The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
A Farewell from President Santomero O by Anthony M. Santomero n March 31, 2006, President Anthony M. Santomero will step down as president of the Federal Reserve Bank of Philadelphia. In this quarter’s message—his last—he reflects on the economic challenges and changes of the past six years and summarizes some of the Bank’s accomplishments. This will be my final contribution to “The Third Dimension” as president of the Federal Reserve Bank of Philadelphia. My tenure as a Fed president has spanned an especially interesting period. I came to the Federal Reserve Bank of Philadelphia in the summer of 2000, on the tail end of one of the greatest bull markets in financial history and in the 10th and final year of the longest economic expansion in U.S. history. We then saw an unprecedented series of disturbances: a terrorist attack on American soil, two wars, numerous financial scandals, even major natural disasters. Meanwhile, longer term trends, such as the blossoming of the Internet age and the globalization of markets, continued to transform the economic and financial landscape. All of these things challenged the Fed—as monetary policymaker, as banking regulator, and as payment system provider. During this time, the Federal Reserve was under the chairmanship of Alan Greenspan, a man who has been called the greatest central banker who ever lived. It has been my privilege to www.philadelphiafed.org work toward meeting these challenges both as a member of the Federal Open Market Committee (FOMC) and as president of this Bank. Along the way, I have shared my perspectives on many of these issues with you here in the Business Review. I hope you have found them interesting. Indeed, I hope you have found — and will continue to find — all of the articles we bring to you through the Business Review worthwhile. They represent our best effort to share with you the work of our dedicated staff of economists in the realms of economic policy, banking and payments, and financial and regional economics. More broadly, the Business Review is one way in which our Bank achieves what I see as its fundamental goal: to serve as a center of central bank knowledge and capability. I want to take this opportunity to summarize some other important ways in which the Bank has been building, sharing, and applying knowledge to contribute to the Fed's effectiveness as the nation's central bank. In the area of economic research and policy, we have built a more vibrant research enterprise. Our economists are producing high-quality research on issues important to central banking and sharing that work with others via participation in professional conferences across the country and around the world. In addition, we have also increased the number of conferences and workshops that we have organized here at the Bank. Indeed, our palpable presence at the nexus of research and policy is perhaps best embodied in our Philadelphia Fed Policy Forum.* This remarkably successful annual event brings together leading academics, policymakers, and market economists for debate and discussion of relevant macroeconomic and monetary policy issues. For a summary of the 2004 Policy Fourm, see the Third Quarter 2005 issue of the Business Review. * Anthony M. Santomero, President, Federal Reserve Bank of Philadelphia Q1 2006 Business Review We are honored that one commentator has called it the “Jackson Hole of the East,” referring to the highly regarded annual conference of economists held in Jackson Hole, Wyoming. All of these activities help deepen the Fed’s understanding of monetary policy and its impact on the national economy. On a personal level, the work of our Research Department has served me well—challenging my thinking as an economist and helping me sharpen my contributions to the discussions at the FOMC. Let me add that participating in the FOMC discussions in Washington has truly impressed upon me the gravity of the Fed’s role in our nation’s economy. The Bank has been active in other areas of central banking as well. The Philadelphia Fed has long been a System leader in the payments arena. We operate one of its premier check processing centers, and recently we have been named as a System consolidation site, absorbing the check volume from New York’s East Rutherford Operations Center. At the same time, we are advancing the System’s knowledge of the evolving payments system with the establishment of our Payment Cards Center. And we are playing a key role in modernizing government payments through the services we provide to the U.S. Treasury. In the area of bank supervision, we established a unit in our Supervision, Regulation and Credit Department to analyze retail credit risk. Now the System has charged us with developing the strategy for imple- Q1 2006 Business Review menting this part of the new Basel II capital requirements. Our Bank is also now responsible for implementing the System’s new discount window lending policy and overseeing requisite upgrades to our technology. At the same time, our Bank has made strong contributions to economic education and financial literacy. We are System leaders in economic education and have developed finan- delphia community and to become an integral part of economic policy discussion in our District and the nation. Our Bank has had the opportunity to contribute to such discussions throughout our region and to establish its expertise in monetary policy on the national stage. In my view, the Philadelphia Fed has established itself as an outstanding component of the Federal Reserve As part of the nation’s central bank, we are an organization with an important niche in public service and a stellar reputation for quality and credibility. cial literacy curricula used locally and nationally. Our programs to promote financial literacy, improve access to credit, end predatory lending, and foster urban development continue to make a real difference in our communities. To increase public understanding of the Fed and its role, our Bank opened its doors to visitors on Independence Mall with the “Money in Motion” exhibit. This interactive financial exhibit has brought over 67,000 national and international visitors to our Bank since it opened. We also sponsored and hosted roughly 572 events in our new state-of-the-art conference center in its first year of operation. Over the six years of my tenure, our Bank has strived to become an even more important part of our Phila- System. As part of the nation’s central bank, we are an organization with an important niche in public service and a stellar reputation for quality and credibility. I am confident that the Bank will continue to move forward under its new leadership and will remain steadfastly committed to the strength and growth of the Third District’s economy. In closing, I would like to express my gratitude for having had the pleasure of national service in a truly outstanding institution and for the opportunity to work under Chairman Greenspan. With Greenspan’s departure, President Bush has made an excellent choice in Ben Bernanke as our new Chairman. I am confident that the reins of the Federal Reserve are in good hands. I wish him, my colleagues, and our readers the very best. BR www.philadelphiafed.org Debt Maturity: What Do Economists Say? What Do CFOs Say? by Mitchell Berlin L ike households, firms that borrow money to finance operations must make decisions about the optimal maturity of their debt. Should a firm take a short-term loan now and refinance later? Or is the firm better off locking in a long-term interest rate now? In this article, Mitchell Berlin discusses recent theories of how firms choose their debt maturity. Some of these theories are very useful for explaining how chief financial officers (CFOs) choose the maturity of their firms’ debt. However, CFOs seem to believe that they can predict future interest rates and time their borrowings accordingly, and this behavior fundamentally conflicts with most economic theories. Any homeowner who has shopped around for a mortgage would recognize many of the concerns facing the chief financial officer (CFO) puzzling over her firm’s optimal debt maturity. A CFO may ask, “Should my firm sell a long-term bond and lock in the current 30-year rate, or should my firm sell a five-year note and refinance in five years?” One of the CFO’s concerns is that the five-year loan may subject Mitchell Berlin is an assistant vice president and economist in the Philadelphia Fed’s Research Department. He is also head of the department’s banking and financial markets section. www.philadelphiafed.org her firm to the risk of refinancing at an inopportune time, for example, when the bond market is skittish and risk premiums are high, or following a string of negative earnings reports. There is now substantial evidence that many CFOs also ask themselves, “Do I think that long-term rates are going to fall soon? If so, maybe we should take out a short-term loan now and refinance at a lower rate five months from now.” While this reasoning may seem sound to some readers, economic models of the relationship between short-term rates and longterm rates say that the CFO is wasting his or her time hoping to lower the firm’s borrowing costs in this way, as I’ll discuss later. Some leading theories of firms’ choice of debt maturity are based on the idea that firms are better informed about their own creditworthiness than are lenders, another consideration that may be familiar to household borrowers. For example, a CFO of a firm with a promising, but untested, new product may reason that by borrowing short-term and reentering debt markets next year, the firm can lower borrowing costs because lenders are likely to raise their projections of firm profitability once initial sales figures come in. Another leading theory says that short-term debt tends to mitigate the conflicting interests of a firm’s stockholders and its bondholders about the firm’s choice of investments. Empirical studies of firms’ debt maturity choices suggest that finance theorists have made significant progress in explaining these matters; at the same time, these empirical studies have uncovered a few interesting puzzles. Sophisticated borrowers’ belief that they can lower their funding costs by timing the maturity of their borrowings based on their forecasts of interest rate movements is one of these puzzles. PRIVATE INFORMATION MAY AFFECT DEBT MATURITY Short-Term Loans Make Funding Costs More Sensitive to New Public Information About Firms. In an influential model by Douglas Diamond, a firm’s insiders (its managers and large stockholders) have information about the firm’s likelihood of default that is superior to that of outsiders, in this case, its creditors. That is, insiders have what economists call private information. This is not to say that creditors are completely uninformed about the firm. They may Business Review Q1 2006 3 have a number of observable indicators of the firm’s credit risk, for example, a credit rating from Moody’s. However, the firm’s managers will often have better information than creditors about the firm’s prospects because of managers’ involvement in the running of the firm. This means that two firms that both have B+ ratings from Moody’s may actually have quite different probabilities of defaulting on their debts. This is a problem not only for creditors but also for the firm that is truly more creditworthy than other, similarly rated firms. Unless the lower risk firm can find some way to signal its private information to its lenders, it will end up borrowing at the same high rate as all the other B+ firms because creditors will be unable to tell them apart. Diamond argues that one possible way for the low-risk firm to lower its borrowing costs is to shorten its debt maturity. Matters known only to management today will gradually become more public in the course of time; for example, a firm that has a low risk of default is more likely to generate a good quarterly earnings report in the future than a higher risk firm. When lenders see a new earnings report, they update their beliefs about a firm’s credit risk, and the firm will be able to borrow at a lower rate than previously. So a manager with private information that his firm is more creditworthy reasons: “With short-term loans we can make our borrowing costs more sensitive to public information as it becomes available to lenders. Since our earnings report is likely to contain good news about our future prospects, our cost of funds is likely to fall.” 1 But Refinancing Short-Term Debt Creates Liquidity Risk. If the future were perfectly predictable, this would be the end of the story. A manager with private information that his firm is low risk would always choose 4 Q1 2006 Business Review the shortest possible maturity. However, even firms with low default risk may temporarily suffer low profits and find themselves trying to refinance their debt at an inopportune time. This might simply lead to higher interest costs for a time, or it might force the firm to cut back business or forgo profitable investments. This is called liquidity risk. Liquidity risk limits even a low-risk firm’s appetite for shortening The Firm’s Optimal Debt Maturity Depends on Observable Measures of Credit Risk. Consider two firms, one of which creditors view as riskier than the other based on observable indicators of credit risk. Although in the real world the information available to creditors is a lot broader than the credit rating alone, I’ll use the shorthand term credit rating to summarize these observables. In Even firms with low default risk may temporarily suffer low profits and find themselves trying to refinance their debt at an inopportune time. the maturity of its debt. While low-risk firms will take account of liquidity risk, high-risk firms will take it even more seriously because they have a higher likelihood of reporting low profits and facing higher borrowing costs. There is empirical evidence that liquidity risk is a real concern for firms and that it affects their choice of debt maturity. CFOs responding to John Graham and Campbell Harvey’s extensive survey of 392 financial executives cite the “cost of refinancing in bad times” as the second most important factor affecting their debt maturity choice. High-risk firms would prefer to make their borrowing costs less sensitive to public information by locking in today’s rate. But in Diamond’s model, they are forced to mimic the low-risk firms and borrow short-term funds or else be revealed as high-risk firms. Economists call this type of equilibrium a pooling equilibrium. Mark Flannery’s paper also highlights private information and low-risk firms’ desire to make funding costs more sensitive to public information as it arrives. In his paper, the countervailing cost of short-term debt is that underwriters must be paid each time firms sell a new debt issue. He presents a separating equilibrium in which managers with private information that their firm is high risk do not mimic the managers of low-risk firms; low-risk firms issue short-term debt and high-risk firms issue long-term debt. 1 Diamond’s model, firms with a higher credit rating are more likely to report strong earnings in the future than those with a lower credit rating. That is, at higher credit ratings it is more likely that the manager of a firm has private information that the firm is low risk. This means that firms with higher credit ratings face less liquidity risk, and thus, Diamond predicts firms with higher credit ratings will use more short-term debt than firms with lower credit ratings. Actually, there is a twist. For some very risky firms, lenders are simply unwilling to lend long term because lenders will lose money too often if they are unable to raise their rate or will refuse to provide further funding based on the most current information. As a result, lenders will provide only very short-term financing for such firms to keep them on a short leash.2 So Diamond predicts that both very low-risk and very high-risk firms will use short-term debt. Lenders use a number of other contractual devices for such borrowers, especially collateral and detailed loan covenants. 2 www.philadelphiafed.org The Empirical Evidence for the Signaling/Liquidity Risk Tradeoff. Studies of large firms with access to public securities markets uniformly support Diamond’s predictions.3 Researchers have found that a firm’s debt maturity increases as its credit rating falls, at least until its credit rating becomes speculative (BB or lower).4 They also find that firms without a credit rating typically use more short-term debt. There are two ways to think about the absence of a credit rating. No credit rating usually means that there is little public information about a firm — which investors view as a source of risk in itself — or it may mean that a firm is smaller and riskier than the typical firm with public ratings. Studies of small firms, which have more limited access to financial markets than large firms, provide less consistent support for Diamond’s model. These firms typically borrow from banks or from finance companies rather than by selling bonds to the public. There are difficulties testing Diamond’s predictions for firms without credit ratings because researchers often can’t observe what information is actually available to lenders. Allen Berger, Marco EspinosaVega, W. Scott Frame, and Nathan Miller take the interesting approach of using the internal risk ratings that banks assign their loan customers as a summary measure of the information available to a firm’s lender and find that debt maturity is longer for riskier loans, as Diamond predicts for firms with low and moderate credit risk. However, there is no switch in the relationship for the riskiest borrowers. One possible explanation for this finding is that bank loan contracts to the riskiest borrowers are likely to include very close monitoring by the lender; thus, the relationship between credit risk and maturity may be more complicated than in Diamond’s model.5 This monitoring includes the extensive use Researchers have found that a firm’s debt maturity increases as its credit rating falls, at least until its credit rating becomes speculative (BB or lower). of covenants that require the borrower to prove its financial health to avoid default; longer-term debt with extensive covenants may be viewed as a relatively close substitute for short-term debt. In itself, the availability of such a substitute may confound the relationship between credit risk and maturity in the Diamond framework. STOCKHOLDER/BONDHOLDER CONFLICTS MAY AFFECT DEBT MATURITY When Firms Have Too Much Long-Term Debt, Managers May Forgo Profitable Investments. Another classic article by Stewart Myers Berger and co-authors also find evidence that supports the empirical significance of private information for the debt maturity of bank loans. Specifically, they show that the relationship between the bank’s rating and the firm’s maturity is weaker for loans in which the bank used credit scoring as part of the loan underwriting process. The authors argue (plausibly) that information asymmetries are less significant for such loans. 5 Articles that provide empirical support for Diamond’s model for larger firms include the ones by Michael Barclay and Clifford Smith; Mark Stohs and David Mauer; Shane Johnson; and Michael Faulkender and Mitchell Petersen. 3 Bonds are rated according to their default risk by ratings agencies, the most prominent of which are Moody’s and Standard & Poor’s. 4 www.philadelphiafed.org begins by making the crucial distinction between investments already in place and growth opportunities, options to make an investment sometime in the future. The key to Myers’s theory is that the way in which the investments in place have been financed — in particular, whether with debt or with equity, and if debt, whether with shortor long-term debt — affects the profits stockholders make from exercising growth options. To see the issues, begin with the simple case where the growth opportunity is profitable on a stand-alone basis — that is, the project has positive net present value — and the firm has no existing debt. In this case, existing stockholders would evaluate the growth opportunity separately from the investment in place and would support exercising the growth opportunity because it is profitable.6 But if the firm has debt outstanding, stockholders will have to share future profits with the bondholders who provided the funds to finance the investment in place. When the outstanding debt is large enough to affect a firm’s investment decisions, it is often referred to as the debt overhang. If the debt overhang is large, the bondholders will capture a relatively large share of the projected revenues from the new profits, and the firm might forgo the profitable growth opportunity. This is known as the underinvestment problem. To see how this can happen, consider a firm that owns a fleet of carts that sell roasted chestnuts in Central Park. The firm is considering whether to purchase a second fleet of ice cream stands. The chestnut carts are profitable only in cold weather, while the ice Myers assumes that managers faithfully carry out the interests of the firm’s existing stockholders. Other prominent models in finance emphasize the conflict between managers’ interests and those of the firm’s stockholders. 