked away in central bank and pension fund vaults” and “arbitrageurs take duration risk in mortgage-backed, corporate, and other markets” and (ii) the Fed’s QE operations are “just a drop in the bucket.” Bauer and Rudebusch (2011, p. 6) make a similar argument, noting that “the scale of the Fed’s purchases of $1.725 trillion of debt securities is small relative to the size of [domestic] bond portfolios” and “the global bond market—arguably, the relevant one—is several times larger.” They conclude that the portfolio balance channel should be weak and suggest that the FOMC’s LSAPs affect long-term yield through the signaling channel. Kocherlakota (2010) suggests that QE merely shifts interest rate risk from bondholders to taxpayers, and as a result “QE ends up having no effects, except for those associated with any new forward guidance that it signals.” PREVIOUS LOW-FREQUENCY RESEARCH Hancock and Passmore (2011) and Stroebel and Taylor (2009) use lower-frequency monthly data to investigate the effect of the FOMC’s purchases of mortgage-backed securities and find mixed results. Hancock and Passmore (2011) find a relatively large impact, while Stroebel and Taylor (2009) find a relatively small or statistically insignificant effect. D’Amico and King (2010) and D’Amico et al. (2012) investigate the effects of QE on the Treasury yield curve using micro-transactions data. D’Amico and King (2010) estimate both flow and stock effects; the former are the response of prices to ongoing purchases and the latter are changes due to expectations about future withdrawals of supply. They find small and temporary flow effects. The stock effect based on a counterfactual yield curve from their model suggests that the nearly $300 billion purchase of Treasury securities would flatten the yield curve in the range of 10 to 15 years by 45 basis points. However, when the observations on key QE announcements days are omitted, only one of the own response or cross-response coefficients is statistically significant. This fact would seem to suggest that their results are critically dependent on an announcement effect. D’Amico et al. (2012) suggest that QE can affect long-term yields and term premiums through three channels. The first is called the scarcity channel, which they define as “a mechanism under which the purchase by the Federal Reserve of assets with a specific maturity leads to higher prices (and lower yields) of securities with similar maturities” (p. 2). The second is called the duration channel, defined as “a mechanism under which the removal…of aggregate duration from the outstanding stock of Treasury debt reduces term premiums on securities across maturities” (p. 2). The duration channel seems to be very similar to the portfolio balance channel. The third is the signaling channel. D’Amico et al. identify scarcity by creating maturity “buckets” consisting of the public’s holdings of Treasury securities of given maturities relative to total Treasury debt outstanding.4 They find that both the scarcity and duration channels are statistically significant; however, the duration channel accounts for only a third or a fourth of their estimate of the total effect. They find no evidence of an important signaling channel. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 57 Thornton Greenwood and Vayanos (forthcoming) focus directly on the portfolio balance channel by organizing their empirical analysis based on Vayanos and Vila’s (2009) segmented-markets model. Specifically, they estimate regressions of bond yields and excess returns on the ratio of maturity-weighted debt to gross domestic product (GDP). They find a positive relationship between their debt measure and both yields and returns, but the estimates are statistically significant only for returns on bonds with maturities of five years or longer. Just as Greenwood and Vayanos (forthcoming) do, Gagnon et al. (2011) investigate the effect of QE on long-term yields using a measure of the public’s holding of Treasury debt. Specifically, Gagnon et al. (2011) estimate the equation as follows: (1) it = α + Xt β + δ pdt + εt . Gagnon et al. (2011) use two measures of i, the 10-year Treasury yield and an estimate of the 10-year Treasury term premium. X is a [1 × K] vector of macroeconomic variables and pd is a measure of the public’s holding of Treasury debt, where a, b, and d are constant coefficients, and e is a random error with a zero mean and a constant variance. They estimate the equation over the period January 1985 through June 2008. Their estimate of the supply effect suggests that the FOMC’s $1.75 trillion asset purchase (QE1) should have reduced the term premium by about 52 basis points and the 10-year Treasury yield by about 82 basis points. Hamilton and Wu (2012) investigate the effect of QE by estimating a three-factor affine term structure model, using assumptions motivated by Vayanos and Vila’s (2009) model. Specifically, they calculate arbitrageurs’ portfolio weights under the assumptions that arbitrageurs comprise the entire private sector and U.S. Treasury debt is held only by arbitrageurs. They use the estimates from their model to infer how changes in the maturity structure of Treasury debt can affect yields. Their estimates of the effect of QE1 on the 10-year Treasury yield and term spreads are smaller than those reported by Gagnon et al. (2011), Greenwood and Vayanos (forthcoming), and D’Amico and King (2010). The effect of QE2 was perverse: The Fed’s asset purchase program increased Treasury yields and term premiums. Hamilton and Wu (2012, p. 38) attribute this to the fact that the “fraction of publicly held debt of more than 10 years maturity continued to increase even as the Fed was implementing its QE2 bond purchases.” THE RELATIONSHIP BETWEEN LONG-TERM YIELDS AND THE SUPPLY OF PUBLIC DEBT This section investigates the relationship between long-term yields and the supply of public debt using the methodology of Gagnon et al. (2011). The analysis differs from previous work in the literature in that I consider several debt supply measures used in the literature and alternative yield measures. The analysis begins with a discussion of the public debt measures used in the literature. 58 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Thornton Figure 1 Public’s Holding of Treasury Debt, Net of SOMA $ Billions 4,500 4,000 3,500 3,000 2,500 2,000 1,500 1,000 500 Gagnon et al. (2011) Hamilton and Wu (2012) 0 Jan-90 Jul-91 Jan-93 Jul-94 Jan-96 Jul-97 Jan-99 Jul-00 Jan-02 Jul-03 Jan-05 Jul-06 Jan-08 Alternative Debt Supply Measures This section analyzes alternative debt measures used in the literature to investigate the portfolio balance channel. The debt measures are those used by Gagnon et al. (2011), Hamilton and Wu (2012), and Greenwood and Vayanos (forthcoming).5 Gagnon et al. (2011) and Hamilton and Wu (2012) use data on the public’s holding of Treasury debt, less that held by the Fed in the System Open Market Account (SOMA). Figure 1 shows these series for the period January 1990 through June 2008. The series are nearly identical until the late 1990s when they begin to diverge. The difference is likely due to the inclusion of Treasury inflationprotected securities (TIPS) in Gagnon et al.’s data. Hamilton and Wu’s data do not include TIPS.6 Gagnon et al. (2011) consider only the public’s holdings of government debt with maturities of one year or longer, net of SOMA. This series is shown in Figure 2. However, they make several adjustments to this series. First, they subtract foreign official holdings of Treasury securities with maturities of one year or longer because foreign governments are unlikely to have a term premium similar to that of the private sector. The resulting series (S2) is also shown in Figure 2. Rather than using the S2 series, Gagnon et al. (2011) also subtract foreign official holdings of agency and private sector debt with maturities of at least one year. This adjustment is inappropriate because agency and private securities are not included in S2. The resulting series (S3), also shown in Figure 2, is negative beginning in November 2007, when foreign official holdings of agency and private sector debt become larger than the public’s holding of Treasury debt.7 As a final adjustment, Gagnon et al. (2011) express S3 as a percent of nominal GDP (S3 gdp). Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 59 Thornton Figure 2 Alternative Measure of the Public’s Holding of Government Debt $ Billions 3,000 2,500 2,000 1,500 1,000 500 S1 S2 S3 0 –500 Jan-90 Jul-91 Jan-93 Jul-94 Jan-96 Jul-97 Jan-99 Jul-00 Jan-02 Jul-03 Jan-05 Jul-06 Jan-08 Figure 3 Gagnon et al. and Greenwood and Vayanos Supply Measures Percent of GDP Maturity-Weighted Debt-to-GDP Ratio 25 4.5 4.0 20 3.5 15 3.0 2.5 10 2.0 5 1.5 1.0 0 Gagnon et al. (left axis) 0.5 Greenwood and Vayanos (right axis) 60 First Quarter 2014 7 06 l-0 Ju 4 nJa 03 l-0 Ju 1 00 l-0 nJa Ju 8 nJa 97 l-9 Ju 5 94 l-9 nJa Ju 2 nJa 91 l-9 Ju 9 l-8 nJa Ju nJa l-8 6 88 0 Ju Ja n- 85 –5 Federal Reserve Bank of St. Louis REVIEW Thornton Figure 4 Average Maturity and Percent of Debt Greater Than 10 Years to Maturity Percent Months to Maturity 80 25 70 20 60 50 15 40 10 30 20 5 P10 AVE 0 10 0 Jan-90 Jul-91 Jan-93 Jul-94 Jan-96 Jul-97 Jan-99 Jul-00 Jan-02 Jul-03 Jan-05 Jul-06 Jan-08 Greenwood and Vayanos (forthcoming) use data from the Center for Research in Securities Prices (CRSP) for every government bond issued between 1940 and 2007 without netting out Fed or agency holdings. Specifically, they construct the maturity structure of the debt by aggregating cash flows across individual bonds—that is, the sum of all principal and coupon payments due over the maturity of each bond issued. They then construct a maturity-weighted debt-to-GDP ratio supply measure. Figure 3 shows this measure with Gagnon et al.’s (2011) S3 gdp measure.8 While the two measures differ greatly in scale, they follow a very similar pattern: Both rise until the mid-1990s and then decline. Gagnon et al.’s measure declines more dramatically because they subtract foreign official holdings of agency and private debt, while Greenwood and Vayanos’s measure is based solely on the public’s holdings of Treasury debt. Hamilton and Wu (2012) do not use the debt measure shown in Figure 1 but rather the average maturity of public debt (AVE) and the percent of public debt with maturities longer than 10 years (P10) (Figure 4). The series behave similarly over the sample period; the correlation is 84 percent. The Data: Term Premiums and Treasury Yields This section considers alternative measures of interest rates used in the literature. Gagnon et al. (2011) evaluate the effectiveness of QE by estimating the effect of the Fed’s purchase of securities on an estimate of the 10-year Treasury term premium (TP) and the zerocoupon 10-year Treasury bond yield (T10).9 Their estimate of TP is obtained from the term structure model of Kim and Wright (2005). Figure 5 shows strong negative trends and similar Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 61 Thornton Figure 5 10-Year Treasury Yield and Term Premium Percent 14 T10 TP 12 10 8 6 4 2 7 06 l-0 Ju 4 nJa 03 l-0 Ju 1 n- 00 l-0 Ja Ju n- 8 97 l-9 Ja Ju 5 n- 94 l-9 Ja Ju 2 nJa 91 l-9 Ju 9 nJa 88 l-8 Ju 6 n- l-8 Ja nJa Ju 85 0 Figure 6 Public’s Holding of Treasury Debt, Net of SOMA Percent 5 SYC TP 4 3 2 1 0 62 First Quarter 2014 7 l-0 06 nJa Ju 4 03 l-0 Ju n- l-0 1 Ja Ju 00 n- l-9 97 8 Ja Ju n- l-9 94 5 Ja Ju Ja n- 2 91 l-9 Ju Ja n- 9 88 l-8 Ju n- l-8 6 Ja Ju Ja n- 85 –1 Federal Reserve Bank of St. Louis REVIEW Thornton Table 1 Estimates of Equation (1) Not Accounting for the Trend (January 1985–June 2008) Variable Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value Const. 0.203 0.775 –3.277 0.000 –2.730 0.000 –2.182 0.000 –2.503 0.000 gap 0.241 0.000 0.220 0.000 0.211 0.000 0.180 0.002 0.192 0.001 cpi 0.320 0.000 0.497 0.000 0.433 0.000 0.307 0.000 0.284 0.000 lrid 0.250 0.015 0.374 0.001 0.381 0.000 0.377 0.001 0.378 0.001 rv 0.492 0.053 1.225 0.000 1.094 0.000 0.943 0.000 1.049 0.000 S1 –0.001 0.003 0.001 0.000 0.001 0.000 0.044 0.000 0.289 0.000 S2 S3 S3 gdp S GV – R2 0.812 0.816 0.842 0.847 0.843 SE 0.402 0.398 0.365 0.363 0.362 NOTE: SE, standard error. cycles for T10 and TP. The correlation between TP and T10 is very high, 94 percent. Given the similarity of these series, it is not surprising that the results are qualitatively similar with either measure. Hamilton and Wu (2012) investigate the effectiveness of the portfolio balance channel using the slope of the yield curve (SYC), measured by the difference between the constant maturity 10-year Treasury zero coupon bond yield and the 6-month T-bill rate. Figure 6 shows SYC and TP over the period January 1985 through June 2008. The Effectiveness of the Portfolio Balance Channel This section reports the estimates of equation (1) using these alternative measures of it (TP and SYC) and a variety of measures of pdt . The macroeconomic variables are those used by Gagnon et al. (2011): the unemployment gap (gap), core consumer price index inflation (cpi), long-run inflation disagreement (lrid), and 6-month realized daily volatility of the onthe-run 10-year Treasury yield (rv).10 Table 1 presents the estimates using TP as the dependent variable and the alternative measures of the public’s holding of Treasury debt discussed previously (see “Alternative Debt Supply Measures”). Contrary to the implication of the portfolio balance channel, the coefficient on the public’s holding of debt net of SOMA, S1, is negative and statistically significant. However, when foreign official holdings of Treasury debt are netted out, the estimate is positive and statistically significant. A coefficient of the same magnitude and statistical significance is obtained when foreign official holdings of agency and private debt are netted out. Hence, despite the abnormal nature of this adjustment, it has no effect on the results: A $600 billion LSAP would reduce the term premium by 40 basis points. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 63 Thornton Table 2 Estimates of Equation (1) Accounting for the Trend (January 1985–June 2008) Variable Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value Const. 1.035 0.058 0.164 0.850 –0.380 0.671 –0.071 0.945 –0.399 0.715 gap 0.205 0.000 0.200 0.000 0.201 0.000 0.192 0.001 0.196 0.001 cpi 0.109 0.097 0.156 0.054 0.207 0.024 0.158 0.056 0.167 0.032 lrid 0.244 0.029 0.276 0.020 0.301 0.009 0.292 0.016 0.297 0.013 rv 0.394 0.056 0.574 0.005 0.665 0.001 0.590 0.003 0.681 0.002 Trend –0.006 0.000 –0.006 0.000 –0.004 0.010 –0.005 0.053 –0.004 0.069 S1 –0.0001 0.533 S2 0.0002 0.409 S3 0.0003 0.145 S3 gdp 0.0160 0.368 S GV 0.1150 0.363 – R2 0.855 0.856 0.856 0.857 0.851 SE 0.353 0.352 0.348 0.352 0.353 NOTE: SE, standard error. The estimates using S3 gdp are, of course, identical to those of Gagnon et al. (2011). While the coefficient on S3 gdp is larger than that on S3, the estimated magnitude of the effect of LSAPs is smaller. A $600 billion LSAP is about 4.0 percent of 2009 GDP, so the estimated effect of the same $600 billion LSAP would be half as large, about 19 basis points. Given the similarity between S3 gdp and SGV, it is not surprising that the estimate of the coefficient also is positive and highly statistically significant when Greenwood and Vayanos’s (forthcoming) measure is used. Unfortunately, the results in Table 1 are the consequence of trends in TP and public debt measures. A simple linear trend accounts for nearly 80 percent of the variation in TP and a small but still relatively large (as much as 60 percent) portion of the variation in the supply measures. The existence of trends can lead to spurious regression in that two variables that trend in the same direction will be positively correlated even if they are independent of each other. The importance of the trend in these measures is shown in Table 2, which reports the estimates of equation (1) when a simple linear trend is included in the equation. The coefficient on S1 remains negative but is not statistically significant. The coefficients on the other debt measures remain positive but much smaller than the estimates in Table 1; more importantly, none is statistically significant at even the 10 percent significance level. When the trend is accounted for, the statistical support for the portfolio balance channel vanishes. This conclusion is the same if the trend is obtained using the Hodrick-Prescott filter or if the equation is estimated in first differences. Hence, there is no statistically significant positive relationship between the term premium and any of the debt measures considered here (i.e., no statistical support for the portfolio balance channel) when the trend is accounted for. 64 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Thornton Table 3 Estimates of Equation (1) Using Alternative Supply Measures and Accounting for the Trend (January 1985–June 2008) Variable Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value Const. 0.0831 0.9198 0.6770 0.2145 2.4388 0.0092 9.0658 0.0000 –0.0500 0.9602 gap 0.2169 0.0002 0.2068 0.0003 0.2055 0.0002 0.3410 0.0000 0.2028 0.0002 cpi 0.1874 0.0006 0.1882 0.0006 0.0923 0.2075 0.0227 0.6266 0.1670 0.0399 lrid –0.0577 0.4817 –0.0611 0.4622 0.1729 0.0904 0.0821 0.1560 0.2898 0.0143 rv Trend AM 0.6981 0.0007 0.6682 0.0012 0.4253 0.0507 0.5823 0.0000 0.6110 0.0027 –0.0060 0.0000 –0.0069 0.0000 –0.0061 0.0000 0.0002 0.7840 –0.0055 0.0001 0.0130 0.1439 0.0246 0.2020 –0.2569 0.0497 –1.1908 0.0000 P10 DUR DUR10 S2duradj 0.0003 0.2970 – R2 0.8366 0.8357 0.8588 0.9227 0.8540 SE 0.2995 0.3003 0.3452 0.2555 0.3511 NOTE: SE, standard error. Table 3 presents the estimates using five alternative supply measures: the average maturity of the debt (AM), the percent of the public’s holding of debt with maturity of 10 years or longer (P10), the duration of the public’s holding of the debt (DUR), the duration of the on-the-run 10-year Treasury securities (DUR10), and the S2 debt measure adjusted for the duration of the debt using Gagnon et al.’s (2011) adjustment procedure (S2duradj). AM and P10 are calculated from Hamilton and Wu’s (2012) data, which cover the period January 1990 through June 2008. Hence, this is the sample period when these variables are used. DUR and DUR10 data were provided by Gagnon et al. (2011). The coefficients on AM and P10 are positive, but neither is statistically significant. The estimates for the two duration measures are negative and statistically significant, suggesting that a shortening of the duration of the public’s holding of government debt as the result of LSAPs would increase the term premium. The coefficient on S2duradj is positive but not statistically significant. Hence, these alternative measures also provide no support for the portfolio balance channel.11 Again, the conclusion is robust to the measure of trend used, whether the equation is estimated using first differences or whether T10 is the dependent variable (Tables 4 and 5). The portfolio balance channel is thought to reduce longer-term rates relative to shorterterm rates, so equation (1) is estimated using SYC as the dependent variable and all 10 supply measures. Unlike TP and T10, there is no significant trend in SYC. However, SYC is highly persistent, so SYCt–1 is included in the regression. