View original document

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

FEDERAL RESERVE BANK OF DALLAS  •  conOIlllC 77Je COJ7zjJaratiue Growth Pel.!C)rmallce ofthe US ECOJ10J77)' in the  Postwar Period Mark A. Wynne  Pree 7r'ade A(Jreenze71!s and tbe Credibili~J! of Trade Re(or1J1s David M Gould  Quant?!.i'ilzg /l1{ln{/geJ7zent~)  Role ill Bank Surviual Thomas F Siems  This publication was digitized and made available by the Federal Reserve Bank of Dallas' Historical Library (  Economic Review Federal Reserve Bank of Dallas Robert D. McTeer, Jr. President and Chief Executive Officer  Tony J. Salvaggio First Vice President and Chief Operating ()fficer  Harvey Rosenblum Senior Vice President and Director of Research  W. Michael Cox Vice President and Associate Director of Research  Gerald P. O'Driscoll, Jr. Vice President and Economic Advisor  Stephen P. A. Brown Assistant Vice President and Senior Economist  Economists Zsolt Becsi Robert T Clair John V Duca Kenneth M Emery Robert W Gilmer David M Gould William C Gruben Joseph H Haslag Evan F Koenig 0'Ann M Ozment Keith R Phillips Fiona 0 Sigalla Lori L Taylor John H Welch Mark A Wynne Kevin J Yeats Mine K YUcel Research Associates Professor Nathan S Balke Southern Methodist Univefsity Professor John Bryant Rice Univefsity Professor Thomas B Fomby Southern Methodist University Professor Scott Freeman University of Texas Professor John H Wood Wake Forest University Editors Rhonda Harris Diana W Palmer Virginia M Rogers The Economic Reviewis published by the Federal Reserve Bank of Dallas This issue matks achange from bimonthly to quarterly publication The views expressed are those of the authors and do not necessarily reflect the positions of the Federal Reserve Bank of Dallas or the Federal Reserve System Subscriptions are available free of charge Please send requests for single-copy and multiplecopy subscriptions, back issues. and address changes to the Public Affairs Department Federal Reserve Bank of Dallas, Station K, Dallas, Texas 75222, (214) 651-6289 Articles may be reprinted on the condition that the source is credited and the Research Department IS provided With a copy of the publication containing the repnnted matenal  Contents Page 1  The Comparative Growth Performance ofthe us. Economy in the Postwar Period Mark A. Wynne  Productivity growth is the single most important detel1llinant of improvements in a country's living standards over time. Accordingly, the u.s. productivity slowdown of the past two decades has caused great concern and sparked much debate. In this article, Mark A. Wynne argues that the problems associated with the U.S. slowdown may be overstated. Wynne shows that the rates of productivity growth experienced in the immediate postwar period were extraordinary in comparison with historical standards. Thus, some slowdown was probably unavoidable. u.s. productivity perfol1llance in comparison with that of other countries, especially Japan's, is also perceived as poor. But this perception may be flawed, Wynne suggests, because higher growth rates abroad reflected convergence of foreign productiVity to U.S. levels.  Page 17  Free Trade Agreements and the Credibility of Trade Reforms David M. Gould  David M. Gould argues that free trade agreements can help developing countries establish the credibility essential to successful trade reform. Credibility, he explains, is necessary if trade reform policies are to entice investment into the economic sectors where the liberaliZing country has its greatest comparative advantage. As Gould explains, a free trade agreement enhances the credibility of trade reform policies by providing evidence of a government's long-term commitment to free trade and by discouraging protectionist policies in foreign markets. Gould concludes with an outlook for U.S.-Mexican free trade.  Contents Page 29  Quantifying Management's Role in Bank Survival Thomas F. Siems  Analysts often regard the quality of bank management as the most important factor in determining whether a bank fails or survives. Applying data envelopment analysis to multiple bank inputs and outputs, Thomas F. Siems presents a new model that quantitatively assesses bank management quality. This new paradigm considers a bank's essential financial intermediation functions (that is, attracting deposits to make loans and investments) to compute a scalar measure of efficiency. Siems' analysis confirms that management's role is important to a bank's survival. Management quality scores for surviving institutions are significantly better than those for failed banks-up to two and one-half years before failure. Banks whose managers poorly allocate resources and disregard the needs of their customers and markets have a greater chance of failing.  Mark A. Wynne Senior Economist Federal Reserve Bank of Dallas  The Comparative Growth Performance of the u.s. Economy in the Postwar Period bservers commonly helieve the u.s. economy has performed poorly in recent years, by both historical and international standards, One can assess economic performance in a variety of ways, depending on \vhat \ve ~l.~sume people care about. On a year-to-year basis, people care about their job prospects and the cost of living, Thus, analysts tend to assess the shOI1-term perform~mce of the economy with measures of int1ation and unemployment. In terms of these short-run measures, the United States has done quite well and continues to do so Unemployment in the United States averaged less than '), ') percent of the labor force from 1988 to 1990. This rate compares favorably with an OEeD average that was closer to 6. ') percent during the same period. I Relative to the economies of Western Europe, where unemployment hovered in the 8 percent to 10 percent range for most of the 1980s, the U.S. experience with job creation has heen even more impressive, In terms of int1ation, the diligence of the Federal Reserve in recent years has decreased the annual rate of price increase from about 12 percent in 1979-80 to annual rates of less than 6 percent since the mid-1980s. Again, this average compares quite favorably with the experience of the other major industrialized countries over the same period, although it is still quite high by the standards of historical experience, However, the growth of the economy determines any sustained improvements in the well-being of U S. citizens. Sustainable increases in living standards can come about only through increases in per capita output. By this measure, the Cnited States seems to have performed poorly in recent years. Since 19')0, real per capita gross domestic product (GDP) in Japan has increased ahout tenfold, which translates into an average  O  Economic Review - January 1992  annual growth rate of 6.1 percent. 2 Per capita GOP in the newly industrializing countries of Southeast Asia (Hong Kong, Singapore, Taiwan, and South Korea) also increased at an average annual rate of more than 6 percent between 1960 and 198'). By contrast, per capita output in the United States increased at only a 2.1 percent annual rate over the same period. More worrisome to some commentators is the fact that performance of the U.S. economy seems to have deteriorated toward the end of this period. Of particular concern is the perceived slowdown in productivity growth over the past two decades. Some commentators say the slowdown began in the mid-1960s, while others focus on the watershed date of 1973. If we opt for the latter date,  I want to thank Edie Adams for excellent research assistance The OECD (Organization for Economic Cooperation and Development) consists of Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Iceland, Ireland, Italy, Japan, Luxembourg, the Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, Turkey, the United Kingdom, and the United States. " GOP is a measure of output produced within a country, whether by domestic or foreign-owned factors of production Gross national product (GNP) is a measure of the output of all domestically owned factors of production, regardless of whether they are located within the country or abroad The two measures are related by the identity GNP = GOP + Net Factor Payments from abroad The US Commerce Department recently switched from emphasizing GNP to using GOP as its principal measure ofaggregate  economic activity  1  we find that between 1960 and 1973, per capita GDP in the United States grew at an average annual rate of 2.7 percent; between 1973 and 1988, growth averaged only 1.7 percent each year. Furthermore, most projections for growth over the next decade are below rates from the golden age of postwar economic growth. For example, a May 1991 long-term forecast from Data Resources Incorporated (1991) predicts that per capita GNP will grow at an average annual rate of only 1.3 percent between 1990 and 1999. 5 Below, I will argue that by comparative standards, whether internationally or historically, recent u.s. growth performance has not been quite as bad as one might believe. First, from the perspective of our own history, the golden age of u.s. growth immediately after World War II is more of an anomaly than the more recent slowdown. In many ways, the recent deceleration of growth can be considered part of a reversion to historical norms. Second, regarding U.S. performance relative to the other major industrialized economies, much of the recent strong performance of the United States' principal competitors can be attributed to a process of catching up with the leader (the United States). Adjusted for the catchup phenomenon, the comparative performance of the U.S. economy is not quite so bad. Indeed, evidence suggests that the United States has performed on a par with other major industrial economies once allowance is made for catch-up. This is not to say that nothing can or should be done to improve the growth potential of the u.s. economy over the coming decades. But before policymakers can address how to increase growth rates, they must know how much of the differences in growth rates observed across countries and over time can be attributed to differences in factors susceptible to influence by policy and how much is beyond the scope of policy to control.  3  4  This figure is from the TRENOLONG0591 forecast, ORl's principal projection, which assumes the economy suffers no major mishaps between 1991 and 2001 For a model that predicts increasing growth rates over time, see Romer (1986)  A century of u.s. economic growth  Let us examine the historical record on economic growth in the United States. I will focus on real GDP, which measures the total final output of labor and capital in the United States. Figure 1 plots the average annual rates of growth of real per capita GDP in the United States since 1870. The rates are expressed as decade averages to smooth short-run movements associated with business cycles. It is hard not to be impressed by the stability of the growth rates over such a long period. Real per capita GDP appears to have grown at an average annual rate of about 2 percent since 1890. There is little evidence of a slowdown in the trend rate of growth of per capita GDP over this period. If anything, there is some evidence of a slight upward trend in the growth rates.' In terms of the more recent experience, it is clear that the economy grew more slowly in the 1970s than in the 1960s or 1980s. However, the slowdown appears to be well within the range of historical experience and thus can be interpreted as a temporary rather than permanent phenomenon. But what underlies sustained increases in per capita GDP and, ultimately, consumption? In the short run, the consumption of every person can be increased in many ways. The first and most obvious way is by reducing saving and investment. But this option can work for only a short time. Lower investment will eventually lead to a decline in the capital stock, thereby lowering our ability to produce and consume in the future. Another possibility is to increase the size of the labor force relative to the population. The problem here is the obvious upper limit on the fraction of the population that can be in the labor force. Also, some disutility is presumably associated with working that would need to be offset against the increase in per capita consumption. This disutility would also offset to some degree the extra output that could be obtained in the short run by simply having everyone work harder. A third possibility is to increase imports without increasing exports. This, too, will work for only a short time, because it entails the accumulation of debts to foreigners that eventually must be repaid with real goods and services. This leaves increases in productivity, the Federal Reserve Bank of Dallas  Figure 1  Growth of Real Per Capita GDP in the United States Average annual growth rate (percent)  5  3  2  o-t----t+------------------I  +---r--.----,.----,-----,----,---r---r--.--------.---.  1870--79  1890-99  1910--19  1930--39  1950--59  1970--79  SOURCES: For 1929-89, data are from Council of Economic Advisers (1991) For 1870-1928, GOP data are from Kendrick (1961, Table A-III); population data are from U S. Department of Commerce (1975, series A-6).  