6 Business Review Q1 2006 5 cream stands are profitable only when the sun is shining. Forecasters are predicting an early spring and a long summer, which means that the chestnut carts are likely to be unprofitable, and the firm may even have to default on its debt if it doesn’t diversify into ice cream sales. How will the firm’s managers reason? According to Myers, they may argue: “Most of the profits from the new ice cream stands are going to go to pay off the old debt used to finance the chestnut carts, rather than to the firm’s stockholders. The profits received by current stockholders are much lower than the profits that would be generated by the ice cream stands alone. Since we are concerned about our existing stockholders, we shouldn’t make the investment, even though it would be profitable.”7 While stockholders might support this decision in the short run, in the long run, they would actually prefer that the firm find a way to avoid underinvestment. To see this, think about a firm that systematically says no to profitable new investments; such a firm would suffer from an endemically low stock price because its profits will be low. So stockholders—especially the stockholders of firms with significant growth opportunities—would support policies to induce managers not to pass up profitable investments. Firms with large growth opportunities can reduce the underinvestment problem in two ways. First, they can borrow less to begin with. The less debt a firm has, the lower the possibility that creditors’ and stockholders’ interests will conflict in a material way. (Microsoft is an example of a firm with The discerning reader may ask why stockholders and bondholders can’t strike some kind of deal to ensure that the profitable investment is made. Myers’s model assumes that there are impediments to renegotiating the terms of the debt. 7 Q1 2006 Business Review no debt outstanding.) Second, for any given amount of debt, the firm can use primarily short-term debt, specifically debt that matures before its existing investments. For example, a firm that uses three-month bank loans or commercial paper that matures in three months can’t shift risks to its creditors because the creditors can insist on a new interest rate in line with current risks every three months. The Evidence for a Relationship Between Underinvestment and Debt Maturity. One of the predictions of Myers’s model mirrors a standard practitioner’s rule of thumb: A firm should try to match the maturity of its assets and liabilities. Indeed, for Graham and Harvey’s CFOs, matching assets and liabilities is the most commonly cited factor determining debt maturity. In all empirical studies of firms’ choices of debt maturity, firms with longer-lived assets have longerterm debt. While there is significant empirical evidence for maturity matching, Myers underinvestment story is not the only theoretical rationale for this practice. (See Enforcement Concerns May Affect Debt Maturity.) Myers’s model also predicts that firms with larger growth opportunities should use more short-term debt. Think of a fast-growing firm or one with substantial investments in R&D as examples of firms with significant growth opportunities.8 Although the literature is not unanimous, most studies support this prediction. Shane Most studies use the ratio of a firm’s market value to its book value as a measure of growth opportunities. The market value includes the firm’s outstanding stock measured at market prices and the value of its debt, while the book value is the original sale price of the stock plus the value of its debt. The idea behind using this ratio as a measure of growth opportunities is that the firm’s stock price will include investors’ valuation of future investments. Many studies also use the firm’s investment in R&D as an indicator of growth opportunities. Johnson’s article is probably the most thorough empirical study so far. First, he finds that firms with larger growth opportunities take on less debt. Second, he finds that firms that use primarily short-term debt have higher debt loads than firms that use primarily long-term debt. These findings are consistent with the idea that firms try to avoid underinvestment both by reducing their reliance on debt and by shortening the maturity of the debt they use.9 MARKET TIMING MAY AFFECT DEBT MATURITY Managers Seem to Believe That They Can Time the Market. The empirical literature has consistently found evidence suggesting that managers time their borrowings in the belief they can use their forecast of interest rate movements to lower their cost of funds. This is the third most common reason given by CFOs in Graham and Harvey’s survey. Specifically, CFOs say that they issue short-term debt when “short-tem rates are low compared to long-term rates,” or when “we are waiting for long-term rates to come down.” This is particularly important for large firms, which have relatively easy access to financial markets. Other studies have consistently found that short-term borrowings are higher when the term spread—the difference between the 10-year and the one-year interest rate on Treasury securities—is high, that is, when long-term rates are relatively high compared to short-term rates.10 Graham and Harvey’s response 8 Johnson’s paper takes explicit account of both the firm’s choice of leverage and of the maturity of its debt as a means of resolving underinvestment. This resolves some contradictory findings in the earlier contributions. 9 These studies include those by Barclay and Smith; Jose Guedes and Tim Opler; and Faulkender and Petersen. 10 www.philadelphiafed.org ENFORCEMENT CONCERNS MAY AFFECT DEBT MATURITY O liver Hart and John Moore present a model of debt maturity in which the borrower can’t fully commit to a stream of future debt payments. In particular, an entrepreneur can always threaten to walk away from her debts. Although she can be compelled to turn over the physical assets of the firm in the event of default, her accumulated skills and knowledge (her human capital) can’t be touched by her creditors. Thus, each debt payment potentially gives rise to bargaining between the lender and the borrower, in which the lender threatens to take the firm’s assets and the borrower answers that assets are worth less in the hands of another manager (so that the lender would be shooting himself in the foot by carrying out the threat). While this scenario may seem a little melodramatic as a description of a routine debt transaction, Hart and Moore argue that, ultimately, the borrower’s threat to walk away and the lender’s threat to seize assets determine the feasibility of a particular stream of debt payments. If the debt maturity is too long term, i.e., if debt payments are postponed until too late in the productive life of the assets they finance, the borrower’s threat to walk away becomes a serious problem. In this case, the contract would be calling upon the borrower to make large repayments precisely when the lender’s threat to liquidate is weakest. When the remaining life of the assets is very short: (1) the borrower’s lost profits from losing control of the assets are small, so the costs of walking away are small;a (2) the value of the assets to the lender is small (because the assets have depreciated), so a the gains to the lender from seizing assets are small. At the same time, the firm’s debt can be too short term. Investments yield profits over time, and the firm’s accumulated cash flow from a project may be too small to cover large debt payments early on. One possibility is to use short-term debt that might be renegotiated if current cash flows are too small to cover promised payments. But efficient renegotiation may be impossible. The problem is that even if future revenues are high enough to shift some payments into the future under a renegotiated agreement, the borrower has only limited ability to make credible commitments to make debt payments out of future revenues. So renegotiations would only lead to promises that would never be kept.b Hart and Moore’s theoretical analysis includes two interesting empirical predictions. The first is that firms will match the maturity of their assets and their debts; thus, Hart and Moore provide another explanation of this businessman’s maxim (borne out by Graham and Harvey’s survey findings). As assets become longer lived, they provide the creditor with the security to wait longer before being repaid; the lender’s threat to seize assets is more credible when assets are longer lived. A second prediction is that more fungible assets — those that can be more readily used by another firm — can more readily support long-term debt. Again, the firm’s ability to commit to making debt payments out of future revenues is enhanced by the strength of the lender’s threat to seize assets. This prediction has recently found empirical support in a historical study of the U.S. railroads by Efraim Benmelech. Of course, borrowers will also be concerned about their reputation and their access to future finance. b An implication of this line of reasoning is that the borrower’s inability to commit to forgo seeking to renegotiate the contract can lead to underinvestment (for reasons different from those emphasized by Myers). That is, some essentially profitable investments simply can’t be financed. This will happen if the borrower has to borrow a large share of the initial investment, if the project yields cash flows too late in the life of the project, and if the project’s liquidation value is too low. www.philadelphiafed.org Business Review Q1 2006 7 to their findings sums up researchers’ typically puzzled response: “[I]t is not clear to us why firms pursue this strategy” (p. 233).11 Michael Faulkender’s article presents striking evidence that firms’ managers believe they can time their borrowings to reduce their cost of funds. He examines financing policies for firms in the chemical industry between 1994 and 1999.12 In particular, Faulkender examines these firms’ use of interest rate swaps undertaken jointly with their new borrowings. In the simplest interest rate swap, one party exchanges the interest payments on its own debt for the interest payments on another party’s debt. So a firm that pays a floating rate on its own debt can exchange these variable payments for another firm’s fixed interest rate payments. When a firm increases its shortterm debt, it will end up paying higher interest costs if interest rates rise. Conversely, when it increases its long-term debt, it will be paying higher interest rates than those prevailing in the market if interest rates fall. This is called interest rate risk. When firms undertake a new borrowing they may take an accompanying swap position to offset, or hedge, their interest rate risk. For example, a firm that takes on new floating rate debt will be hurt if interest rates rise. The firm can hedge One theory of debt maturity by Ivan Brick and Abraham Ravid does predict that the term premium should affect firms’ optimal debt maturity through tax effects. But their model has been consistently rejected by the data, with the puzzling exception of Stohs and Mauer’s paper. 11 Focusing on firms in a single industry is attractive for two reasons. First, since firms in a single industry face similar operating environments, it is less likely that empirical findings are driven by unobserved differences between firms. Second, it is easier to find comparable measures for factors that researchers expect will affect firms’ borrowing policies, for example, factors affecting industry risk. 12 8 Q1 2006 Business Review this risk by exchanging the interest rate payments on its floating rate debt for fixed interest rate payments in the swap market.13 However, a CFO who firmly believes that he or she can forecast interest rates might not hedge against the risk of interest rates rising but might purchase a swap that amplifies the firm’s exposure to rising rates. This behavior is called speculation. affecting the relationship between short-term and long-term interest rates is investor expectations about future interest rates. For example, according to the expectations theory of the yield curve, the 10-year Treasury rate is simply the average of investors’ expected one-year T-bill rates over the next 10 years. While it is plausible that corporate CFOs might have private The belief that CFOs can time the market is equivalent to the belief that sophisticated lenders are systematically taken to the cleaners by corporate CFOs. Faulkender finds that the swaps undertaken in conjunction with new borrowings are not taken for hedging purposes. That is, a firm that takes on new floating rate debt is not typically swapping floating interest payments for fixed interest payments. In itself, this is surprising. Most academic observers have simply assumed that these swaps were undertaken to hedge the new debt. Further, the likelihood that a firm takes a speculative position depends on the term premium. So a firm is not only more likely to borrow using floating rate debt when the term premium is high; it is also more likely to swap fixed interest payments for floating interest payments at the same time as the debt offering when the term premium is high. Most Economists Believe That CFOs Can’t Time the Market. To a first approximation, the main factor The reader may wonder why a firm would ever want to borrow short term using floating rate debt if it preferred a fixed interest rate (or vice versa). A number of explanations are possible. One is that the firm chooses its debt maturity to avoid underinvestment. While short-term debt might be attractive for this reason, a firm may not want to bear the additional interest rate risk. 13 information about their firms’ financial condition — as in Diamond’s model — they are very unlikely to have superior information about future interest rate movements, and thus, they are not likely to produce systematically better interest rate forecasts than other market participants.14 If CFOs don’t have superior forecasts, economic theory says that borrowing short term and refinancing should lead to the same (risk-adjusted) borrowing costs as borrowing long term.15 One way to see this is to imagine that CFOs could systematically reduce borrowing costs by borrowing short term when short-term rates are I am simplifying here. Factors other than expectations affect the precise shape of the yield curve and the theory of the yield curve is a venerable and continuing controversy in economics. Nonetheless, the main point still holds. There is little reason to imagine that corporate treasurers have systematically better information than other market participants about other factors affecting the supply and demand for funds at different maturities. 14 Just to be clear, in this discussion I am not taking account of the issues considered by Diamond. Think about a world in which all information about a firm’s creditworthiness is public information. 15 www.philadelphiafed.org unusually low and switching to longterm rates when long-term rates are unusually low. This would mean that borrowers are systematically profiting at lenders’ expense. It should be kept in mind that these lenders are large banks, insurance companies, money market funds, and so forth. Thus, the belief that CFOs can time the market is equivalent to the belief that sophisticated lenders — whose business is to make money by borrowing and lending — are systematically taken to the cleaners by corporate CFOs. Why do CFOs believe they can time the market? One possibility is that CFOs are simply wrong and that their firms’ cost of funds is not lowered by market timing. Indeed, Graham and Campbell’s survey uncovers a number of capital budgeting and financing policies that are very common, yet don’t appear rational from the standpoint of a financial economist. Another possibility is that CFOs are actually able to lower their cost of funds, but for reasons other than their ability to forecast interest rate movements. Perhaps managers have hit upon a rule of thumb that has actually worked, but not because CFOs have better models of the yield curve. In their article, Malcolm Baker, Robin Greenwood, and Jeffrey Wurgler, using U.S. data between 1953 and 2000, present empirical evidence that large firms do indeed borrow short term when long-term borrowing would have been more expensive (and vice versa). The authors suggest that managers may actually be able to exploit inefficiencies in debt markets to lower their borrowing costs, although they do not identify a particular type of market inefficiency that would explain this possibility. Alexander Butler, Gustavo Grullon, and James Westen have argued that Baker and coauthors’ results are www.philadelphiafed.org flawed on econometric grounds.16 Apart from these concerns, most economists will remain unconvinced about the profitability of market timing without a plausible economic mechanism to explain corporate treasurers’ success. Ultimately, Baker and coauthors’ main argument for taking the possibility of profitable market timing seriously is that corporate treasurers believe it is profitable. But essentially irrational practices can persist as long as the available data do not provide strong evidence that the practice is losing money. The correlations unearthed by Baker and coauthors suggest that during the postwar period, corporate treasurers could convince themselves that they were not losing money for their firms, even if their dreams of timing the market were delusory. CONCLUSION Financial economists have made significant progress in understanding firms’ debt maturity decisions. Substantial empirical evidence supports the view that firms’ private information about their credit risk is an important determinant of debt maturity. In particular, the evidence is broadly consistent with a model in which firms balance two opposing factors. Shortterm debt makes borrowing costs more sensitive to public information but may force a firm to borrow at an inopportune time. Substantial evidence also supports the view that firms with significant growth opportunities will choose the maturity of their debt to Butler, Grullon, and Westen’s article makes a convincing argument that the empirical patterns in Baker et al. are spuriously driven by structural shifts during the 1980s. In particular, they argue that both excess returns and firms’ debt maturity policies changed in response to changes in monetary and fiscal policy in the 1980s, leading to a spurious correlation in Baker et al.’s data. 16 avoid debt overhang, which can lead the firm to forgo profitable investments. While it is not the business of economists to slavishly produce models that reinforce businessmen’s prejudices, both views find support in survey responses by CFOs, who state that they choose debt maturity to match the maturity of their assets and liabilities and that their borrowing choice reflects their desire to avoid having to borrow at an inopportune time. CFOs’ own statements provide financial economists with some comfort that they are not theorizing about debt maturity in a vacuum. While the theories seem to have been successful in explaining the borrowing choices of large firms, financial economists have made less headway in understanding maturity decisions for smaller firms. However, CFOs also state that their debt maturity choices are partly driven by the desire to borrow short term when short-term rates are unusually low or to lock in a long-term rate when they believe long-term rates are likely to rise. There is also substantial empirical evidence that firms’ financing decisions do, in fact, reflect this motive. Here, there is less comfort for economists because economic models do not support the idea that firms can systematically reduce borrowing costs this way. While economists are often puzzled and challenged to explain business practices — is the practice simply irrational, or is there some logic to it? — CFOs’ belief that they can reduce borrowing costs by timing the maturity of their borrowings is even more puzzling, because there is some recent evidence that such timing may actually work. Unsurprisingly, this evidence has been forcefully challenged and remains an open area for