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 65 Thornton Table 4 Estimates of Equation (1) Using Dependent Variable T10 and Not Accounting for the Trend (January 1985–June 2008) Variable Const. Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value 4.8405 0.0000 4.0288 0.0018 4.3217 0.0036 4.4339 0.0059 gap –0.2472 0.0037 –0.2529 0.0035 –0.2477 0.0036 –0.2576 0.0048 cpi 0.3100 0.0044 0.3770 0.0029 0.3734 0.0100 0.3389 0.0092 lrid 0.6654 0.0003 0.7113 0.0001 0.7003 0.0001 0.7002 0.0001 rv Trend S1 0.2827 0.4145 0.4109 0.2282 0.3425 0.3308 0.3032 0.3738 –0.0146 0.0000 –0.0123 0.0000 –0.0119 0.0000 –0.0117 0.0016 0.0003 0.3116 0.0004 0.2610 0.0002 0.4527 0.0165 0.5699 S2 S3 S3 gdp – R2 0.8999 0.9007 0.8998 0.8995 SE 0.5536 0.5513 0.5539 0.5548 NOTE: SE, standard error. Table 5 Estimates of Equation (1) Using Dependent Variable T10 and Accounting for the Trend (January 1985–June 2008) Variable Const. Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value 6.6212 0.0000 6.1643 0.0000 11.0086 0.0000 19.8398 0.0000 4.4411 0.0012 gap –0.2560 0.0021 –0.2567 0.0021 –0.2396 0.0039 –0.0044 0.9117 –0.2462 0.0035 cpi 0.3168 0.0034 0.3003 0.0045 0.2136 0.0402 0.1312 0.0259 0.3490 0.0052 lrid 0.5869 0.0009 0.5998 0.0009 0.3925 0.0088 0.3632 0.0000 0.6992 0.0001 rv 0.0956 0.7541 0.1266 0.6949 0.0761 0.8035 0.3908 0.0098 0.3286 0.3340 Trend –0.0143 0.0000 –0.0139 0.0000 –0.0125 0.0000 –0.0019 0.0741 –0.0126 0.0000 AM –0.0146 0.4315 –0.0364 0.3105 –0.8946 0.0000 –2.0904 0.0000 P10 DUR DUR10 S2duradj 0.0004 0.4904 – R2 0.8999 0.9001 0.9238 0.9594 0.8996 SE 0.5537 0.5532 0.4831 0.3527 0.5544 NOTE: SE, standard error. 66 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Thornton Table 6 Estimates of Equation (1) with SYC as Dependent Variable (January 1985–June 2008) Variable Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value Const. 0.0568 0.8479 0.2655 0.1046 0.1996 0.0456 0.1632 0.0646 0.1640 0.1060 gap 0.1086 0.0071 0.1135 0.0050 0.1088 0.0053 0.1137 0.0034 0.1108 0.0038 cpi 0.0338 0.2599 0.0264 0.1451 0.0326 0.0637 0.0486 0.0148 0.0422 0.0528 lrid –0.1099 0.0001 –0.1194 0.0001 –0.1189 0.0000 –0.1212 0.0000 –0.1160 0.0001 rv 0.2528 0.0013 0.2125 0.0037 0.2214 0.0027 0.2343 0.0015 0.2164 0.0026 SYC t–1 0.9085 0.0000 0.9066 0.0000 0.9112 0.0000 0.9112 0.0000 0.9119 0.0000 S1 0.0000 0.8088 –0.0001 0.3315 0.0000 0.1937 –0.0051 0.0795 S2 S3 S3 gdp S GV –0.0170 0.4623 – R2 0.9686 0.9687 0.9688 0.9691 0.9703 SE 0.2024 0.2018 0.2015 0.2007 0.1984 NOTE: SE, standard error. Table 7 Estimates of Equation (1) with SYC and Alternative Supply Measures (January 1985–June 2008) Variable Const. gap Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value –0.4507 0.0115 –0.5291 0.0013 –0.8071 0.0531 –0.5240 0.3026 0.1946 0.1944 0.1708 0.0021 0.2173 0.0002 0.1289 0.0013 0.0998 0.0119 0.1105 0.0049 cpi 0.0257 0.1834 0.0189 0.2795 0.0482 0.0124 0.0568 0.0414 0.0282 0.1149 lrid –0.0387 0.4037 –0.0399 0.3751 –0.0709 0.0150 –0.0998 0.0005 –0.1161 0.0001 rv 0.5325 0.0001 0.5211 0.0001 0.2856 0.0006 0.2563 0.0006 0.2257 0.0022 SYC t–1 0.8471 0.0000 0.8177 0.0000 0.8905 0.0000 0.9112 0.0000 0.9075 0.0000 AM 0.0027 0.3846 0.0201 0.0959 0.1375 0.0268 0.0705 0.2073 P10 DUR DUR10 S2duradj 0.0000 0.6032 – R2 0.9740 0.9741 0.9699 0.9689 0.9686 SE 0.1948 0.1940 0.1980 0.2014 0.2023 NOTE: SE, standard error. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 67 Thornton Table 6 presents the estimates using S1, S2, S3, S3 gdp, and SGV. None of these measures provides support for the portfolio balance channel. The coefficients on S1 and S3 are positive but not statistically significant. The coefficients on S2, S3 gdp, and SGV are negative; however, none is statistically significant at the 5 percent significance level. Table 7 presents the estimates using the five alternative supply measures. These results are somewhat more supportive of the portfolio balance channel. The coefficient estimates for AM and P10 are positive but not statistically significant at the 5 percent level. The coefficient on P10 is marginally significant at the 10 percent level but is small: A 1-percentage-point increase in the percent of debt with maturity of 10 years or longer increases SYC by only 2 basis points. The coefficient on DUR is positive and statistically significant. A 1-year increase in DUR would increase the SYC by nearly 14 basis points. However, the standard deviation of DUR over the sample period is about 0.5 years. Consequently, a relatively large change in DUR would be required to have a very large effect on SYC. D’Amico et al. (2012) indicate that the average duration of the Treasury securities held by the public declined from 4.42 years to 4.30 years during the first LSAP. Hence, this would account for only about a 2-basis-point flattening of the yield curve during this period. They note that QE1 removed only 0.10 years of duration from the market, so the duration effect of QE2 would be even smaller. The duration-adjusted S2 supply measure also provides no support for the portfolio balance channel. The estimated coefficient is positive, but it is also very small and not statistically significant. Overall, the evidence is more supportive of the portfolio balance effect using SYC. Two of the 10 supply measures yield coefficients that are correctly signed and statistically significant. However, in either case the effect on SYC is modest. Consequently, the measures cannot account for the well-documented decline in long-term interest rates and the term premium reported in the event study literature. CONCLUSION With its principal policy tool—the federal funds rate—effectively at zero, the FOMC has attempted to stimulate aggregate demand by reducing longer-term rates through the so-called signaling and portfolio balance channels of policy. The portfolio balance channel assumes that the market for long-term Treasury securities is segmented from the rest of the financial market and hypothesizes a positive relationship between the term premium in long-term bonds and the quantity of long-term debt held by the public. By implication, the portfolio balance channel suggests that term premiums, and consequently long-term Treasury yields, can be reduced through LSAPs or by purchasing longer-term securities while simultaneously selling an equal quantity of shorter-term securities. This article uses the reduced-form methodology of Gagnon et al. (2011) and Krishnamurthy and Vissing-Jorgensen (2012) to investigate the portfolio balance channel with 3 interest rate measures and 10 public debt supply measures from the literature. The results indicate there is 68 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Thornton no statistically significant positive relationship between either the term premium or the 10-year Treasury yield and any of the 10 supply measures. However, there is a statistically significant relationship between the SYC and 2 of the 10 supply measures. The estimate of the effect is very small and cannot account for the estimates of the effect of LSAPs on long-term yields reported in the event study literature. Hence, there appears to be no empirical support for the idea that these purchases reduced long-term yields or flattened the yield curve by reducing the public’s holdings of long-term debt as the portfolio balance channel suggests it should have. ■ Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 69 Thornton NOTES 1 See Thornton (2010). 2 See Thornton (2013) for an analysis of the extent to which the announcement effects in the event study literature are identified. 3 Specifically, Bernanke notes that “The channels through which the Fed’s purchases affect longer-term interest rates and financial conditions more generally have been subject to debate. I see the evidence as most favorable to the view that such purchases work primarily through the so-called portfolio balance channel, which holds that once short-term interest rates have reached zero, the Federal Reserve’s purchases of longer-term securities affect financial conditions by changing the quantity and mix of financial assets held by the public.” 4 This formulation seems at odds with the concept of scarcity. A more natural measure would seem to be the public’s holding of Treasury securities of given maturities relative to the total supply of those maturities. 5 I would like to thank the authors of these studies for providing the data. The data for the article by Hamilton and Wu (2012) can be found at http://research.stlouisfed.org/econ/thornton/. 6 The results are quantitatively similar and the qualitative conclusions identical if the Hamilton and Wu base series is used, suggesting that including or excluding TIPS has only a minor effect on the results. 7 They provide no reason for making this adjustment. They merely note that they made it. 8 The Greenwood and Vayanos measure is available only through December 2007. Hence, estimates of equation (1) using their measure are based on monthly data over the period January 1985 through December 2007. 9 Bauer and Rudebusch (2011) have an alternative estimate of the risk premium. However, their measure behaves similarly to that of Gagnon et al. (2011). Indeed, the qualitative conclusions presented in this section are the same when the Bauer and Rudebusch measure is used. 10 See Gagnon et al. (2011) for the precise definitions of these variables. 11 As before, the qualitative conclusions are unchanged if the 10-year Treasury yield or Bauer and Rudebusch’s (2011) measure of the term premium is used. For completeness, Hamilton and Wu’s (2012) supply factors were also used. These factors are available for the period January 1990 through July 2007. None of these supply factors was statistically significant when either TP or T10 was the dependent variable. REFERENCES Bauer, Michael and Rudebusch, Glenn. “The Signaling Channel for Federal Reserve Bond Purchases.” Working Paper No. 2011-21, Federal Reserve Bank of San Francisco, September 2011; http://www.frbsf.org/economic-research/files/wp11-21bk.pdf. Bernanke, Ben S. “The Economic Outlook and Monetary Policy.” Speech at the Federal Reserve Bank of Kansas City Economic Policy Symposium, Jackson Hole, Wyoming, August 27, 2010; http://www.federalreserve.gov/newsevents/speech/bernanke20100827a.htm. Cochrane, John H. “Sense and Nonsense in the Quantitative Easing Debate.” VOX, December 7, 2010; http://www.voxeu.org/index.php?q=node/5900. Cochrane, John H. “Inside the Black Box: Hamilton, Wu, and QE2.” Unpublished manuscript, University of Chicago Booth School of Business, March 3, 2011; http://faculty.chicagobooth.edu/john.cochrane/research/papers/hamiton_wu_term_structure.pdf. D’Amico, Stefania; English, William B.; Lopez-Salido, J. David and Nelson, Edward. “The Federal Reserve’s LargeScale Asset Purchase Programs: Rationale and Effects.” Finance and Economics Discussion Series No. 2012-85, Board of Governors of the Federal Reserve System, December 2012; http://www.federalreserve.gov/pubs/feds/2012/201285/201285abs.html. 70 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Thornton D’Amico, Stefania and King, Thomas B. “Flow and Stock Effects of Large-Scale Treasury Purchases.” Finance and Economics Discussion Series No. 2010-52, Board of Governors of the Federal Reserve System, September 2010; http://www.federalreserve.gov/pubs/feds/2010/201052/201052abs.html. Gagnon, Joseph; Raskin, Matthew; Remache, Julie and Sack, Brian. “The Financial Market Effects of the Federal Reserve’s Large-Scale Asset Purchases.” International Journal of Central Banking, March 2011, 7(1), pp. 3-43. Greenwood, Robin and Vayanos, Dimitri. “Bond Supply and Excess Bond Returns.” Review of Financial Studies (forthcoming). Advance Access published online January 14, 2014; http://rfs.oxfordjournals.org/content/early/2014/01/14/rfs.hht133.full.pdf+html. Hamilton, James D. and Wu, Jing Cynthia. “The Effectiveness of Alternative Monetary Policy Tools in a Zero Lower Bound Environment.” Journal of Money, Credit, and Banking, February 2012, 44(Suppl. s1), pp. 3-46. Hancock, Diana and Passmore, Wayne. “Did the Federal Reserve’s MBS Purchase Program Lower Mortgage Rates?” Journal of Monetary Economics, July 2011, 58(5), pp. 498-514. Joyce, Michael; Lasaosa, Ana; Stevens, Ibrahim and Tong, Matthew. “The Financial Market Impact of Quantitative Easing.” Working Paper 393, Bank of England, August 2010; http://www.bankofengland.co.uk/research/Documents/workingpapers/2010/wp393.pdf. Kim, Don H. and Wright, Jonathan H. “An Arbitrage-Free Three-Factor Term Structure Model and the Recent Behavior of Long-Term Yields and Distant-Horizon Forward Rates.” Finance and Economics Discussion Series No. 2005-33, Board of Governors of the Federal Reserve System, August 2005; http://www.federalreserve.gov/pubs/feds/2005/200533/200533pap.pdf. Kocherlakota, Narayana. “Economic Outlook and the Current Tools of Monetary Policy.” Speech at the European Economics and Financial Centre, London, England, September 29, 2010; http://minneapolisfed.org/news_events/pres/speech_display.cfm?id=4555. Krishnamurthy, Arvind and Vissing-Jorgensen, Annette. “The Effects of Quantitative Easing on Long-Term Interest Rates.” Brookings Papers on Economic Activity, Fall 2011, pp. 215-65. Krishnamurthy, Arvind and Vissing-Jorgensen, Annette. “The Aggregate Demand for Treasury Debt.” Journal of Political Economy, April 2012, 120(2), pp. 233-67. Neely, Christopher J. “Unconventional Monetary Policy Had Large International Effects.” Working Paper No. 2010-018D, Federal Reserve Bank of St. Louis, July 2010, revised August 2013; http://research.stlouisfed.org/wp/2010/2010-018.pdf. Stroebel, Johannes and Taylor, John B. “Estimated Impact of the Fed’s Mortgage-Backed Securities Purchase Program.” NBER Working Paper No. 15626, National Bureau of Economic Research, December 2009; http://www.nber.org/papers/w15626.pdf?new_window=1. Swanson, Eric T. “Let’s Twist Again: A High-Frequency Event-Study Analysis of Operation Twist and Its Implications for QE2.” Working Paper No. 2011-08, Federal Reserve Bank of San Francisco, February 2011; http://www.frbsf.org/economic-research/files/wp11-08bk.pdf. Thornton, Daniel L. “Can the FOMC Increase the Funds Rate Without Reducing Reserves?” Federal Reserve Bank of St. Louis Economic Synopses, 2010, No. 28, October 6, 2010; http://research.stlouisfed.org/publications/es/10/ES1028.pdf. Thornton, Daniel L. “An Evaluation of Event-Study Evidence on the Effectiveness of the FOMC’s LSAP Program: The Reasonable Person Standard.” Working Paper No. 2013-033A, Federal Reserve Bank of St. Louis, October 2013; http://research.stlouisfed.org/wp/2013/2013-033.pdf. Vayanos, Dimitri and Vila, Jean-Luc. “A Preferred-Habitat Model of the Term Structure of Interest Rates.” NBER Working Paper No. 15487, National Bureau of Economic Research, November 2009; http://www.nber.org/papers/w15487.pdf?new_window=1. Woodford, Michael. “Monetary Policy in the Information Economy,” in Economic Policy for the Information Economy. Proceedings of the Economic Policy Symposium sponsored by the Federal Reserve Bank of Kansas City, Jackson Hole, Wyoming, August 20-September 1, 2001, pp. 297-370; http://www.kansascityfed.org/Publicat/sympos/2001/papers/S02wood.pdf. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 71 Thornton Woodford, Michael. “Methods of Policy Accommodation at the Interest-Rate Lower Bound,” in The Changing Policy Landscape. Proceedings of the Economic Policy Symposium sponsored by the Federal Reserve Bank of Kansas City, Jackson Hole, Wyoming, August 20-September 1, 2012, pp. 185-288; http://www.kansascityfed.org/publicat/sympos/2012/Woodford_final.pdf. Wright, Jonathan H. “What Does Monetary Policy Do to Long-Term Interest Rates at the Zero Lower Bound?” Economic Journal, November 2012, 122(564), pp. F447-66. 72 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW The Evolution of Federal Reserve Policy and the Impact of Monetary Policy Surprises on Asset Prices Brett W. Fawley and Christopher J. Neely This article describes the joint evolution of Federal Reserve policy and the study of the impact of monetary policy surprises on high-frequency asset prices. Since the 1970s, the Federal Open Market Committee has clarified its objectives and modified its procedures to become more transparent and predictable. Researchers have had to account for these changes to procedures and perceived objectives in developing methods to study the effects of monetary surprises. Unexpected changes to the Committee’s federal funds target and postmeeting statements strongly and consistently affect asset prices, including interest rates, exchange rates, and (for target changes) stock prices. The study of monetary surprises on asset prices provides important insight for policymakers, financial market participants, and economic models. (JEL E52, E58, G14) Federal Reserve Bank of St. Louis Review, First Quarter 2014, 96(1), pp. 73-109. ince the 1970s, monetary policy has been the primary macroeconomic stabilization instrument. In light of this fact, many researchers have studied how monetary policy affects asset prices, consumer prices, output, and employment to improve such policy. This large literature has used two main methods to study the effect of monetary shocks on macroeconomic variables: vector autoregressions (VARs) and studies of high-frequency monetary shocks on asset prices.1 VARs offer the advantage of directly studying the effects of monetary policy shocks on key variables—prices, output, and employment—rather than indirectly studying them through their effects on asset prices (see Litterman and Weiss,1985; Strongin,1995; Edelberg and Marshall, 1996; Evans and Marshall, 1998; Bernanke and Mihov, 1998; and Christiano, Eichenbaum, and Evans, 1999). It is difficult to isolate the effects of policy-induced changes in interest rates on monthly or quarterly macroeconomic variables from changes induced by other factors, however, and it is equally difficult to definitively differentiate the effects of monetary policy shocks from the effects of variables to which monetary policy reacts. That is, VAR S Brett Fawley is a quantitative analytics and reporting analyst at Rosen Consulting Group. Christopher J. Neely is an assistant vice president and economist at the Federal Reserve Bank of St. Louis. This article was written while Fawley was a senior research associate at the Federal Reserve Bank of St. Louis. The authors thank Rasmus Fatum, Ken Kuttner, Ed Nelson, Carlo Rosa, Juan Sánchez, Eric Swanson, and Dan Thornton for helpful comments and Sean Grover for research assistance. © 2014, The Federal Reserve Bank of St. Louis. The views expressed in this article are those of the author(s) and do not necessarily reflect the views of the Federal Reserve System, the Board of Governors, or the regional Federal Reserve Banks. Articles may be reprinted, reproduced, published, distributed, displayed, and transmitted in their entirety if copyright notice, author name(s), and full citation are included. Abstracts, synopses, and other derivative works may be made only with prior written permission of the Federal Reserve Bank of St. Louis. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 73 Fawley and Neely analysis requires controversial identification assumptions to identify simultaneous causality because time aggregation of data to lower frequencies—such as the monthly or quarterly data used in VAR analysis—generally produces simultaneous causality in economic data even if there is unidirectional causality at very high frequencies. In other words, although it is unlikely that macro variables or asset price changes within the meeting day influence Federal Open Market Committee (FOMC) policy decisions, asset price changes in the weeks before such decisions very likely have influenced such decisions. The combination of simultaneity and the omission of many variables that affect asset prices inherently leave a great deal of uncertainty about the effect of monetary policy on monthly prices, output, and employment. It is far easier to identify the effect of high-frequency (daily, hourly) monetary shocks on asset prices. If the monetary policy instrument and market expectations for its value are known, then it is possible to characterize the impact of monetary policy shocks—deviations from expectations—on asset prices, which react quickly to news and transmit monetary policy to the economy. Because financial markets are forward looking, one would normally expect asset prices to react only to the unexpected portion of monetary policy changes, as the expected portion would already be priced into assets. Such high-frequency studies of the effect of monetary shocks on asset prices interest both market participants and economists and constitute a useful first step to answering larger questions about the effects of monetary policy on macro variables. Why study the effect of monetary policy shocks when systematic monetary policy presumably has greater total effects? Both systematic and unsystematic policy actions might be expected to affect asset prices. However, the effects of the systematic policy arise as new information (e.g., data releases, policy statements) becomes available and reshapes market expectations about the economy and the likely policy reaction. These expectations about economic conditions and the central bank’s reaction function form over time and are influenced by both monetary and nonmonetary events. Therefore, it is very difficult to estimate the effects of these unobservable changes in expectations of systematic monetary policy on asset prices, though they will have an effect. In contrast, when a central bank makes a discrete change to policy, the monetary surprise changes expectations immediately—by definition— and it is easy to determine the effects of such surprises on asset prices, which inform us about the transmission of all monetary policy. Such investigations have proliferated since the seminal work of Kuttner (2001), but efforts to draw those lessons together have been only limited and fragmented. This article remedies that deficiency by reviewing the literature that has sought to both characterize the response of asset prices to high-frequency monetary policy shocks and—more ambitiously—to explain those reactions.2 A central theme in this literature is that financial markets’ expectations of monetary policy have become more accurate as Federal Reserve policy has become more transparent in its objectives and procedures. In the 1970s, the FOMC allowed inflation to rise to intolerable levels, which was symptomatic of the lack of clarity in the Fed’s ultimate objectives. The fact that Cook and Hahn (1989) were able to link federal funds target changes—not just surprises— in the 1970s to asset price changes suggests that the Fed’s lack of clarity produced such poor 74 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Fawley and Neely expectations of funds target changes that such changes were largely unexpected, nearly equivalent to surprises. In 1979, however, the Federal Reserve’s new Chairman, Paul Volcker, clarified that one of the Fed’s long-run objectives was price stability, as he sought to dramatically reduce the inflation rate. By the late 1980s, the Fed had reestablished its commitment to price stability and was functionally again using the federal funds rate to achieve its objectives.3 As markets came to understand Fed objectives and its likely reactions, only the unexpected portion of federal funds target changes (the surprise) affected asset prices by the late 1980s (Kuttner, 2001). Researchers exploited a liquid federal funds futures market to accurately estimate market expectations and reduce the amount of measurement error in expectations (and, thus, surprises). The Fed continued to become more transparent in its procedures in 1994-95 by greatly reducing unscheduled changes in the funds target, explicitly announcing the funds target at the conclusion of FOMC meetings, and describing the FOMC’s view of economic conditions and monetary policy after each policy meeting. Each of these actions improved the market’s ability to anticipate or react to Fed actions. Reducing intermeeting target changes made the timing of those changes easier to estimate and reduced the problem of omitted variables.4 Announcing the new target allowed market prices to reach the new target quickly. FOMC meeting statements allowed the FOMC to influence long rates by communicating its view of the economy and policy and presumably improved market forecasts of future policy (Poole, Rasche, and Thornton, 2002; Swanson, 2006). The influence of FOMC statements on the yield curve prompted Gürkaynak, Sack, and Swanson (2005) to develop a two-factor model of monetary policy shocks that has influenced further research. In brief, researchers have found that federal funds rate surprises have consistent and sizable effects on other asset prices, including long-term interest rates, foreign exchange, and equities. For example, Kuttner (2001) estimates that a 100-basis-point surprise increase in the federal funds rate would raise 10-year interest rates by about 30 basis points. Andersen et al. (2003) calculate that the same increase would raise the value of the dollar by 66 to 107 basis points, and Bernanke and Kuttner (2005) find that such a surprise would reduce equity prices by about 4 percent. The rest of this article reviews the challenges involved in defining monetary policy shocks and accurately estimating their impact on asset prices. RESEARCH ON MONETARY POLICY SHOCKS AND ASSET PRICES The efficient markets hypothesis implies that, because financial markets are forward looking, only the unexpected portion of a monetary policy change should influence asset prices and it should do so very quickly (Fama, 1970). Therefore, any study of monetary policy must decompose actions into expected and unexpected portions; that decomposition depends on market perceptions of Federal Reserve objectives, procedures, and communications. A central theme of this article is that research on monetary policy shocks has evolved jointly with those FOMC objectives, procedures, and communications. As the Federal Reserve has become Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 75 Fawley and Neely more transparent in its objectives and communications, it has become better able to influence asset markets without disrupting them (Bernanke and Kuttner, 2005; Gürkaynak, Sack, and Swanson, 2005). Early Research on Effects of Monetary Shocks Any study of the effects of monetary policy must first define the monetary policy instrument, the primary tool that a central bank uses to influence the economy. Monetary instruments are closely related to, but distinct from, the stance of monetary policy—the contribution that monetary policy makes to economic and financial conditions—because any change in the instrument is likely to imply a change in the stance of monetary policy in the short run, though the stance can certainly change without a change in the instrument in the longer run. That is, researchers are interested in the monetary policy instrument because unexpected changes in the instrument equate to unexpected changes in the stance of monetary policy, though the stance of policy can change without a change in the instrument. It is natural to think of monetary policy as operating through changes in measures of money and monetarists, such as Friedman, commonly characterized the stance of monetary policy with reference to the growth rate of money.5 Indeed, the Fed described its own policies in the 1970s and 1980s in terms of targeting various measures of reserves or monetary aggregates. Thus, empirical studies of monetary policy during the 1970s and 1980s typically associated changes in monetary policy with changes in measures of the money supply. Sims (1980), for example, describes the monetarist view that “the time path of the money stock is a good single index of monetary policy.” He questions the efficacy of monetary policy on the grounds that money adds no additional explanatory power to a system that includes output, prices, and a short interest rate. The immediate predecessor of the literature examining high-frequency monetary policy shocks was the literature that searched for a liquidity effect: an injection of reserves that decreases the nominal interest rate.6 A large literature searched for this liquidity effect in various ways: estimating investment-savings, liquidity-preference-money-supply (IS-LM) models; regressing interest rates on functions of money growth; and studying asset price behavior around monetary stock announcements. The results were mixed, at best, with many studies finding negative or perverse results with aggregate data. Reichenstein (1987) surveys this literature and concludes that “the Fed appears to have little control over month-to-month changes in [short-term] interest rates.” Despite the lack of clear success, researchers used VARs with aggregate data to continue to look for a liquidity effect well into the 1990s (Thornton, 2001b). The increasing adoption and recognition of overt interest rate targeting by central banks ultimately brought an end to this line of research, however (Friedman and Kuttner, 2011). Measuring Monetary Policy with the Federal Funds Rate Although the Federal Reserve described its policy objectives in the 1970s and much of the 1980s in terms of targets for some measure of money or reserves, most central banks normally 76 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Fawley and Neely conduct monetary policy by trading short-term securities or managing short-term loans to banks to target a short-term interest rate. Observing that central banks usually conduct monetary policy through a short-term interest rate and that monetary aggregates do not predict output or interest rates very well, McCallum (1983) reasoned that if the Fed’s behavior is determined by an interest rate rule, “policy innovations—i.e., the unsystematic portion of the Fed’s behavior—are then represented as the stochastic disturbances appearing in this interest rate rule, not in some rule determining the value of the money stock.” Building on such observations, researchers began to equate the federal funds rate with the U.S. monetary policy instrument by the late 1980s. Cook and Hahn (1989) argue that, because the Fed uses the federal funds rate as its instrument and does not quickly reverse target changes, the funds rate should influence other interest rates.7 These authors regress changes in bill or bond rates on 75 federal funds target changes from September 1974 through September 1979, a period in which the Fed tightly controlled the funds rate: (1) ∆Rt = β1 + β 2 ∆RFFt + ut , where DRt is the change in a bond or bill rate and DRFFt is the federal funds target change. Contrary to Reichenstein’s (1987) conclusions surveying studies of the impact of money on interest rates, Cook and Hahn’s results imply that a 1-percentage-point increase in the federal funds rate was associated with a 55-basis-point increase in 3-month Treasury bills and a 13basis-point increase in 10-year bonds.8 Table 1 summarizes the empirical results from important papers in this literature while Table 2 describes their contributions. Bernanke and Blinder (1992) implicitly support this use of the funds rate, arguing that the federal funds rate is “a good indicator of monetary policy” because it forecasts macro variables better than any other variables—suggesting that it might affect these macro variables—and responds systematically to unemployment and inflation, which reflect the Fed’s dual mandate. Furthermore, they argue that because the funds rate is insensitive to changes in demand for reserves, it “is mostly driven by policy decisions.” Some would criticize the use of regressions such as equation (1) to determine the effect of changes in the federal funds target on the grounds that the effect of the announcement change on asset prices is measured only over one day—or a few days—and might be temporary. Such criticisms are misplaced. Because uncertainty about asset prices usually rises with the forecast horizon, no one can know the long-term effects of any event on asset prices. The efficient markets hypothesis implies that the market’s best guess must have been that the effects of the federal funds target change would persist. Otherwise, expectations of a temporary impact of a policy announcement would create a risk-arbitrage opportunity for investors to bet on the reversal of the policy’s effects. The Importance of Expectations Studies on later samples failed to confirm the Cook and Hahn (1989) results, however. In particular, Radecki and Reinhart (1994) and Roley and Sellon (1995) failed to find any significant relationship between federal funds target changes and interest rates over later samples.9 Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 77 First Quarter 2014 Estimated Impact of Monetary Policy Surprises on Interest Rates Kuttner (2001) Table number in original article Sample Poole, Poole, Poole and Rasche, and Rasche, and Rasche (2000) Thornton (2002) Thornton (2002) Rigobon and Sack (2004) Rigobon and Sack (2004) Gürkaynak, Fleming and Sack, and Piazzesi (2005) Swanson (2005) Hamilton (2008) Table 3 Table 3 Table 7 Table 3 Table 4 Table 4 Table 1 Table 1 Table 2 1989:06– 2000:02 1988:10– 2000:02 1987:08– 1993:12 1994:03– 2001:05 1994:01– 2001:11 1994:01– 2001:11 1994:01– 2004:12 1991:07– 2004:12 1998:10– 2006:12 4,552 Observations 42 53 38 62 73 73 93 120 Daily Daily Daily Daily Daily Daily 1-hour 30-minute Daily Expectations measure Currentmonth FF Nextmonth FF Nextmonth FF Nextmonth FF Currentmonth FF Currentmonth FF Currentmonth FF Expected and surprise? With anticipated With anticipated With anticipated With anticipated No correction No correction Data frequency Correction? Errors-in-variables Errors-in-variables correction correction Current 3-month Current 3-month eurodollar eurodollar future future Surprise only Surprise only Surprise only Surprise only Surprise only ID through ID through Intraday data Intraday data No correction 66.6 (4.4) 53.7 (4.0) 65.8 (2.2) 60.9 (7.4) 52.2 (5.7) 70.6 (2.1) heteroskedasticity heteroskedasticity (IV) (GMM) Maturity of regressand Federal Reserve Bank of St. Louis REVIEW 3-month 79.1 (9.4) 6-month 71.6 (8.4) 12-month 71.6 (9.2) 2-year 61.4 (10.2) 73 (9.0) 78 (9.0) 82.3 (10.0) 80.8 (28.0) 89.9 (11.0) 63.5 (20.0) 87.6 (11.5) 47.1 (13.0) 91.8 (10.0) 54.6 (19.0) 75.6 (9.3) 27.6 (12.7) 76.1 (14.0) 36.4 (20.0) 79 (11.2) 15.5 (11.6) 74.8 (2.3) 48.3 (13.0) 45.5 (8.7) 3-year 68.5 (2.9) 64.1 (3.0) 5-year 48.1 (11.2) 10-year 31.5 (10.2) 30-year 19.4 (8.4) 48 (10.0) 56.9 (14.0) 18.2 (23.0) 93 (12.6) 12.5 (13.9) 31.3 (14.8) 26.4 (8.1) 42.6 (11.0) 2.7 (22.0) 61.1 (13.7) 0.8 (10.2) 8.3 (15.6) 12.5 (5.8) 27.7 (10.0) –7.5 (13.0) 35.2 (13.6) –13.3 (8.3) 42.6 (2.8) NOTE: The lower half of the table displays regression coefficients (yield changes in basis points) and standard errors (in parentheses) from a number of articles on the impact of monetary policy (federal funds rate target) surprises on changes in interest rates of varying maturities. FF, federal funds target rate; GMM, generalized method of moments; ID, identification; IV, instrumental variables. Fawley and Neely 78 Table 1 Federal Reserve Bank of St. Louis REVIEW Table 2 Summary of the Most Important Papers in the Literature on Monetary Policy Surprises Article Method/Innovation Results Regression on federal funds target changes “We find that changes in the target caused large movements in short-term rates and smaller but significant movements in intermediate- and long-term rates.” Kuttner (2001) Identification of federal funds target surprises using futures contracts “Interest rates’ response to anticipated target rate changes is small, while their response to unanticipated changes is large and highly significant.” Poole, Rasche, and Thornton (2002) Correction for errors-in-variables bias “The estimates…suggest that Treasury rates respond significantly to unexpected changes in the Fed’s funds rate target…[T]he response of the Treasury rate to unexpected target changes declines as the term lengthens…We find that the response of the 3-month T-bill rate is nearly identical before and after [the 1994] procedural change. The magnitude and significance of the response of longer-term rates, however, declines after this procedural change.” Rigobon and Sack (2004) Correction for simultaneity/omitted variables “The results indicate that an increase in short-term interest rates results in a decline in stock prices and in an upward shift in the yield curve that becomes smaller at longer maturities. The findings also suggest that the event-study estimates contain biases that make the estimated effects on stock prices appear too small and those on Treasury yields too large.” Bernanke and Kuttner (2005) Full sensitivity analysis for equities/Decomposition of equity return “We find that, on average, a hypothetical unanticipated 25-basis-point cut in the Federal funds rate target is associated with about a 1% increase in broad stock indexes. Adapting a methodology due to Campbell and Ammer, we find that the effects of unanticipated monetary policy actions on expected excess returns account for the largest part of the response of stock prices.” Gürkaynak, Sack, and Swanson (2005) Two-factor model “We test whether [monetary policy] effects are adequately captured by a single factor— changes in the federal funds rate target—and find that they are not. Instead, we find that two factors are required. These factors have a structural interpretation as a ‘current federal funds rate target’ factor and a ‘future path of policy’ factor, with the latter closely associated with Federal Open Market Committee statements...According to our estimates, both monetary policy actions and statements have important but differing effects on asset prices, with statements having a much greater impact on longer-term Treasury yields.” Fleming and Piazzesi (2005) Tick-by-tick analysis “Analysis of high-frequency data shows that Treasury note yields are highly volatile around FOMC announcements, even though the average effects of fed funds target rate surprises on such yields are fairly modest.” Hamilton (2008) Daily model with uncertainty over event days “This paper develops a generalization of the formulas proposed by Kuttner (2001) and others for purposes of measuring the effects of a change in the federal funds target on Treasury yields of different maturities…Although the methods are new, the conclusion is quite similar to that reported by earlier researchers—changes in the fed funds target seem to be associated with quite large changes in Treasury yields, even for maturities of up to 10 years.” Hausman and Wongswan (2011) Two-factor model applied to full set of international assets “This paper analyzes the impact of U.S. monetary policy announcement surprises on foreign equity indexes, short- and long-term interest rates, and exchange rates in 49 countries… Global equity indexes respond mainly to the target surprise; exchange rates and long-term interest rates respond mainly to the path surprise; and short-term interest rates respond to both surprises.” 