amount of final output obtained per unit of input or some combination of inputs, as the only feasible route for raising living standards over time. When evaluating productivity performance, it is important to take a long-run or historical perspective to get a clear idea of what is occurring, because productivity measures are very volatile over short periods. The palticular measure of productivity on which I focus is labor productivity, defined as real GDP per hour worked. Two considerations dictate this choice. First, estimates are available for long periods, covering a variety of countries, thus facilitating intertemporal and international comparisons. Second, one can argue that this measure of productivity is more appropriate for assessing the potential for sustained increases in living standards than more broadbased measures, such as total factor productivity.' Maddison (982) is the primary data source used by most researchers in this field. The appendix to this book contains productivity estimates for sixteen countries, including the United States, covering 1870-1979. These data are extended to 1984 for a subset of six countries in a more recent article by Maddison (987). However, even with Economic Review-january 1992  the extra data provided by Maddison (987), we are left with an incomplete idea about what has been happening to productivity growth in the United States since the early 1980s. Figure 2 plots the growth rate of productivity in the United States until 1988 by consistently extending Maddison's series (for details on the extension, see the Appendix). Several points can be made by examining Figure 2. First, there is a sense that something around a 2 percent annual rate of productivity growth can be considered a long-run norm for the U_S. economy. The big deviations from this norm occurred during the Great Depression and World War II, periods that are extraordinary in terms of aggregate economic activity. Between 1929 and 1938, productivity grew at an average annual rate of less than 1 percent. The collapse in business investment during the Great Depression can be taken to be the major cause of the productivity slowdown during this period. Between 1938 and 1950, a period that incorporates World War II, productivity grew at average annual rates of more than 4 percent. The immunity of the U.S. economy to war damage and the labor shortage associated with conscription played important roles in boosting U.S. productivity growth during this period. Private investment contributed little to productivity growth during the war years. Gross private investment had just returned to its preDepression level when the United States entered the war in 1941. Investment declined during the war, although government-owned, privately operated capital enhanced manufacturing productivity during the war years (Gordon 1969). The rapid rates of productivity growth during the war and immediate postwar period essentially reversed the effects of the extremely low rates of productivity growth recorded during the Great Depression. Averaging over 1929-50, productivity grew at an annual rate of 2.7 percent. Growth  5  Baumol, Blackman, and Wolff (1990, 226-28) make a strong argument for focusing on labor productivity despite its shortcomings Their argument essentially boils down to the notion that increases in labor productivity are indicative of increases in the potential reward, in terms of consumption, for an hour of work  Comparative perfonnance of the u.s. economy  Figure 2  Growth Rate of GOP Per Hour in the United States Average annual growth rate (percent) 45 4  35 3 25 2  15  5  O+---r---r--...---...----r---r--,...---,...---,.---------, 1880  1890  1900  1913  1929  1938  1950  1960  1970  1980  1988  SOURCES: Data are from Maddison (1982) and the calculations described in the Appendix.  rates in the 1950s and 1960s were closer to the historical norm. Evidence clearly shows a slowdown during the 1970s that brought productivity growth rates to near Great Depression levels. Baumol, Blackman, and Wolff (1990, 69) present a similar figure that shows the slowdown persisting into the 1980s. They discount the significance of this persistence, arguing that their data are distorted by cyclical factors and do not capture the productivity gains that came later in the decade. Figure 2 shows a greater deterioration in productivity growth in the 1970s than is indicated in Maddison's raw data. This difference may be attributable to data revisions or to the updated and historical series not being strictly comparable. As the Appendix shows, the manner in which the productivity figures were estimated for the 1980s seems to reproduce the historical series reasonably well, so the difference in the estimates for the 1970s probably is due to data revisions. Evidence of an increase in productivity growth in the 1980s is significant. The data that form the basis of these estimates go only as far as 1988 and are subject to some revision. This limitation, in conjunction with evidence of vigorous productivity growth in U.S. manufacturing during the 1980s, could lead to further upward revisions of productivity growth during this period. 4  One might argue the importance of recent productivity growth in the United States not being far out of line with historical experience. What really matters is the performance of the United States relative to the other major industrial nations--especially, in the minds of some commentators, Japan. Figure 3 shows the productiVity performance of the United States relative to a small group of industrialized nations since 1870. The data are from Maddison (1987) and show average annual rates of productivity growth during four phases of recent economic history. The first phase, 1870-1913, corresponds to what many analysts consider the golden age of American ca pitalism. During this period, the United States experienced the highest rates of productivity growth of the five countries shown. But even during this period of leadership, productivity grew at only a 2 percent annual rate. The period from 1913 to 1950 covers the two world wars and the Great Depression. Again, the United States grew faster than any of the other countries in this group, although the choice of dates no doubt exaggerates the size of the gap between the United States and the follower countries. During the golden age of postwar economic growth, from 1950 to 1973,  Figure 3  Phases of Growth in GOP Per Hour Worked Average annual growth rates (percent)  7 6  ,,. .,.  •"  ...... France - - Gennany _.- Jilpan  -U.K  3  /  ~,  ,  5  .  /  '.  . " .,'",/. . . . I  -U.S.A-  ".  '  ,'"  '"  ""'... ...  I ?~ ,,", 1.·- ,  "  \. "  :.~.~>'. ~~ ,~ •  ,.  .:... "  O+---------,-------r---------, 187(H913  1913-50  1950-73  197:Hl4  NOTE: The figures for Germany refer to the former West Germany. SOURCE: Maddison (1987).  Federal Reserve Bank of DaUas  Figure 4  Growth of Real Per Capita GDP in the G-7 Average annual growth rate (percent) 10  ...... ,  9  ,  ...... _....... - - -  , "  "  7  .. .  6  '"  5  ----~~~ __ ~  4  .•..••... ~.~.:.~  ,  France Germany Italy UK.  ,  --.":.~.::.::.~...... ......  . .·. -~-.4. ~. ~:~:~:~:;;=~,~.~.~ .-.-._._._  .............,  3  2  "  Canada US.A Japan  ",  ...  r-'7~::::......._=::::::::.-..~~5:::;.~ ...:-?f ..::.;:~..:....;  O-f---------,-------r--------, 1950-59  1960-69  1970-79  I9SD-89  NOTE: The figures for Germany refer to the former West Germany SOURCE: Summers and Heston (1991)  productivity in the United States grew 2.5 percent a year, but this rapid rate of growth was not enough to put the United States at the head of the group In fact, the United States placed last in terms of productivity growth during this period, being outperformed even by the enited Kingdom. Following the oil-price shock in 1973, all the major industrial economies experienced a productivity slowdown. The United States remained in last place, achieving productivity growth of only 1 percent annually between 1973 and 1984. Productivity growth rates that were already low by international standards fell further; the fact that the other industrial economies were also experiencing a slowdown was little consolation. Unfortunately, these figures cover only the early 1980s. We know that there is some evidence of a pickup in U.S. productivity growth during the 1980s, so it would be interesting to know how the U.S. petformance compared with that of the other major economies during this period. Consistent productivity estimates are not available for the full decade, but we have information on real per capita GDP until 1988. Figure 4 summarizes the comparative growth performance of the United States during the postwar period based on the estimates in the Penn World Table Mark 5 (Summers and Heston Economic Review-January 1992  1991). These figures are similar to Figure 3 in tem1S of growth rates duting the postwar period. Figure 4 piaL'; the average annual growth rate of real per capita GDP for each of the G-7 countries (United States, Canada, United Kingdom, France, Germany, Italy, and Japan) during the past four decades. 6 In terms of real per capita GDP growth, the United States was outperformed by almost all the other countries for most of the period. The stability of the U.S. growth rate around 2 percent over the entire sample is noteworthy. Also, whereas all countries experienced a slowdown in their growth rates in the 1970s and 1980s, the slowdown in the United States was less dramatic, and, in fact, the United States experienced a modest rebound in the 1980s. Again, one must interpret these figures with caution, because the dates correspond with the longest peacetime expansion in U.S. history. Let us now turn from performance at the aggregate level to the performance of the manufacturing sector. Many commentators consider the maintenance of a healthy and vigorous manufacturing sector as crucial to the long-term health of the nation. Figure 5 shows the average annual growth rates of productivity in the manufacturing industry by decade for the G-7 countries. 7 Productivity growth in manufacturing in all the countries accelerated between the 1950s and the 1960s, except in Germany. A dramatic slowdown in productivity growth occurred in the 1970s in all countries. In all three decades, the United States shared last place with the United Kingdom, averaging productivity increases of slightly more than 2 percent a year.  6  In this article, Germany refers to the former West Germany  7  One must interpret the figures for productivity in manufacturing with caution Note that the measure of hours used in the calculation ofthese productivity figures is hours worked, except for the United States, where the measure is hours paid However, Jablonski, Kunze, and Flohr Otto (1990) show that the use ofhours paid versus hours worked makes little difference to long-term average annual rates of productivity growth in the United States Note also that these measures of productivity across countries are limited to trend (growth rate) comparisons rather than level comparisons, because the levels of manufacturing output across countries are not strictly comparable  5  Figure 5 Growth of Manufacturing Output Per Hour in the G-? Average annual growth rates (percent) 12 _•••• Canada  U.SA. _.- Japan  10  ••.... France  ~.  '.'..  - - Germany ,  8 ~~~  ~'~~  - - Italy , - UK  :  ':::::~:;:,:::.:;.;;;;;~":~:~:.:.• :. ....~ ~ ......................................................................  ""..  ~~~  2  0+--------,-------.--------; 1950-59  1960-69  1970-79  1980-89  NOTE: The figures for Germany refer to the former West Germany SOURCE: Monthly Labor Review, various issues.  The most remarkable feature of Figure 5 is what it tells us about the 1980s Whereas manufacturing productivity continued to decline in the other G-7 countries, both the United States and the United Kingdom experienced a noticeable acceleration in productivity growth. For the decade as a whole, productivity in U.S. manufacturing grew at the same rate as productivity in Japanese industry. This review of the evidence raises two questions. First, why has the United States grown at a slower pace than the other major industrial economies in the postwar period? Second, why did all the industrial economies experience a dramatic slowdown after 1973? One of the oldest explanations of differences in growth rates between rich and poor countries attributes it to catch-up or convergence This basic idea is that when comparing countries over long periods, an inverse relationship exists between a country's initial level of output, or productivity,  • Baumol, Blackman. and Wolff (1990)  6  and its subsequent growth rate. The mechanism whereby poor countries tend to grow more rapidly than rich countries is usually identified as diffusion oftechnolo~the technical know-how of the advanced economies can be freely obtained and copied by the poorer countries, thereby enabling them to catch up with the leaders. Differences in rates of return to capital across countries will also tend to produce convergence. If the marginal product of capital falls as capital is accumulated, poor countries will offer better investment opportunities than will rich countries, other things being equal. Higher rates of capital accumulation in poorer countries translate into more rapid rates of growth. Over time, the return to investment falls as capital is accumulated, bringing the rates of growth in poor countries closer to those in the rich countries. The convergence hypothesis helps answer both questions. The surge in U.S. productivity growth in the 1940s and 1950s can be interpreted as catching up on the backlog of technical innovations that occurred during the Great Depression but were not fully exploited at the time H The rapid rates of growth of the other industrial economies in the postwar period can be interpreted as catching up with the world leader, the United States Even the slowdown in the 1970s and 1980s is consistent with a narrowing of the gap between the leader and the followers, although this surely is not the complete explanation; the two oil-price shocks during the 1970s undoubtedly played an important role in slowing growth. Figure 6 clearly illustrates the convergence phenomenon among the G-7 countries. This figure shows the log of per capita GDP for each of the G-7 countries over 1950-88 An obvious sample selection problem is that the sample consists of ex post rich countries. A more appropriate test of the convergence hypothesis would require examination of the comparative performance of countries that are rich and poor in an ex ante sense. I will return to this issue below. The level of real per capita GDP in the United States in 1950 was $8,665, measured in 1985 international prices. By 1988, this level had increased to $18,339. In terms of levels, the United States began and ended the period in first place. Canada finished in second place at the beginning and end of the period, growing at a rate very similar to that of the Federal Reserve Bank of Dallas  Figure 6 Real Per Capita GDP in the G-7 Log levels, 1985 international prices 10  9.5  :::~:.~._._._.~..  USA 9  Canada .... ~.......... •...  , ,. ..' ,"  85  .,....(.~ •• ~ ..~,.,.  UK  France ..... ~.;; , , 8  lIaIY"  ,. ,  ;/  ,. ,  ~,  "'.  ~~ .. ~  ".'  "  Genn ny','  75  ",.~-r'"  .  ,  Japan'  7-1----.-----.-----,-------,.---------, 1950  1960  1970  1980  1988  NOTE: The figures for Germany refer to the former West Germany SOURCE: Summers and Heston (1991)  United States. Japan's rapid growth rates took per capita GDP from $1,275 in 1950 to S12,209 in 1988, moving Japan from last place at the beginning of the period to fourth place at the end of the period. The big loser in relative terms is the United Kingdom, which slipped from third to sixth place over the postwar period.  