research. BR Business Review Q1 2006 9 REFERENCES Baker, Malcolm, Robin Greenwood, and Jeffrey Wurgler. “The Maturity of Debt Issues and Predictable Variation in Bond Returns,” Journal of Financial Economics (November 2003). Butler, Alexander, Gustavo Grullon, and James Westen. “Can Managers Successfully Time the Maturity of Their Debt Issues?” Working Paper, Rice University (July 2004). Barclay, Michael J., and Clifford W. Smith Jr. “The Maturity Structure of Corporate Debt,” Journal of Finance, 50 (June 1995), pp. 609-31. Diamond, Douglas. “Debt Maturity Structure and Liquidity Risk,” Quarterly Journal of Economics (August 1991), pp. 709-37. Barclay, Michael J., Leslie M. Marx, and Clifford W. Smith Jr. “The Joint Determination of Leverage and Maturity,” Journal of Corporate Finance, 9 (2003), pp. 149-67. Faulkender, Michael. “Hedging or Market Timing? Selecting the Interest Rate Exposure of Corporate Debt,” Journal of Finance, 60 (April 2005), pp. 931-62. Benmelech, Efraim. “Asset Salability and Debt Maturity: Evidence from the 19th Century American Railroads,” Working Paper, Harvard University, (2004). Faulkender, Michael, and Mitchell A. Petersen. “Does the Source of Capital Affect Capital Structure?” Review of Financial Studies, 19, 1 (Spring 2006), pp. 45-79. Berger, Allen N., Marco A. EspinosaVega, W. Scott Frame, and Nathan H. Miller. “Debt Maturity, Risk, and Asymmetric Information,” Journal of Finance, 60, 6 (December 2005), pp. 2895-2923. Flannery, Mark J. “Asymmetric Information and Risky Debt Maturity Choice,” Journal of Finance, 41 (March 1986), pp. 19-37. Brick, Ivan, and Abraham Ravid. “Interest Rate Uncertainty and the Optimal Debt Maturity Structure,” Journal of Financial and Quantitative Analysis, 26 (March 1991), pp. 63-81. www.philadelphiafed.org 10 Q1 2006 Business Review Guedes, Jose, and Tim Opler. “The Determinants of the Maturity of Corporate Debt Issues,” Journal of Finance, 51 (December 1996), pp. 1809-33. Hart, Oliver, and John Moore. “A Theory of Debt Based on the Inalienability of Human Capital,” Quarterly Journal of Economics, 109 (1994), pp. 841-79. Johnson, Shane A. “Debt Maturity and the Effects of Growth Opportunities and Liquidity Risk,” Review of Financial Studies, 16 (Spring 2003), pp. 209-36. Myers, Stewart C. “Determinants of Corporate Borrowing,” Journal of Financial Economics, 5 (1977), pp. 147-75. Stohs, Mark H., and David C. Mauer. “The Determinants of Corporate Debt Maturity,” Journal of Business, 69 (July 1996), pp. 279-312. Graham, John R., and Campbell R. Harvey. “The Theory and Practice of Corporate Finance: Evidence from the Field,” Journal of Financial Economics, 60 (2001), pp. 187-243. www.philadelphiafed.org Business Review Q1 2006 10 What a New Set of Indexes Tells Us About State and National Business Cycles by theodore m. crone M any people are interested in comparing the pattern of economic growth in their state with growth in other states or in the nation. Although the National Bureau of Economic Research sets dates for peaks and troughs of national business cycles, we lack official dates for turning points in state economies. Some states have suffered recessions when the nation did not, and some avoided recessions during some national downturns. In this article, Ted Crone presents information on a recently constructed set of coincident indexes for the 50 states. These indexes can be used to define business cycles at the state level and can tell us how business cycles and the overall patterns of growth have differed among the states. Workers, business owners, and policymakers are typically interested in how the pattern of economic growth in their state compares with growth in other states or in the nation. Often their job prospects, their profits, or their tax revenues are sensitive to the local business cycle. They may want to know if recessions are more frequent Ted Crone is a vice president in the Research Department of the Philadelphia Fed. He is also head of the department’s regional economics section. www.philadelphiafed.org in their state than in other states or if their recessions are more severe or last longer. They may also be interested in how well the information they have about the local economy reflects national conditions. At the national level, we have a commonly accepted definition of business cycles. A committee of the National Bureau of Economic Research (NBER) sets dates for peaks at the end of expansions and troughs at the end of recessions.1 The economies of the individual states, however, do An explanation of the committee’s procedure for determining the dates of business-cycle turning points can be found at www.nber.org/cycles. html. 1 not march in lock-step with the national economy, and there are no official dates for turning points in state economies. A casual glimpse at state economic data reveals that some states have suffered recessions when the nation did not and some states have avoided recessions when the nation was in a downturn. Using a recently constructed set of coincident indexes for the 50 states, we can more clearly define business cycles at the state level.2 We can also learn about the course of the national economy from what is happening in the states. For example, by following the states whose indexes are declining we can trace the spread of national recessions across the country. Finally, by calculating an index based on the number of states in decline versus the number expanding we can get an early signal of national recessions. WHAT IS A BUSINESS CYCLE ANYWAY? The popular notion of a business cycle and the one used by the NBER dating committee goes back to the work of Arthur Burns and Wesley Mitchell. They identified four phases See the article I wrote with Alan ClaytonMatthews. The historical series for these indexes can be found at www.philadelphiafed. org/econ/stateindexes. This article is based on the indexes from 1979 to 2004. A complete set of state indexes is available only from 1979 because some data series needed to construct the indexes are not available before then. For consistency, each state’s index is constructed from the same set of variables. Using a different set of variables for different states could affect the timing and magnitude of changes in the index so comparisons across the states would not be valid. 2 Business Review Q1 2006 11 of the business cycle: an expansion followed by recession and contraction and then a revival of economic activity leading to the next expansion phase. These four phases are commonly collapsed into two periods: a period of growth (revival and expansion) and a period of widespread and significant decline in economic activity (recession and contraction). The NBER dating committee looks at a number of indicators, such as personal income, employment, wholesale and retail sales, and industrial production, when it sets the dates for peaks in the expansion and troughs in the recession. These data are not all available at the state level. But the new state indexes combine several monthly and quarterly data series that are available for all 50 states — nonfarm employment, average hours worked in manufacturing, the unemployment rate, and wages and salaries adjusted for inflation. The indexes represent a composite measure of the underlying “state of the economy” in each of the 50 states, and we use changes in the indexes to define state business cycles. To compare business cycles at the state level with national business cycles, we need a common measure of the underlying “state of the economy.” For this purpose we have constructed a national index of economic activity based on the same economic series as the state indexes. (See A National Index of Economic Activity, pages 2223.) Over the past 25 years, all of the monthly declines in the national index have occurred in unbroken time intervals that we can identify as national recessions. The four periods of decline in this index correspond closely to the four official recessions defined by the NBER. When we refer to national recessions in the remainder of this article, we will be referring to these periods of decline in the national index of economic activity. 12 Q1 2006 Business Review BUSINESS CYCLES DIFFER WIDELY AMONG THE STATES The state indexes do not trace out recessions and expansions as clearly as the national index. During state expansions, the indexes sometimes register a month or two of decline that is neither sharp enough nor long enough to indicate a separate state recession. During state recessions, the indexes sometimes register a month or two of increases that do not indicate the beginning of a recovery. The data at the state level are more volatile than the national data, and single events, have been in recession every time the nation has, and 28 states have not had a recession independently of a national recession. Fifteen states belong to both groups. They have had recessions that correspond to all four national downturns since 1979 and have had no other recessions (Table 1, Column 1).5 Missouri and Pennsylvania are good examples of states whose business cycles follow the national pattern (Figure 1). Recessions in both states have occurred at the same time as in the nation. But the state recessions in Missouri and Pennsylvania have The state indexes do not trace out recessions and expansions as clearly as the national index. such as hurricanes, plant shutdowns, or temporary spikes or declines in demand for a particular product, can affect the state economies more strongly than the national economy. We use the following criteria to define recessions at the state level. The cumulative decline in the state’s coincident index must be at least 0.5 percent, which is the smallest decline in the national index for any recession in the last quarter century. The period from the state index’s peak to its trough must be at least three months.3 Based on these criteria, at least 36 states and as many as 44 states have been in recession during each of the four national recessions since 1979.4 The Number and Timing of Recessions Varies Among the 50 States. Only about half the states (24) These criteria were chosen to meet Burns and Mitchell’s conditions for a recession: The decline in the economy must be diffuse, last a sufficient length of time, and be sufficiently large. 3 Thirty-six states were in recession during the brief national recession in 1980, and 44 states were in recession at some point in the long national recession in 1981-82. 4 been deeper and lasted longer than the national recessions. In part because of the longer and deeper recessions, the average monthly growth in the indexes for these two states has only been about three-quarters as great as the average for the nation (Table 2). Among the 50 states, the average monthly increases in the state indexes have ranged from 1.8 times the U.S. average (Nevada) to slightly more than onethird the U.S. average (Louisiana). Not surprisingly, the states with the highest average economic growth as measured by the change in their indexes also had some of the greatest increases in population (Nevada, Arizona, Georgia, Florida, and Utah), and states with the weakest economic growth had some of the slowest population growth over the past 26 years (Louisiana, West Virginia, Michigan, North Dakota, Ohio, and Iowa). Two states in recession during all four national recessions (Delaware and Illinois) had no recovery between the two national recessions in the early 1980s. Seven states were in recession during three of the four national recessions and had no other recessions (Table 1, Column 2). 5 www.philadelphiafed.org FIGURE 1 Three-Month Change in State Index Percent 3 Missouri 2 2 1 1 0 0 -1 -1 -2 -2 -3 Pennsylvania Percent 3 Jan JanJan JanJanJan JanJan JanJanJan Jan Jan Jan Jan JanJan Jan Jan JanJanJanJan JanJan Jan 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 00 01 02 03 04 National Recessions -3 Jan JanJan JanJanJan JanJan JanJanJan Jan Jan Jan Jan JanJan Jan Jan JanJanJanJan JanJan Jan 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 00 01 02 03 04 3-Month Change in State Index Shaded areas represent periods of decline in the national index described in A National Index of Economic Activity. TABLE 1 States in Recession During Most National Recessions Since 1979 and Experiencing No Other State Recessions States in Recession During All Four National Recessions Since 1979 States in Recession During Three of the Four National Recessions Since 1979 Alabama Georgia California Iowa Delaware* Maine Illinois* Maryland Indiana Rhode Island Kansas Utah Kentucky Virginia Massachusetts Minnesota Missouri Nevada Pennsylvania South Carolina Tennessee Wisconsin *These states had no recovery between the 1980 and 1981-82 recessions. www.philadelphiafed.org While state recessions generally occur around the same time as national recessions, 22 states have had at least one recession that did not correspond to a national recession. Texas is a good example. Figure 2 shows the three-month change in the coincident index for Texas, with periods of decline in the national index shaded in gray. Texas has had three recessions since 1979, but only two of them occurred during national downturns. The third recession in Texas occurred in the mid-1980s when all the major energyproducing states suffered an economic downturn. This was a period of general decline in oil prices.6 At some time between 1984 and 1986, 13 states were in recession, including all nine states with the highest proportion of output (gross state product) in mining and natural resources. These nine states are Wyoming, Alaska, Louisiana, New Mexico, West Virginia, Oklahoma, Texas, North Dakota, and Montana.7 There was a 50 percent decline in the refiners’ acquisition costs of crude oil between the fourth quarter of 1985 and the fourth quarter of 1986. 6 The four other states that suffered a recession in the mid-1980s were Colorado, Idaho, Nebraska, and South Dakota. 7 Business Review Q1 2006 13 Industrial Structure Explains Some of the Differences in the Pattern of Growth Among the States. In general, changes in the indexes for states with a relatively high dependence on natural resources are not highly correlated with changes in the U. S. index, and changes in these states tend to lag changes in the national economy. That is, over the business cycle, the U.S. economy tends to accelerate or decelerate before the economy in these states. To illustrate this point, we calculated the correlations of the change in the national index with the change in the state index for the same month and for each of the six months preceding and following the change in the national index.8 Table 3 shows the highest correlation for each state in this 13-month span and the month in which the highest correlation occurs.9 For the states in the column marked “t,” the highest correlation was between changes for the state and the nation in the same month. Almost half the states (23) had their highest correlation with the contemporaneous change in the national index. The columns to the left of “t” show states with the highest correlations between the national changes and previous months’ changes in the states. Thus, for Arkansas the highest correlation (0.73) was between the The correlation of the change in the indexes measures the degree of co-movement between changes in the state index and the national index. The correlations will not be affected by differences in trend growth. 8 The correlations at neighboring leads and lags are often very similar, but the correlations continually decline as one moves farther away from the lead or lag with the highest correlation. The correlations for Alaska were negative for all the leads and lags in the 13-month span we considered. For Alaska, we report the correlation closest to zero during that time span. Alaska’s business cycles have not been in sync with the national cycles; in terms of timing the closest positive correlation is between the change in the national index and the change in the state index 26 months prior. 9 14 Q1 2006 Business Review change in the national index and the change in the state index two months earlier. In other words, Arkansas’ growth leads that of the nation. The columns to the right of “t” show states with the highest correlations between the changes at the national level and future months’ changes for the states. For example, the highest correlation with the change in the Illinois state index (0.84) was one month after the change in the national index. We used the proportion of state output, or gross state product, for nine different sectors to estimate the effect of industrial structure on the timing of changes in each state’s economy. Other things equal, economic growth in states with higher percentages of output in agriculture and construction tends to lead growth in the nation. The opposite is true for states with TABLE 2 Average Monthly Increase in State Indexes 1979 to 2004 (Average Increases in U.S. Index = 0.24) State Nevada New Hampshire Arizona Georgia Florida Idaho Oregon Utah North Carolina Colorado California Massachussetts New Mexico Washington Delaware South Carolina Vermont Virginia Texas New Jersey Minnesota Tennessee Connecticut Maryland Rhode Island Average Monthly Growth in State Index 0.44 0.40 0.40 0.35 0.34 0.34 0.32 0.32 0.31 0.30 0.30 0.30 0.29 0.29 0.29 0.29 0.29 0.28 0.28 0.28 0.27 0.27 0.27 0.25 0.24 State Maine South Dakota Arkansas Wisconsin Kentucky Nebraska Alabama Indiana Mississippi New York Missouri Pennsylvania Kansas Illinois Hawaii Iowa Wyoming Ohio North Dakota Michigan Oklahoma Montana West Virginia Alaska Louisiana Average Monthly Growth in State Index 0.23 0.23 0.23 0.23 0.21 0.21 0.21 0.20 0.20 0.19 0.19 0.18 0.18 0.18 0.17 0.17 0.17 0.17 0.14 0.14 0.14 0.13 0.12 0.11 0.09 www.philadelphiafed.org FIGURE 2 Three-Month Change in Texas State Index Percent 3 2 1 0 -1 -2 -3 Jan 1979 Jan 1981 Jan 1983 Jan 1985 Jan 1987 Jan 1989 Jan 1991 Jan 1993 Jan 1995 Jan 1997 Jan 1999 Jan 2001 Jan 2003 Shaded areas represent periods of decline in the national index described in A National Index of Economic Activity. higher percentages of output in mining and natural resources and in wholesale trade.10 Thus, differences in industrial structure help explain differences in the timing of growth among the states. INFORMATION FROM THE STATES ADDS TO OUR UNDERSTANDING OF THE NATIONAL ECONOMY The 50 state indexes in this article were designed to track economic conditions at the state level, We used a statistical technique known as an ordered probit to estimate the extent to which the industrial structure of the states influenced the timing of changes in their economic activity relative to the nation. States were given values between -2 and +6 based on the timing of the highest correlation of the change in their indexes with the change at the national level reported in Table 3. The other sectors included in the ordered probit analysis were manufacturing, transportation and utilities, financial services, other services, and retail trade. None of these had a significant effect on the timing of changes in the states’ economies. The government sector was omitted. 10 www.philadelphiafed.org and we use them to define state business cycles. But they are also useful in describing the geographic scope of national expansions and contractions, and they provide information about the near-term outlook for the national economy. Changes in the State Indexes Track the Geographic Progression of National Recessions and Recoveries. The maps in Figure 3 show the progression of the most recent recession and recovery. The shading indicates which states experienced an increase, a decrease, or no change in their economic activity indexes at three-month intervals between March 2001 and June 2002—a period that spans the recession and early recovery. In March 2001 declines were concentrated in a limited number of states, mostly in the Midwest and South. By September, the recession had spread to almost every state in the nation, and most remained in recession through the end of 2001. By March 2002, however, most of the states in the West, Rocky Mountains, and Southeast were in recovery. By June 2002 almost all the states were in recovery.11 The other three national recessions since 1979 developed in a similar manner. Declines began in a relatively small group of states, most of them in one or two geographic areas. The initial group of states has not always been the same, but 12 states have consistently gone into recession before the nation, even if they have not been in the initial group.12 Eventually, the economic decline spreads to almost all of the states and practically every section of the country.13 Once the national recession is over, the number of states in decline drops quickly to just a few, and most states enter their recovery phase. A Diffusion Index Summarizes the Pattern of Growth and Decline Among the States. The geographic Researchers at the St. Louis Fed have also used the 50 state indexes with a slightly different definition of state recessions to illustrate the geographic spread of the four national recessions since 1979. See the article by Michael Owyang, Jeremy Piger, and Howard Wall. The authors use the state indexes in what is known as a regime-switching model to estimate whether a state is in the recession or expansion phase of the business cycle in every quarter from 1979 to 2002. In my 2005 article I used similarities among the cyclical components of these state indexes to redefine economic regions in the U.S. 11 The 12 states are California, Indiana, Massachusetts, Michigan, Missouri, North Carolina, Ohio, Oregon, Pennsylvania, South Carolina, Tennessee, and Washington. The lead times have varied, however, for each state and each recession. This pattern is not likely to be the result of mere chance. If, for every national recession, each state had a 50 percent chance of going into recession before the nation, the probability of a state going into recession early all four times since 1979 would be 0.0625 = (0.5)4. This would imply that only about three states, not 12, would have entered every recession since 1979 before the nation as a whole. 