79 Fawley and Neely First Quarter 2014 Cook and Hahn (1989) Fawley and Neely Thornton (1998) criticizes Cook and Hahn’s target change series as being endogenous— that is, partially expected and therefore not true policy surprises. Likewise, Kuttner (2001) argues that the failure to confirm Cook and Hahn’s results on later samples is due to the failure to decompose monetary policy shocks into their expected and surprise components.10 Expected funds rate changes should not influence other asset prices because financial markets are forward looking; only the unexpected component of changes in the federal funds target should change other asset prices.11 In the 1970s, funds rate expectations were apparently so poor that changes in the funds target were good proxies for their unexpected component. By the late 1980s, however, clearer Fed objectives and procedures allowed markets to anticipate a large portion of federal funds target changes, so changes in the funds target became poor proxies for their unexpected component. Therefore, researchers began to consider how to decompose federal funds rate changes into expected and unexpected components to test the effect of the latter with high-frequency data. The most common method to compute expectations of the federal funds target is due to Kuttner (2001), who used prices from the federal funds futures market. The basic idea is that the futures market implies an average federal funds rate for a particular contract month and— because the New York Open Market Desk can keep the average federal funds rate near the target—this implies an average federal funds target for the contract month. Thus, the market’s expectation of the change in the target on the FOMC meeting date can be calculated if the target at the start of the month and the date on which the target might be changed are known. Appendix A on the federal funds futures market details this procedure. Using these procedures to decompose the surprise and expected components of federal funds target changes from June 1989 to February 2000, Kuttner (2001) estimates (2) ∆R tn = α + β1∆ rt e + β 2 ∆ rtu + ε t , where DRtn is the change in the yield of an n-year Treasury bond on date t, and Dr̃ te and Dr̃ tu are the expected and surprise components of the federal funds target change on day t. Using 42 days of changes in the FOMC target, Kuttner finds that an unexpected 1-percentage-point increase in the federal funds target raises 3-month Treasury yields by 79 basis points and 10year yields by 32 basis points. The reaction to the surprise component is significant at all maturities analyzed, while the impact of expected changes is always small and insignificant. Table 3 illustrates Kuttner’s results on the set of all FOMC events (all FOMC meetings plus intermeeting rate changes) from October 1988 to June 2007. Some anticipated target changes are statistically significant at the very short end of the yield curve, presumably because the central bank can control the very short end of the yield curve through the open market operations that pin down the federal funds rate. The unanticipated component, however, clearly has the much stronger and more statistically significant impact on interest rates at all horizons. Kuttner (2001) claims that federal funds futures offer three advantages over other procedures to identify expectations of monetary policy: (i) Futures require no model; (ii) futures data are not revised and so there is no “data vintage” problem; and (iii) futures do not entail an errors-in-variables problem as do VARs.12 In addition, Gürkaynak, Sack, and Swanson 80 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Fawley and Neely Table 3 Response of Interest Rates to Expected and Unexpected Components of Federal Funds Target Surprises (1988-2007) Maturity Intercept Anticipated Unanticipated R2 3-month –0.01 (–3.7) 0.05 (2.2) 0.66 (15.2) 0.60 6-month –0.02 (–3.8) 0.05 (2.3) 0.64 (14.5) 0.58 12-month –0.01 (–1.9) 0.03 (1.5) 0.62 (13.3) 0.53 2-year –0.01 (–1.1) 0.05 (1.7) 0.46 (8.1) 0.30 5-year 0.00 (–0.8) 0.02 (0.7) 0.32 (5.2) 0.15 10-year 0.00 (–0.5) –0.01 (–0.2) 0.17 (3.2) 0.06 30-year –0.01 (–1.4) –0.01 (–0.6) 0.08 (1.9) 0.02 NOTE: The table shows the results of a regression similar to that of Kuttner (2001, equation (2)) that uses all FOMC events from October 1988 to June 2007. The event set includes 177 events, including all regularly scheduled FOMC meetings, plus intermeeting rate changes. Shaded numbers denote significantly positive (blue) or negative (red) t-statistics. (2007) show that federal funds futures are the best-available forecasts of federal funds targets at horizons of up to 6 months.13 These advantages have made federal funds futures the standard metric for computing expectations of the federal funds target. The most attractive alternative to using federal funds futures data to calculate expectations for the federal funds target is to use some sort of survey data. Andersen et al. (2003), for example, use Money Market Services (MMS) survey data in their broad study of the effects of macro announcements on foreign exchange returns and volatility. Using data from January 1992 to December 1998, they find that positive funds target surprises significantly appreciate the dollar for four of the five exchange rates. The use of expectations from the federal funds futures market, rather than from MMS survey data, implies similar effects of funds target surprises on the foreign exchange market. Fatum and Scholnick (2006) determine that changes in 2-month-ahead federal funds futures on days without monetary policy news, which the authors interpret as policy expectations, are significant predictors of three exchange rates at the daily frequency and the response is rapid, within the day. Faust et al. (2007) find somewhat larger results with FOMC meetingday data from March 1995 through December 2002 and expectations from the federal funds futures market. Their regressions using 20-minute windows imply that a 100-basis-point surprise increase in the funds target would depreciate the dollar 123 basis points against the deutsche mark/euro (DEM/EUR) and 66 basis points against the British pound (GBP). The impacts are highly statistically significant. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 81 Fawley and Neely Noisy Measures of Monetary Policy Computing shocks to the federal funds target entails confronting the fact that expectations of the federal funds target—and therefore the surprises in the federal funds target—are measured with some error because of bid-ask spreads, stale expectations, and risk premia.14 When regressors are measured with error, the coefficients on those regressors are biased and the expected values are attenuated toward zero. Poole, Rasche, and Thornton (2002) correct for such errors-in-variables bias by estimating the size of the measurement error from policy actions that were correctly anticipated. The authors gauge the degree of correct anticipation from surveys and commentary in the “Credit Markets” section of the Wall Street Journal published on the days before FOMC meetings. Using this correction and with data from March 1994 through May 2001, the authors report that a 1-percentage-point reduction in the federal funds target reduces 3-month Treasury yields by 77 to 82 basis points and 10-year Treasury yields from 40 to 43 basis points. The corrected estimates are somewhat larger than (but fairly close to) the uncorrected estimates, indicating that measurement error is not a major problem.15 Hamilton (2008) develops a method to measure monetary policy surprises to account for both noise induced by deviations in the effective federal funds rate from its target and uncertainty about the date of policy actions (as may have been important prior to 1994).16 The method extracts the monetary policy signal from daily federal funds futures changes under the assumption that the econometrician does not know the dates of policy actions and must take them to be equally likely on all dates in the sample. Using this procedure, Hamilton (2008) estimates his regression over all business days, not just those with FOMC meetings or known target changes, to find that a 1-percentage-point surprise to the federal funds target increases 3-month Treasury yields by 66 basis points and 10-year yields by 43 basis points. Simultaneity and Omitted Variables in Asset Price Reactions to Monetary Policy If nonmonetary news influences monetary policy and asset prices within the event window, or if monetary policy responds contemporaneously to asset price changes within the event window, then that complicates the study of monetary policy’s effect on asset prices. The first problem is omitted variables bias, while the second problem is simultaneity bias.17 Appendix B describes these problems in some detail. In either case, a naive event study estimate of the impact of target rate changes on asset prices will inconsistently estimate the true impact of those target rate changes. To understand these biases, consider a linear system in which monetary policy (Dmt) and asset prices (Dpt) are determined simultaneously as follows: (3) ∆pt = a0 + a1∆mt + a2newst + e p ,t (4) ∆mt = b0 + b1∆pt + b2newst + em ,t , where newst denotes macro releases that potentially affect monetary policy and asset prices. 82 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Fawley and Neely Omitted variables bias occurs when nonmonetary news within the event window moves financial prices but the nonmonetary regressor is incorrectly excluded. That is, a2 ≠ 0 but the estimated regression imposes a2 = 0. In that case, the event study estimate of the impact of target rate changes will be biased. If monetary policy reacts to asset price changes within the observation interval for the data—for example, within the day for daily data—then b1 will be nonzero in equation (4) and an ordinary least squares (OLS) estimate of a1—the effect of the monetary policy change on the asset price—will be inconsistent. As the observation interval around a policy change becomes arbitrarily small, the parameter b1 will tend to go to zero—equation (4) will contain only lagged values of Dpt , which present no simultaneity problem—and the OLS estimator of a1 becomes consistent. Although the FOMC rarely directly reacts to asset price changes within a day—that is, it seems likely that b1 = 0 but there are lags of Dpt in equation (4)—it is more plausible that omitted variables bias presents a problem. Specifically, prior to February 1994, Chairman Alan Greenspan changed the federal funds target on several days of weak employment reports, presumably in response to those releases (i.e., a2 ≠ 0 in equation (3)).18 A negative employment report will tend to directly reduce equity prices and interest rates, but it will also tend to make the FOMC reduce interest rates, which will tend to increase equity prices but reduce longerterm interest rates. If the whole effect were naively ascribed to the policy shock—ignoring possible joint-response bias—then equity responses would be underestimated and yield responses to policy shocks would be overestimated. Bernanke and Kuttner (2005) report that failing to account for the omitted variables bias related to target changes that occur on the same day as employment reports attenuates the average stock market response in their sample by 1 percentage point. To address these occasions when policy surprises react to asset price changes or (much more frequently) react to other news, researchers have sought to identify the effects of monetary policy shocks with (i) explicit identification schemes with daily data or (ii) higher-frequency data and narrow event windows.19 Rigobon and Sack (2004) pursue the first strategy: These authors identify monetary shocks from FOMC meetings and the semiannual monetary policy testimony from the daily conditional heteroskedasticity of nearby futures on 3-month eurodollar interest rates over a sample from 1994 through November 2001. The nearby futures on 3-month interest rate contracts are used to reduce problems with timing issues (as discussed more extensively in the following subsections). These authors exploit the fact that policy shocks display greater variance on announcement days than nonannouncement days to identify the effect of monetary policy shocks, solving the usual problems of simultaneity/omitted variables in daily interest rate data in a way that does not require the strong assumptions of daily event studies.20 The two heteroskedasticity estimators employed by Rigobon and Sack (2004) diverge widely in the degree to which they indicate that event studies overstate yield responses. Rigobon and Sack (2004) find that a 1-percentage-point surprise federal funds target cut (i) increases broad stock indexes by 4.85 to 10.06 percentage points and (ii) lowers 6-month Treasury yields by 47 to 88 basis points and 10-year Treasury yields by 1 to 61 basis points. The authors attribFederal Reserve Bank of St. Louis REVIEW First Quarter 2014 83 Fawley and Neely ute the large effects on stock prices to the lack of downward bias in their estimates compared with event study estimates, but they also report that using federal funds futures instead of futures on 3-month eurodollars produces smaller estimates of the impact of monetary shocks. Rosa (2011b) reexamines the biases that Rigobon and Sack’s (2004) identification-throughheteroskedasticity methods are designed to confront. Rosa (2011b) finds a small but statistically significant bias in event study estimates of asset market reactions. Nevertheless, Rosa recommends the event study estimator because its bias is small and it outperforms the heteroskedasticity-based estimator. The second strategy to avoid omitted variables and/or simultaneity is to use highfrequency (intraday) data to estimate the effect of monetary policy on asset prices (a1). At a sufficiently high frequency, asset prices presumably have no effect on monetary policy (b1 = 0) and even news variables are predetermined, so the relation can be simply estimated by OLS. Gürkaynak, Sack, and Swanson (2005) and Fleming and Piazzesi (2005) use intraday data from 1994 through 2004 and 30-minute event windows to estimate the effects of policy surprises on yield changes. The Fleming and Piazzesi (2005) results generally confirm those from daily event studies: A 1-percentage-point cut in the federal funds target lowers 3-month Treasury yields by 67 basis points and 10-year Treasury yields by 8.3 basis points (although the latter response is insignificant).21 Federal Funds Surprises and Equity Prices A large literature debates the extent to which monetary policy affects the economy through interest rate channels and/or credit channels. Monetary policy can affect consumption, investment, and the international competitiveness of domestic goods by influencing the foreign exchange value of the domestic currency and stock prices. Credit channels exploit the fact that easier monetary policy can improve consumers’ and firms’ balance sheets and thus reduce the effect of certain financial frictions—adverse selection and moral hazard—that hinder borrowing (Mishkin, 1995).22 Thus, monetary policy is often thought to generate a substantial portion of its effects through equity markets. Therefore, economists study how much and why monetary surprises affect stock prices and what these answers imply for monetary policy channels. To answer these questions, Bernanke and Kuttner (2005) studied monetary policy effects on equity prices using data from May 1989 through December 2002. This study carefully considered factors that could affect the response, such as distinguishing scheduled from unscheduled target changes, simultaneity between news and target changes, and the timing of target changes—an advancement or postponement of an expected action—or whether they signal persistent changes in interest rates. Positive (negative) federal funds target shocks significantly reduced (raised) equity prices; high-tech sectors reacted more strongly than did broad indexes. The authors found no evidence for asymmetry in the magnitude of positive/negative target shocks, but they did find that reversals—for example, a target increase after a series of decreases— had particularly strong effects on equities, although they cautioned there are only five such observations in the sample. The authors also showed that six “outliers”—four of which are intermeeting moves—strongly influenced the estimated impact coefficients. The authors speculated 84 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Fawley and Neely that these very strong effects were due to the fact that intermeeting moves convey much more urgency than equally sized surprises at regularly scheduled FOMC meetings. Perhaps the main contribution of Bernanke and Kuttner (2005), however, is their use of federal funds surprises, measured with futures data, to study the source of the equity returns with a Campbell and Shiller (1988) decomposition of excess equity returns into expectations of future dividends, interest rates, and excess returns.23 This methodology implies that positive federal funds surprises reduce expected excess returns or dividends—depending on the sample—but not real interest rates. The authors interpret an effect on expected excess returns as arising from an increase in the riskiness