A growth model to illustrate convergence 9 The convergence hypothesis states that, other things being equal, poor countries will tend to grow faster than rich countries. This idea is also sometimes referred to as the benefits ofbackwardness. The idea, which originates in the literature on economic history and development, was first formalized in the neoclassical growth model of Solow (1956). The key assumption that predicts convergence in this model is diminishing returns to the factor that can be accumulated, namely capital. Poorer countries-that is, those with low levels of capital per worker-will offer higher rates of return to investment than will richer countries. This leads to higher rates of investment, and thereby higher growth rates, in the poorer countries. Next I will discuss the key elements of the neoclassical model that produce the convergence result. Economic Review-January 1992  The standard neoclassical growth model is formulated in a closed economy setting. I will follow this convention below, but it raises the question of how allowance for open economy considerations alters predictions about convergence. The answer is that for any given values of the parameters of tastes and technology, a variety of considerations leads us to predict that convergence will be even more rapid in an open economy than in a closed economy. The factors that are usually singled out are the mobility of capital in response to return differentials, the mobility of labor in response to wage differences, and the scope for poor countries to imitate the technology of the rich countries. Thus, convergence could occur even in a world of closed economies. In other words, even without investment in poorer countries by developed countries, even without scope for labor from poor countries to move to rich countries, and even without the possibilities for poor countries to imitate the technology of rich countries, the neoclassical model would lead us to expect more rapid growth, on average, in the poorer countries than in the richer countries. Thus, the convergence predictions made by the model outlined below are conservative estimates, given the reality of extensive linkages between the countries of the world. Let us start with the standard assumption that aggregate output is produced with capital and labor. lO A common specification of the production function that relates inputs to final output is the Cobb-Douglas specification, (1)  Here, Y denotes aggregate output, A denotes the state of technology (or alternatively, the level of multifactor productivity), K denotes the capital stock, Le RI denotes the effective labor force, and a is a fixed parameter. Note that the effective labor force differs from the actual labor force, L (assumed  • TnlS seclion ;s concerned wilh technical delsHs of the argumfJnI and can be skipped without loss of continUity. The anatysls that fol/owsls basedon BarroandSa/a·/-Marlln (1990)  7  for simplicity to be equal to the population), by the factor e~t, which denotes the productivity of each worker. Each worker becomes more productive over time as a result of exogenous technological progress occurring at the rate g. For simplicity, we will assume that labor is supplied inelastically. We can rewrite the production function in intensive form as (2)  where j7 denotes output per effective worker (Y/ Le Mt ) and Ii denotes capital per effective worker (K! Le~I).  Output of the single final product can be used for either consumption or investmentY The aggregate resource constraint can be written in terms of quantities per effective worker as (3)  ji=c+i,  c 1- a -1  u(c)=---,  1- ()  .  where (}>O is a constant parameter. Households are infinitely lived, care about both current and future consumption, and are assumed to maximize the lifetime utility function  (8)  K = I - OK,  where a dot over a variable denotes a time derivative (x = dx/ dO and 0 is the rate at which capital depreciates. From the definition of capital per effective worker we obtain the following expression for the evolution of Ii:  <SI  The interpretation of this equation is straightforward. The growth rate of the capital stock per effective worker is equal to the excess of output per effective worker over the desired level of consumption per effective worker and the amount of investment per effective worker required to maintain the capital-labor ratio, (n+g)1i, and replace worn-out capital, oli. Household preferences are defined over per capita consumption, c (=C! L), rather than consumption per effective worker. At any point in time, the utility derived from a given level of per capita consumption is given by the function (7)  where T denotes gross investment per effective worker. Growth in the aggregate capital stock is determined by net investment, which is simply gross investment less depreciation; that is,  (4)  following expression:  (.  ]  .!.K k=k K-n-g,  where n is the growth rate of the labor force. Consolidation of these equations gives us the  " I am abstracting from the existence of government for simplicity. Government spending and taxing policies have no effect on the steady-state rate ofgrowth of this economy They may augment or impede the transitional dynamics that underlie the convergence result but are unlikely to change them qualitatively  The term en! measures the size of the household, which is assumed to grow at the same rate as the labor force, 11. The term e pt reflects discounting of future consumption by the household; the parameter p>O is the pure rate of time preference. The outcome of the household's constrained optimization problem is the following equation for the evolution of per capita consumption over time:  (9)  1 ( aA k- a-l c =-  c  ()  s: ) -u-p.  The growth rate of per capita consumption outside of the steady-state equilibrium is determined by the difference between the net marginal product of capital and the rate of time preference. The rate of time preference is the same as the real rate of interest in steady-state equilibrium. When the rate of return to physical capital accumulation exceeds the steady-state real interest rate, net investment is positive. Federal Reserve Bank of Dallas  In steady-state equilibrium, the quantities of output, capital, and consumption per effective worker are constant. Using equation 9 above, the steady-state capital stock per effective worker is determined by  (0)  Marginal product of capital  aACk*y,-l=o+p+ag,  where the star C*) denotes a steady-state value. Figure 7 illustrates this equilibrium. 12 If the initial value of the capital stock, ~)' is below the steady-state level, Ii*, equations 2, 6, and 9 will determine the evolution of capital, output, and consumption over time. Because the effective capital stock rises during the transition to the steady state, the per capita capital stock must grow at a higher rate than the steady-state growth rate, g Therefore, the growth rate of Ii is declining near the steady state. The growth rate of Ii, however, is not monotonically declining over the entire range of Ii. The possibility of a nonmonotonic relationship stems from the dependence of the saving rate on Ii. If the saving rate is assumed fixed Cas in Solow's (956) original exposition of the model), then the growth rates of Ii and yare monotonically decreasing in Ii. In other words, poor countries always grow faster than rich countries. Despite the relative simplicity of this model, it is not possible to solve explicitly for the paths of capital, output, and consumption. Rather, one must use some approximation to the exact solution. Log-linearizing the model about the steady-state equilibrium, we obtain the following expression for output per unit of effective labor:  where the parameter [3>0 determines the speed of convergence to the steady state and is a complicated function of the underlying parameters of tastes and technology. The average growth rate of per capita output over an interval of time from ~) to q, + T) can then be written as  (2)  Figure 7  Determination of Steady-State Equilibrium  ~ln[YI"+T)=g+o-e-fJ1)ln[z.::). T  Y,,,  T  Y,,,  Thus, given the steady-state level of output per effective worker, which is determined by the steady-state level of capital per effective worker, Economic Review - January 1992  o+P+crgl---...-----"'''o::_--------  Ii Capital per eHective worker  the growth rate of per capita output will be higher the lower the level of per capita output is at the beginning of the period. Convergence to the steady state is more rapid when [3 is greater. The relationship between [3 and the fundamental parameters of tastes and technology is discussed further in Bano and Sala-i-Martin (990).  Empirical studies of convergence We can test for convergence in two ways. The first and most common approach involves looking at the relationship between the level of GDP or productivity in a country at some initial date and the subsequent growth rate of GDP or productivity over some interval. If convergence is present, this relationship should be negative. Another approach is to look at the dispersion of the log levels of GDP or productivity across countries and how it changes over time Convergence predicts that the dispersion should fall over time. 15 I will focus on  " For a more extensIVe diSCUSSIon of the determination of steady-slale equilibrium in the one-sector neoclassical model see Wynne (1991). Barro and Sata-l-Manm (1990) reler 10 Ine IWO types 01 convergence as /3-convergence and (J'-convergence, respectively, and show that the lormer does not necessarily  imply the latter Most 01 the empirical studies that we will consider locus on {3-convergence.  9  studies that use the first approach, because they are more relevant to assessing the role of convergence in the comparative growth performance of the United States. This simple approach to testing the convergence hypothesis entails estimating the unconditional convergence relationship (3)  ~ln(X;";))=a+bln(x:), T  XI"  "  where Xi denotes real GDP per some measure of population (usually either total population or the number of adult equivalents) or some measure of productivity in country i. If convergence is present, the estimate of h should be negative. Mankiw, Romer, and Weil (1990) report estimates of this type of relationship for a sample of ninety-eight countries over 1960-85. They estimate h=O 09, with a standard error of 0.05. Because this estimate is not statistically different from zero, it implies that we do not see convergence across all countries. When the sample is restricted to include only the developed countries of the OECD or the G-7, the estimate of b turns significantly negative. Grier and Tullock (1989) report similar findings. The finding that convergence seems to be occurring among only the developed countries could reflect one of two things. First, it may be the case that a little bit of backwardness is a good thing but too much is a hindrance. Thus, only relatively developed economies are in a position to free-ride off the technological know-how of the leader, because only they have what Abramowitz (1986) terms the "social capability" (that is, skilled labor force, institutions, etc.) to exploit the most advanced technology. The alternative is that convergence among the rich countries reflects sample selection bias. De Long (1988) made this point about Baumol's (1986) study of convergence among sixteen major industrial economies since 1870 and showed that convergence did not appear to hold in a sample of ex ante developed countries. When proxies for human capital are added to the basic convergence relationship, Barro (1991), Mankiw, Romer, and Weil (1990), and Baumol, Blackman, and Wolff (1990) find negative coefficients on the initial level of output or productivity. This finding of conditional convergence supports both models that emphasize the 10  need for a broader concept of capital accumulation when studying growth (that is, models that include human capital accumulation) and the social capability notion of Abramowitz These and other empirical studies show that convergence seems to be present among the richer (usually OECD) countries and to have occurred more rapidly after World War II than before. It is not clear, however, how much of these findings rd1ect a sample selection bias. We also do not have a clear idea of the relative importance of technological diffusion and capital accumulation in the convergence process. It is possible to find convergence across a broader group of countlies once we control for some measure of human capital. Can we estimate how much of the shortfall in U.S. growth is attributable to catching-up by the other industrialized countries? Baumol, Blackman, and Wolff (990) use the result.<; from convergence relationships estimated over 1870-1979 and 1950-79 and Maddison's (982) productivity figures to sUppOl1 their thesis that the United States has performed better than commonly thought. They calculate that over the longer sample, u.s. productivity growth was some 20 percent greater than that predicted by the simple convergence relationship. Even in the postwar sample, u.s. productivity growth was some 15 percent greater than what the simple relationship predicted. Using Summers and Heston's (1991) data for real per capita GDP covering 1950-88, I estimated the following unconditional convergence relationship for the OECD countries: GR5088  =  0.136 - 0.013 In(RGDPI950), (0.021) (0.003) R2  =  0.54  where GR5088 is the average annual rate of real per capita GDP growth between 1950 and 1988 and RGDP1950 is the level of real per capita GDP in 1950. Standard errors are in parentheses. This regression tells us that more than half the crosscountry variation in GECD growth rates over this period can be explained by the starting positions of the different countries. Real per capita GDP in the United States was $8,665 in 1950. The unconditional convergence relationship predicts that per capita GDP should have grown at an average annual rate of 1.8 percent between 1950 and 1988. The actual growth rate was 2 percent over Federal Reserve Bank of Dallas  Table 1  Excess of Actual over Predicted Growth Rates of GOP Per Adult, 1960-85 (Percent) Unconditional 98 Countries OECD United States Japan Germany  1.1 5.1 2.2  1.2 4.7 2.2  Conditional 98 Countries OECD United States Japan Germany  -.6 26 - 2  1.4 4.5 1.4  NOTE: Figures are based on results reported in Mankiw, Romer, and Weil (1990).  