12 13 The 1990-91 recession was somewhat different. Even though most states went into recession, many in the Southwest and Rocky Mountain regions did not. Business Review Q1 2006 15 TABLE 3 Highest Correlation of Change in the State Index With Change in the National Index* Period Relative to U.S. (= t) Number of States t-2 t-1 t t+1 t+2 t+3 t+4 t+5 t+6 7 13 23 2 1 1 1 1 1 LA (0.31) OK (0.31) WY (0.23) AR (0.73) GA (0.83) AL (0.73) IL (0.84) WV (0.63) TX (0.55) DE (0.72) IN (0.74) AZ (0.81) UT (0.71) ID (0.54) MD (0.77) CA (0.65) MT (0.34) ME (0.71) CO (0.56) OR (0.68) MI (0.72) CT (0.75) SD (0.52) MO (0.83) FL (0.77) WA (0.74) MS (0.77) HI (0.21) NE (0.67) IA (0.72) NH (0.71) KS (0.71) NV (0.69) KY (0.78) OH (0.77) MA (0.70) RI (0.67) MN (0.79) SC (0.84) NC (0.84) ND (0.35) NJ (0.76) NM (0.64) NY (0.80) PA (0.82) TN (0.84) VA (0.80) VT (0.72) WI (0.78) AK (-0.15) * The correlation indicates the degree to which the change in the state’s index moves with the change in the national index. For example, if the highest correlation occurs at t-1, this means that economic growth or decline in the state precedes growth or decline at the national level. 16 Q1 2006 Business Review www.philadelphiafed.org FIGURE 3 March 2001 June 2001 Decreased No Change Increased September 2001 Decreased No Change Increased December 2001 Decreased No Change Increased March 2002 June 2002 Decreased No Change Increased www.philadelphiafed.org Decreased No Change Increased Decreased No Change Increased Business Review Q1 2006 17 dispersion of national recessions and expansions like that shown in Figure 3 can be summarized by a diffusion index of the 50 states. This index is simply the percentage of states in which the economy is expanding minus the percentage in which it is declining.14 Diffusion indexes can be calculated using changes over any interval of time, although one-, three-, and six-month changes in the indexes are the most common. Figure 4 presents the one-month and three-month diffusion indexes for the 50 states from 1979 to 2004. The one-month diffusion index represents the percentage of states whose indexes have increased in the last month minus the percentage whose indexes have declined. The three-month diffusion index represents the percentage of states whose indexes have increased over the most recent three-month period minus the percentage of states whose indexes have declined over that period. These indexes do not measure the magnitude of the change but only the scope of change across the states. The degree of increase or decrease in a state’s index does not affect the diffusion index. Diffusion indexes are commonly used to measure the breadth of a downturn or of an expansion in the overall economy or in a particular sector.15 For example, the Bureau of Labor Statistics (BLS) produces a diffusion index for payroll employment, and the Federal Reserve Board produces one for industrial production by subtractReported percentage changes in these indexes, like changes in most statistical series, are rounded to the first decimal place. Thus, any change less than 0.05 percent in either direction is recorded as no change. 14 Diffusion indexes are also a standard way to summarize the responses to qualitative surveys in which respondents are asked whether some aspect of their business has increased, decreased, or remained unchanged. See the article by Michael Trebing and the OECD handbook. 15 18 Q1 2006 Business Review ing the percentage of subsectors that is declining from the percentage that is increasing.16 These types of indexes received considerable attention in the 1950s when Geoffrey Moore argued that they could be used as leading indicators because they tend to decline before the aggregate series in economic downturns and rise before the aggregate series in recoveries.17 But this is not The diffusion index of the 50 state economies differs in a significant way from these two. The state indexes are not components of the national index. The national index is estimated separately; it is not the sum or a weighted average of the 50 states, as in the case of employment and industrial production. 16 See the two articles by Moore. The suggestion that diffusion indexes decline before the peak in the aggregate series and rise before the trough is only a historical observation, not a statement about the mathematical properties of diffusion indexes. See the article by Stefan Valavanis. true for all diffusion indexes. Patricia Getz and Mark Ulmer compared turning points or peaks and troughs in the diffusion indexes for total private employment and for manufacturing employment to turning points in the two overall series from 1977 to 1989. They found some evidence that turning points in the diffusion index for manufacturing employment signaled turning points in overall manufacturing employment, but they found no such evidence for total employment.18 James Kennedy at the Federal Reserve Board examined break-even or reference points in the diffusion index for industrial production, that is, points at 17 The authors examined the relationship between the diffusion indexes and the levels of employment. But the logical comparison is with growth rates. See the article by Arthur Broida and the one by H.O. Stekler. 18 FIGURE 4 One-Month and Three-Month Diffusion Indexes for the 50 States 120 100 80 60 40 20 0 -20 -40 -60 -80 Aug 1979 Aug 1981 Aug 1983 Aug 1985 Aug 1987 Recessions Aug 1989 Aug Aug 1991 1993 Three-Month Aug 1995 Aug 1997 Aug 1999 Aug 2001 Aug 2003 One-Month Shaded areas represent periods of decline in the national index described in A National Index of Economic Activity. www.philadelphiafed.org which the number of components that were increasing equaled the number decreasing. These break-even points rarely preceded turning points in industrial production, and more often than not, they lagged the turning points in the overall index. Our Diffusion Indexes of Economic Activity in the 50 States Do Better as Predictors of National Recessions. The one-month diffusion index has turned negative before the decline in the national index in all four recessions since 1979, with lead times of one to four months (Figure 4). The three-month diffusion index has not provided as much lead time as the one-month index. In three of the four recessions it turned negative between one and three months before the national index. In the other recession, it turned negative in the same month as the decline in the national index.19 The ability of a diffusion index to predict a coming recession can be formalized with a statistical model that uses the index to predict the probability of being in recession in the near future. It is obvious from Figure 4 that recessions are preceded by low readings of the diffusion index and by sharp declines in the index. We used the three-month diffusion index and the three-month change in that index to predict the probability of being in a national downturn three months in the future (Figure 5).20 We would expect that when the probability climbs above 50 percent, the nation would be in a recession sometime in the near future. Indeed, the probability climbs above 50 percent before every national recession, with a lead time of one to four months. Moreover, there has been no occasion since 1979 when the probability climbed above 50 percent and the nation did not go into recession. At the end of recessions, the model’s record of predicting recoveries is good but not perfect. Before the end of every recession except the one in 1980, the probability of being in recession in the near future drops below 50 percent. After the 1980 recession, the probability dropped below 50 percent in the first month of the recovery. Diffusion Indexes Also Contain Information on the Course of the National Economy Beyond Turning Points. In his study of industrial production, James Kennedy found that the diffusion index provided valuable information for forecasting near-term growth in industrial production. We repeated Kennedy’s exercise with the one-month and three-month diffusion indexes for the states and the monthly change in the national index. We got results similar to Kennedy’s. (See Information in the Diffusion Indexes about Changes in the National Index, page 24.) Past changes in the national economic activity index provide information about the current month’s change. If we add past values of the diffusion index for the 50 states, we get a better estimate of the current month’s change in the national index. Thus, the diffusion index of the 50 states not only confirms the information in the national index, but it also provides independent information about the future course of the national economy. FIGURE 5 Probability of Being in a National Recession in Three Months Based on the Three-Month Diffusion Index for the 50 States* 1 0.9 0.8 0.7 0.6 0.5 Both the one-month and the three-month diffusion indexes had a one-month negative reading in early 2003 that was not followed by a recession. This negative reading may have been associated with the uncertainty surrounding the buildup to and the beginning of the war in Iraq. 19 We estimated the probability with a standard probit model. See the article by Andrew Filardo for the use of probit and other types of models to predict recessions. The one-month diffusion index and its three-month change send signals of recession using the probit model, but the signals are somewhat weaker. The one-month change also produces a false signal in February 2003 when the recession probability was slightly above 50 percent. 20 www.philadelphiafed.org 0.4 0.3 0.2 0.1 0 Jan 1979 Jan Jan 1981 1983 Jan Jan Jan Jan 1985 1987 1989 1991 Jan Jan Jan 1993 1995 1997 Jan 1999 Jan Jan 2001 2003 Shaded areas represent periods of decline in the national index described in A National Index of Economic Activity. * The probability is based on the three-month diffusion index for the 50 states and the three-month change in that index. Business Review Q1 2006 19 THE VIEW FROM THE STATES: A FULLER PICTURE OF REGIONAL AND NATIONAL BUSINESS CYCLES The new indexes for the 50 states were developed as summary measures of state economic conditions. They provide valuable information not only about the economies of the individual states but also about the national economy. The indexes help us identify state business cycles. We can compare the state cycles with national cycles in terms of their timing and severity, and we can compare business cycles across states. The state indexes also allow us to track the geographic development of national recessions and recoveries. 20 Q1 2006 Business Review Furthermore, diffusion indexes for the 50 states can signal the near-term onset of a national recession. This ability to forecast recessions is formalized in a model that predicts recession probabilities rather accurately. Furthermore, the diffusion indexes contain information about the course of the national economy beyond these turning points; they provide independent information about the next month’s increase in the national index. More than a half century ago, Arthur Burns and Wesley Mitchell argued that we should look at a large number of indicators when judging the condition of the U. S. economy. The NBER dating committee looks at a number of national series to set the dates for recessions and expansions, but they do not determine these dates until recessions or expansions are well underway. The new state indexes add another set of indicators for researchers and economic forecasters to look at. The individual state indexes and the diffusion indexes for the 50 states are available within a month of the time the data are collected. The indexes can confirm the information in the national data that are available at the time; they can illustrate the breadth of expansions and recessions; and they can provide valuable information about the near-term course of the national economy. BR www.philadelphiafed.org REFERENCES Broida, Arthur L. “Diffusion Indexes,” American Statistician, 9, 3 (June 1955), pp. 7-16. Burns, Arthur F., and Wesley C. Mitchell. Measuring Business Cycles. N.Y.: NBER, 1946. Crone, Theodore M. “The Long and the Short of It: Recent Trends and Cycles in the Third District States,” Federal Reserve Bank of Philadelphia Business Review (Third Quarter 2003), pp. 29-37. Crone, Theodore M. “An Alternative Definition of Economic Regions in the United States Based on Similarities in State Business Cycles,” Review of Economics and Statistics (November 2005), pp. 617-626. Crone, Theodore M., and Alan Clayton-Matthews. “Consistent Economic Indexes for the 50 States,” Review of Economics and Statistics (November 2005), pp. 593-603. Filardo, Andrew J. “How Reliable Are Recession Prediction Models?” Economic Review, Federal Reserve Bank of Kansas City (Second Quarter 1999), pp. 35-55. www.philadelphiafed.org Getz, Patricia M., and Mark G. Ulmer. “Diffusion Indexes: A Barometer of the Economy,” Monthly Labor Review (April 1990), pp. 13-21. Rudebusch, Glenn D. “Has a Recession Already Started?,” Federal Reserve Bank of San Francisco Economic Letter 2001-29 (October 19, 2001). Kennedy, James E. “Empirical Relationships between the Total Industrial Production Index and Its Diffusion Indexes,” Finance and Economics Discussion Series No. 163, Federal Reserve Board, Washington, D.C., July 1991. Stekler, H.O. “Diffusion Index and First Differencing,” Review of Economics and Statistics, 43 (1961), pp. 201-208. Moore, Geoffrey H. “Analyzing Business Cycles,” American Statistician, 8, 2 (April-May 1954), pp. 13-19. Moore, Geoffrey H. “Diffusion Indexes: A Comment,” American Statistician, 9, 4 (October 1955), pp. 13-17, 30. Organization for Economic Cooperation and Development. Business Tendency Surveys: A Handbook. Paris: OECD, 2003. Owyang, Michael T., Jeremy Piger, and Howard J. Wall. “Business Cycle Phases in U.S. States,” Review of Economics and Statistics (November 2005), pp. 604-30. Stock, James H., and Mark W. Watson. “New Indexes of Coincident and Leading Economic Indicators,” NBER Macroeconomics Annual, 1989, pp. 351-94. Stock, James H., and Mark W. Watson. “Business Cycle Fluctuations in US Macroeconomic Time Series,” Handbook of Macroeconomics, Vol I. NY: Elsevier, 1999, pp. 3-64. Trebing, Michael E. “What’s Happening in Manufacturing: ‘Survey Says . . .’,” Federal Reserve Bank of Philadelphia Business Review (September/October 1998), pp. 15-29. Valavanis, Stefan. “Must the Diffusion Index Lead?” American Statistician, 11, 4 (October 1957), pp. 12-16. Business Review Q1 2006 21 A National Index of Economic Activity I hours worked, the change in the unemployment rate, and the change in real wages and salaries. ∆ct is the change in the log of the unobserved state of the economy or the coincident index that is to be estimated. The trend in the index is set to equal the trend in total gross state product for the 50 states. The figure below shows the monthly percentage change in the national coincident index estimated from this model. There have been four periods since 1979 when changes in the national index were negative for several consecutive months; these four periods correspond closely to the four official recessions since 1979. Except for these four periods there has never been a monthly decline in the index. Thus, we can use the index to designate a set of business-cycle peaks and troughs for the national economy. The cumulative declines in the index from peak and trough have ranged from 0.5 percent (1990-91) to 1.8 percent (1981-82). n the late 1980s James Stock and Mark Watson developed a coincident index for the U.S. economy by identifying a common unobserved factor underlying the observed measures of economic activity for the nation.a The U.S. index in this paper is estimated by a Stock/Watson type model using national data that are also available at the state level.b Thus, the model produces a national index comparable to the state indexes. The Stock/Watson type model is commonly referred to as a single dynamic factor model and is based on the following set of equations. For each of the observed variables: ∆xt = a + b ∆ct + ut For the unobserved state of the economy: ∆ct = d + f ∆ct-1 + g ∆ct-2 + et In the model developed for this article, ∆xt is the change in the log of employment, the change in the log of average Figure One-Month Change in National Index (Shaded Areas Represent Recessions Based on NBER Dates) Percent change 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 Jan 1979 Jan 1981 Jan 1983 Jan 1985 Jan 1987 Jan 1989 Jan 1991 Jan 1993 Jan 1995 Jan 1997 Jan 1999 Jan 2001 Jan 2003 1- month See the article by Stock and Watson. The Stock and Watson coincident index was developed as an alternative to the coincident index currently published by the Conference Board. The NBER no longer publishes the alternative Stock and Watson index; it was discontinued at the end of 2003. a b This is the same model referred to in my 2003 Business Review article. 22 Q1 2006 Business Review www.philadelphiafed.org A National Index of Economic Activity (continued) The table below compares the peaks and troughs based on this national index to the peaks and troughs designated by the NBER dating committee. The peaks in the business cycle based on the index are one to two months later than the official dates set by the NBER dating committee. The troughs of the recession are either simultaneous with the NBER dates or two months later. This difference in timing between the index and the NBER dates may be due to the fact that the timing of the index is primarily based on the timing of nonfarm employment, which slightly lags the overall economy.c Despite the differences in timing, the index contains valuable information about national recessions and expansions. The NBER typically announces the end of an expansion or recession five months or more after the end has occurred because it wants to make sure that a new phase of the business cycle has begun and that the original data do not simply represent the normal variation in the series. Therefore, any information before the announcement can be helpful in evaluating the state of the national economy.d TABLE Business-Cycle Peaks and Troughs 1979-2004 Coincident Index of U.S. Economic Activity NBER Business Cycle Dates NBER Announcement Dates Peak March 1980 January 1980 June 1980 Trough July 1980 July 1980 July 1981 % change in the index -0.7% Peak Trough % change in the index Peak Trough % change in the index Peak Trough % change in the index August 1981 July 1981 November 1982 November 1982 January 1982 July 1983 -1.8% September 1990 May 1991 July 1990 March 1991 April 1991 December 1992 -0.5% May 2001 March 2001 November 2001 January 2002 November 2001 July 2003 -0.6% c See Stock and Watson’s article. Each of the other variables besides employment contributes significantly to the estimation of the coincident index for the nation. This is also true for most of the state indexes. d See the article by Glenn Rudebusch for an illustration of the difficulty of timing recessions. www.philadelphiafed.org Business Review Q1 2006 23 Information in the Diffusion Indexes about Changes in the National Index I n his examination of the diffusion indexes for industrial production, James Kennedy tested whether the diffusion indexes provide any independent information about future changes in overall industrial production. He estimated a regression of the one-month change in industrial production on 12 lags of changes in industrial production and 12 lags of various diffusion indexes for industrial production. He found that the lags of the diffusion indexes provided information beyond that found in past changes in industrial production itself. We repeated Kennedy’s experiment with the national economic index and the one-month and three- month diffusion indexes of the 50 states. The results of our regressions are found in the table below. A standard statistical test (an F-test) confirms that the 12 lags of the diffusion indexes add information to past changes in the national index that helps predict the current change in the national index. The statistics reported in the table are based on equations of the following form: 12 12 i=1 i=1 ∆ln(US)t = α + ∑ ßi ∆ln(US)t-i + ∑ γi DIFFt-i where (US)t is the national index of economic activity described in this article and DIFF is either the one-month or three-month diffusion index for the 50 states. table Results of Regression of One-Month Change in the National Index on 12 Lags in the Change in the National Index and 12 Lags in the Diffusion Indexes for the 50 States* Adjusted R-squared (R2) Probability that all the coefficients on the lags of the diffusion index equal zero (based on an F-test) 12 lags of changes in the national index 0.89 — 12 lags of changes in the national index and 12 lags of the one-month diffusion index 0.91 < 0.001 12 lags of changes in the national index and 12 lags of the three-month diffusion index 0.92 < 0.001 Dependent variables in the equation * The adjusted R2 is a measure of the goodness of fit of the regression. A higher R2 for the estimated equations with the diffusion indexes indicates a better fit with the inclusion of these variables. The probability measure from the F-test indicates that the improvement in the R2 is statistically significant. 