of stocks or the willingness of investors to bear stock risk. Alternatively, the change in expected excess returns might stem from overreaction or excess sensitivity to policy. The aggregated equity returns used by Bernanke and Kuttner (2005) cannot shed much light on differential responses by individual stocks and industry portfolios to monetary policy surprises. To investigate such heterogeneity, Ehrmann and Fratzscher (2004) regress daily returns for individual U.S. firms and industry portfolios on monetary policy surprises from 79 FOMC meetings from 1994 through January 2003. The expectations are computed using Reuters’ surveys of market participants. The results have mixed implications for the credit and interest rate channels, however. The fact that stock prices of financially constrained firms react strongly to monetary policy tends to support the credit channel, while the strong reaction of firms in cyclical and capital-intensive industries tends to support the interest rate channel. Basistha and Kurov (2008) investigate the reactions of individual stocks to find that all stocks tend to react particularly strongly to monetary policy in recessions and tight credit conditions. Stock prices of financially constrained firms display particularly strong asymmetry, however, which the authors interpret as further evidence for the credit channel. Two papers extend this research to investigate the impact of U.S. funds target surprises on international equity prices. Ehrmann and Fratzscher (2009) consider how national characteristics—openness, exchange rate regime, and so on—determine the strength of the transmission process. On average, a 100-basis-point surprise U.S. tightening reduces equity prices by 2.7 percent, although there is a great deal of heterogeneity in responses to U.S. monetary policy surprises, both across countries and across sectors. Financially open and more integrated countries have stronger equity market reactions. The reaction of U.S. short-term interest rates to the target surprise appears to govern the strength of the overall transmission to foreign equities, and countries with strong equity market reactions to U.S. target surprises also tend to exhibit strong exchange rate and interest rate reactions. Ammer, Vega, and Wongswan (2010) examine the impact of U.S. target surprises on firmlevel equities from the United States and 21 foreign countries.24 Consistent with Bernanke and Kuttner’s (2005) results for U.S. equities, they find that an unexpected tightening of 100 basis points reduces U.S. and foreign equity prices by 6.4 percent and 6.8 percent, respectively. The authors interpret the sensitivity of cyclical industries as supporting the demand (interest rate) channel of monetary policy. In contrast to the Ehrmann and Fratzscher (2009) results, Ammer, Vega, and Wongswan (2010) find that countries with fixed exchange rates respond more strongly to U.S. target surprises than do countries with flexible exchange rates. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 85 Fawley and Neely Increasing Predictability in Fed Policy In 1994, the FOMC greatly reduced the frequency of intermeeting funds target changes and began to issue statements after policy meetings. Such statements could potentially contain three distinct pieces of information: (i) the current policy action in terms of the federal funds target or asset purchases; (ii) a statement on current or future economic conditions; and (iii) a forecast for the path of policy.25 Table 4, excerpted from Middeldorp (2011), describes changes in FOMC communication policy from 1993 to 2007. The information in the announcements was initially very simple—a qualitative description of the change in the funds target: Chairman Alan Greenspan announced today that the Federal Open Market Committee decided to increase slightly the degree of pressure on reserve positions. The action is expected to be associated with a small increase in short-term money market interest rates. (FOMC, 1994a) Such postmeeting statements later became more elaborate, with numerical changes to the target and brief discussions of economic conditions, as in this excerpt from the August 1994 FOMC press release (FOMC, 1994b): The Federal Reserve announced today the following monetary policy actions: • The Board of Governors approved an increase in the discount rate from 3½ percent to 4 percent, effective immediately. • The Federal Open Market Committee agreed that this increase would be allowed to show through completely into interest rates in reserve markets. These measures were taken against the background of evidence of continuing strength in the economic expansion and high levels of resource utilization. The actions are intended to keep inflationary pressures contained, and thereby foster sustainable economic growth. The practices of reducing the frequency of intermeeting target changes and issuing statements allowed markets to better anticipate the timing and direction of policy target changes (see the boxed insert). In the pre-February 1994 sample, Poole, Rasche, and Thornton (2002) “find few instances where there was a widespread expectation that the Fed would take an action on a particular day.” In contrast, Poole, Rasche, and Thornton (2002) identify only 18 “surprise” events in the set of 62 policy events from March 1994 to May 2001, using articles from the Wall Street Journal to assess expectations. Swanson (2006) demonstrates that the private sector has become better at forecasting interest rates and less surprised by Fed actions since the late 1980s. He attributes this improvement to increased Fed transparency, showing as a control that GDP and inflation forecasts did not improve over the same period. These factors reduced the measured effects of policy surprises on longer-term interest rates (Poole and Rasche, 2000; Poole, Rasche, and Thornton, 2002). For example, compare the decline in coefficients in the “Target” columns from the first to the second panel in Table 5.26 Multidimensionality: Two Monetary Policy Factors The increasing predictability of Fed policy, as documented by Swanson (2006), might well be partly due to the FOMC’s strategy of releasing statements regarding the economic outlook 86 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Fawley and Neely Table 4 Changes in FOMC Communication Policy (1993-2007) Date Label Nature of change March 1993 Minutes Merging of FOMC “Minutes of Actions” and “Policy Record” into one new document labeled the “Minutes of the FOMC” February 1994 Statement First postmeeting statement with qualitative description of change in policy August 1994 Rationale Some reasoning behind the decision is added to the statement July 1995 Federal funds target Inclusion of actual numerical federal funds target in statement May 1999 Bias Inclusion of FOMC’s asymmetric policy directive in statement January 2000 Balance of risks Revised statement language discussing balance of risks toward growth or inflation rather than bias for federal funds target March 2002 FOMC vote Inclusion of vote with name(s) of dissenters in statement August 2003 Guidance Statement language explicitly indicating the likely direction of rates over extended period January 2005 Earlier minutes Minutes released three weeks after meeting November 2007 Enhanced projections More detailed, frequent, and extended projections NOTE: This table is excerpted from Middeldorp (2011). and likely future policy after its meetings. This suggests that the FOMC might be able to influence the whole yield curve—not just the very short end—by communicating its intentions to markets. The reason for this is that long-term interest rates depend, in part, on the expected path of future short-term interest rates, which the FOMC can influence. Fleming and Piazzesi (2005) and Gürkaynak, Sack, and Swanson (2005) identify specific cases where the response of long rates appears to be unrelated to the measured surprise to the funds target. Specifically, Gürkaynak, Sack, and Swanson (2005) identify large responses in 2- to 5-year Treasury yields after the January 28, 2004, policy meeting, despite the fact that there was almost no surprise in the federal funds target. The authors attribute these unusual responses to changes in expectations shaped by the FOMC’s meeting statements rather than the policy action itself. The independence of long rate changes and federal funds surprises motivates Gürkaynak, Sack, and Swanson (2005) to develop a two-factor model of monetary policy shocks (see Appendix C).27 A single factor fails to adequately describe yield curve reactions (up to one year ahead) to monetary policy shocks, but statistical tests cannot reject the same null of two factors for yields up to a year. The first factor (target or timing) is closely related to current-month federal funds futures surprises, while the second factor (level or path) correlates strongly with 1-year eurodollar futures (i.e., 1-year-ahead policy expectations). The second factor significantly increases the power of monetary policy actions to explain medium- to long-term interest rate changes, with the largest improvements at the longest maturities: The 10-year yields respond almost three times more to the path factor than the target factor, and the R-squared rises from 8 percent to 74 percent (Gürkaynak, Sack, and Swanson, 2005). The path factor is less important for equity returns, however: The coefficient Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 87 Fawley and Neely The Evolution of Fed Policy One needs to define a monetary policy instrument to study the effects of monetary policy surprises on asset prices. The nature of Federal Reserve monetary policy has changed over time, however, from a hybrid of interest rate targeting to hit ranges for monetary aggregates, to targeting for various categories of bank reserves, and back to explicit interest rate targeting. This evolution of Fed policy can be interpreted as a series of moves toward greater transparency in objectives and procedures that have strongly influenced the methods and assumptions of monetary policy researchers, as well as the impact of monetary policy on asset prices. From 1970 to October 1979, the Federal Reserve targeted growth rates for monetary aggregates, primarily M1 but also M2 (Meulendyke, 1998). The Full Employment and Balanced Growth Act of 1978 (the Humphrey-Hawkins Act) enshrined monetary targeting into law by requiring the Federal Reserve to set targets for calendar years and to explain any deviations. To realize these money supply growth targets, the FOMC chose a target for the federal funds rate and instructed the New York Fed Trading Desk to trade appropriately to correct deviations of the funds rate from this target. Gradually, the Desk began to control the federal funds rate fairly closely (Meulendyke, 1998). Unfortunately, Federal Reserve monetary policy in the 1970s failed to control inflation.1 Annual consumer price index inflation reached 12.2 percent per year in September 1979, eventually peaking at 14.8 percent per year in March 1980. Many analysts blamed federal funds rate targeting for producing too much inertia in policy (Meulendyke, 1998). As a result, in October 1979, Chairman Volcker announced that the FOMC would target nonborrowed reserves, rather than the funds rate, to achieve desired M1 growth. In 1983, the lack of a stable relation between M1 growth and economic activity led the FOMC to change procedures again, this time to targeting borrowed reserves (total reserves less nonborrowed reserves).2 In practice, analysts interpreted this procedure as a move back toward “soft” interest rate targeting (Thornton, 1988). This procedure, however, depended on a stable function for bank borrowing from the Fed. Banking problems—particularly those of Continental Illinois in 1984—soon led banks to become very wary of borrowing from the Federal Reserve, lest investors, depositors, and/or regulators conclude they were in financial trouble (Meulendyke, 1998). As with M1 targeting, the policy of targeting borrowed reserves failed because the borrowing function was unstable; banks became less willing to borrow from the Fed and borrowed reserves were not closely related to prices or economic activity (Thornton, 1988). 1 Researchers broadly agree that changing ideas on the objectives and scope of monetary policy were critical to the development of the Great Inflation and the subsequent Great Disinflation, but they disagree about why the Fed failed to act. DeLong (1997) believes that the Great Depression left the Federal Reserve with no mandate to control inflation at the expense of unemployment. In contrast, Romer and Romer (2002) implicitly argue that the Fed used a fairly sophisticated but deeply flawed model that claimed to offer an exploitable inflation-unemployment trade-off. Nelson (2005a,b) and Nelson and Nikolov (2004) argue that “monetary neglect”—emphasis on nonmonetary factors in inflation—largely explains the Great Inflation not only in the United States but also in Canada, Australia, New Zealand, and the United Kingdom. 2 Banks can borrow reserves directly from the Fed through the discount window to meet reserve requirements, avoid overnight overdrafts, or meet seasonal funding needs. The demand for borrowed reserves theoretically reflects the tightness of credit conditions in the nonborrowed reserves market. 88 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Fawley and Neely The Evolution of Fed Policy (cont’d) By the late 1980s, the FOMC was effectively targeting the federal funds rate, although it did not announce changes in the federal funds target immediately after FOMC meetings. Instead, markets had to infer new targets from the Desk’s trading patterns, which could take a day or two. In addition, during the late 1980s and early 1990s, the FOMC’s standard practice was to change the funds target between scheduled meetings, often in response to incoming economic news. In 1994, the FOMC changed its procedures in three ways. First, the FOMC began publicly announcing changes to the target immediately after the meeting or decision. Second, the FOMC almost eliminated intermeeting target changes, which had been common in prior years.3 From October 1988 to December 1993, the FOMC changed the federal funds target 32 times, but only 9 times at a scheduled meeting. From 1994 to 2012, the FOMC has changed the target 60 times, 53 of which were at a scheduled meeting.4 Third, the FOMC began to issue statements on the state of the economy and monetary policy immediately after meetings. The statements initially were short and qualitatively described the federal funds target policy but became more detailed in August 1994 and included a numerical description of the funds target by July 1995. FOMC communication policy continued to evolve. In May 1999, the FOMC began to issue a statement of its “bias,” or the direction in which its next move was most likely. But this approach was replaced by a “balance of risks” statement in January 2000 that switched the statement’s emphasis to the likely risks to growth or inflation. From August 2003 to December 2005, the FOMC added implicit forward guidance about the likely future course of the funds target. This practice began again in December 2008 in the wake of the financial crisis. Of course, the financial crisis in the summer and fall of 2008 also motivated the FOMC to institute many temporary special programs to support credit markets and to announce the first of several asset purchase programs in November 2008. 3 The FOMC had been reducing the frequency of intermeeting target changes prior to February 1994. The seven inter- meeting target changes since 1994 occurred on 4/18/1994, 10/15/1998, 1/3/2001, 4/18/2001, 9/17/2001, 1/22/2008, and 10/8/2008. The 10/15/1998 target change came on the heels of the Long-Term Capital Management collapse, which had been affected by the Russian default (8/17/1998); and the 9/17/2001 change followed the 9/11 terrorist attacks. 4 FOMC transcripts reveal that Chairman Greenspan came to consider the practice of making decisions only at scheduled meetings to be a “very useful procedure.” The Chairman would sometimes hint at policy decisions between meetings, however (FOMC, 1998). The FOMC has also changed the federal funds target much less frequently since 1994. The FOMC did not begin making target changes in multiples of 25 basis points until late 1989; previously, changes in multiples smaller than 25 basis points were common (Poole, Rasche, and Thornton, 2002). on the target factor is four times that of the path factor in a regression of the Standard and Poor’s (S&P) 500 index returns on the two factors, and the R-squared improves little. Table 5 shows results of an exercise similar to that in Gürkaynak, Sack, and Swanson (2005), in which one regresses daily interest rate changes on target and path factors, plus a constant (coefficient not shown). The top panel shows the results for all FOMC meetings; the bottom panel shows the results with employment reports removed. The target coefficients in the 1988-93 subsample are uniformly larger than those in the 1994-2007 subsample, particularly at longer horizons, and tend to be more statistically significant despite the shorter sample. At the same time, the coefficients on the path factor become larger and more statistically significant in the second subsample. This pattern presumably occurs because the introducFederal Reserve Bank of St. Louis REVIEW First Quarter 2014 89 Fawley and Neely Table 5 Response of Interest Rates to Target and Path Shocks 1988-93 Target 1994-2007 Path R 2 Target 1988-2007 Path R 2 Target Path R2 All meeting observations 3-month 0.94 (14.9) 0.18 (3.3) 0.79 0.63 (12.3) 0.35 (7.7) 0.66 0.74 (20.1) 0.18 (8.5) 0.73 6-month 0.98 (14.5) 0.17 (2.8) 0.78 0.58 (16.1) 0.48 (15.1) 0.82 0.73 (23.4) 0.24 (13.4) 0.81 12-month 1.07 (22.8) 0.14 (3.4) 0.90 0.47 (14.7) 0.56 (19.9) 0.85 0.69 (27.2) 0.31 (20.8) 0.87 2-year 1.00 (22.4) 0.10 (2.4) 0.89 0.27 (8.3) 0.73 (25.5) 0.87 0.52 (20.8) 0.41 (28.2) 0.88 5-year 0.86 (13.5) 0.04 (0.7) 0.75 0.13 (3.6) 0.73 (22.7) 0.83 0.37 (12.5) 0.43 (25.2) 0.82 10-year 0.64 (9.6) 0.03 (0.5) 0.60 –0.01 –(0.4) 0.58 (17.4) 0.73 0.20 (6.4) 0.35 (19.8) 0.71 30-year 0.50 (7.6) –0.01 –(0.1) 0.48 –0.08 –(2.1) 0.36 (10.3) 0.50 0.10 (3.1) 0.25 (13.6) 0.53 With meetings on days of employment reports omitted 3-month 0.78 (10.2) 0.19 (4.0) 0.69 0.67 (12.0) 0.27 (7.7) 0.65 0.73 (17.0) 0.15 (8.3) 0.68 6-month 0.83 (10.0) 0.17 (3.4) 0.67 0.61 (15.5) 0.36 (14.9) 0.81 0.69 (19.0) 0.20 (13.1) 0.76 12-month 0.91 (16.6) 0.17 (4.9) 0.85 0.49 (14.0) 0.43 (19.7) 0.84 0.62 (21.2) 0.26 (20.3) 0.84 2-year 0.88 (17.4) 0.18 (5.8) 0.86 0.26 (7.5) 0.55 (25.1) 0.86 0.43 (15.0) 0.34 (27.3) 0.85 5-year 0.71 (9.9) 0.17 (3.8) 0.67 0.11 (2.9) 0.55 (22.3) 0.82 0.26 (8.0) 0.35 (24.6) 0.80 10-year 0.51 (7.0) 0.14 (3.1) 0.52 –0.04 –(1.0) 0.43 (17.0) 0.73 0.10 (2.9) 0.28 (19.4) 0.70 30-year 0.38 (5.3) 0.10 (2.2) 0.38 –0.11 –(2.4) 0.27 (10.0) 0.50 0.00 (0.0) 0.20 (13.5) 0.52 NOTE: The table shows results of an exercise similar to that in Gürkaynak, Sack, and Swanson (2005), in which one regresses daily interest rate changes on target and path factors, plus a constant (coefficient not shown). Columns 2 through 4 show results using all FOMC events for the October 1988–December 1993 period; columns 5 through 7 show similar results for the January 1994–June 2007 period; and columns 8 through 10 show results for the whole sample. The top panel shows the results for all FOMC meetings; the bottom panel shows the results with employment reports removed. For the top (bottom) panel, the first subsample includes 65 (57) events and the second subsample 112 (111) events, for a total of 177 (168) events. Shaded numbers denote significantly positive (blue) or negative (red) t-statistics. 