lhi.~ period. suggesting thaI the United States did slightly hetter than expected. The difference het\\ een the t\\"o grO\\"th rates. \\"hile seemingly insignificant. translates inlo a difference of more than Sl,OOO in real per capita GDP in 19HH. We can also use lhe results reported in Manki,,', Romer, and Wl'il ( 1990) to assess the comparati\"e performance of the United States since 1960. Using their con\"ergence relationships, I calculated the excess of actual o\"er predicted grO\\"th rates for lhe l'nited Stales, Japan. and Germany. For the~e calculations, I used the relationships estimated for the full sample of ninety-eight countries and for the smaller sample of twenty-two OECD countries (Iceland and Luxembourg are omitted from Mankiw, Romer, and Weil's (1990) .sample). Tahle 1 presents the results of these cilculations. The interpretation of the results in Tahle 1 is straightforward. The United States grew ahout 1 percentage point faster o\"er this period than \\"e "'ould hm'e predicted hased on its 1960 !e\"el of GDP per working-age person. Japan and Germany. hcm"e\"er. posted e\"en more impressi\'e gains. with Japan growing ahout 5 percentage points more than predicted hI' the unconditional cC)\1\"ergence relation and Germany growing 2.2 percentage points more. Controlling for invest-  Economic Review - January 1992  Illent share, the growlh rale of lhe effective lahor force, and school enrollmenl, both Germany and the I Jnited States underperformed in this period relati\"e to the large group of countries. Relative to lhe narro\\"er OECD grouping. ho\\"ever, hoth countries did hetter llun expected. although not a.s well as Japan The reason for the discrepancy hetween the results from lhe conditional convergence relationships for lhe full set of countries ami those for the OECD suhset is not immediately obvious, hut it may he related to the inclusion of lhe newly industrializing Pacific countries (Tai\\"~1l1. Hong Kong, Singapore, and South Korea) in the larger grouping. Do\nick amI !\guyen (19H9), in their study of cOI1\'ergence of per capita GDP levels among OECD countries in the post,\ ar period. show that, once we allow for the calch-up phenomenon, several of the richer countries that appear to have heen laggards in terms of their growth rates really grew at rates closer to the a\"erage for the group as a whole. From 19'10 lo 1960, the average annual growth rate of per capiu GDP for the OECD as a \\"hole (excluding JljXll1) was 3.12 percent. In the United States, the average annual growth rate of per capita GDP ,,"as 1.79 percentage points 100\"er o\"er the same period. But adjusted for cyclical bias and catch-up, the shortfall in the United States was only 0.26 percentage points. From 1960 to 1973, when the OECD grew at an average annual rate of 3.9'1 percent, the United States grew at an average annual rate that was 1. IS percentage points lower. Adjusted for cyclical bias and catch-up, the United States grew 0.02 percent faster than the OECD a\"erage. The shcJl1fali in U.S. growth over 197~'1 translates into a rate that is 0.61 percentage points higher when adjusted for catch-up. The evidence presented in this section supports the convergence hypothesis, alheit in modified form. The fact that human capital plays an imp011ant role in long-run growth is hardly controversial, but it is not clear from the existing empirical studies whether its impOt1ance stems from its role in enhancing a countty's social capability (and thereby the ease with which the technology of the leader may be copied) or whether it plays a role more akin to that of physical capital accumulation. From the U.S. point of view, the existence of convergence puts a somewhat hetter light on its recent growth performance. 11  Prospects for the future Productivity growth is the key to achieving sustained increases in standards of liVing, which inevitably raises the question of what determines productivity growth. A myriad of determinants have been suggested, among them accumulation of human capital, technological innovation, quality and quantity of publicly provided inputs, and so on. The proposition that education plays an important role in raising productivity is noncontroversial, and current concerns ahout the quality of education in the United States reflect this fact. The issues are well known and do not warrant repeating here. Recently, several authors have emphasized the role of publicly provided capital in the form of infrastructure as a key determinant of productivity performance. I This idea is that certain key inputs to production must hy their nature he publicly provided; standard examples are roads and air-traffic control systems. The adverse effects of highway congestion or airport delays on productivity are ohvious. The state of neither education nor the public capital stock gives grounds for optimism about achieving higher rates of productivity growth in the near future. Greater grounds for optimism may be found in some of the changes in the U.S. economy over the past ten years. It is possible to argue that the corporate restructurings of the 1980s have resulted in more efficient organizational structures and have set the stage for rapid productivity growth over the next decade. Likewise, technological innovations, such as the spread of personal computers, may be reaching a stage in which rapid productivity gains can be realized. But the bottom line must be that because we have only a limited understanding of what caused productivity movements in the past, we can say little with certainty about the course of productivity in the future. The convergence hypothesis can explain some of the difference between the growth rates of the United States and Japan over the postwar period. But Japan has grown at an even faster rate than the already rapid rates predicted by condiI  .. See Aschauer (19898 1989b)  12  tional and unconditional convergence relationships. The relevant question, then, is how much of this excess can be attributed to superior public policy in Japan and thus achievable by the United States' The fact that the United States seems to have done somewhat better than convergence would lead us to have expected raises the question of whether we could have done even better.  Conclusion A thorough understanding of where we have been is essential to an understanding of where we are headed. I have tried to show that where we have been is not quite as bad as where we thought we were. Several stylized facts emerged from our review of the long-run growth experience of the United States. First, the growth rates of both labor productiVity and per capita GDP seem remarkably stahle around average annual rates of about 2 percent. This stability also seems to characterize productivity growth in manufacturing in the postwar period. During the 1970s, a global slowdown occurred in productivity growth in the developed countries, not just in the United States. During the 1980s, U.S productivity growth rebounded, especially in manufacturing. Much of the cross-country variation in growth rates is attributable to convergence. Thus, it is not very surprising that countries like Japan and Germany, which had a significant proportion of their capital stock destroyed during World War II, grew more rapidly than the United States for most of the postwar period. However, despite the fact that the United States did not lead in terms of growth rates, it did lead in terms of levels of output per capita and continues to do so. To make good policy, it is necessary to understand just how much of U.S. historical experience and comparative experience can be attributed to differences in policy and how much is beyond the policymakers' control. While catchup explains some of the strong performance of Japan and Germany, so do factors that are more amenable to policy, factors such as capital deepening-that is, increasing the amount of physical capital per worker-and increasing the amount of human capital embodied in each worker through education. Federal Reserve Bank of Dallas  Appendix  Forthe sake of comparability, it is important that the extension of the productivity estimates to 1988 be done in a manner that is as consistent as possible with that used by Maddison (1982) in constructing the historical series. The three components of the productivity figure are an estimate of real GOP, an estimate of the average number of hours worked, and an esti mate of total employment. While estimates of each of these series are readily available for the United States, Maddison makes several adjustments to them to allow comparisons across countries and over time. The appendix to Maddison (1982) explains the details of these adjustments. Maddison's GOP figures are from OEeO sources for 1950 on. These figures are adjusted so that output is measured at 1970 U.S. relative prices-thus, output and productivity levels can be compared across countries. I use the estimate of GOP presented in Summers and Heston (1991), in which real GOP is evaluated at 1985 purchasing power parities. Maddison calculates total employment as the size of the labor force less the number unemployed. He includes the number of 14and 15-year-olds employed in the total employment figure, whereas the standard definition used in the United States includes employment of only 16-year-olds and older. I calculate two measures of employment: one that includes and one that excludes employment of 14- and 15-year-olds. The biggest difficulties are with the hoursworked measure. The measure of hours presented in standard sources in the United  Economic Review - January 1992  States is a measure of hours paid rather than hours actually worked. Maddison makes two adjustments to the standard series to exclude paid holidays, vacation, and time lost due to sickness. I use the standard series on hours worked (that is, hours paid) and adjust it using the adjustment ratios provided in Jablonski, Kunze, and Flohr Otto (1990). Table A1 compares my estimates of GOP growth and productivity growth (including and excluding employment of 14- and 15Table A1 GOP Growth (Percent)  Productivity Growth (Percent)  1950-60  3.2* 3.2  2.4* 2.3/2.3  1960-70  3.9* 3.9  2.6* 2.6/2.6  1970-79  3.4* 3.1  1.9* 1.3/1.2  1950-73  3.7* 3.6  2.6* 2.4/2.4  Interval  NOTE: The numbers with stars (') are from Maddison (1982), and those without stars are the author's estimates, except for the GDP growth figures, which are from Summers and Heston (1991)  year-olds from the labor force) over various intervals with estimates presented in Maddison (1982). Some of the differences in this table may be attributable to differences in the way growth rates are calculated. It is not clear (Continued on the next page)  1  Appendix-Continued whether Maddison (1987) uses exact growth rates or logarithmic approximations. I use exact growth rates in all my calculations, including those using levels data from Maddison (1982). Thus, the average annual growth rate of a variable x from t to (t + T) is calculated as  (x;r  r 1  -1.  The numbers with stars (*) are from Maddison (1982), while those without stars are my own estimates. The first number in the productivity growth column is an estimate based on the inclusion of 14- and 15-yearolds in the employment figure, while the second number excludes them. Inspection of Table A1 reveals that the estimates of GOP growth are identical for 1950-70. For 195073, there is only a 0.1 percentage point difference, but for 1970-79 the difference is 0.3 percentage points. This difference may be due to revisions to the estimate of 1979 GOP used by Maddison. The productivity growth estimates are the same for 1960-70. Table A2 makes similar comparisons between the figures I use for GOP and pro-  1'1  Table A2 GOP Growth (Percent)  Productivity Growth (Percent)  1950-73  3.7* 3.6  2.5' 2.4/2.4  1973-84  2.3' 2.4  1.0' 1.0/1.0  Interval  n  NOTE: The numbers with stars are from Maddison (1987), and those without stars are the author's estimates, except for the GDP growfh figures, which are from Summers and Heston (1991).  ductivity and those reported in Maddison (1987). The Summers-Heston (1991) estimates of GOP put growth over 1950-73 0.1 percentage points lower than Maddison's estimate but add 0.1 percentage points to the estimate of growth over 1973-84. Productivity growth over 1973-84 is 1 percent based on Maddison's estimate and my estimates, regardless of whether the employment of 14and 15-year-olds is included or excluded. On the basis of this comparison of estimates, I am confident that the extension of Maddison's productivity series to 1988 is consistent with the earlier figures.  Federal Reserve Bank of Dallas  References Abramowitz, Moses (986), "Catching Up, Forging Ahead, and Falling Behind," Journal ofEconomic History 46 (2/June): 385-406, reprinted in Moses Abramowitz (989), Thinking About Growth (Cambridge, Mass.: Cambridge University Press, 220-42). Aschauer, David A. 0989a), "Public Investment and Productivity Growth in the Group of Seven," Federal Reserve Bank of Chicago Economic Per>,pectives 13 (September/October): 17-25. - - - 0989b), "Is Public Expenditure Productive?" Journal ofMonetary Economics 23 (2/March): 177-200. Barro, Robert]. (991), "Economic Growth in a Cross Section of Countries," Quarterly Journal ofEconomics CVI (2/May): 407-43. - - - , and Xavier Sala-i-Martin (990), "Economic Growth and Convergence Across the United States," NBER Working Paper No. 3419 (Cambridge, Mass.: National Bureau of Economic Research). Baumol, William]. (986), "Productivity Growth, Convergence, and Welfare: What the LongRun Data Show," American Economic Review 76 (5/December): 1072-85. - - - , Sue Anne Batey Blackman, and Edward N. Wolff (990), Productivity and American Leadership' The Long View (Cambridge, Mass.: MIT Press).  