24 Q1 2006 Business Review www.philadelphiafed.org Your House Just Doubled in Value? Don’t Uncork the Champagne Just Yet! by wenli li and rui yao M ore and more the United States is becoming a nation of homeowners. Along with this rise in ownership, an increasing share of households’ wealth is invested in housing. However, house prices fluctuate over time. Some studies offer evidence that changes in house prices have had a large effect on total output and total consumption. In this article, Wenli Li and Rui Yao present their recent research, which tries to quantify the effects of houseprice changes on both consumption and the well-being of American households. Their study looks at the economy as a whole, as well as different demographic groups. The United States is increasingly becoming a country of homeowners. As reported recently by the Census Bureau, close to 70 percent of households now own their primary residence. Homeownership is no longer just an American dream; watering lawns, sweeping sidewalks and cleaning drain gutters is no longer the sole privilege of middle-income and affluent households. With the rise in the homeownership rate, an increasing share of household wealth is tied to Wenli Li is an economic advisor and economist in the Research Department of the Philadelphia Fed. www.philadelphiafed.org housing.1 According to the Federal Reserve Board’s Flow of Funds account, residential property accounts for over 30 percent of total household assets, and home equity accounts for over 20 percent of total household net worth. House prices, however, fluctuate over time (Figure 1). As can be seen in the figure, since 1975 the country has had two episodes of prolonged negative returns (adjusted for inflation) on housing: one from the late 1970s to early 1980s and the other in the early 1990s. Since 1996, however, the return on housing has been moving up steadily, exceeding 15 percent in real terms in the second quarter of 2004. The rise of housing wealth’s share in total household net worth started in 2000 after the stock bubble burst. 1 There has been some speculation and some evidence that house-price movements have had a large impact on total output and total consumption. For instance, Global Economic Research wrote in its recent publication that “the U.S. housing market has been a significant driver of U.S. economic growth in recent years and has contributed more to GDP than many analysts expected. We estimate that housing accounted for 10 percent to 15 percent of the growth in real GDP in 2004, more than double its normal share in the economy.”2 Some academic researchers have found that house-price appreciation also has a sizable effect on consumption. Economists Karl Case, John Quigley, and Robert Shiller find that an additional dollar of housing wealth increases total household consumption by 3 to 15 cents. Similarly, John Benjamin, Peter Chinloy, and Donald Jud find that housing wealth increases household consumption by 8 cents. In their recent working paper, Fed Chairman Alan Greenspan and Federal Reserve Board staff economist James Kennedy estimate that cashouts through mortgage refinancing and net extensions of home equity loans less 2 See the article by Joseph Carson. Rui Yao is an assistant professor of economics and finance, Zicklin School of Business, Baruch College, New York City. Business Review Q1 2006 25 unscheduled repayments amounted to about $300 billion between 2003 and 2004, or roughly 13 percent of the increase in the household real estate assets during that same period as reported in the U.S. flow of funds account. Our recent research tries to quantify the effects of changes in house price on both consumption and the well-being of American households, both in the economy as a whole and across different demographic groups.3 We find that the effects are small for the economy as a whole, but they vary substantially across households of different ages, homeownership status, and amount of assets. WHAT’S SPECIAL ABOUT HOUSING AND HOUSING WEALTH GAINS? As discussed in a previous Business Review article, residential housing 3 See our working paper. is unique because it combines a flow of services with an investment good.4 The homeowner gets to live in the house in lieu of renting and receives a potential return on the equity in the house. As of yet, there is no financial product that permits a complete separation of these two functions of residential houses—consumption and investment—for homeowners. To put it simply, when households live in their own house, they necessarily bear the risk that the value of their house will fluctuate over time. In that sense, owner-occupied housing is also an investment. In principle, renting would accomplish this separation. More precisely, households can live in a rented house and then invest in other properties. But there are good reasons why most renters wish to become homeowners. First, rents may rise. As economists Todd Sinai and Nicholas Souleles have argued, owning 4 See Wenli Li’s Business Review article. FIGURE 1 Real Rate of Return of House Prices Percent 20 15 10 5 0 -5 -10 19 7 19 52 7 19 62 7 19 72 7 19 82 7 19 92 8 19 02 8 19 12 8 19 22 8 19 32 8 19 42 8 19 52 8 19 62 8 19 72 8 19 82 8 19 92 9 19 02 9 19 12 9 19 22 9 19 32 9 19 42 9 19 52 9 19 62 9 19 72 9 19 82 9 20 92 0 20 02 0 20 12 0 20 22 0 20 32 04 2 -15 one’s home provides insurance against the risk that rents will rise by purchasing future housing services at today’s price. Second, housing services are, in general, cheaper for homeowners, since renters have few incentives to maintain the rental unit and landlords charge a higher price to cover for possible damages. Finally, homeowners also get tax breaks. They can deduct their interest payment on mortgage loans on first and second homes from their income tax, as long as these loans total less than $1.1 million, and they do not pay tax on the implicit income they receive from being their own landlords. Housing wealth is less liquid than many other types of financial wealth. The conventional fee for selling one’s house is 6 percent of the house value. And the time and effort involved far exceeds a trip to the ATM or a call to your mutual fund company or broker. In addition, saving in the form of housing exposes households to idiosyncratic (household-specific) risks, unlike the well-diversified portfolio of financial assets in most models of lifetime savings and consumption used by economists. For example, if employers relocate, households may need to move at a time when they are unwilling or unable to sell their houses.5 These differences suggest that economists may need to modify their models of savings and consumption in a world populated by homeowners. The reason is three-fold. First, in traditional models, there is a sharp distinction between consumption and savings, but housing is both a consumption good and a savings vehicle. Second, in traditional models, consumers save time annualized real house price return, quarter over quarter In the short run, as demand for housing goes up, rents may come down. But eventually, rents will go up as property value goes up. Since we all need to live somewhere, Sinai and Souleles argue that the actual idiosyncratic risk of owning a house is somewhat lower because ownership hedges against the risk of rents rising. 5 Note: The rates of return are calculated using the house price index, a weighted repeated sales index constructed by the Office of Federal Housing Enterprise Oversight, deflated by the core consumer price index (which excludes the often volatile food and energy prices) from the Bureau of Labor Statistics. The core CPI compares prices for a fixed list of goods and services to a base period. Currently, the base, which equals 100, is average prices in the period 1982-84. Returns adjusted for inflation are called real returns. 26 Q1 2006 Business Review www.philadelphiafed.org to insure against a rainy day, so-called precautionary savings. These models do not take into account the transaction cost of using past savings to finance current consumption needs, as is the case with accessing home equity. Third, in traditional models, savings decisions are not affected by constraints on a household’s wealth. Current institutional arrangements, however, often constrain homeowners from borrowing a mortgage that exceeds 80 percent of the house value unless private mortgage insurance is purchased.6 The new set of models that economists have developed in recent years to deal with housing issues explicitly model households that plan over their life cycle and make housing decisions along two dimensions: renting versus owning, and the size of the house for those who choose to own. These households also make mortgage decisions and other financial investment decisions, such as buying stocks and bonds. Notably, these households face borrowing constraints that preclude them from borrowing today against their future income. Researchers have used variations of such models to study, among many other things, households’ optimal mortgage choice, portfolio decisions in the presence of housing, and the cost of subsidizing mortgage interest payments.7 Our paper is the first to study directly the consumption and welfare consequences of changes in house prices using a Financial innovations have relaxed, but not abolished, the down payment requirement for buying a house. 6 See, respectively, the articles by John Campbell and Joao Cocco; the article by Joao Cocco and the one by Rui Yao and Harold Zhang; and the article by Martin Gervais. Patrick Bajari, C. Lanier Benkard, and John Krainer also study the welfare implications of house-price changes but in a complete market setting with perfect information. 7 www.philadelphiafed.org model in which households explicitly maximize their lifetime welfare.8 HOUSE PRICES AND CONSUMPTION Middle-Aged Homeowners’ Consumption Is Not Sensitive to Changes in House Price. Contrary to many accounts in the popular and business press, some economists, most prominently Ed Glaeser, have argued that changes in housing price should have little impact on the consumption and welfare of an average homeowner, even if housing is a large share of the average individual’s wealth. The rea- who has lived for some years in Manhattan. He has probably reaped an enormous capital gain from his house, but he would also find it very expensive to buy another house in Manhattan if he decides to sell his current house to capture the capital gains. However, this logic holds only if today’s house is much like tomorrow’s house, i.e., the homeowner lives in the same house, as is the case with an average household: a middle-aged homeowner who has accumulated adequate savings to buy his desired house and who has yet to retire to Palm Beach, Florida. For these households, Contrary to many accounts in the popular and business press, some economists have argued that changes in housing price should have little impact on the consumption and welfare of an average homeowner. son is that when house prices increase, the assets of the household that owns a home obviously increase. At the same time, however, the price of housing services—the price the household would have to pay in order to rent the same house—has also increased, thus offsetting that gain. The household is not richer, since housing expenses appear on both sides of the household’s balance sheet. Imagine a homeowner Economists refer to this class of models—which have become standard in modern macroeconomics—as dynamic stochastic equilibrium models. Every modeling exercise requires simplifying assumptions to keep things manageable. To single out the impact of house-price changes, we assume that household income does not change with house prices. In reality, however, house prices and household income are positively correlated—that is, they tend to rise or fall together—especially in a local market. Our argument remains true in this new environment, with the added effect that the changes in income will reinforce the impact of changes in house price, in both an economic upturn and a downturn. 8 the increase in housing wealth cancels out the increase in the price of their housing services. The net cancellation, however, does not accurately reflect the decisions faced by homeowners who are either young or old, nor those faced by renters. Young Homeowners’ Consumption Is More Sensitive to Changes in House Price. Young households’ main wealth lies in their future earnings, which are not easy to borrow against. To a significant extent, young households’ housing choices are limited by the down payment they can afford. Economists say that such homeowners are borrowing constrained. House-price appreciation effectively increases their assets and relaxes their borrowing constraint. Since they now have more home equity, young homeowners typically respond to the increase in house price by cashing out some of the home equity through mortgage refinancing so that they can spend the money on Business Review Q1 2006 27 other consumption goods. Imagine a young couple who owns a $150,000 home with 20 percent equity ($30,000). The household does not have any other financial wealth, and it lives on a budget. Suppose that the price of the house increased 10 percent. The household now has $15,000 more in home equity. The couple may decide to cash out $900—through refinancing, home equity loans, or second mortgages—and buy that big TV they wanted but couldn’t afford. In other words, the 10 percent appreciation in house price increased their nonhousing spending by $900. It is worth pointing out that the consumption consequences of houseprice appreciation depend on the extent of house-price increases and their persistence. For example, the initial house-price appreciation eases young homeowners’ borrowing constraint and thus increases their consumption. Once these young homeowners overcome the liquidity problems, future house-price appreciation will not help them much in that regard. Furthermore, young households will respond more to more persistent house-price changes, given the transaction cost. Old Homeowners’ Consumption Is Also More Sensitive to Changes in House Price. Since old homeowners face a relatively short remaining life span, they are less concerned about long-term consumption. The risk of fluctuating rents in the future is not as important to them as it is to middleaged and young homeowners. In other words, the increase in the price of the house they own increases their wealth more than it does the cost of housing consumption over their remaining lives. Therefore, they are more likely to sell their houses or downsize when house prices increase in order to capture the wealth gains. As a result, their nonhousing consumption, for example, travel, increases correspondingly. 28 Q1 2006 Business Review Consider an 80-year-old couple who, according to the National Center for Health Statistics, have a life expectancy of another five to six years. Suppose the couple owns a $100,000 house with 100 percent equity. Suppose the price of their house rises 10 percent. The couple may decide to sell the house and get $110,000—$10,000 more than they would have gotten otherwise—and then rent a similar house at an annual cost of 6 percent of the house value. If house prices and The consumption consequences of house-price appreciation depend on the extent of house-price increases and their persistence. rents stay unchanged for the next five years and assuming a zero discount rate, the couple gains $110,000 by selling the house, pays $33,000 for renting a similar house for five years (=$110,000*0.06*5), and still winds up with $77,000 in cash.9 With the additional cash in hand, the couple can go on a cruise to celebrate their anniversary, a choice that may have been too expensive for them before, instead of waiting till the end of the fifth year to cash out the $110,000 from the house. Having said this, the couple does have to sell the house to get the cash right now. Therefore, their balance sheet may look worse than before, a price they pay to obtain the liquidity. Of course, this effect on old homeowners may be weakened to the extent that old people may prefer Obviously, the longer the couple needs to rent, the smaller will be the financial benefits of selling their house now to capture the capital gains. 9 leaving a house to their heirs rather than leaving money, since a house has more sentimental value for family members than for strangers who buy the house. Renters Reduce Consumption When House Price Increases. Since rental housing and owner-occupied housing are substitutes, their prices typically move in the same direction.10 Thus, if house prices increase, rents are also likely to increase. As a result, renters may reduce their consumption in order to pay the higher rent and increase their savings for a down payment on a house. Consider a renter who rents a $100,000 house and pays $6,000 a year for rent. Suppose the house appreciates 10 percent, to $110,000, and the landlord adjusts the rent accordingly. The renter ends up paying $600 more a year for renting the same house. This extra money will have to come from someplace; the renter will have to cut his current consumption or dip into his (existing) savings. If the renter still plans to buy a house, he will have to cut current consumption in order to continue living in the same house and still save for the down payment on a future house, which has also gotten more expensive. When house prices rise sufficiently, renters may decide that they may never have enough wealth to buy a house. As a result, renters may stop saving for a down payment. Nevertheless, they still need to cut current consumption to pay for the higher rents if they choose to stay in the same house. Summarizing the Demographic Effects. In our working paper, we Substitutes are goods purchased in place of another good when its price rises. For example, if the price of houses goes up, the demand for rental units may rise and so rents would rise. If the price of houses falls, the demand for rental units may decline, so rents fall. 10 www.philadelphiafed.org find that an additional dollar of housing wealth raises the consumption of young homeowners (those in their twenties and early to mid-thirties) by 5 to 6 cents, middle-aged homeowners (those in their late thirties to mid-sixties) by 4 cents, and old homeowners (those in their mid-sixties and above) by 8 cents (Figure 2). These are called the marginal propensities to consume out of housing wealth. These numbers are largely in line with those found in recent empirical studies. Using data on UK households, John Campbell and Joao Cocco estimate the largest effect of house prices on consumption—11 cents on the dollar—for old homeowners, and almost no effect for young renters.11 Using data on U.S. households, Andreas Lehnert also finds that the percentage increase in consumption for a given change in wealth depends crucially on a household’s age, ranging from 0 percent for middle-aged homeowners to 4 percent for young homeowners and 8 percent for relatively old homeowners. As a comparison, the marginal propensities to consume out of housing wealth found in our paper and those cited above are slightly larger than the marginal propensities to consume out of stock market wealth. Econometric specifications of total consumption such as those included in the Federal Reserve Board’s model, as outlined in the article by Flint Brayton and Peter Tinsley, generally show that an additional dollar of stock market wealth raises the level of consumer spending by 3 to 5 cents. housing wealth rises, these increases in consumption do not necessarily make all homeowners better off. But how do economists measure well-being? One way is to ask a household a hypothetical question: How much would your lifetime consumption have to increase to make you indifferent between the current situation where house prices remain unchanged and a permanent increase in house prices of a certain percentage (10 percent as in the case of our working paper)?12 Since households of different homeownership status (rent versus own), ages, or assets have different needs for housing services, the answers will clearly be very different for different households. Young Homeowners Are Worse Off. Young households’ income is likely to rise steeply over most of their lifetime, and the size of their families Economists call this the change in compensated demand. 