90 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Fawley and Neely Figure 1 Relationship Between Interest Rate Futures and Target and Path Surprises (1994-2007) MP1 ED12 50 50 40 40 30 30 20 20 10 10 0 0 –10 –10 –20 –20 –30 –30 –40 –40 –50 –50 –50 –40 –30 –20 –10 0 10 Target Factor 20 30 40 50 –100 –80 –60 –40 –20 0 20 40 60 80 100 Path Factor NOTE: MP1 denotes the Kuttner (2001) federal funds shock measured from near-month federal funds futures contracts and ED12 denotes the change in 12-month-ahead eurodollar futures rates, using all FOMC events from 1994 through 2007. The 45-degree blue lines denote a one-toone relationship; the dashed red lines denote the least squares fit from regressing the y variable on the x variable and a coefficient. The slope is determined by construction (see Appendix C). tion of FOMC meeting statements in 1994 not only improved market confidence in its expectations of the path but also made target surprises less informative. The bottom panel of Table 5 shows the same regression results but with days of employment reports removed. The target coefficients in the top panel are bigger (artificially inflated) because omitting employment reports from the regression tends to increase the estimated impact of monetary policy. In the bottom panel, the days of employment reports are completely omitted from the sample so there is no omitted variables bias.28 Gürkaynak, Sack, and Swanson (2005) emphasize that the path factor is not an independent monetary policy tool, but rather exists because the FOMC can influence medium-term rates by shaping expectations of the target’s path. The ability to influence medium-term rates is important because much of monetary policy’s effect on the economy occurs through mediumterm rates and the incentives they provide for cyclical spending, such as business and residential investment. In addition, the ability to influence longer-term rates would prove particularly valuable near the zero lower nominal bound. Although the FOMC does not directly choose the path surprise in the same way that it chooses the federal funds target, the FOMC can strongly influence the path with its meeting statement (Gürkaynak, Sack, and Swanson, 2005). This raises the question of whether such a Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 91 Fawley and Neely Figure 2 Relationship Between Federal Funds Surprises and Eurodollar Futures 1988-1993 3-Month Eurodollar 20 3-Month Eurodollar 20 10 10 0 0 –10 –10 –20 –20 –30 –30 1994-2007 –40 –40 Slope = 0.83 –50 –50 –40 –30 –20 –10 0 10 20 Slope = 0.65 –50 –50 –40 –30 –20 1988-1993 12-Month Eurodollar –10 0 10 20 Federal Funds Surprise Federal Funds Surprise 1994-2007 12-Month Eurodollar 30 20 20 10 10 0 0 –10 –10 –20 –20 –30 –30 –40 Slope = 0.76 –50 –50 –40 –30 –20 –10 0 10 20 –40 Slope = 0.35 –50 –50 –40 –30 –20 Federal Funds Surprise One-for-One –10 0 10 20 30 Federal Funds Surprise Less than One-for-One More than One-for-One Perverse NOTE: The figure plots changes in 3-month-ahead eurodollar futures (top panel) and 12-month-ahead eurodollar futures (bottom panel) against federal funds rate surprises measured using the methodology of Kuttner (2001). The left (right) column plots all FOMC events for the October 1988–December 1993 period (January 1994–June 2007 period). The solid blue line denotes a one-for-one change in the x and y variables; the dashed red line denotes the OLS fitted line from regression of the y variable on the x variable plus a constant. The slope of this line is identified in the lower-right corner of the figure. statement reflects a commitment by the FOMC to pursue a certain policy that is inconsistent with its expected reaction function or whether it simply reflects the FOMC’s view of the normal policy response to likely economic conditions. Campbell et al. (2012) argue that the FOMC has sufficient experience with communication to influence the economy by committing itself to an unusual policy path. These authors term forward policy guidance that links policy to the forecast of economic activity in the normal way as “Delphic” forward guidance and policy that commits the FOMC to a particular policy as “Odyssean” forward guidance. 92 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Fawley and Neely How should the path factor be interpreted? Hausman and Wongswan (2011) show that the somewhat complex Gürkaynak, Sack, and Swanson (2005) transformation of the yield curve to target and path shocks is nearly equivalent to simply using federal funds surprises and changes in the 12-month-ahead eurodollar futures rate. Figure 1 shows this close relation as scatterplots of the federal funds surprises and changes in the 12-month-ahead eurodollar futures rate versus the target and path factors. The data points in the first panel have a 45degree slope by construction (see Appendix C). The intimate link between changes to year-ahead policy expectations and path surprises raises the question of how the relationship between current target surprises and near-term expectations shifted in 1994. Figure 2 shows a scatterplot of federal funds surprises and corresponding changes in 3- and 12-month-ahead eurodollar rates, along with a 45-degree line denoting one-for-one changes, for two different subsamples. Figures from 1988 to 1993 are shown in the left panel, while figures from 1994 to 2007 are shown in the right panel. Squares (circles) denote surprises that changed eurodollar rates more (less) than one for one. If a surprise to the federal funds target simply moves an expected policy action forward in time, one would expect it to have a less than one-for-one effect on 3- to 12-month interest rates, such as the 3-month eurodollar. In contrast, if a surprise increases expectations of further moves in the same direction, one might expect a more than one-for-one effect on 3- to 12-month interest rates. The panels show that both 3-month and 12-month eurodollar rates responded less strongly—the fitted lines have flatter slopes—to a given target surprise after the FOMC began issuing statements in 1994. This suggests that FOMC target changes produced expectations of less-persistent effects in the latter period. How do the target and path factors affect asset prices? Tables 6 and 7 show the results of regressing equity and foreign exchange returns and interest rate changes, respectively, on the target and path factors for different samples.29 Consistent with Gürkaynak, Sack, and Swanson (2005), the second panels in each table show that the path factor emerges following the FOMC’s 1994 decision to issue postmeeting statements. As in Table 5, Table 7 also documents a significant decline in the importance of target surprises to international interest rate changes in the post-1994 sample as FOMC meeting statements made policy surprises relatively less informative. Note that, in both tables, the coefficients on the path factor in the 1988-93 sample are very large—but often statistically insignificant—because of the lack of path factor variation and lack of information in the path factor during that sample. Statistical methods are not necessarily the only way to identify target and path factors, however. Rosa (2011a,c) attempts to identify the second factor using a “narrative” approach in which he first summarizes the tone of each FOMC meeting statement about the future direction of monetary policy and then approximates the unexpected components of the statement by estimating forecasting regressions. Both the surprise component of policy decisions and the statement’s tone significantly influence stock prices (Rosa, 2011a) and dollar exchange rates (Rosa, 2011c). The surprise component of the statements accounts for most of monetary policy’s effect on asset returns. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 93 Fawley and Neely Table 6 Response of the S&P 500 and Exchange Rates to Target and Path Shocks 1988-93 Target 1994-2007 Path R 2 Target Path 1988-2007 R 2 Target Path R2 S&P500 –3.38 –(2.8) 1.53 (1.4) 0.13 –6.27 –(5.6) –1.68 –(1.7) 0.24 –3.73 –(4.4) –1.06 –(2.2) 0.12 CAD/USD –0.03 –(0.1) 0.10 (0.3) 0.00 –0.57 –(1.3) 0.75 (1.9) 0.05 –0.30 –(1.0) 0.35 (2.1) 0.03 DEM-EUR/USD 1.48 (1.1) 1.60 (1.4) 0.05 –1.42 –(2.2) 1.94 (3.3) 0.13 0.01 (0.0) 1.12 (3.2) 0.06 GBP/USD 0.95 (0.8) 0.72 (0.7) 0.02 –0.58 –(1.1) 1.33 (2.8) 0.08 0.15 (0.3) 0.74 (2.4) 0.03 CHF/USD 1.91 (1.6) 1.48 (1.4) 0.07 –0.68 – (1.0) 2.15 (3.6) 0.12 0.56 (1.0) 1.16 (3.4) 0.07 JPY/USD 1.28 (1.4) 0.81 (0.8) 0.04 0.85 (1.2) 2.17 (3.4) 0.11 0.99 (1.8) 0.97 (3.1) 0.07 NOTE: The table shows results of an exercise similar to that in Gürkaynak, Sack, and Swanson (2005), in which one regresses daily S&P 500 and exchange rate returns on target, path factors, and a constant (coefficient not shown). Columns 2 through 4 show results using all FOMC events for the October 1988–December 1993 period; columns 5 through 7 show similar results for the January 1994–June 2007 period; and columns 8 through 10 show results for the whole sample. The first subsample includes 65 events and the second subsample 112 events, for a total of 177 events. Shaded numbers denote significantly positive (blue) or negative (red) t-statistics. CAD, Canadian dollar; CHF, Swiss franc; DEM, German deutsche mark; EUR, euro; GBP, British pound; JPY, Japanese yen; USD, U.S. dollar. International Effects of Monetary Shocks The literature on the effects of monetary shocks spurred similar research into the effects of monetary policy shocks on international asset prices. Craine and Martin (2008) extend Rigobon and Sack’s (2004) identification-through-heteroskedasticity method to study spillovers of monetary shocks between the United States and Australia, with a special emphasis on accounting for nonmonetary shocks. The authors find that U.S. monetary policy shocks spill over to Australian interest rate and equity markets, but Australian shocks do not seem to affect U.S. financial markets.30 Nonmonetary surprises are more important than monetary surprises for long maturity yields and in equities. Valente (2009) follows Craine and Martin’s work with a study of the interaction between U.S. monetary policy and the yield curve in Hong Kong and Singapore. Two-factor methods have also been applied to international data. Hausman and Wongswan (2011) regress interest rate changes, equity index returns, and exchange rate returns in 49 different countries on target and path factors for U.S. monetary policy surprises and obtain results similar to those of Gürkaynak, Sack, and Swanson (2005): Short-term interest rates respond to both path and target surprises, exchange rates and long-term rates respond primarily to path surprises, and equity indexes respond primarily to target surprises. The insensitivity of equity prices to path shocks is somewhat surprising as equity prices should reflect the present value of profits into the infinite future and thus equities should be sensitive to the whole yield 94 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Fawley and Neely Table 7 Response of International Long Yields to Target and Path Shocks 1988-93 1994-2007 Path R U.S. 10-year 0.64 (9.6) 0.03 (0.5) 0.60 –0.01 –(0.4) 0.58 (17.4) 0.73 0.20 (6.4) 0.35 (19.8) 0.71 Canadian10-year 0.37 (4.1) 0.04 (0.4) 0.23 0.04 (0.7) 0.50 (10.8) 0.54 0.15 (3.5) 0.26 (10.5) 0.43 0.05 (1.0) 0.21 (5.1) 0.21 0.11 (2.2) 0.12 (5.1) 0.20 German 10-year Target Path 1988-2007 Target 2 R 2 Target Path R2 French 10-year 0.15 (1.9) –0.02 –(0.3) 0.06 –0.09 –(2.0) 0.03 (0.8) 0.04 0.00 (0.1) 0.04 (1.8) 0.02 U.K. 10-year 0.13 (1.5) –0.03 –(0.3) 0.03 0.00 –(0.1) 0.14 (2.6) 0.06 0.07 (1.5) 0.06 (2.1) 0.04 Japanese 10-year 0.29 (3.9) 0.02 (0.4) 0.22 0.03 (0.6) 0.12 (2.8) 0.08 0.08 (2.1) 0.11 (4.8) 0.16 NOTE: The table shows results of an exercise similar to that in Hausman and Wongswan (2011), but using the Gürkaynak, Sack, and Swanson (2005) two-factor methodology, in which one regresses daily international long yields on target, path factors, and a constant (coefficient not shown). Columns 2 through 4 show results using all FOMC events for the October 1988–December 1993 period; columns 5 through 7 show similar results for the January 1994–June 2007 period; and columns 8 through 10 show results for the whole sample. The first subsample includes 65 events and the second subsample 112 events, for a total of 177 events. Shaded numbers denote significantly positive (blue) or negative (red) t-statistics. curve. As in Tables 5 and 6, Table 7 shows a strong path factor effect on foreign long rates after 1993. CONCLUSION The speed and flexibility with which monetary policy can be employed has made it the primary policy for macroeconomic stabilization. The importance of this policy for inflation, output, and employment has motivated researchers to explore its impact on the economy. Because of the difficulty in directly discerning the impact of monetary policy on monthly and quarterly macroeconomic variables, researchers have studied the effect of such policies on quick-reacting asset prices as a first step in understanding the broader implications of monetary policy for macro variables. A key lesson from this literature is that researchers must properly account for Federal Reserve procedures to draw the correct inference about the impact of monetary surprises. Specifically, the increasing transparency of Fed procedures and objectives, coupled with credibility won over several decades, has enabled the FOMC to influence asset prices with statements rather than large and disruptive surprises in overnight interest rates. Early research on the effects of low-frequency changes in monetary or reserve aggregates found no consistent effect of these variables on asset prices, despite the Fed’s description of Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 95 Fawley and Neely its own objectives in terms of the former variables. But Cook and Hahn (1989) found strong effects of federal funds target changes—the FOMC’s monetary policy instrument—in the 1970s. Kuttner (2001) established the importance of decomposing policy actions into expected and unexpected components with data from federal funds futures markets, as asset prices should respond only to the unexpected component. Other researchers (e.g., Poole, Rasche, and Thornton, 2002, and Hamilton, 2008) advocated methods that are robust to the measurement error inherent in constructing monetary surprises. FOMC procedural changes, particularly those in 1994-95, have influenced researchers’ methods. The FOMC greatly reduced the frequency of intermeeting target changes, announced target changes as they were decided, and began to issue postmeeting statements to guide public understanding of the economy and likely future monetary policy. These changes reduced the necessity of accounting for other news that might affect asset prices and the simultaneity of monetary policy and asset price changes (Rigobon and Sack, 2004). In addition, the resultant increased Fed transparency has greatly improved the market’s ability to forecast and price in monetary policy actions well before they happen (e.g., Poole, Rasche, and Thornton, 2002, and Swanson, 2006). An apparent disconnect of movements in longer-term interest rates from federal funds surprises led researchers to realize they should account for the effect of FOMC policy statements on year-ahead interest rates (Bernanke and Kuttner, 2005; Gürkaynak, Sack, and Swanson, 2005). In the post-1994 sample, these path surprises significantly affect exchange rates and international interest rates across the yield curve; however, they do not affect U.S. or international equity prices. Gürkaynak, Sack, and Swanson (2005) argue that the influence of these path surprises indicates that the Fed can credibly commit to a policy path. Researchers have used such institutional knowledge to establish that monetary policy surprises influence the prices of many asset classes, including fixed income, equity, and foreign exchange. They have also investigated the impact of monetary policy surprises on equity prices from different types of firms and industries and what that reveals about the relative importance of the asset price and credit channels of monetary policy (Ehrmann and Fratzcher, 2004, 2009; Basistha and Kurov, 2008). Most recently, researchers have found considerable international impact of monetary policy surprises (Craine and Martin, 2008; Hausman and Wongswan, 2011). ■ 96 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Fawley and Neely APPENDIXES Appendix A: Expectations of the Federal Funds Target from Futures Prices The Chicago Board of Trade has sponsored trading in the 30-day federal fund futures contract—an interest rate derivative—since October 3, 1988. Emmons, Lakdawala, and Neely (2006) describe federal funds futures and options on those futures in some detail. The federal funds futures market is a derivatives market whose final settlement price is determined by the average federal funds rate over the contract month. Thus, the final settlement price on the March 2013 contract is determined by the average federal funds rate during that month. If the contract month has M trading days, the final settlement price (price settle,t ) will be price settle ,t = 1 − ff settle ,t = 1 − (A.1) 1 ∑ Mj=1 ff jt , M where ff settle,t is the average federal funds rate implied by the final settlement price. On day i, the federal funds futures rate implied by the day i future price (ffit = 1 – Priceit ) for a given contract month, t, is equal to the expected final settlement price plus a risk premium: ffit = Ei ff settle ,t + rpit = Ei (A.2) 1 M t t ∑ j =1 ff j + rpi . M Piazzesi and Swanson (2008) show that implied federal funds rates from futures are modestly biased predictors, likely to slightly overpredict the rate implied by final futures settlement by 3 to 6 basis points per month of forecast horizon, on average. The authors estimate the following t regression with heteroskedastic-consistent errors: ftn – rt+n = a (n) + et+n , where ftn is the n-periodahead rate implied by the federal funds futures price in month t and rt+n is the actual (realized, ex post) average funds rate