DeLong, ]. Bradford (988), "Productivity Growth, Convergence, and Welfare: Comment," American Economic Review 78 (5/December): 1138-54. Dowrick, Steve, and Due-Tho Nguyen (989), "OECD Comparative Economic Growth 195085: Catch-Up and Convergence," American Economic Review 79 (5/December): 1010-30. Gordon, Robert]. (969), "$45 Billion of U.S. Private Investment Has Been Mislaid," American Economic Review 59 (3/June): 221-38. Grier, Kevin B., and Gordon Tullock (989), "An Empirical Analysis of Cross-National Economic Growth, 1951-80," Journal ofMonetary Economics 24 (2/September): 259-76. Jablonski, Mary, Kent Kunze, and Phyllis Flohr Otto (990), "Hours at Work: A New Base for BLS Productivity Statistics," Monthly Labor Review 113 (2/February): 17-24. Kendrick, John W. (961), Productivity Trends in the United States (Princeton: Princeton University Press). Maddison, Angus (987), "Growth and Slowdown in Advanced Capitalist Economies: Techniques of Quantitative Assessment," Journal ofEconomic Literature 25 (2/March): 649-98. - - - (982), Phases of Capitalist Development (Oxford: Oxford University Press).  - - - , and Edward N. Wolff (988), "Productivity Growth, Convergence, and Welfare: Reply," American Economic Review 78 (5/December): 1155-59.  Mankiw, N. Gregory, David Romer, and David N. Weil (990), "A Contribution to the Empirics of Economic Growth," NBER Working Paper No. 3541 (Cambridge, Mass.: National Bureau of Economic Research).  Council of Economic Advisers (991), Economic Report ofthe President, February 1991 (Washington, D.C.: Government Printing Office).  Romer, Paul M. (986), "Increasing Returns and Long-Run Growth," Journal ofPolitical Economy 94 (5/0ctober): 1002-37.  Data Resources Incorporated (991), Review of the U.S. Economy: Ten-Year Projections (Lexington, Mass.: Data Resources Incorporated).  Solow, Robert M. (956), "A Contribution to the Theory of Economic Growth," Quarterly Journal ofEconomics LXX O/February): 65-94.  Economic Review - January 1992  15  Summers, Robert, and Alan Heston (991), "The Penn World Table (Mark 5): An Expanded Set of International Comparisons," Quarterly Journal ofEconomics CVI (2/May): 327-68. U.S. Department of Commerce, Bureau of the Census (975), Historical Statistics ofthe United States: Colonial Times to 1970 (Washington, D.C.: Government Printing Office). Wolff, Edward N. (991), "Capital Formation and Productivity Convergence Over the Long Term," American Economic Review 81 (3/]une): 565-79. Wynne, Mark A. (991), "The Long-Run Effects of a Permanent Change in Defense Purchases," Federal Reserve Bank of Dallas Economic Review, January, 1-17.  Federal Reserve Bank of Dallas  David M. Gould Economist Federal Reserve Bank of Dallas  Free Trade Agreements and the Credibility of Trade Reforms  T  he economic success of outward-oriented, protrade countries such as South Korea and Taiwan has led other developing countries to attempt trade Iilx:ralization More countries arc trying to become integrated into world markets and are dropping old notions that economic development can only be achieved in isolation. Indeed, in the I9HOs a number of Latin American countries moved toward more open, liberalized economics. IIo\Vever, of those countries that have attempted trade liberalization. some have achie\'ed great success. \\·hile others have failed miserably. A common thread in unsuccessful trade liberalization attempts is the failure of the domestic gO\"l'rt1ment to create a believable trade Iiber~diza­ tion policy. If the private sector perceives trade reform as only temporary, investment will not move from the protected impol1-competing sectors to the more crficient expOl1 sectors. In fact, a noncredible liberalization attempt may be worse than no attempt ~lt all. I For example, Peru's noncredible tra(k~ liberalization attempt in the early 1980s was extremely costly because investors, believing that taritr~ would rise again, impol1ed massive quantities of foreign goods and decreased domestic investment. lvlost economic analysis of trade liberalization credibility has focused on domestic prohlems, such as the rrivate sector's difficulty understanding the government's true motives. It is sometimes the case that developing countries shm\" interest in trade reform only because reform is a precondition to assistance from international organizations such as the World Rank and the International MOnelalY Fund (IMF). The economic literature, however, has yet to address the role that foreign markets can play in either enhancing or diminishing domestic credihility. The more open foreign markets ~lre, the Economic Review - January 1992  greater will be the benefits of trade reform, and the greater will be the likelihood of a successful liheralization. On the other hand, if foreign markets arc closed, the benefits of trade reform and the probability of a successful liberalization will be smaller Indeed, a developing country may eliminate all protection but still fail to entice some investment from the protected import-competing industries to the export industries if investors anticipate a protectionist response in foreign markets. Mexico, for example, is liberalizing trade and increasing incentives to shift resources out of the protected sectors and into the more efficient export sectors. Yet. some producers may be hesitant to increase exports to the United States in fear that exporting too much will trigger a protectionist response from the United States. In this article, I explore the role of foreign markets in credible trade liberalization, and I argue that credible trade reform in develoring countries may require two clements: a credible domestic policy and a credible foreign policy that discourages protection in foreign markets. I also discuss how pa.I1icipation in a multilateral or bilateral free trade 3greement can enhance credibility on both the domestic and foreign fronts of trade liberalization. Using the proposal for a free trade  I wish to thank Stephen P A Brown, John K Hill, Miguel Savastano, William C Gruben, and Evan F. Koenig for helpful comments Any remaining errors are solely my responsibility , For example. see Calvo (1987) and Edwards and van Wijnbergen (1985)  agreement between the United States and Mexico as an example, I explain two ways such agreements heip maintain open trade. First, a free trade agreement can weaken internal political pressures to reverse liberalization. Second, an agreement can defuse the political trigger mechanism that creates barriers to trade in foreign markets.  Problems with credibility To be effective, economic policy must be believable. In fact, both theory and evidence suggest that lack of credibility can have very high costs. Consider, for example, the Federal Reserve's policy of stopping the high inflation rate in the late 1970s by reducing money supply growth. Some economists argue that because the private sector at first did not believe the Federal Reserve's resolve to decrease inflation, prices and wages were set too high relative to the future monetary base. As the nominal demand for money exceeded its supply, the lack of liquidity unleashed recessionary forces in the early 1980s. Only when the private sector began to believe the Federal Reserve's commitment to low inflation did the U.S. economy begin to grow again. Similar problems can result in the case of a trade liberalization when the private sector believes the policy is only temporary. Calvo (1987) shows that significant economic welfare cost,> can result from temporary trade liberalization. In his analysis, temporary liberalization decreases economic welfare because it works like a tax on future consumption that distorts prices between time periods. The cost of these price distortions is from the misallocation of consumption opportunities between time periods. Edwards and van Wijnbergen (1985) examine a similar problem and show that opening  2  3  4  18  Rodrik (1989) For example, see Edwards (1989) and Choksi and Papageorgiou (1986) In formal economic terms, the government's objective is to redistribute income from individuals with a low marginal utility of income to those with a high margmal utility 01 mcome  an economy to foreign capital flows, in the absence of a credible trade liberalization policy, can exacerbate preexisting distortions in trade, which, in turn, decrease economic well-being. The theoretical literature about trade liberalization has addressed the problem of credibility from three angles: the inconsistency between trade reform policies and other government policies, the lack of incentive over time for the government to adhere to trade reform, and the lack of private sector information about the government's incentives. 2 The first credibility problem arises from the inconsistency between government trade liberalization policies and other macroeconomic policies. An example of this problem is trade reform in a country in which the government tries to keep its exchange rate fixed while it also expands its fiscal deficit and continues rapid money growth. By maintaining the exchange rate fixed while pursuing expansionary fiscal and monetary policies, the government is artificially maintaining a high value of its currency and distorting the price of foreign goods relative to domestic goods. In this case, allowing free trade will eventually lead to a balance of payments crisis as consumers deplete the country's foreign currency reserves to purchase artificially cheap imports. Because the private sector will eventually realize that the government's free trade policies are incompatible with its other macroeconomic policies and that this combination will lead to a crisis, the trade reform will lack credibility. This issue is perhaps one of the more common sources of credibility problems, particularly in Latin America.' A possible solution to this problem of inconsistent policies is to first stabilize the macroeconomic environment and then proceed to trade liberalization. Second, government liberalization policies may be time-inconsistent. A time-inconsistent policy is one in which the government, at some date after liberaliZing trade, has an incentive to break its promises. For example, Staiger and Tabellini (1987) analyze the case in which a government wants to redistribute income from the rich to the poor 4 Under this objective, a free trade policy will never be credible because the government will always have the incentive to provide more protection than expected to import-competing Federal Reserve Bank of Dallas  firms whenever the relative price of imports decreases. When the price of imports falls, the import-competing sector becomes relatively poor; consequently, the government has an incentive to renege on its free trade promise and redistribute income though protection to these sectors. Free trade, then, is not a credihle policy hecause the private sector understands the government's incentive structure. Creating a credible trade reform that is timeconsistent is problematiC because it depends on the government's ability to precommit to a particular trade regime. If a governmem cannot precommit to free trade, it may have to pursue a time-consistent but second-best policy of partial tariff protection. In other words, the government may never he ahle to create a credihle policy committed to complete free trade; it may, however, be able to create a credible policy with less protection. A third type of credibility problem develops when the private sector does not know enough about the government's true incentives. Whereas the problem of credibility in the previous examples developed when the private sector knew too well the behavior and incentives of the government, here not enough is known. In some circumstances, individuals cannot determine how sincere the government is about its willingness to liberalize. For instance, often governments show imerest in free trade only because it is a precondition to foreign assistance from the World Bank, IMF, or another source. Furthermore, the political process may determine that the government committed to free trade today may not be the government in power tomorrow. It may also be the case that individuals cannot determine the relative power of constituent groups that influence future policy. To overcome the credibility problem caused by inadequate information about the policies of tomorrow's governments, the government may attempt to signal that it intends to reform trade by partaking in behavior that would be too costly for a government only faking reform. Rodrik (1989) demonstrates that overshooting free trade by actually subsidizing imports may achieve this goal, while Aizenman (1991) shows that public investmem in the export sectors can signal the government's commitment to trade reform. Economic Review - January 1992  The role of foreign markets in a credible trade reform The economic literature discussed in the previous section has addressed the domestic issues concerning credibility problems but has yet to examine the role that foreign markets can play in either enhancing or diminishing credibility. A fully informed private sector in the liberalizing country may be unwilling to shift investment out of the protected import-competing sectors and into the export sectors for fear that a protectionist response in the foreign countries will hamper access to foreign markets. Even if present trade barriers in foreign markets are low, the belief that they may increase can deter long-term investment commitments in the exports sector In this case, free trade lacks credibility not because the private sector believes the government's policies will change in the future, but rather because the private sector does not believe foreign markets will remain open. For example, consider a country that liberalizes its trade sector to eliminate protectionist policies that favor investment in the importcompeting sectors and dissuade investment in the export sectors.' A firm-a shoe manufacturer, for example-that once only sold its products domestically will discover that its input costs have declined and it can now make a profit selling its products in foreign markets. Suppose also that the firm considers the domestic government's commitment to liberalization to be sincere, but believes that foreign markets may impose tariffs on its shoe exports. 6 The firm may export some shoes, but not enough to maximize profits at the current foreign price of its product. The firm will limit additional long-term capital commitments required to expand capacity if it believes that its current comparative advantage will be eliminated by a  5  Protection that favored the import-competing industries before liberalization would simultaneously discourage other industries. including exports  6  These concerns are not hypothetical Developed countries have often shown a considerable willingness to protect their industries against import competition See the box titJed "The Rise of Protectionism in Developed Countries ..  