12 is likely to expand over time. Therefore, their desired house size–one that matches their future income and family size–exceeds what they can afford with their current income and wealth. Thus, they will upgrade over time. The typical housing ladder for young households is an apartment, a starter house, and then a larger house. Even if all houses increase in value at the same rate, a big house typically appreciates more in dollars than does a small house.13 As a result, the wealth gains young homeowners receive from their current houses are not enough to offset the cost of acquiring future housing services that are likely to exceed their current services. Consider a young homeowner in Joseph Gyourko and Joseph Tracy find that between the mid-1980s and the late 1990s, high-end houses tended to have a higher real rate of appreciation than low-end houses. However, they argue that some of it is because the quality of high-end houses has improved more than that of lower-end houses. 13 FIGURE 2 Consumption Changes as a Percentage of Housing Wealth Changes Percent 9 8 7 6 5 4 WHEN CONSUMPTION RISES BECAUSE OF RISING HOUSE PRICES, HOUSEHOLDS AREN’T ALWAYS BETTER OFF Although homeowners will increase their consumption when their 11 See Campbell and Cocco’s 2004 article. www.philadelphiafed.org 3 2 1 0 Renter Young Homeowner Middle-aged Homeowner Old Homeowner Note: Renters do not participate in housing wealth gains since they do not own houses. As a result, their consumption out of housing wealth gains is not defined. Business Review Q1 2006 29 his late twenties who owns a condo worth $100,000 and plans to move to a house worth $200,000 after marriage. Assume that house price appreciates 10 percent, and the condo is now worth $110,000 and the house, $220,000. Before the house-price increase, the young homeowner needed only $100,000 in addition to his gains from selling his own house to buy the new house; now he needs $110,000, or $10,000 more. The young homeowner will have to either postpone the purchase of the new house or buy a smaller one. Middle-Aged and Old Homeowners Benefit from House-Price Appreciation as Wealth Effects Dominate Their Consumption Needs. For middle-aged homeowners, income has peaked and family size has stabilized. Their house size matches their income and wealth profiles as well as their families’ needs. Therefore, they will not change their house size for the foreseeable future. House-price appreciation thus increases their assets without changing their cost of acquiring future housing services. For example, consider a middleaged couple who have two children, both in primary school, and who own a house worth $350,000. The couple most likely will not experience dramatic changes in its income, since both members have been working at their careers for a number of years and their children will remain at home for the next five to 10 years. Since this family does not have any plans to move, a 10 percent house-price appreciation increases its assets by $35,000 without increasing its cost of acquiring housing services. The household can then spend the additional wealth on other consumption goods over its remaining life span. Old homeowners, who are generally looking to downsize, also benefit from house-price appreciation. House-price appreciation allows them to capture additional housing wealth 30 Q1 2006 Business Review gains and reduce their housing costs. For example, imagine a 65-year-old couple who own a $300,000 house and who intend to move to a $100,000 condo. After the 10 percent house- more for the same housing services. To make it worse, suppose the renter plans to buy the apartment, which is being turned into a condo by his landlord. Before he needed to pay The effects of house-price changes on total consumption and consumer welfare obviously depend on the demographics of the economy. price appreciation, the couple sell the house for $330,000 and buy the condo for $110,000. Their wealth gain is $220,000, $20,000 more than the wealth gain they would have received before the house-price appreciation. This additional wealth will help boost their other consumption as well as their bequest. Renters Are Strictly Worse Off When Housing Price Increases. Think of a renter who pays $6,000 a year for an apartment worth $100,000. If we assume rent is 6 percent of the house value, after house-price appreciation, he pays $6,600 a year, $600 only $100,000; now he needs to pay $110,000, or $10,000 more. As with the young homeowner, the renter will have to either cut current consumption to continue to live in the apartment or buy a smaller condo. As we show in our working paper, house-price appreciation of 10 percent yields a loss of 4.5 percent in welfare for renters and a 2 percent loss in welfare for young homeowners. At around age 50, the welfare gains reach breakeven. Only households beyond the age of 65 receive a welfare gain exceeding 2 percent. FIGURE 3 Change in Welfare from a 10 Percent House-Price Appreciation Percent 6 4 2 0 -2 -4 -6 Renter Young Homeowner Middle-aged Homeowner Old Homeowner Note: The welfare consequence is defined as changes in a household’s lifetime consumption so as to make it indifferent between the current situation where house prices remain unchanged and a permanent 10 percent house-price appreciation. For example, an old homeowner would need to be given 5 percent more lifetime consumption to be as well off without an increase in house prices as with a 10 percent increase in house prices. In contrast, a renter would be willing to pay 4.4 percent in consumption to avoid the 10 percent increase in house prices. www.philadelphiafed.org An Example of the Effects of a Change in Housing Wealth C onsider an economy that consists of one renter, one homeowner in his early 30s, one homeowner in his mid-50s, and one homeowner in his 80s. The renter rents a $150,000 house, the young homeowner lives in a $200,000 house, the middle-aged homeowner owns a $350,000 house, and the old homeowner has a $300,000 house (to keep things simple, let’s forget about the landlord). Now let’s assume that all houses appreciate 10 percent. As a result, the young homeowner gains $20,000, the middle-aged homeowner gains $35,000, and the old homeowner gains $30,000. The total housing wealth gains for households in this economy is thus $85,000 (=$20,000 + $35,000 + $30,000). Further assume that the marginal propensity to consume that results from this increase in housing wealth—that is, the increased consumption arising from a $1 increase in wealth—is 0.06 for homeowners younger than 35, 0.04 for homeowners between 35 and 75, and 0.10 for homeowners above age 75. The total consumption gain for the homeowners is $5,600 (=20,000 * 0.06 + 35,000 * 0.04 + 30,000 * 0.1). Assuming a rental cost of 6 percent of the house value, the renter needs to pay $900 more in rent because of the house-price appreciation (=0.1*150,000*0.06). He will need to find this rent money. Suppose the money comes from lower current consumption and the renter increases savings by $100 for the furure purchase of a house. Then the total consumption of the homeowners and the renter has increased by $4,600 (=$5,600 -$900 - $100), implying a marginal propensity to consume out of the $85,000 increase in housing wealth of 5.4 percent (=$4,600/$85,000). Suppose house prices stay high, and households in the economy grow older and move into the houses formerly occupied by the next oldest household. In addition, the old homeowner dies, and a new renter is born into the economy. Compared to the economy before house prices appreciated, the new renter pays $900 more in rent. The new young homeowner pays $20,000 more for the new house. The new middle-aged homeowner loses $15,000 (he made $20,000 but pays $35,000 more for his new house). The new old homeowner gains $5,000. The old homeowner’s heirs are better off by $30,000, the appreciation of his house. We illustrate the points in Tables 1 and 2. TABLE 1 Housing Wealth and Consumption Renter Rents/house value before house-price appreciation House value: $150,000 Annual rent: $9,000=$150,000*0.06 Rents/house value $900 gains after house =$150,000*0.1*0.06 price appreciates 10% Marginal propensity to consume out of housing wealth Consumption gains Young homeowner Old homeowner House value: $200,000 House value: $350,000 House value: $300,000 $20,000 =$200,000*0.1 $35,000 =$350,000*0.1 $30,000 =$300,000*0.1 0.06 -$1000 =-$900-$100 assuming the renter increases savings by $100 for future house purchase Middle-aged homeowner $1200 =$20,000*0.06 0.04 $1400 =$35,000*0.4 0.1 $3000 =$30,000*0.1 continued on page 32 www.philadelphiafed.org Business Review Q1 2006 31 An Example of the Effects ... (continued) TABLE 2 Housing Wealth and Consumer Welfare Renter Rents/house value before house-price appreciation Middle-aged homeowner Old homeowner House value: $200,000 House value: $350,000 House value: $300,000 Rents/house value $900 gains after house =$150,000*0.1*0.06 price appreciates 10% $20,000 =$200,000*0.1 $35,000 =$350,000*0.1 $30,000 =$300,000*0.1 Wealth gains/losses -$20,000 after upgrading to the next house -$15,000 =$20,000-35,000 $5,000 =$35,000-30,000 $30,000 memo House value: $150,000 Annual rent: $9,000=$150,000*0.06 Young homeowner New renter loses $900 IMPLICATIONS FOR THE ECONOMY AS A WHOLE The effects of house-price changes on total consumption and consumer welfare obviously depend on the demographics of the economy. One thing that is certain is that although individual groups—notably the old, the young, and renters—may experience significant changes in their wealth, the overall change in wealth may be small, since the individual effects may, to some degree, cancel each other in aggregation. (See An Example of the Effects of a Change in Housing Wealth.) In our analysis, we find that a permanent house-price increase of 10 percent leads to a slight decrease (0.9 percent) in overall welfare. So far in our analysis, we assumed that house-price appreciation is the same for all places. Obviously, housing 32 Q1 2006 Business Review markets are local markets. If you live in an area where house-price appreciation is strong and move into an area with low appreciation, you can gain. In the longer term, however, appreciation across areas will likely equalize precisely because of this type of movements. We do not consider these differential regional markets here. (See What If Housing Prices Fall? for a brief discussion of the situation in which housing prices depreciate.) CONCLUSIONS Homeownership occupies a pedestal next to apple pie and motherhood as part of the American dream. Spurred by demographic trends, a strong economy, and preferential government policy, the homeownership rate has increased significantly in recent years. Today, close to 70 percent of households own their houses, and a substantial amount of households’ wealth is now tied to housing. Do these statistics imply that changes in housing prices will have significant effects on households’ consumption and welfare? The answer is, it depends on whom you’re asking. Since a house is both an asset and a necessary outlay (we all need to live somewhere), house-price increases do not make a typical household richer. In other words, changes in house price have limited effects for a typical household and for the overall economy. The distributional effects, however, can be large. In particular, increases in house prices effectively transfer wealth from renters to homeowners and from young to old. By contrast, decreases in house prices transfer wealth from homeowners to renters and from old to young. BR www.philadelphiafed.org What If Housing Prices Fall? S o far, we have focused our attention on the effects of house-price appreciation, drawing on the recent experience of the residential housing market. A natural question is: What happens if house prices depreciate? This question is especially important in light of recent concerns of a possible housing bubble in the U.S. A housing bubble here means that house price is significantly higher than its fundamental value. There are several common ways of thinking about housing’s fundamental value. One is to consider the ratio of housing prices to rents, an equivalent to the price-to-dividend ratio for stocks. Since rent is a measure of the flow of housing services, in the long run, there should be a stable relationship between rents and housing prices. Another way is to consider the ratio of housing prices to household income. Of course, regulatory and tax changes can alter the long-run relationship between rents and housing prices as well as income and housing prices. Interested readers can read articles by Joshua Gallin (2003, 2004), and Charles Himmelberg, Christopher Mayer, and Todd Sinai (2005), among many others. Our argument applies equally to the situation with house-price depreciation. Middle-aged homeowners’ consumption will remain least responsive to the decline in house price for the same reasons discussed earlier. Young homeowners have to cut their consumption, since they can no longer rely on home equity to help smooth con- www.philadelphiafed.org sumption. Note that these homeowners do not have much liquid savings they can cut. Similarly, old homeowners, who are already depleting their savings to support consumption, also need to cut their consumption now that they are not as wealthy as they used to be. Renters, by contrast, will increase their consumption, since they now pay less in rent and do not need to save as much as they used to in order to buy a house. Despite the decline in consumption, young homeowners may still benefit from a depreciation in house prices if the decline in their future housing cost is significant enough to offset the short-term drop in consumption due to their worsened liquidity. Middle-aged and especially old homeowners, on the other hand, are worse off because the decline in housing costs for their remaining life may not be enough to compensate them for the decline in their wealth. Renters, by contrast, are strictly better off, since they suffer no wealth loss, yet benefit from lower future housing cost. If the house-price depreciation becomes too severe, however, many homeowners may choose to default on their mortgage. Things would then become more complicated. Those who had more equity before the depreciation would obviously lose more financially. However, since lenders may decide not to lend to these people in the future, those households with longer expected life spans (generally the young and middle-aged) will suffer more from the reduced access to future credit. Business Review Q1 2006 33 REFERENCES Bajari, Patrick, C. Lanier Benkard, and John Krainer. “House Prices and Consumer Welfare,” Journal of Urban Economics, forthcoming. Benjamin, John, Peter Chinloy, and Donald G. Jud. “Real Estate Versus Financial Wealth in Consumption,” Journal of Real Estate Finance and Economics, 29 3 (2004), pp. 342-54. Brayton, Flint, and Peter Tinsley. “A Guide to FRB/US,” Federal Reserve Board Finance and Discussion Series Working Paper 1996-42 (October). Cocco, Joao F. “Portfolio Choice in the Presence of Housing,” Review of Financial Studies (2005), pp. 535-67. Gallin, Joshua. “The Long-Run Relationship between House Prices and Rents,” Federal Reserve Board Finance and Economics Discussion Series 2004-50. Gallin, Joshua. “The Long-Run Relationship between House Prices and Income: Evidence from Local Housing Markets,” Federal Reserve Board Finance and Economics Discussion Series 2003-17. Himmelberg, Charles, Christopher Mayer, and Todd Sinai. “Assessing High House Prices: Bubbles, Fundamentals and Misperceptions,” Columbia University Working Paper (September 2005). Lehnert, Andreas. “Housing, Consumption, and Credit Constraints,” Federal Reserve Board Finance and Economics Discussion Series 2004-63. Li, Wenli. “Moving Up: Trends in Homeownership and Mortgage Indebtedness,” Federal Reserve Bank of Philadelphia Business Review (First Quarter 2005), pp. 26-34. Campbell, John, and Joao F. Cocco. “Household Risk Management and Optimal Mortgage Choice,” Quarterly Journal of Economics, 118 (2004), pp. 1149-94. Gervais, Martin. “Housing Taxation and Capital Accumulation,” Journal of Monetary Economics, 49, 7 (2002), pp. 1461-89. Campbell, John, and Joao F. Cocco. “How Do House Prices Affect Consumption? Evidence from Micro Data,” Harvard University and London Business School Working Paper. Glaeser, Edward. “Comments and Discussion on Karl E. Case’s Real Estate and the Macroeconomy,” Brookings Papers on Economic Activity, 2 (2000), pp. 146-50. Sinai, Todd, and Nicholas S. Souleles. “Owner-Occupied Housing as a Hedge Against Rent Risk,” Quarterly Journal of Economics, forthcoming. Carson, Joseph G. “Home Appraisal: Housing Cycle Should Continue to Fuel Consumer Spending,” Global Economic Research, www. alliancebernstein.com. Greenspan, Alan, and James Kennedy. “Estimates of Home Mortgage Originations, Repayments, and Debt on One-to-Four Family Residences,” Federal Reserve Board Finance and Economics Discussion Series 2005-41. Yao, Rui, and Harold H. Zhang. “Optimal Consumption and Portfolio Choices with Risky Housing and Borrowing Constraints,” Review of Financial Studies, 18 (Spring 2005), pp. 197-239. Case, Karl E., John M. Quigley, and Robert T. Shiller. “Comparing Wealth Effects: The Stock Market versus the Housing Market,” University of California Working Paper. 34 34 Q1 Q1 2006 2006 Business Business Review Review Li, Wenli, and Rui Yao. “The LifeCycle Effects of House Price Changes,” Federal Reserve Bank of Philadelphia Working Paper 05-7 (2005). Gyourko, Joseph, and Joseph Tracy. “A Look at Real Housing Prices and Incomes: Some Implications for Housing Affordability and Quality,” Federal Reserve Bank of New York Economic Policy Review, 5 (1999), pp. 63-77. www.philadelphiafed.org www.philadelphiafed.org Recent Developments in Consumer Credit and Payments O conference summary n September 29 and 30, 2005, the Federal Reserve Bank of Philadelphia’s Research Department and Payment Cards Center organized the fourth in a series of conferences exploring new academic research on the topic of consumer credit and payments. Nearly 100 participants attended the conference, which included seven research papers on topics such as the design of consumer bankruptcy law, predatory lending, consumers’ choice of borrowing terms and indebtedness, the function of credit reporting agencies, and pricing in credit card and ATM networks. Keynote speaker Gary H. Stern, president of the Federal Reserve Bank of Minneapolis and current chairman of the Federal Reserve System’s Financial Services Policy Committee, opened the conference. Stern began his remarks by pointing to the increasing quantity and quality of research on consumer credit This summary was prepared by Ronel Elul, Joanna Ender, Bob Hunt, and James McGrath. Elul and Hunt are senior economists in the Research Department of the Philadelphia Fed. Ender is a research analyst in Research. McGrath is an industry specialist in the Bank’s Payment Cards Center. The conference agenda, papers, and presentations can be found at www.philadelphiafed.org/econ/conf/ consumercreditandpayments/index.html. www.philadelphiafed.org and payments. While the Federal Reserve System is a significant producer of research in this area, it is also an important consumer because it acts as a provider and, in some instances, a regulator of payment services. As with monetary economics, good research informs good policy decisions, and this can be especially important when research challenges the conventional wisdom. Next Stern described some of the differences between the objectives of private providers of payment services (profit maximization) and the Fed, which is to maximize social welfare. In particular, the Fed’s mission is to encourage the efficiency, accessibility, and integrity of the payment system. Its ability to make improvements along these dimensions depends on the nature of competition in these markets, the significant network features of most payment systems, and any publicgood aspects that arise in facilitating payments. Thus, one rationale for the Fed’s involvement in a payment market might be the existence of significant market failure—too little competition or too little investment in security or reliability, for example—that cannot be more easily addressed by other means (such as regulation).1 When such conditions no longer exist, however, perhaps the Fed should gradually exit the market. Does economic reasoning inform the Fed’s choice of which payment services to provide and on what scale? Stern argued yes, pointing to the Fed’s recent decision to reduce its check-processing operations, which accounts for the majority of the System’s staffing. The national check-processing market is declining about 10 percent each year. If the Fed does not downsize, it will account for an ever-growing share of the business. But the Fed has determined that there is no market failure in this market that would justify its becoming an increasingly important provider. Nor are there significant economies of scope between its checkprocessing operations and its other payment businesses. In response, the Fed has decided to reduce its check-processing capacity Stern pointed out an additional rationale would be the existence of significant economies of scope between the Fed’s retail and wholesale payment business (Fedwire), but such economies must be rigorously demonstrated. 