in month t + n. For values of n = {1,2, 3, 4, and 5}, a (n) = {3.4, 7.4, 12.5, 19.2, 27.6, and 36.7}, respectively. Some analysts argue that such risk premia are too large to be plausible (Carlson, Craig, and Melick, 2005). But because the risk premium changes very slowly—at business cycle frequencies—daily changes in federal funds futures prices closely approximate daily changes in the market’s expectation of the funds target (Piazzesi and Swanson, 2008). The New York Fed is able to keep the actual federal funds rate “close” to the federal funds target desired by the FOMC. Therefore, the average federal funds rate over a month will be very close to the average federal funds targets for that month: M M 1 1 ff jt ≈ ∑ ff jtarget ,t . ∑ j = 1 M M j =1 (A.3) If there is an FOMC meeting within a month, on day d of the month, the average federal funds target will be a weighted average of the (known) target prior to the FOMC meeting and the new target chosen at the FOMC meeting: (A.4) d 1 1 1 M t ff jtarget ,t + ∑ Mj =d+1 ff jtarget ,t . ∑ ∑ j =1 ff j ≈ M j =1 M M Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 97 Fawley and Neely target ,t If we denote the known target going into the meeting as ff start and the target chosen at the target ,t FOMC meeting, on day d of the month, as ffmeeting , then the average funds rate over the month is equal to 1 d target ,t M − d target ,t M t ff start + ffmeeting ∑ j =1 ff j ≈ M M M (A.5) or, solving for the target to be chosen at the FOMC meeting: target ,t ffmeeting ≈ (A.6) M 1 d target ,t M t ff start . ∑ j =1 ff j − M −d M M Using the fact that the average expected funds rate for a target month is equal to the funds rate 1 implied by the current futures prices ffit = Ei ∑ Mj =1 ff jt + rpit , the expected target, as of M day i, at the FOMC meeting in month t can be written as follows: target ,t Ei ( ffmeeting )≈ (A.7) M t d target ,t t ffstart . ffi − rpi − M −d M If the FOMC meeting day is near the end of the contract month, then M–d will be small and the calculation will be very sensitive to small errors in the data, perhaps caused by bid-ask spreads. In this case, it is better to use the next month’s contract—when there will be no FOMC meeting—to estimate the market’s expectation of the FOMC decision. For relatively short forecast horizons, researchers commonly ignore the risk premium in equation (A.7), assuming that it will be very small. Equation (A.7) can be used to decompose FOMC target changes into expected and surprise components. The expected change in the federal funds target is the expected target at the meeting less the current (start of month) target, expressed as target ,t target ,t target ,t Ei ( ∆ffmeeting ) = Ei ( ffmeeting ) − ffstart M t d target ,t target ,t ffstart − ff start ffi − rpit − − M d M M d target ,t M − d target ,t t t ≈ ffstart − ff start ffi − rpi − M −d M M M t M target ,t t ≈ ffstart ffi − rpi − M M −d M target ,t ≈ ( ffit − rpit − ff start ). M −d ≈ (A.8) (A.9) target ,t target ,t target ,t ≈ Ei ( ∆ffmeeting ) = Ei ( ffmeeting ) − ffstart ( M target ,t ffit − rpit − ff start ( ). M −d ) target , t The surprise component ffmeeting is the actual change less the change that was expected just before the meeting: 98 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Fawley and Neely (A.10) target ,t target ,t target ,t ∆ ffmeeting = ∆ffmeeting − Ei ( ∆ffmeeting ). Kuttner (2001) recognizes that a time-varying risk premium, rpit, potentially contaminates the federal funds shocks calculated from equations (A.9) and (A.10), and that the multiplier M amplifies this measurement error, particularly near the end of the month. Therefore, M −d he proposes a more robust estimate of federal funds shocks using changes in futures prices around FOMC events. Specifically, he observes that the futures price in equation (A.7) can be solved for to obtain the following: (A.11) ffit = M −d d target ,t target ,t + Ei ( ffmeeting ffstart + rpit , ) M M which implies that a daily difference in the futures price should estimate the sum of the daily change in the target for month t plus the daily difference in the risk premium: (A.12) ffi t − ffit−1 = M −d target ,t ∆Ei ( ffmeeting ) + rpit − rpit−1 . M If the day-to-day change in the risk premium is small, then he can estimate the policy surprise—that is, the change in the expected target on the day of the meeting—as follows: (A.13) target ,t ∆ ffmeeting = M ffit − ffi−t 1 ) . ( M −d Changes in futures prices around FOMC events reflect unanticipated changes to the funds rate. While surprises calculated from equation (A.13) may still suffer from measurement error, they remain robust under much weaker assumptions about the sources of such error. Appendix B: Simultaneity and Omitted Bias The monetary policy shock literature typically seeks to determine the effect of monetary policy on some function of asset prices. Researchers have been primarily concerned with the effect of some function of monetary policy (Dmt) on some function of asset prices (Dpt), which could include changes in bond yields, stock prices, or exchange rates, but the relation could also include other news, such as macroeconomic releases (newst ), and/or other variables: (B.1) ∆pt = a1∆mt + a2newst + e p ,t . Although we would like to estimate the parameter, a1, if monetary policy reacts to asset price changes within the observation interval for the data (e.g., within the day for daily data), then b1 will be nonzero in equation (B.2) and this will affect the estimation of equation (B.1): (B.2) ∆mt = b1∆pt + b2newst + em ,t . For simplicity of notation, we assume that all variables are known to be mean zero and that the structural errors, ep,t and em,t , are uncorrelated over time and contemporaneously. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 99 Fawley and Neely They have diagonal covariance matrix V and are also uncorrelated with exogenous news shocks, whose variance is normalized to equal 1. Equations (B.1) and (B.2) can be written in terms of the matrixes, M, P, N, Em, and Ep, which represent the T × 1 vectors whose tth elements are Dmt , Dpt , newst , em,t , and ep,t , respectively: (B.3) P = Ma1 + Na2 + E p (B.4) M = Pb1 + Nb2 + E m . Or, this can be written in matrix form: 1 −b 1 P M = N a2 b2 + E p Em , 1 a − 1 (B.5) and equation (B.5) can be rewritten as follows: P M = NDC −1 + E E C −1 , p m (B.6) 1 −b 1 1 b1 1 , and so C −1 = where D = a2 b2 , C = . The following matrixes 1 − a1b1 a1 1 −a1 1 are used to define the unconditional covariance matrix of the endogenous variables: (B.7) 2 2 1 (a2 + a1b2 ) (b1a2 + b2 ) (a2 + a1b2 ) −1 ′ C D′N ′NDC = N ′N 2 1 − a1b1 (b1a2 + b2 )(a2b1 + b2 ) (b1a2 + b2 ) −1 C ′VC −1 = −1 (B.8) V + a 2V 1 11 1 22 2 (1 − a1b2 ) b1V11 + a1V22 b1V11 + a1V22 , b12V11 + V22 where V11 and V22 are the variances of the structural errors of the price and money equations. The OLS estimates of the parameters in equation (B.3) are as follows: (B.9) −1 −1 â M ′P M ′M M ′N M ′Ma1 + M ′Na2 + M ′E p M M M N ′ ′ 1 = = â2 M ′N N ′N N ′P M ′N N ′N N ′Ma1 + N ′Na2 + N ′E p . Taking the inverse, multiplying through, and eliminating terms that are identically zero yields the following expression for the OLS estimates: ( N ′NM ′M − M ′NN ′M ) a + ( N ′NM ′− M ′NN ′) E â 1 1 P 1 = (B.10) â2 M ′MN ′N − ( M ′N ) 2 ( M ′MN ′N − M ′NM ′N ) a2 + ( M ′MN ′− M ′NM ′) EP 100 First Quarter 2014 . Federal Reserve Bank of St. Louis REVIEW Fawley and Neely The behavior of the coefficients in large samples is given by the probability limits (plims) of the expressions in (B.10), if the limits exist: (B.11) â a 1 = 1 plim T →∞ â2 a2 N ′NM ′Ep 1 + plim 2 T →∞ M ′MN ′N − ( M ′N ) − M ′NM ′E p . Fortunately, the rules of probability limits allow us to pass the plim function through products and quotients. That is, the plim of a product is the product of its plims, and the plim of a quotient is the quotient of the plims. This allows us to determine the plims of the coefficient estimators in equation (B.11). We use the following limits: (B.12) plim T →∞ (B.13) plim T →∞ (B.14) 1 N ′N = 1 T 1 b 2V + V 2 M ′MN ′N − ( M ′N ) = 1 11 222 2 T (1 − a1b1 ) ( plimT →∞ ) 1 V b M ′E p = 11 1 T 1 − a1b1 (B.15) (b a + b ) 2 + b 2V + V 1 11 22 plim [ M ′M ] T = 1 2 2 2 T →∞ (1 − a1b1 ) (B.16) 1 a b +b plim M ′N = 2 1 2 . 1 − a1b1 T →∞ T Equation (B.14) uses the assumption that the structural errors are uncorrelated with the news 1 shocks—that is, plim N ′E p = 0. Using these probability limits in (B.11), we obtain the T →∞ T following: V11b1 2 â a 1 − a1b1 a1 1 a b − b1 ( ) 1 1 1 1 (1 − a1b1 ) + = (B.17) plim = + 2 2 T →∞ â2 a2 (b1 V11 + V22 ) − a2b1 + b2 V11b1 a2 (b1 + V22 V11 ) − (a2b1 + b2 ) 1 − a1b1 1 − a1b1 . Note that the OLS estimation of an equation with an endogenous regressor (Dmt)—that is, b1 ≠ 0—will generally produce inconsistent estimates of both regressors.31 In this case, the OLS estimate, â1, is consistent if the variance of the structural asset price shocks equals zero (V11 = 0) or if Dpt has no contemporaneous effect on Dmt; that is, b1 = 0, in which case Dpt would be considered predetermined. Otherwise, if the ratio of variance of monetary policy shocks to the variance of price shocks (V22 /V11) gets arbitrarily large, then â1 will converge in probability to a1. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 101 Fawley and Neely One way to drive the variance ratio (V22 /V11) of the structural shocks arbitrarily large is to take an arbitrarily short interval around the monetary policy event—an FOMC announcement, for example. In such an interval, the variance of monetary shocks will be very large and the variance of prices can be made arbitrarily small. Also, in such a short interval, the monetary authorities are very unlikely to be reacting to any changes in asset prices, which means that Dpt has no contemporaneous effect on Dmt ; that is, b1 = 0. Note too that if the news variables in equation (B.4) are omitted from the estimated regression, there is a different problem: omitted variables bias. In this case, the OLS estimate of the coefficient on the monetary shock, a1, is as follows: (B.18) −1 −1 â1 = [ M ′M ] M ′P = [ M ′M ] M ′Ma1 + M ′Na2 + M ′E p . Again, we can use the plims that exist for (B.19) plim â1 = a1 + T →∞ 1 1 1 M ′M , M ′N , and M ′E p to show that T T T (1 − a1b1 ) ((a2b1 + b2 ) a2 + V11b1 ) . 2 (b1a2 + b2 ) + (b12V11 + V22 ) If a2 = 0, there is no omitted variable, but simultaneity (b1 ≠ 0) still creates inconsistency: (B.20) plim â1 = a1 + T →∞ (1 − a1b1 ) b1 b V11 + b12 +V22 V11 2 2 . If b1 = 0, then the monetary policy shock is predetermined; there is no simultaneity. In this case, if either b2 or a2 were also zero—that is, news does not contemporaneously affect both asset prices and monetary—then equation (B.19) shows that â1 would be consistent. Otherwise—still assuming that b1 = 0—then as the effect of news on monetary policy (b2) or the variance of monetary policy shocks (V22) gets arbitrarily large, the estimate approaches consistency (see below): (B.21) plim â1 = a1 + T →∞ b2 a2 a2 . = a1 + b + V22 b2 + V22 b2 2 2 Appendix C: Target and Path Surprises Gürkaynak, Sack, and Swanson (2005) identify monetary policy shocks as the unobserved monetary factors F from (C.1) X = FΛ + η , where h are white-noise disturbances and L contains the loadings of F on X. Gürkaynak, Sack, and Swanson (2005) include five variables in X: (i) the surprise to the federal funds target measured from current-month federal funds futures; (ii) the surprise change in expectations of the federal funds target two FOMC meetings ahead, measured from the appropriate federal 102 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Fawley and Neely Table C1 Response of Federal Funds Surprises and 12-Month Eurodollar Futures to the Normalized Target and Path Factors 1988-93 1994-2007 Target Path R MP1 1.12 (15.3) 0.00 (0.0) 0.79 ED12 1.22 (25.0) (0.02) (0.5) 0.91 2 Target 1988-2007 Path R 0.93 (40.5) 0.00 (0.0) 0.94 0.31 (16.6) 1.01 (59.9) 0.97 2 Target Path R2 1.00 (41.6) 0.00 (0.0) 0.91 0.58 (41.0) 0.58 (70.9) 0.97 NOTE: The table shows the result of regressing the federal funds surprise (MP1) and the daily change in 12-month eurodollar futures (ED12) on the target and path factors used as regressors in Tables 5 through 7. The full sample results (1988-2007) show the effect of the normalization procedure, while the subsample results (1988-93 and 1994-2007) show the effect of sampling variation. funds future contract; and the price change in (iii) 6-month, (iv) 9-month, and (v) 12-month eurodollar futures contracts. Gürkaynak, Sack, and Swanson (2005) structurally identify the two monetary policy factors, F, as a linear transformation of the first two principal components (Z) of X. Specifically, (C.2) F = ZU , where Z is the first two principal components of X, and U is a 2 × 2 matrix whose elements are identified by imposing the following restrictions: (i) the columns of U have unit length, (ii) the columns of F are orthogonal, and (iii) F2 , the second column of F, does not influence the current federal funds shock. The last restriction, which implies two equations, provides the structural interpretation of F1 and F2 as the target and path surprise, respectively. In other words, F1 contains all information from the first two principal components that explains the current federal funds surprise, and F2 contains all residual information. Finally, Gürkaynak, Sack, and Swanson (2005) rescale F1 and F2 to provide comparability between coefficients and against earlier studies. Specifically, they scale F1 to move one for one with the current federal funds surprise (MP1) and F2 to have the same magnitude effect on 12-month eurodollar futures (ED12) as F1. Table C1 illustrates the relationship between MP1 and ED12 and the normalized target and path factors used as regressors in Tables 5 through 7. Please note the effect of sampling variation when comparing the size of subsample coefficients in Tables 5 through 7 with previous studies. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 103 Fawley and Neely NOTES 1 The Romer and Romer (1989, 1994) narrative approach is another method to identify monetary policy shocks. The terms “shocks” and “surprises” are closely related but not quite synonymous. A “shock” denotes the unexpected component of a variable in a statistical model. Some researchers reserve the term for the unexpected component of a variable in a structural statistical model. In contrast, a “surprise” denotes any unexpected occurrence, particularly an event that markets did not expect. Most of the research discussed in this article uses statistical methods in which expectations are derived from futures market prices, in which case there is no distinction between shocks and surprises. 2 This paper focuses on the literature studying the reactions to conventional monetary policy shocks, not on reactions to the unconventional policies, including long-term security purchases, which debuted in 2008. 3 The federal funds market is an overnight market in which depository institutions lend reserve balances to other depository institutions. 4 In a regression, the omission of relevant explanatory variables will generally bias the estimated coefficients on included variables, unless the omitted explanatory variables happen to be uncorrelated with the included variables. 5 Friedman (1968) argued that “interest rates are such a misleading indicator of whether monetary policy is ‘tight’ or ‘easy.’” He viewed nominal interest as reflecting the stance of past monetary policy via inflation expectations. 6 Early studies measured the unexpected component of money growth as the residuals from a univariate or multivariate autoregressive model of money growth (Barro, 1978) or as the difference between announced money growth and Money Market Services (MMS) median survey forecasts (Cornell, 1982, 1983; Hardouvelis, 1984). 7 The Federal Reserve did not begin announcing a numerical federal funds target in its statement until July 1995 (Middeldorp, 2011), but it has published an official federal funds target dating back to 1982 (http://research.stlouisfed.org/fred2/series/DFEDTAR?cid=118). Although a monetary policy instrument is not the same as the monetary policy stance—for example, “high” nominal interest rates do not necessarily indicate tight policy if inflation expectations are sufficiently high—a discrete change in the instrument typically implies a discrete change in the stance as price levels and inflation expectations typically do not “jump” to offset a change in the instrument. 8 The end of Cook and Hahn’s (1989) sample coincides with Chairman Volcker’s decision to deemphasize the role of the federal funds rate in setting monetary policy. Cook and Hahn (1989) document the market’s adeptness at interpreting pre-1994, unannounced policy changes from open market operations. Cook and Hahn (1988) find that when discount rate changes signaled federal funds changes, they also moved interest rates. This result held in their 1973-79 and 1979-85 subsamples. 9 Roley and Sellon (1995) do find some evidence that 30-year Treasury yields anticipate future policy. 10 Fatum and Scholnick (2008) confirm this finding and argue that failing to correctly model the unexpected com- ponent of monetary policy news leads to underestimating the effects of such news. They also confirm that systematic reactions occur rapidly, within the day of the announcement. 11 Poole, Rasche, and Thornton (2002) present evidence that markets were commonly able to predict policy actions at least two weeks ahead after 1994. 12 Surprises measured with futures data also contain some measurement error from bid-ask spreads and risk pre- mia, but probably much less than other methods. 13 Some researchers have used longer-term interest rates to identify shocks on the grounds that they better meas- ure surprising actions by the Fed while minimizing measurement error when only the timing of the action is a surprise. For example, if the Fed surprises markets by lowering interest rates one meeting earlier than expected, then near-month federal funds futures will measure a large surprise, while 3-month eurodollar futures might remain unchanged. Cochrane and Piazzesi (2002) measure policy surprises from 1-month eurodollar deposit rates; Ellingsen and Soderstrom (2004) use the 3-month Treasury bill rate; Rigobon and Sack (2004) use the 3-month eurodollar futures rate; and Bomfim (2003), Poole and Rasche (2000), and Poole, Rasche, and Thornton (2002) use month-ahead federal funds futures. Bernanke and Kuttner (2005) check the robustness of their results to those obtained with 3-month eurodollar futures. 104 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Fawley and Neely 14 Rigobon and Sack (2008) argue that correcting for the noise in macroeconomic announcements significantly increases the implied information content of those announcements. 15 Piazzesi and Swanson (2008) argue that risk premia on federal funds futures are small and fairly stable at a daily frequency (also see Hamilton, 2009). 