19  foreign protectionist response. This situation may be especially important for manufacturing industries that require large, specialized capital investments to increase output, rather than some types of industries that can expand output with small long-term capital investments. Consequently, even when a trade liberalization policy is credible in terms of being compatible with other macroeconomic policies, time-consistent, and fully known and compatible with government incentives, the threat of a protectionist response in foreign markets reduces credibility.  Free trade agreements and the credibility of trade reform By limiting the protectionist response in foreign countries, an international free trade agreement can help enhance the credibility of trade refonns. A free trade agreement also reduces the risk of changes in the domestic stance toward trade liberalization by limiting the influence of constituent groups on future government policies. A free trade agreement can enhance credibility by signaling a government's commitment to free trade and by increasing the costs of a future policy reversal. The sustainability of free trade agreements, as in any legal document or negotiated contract, depends on how comprehensive and well written their terms are, as well as on their ability to resolve future unforeseen conflicts. Unlike most legal contracts, enforcement of these agreements is entirely voluntary, and their credibility does not depend on the objectives and interests of only two parties, but on the relative power of competing interests within two or more subscribing countries. Protectionist measures usually develop not because they are in the best interest of a country, but because the beneficiaries of protection are typically few and individually have much to gain. (See the box titled "The Rise ofProtectionism in Developed Countries. ") Industries will lobby for protection up to the point at which the last dollar spent on lobbying equals the expected additional benefits of increased protection. 7  , See Krueger (1974)  20  According to the political-economy approach to trade policy, the level of protection reflects  an equilibrium outcome of competing pro- and antitrade interests. Magee, Brock, and Young 0989, 3), for instance, explain the development of trade policies in the following way: "Policies play the same role in politics as prices play in an economy: both are equilibrating variables that adjust until opposing forces are balanced." According to political-economy models, changes in trade policies can tell us something about the relative power of pro- and antitrade groups. Thus, the negotiation of a free trade agreement (which is like a contest between groups) signals something that may not have been directly observable in the economy, namely an increase in the relative power of protrade interests and the sustainability of free trade This signal will help increase investors' confidence that free trade will not be temporary. Unlike the typical piecemeal increase in protection that results from the lobbying activity of individual industries, a free trade agreement may actually imply a general change in the rules of the game in favor of protrade interests. In other words, the agreement may not only reduce tariffs and nontariff barriers to trade but also decrease the benefits of lobbying for protection, enhancing the long-term sustainability of the reform. A free trade agreement changes the incentives and returns of protectionist lobbying in two basic ways. First, once implemented a free trade agreement will bind together diverse export industries in their opposition to increases in protection. These industries will devote more resources to oppose any potential increase in domestic protection because of the threat of a retaliatory response and possible abrogation of the entire agreement. Usually, it does not pay for anyone exporter to lobby against a single protective policy if the costs of such a policy to that firm are relatively small. However, with a free trade agreement, an exporter will have the incentive to lobby actively against any increases in domestic protection in fear that an increase in protection would induce a retaliatory response against its own products from the other members of the agreement. Second, by making protection more visible, a free trade agreement increases the costs of lobbying for protection. The inclusion of a Federal Reserve Bank of Dallas  mechanism to settle disputes makes hiding and implementing protectionist policies, which lower national income, more difficult The free trade agreement between the United States and Canada illustrates this effect. The u.S.-Canada agreement contains two mechanisms to address disputes. The first mechanism requires that each nation inform the other in writing of any proposed or actual change that might affect the agreement. Each country can request discussions on any matter of concern, and any problem that is not resolved within thirty days can be referred to a joint u.S.-Canada trade commission. The second mechanism deals with antidumping and countervailing duty cases. Each country still applies its own antidumping and countervailing statutes, but-at the request of either government-a binational panel can be set up to review the decisions of each country's administrative agencies. In contrast to the disputesettlement mechanisms of the General Agreement on Tariffs and Trade, problems encountered under the U.S.-Canadian free trade agreement are solved rather quickly Furthermore, smaller firms are more likely to contest foreign protection because the two governments, not the firm, hear the costs of the appeal process Thus, the inclusion in a free trade agreement of adjudication clauses that allow foreign and domestic industries to bring complaints about trade protection to an arhitrage or judicial review committee creates an environment in which protection is more likely to be noticed and more difficult to impose. In summary, the negotiation of a free trade agreement signals an increase in the relative power of groups that favor free trade and enhances the credibility of trade reform policies. Furthermore, approval of a free trade agreement changes the rules of the protectionist game by altering the reward stmcture of lohbying. Free trade agreements unite groups that favor free trade in opposition to protectionist lobbying and make protectionist lobbying more likely to be noticed.  Implications for  credibility to Mexico's trade liheralization policy. The agreement can do this in two ways. First, it can weaken domestic political pressures from special interest groups to reverse trade liberalization. Second, it can defuse the political trigger mechanism in the United States that creates barriers to trade. An executive from a large pulp mill in Chihuahua, Mexico, recently said, "Policies in Mexico have always changed when presidents did, but free trade gives a sense of permanence to the very sound policies of this administration" H Indeed, a free trade agreement lends credibility to the Mexican trade reform because it signals an increase in the relative strength of the protrade groups within Mexico and the United States and raises the costs of protectionist lobbying. Currently, tariffs and other barriers to trade between Mexico and the United States are relatively low, but fears of a future increase in trade barriers dissuade investors from making long-term investments in Mexican exportoriented industries. U.S. and Mexican investors will make more long-term commitments if they expect markets to remain open. A credihle free trade policy will induce a substantial increase in trade of those products that require long-term capital investments-greater even than what would be indicated simply on the basis of decreases in tariff rates. Although Mexico's gains from trade are likely to be relatively larger than those of the United States, the type of gains will be similar. Both countries will benefit from a reallocation of resources to industries that reflect each country's comparative advantage. The United States' greatest gain will likely be in the export of services to Mexico, whereas Mexico's greatest gain will likely be in the exports of manufactured goods to the United StatesY Of course, the terms of the free trade agreement are yet to be seen, but even if the current barriers to trade are not changed, the mere existence of an agreement will improve expectations that the barriers will not increase.  u.s. and Mexican trade  Perhaps from Mexico's point of view, the most significant feature of the proposed North American Free Trade Agreement is that it will lend Economic RevieW-January 1992  • Rodo/fo Figueroa of Grupo Chihuahua. quoted in "Latin Turnaround:' Wall Street Journal. May 24, 1991, P 1 9  US International Trade Commission (1991)  21  As Harberger (1991) notes, "The big message of the [free trade agreement] is that 'this border is open and is going to stay open.' "  Conclusion In this article, I explored the role of free trade agreements in trade liberalization and argued that credible trade reform in developing countries may require two elements: a credible domestic policy and a credible foreign policy that discourages protection in foreign markets. Participation in multilateral or bilateral free trade agreements can enhance credibility on both the domestic and foreign fronts of trade liberalization.  First, a free trade agreement can weaken internal political pressures to reverse liberalization. Second, an agreement can defuse the political trigger mechanism that creates barriers to trade in foreign markets. The benefits of the proposed free trade agreement between the United States, Canada, and Mexico will derive from not only a decrease in tariff and nontariff barriers but also from a credible commitment that future trade barriers will not be erected. It is the expectation of lasting free trade, in addition to low trade barriers, that will entice long-term investment away from protected importcompeting sectors and into the export sectors where a country's comparative advantage lies.  Federal Reserve Bank of Dallas  The Rise of Protectionism in Developed Countries  Despite a substantial drop in tariff rates since the creation of the General Agreement on Tariffs and Trade (GATT), protection increased dramatically in the 1980s. 1 Most new protection is based on administrative regulations that are usually implemented through bureaucratic procedures rather than enacted by law. Administered protection has originated primarily in developed countries, particularly the United States, and is typically directed at imports of manufactured goods. This trend is a consequence of the increased competition in the international market for manufactured goods, where a large number of developing countries have gained a comparative advantage. Increased competition and the ease with which recent protection has been imposed can create problems for developing countries attempting to liberalize trade. Developed countries initiated more than 1,700 actions of administered protection between 1980 and 1985 (Table 1). These actions included countervailing duties, antidumping suits, safeguard actions, and other complaints about unfair trading practices. Under GATT, safeguard actions can be taken when imports threaten or cause serious injury to domestic producers. Antidumping actions are imposed when exports are sold at less than their cost and countervailing duties are levied to counteract the use of subsidies. GATT provided these protective actions to level the playing field; however, the ease with which countries can impose restrictions has led to their use in ways that undermine free trade. Problems arise largely because administered protection is imposed by the country  Economic Review - January 1992  claiming to be harmed, not through international procedures. For example, an antidumping complaint by a U.S. computer firm is investigated by the U.S. government, which determines the U.S. response. The ability to impose administrative trade restrictions-without a legislative process or an international procedure-offers extraordinary power to vocal domestic constituents in favor of protection. Because administered protection is not usually subject to political debate, its costs often remain hidden. That makes administered protection an especially attractive means for politicians to cater to special interests without protests from other groups harmed by protection. 2 (Continued on the next page)  1  GATT was created after World War II as a way to provide the rules and procedures for countries to dismantle trade barriers, particularly tariffs. In practice, GATT provided seven rounds, or meetings, to negotiate international trade policy. An eighth round, the Uruguay round, was added in 1986 but has not yet been concluded.  , There are circumstances, although uncommon, in which free trade is not the optimal policy An example is the case in which a country has monopoly power in world markets. In this in· stance, a country can use tariffs to extract monopoly rents from the rest of the world. Another case is an industry that is sUbject to increasing returns to scale or has technological spillovers to other industries. In this case, however, the optimal strategy is to subsidize the industry, not to impose tariffs Nevertheless, even in such unusual circumstances, free trade is usually the best policy because of the difficulty in determining which industries justify the need for protection, as well as the potential for the political abuse of protection to satisfy special interest groups. See Krugman (1990) for more detailed arguments on this subject  2  The Rise of Protectionism in Developed Countries-Continued Protectionism, however, is not limited to administrative regulations. In general, protectionist policies (whether administered or enacted by law) arise because the costs and benefits of lobbying are different for the gainers and losers offree trade. As Pareto (1927), Olson (1965), and others more recently have noted, if the benefits of a protectionist policy are concentrated among a small number of firms and the costs are spread over a large number of consumers, it may not pay for any one consumer to incur the cost of actively opposing such a policy.3 In other words, the benefits of protection are concentrated, whereas the costs are diffuse. 