1 Business Review Q1 2006 35 while adjusting its prices to ensure that it recovers the full cost of providing these services. The Fed also supported the recently enacted Check 21 law, which will facilitate the electronic presentment of checks, thereby reducing the need to process and ship paper checks. The market for automated clearinghouse (ACH) transactions has also experienced significant change, and the Fed is adapting. On the one hand, demand has grown dramatically, a situation that requires significant ongoing investment. On the other hand, private-sector providers have consolidated and are now increasingly competitive. While the Fed remains a dominant provider, its market share has fallen over time. Improvements in information-processing technology, combined with significant economies of scale, have reduced the cost of ACH transactions, leading the Fed to reduce prices 66 percent over the past decade. In each of these cases, economic research has aided the Fed’s decisionmaking. Stern offered some examples of how economic research could influence the Fed’s policy decisions in the payments arena in the future. First, what is the efficacy of alternatives to the Fed’s provision of retail payment services when there are market failures? For example, should the Fed play a more significant role in standard setting, even where it is not an active service provider? Second, how will the electronification of checks affect the market structure and competitive conditions of the check processing business? Third, are the existing theoretical models of payment networks adequate for making policy decisions about whether and how to regulate interchange and other fees that arise in credit and debit card transactions? More economic research in each of these areas would help to inform policymakers and improve social welfare. 36 Q1 2006 Business Review Do Consumers Choose the Right Credit Contracts? In the first paper presented, Nicholas Souleles, of the University of Pennsylvania, reported the results of a study (with Sumit Agarwal, Souphala Chomsisengphet, and Chunlin Liu) that examined consumers’ choice between two credit card contracts and their subsequent borrowing decisions in the period 1997 to 1999.2 A large more than consumers who chose the card without the annual fee. Almost half of these borrowers (44 percent) paid interest on an average balance of $500 or more during the period studied. More than half of consumers choosing the no fee contract did not carry a balance at any time during the two-year period. On average at least, consumers would appear to be making rational choices about loan contracts. Improvements in information-processing technology have reduced the cost of ACH transactions, leading the Fed to reduce prices 66 percent over the past decade. U.S. bank offered consumers a choice between two credit cards: one with an annual fee (about $20) but a lower interest rate and another with no annual fee but a higher interest rate (about three percentage points higher). Consumers were free to switch from one contract to the other at any time. To minimize their total borrowing costs, consumers expecting to borrow a large amount should choose the contract with the annual fee and a lower rate. Conversely, consumers who do not expect to borrow very much should choose the card without an annual fee. Did consumers choose rationally? When consumers chose a contract that turned out to be more expensive for them, how likely were they to switch contracts? Souleles and his co-authors found that, on average, consumers who chose the card with an annual fee (and lower interest rate) subsequently borrowed “Do Consumers Choose the Right Credit Contracts?,” mimeo, University of Pennsylvania (2005). This paper was previously circulated under the title “How Well Do Consumers Forecast Their Future Borrowing?” 2 After the fact, however, some consumers would have done better had they chosen the other contract. For example, 24 percent of consumers who paid the annual fee never borrowed at all. Among consumers who did not pay the annual fee, 12 percent paid interest on an average balance of $1,200 a month or more. In total, about 40 percent of consumers chose a contract that turned out to be more expensive (56 percent paying the annual fee and 21 percent who didn’t). Are these mistakes? Or is it that consumers’ borrowing was not what they anticipated it would be? To explore the possibility that consumers are making mistakes, Souleles and his co-authors examined a subset of consumers who also had substantial deposits at the bank. The idea is that these customers have ample liquid funds to help them manage an expense shock, so we would not expect them to borrow much on their cards or to pay the annual fee. Not surprisingly, only 22 percent of these consumers do pay the annual fee (compared to 55 percent for the entire sample). What is surprising, however, is that 10 percent www.philadelphiafed.org of these liquid customers who did not pay the annual fee also paid interest on an average balance of $1,200 a month or more. These customers chose a contract that turned out to be more expensive because they borrowed a significant amount, and yet it seems unlikely this was due to unanticipated shocks. The authors concluded that unanticipated borrowing does not explain all of the patterns in the data Next, the authors explored whether consumers are likely to choose the more affordable contract when the cost of mistakes is higher. In particular, they calculated the interest consumers would have saved if they had paid the annual fee to benefit from the lower interest rate on their card. When the interest savings (net of the annual fee) was less than $26, about 37 percent of consumers chose the wrong contract. But when the interest savings exceeded $300, only 7 percent of consumers chose the wrong contract. Examining the small share of consumers who changed their contracts, the authors found that the majority of those initially chose a contract that turned out to be more costly than the alternative. They also found that the probability that a consumer changes his or her contract is significantly affected by the net savings that result after the switch. The discussant, John Leahy, of New York University, suggested that Souleles and his co-authors present a formal model of consumers’ contract choices to help interpret the pattern of mistakes they report. Leahy suggested that borrowers might choose the more costly contract to discourage themselves from borrowing in the future. This is an example of a commitment problem explored by other researchers in the literature. Leahy also noted that only 5 percent of borrowers’ errors cost them more than $25 a year; only 1 percent made errors that cost them more than $100 a year; and even www.philadelphiafed.org less, 0.1 percent, made errors that cost more than $300 a year. Such low costs suggest that many errors may simply be due to consumers’ inattention. It is even possible that some consumers forgot they had the option to switch. For those who did pay the annual fee, such costs are sunk until the fee comes due a year later. Explaining the Rise in Consumer Bankruptcies in the U.S. Igor Livshits, of the University of Western Ontario, presented the results of his research (co-written with James MacGee and Michèle Tertilt), which tested a variety of explanations for the dramatic rise in bankruptcy filings in the U.S. over the last quarter century.3 The basic facts are as follows: (1) the number of filings increased from 1.4 per thousand adults in 1970 to 8.5 per thousand in 2002; (2) filers’ ratio of unsecured debt-to-income has increased; and (3) the average real interest rate on unsecured credit hardly changed. The authors constructed a lifecycle model of consumers who borrow and sometimes default and calibrated it to match the behavior of borrowers in the U.S. economy during the late 1990s. They used the model to explore the effects of many proposed explanations for the rise in the bankruptcy filing rate that occurred after 1980. They considered a variety of possible explanations for the rise in the bankruptcy filing rate and concluded that while no single explanation is fully consistent with the evidence, a combination of factors, including a decline in stigma associated with filing for bankruptcy, comes reasonably close. “Accounting for the Rise in Consumer Bankruptcies,” mimeo, University of Western Ontario (2005). 3 Livshits and his colleagues first considered whether an increase in “uncertainty” can explain the patterns in the data. They found that increases in the magnitude or likelihood of expense shocks (such as out-of-pocket medical expenses) or income shocks (such as unemployment spells) would increase the bankruptcy filing rate, but it would also reduce the ratio of unsecured debt-to-income, which did not happen. The authors also considered shocks to family structure (such as divorce or an unplanned pregnancy) but found that these did not rise after the early 1980s. They did find that the decline in the share of the adult population that is married would explain a small part of the rise in the filing rate. They found no effect from changes in age structure of the population. Livshits and his co-authors then turned to changes in the credit market environment. They rejected the potential effect of the changes in U.S. bankruptcy law introduced in 1978, arguing that Canada also experienced a rise in bankruptcy filings in the absence of a change in its laws. They found the relaxation of binding usury ceilings after 1978 can explain a significant rise in bankruptcy filings and an increase in the debt-to-income ratio, but it would also imply an increase in the real cost of unsecured credit that is not observed in the data.4 They are also skeptical that in practice the usury ceilings are sufficiently restrictive to generate such effects. They did find two factors that seem to be important in explaining the In Marquette National Bank v. First Omaha Service Corp, 439 U.S. 299 (1978), the Supreme Court determined that lenders could charge interest rates permitted under the laws of the state where they were located, rather than where their customers were located. Thereafter, states competed to attract lenders to their jurisdiction by raising their usury ceilings. 4 Business Review Q1 2006 37 rise in bankruptcy filings but which cannot individually explain the patterns in the data. First, a decline in the cost of underwriting unsecured credit (perhaps due to rapid improvements in information technology) would increase borrowing but would have little effect on bankruptcy filing rates and is associated with a significant decline in average real interest rates. On the other hand, a decline in the stigma associated with filing for bankruptcy would indeed explain a significant share of the increase in filings but would also increase real interest rates and reduce the ratio of debt-to-income.5 In short, no single explanation seems to fit the trends observed in the U.S. economy over the last two decades. Livshits and his colleagues then asked what combination of factors would explain the observed trends. They argue that increases in both expense and income uncertainty, combined with a decline in underwriting costs and stigma, fit the data fairly well. In their simulation, increases in uncertainty play a relatively small role, while a decline in stigma is the primary driver of the rise in bankruptcy filings. At the same time, a decline in underwriting costs offsets the effect of stigma on interest rates and the ratio of debt-to-income. The authors concluded that a decline in stigma plays a very important role in the story and suggested that it should be the focus of future research. The discussant, Satyajit Chatterjee, of the Federal Reserve Bank of Philadelphia, argued that the paper is an important advance but its results should be interpreted cautiously. For example, when the model is calibrated to the data, the implied recovery rate for debt in bankruptcy is about 28 percent, which seems rather high for a model that seeks to explain filings under Chapter 7 (discharges) rather than Chapter 13 (workouts). In the paper, this is explicitly modeled as In their calibrations, the level of stigma required to explain the filing rate of the early 1980s is equivalent to the welfare lost from a 28 percent decline in consumption. 6 5 38 Q1 2006 Business Review Predatory lending is less likely to occur as the lending market becomes more competitive. a wage garnishment, but Chatterjee suggested that its high value probably reflects other costs omitted from the model, such as the nonexempt assets subject to liquidation by the court. He also wondered how the results would change if the assumption of a perfectly competitive loan market was relaxed. Predatory Lending The next speaker, Bilge Yilmaz, of the University of Pennsylvania, presented the results of his research (with Philip Bond and David Musto) on the topic of predatory lending. They begin by offering a definition of the practice and investigating the conditions under which it can occur.6 The authors define predatory lending as a loan the lender knows will, on average, make the borrower worse off. But why would a borrower choose a loan that was likely to make him or her worse off? In their model of a rational loan market, it must be Philip Bond, David Musto, and Bilge Yilmaz, “Predatory Lending in a Rational World,” Federal Reserve Bank of Philadelphia Working Paper 06-2 (2006). the case that lenders know more about a borrower’s future income prospects than does the borrower. Predatory lending has two obvious policy implications. First, if borrowers are choosing loans that are likely to make them worse off, credit is being misallocated in a way that may be socially wasteful. Second, predatory lending may increase the inequality in the distribution of wealth. Yilmaz and his co-authors develop a model in which a borrower applies for a loan using his or her home as collateral. The lender has some information about whether the borrower is more likely a “good” or “bad” risk. Borrowers who are good risks are more likely to earn sufficient income in the future to repay the loan than are borrowers who are bad risks. Based on that knowledge, the lender makes a loan offer, which the borrower either accepts or declines. If the loan cannot be repaid, the lender recoups at least some of the proceeds by foreclosing on the borrower’s home. Their first insight, according to Yilmaz, is that in order for predatory lending to occur, it must be the case that good and bad risks receive the same loan terms. In other words, the equilibrium must be a pooling equilibrium. If that were not the case, the lender’s superior information would be revealed by his offer: The bad risks would realize they faced a higher risk of defaulting on the loan than they originally thought. In that case, the bad risks would not take out the loan. When is predatory lending likely to occur? Yilmaz and his co-authors show that several conditions are required. First, predatory lending requires that lenders be better informed than borrowers about the riskiness of the loan. Second, collateral values must be sufficiently high, so that lenders do not lose too much if they lend to bad borrowers who subsequently www.philadelphiafed.org default.7 Third, predatory lending is less likely to occur as the lending market becomes more competitive because rival lenders tend to cherry-pick the best borrowers, unraveling the pooling equilibrium.8 The authors examine three policies that may affect predatory lending. They argue that interest-rate ceilings (usury laws) can sometimes help reduce predatory lending. If the ceiling is set sufficiently low, lenders cannot recoup the cost of their inefficient loans to the bad risks. Of course, such a benefit must be weighed against the other distortions usury ceilings can cause. Next, they consider the Community Reinvestment Act, which requires banks to lend in underserved and underprivileged areas. The authors suggest that this can also help break down predatory lending if it increases competition in the lending market in such areas. Note this might imply less actual lending in these areas, rather than more, because the bad risks choose not to borrow. Finally, they consider the Equal Credit Opportunity Act, which specifies that certain factors (for example, age, race, or gender) may not be considered in underwriting or pricing loans. If such restrictions do facilitate a pooling equilibrium, predatory lending may become more likely. Discussant Andrew Winton, of the University of Minnesota, pointed to some alternative explanations of why a borrower might accept a predatory loan. For example, borrowers might not understand the “fine They point out that loans that increase the value of collateral, such as home-improvement loans, may therefore increase the prospects for predatory lending. 7 Still, as long as loans are fully collateralized, the authors show that predatory lending remains a possibility even under highly competitive conditions. print” of loan contracts, or lenders may misrepresent loan terms. Borrowers may exhibit excessive optimism or too heavily discount the costs of a loan contract that occur in the future. Each is an example of predatory lending in a less than rational world. Winton also suggested that, in addition to foreclosures, the authors should examine other costs of predatory loans, including excessive loan payments. Credit Bureaus, Relationship Banking, and Loan Repayment Martin Brown, of the Swiss National Bank, discussed his work with Christian Zehnder on the function and effects of credit reporting agencies.9 In particular, they studied the extent to which credit registries improve repayment behavior, an idea that is widely accepted but has not been rigorously tested in empirical work. They also examined another mechanism for disciplining borrowers—relationship lending, which involves repeated interactions between a specific borrower and lender. One question they sought to answer was the degree to which these two mechanisms are substitutes or complements. Brown and Zehnder developed an experiment in which multiple borrowers and lenders interact with each other in a computerized lending game. There are more lenders than borrowers, so the loan market is relatively competitive. The authors examined the performance of their experimental loan market along two dimensions: whether or not a credit bureau exists and whether or not borrowers and lenders can recognize each other. Note that if borrowers and lenders cannot 8 www.philadelphiafed.org “Credit Registries, Relationship Banking, and Loan Repayment,” IEW Working Paper 240, University of Zurich (2005). 