16 Hamilton’s (2008) methodology nests Kuttner’s (2001) specification when the effective federal funds rate always equals the target and the dates of policy actions are known. 17 Gürkaynak, Sack, and Swanson (2005) note a number of instances in which major news was released on days of FOMC meetings or target changes. 18 Thornton (2001a) describes the shift away from unilateral discretion for the chairman to adjust the funds target in favor of consultation with the FOMC before making any changes. 19 In this context, “identification” means that the assumed model permits one to sort out the two-way causality in asset price changes and FOMC actions. That is, the assumed data-generating process in equations (3) and (4) must permit consistent estimation of a1 and b1 if the contemporaneous effects are to be considered identified. Gürkaynak, Sack, and Swanson (2005) find that policy surprises constructed from federal funds futures in 30minute and 1-day windows around policy events are nearly identical; the only notable deviations occur on pre1994 event days that coincide with the release of the employment report. 20 The results are somewhat sensitive to the inclusion of the semiannual report to Congress. 21 Narrow event windows might not capture the full impact on asset prices. Fleming and Piazzesi (2005) report that, with the exception of intermeeting moves, asset prices respond very quickly to FOMC announcements. 22 In the context of borrowing and lending, adverse selection is the tendency of individuals and firms with bad credit to be more likely to seek loans from banks. Moral hazard is the tendency of borrowers to engage in risky activities that will make it less likely they will repay their loans. Both adverse selection and moral hazard are problems because of the existence of asymmetric information, which means that borrowers know things about their ability to repay that lenders do not. 23 Bernanke and Kuttner (2005) follow Patelis (1997) in using the Campbell and Ammer (1993) methodology to decompose stock returns into changes in the discount factor (interest rates), expected dividends, or expected excess returns. 24 Ammer, Vega, and Wongswan (2010) also consider the impact of path shocks described by Gürkaynak, Sack, and Swanson (2005). But, consistent with Gürkaynak, Sack, and Swanson’s (2005) results, path shocks have little effect on equity, so Ammer, Vega, and Wongswan (2010) report only the results for the target shocks. 25 If the FOMC follows a “policy rule” that links policy to economic conditions (e.g., a Taylor rule), then a forecast of economic conditions would imply a policy forecast and vice versa. Faust, Swanson, and Wright (2004) find little evidence that Federal Reserve policy surprises can be used to improve forecasts of statistical releases, which suggests that the Federal Reserve does not necessarily have superior information about the state of the economy. 26 Kool and Thornton (2012), however, argue for a more skeptical view. Their study of forward guidance in New Zealand, Norway, Sweden, and the United States finds limited evidence that forward guidance improves the private sector’s ability to forecast monetary policy. 27 Gürkaynak, Sack, and Swanson’s (2005) two-factor model is arguably very similar to Bernanke and Kuttner’s (2005) use of level and timing shocks in their study of equity reactions. In fact, the two sets of explanatory variables span the same space. 28 The authors thank Brian Swanson for suggesting this exercise and interpretation. 29 The negative “Target” coefficients in Table 6, which would imply a perverse exchange rate response to interest rates, are the product of the use of noisy daily data. Use of intraday exchange rate data produces positive coefficients, as expected. 30 Neely (2013) characterizes the impact of the Federal Reserve’s unconventional policies on international bond yields and exchange rates. Bauer and Neely (2013) investigate the channels through which such effects occur. 31 An estimator is consistent if it converges in probability to the parameter as the sample size increases. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 105 Fawley and Neely REFERENCES Ammer, John; Vega, Clara and Wongswan, Jon. “International Transmission of U.S. Monetary Policy Shocks: Evidence from Stock Prices.” Journal of Money, Credit, and Banking, September 2010, 42(Suppl. s1), pp. 179-98. Andersen, Torben G.; Bollerslev, Tim; Diebold, Francis X. and Vega, Clara. “Micro Effects of Macro Announcements: Real-Time Price Discovery in Foreign Exchange.” American Economic Review, March 2003, 93(1), pp. 38-62. Barro, Robert J. “Unanticipated Money, Output, and the Price Level.” Journal of Political Economy, August 1978, 86, pp. 549-80. Basistha, Arabinda and Kurov, Alexander. “Macroeconomic Cycles and the Stock Market’s Reaction to Monetary Policy.” Journal of Banking and Finance, December 2008, 32(12), pp. 2606-16. Bauer, Michael D. and Neely, Christopher J. “International Channels of the Fed’s Unconventional Monetary Policy.” Federal Reserve Bank of St. Louis Working Paper No. 2012-028D, December 12, 2013; http://research.stlouisfed.org/wp/2012/2012-028.pdf. Bernanke, Ben S. and Blinder, Alan S. “The Federal Funds Rate and the Channels of Monetary Transmission.” American Economic Review, September 1992, 82(4), pp. 901-21. Bernanke, Ben S. and Kuttner, Kenneth N. “What Explains the Stock Market’s Reaction to Federal Reserve Policy?” Journal of Finance, June 2005, 60(3), pp. 1221-57. Bernanke, Ben S. and Mihov, Ilian. “Measuring Monetary Policy.” Quarterly Journal of Econometrics, August 1998, 113(3), pp. 869-902. Bomfim, Antulio N. “Pre-Announcement Effects, News Effects, and Volatility: Monetary Policy and the Stock Market.” Journal of Banking and Finance, January 2003, 27(1), pp. 133-51. Campbell, Jeffrey R.; Evans, Charles L.; Fisher, Jonas D.M. and Justiniano, Alejandro. “Macroeconomic Effects of Federal Reserve Forward Guidance.” Brookings Papers on Economic Activity, Spring 2012, pp. 1-80. Campbell, John Y. and Ammer, John. “What Moves the Stock and Bond Markets? A Variance Decomposition for Long‐Term Asset Returns.” Journal of Finance, March 1993, 48(1), pp. 3-37. Campbell, John Y. and Shiller, Robert J. “The Dividend-Price Ratio and Expectations of Future Dividends and Discount Factors.” Review of Financial Studies, 1988, 1(3), pp. 195-228. Carlson, John B.; Craig, Ben R. and Melick, William R. “Recovering Market Expectations of FOMC Rate Changes with Options on Federal Funds Futures.” Journal of Futures Markets, December 2005, 25(12), pp. 1203-42. Christiano, Lawrence J.; Eichenbaum, Martin and Evans, Charles L. “Monetary Policy Shocks: What Have We Learned and to What End?” in John B. Taylor and Michael J. Woodford, eds., Handbook of Macroeconomics. Volume 1A. Amsterdam: Elsevier B.V., 1999, pp. 65-148. Cochrane, John H. and Piazzesi, Monika. “The Fed and Interest Rates: A High-Frequency Identification.” American Economic Review, May 2002, 92(2), pp. 90-95. Cook, Timothy and Hahn, Thomas. “The Information Content of Discount Rate Announcements and Their Effect on Market Interest Rates.” Journal of Money, Credit, and Banking, May 1988, 20(2), pp. 167-80. Cook, Timothy and Hahn, Thomas. “The Effect of Changes in the Federal Funds Rate Target on Market Interest Rates in the 1970s.” Journal of Monetary Economics, November 1989, 24(3), pp. 331-51. Cornell, Bradford. “Money Supply Announcements, Interest Rates, and Foreign Exchange.” Journal of International Money and Finance, August 1982, 1, pp. 201-08. Cornell, Bradford. “The Money Supply Announcements Puzzle: Review and Interpretation.” American Economic Review, September 1983, 73(4), pp. 644-57. Craine, Roger and Martin, Vance L. “International Monetary Policy Surprise Spillovers.” Journal of International Economics, May 2008, 75(1), pp. 180-96. DeLong, J.B. “America’s Peacetime Inflation: The 1970s,” in Christina D. Romer and David H. Romer, (eds.), Motivation and Strategy. NBER Studies in Business Cycles. Volume 30. Chicago: University of Chicago Press, 1997. 106 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Fawley and Neely Edelberg, Wendy and Marshall, David. “Monetary Policy Shocks and Long-Term Interest Rates.” Federal Reserve Bank of Chicago Economic Perspectives, March 1996, 20(2), pp. 2-17; http://www.chicagofed.org/digital_assets/publications/economic_perspectives/1996/epmar96a.pdf. Ehrmann, Michael and Fratscher, Marcel. “Taking Stock: Monetary Policy Transmission to Equity Markets.” Journal of Money, Credit, and Banking, August 2004, 36(4), pp. 719-37. Ehrmann, Michael and Fratzscher, Marcel. “Global Financial Transmission of Monetary Policy Shocks.” Oxford Bulletin of Economics and Statistics, December 2009, 71(6), pp. 739-59. Ellingsen, Tore and Soderstrom, Ulf. “Monetary Policy and the Bond Market.” IGIER Working paper, Innocenzo Gasparini Institute for Economic Research, 2004. Emmons, William R.; Lakdawala, Aeimit K. and Neely, Christopher J. “What Are the Odds? Option-Based Forecasts of FOMC Target Changes.” Federal Reserve Bank of St. Louis Review, November/December 2006, 88(6), pp. 543-61; http://research.stlouisfed.org/publications/review/06/11/Emmons.pdf. Evans, Charles L. and Marshall, David A. “Monetary Policy and the Term Structure of Nominal Interest Rates: Evidence and Theory.” Carnegie-Rochester Conference Series on Public Policy, December 1998, 49(1), pp. 53-111. Fama, Eugene F. “Efficient Capital Markets: A Review of Theory and Empirical Work.” Journal of Finance, May 1970, 25(2), pp. 383-417. Fatum, Rasmus, and Scholnick, Barry. “Do Exchange Rates Respond to Day-to-Day Changes in Monetary Policy Expectations When No Monetary Policy Changes Occur?” Journal of Money, Credit, and Banking, September 2006, 38(6), pp. 1641-57. Fatum, Rasmus and Scholnick, Barry. “Monetary Policy News and Exchange Rate Responses: Do Only Surprises Matter?” Journal of Banking and Finance, June 2008, 32(6), pp. 1076-86. Faust, Jon; Rogers, John H.; Wang, Shing-Yi B. and Wright, Jonathan H. “The High-Frequency Response of Exchange Rates and Interest Rates to Macroeconomic Announcements.” Journal of Monetary Economics, 2007, 54, pp. 1051-68. Faust, Jon; Swanson, Eric T. and Wright, Jonathan H. “Do Federal Reserve Policy Surprises Reveal Superior Information about the Economy?” Contributions in Macroeconomics, October 2004, 4(1), pp. 1-31. Federal Open Market Committee. “Press Release.” February 4, 1994a; http://www.federalreserve.gov/fomc/19940204default.htm. Federal Open Market Committee. “Press Release.” August 16, 1994b; http://www.federalreserve.gov/fomc/19940816default.htm. Federal Open Market Committee. “Conference Call.” September 21, 1998; http://www.federalreserve.gov/monetarypolicy/files/FOMC19980921confcall.pdf. Fleming, Michael J. and Piazzesi, Monika. “Monetary Policy Tick-by-Tick.” Unpublished manuscript, Federal Reserve Bank of New York, August 2005. Friedman, Benjamin M. and Kuttner, Kenneth N. “Implementation of Monetary Policy: How Do Central Banks Set Interest Rates?” in Friedman, Benjamin M. and Woodford, Michael (eds.)., Handbook of Monetary Economics. Volume 3B. Amsterdam: North-Holland, 2011, pp. 1345-438. Friedman, Milton. “The Role of Monetary Policy.” American Economic Review, March 1968, 58(1), pp. 1-17. Gürkaynak, Refet S.; Sack, Brian P. and Swanson, Eric T. “Do Actions Speak Louder Than Words? The Response of Asset Prices to Monetary Policy Actions and Statements.” International Journal of Central Banking, May 2005, 1(1), pp. 55-93. Gürkaynak, Refet S.; Sack, Brian P. and Swanson, Eric T. “Market-Based Measures of Monetary Policy Expectations.” Journal of Business and Economic Statistics, April 2007, 25(2), pp. 201-12. Hamilton, James D. “Assessing Monetary Policy Effects Using Daily Federal Funds Futures Contracts.” Federal Reserve Bank of St. Louis Review, July/August 2008, 90(4), pp. 377-93; http://research.stlouisfed.org/publications/review/08/07/Hamilton.pdf. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 107 Fawley and Neely Hamilton, James D. “Daily Changes in Fed Funds Futures Prices.” Journal of Money, Credit, and Banking, June 2009, 41(4), pp. 567-82. Hardouvelis, Gikas A. “Market Perceptions of Federal Reserve Policy and the Weekly Monetary Announcements.” Journal of Monetary Economics, September 1984, 14(2), pp. 225-40. Hausman, Joshua and Wongswan, Jon. “Global Asset Prices and FOMC Announcements.” Journal of International Money and Finance, April 2011, 30, pp. 547-71. Kool, Clemens J.M. and Thornton, Daniel L. “How Effective Is Central Bank Forward Guidance?” Federal Reserve Bank of St. Louis Working Paper No. 2012-063A, December 2012; http://research.stlouisfed.org/wp/2012/2012-063.pdf. Kuttner, Kenneth N. “Monetary Policy Surprises and Interest Rates: Evidence from the Fed Funds Futures Market.” Journal of Monetary Economics, June 2001, 47(3), pp. 523-44. Litterman, Robert B. and Weiss, Laurence. “Money, Real Interest Rates, and Output: A Reinterpretation of Postwar U.S. Data.” Econometrica, January 1985, 53(1), pp. 129-56. McCallum, Bennett T. “A Reconsideration of Sims’ Evidence Concerning Monetarism.” Economic Letters, 1983, 13(2-3), pp. 167-71. Meulendyke, Ann-Marie. U.S. Monetary Policy and Financial Markets. Federal Reserve Bank of New York, 1998; http://research.stlouisfed.org/aggreg/meulendyke.pdf. Middeldorp, Menno. “FOMC Communication Policy and the Accuracy of Fed Funds Futures.” Federal Reserve Bank of New York Staff Report No. 491, April 2011; http://www.newyorkfed.org/research/staff_reports/sr491.pdf. Mishkin, Frederic S. “Symposium on the Monetary Transmission Mechanism.” Journal of Economic Perspectives, Fall 1995, 9(4), pp. 3-10. Neely, Christopher J. “Unconventional Monetary Policy Had Large International Effects.” Federal Reserve Bank of St. Louis Working Paper No. 2010-018E, July 2010, updated August 2013; http://research.stlouisfed.org/wp/2010/2010-018.pdf. Nelson, Edward. “The Great Inflation of the Seventies: What Really Happened?” B.E. Journal of Macroeconomics: Advances in Macroeconomics, July 2005a, 5(1), pp. 1-50. Nelson, Edward. “Monetary Policy Neglect and the Great Inflation in Canada, Australia, and New Zealand.” International Journal of Central Banking, May 2005b, 1(1), pp. 133-79. Nelson, Edward and Nikolov, Kalin. “Monetary Policy and Stagflation in the U.K.” Journal of Money, Credit, and Banking, June 2004, 36(3), pp. 293-318. Patelis, Alex D. “Stock Return Predictability and the Role of Monetary Policy.” Journal of Finance, December 1997, 52(5), pp. 1951-72. Piazzesi, Monika and Swanson, Eric T. “Futures Prices as Risk-Adjusted Forecasts of Monetary Policy.” Journal of Monetary Economics, May 2008, 55(4), pp. 677-91. Poole, William and Rasche, Robert H. “Perfecting the Market’s Knowledge of Monetary Policy.” Journal of Financial Services Research, December 2000, 18(2-3), pp. 255-98. Poole, William; Rasche, Robert H. and Thornton, Daniel L. “Market Anticipations of Monetary Policy.” Federal Reserve Bank of St. Louis Review, July/August 2002, 84(4), pp. 65-93; http://research.stlouisfed.org/publications/review/02/07/65-94PooleRasche.pdf. Radecki, Laurence and Reinhart, Vincent R. “The Financial Linkages in the Transmission of Monetary Policy in the United States,” in National Differences in Interest Rate Transmission. Basel: Bank for International Settlements, 1994, pp. 291-337. Reichenstein, William. “The Impact of Money on Short-term Interest Rates.” Economic Inquiry, January 1987, 25(1), pp. 67-82. Rigobon, Roberto and Sack, Brian P. “The Impact of Monetary Policy on Asset Prices.” Journal of Monetary Economics, November 2004, 51(8), pp. 1553-75. 108 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW Fawley and Neely Rigobon, Roberto and Sack, Brian P. “Noisy Macroeconomic Announcements, Monetary Policy, and Asset Prices,” in John Y. Campbell, ed., Asset Prices and Monetary Policy. Chicago: University of Chicago Press, 2008, pp. 335-70. Roley, V. Vance and Sellon, Gordon H. Jr. “Monetary Policy Actions and Long Term Interest Rates.” Federal Reserve Bank of Kansas City Economic Quarterly, Fourth Quarter 1995, 80, pp. 73-89; http://www.kc.frb.org/publicat/econrev/pdf/4q95role.pdf. Romer, Christina D. and Romer, David H. “Does Monetary Policy Matter? A New Test in the Spirit of Friedman and Schwartz,” in Olivier Jean Blanchard and Stanley Fischer, eds., NBER Macroeconomics Annual 1989. Volume 4. Cambridge, MA: MIT Press, 1989, pp. 121-84. Romer, Christina D. and Romer, David H. “Monetary Policy Matters.” Journal of Monetary Economics, August 1994, 34(1), pp. 75-88. Romer, Christina D. and Romer, David H. “A Rehabilitation of Monetary Policy in the 1950s.” American Economic Review, May 2002, 92(2), pp. 121-27. Rosa, Carlo. “Words That Shake Traders: The Stock Market’s Reaction to Central Bank Communication in Real Time.” Journal of Empirical Finance, December 2011a, 18(5), pp. 915-34. Rosa, Carlo. “The Validity of the Event-Study Approach: Evidence from the Impact of the Fed’s Monetary Policy on U.S. and Foreign Asset Prices.” Economica, July 2011b, 78(311), pp. 429–39. Rosa, Carlo. “The High-Frequency Response of Exchange Rates to Monetary Policy Actions and Statements.” Journal of Banking and Finance, February 2011c, 35(2), pp. 478-89. Sims, Christopher A. “Comparison of Interwar and Postwar Business Cycles: Monetarism Reconsidered.” American Economic Review, May 1980, 70, pp. 250-57. Strongin, Steven H. “The Identification of Monetary Policy Disturbances Explaining the Liquidity Puzzle.” Journal of Monetary Economics, June 1995, 35(3), pp. 463-97. Swanson, Eric T. “Have Increases in Federal Reserve Transparency Improved Private Sector Interest Rate Forecasts?” Journal of Money, Credit, and Banking, April 2006, 38(3), pp. 791-819. Thornton, Daniel L. “The Borrowed-Reserves Operating Procedure: Theory and Evidence.” Federal Reserve Bank of St. Louis Review, January/February 1988, 70(1), pp. 30-54; http://research.stlouisfed.org/publications/review/88/01/Borrowed_Jan_Feb1988.pdf. Thornton, Daniel L. “The Information Content of Discount Rate Announcements: What Is Behind the Announcement Effect?” Journal of Banking and Finance, January 1998, 22(1), pp. 83-108. Thornton, Daniel L. “The Codification of an FOMC Procedure.” Federal Reserve Bank of St. Louis Monetary Trends, March 2001a; http://research.stlouisfed.org/publications/mt/20010301/cover.pdf. Thornton, Daniel L. “The Federal Reserve’s Operating Procedure, Non-Borrowed Reserves, Borrowed Reserves and the Liquidity Effect.” Journal of Banking and Finance, September 2001b, 25(9), pp. 1717-39. Valente, Giorgio. “International Interest Rates and U.S. Monetary Policy Announcements: Evidence from Hong Kong and Singapore.” Journal of International Money and Finance, October 2009, 28(6), pp. 920-40. Federal Reserve Bank of St. Louis REVIEW First Quarter 2014 109 110 First Quarter 2014 Federal Reserve Bank of St. Louis REVIEW