4 For example, in 1984 the U.S. Federal Trade Commission estimated that quotas and tariffs on sugar imports cost U.S. consumers $1.266 billion and benefited the sugar industry and government by $783 million. s Clearly, this policy was not in the national interest; it represented a national net loss of $483 million. The loss to each consumer, however, was only about $5 a year. That amount is hardly enough to be noticed by most consumers or to motivate them to lobby against sugar protection. To each individual sugar producer, however, the protection represented hundreds of thousands of dollars, which more than compensated for the costs of lobbying for protection.  The current rise of protectionism in developed countries can be seen as a political response by their industries to economic pressures in the world trade system, what Shagwati (1988, 62) calls the double squeeze. The first squeeze has come from Japan and advanced to newly industrialized countries (NICs), such as Taiwan and Hong Kong, which are competing in the market for hightechnology manufactured goods. The second squeeze has come from other NICs and developing countries, such as Malaysia and Thailand, which are competing in the laborintensive manufactured goods market. Since the 1980s, both forces have prompted a structural adjustment in developed countries away from manufacturing and toward other sectors, which, in turn, has touched off strong political pressures to create trade barriers.  (Continued on the next page)  , See, for example. Brock and Magee (1978) and Mayer (1984). 4  Endogenous tariffs may arise as a result of other asymmetries. such as asymmetries in the representation of economic interests within the government organization. See Hillman (1989) or Week-Hannemann (1990) for a summary of this literature.  5  Tarr and Morkre (1984) This example is also presented in Krugman and Obstfeld (1988,189).  Federal Reserve Bank of Dallas  The Rise of Protectionism in Developed Countries-Continued Table 1 Number of Administered Protection Cases Initiated, 1980-85 Protection measure Safeguards United States· Australia Canada EEC' Countervailing duties United States b Australia Canada EEC Antidumping actions United States C Australia Canada EEC Other unfair trading practices United States· Total United States Australia Canada EEC  1981  1982  1983  1984  1985  2 1 0 3  6 0 1 1  4 1 2 1  2 2 0 1  6 0 0 1  3 0 1  23 4 4 7  8 0 3 0  10 0 0 1  123 2 1 3  21 7 3 2  51 6 2 1  39 3 3 0  252 18 12 7  22 62 25 25  14 50 19 47  61 78 72 55  47 87 36 36  71 56 31 49  65 60 36 42  280 393 219 254  l8  19  73  39  33  39  231  60 63 28 28  49 50 20 49  261 81 75 59  109 96 39 39  161 62 33 51  146 63 40 42  786 415 235 268  1980  198G-85  'European Economic Community - Not available a US Trade Act, Section 201. • U S Trade Act, Section 701. , US Trade Act. Section 731 • As defined in the U S. Trade Act, consist of unfair importing practices (Section 337), unfair trading practices (Section 301), and market disruption (Section 406). SOURCE: Finger and Nogues (1987), p 708, Table 1.  Economic Review - January 1992  2  References Aizenman, Joshua (991), "Trade Refom1s, Credibility, and Development," NBER Working Paper Series, no. 3600 (Cambridge, Mass.: National Bureau of Economic Research, January). Barro, Robert, and David Gordon (983), "A Positive Theory of Monetary Policy in a Natural Rate Model," Journal ofPolitical Economy 91 (August): 589--610. Bhagwati, Jagdish (988), Protectionism (Cambridge, Mass.: MIT Press). Brock, William, and Stephen Magee (978), "The Economics of Special-Interest Politics: The Case of the Tariff," American Economic Review 68 (May, Papers and Proceedings, 1977): 246-50. Calvo, Guilermo (987), "On the Costs of Temporary Policy," Journal ofDevelopment Economics 27 (October): 245--61. Choksi, Am1eane, and Demetris Papageorgiou, eds. (986), Economic Liberalization in Developing Countries (Oxford: Basil Blackwell). Edwards, Sebastian (989), Exchange Rates, Devaluation, and Adjustment: Exchange Rate Policy in Developing Countries (Cambridge, Mass.: MIT Press). - - - , and Sweder van Wijnbergen (985), "The Welfare Effects of Trade and Capital Market Liberalization," International Economic Review 27 (February): 141-48. Finger, ). Michael, and Julio Nogues (987), "International Control of Subsidies and Countervailing Duties," World Bank Economic Review 1 (September): 707-25. Harberger, Arnold (991), "A Study of Mexico's Real Exchange Rate" (University of California, Los Angeles, mimeographed).  26  Hillman, Arye (989), The Political Economy of Protection (London: Harwood Academic Publisher). Krueger, Anne (974), "The Political Economy of the Rent-Seeking Society," American Economic Review 64 (June): 291-303. Krugman, Paul (990), Rethinking International Trade (Cambridge, Mass.: MIT Press). - - - , and Maurice Obstfeld (988), International Economics: Theory and Policy (Glenview, Ill.: Scott, Foresman and Company). Magee, Stephen (978), "Three Simple Tests of the Stomper-Samuelson Theorem," in Issues in International Economics, ed. P. Oppenheimer (London: Oriel Press) 138--53. - - - , William Brock, and Leslie Young (989), Black Hole Tariffs and Endogenous Policy Theory: Political Economy in General EqUilibrium (New York: Cambridge University Press). Mayer, Wolfgang (984), "Endogenous Tariff Formation," American Economic Review 74 (December): 970-85. Olson, Mancur (965), The Logic of Collective Action: Public Goods and the Theory ofGroups (Cambridge, Mass.: Harvard University Press). Pareto, Vilfredo (927), Manual ofPolitical Economy (New York: A.M. Kelly). Rodrik, Dani (989), "Promises, Promises: Credible Policy Reform via Signalling," Economic Journal 99 (September): 756-72. Staiger, Robert, and Guido Tabellini (987), "Discretionary Trade Policy and Excessive Protection," American Economic Review 77 (December): 823-37.  Federal Reserve Bank of Dallas  Tarr, David, and Morris Morkre (984), Aggregate Costs to the United States of Tariffs and Quotas on Imports (Washington, D. c.: Federal Trade Commission).  Week-Hannemann, Hannelore (990), "Protectionism in Direct Democracy," Journal of Institutional and Theoretical Economics 146 (September): 389-418  US. International Trade Commission (991), The Likely Impact on the United States ofa Free Trade Agreement with Mexico, USITC Publication 2353, Investigation No. 332-297 (Washington, D.C.).  World Bank (987), World Development Report (New York: Oxford University Press).  Economic Review -January 1992  27  Thomas F. Siems Senior Economist Federal Reserve Bank of Dallas  Quantifying Management's Role in Bank Survival anking liter~lture often cites mismanagement as the most imporunt cause of bank failures. Pantalone and Platt ( 19i"i7l state that "it is the management of the hank that determines success or failure Most often, hanks fail hecause they have chosen paths that are excessively risky for the returns that they receive and hecause these paths make them particularly vulnerahle to adverse economic conditions." Sehallos and Thomson (1 (90) recently wrote that "the ultimate determinant of whether or not a hank fails is the ability of its management to operate the institution efficiently and to e\'aluate and m~lllage risk" Additionally. in a study hy the Office of the Comptroller of the Currency to uncover specific reasons for bank failures, Graham and Horner (1988) concluded that "the difference hetween the f~liled hanks and those that remained healthy or recovered from problems was the caliber of management.·' No specific quantit~ltive measure currently exists to assess the quality of hank management. Rather. generally bank examiners regularly \'isit hanks and conduct on-site examinations to assess them. From these examinations. examiners give hanks a CAVlHf, rating. which is an overall evaluation of a bank's health and is an acronym based on the follOWing five factors:  B  Capital adequacy, A<;set quality. Management qu~dity, Earnings ability. and Liquidity. Examiners score each of these factors as a single number from one to five. with one being the strongest rating, and develop an overall CAMEL rating from one to five from the factor scores. The Economic Review - January 1992  use of CAMEL factors in evaluating a hank's health has become widespread because of hoth its simplicity and use hy regulators. Financial data and relationships are the principal ingredients for scoring capital, asset quality, earnings, ;ll1d liquidity. Assessing management quality, however, is considered qualitative and therefore requires professional judgment of a bank's compliance with policies and procedures, aptitude for risk-taking, development of strategic plans, and the degree of involvement hy the bank's officers and directors in the decisionmaking process. In this article. I present a new model to quantitatively measure bank management quality. This model considers the essential intermediation functions of a hank and uses multiple inputs and outputs to compute a scalar measure of efficiency. [ compute the dficiency metric, a proxy for management quality, using a linear programming technique known as data enuelopment ana~Y:-il's (OEA). DEA has been used successfully to provide a new definition of efficiency in many applications. including schools (Bessent and others 1982), courts (Lewin. Morey, and Cook 1982), strip mines (Hyrnes, Eire, and Grosskopf 1984), and health care (Nunamaker 19H'S and Sherman 1984). I huilt the efficiency model presented in this article on the notion of using the metric as a variable in a bank-failure prediction equation. Given that we wish to differentiate banks that fail from those that survive, I include in the model variables that I believe capture the importance of  I wish to thank Robert T. Clair, Gerald P O'Oriscoll, and Kevin J Yeats for helpfUl comments and suggestions Any remaining errors are solely my responsibility  29  management in a bank's survival, that is, the necessary input allocation and product mix decisions needed to acquire deposits and subsequently make loans and investments. While this metric, which is based strictly on publicly available financial information, may not replace an examiner's assessment of management completed during an on-site examination, it can assist examiners as an early warning tool. I While others may disagree with my choice of input and output variables for this model, the empirical results support the argument that management quality is very important to a bank's survivaV  The importance of management An institution's management quality is important to its long-term survival Cates (985) states, "[Blank failure, which affects only a handful of banks, is caused by mismanagement; mismanagement, furthermore, of the basic, old-fashioned risks of banking like lending, liquidity, and control ,. Bank managers make the decisions and fashion the plans that define the direction for the institution. Management determines allocation of the bank's resources, establishes the internal controls and procedures, organizes strategic plans, and responds to changes in the external environment. Because banks operate in a competitive, uncertain, changing environment, bank managers must learn to deal with and manage the inherent risks. As Kaufman (986) states, "[Tlo survive in a risky world, banking firms must cope with risk and manage it."  I  2  Siems (1991) found the management qualily metric to be a highly slgmficant variable In both one-year-ahead (twelvemonth to eighteen-month) and two-year-ahead (twentyfour-month to thirty-month) bank-failure prediction models Furthermore, the predictive accuracy of these models significantly improved with the inclusion of this variable The objective is not so much the theoretical development of  a model to prove efficiency but rather the construction of a measure to help distinguish banks that survive from those that fail The model developed herein is a useful proxy for management quality because the variables focus on bank management's basic allocation. control, and product mix decisions  30  Measuring bank management quality To proxy bank management quality, I present a model to measure the efficiency by which management can transform inputs into outputs. I built the model on the notion of using the measure to help discriminate banks that survive from those that fail. Hence, the model includes variables that are most descriptive of management's decisionmaking role in a bank's intermediation process. Because of the multitude of functions performed and decisions made by management, a descriptive model of bank management quality must contain several inputs and outputs. Single ratios, such as total operating income divided by total operating expense, suffer from several limitations. For example, while such ratios may provide an overall measure of operational efficiency, they fail to indicate the resource allocation and product decisions made by management because the numerator and denominator are aggregate measures. Moreover, when several nonaggregated single input-output ratios are used to assess the myriad of decisions made by management, the ratios collectively present a morass of numbers that give no clear evidence of the efficiency of a bank. One ratio may show that the bank is highly efficient, while another displays a highly inefficient operation. Sexton (986) argues that such ambiguity makes ratio analysis ineffective in measuring true efficiency. Clearly, a model that captures bank management's allocation and control decisions is needed. Such a model requires the identification of several inputs and outputs. What are the allocation and control decisions that managers make to operate a bank?  Essential functions in banking The banking industry has changed over the years, but the functions of operating as a financial intermediary have remained basically unchanged. For the model presented herein, commercial banks are represented in a two-stage process in which they first acquire deposits and then bundle together these monies to make loans and investments. In a manner similar to that of Berg, Forsund, Federal Reserve Bank of Dallas  Figure 1 Commercial Bank Decisionmaking Model: Two-Stage Representation  ..  t  I  /  -~.~~------.---_  .. - . __..- - _ _ _-_ ...  ...----"-, Stage 1  Core Deposits  _A_nrad_--, . _OI=h:.;e,..:Oe=IlOS'=ts:....Oepo<.1S '-  Total Interest Exoense  _  ... ..... - ..... _..  