9 recognize each other, they cannot engage in relationship lending. Suppose that borrowers and lenders cannot recognize each other. This is consistent with a lending market in which borrowers are highly mobile. If there is no credit bureau, borrowers are essentially anonymous. In that case, the experimental results show the market performs extremely poorly – borrowers frequently default so few lenders offer any funds. Next, Brown and Zehnder introduce a credit bureau. This consists of a list lenders receive in every period that documents each borrower’s previous loans and repayment behavior (no other information is provided). With the bureau in place, the market functions dramatically better, for most rounds of the game. Repayment rates and lending volume are significantly higher. Brown and Zehnder attribute this improvement in results to the disciplining effect of credit registries; borrowers are willing to repay in order to maintain reputations and hence retain access to future credit. As further evidence, they point to the following detail from their experimental results. In the final periods of the game, the market breaks down even in the presence of a credit bureau. Borrowers recognize that they have no further need to maintain their reputation and lenders, recognizing this, decline to lend. Next, Brown and Zehnder considered the case where borrowers and lenders can recognize each other, which makes ongoing lending relationships possible between specific borrowers and lenders. They found that even in the absence of a credit bureau, the loan market functions very well. Thus lending relationships also appear to act as an effective mechanism for disciplining borrowers. When a credit bureau is introduced in this environment, there is a slight increase in performance, but the difference is Business Review Q1 2006 39 not statistically significant. Brown and Zehnder conclude that credit bureaus and relationship lending are largely substitutes. Discussant Paul S. Calem (LoanPerformance) argued that the paper raises several potential policy implications. It clearly provides evidence of the contributions that credit bureaus can make—they make it possible for consumers to invest in their reputations as good borrowers. This, in turn, increases the availability and pricing of credit. But Calem pointed out that the experiment is highly stylized so it is important to place the results in the context of actual credit markets. For example, in the U.S. at least, there are markets in which credit bureaus dominate (consumer credit) and other markets where relationship lending is more important (small-business lending). In addition, he pointed out that while relationship lending may serve as another mechanism for enforcing repayment, it does have some drawbacks. For example, it may suffer from “lockin” where the cost of changing lending relationships results in less competitive pricing. Returning to Brown and Zehnder’s experiment, Calem noted it would be interesting to know whether the presence of a credit bureau has a significant effect on the pricing of loans or whether the incremental contribution of credit bureaus depends on competitive conditions. Finally, while the paper is silent on these questions, Calem pointed out that the actual content of credit bureau files may be important factors. Brown and Zehnder’s credit bureaus include both positive and negative credit information, but many bureaus around the world include only negative information. In addition, the optimal length of credit histories included in bureau files is open to debate. If records are kept too long, marginal bor- 40 Q1 2006 Business Review rowers may feel that their record can never be rehabilitated, and this would weaken the discipline that credit bureaus are supposed to enable. The Effects of Incomplete Information on Consumer Credit Jonathan Zinman (Dartmouth College) presented the results of his work with Dean Karlan. They have designed an empirical study that seeks to identify adverse selection and moral hazard in loan markets.10 In other words, do higher interest rate loans attract riskier clients? (This is known as adverse selection.) Do higher inter- among the clients who respond to the offer, approximately 40 percent were randomly given a low contract rate instead (the remainder received the original offer rate). Finally, half of the applicants were randomly given a dynamic repayment incentive—assuming the borrower repaid the current loan, he or she would receive a favorable interest rate on subsequent loans over the next year. To test for adverse selection, Karlan and Zinman compared the repayment performance of two groups: borrowers who responded to the low offer rate and borrowers who responded to the high offer rate but subsequently Credit bureaus make it possible for consumers to invest in their reputations as good borrowers. This, in turn, increases the availability and pricing of credit. est rate loans induce borrowers to take more risks (i.e., moral hazard)? How can the two be separately measured? Despite an abundance of theoretical work, there is remarkably little empirical research on these questions. Karlan and Zinman implemented their experiment through a South African lender specializing in providing unsecured credit to the working poor. Their typical loans are small ($150) and the term is rather short (four months). Their experiment consisted of three stages. In the initial stage an interest rate (the offer rate) was randomly assigned to a pool of potential borrowers with similar observable characteristics. This rate could be either high or low. In the next stage, “Observing Unobservables: Identifying Information Asymmetries with a Consumer Credit Field Experiment,” Working Paper (2005). 10 received the lower contract rate. This is the test for adverse selection. Since both groups actually received the low interest rate in this experiment, there should be no effect of moral hazard. The question remains: Do higher interest rates attract riskier borrowers who care less about high rates because they are less likely to repay the loan? Next, the authors constructed two tests for moral hazard. Recall that moral hazard exists when the terms of credit affect an individual’s incentives to repay his or her loan. Karlan and Zinman begin by focusing only on those borrowers who responded to the high offer rate. This should remove the effects of adverse selection, because these borrowers should initially have the same expectations about their prospects for repaying the loan. In the first test, Karlan and Zinman compared the repayment performance of borrowers who actually received a www.philadelphiafed.org lower contract rate with those who paid at the original offer rate. Moral hazard would then show up if the second group—that with the higher interest rate—is more likely to default. Karlan and Zinman also considered a second, potentially cleaner test for the effects of moral hazard. Under the first test the fact that one group is paying a higher interest rate than the other implies there is a higher repayment burden, which in itself may lead to differences in subsequent repayment behavior, even in the absence of moral hazard. In their second test, Karlan and Zinman compared the repayment behavior of borrowers who were offered the favorable rate on future loans with those who were not.11 If those offered this dynamic repayment incentive perform better, this would also provide evidence of moral hazard (since it reflects the effect of incentives on repayment behavior). Karlan and Zinman found the problem of asymmetric information to be relevant in these loan markets. They estimate that about 20 percent of the overall default rate can be attributed to a combination of adverse selection and moral hazard. Moreover, the strongest evidence of moral hazard is identified when examining the effect of the dynamic repayment incentive— a one-percentage-point decrease in the cost of future loans reduces the default rate on the current loan by about 4 percentage points. Interestingly, they found that the particular type of information problem depended on the gender of the borrower. Lending to female borrowers appeared to suffer from the adverse selection problem, while lending to male borrowers appeared to suf- Since both groups are currently paying the same interest rate, there is no difference in repayment burden that may cloud the interpretation of the results. fer from the moral hazard problem. The discussant, Pierre-Andre Chiappori (Columbia University), stated this was extremely important research. Distinguishing between adverse selection and moral hazard is important because each has distinct welfare implications and policy prescriptions. He suggested the analysis might benefit from a structural model. In particular, he wondered about how the competitive structure of the loan markets might influence the results Do higher interest rate loans attract riskier borrowers? Do higher interest rate loans induce borrowers to take more risks? and even the form of loan contracts. Some people might not respond to high rate offers because they receive better offers elsewhere. What alternatives are available to potential borrows? Do these depend on gender? Can that explain the differences in results for men and women? Pricing in Consumer Payment Networks Alexander Tieman (International Monetary Fund) presented a paper co-authored with Wilko Bolt that examines pricing behavior in two-sided markets.12 A two-sided market is one in which there are two distinct types of end users that derive benefits from interacting with each other, which is For an accessible review of the literature, see Bob Hunt’s 2003 Business Review article at http://www.philadelphiafed.org/files/br/ brq203bh.pdf. 13 11 www.philadelphiafed.org typically facilitated by a network or platform. They focus on the concrete example of a consumer payment network, such as Visa or MasterCard, which facilitates transactions between merchants and consumers. Two-sided markets often exhibit positive externalities. In the case of payment networks, the value of holding a card for consumers is increasing in the number of merchants willing to accept the card. Conversely, the value to merchants of agreeing to accept a payment card is increasing the number of consumers that are willing to use it. Thus participants on each side of the market would benefit from subsidies that increase demand among participants on the other side. One role of payment networks, then, is to coordinate the incentives offered to consumers and merchants. Bolt and Tieman point out that in such markets there is both a price and a price structure. In this case, price refers to the total cost of transactions paid by the merchant and the consumer, while price structure refers to the share of the total price that is paid by each party. Both are set directly, or indirectly, by the network. The distinction is important because it is possible that one party, perhaps the consumer, may not pay anything for the transaction or may even receive a subsidy for using a payment card. This appears to be the case for debit cards in the Netherlands, for example. Such skewed pricing structures are receiving a good deal of scrutiny by antitrust authorities around the world and are the focus of a number of lawsuits in the U.S. The economic literature on twosided networks is relatively new and underdeveloped.13 Tieman points out “Skewed Pricing in Two-Sided Markets: An IO Approach,” DNB Working Paper 2004/13, De Nederlandsche Bank, Amsterdam (2004). 12 Business Review Q1 2006 41 that in many theoretical models of these markets, the equilibrium price structure does not look like what we often observe in consumer payment networks. Instead of skewed pricing, where one side of the market pays all (or more) of the cost of a transaction, these models tend to generate interior pricing, where each side of the market contributes to the cost of a transaction. In addition the share of total transaction costs paid by one side of the market is inversely related to the relative price elasticity of demand.14 Put more simply, the side of the market whose demand is most sensitive to changes in price bears the larger share of the total cost of the transaction. This is exactly opposite the intuition learned from the microeconomic analysis of a traditional market. In their paper, Bolt and Tieman report that such results follow from a particular assumption about the properties of the demand curves (log concavity). If a more traditional assumption about demand curves (constant elasticity of substitution) is used instead, the results are very different. In that case, the side of the market that is least sensitive to price changes will bear the larger share of total transaction costs. And if one side of the market (e.g., consumers) is sufficiently more sensitive to changes in prices than the other (e.g., merchants), it will bear none of the transactions costs. Indeed, a profit-maximizing network would choose to subsidize consumers, financing the subsidy at least in part by raising the price paid by merchants. In short, their model derives a price structure that looks like what is observed in many consumer payment networks. By price elasticity, we mean the decline in transaction volume, expressed in percentage terms, induced by an increase in transaction price, also expressed in percentage terms. 14 42 Q1 2006 Business Review Next, Bolt and Tieman turned to policy questions. How does the pricing strategy of a profit-maximizing network compare to that of a benevolent social planner? They found that a social planner would also choose a highly skewed price structure, but a lower total price than would a profit-maximizing network. Thus, a monopoly payment network would result in too few, rather than too many, transactions. In contrast, a social planner would run the network at a loss, which would require ongoing subsidies from One role of payment networks is to coordinate the incentives offered to consumers and merchants. some other part of the economy. If the network was required to break even, it is likely that all the costs would be recovered from prices charged on only one side of the network. Bolt and Tieman concluded that the existence of skewed pricing in itself does not justify intervention by antitrust authorities, but a concern for the overall price charged might. This stands in contrast to the public debate, which focuses primarily on skewed pricing rather than on the total prices charged by consumer payment networks. The discussant, Rafael Rob (University of Pennsylvania), distinguished between the two types of equilibria explored in models of this sort. Most papers in the literature focus on an interior equilibrium where not all consumers and merchants adopt the payment technology. Bolt and Tieman, on the other hand, focus on the cor- ner solutions where there is universal adoption by one or both sides of the market. Rob pointed out that models in the existing literature can also generate corner solutions if the disparities in price elasticities are sufficiently great, but they may not have the same properties as the ones explored by Bolt and Tieman. Rob also pointed out that this is a model of a monopoly provider of payment services. While this is a good approximation of the Dutch debit card market, the U.S. credit and debit card networks are a duopoly. It would be interesting to explore whether the results are sensitive to this distinction. ATM Surcharges and Consumer Welfare Gautam Gowrisankaran (Washington University, St. Louis) presented his paper with John Krainer that explores the potential gains and losses associated with the introduction of ATM surcharges in the 1990s.15 Surcharges are fees charged to consumers by owners of an ATM. Prior to 1996, ATM surcharges were extremely rare, but thereafter they became very common. This change had two effects. On the one hand, ATMs became more profitable, which stimulated the deployment of ATMs and reduced the distance consumers must travel in order to access their deposit accounts. On the other hand, consumers were now required to pay for the privilege of using at least some ATMs. In addition, the increase in ATMs exceeded the increase in transaction volume so that the average number of transactions per machine fell. Since most of the cost of operating an ATM is fixed, the decline in transaction volume implies that the 15 “The Welfare Consequences of ATM Surcharges: Evidence from a Structural Entry Model,” Federal Reserve Bank of San Francisco Working Paper 2005-01 (2005). www.philadelphiafed.org average cost of each transaction rose significantly. Gowrisankaran and Krainer asked whether, on balance, consumers and society were made better or worse off by the introduction of ATM surcharges. In practice, this simple question is very difficult to answer. To do so, Gowrisankaran and Krainer painstakingly gathered a data set of ATM locations, potential ATM locations (grocery stores and banks), and population in 32 counties along the border of two states, Minnesota and Iowa. They chose this area because, unlike Minnesota, Iowa enforced a no-surcharge law throughout the 1990s. In principle, differences in the deployment and use of ATMs in these border counties can be used to estimate the effects of a surcharge ban. But to do so, Gowrisankaran and Krainer also had to develop a structural model of the ATM market and some novel approaches to estimating the parameters of the model. To estimate their model efficiently, the authors needed to avoid calculating equilibrium outcomes for every possible combination of parameter values. While this has been done for other models of entry, Gowrisankaran and Krainer were at a disadvantage—they did not know what the prices (surcharges) were in Minnesota. Their insight was to estimate the entry model using data from Iowa counties (where prices = 0) and, using those coefficients, estimate the effects of nonzero prices using data from Minnesota counties. Assuming that the fixed cost of deploying ATMs and consumer preferences are similar in counties on either side of the Minnesota-Iowa border, the difference in the relative number www.philadelphiafed.org and geographic dispersion of ATMs between the two states can be used to infer something about the price elasticity of demand. All else equal, the greater these differences, the less elastic is the demand curve for ATMs. In the actual estimation, they found that the probability a consumer will use a given ATM falls equally as much if the total surplus is 14 percent higher. The discussant, James McAndrews (Federal Reserve Bank of New York), pointed to one of the simplifying assumptions of the paper—that the market for ATM transactions is independent of the market for other bank services. If that assumption is relaxed, differences in the market structure of Since most of the cost of operating an ATM is fixed, the decline in transaction volume implies that the average cost of each transaction rose significantly. ATM is moved 1 kilometer away or she is required to pay 8 to 10 cents more to use it. They conclude that consumer demand for transactions at ATMs is price elastic. Using estimates from their model, Gowrisankaran and Krainer calculated measures of consumer and producer surplus that result under a no surcharge regime and one that permits surcharging. They reported little difference in the total surplus generated but significant differences in its distribution. While fewer ATMs are deployed in a no surcharge regime, the estimated consumer surplus is about 10 percent higher (and producer surplus 10 percent lower) than in a regime that permits surcharging. Transaction volume is also about 16 percent higher in the no surcharge regime. They also derived the first best outcome, where consumers are charged only the marginal cost of a transaction and fixed costs are recovered via lump sum taxes. Compared to the surcharge regime, there are 50 percent more ATMs and 38 percent more transactions, and the banking between the two states might influence ATM deployment and pricing decisions. It is then possible that at least some of the effects attributed to a surcharge ban might actually be driven by differences in banking structure. McAndrews presented evidence that banking markets in the Minnesota border counties are indeed different from those in the Iowa border counties. He conjectured that Minnesota’s single-office banks were likely to charge lower foreign fees.16 On the other hand, he conjectured that since the banking market in Iowa was more concentrated, surcharges may be lower because competition for deposits is less intense.17 The net effect, McAndrews argued, is that the benefits of surcharging may be exaggerated. BR 16 Foreign fees are fees a bank charges its own customers when they use an ATM the bank does not own. One reason banks may surcharge consumers that are not their own customers is to encourage them to become customers. 17 Business Review Q1 2006 43