Full-Time Equivalent Employees Salary Expenses Premises and Fixed Assets Other Noninterest Expenses  @ C:;:lS ------I  Sla98 2; - - - EamlflQ Assets MaKe LQans and - - - Total In aOlSI Income Investments I-----------'  Purchased Funds  and Jansen (989), the model interprets deposits as intermediary outputs, which is also consistent with the view advocated by Kolari and Zardkoohi (1987). This two-stage representation aids in interpretation because deposits are produced in the first stage and used as inputs to the production of loans and investments in the second stage.  A new model to measure bank management quality I have identified a subset of inputs (resources) and outputs (products and services) that I believe model the quality of commercial bank management from a failure-prediction perspective. That is, the model includes certain inputs and outputs that I feel are critically important to the management of a bank. 5 Considering again the two main functions of a bank-acquiring deposits and making loans and investments-I developed a two-stage model that employs multiple inputs and outputs to assess management efficiency. As Figure 1 shows, both stages utilize four inputs that primarily represent operating expenses. These four inputs are the number of full-time equivalent employees, salary expenses, value of premises and fixed assets, and other noninterest expenses as reported on a bank's Consolidated Report of Condition and Income to its primary regulator. The operation of all bank activities involves labor, materials, Economic Review-January 1992  machines, and buildings, and management certainly has a great deal of discretion concerning the allocation of these resources. Management determines the number of employees needed to perform desired functions at a desired level of service. They establish salary levels, and they determine the types of facilities to build, where to build them, and how to furnish and operate them. 4 Management also decides (possibly as a result of previous decisions) what other noninterest expenses to incur, such as legal  3  4  The number of DEA efficiency models applied to banking are relatively few and recent. In general. researchers have developed models to measure either the relative efficiency of bank branches or the overall efficiency of the banking industry Charnes and others (1990), Berg, Forsund, and Jansen (1989), Rangan and others (1988), Parkan (1987), and Sherman and Gold (1985) provide five different DEA models applied to banking Another useful reference concerning efficiency in banking is Evanoff and fsrailevich (1991), in which they discuss the concept of efficiency and define the means to measure it Their article also includes a review of relevant literature regarding inefficiency in the banking industry Some of these input factors, such as salary expenses, are largely determined by market forces, however. bank man· agement ultimately makes decisions regarding the overall level of salaries, which can influence past-due collections, loan portfolio quality decisions, etc  31  assistance and administrative expenditures related to maintaining and liquidating foreclosed real estate and other assets. Another input associated with acquiring deposits in stage one is total interest expense. Management establishes the types of deposits they wish to attract and the interest rate levels offered to depositors. \V'hile interest rates are largely influenced by market forces and monetary policy, management makes decisions regarding the composition of deposits, which directly influences total interest expenditures. Purchasedfunds, the final input in the model, represents funds needed in addition to all other deposits to adequately service the bank's investments and provide needed liquidity.) High purchased funds signal that management has not attracted enough stable or core deposits for the volume of loans it is currently servicing. When this happens, the bank must buy funds and subject itself to increased liquidity risk.!> Humphrey (1991) argues that the user costs of demand deposits, small time and savings deposit,>, and purchased funds must be included as appropriate inputs along with labor and physical capital. He states that operating costs are less than onethird of total banking costs at typical banks and therefore do not give an overall picture of productivity in banking. Other models that show purchased funds as an appropriate banking input include Rangan and others (1988) and Triplett (1991). The obvious output from the first stage of the model is total deposits. For a bank, deposits can be considered as either an input or an output. In this model, core deposits are interpreted as an intermediary output These are relatively stable deposits obtained by the bank. Finally, the second  5  Purchased funds include federal funds purchased and securities sold under agreements to repurchase, demand notes issued to the U S Treasury. other borrowed money, lime certificates of deposit of $100,000 or more, and openaccount time deposits of $100,000 or more  stage has two outputs: earning assets, which include all interest-earning assets, and total interest income. The outputs in this new model also appear to be a direct result of management's decisions. Core deposits represent stable funds desired by the organization for lending and investment purposes. Earning assets and total interest income result from management's decisions regarding where to invest fund'>. Management makes decisions concerning the relative riskiness of each asset in which it invest'). The most etTicient banks allocate resources and control internal processes by effectively managing the number of employees, salary expenses, facilities, other noninterest expenses, total interest expenses, and purchased funds while working to maximize core deposits, earning assets, and total interest income. To do this, efficient bank managers establish controls and procedures that keep operating expenses relatively low while still attracting an adequate volume of core deposits (so that purchased funds remain low). Prudent managers also devise loan policies that discriminate creditworthy borrowers from those in danger of default to increase the value of earning assets and operating income. By evaluating the riskiness of potential loans, management is better able to choose which loans to make. Failed banks historically exhibited more lending problems (that is, mismanagement of the loan portfolio) than other operating inefficiencies. Graham and Horner (1988) found that 86 percent of the failed banks they studied had inappropriate lending policies, including liberal repayment terms, collection practices, or credit standards Overall, bank managers must integrate policies and techniques for managing the money position, providing liquidity, lending profitably, and investing rationally in a practical asset/liability management framework. The most efficient banks do this by controlling operating expenses, managing interest rate sensitivity, utilizing risk management techniques, and strategically planning for the bank and its markets for the future.  Quantifying bank management quality , In addition, Gunther (1989) states that "a high reliance on purchased or wholesale funds, such as large certificates of deposit, federal funds purchased, and securities sold under agreement to repurchase, is often associated with high asset growth and aggressive lending strategies "  32  Data envelopment analysis computes a bank's efficiency in transforming inputs into outputs, relative to its peers. First developed by Federal Reserve Bank of Dallas  Charnes, Cooper, and Rhodes (978), who built on the concept of technical efficiency by Farrell (957), DEA provides a new definition of efficiency DEA is a linear programming technique that converts multiple inpurs and outputs into a scalar measure of efficiency.' This conversion is accomplished by comparing the mix and volume of services provided and the resources used by each bank compared with all other banks. Each bank is evaluated against a hypothetical bank with an identical output mix that is constructed as a combination of efficient banks. DEA identifies the most efficient banks in a population and provides a measure of inefficiency for all others. The most efficient banks are rated a score of one, while the less efficient institutions score between zero and one. DEA does not give a measure of optimal efficiency; it '''ill only differentiate the least efficient banks from the set of all banks (even where all hanks might be inefficient). Thus, the efficient institutions calculated using DEA establish the best practice frontier. DEA was designed specifically to measure relative efficiency using multiple inputs and outputs with no a priori infonnation regarding which inputs and outputs are most important in determining an efficiency score. The relative efficiency of a bank is defined as the ratio of its total weighted output to its total weighted input. Mathematically, this is represented as  weighted input. In general, banks will have higher weights on those inputs that it uses least and those outputs that it produces most. H  Graphical representation of DEA In this section, I illustrate graphically a small problem VisualiZing the concepts underlying DEA will assist with its interpretation for larger and more complex problems Consider five banks, each using one input to produce two outputs. Table 1 shows the levels of the input required to produce the outputs for each bank. Single inpUHJutput analyses can be used to characterize each bank using the single input (INPUT and the two outputs (OUTPUT] and OUTPUT).? In fact, by normalizing each output relative to the level of input required to produce it, each bank can be graphically represented in a two-dimensional space, as Figure 2 shows. In this figure, each bank is represented by a point whose coordinates are simply the normalized output levels shown in Table 1. Any hank located both above and to the right of another bank is clearly more efficient, because it is producing higher levels of both j  1  .'  )  See Seiford (1990) for an eXlensive bibliography of DEArelated publicatIons  L, u,kOUFPUT,k EFFICIENCY I,  =-'~-~-,- - - - -  8  The DEA model for a specific bank can be formulated as a linear fractional problem. which can be eaSilY solved if it is transformed into an equivalent linear program in which the bank's input and output weights are the decision variables A complete DEA solution requires that one such linear program be solved for each bank See Barr and Siems (1991a) for technical details regarding the mathematical formulation of DEA  9  While ratios can provide useful information about efficiency, they fail to accommodate multiple inputs and outputs when accurate objective weights for the inputs and outputs are not known tn this example, if OUTPUT/INPUT, was the only ratio used to measure efficiency, bank A would be rated as the least efficient and bank C as the mosl efficient If management decided to measure efficiency by using a weight of 0 8 for OUTPUT/INPUT, and 02 for OUTPUT/ INPUT,> then bank A would be rated as the most efficienl and bank C as the second least efficient Clearly, with more inputs and outputs the task of measuring efficiency in this manner becomes even more complex and subjective  L,viJNPUF". ;=1  where It ... is the unit weight placed on output r and Vi. is the unit weight placed on input i by the kth bank in a population of banks Using this notation, there are s output variables and m input variables used to calculate efficiency. Now, how should the weights (the u's and v's) be selected? DEA selects the weights that maximize each bank's efficiency score as long as no weight is negative and the weights are universal (that is, any bank should be able to use the same set of weights to evaluate its own efficiency ratio, and the resulting ratio must not exceed one). That is, for each bank, DEA will maximize the ratio of its own weighted output to its own Economic Review -January 1992  33  Table 1  Sample Data for DEA Example Bank  INPUT,  OUTPUT,  OUTPUT2  OUTPUT/INPUT,  OUTPUT/INPUT,  A B  5 6  10 24 40 12 21  5 4 2 4 1  2 4 5 3 3  C  8  D  4  25 24 16 16  E  7  7  outputs while using the same 1<.:\'(.:'1 of input. Figure 3 sh(l\\ s th~lt hanks A. B. and C define the efficient frontier (representcd hy the thick. hlack line) as a piccewise linear curve These hanks have the property that no other hank is supcrior on both dimensions Every l);Ink is in one of three places: on the efficient frontier. on one of the extensions frol11 the frontier to one of the axes. or some\\'here hem'een the origin and the frontier A hank on an extension is called a lI'mkl)' cllicicllt hank That is. the hank \\'ill han' an efficiency score of one. hut it is not technically on the clTicicnt frontier hecause another hank is superior on at least one dimension. Banks that arc inside the piece\vise linear curve that forms the efficient frontier (including the extensions) arc CIlIY>/O!)cc!; hence this analytical technique is called data ClllYl/opment ana/)'si.\~ Each hank gets an efficiency score in terms of its position relative to the frontier. 1" Banks on the frontier arc the !Jcs/ JJILlc/icC' firms. \yhich usc current technologies and mdhods in the hanking industry most efficiently ~l11d therefore hm'e dliciency scores of one. Each enveloped hank is compared with a hypothetical comparison hank.  \\ hich is ~l theoretical point on the rmntier having thc' s~lme output mix as the hank under e\'aluation. The efficiency score for each enveloped hank is simply the ratio of the actu~i1 output levels rml11 the hank to the theoretical k'\'c!s of its hypothetical comparison hank, Figure 4 sho\\'s ho\\' the en iciency scores are computed for each hank. Each h~lnk that forms the frontier (for example. hanks /\. B. and C) has  Figure 2  Normalized Output Levels for Five Banks Output/Input, 6  1\  5  • o  e  • • 3  c  •  2  • E  m Efficiency, as used here refers to a bank's ability to trans-  form a set of inputs into a set of outputs This ability differs from models designed to measure profit maximization A bank might be able to increase proMs by changing J/s product miX but thiS change may not necessarily make the bank more efficient given the mputs and outputs selected for this model  34  o  2  3  4  5  6  Outputilnput,  Federal Reserve Bank of Dallas  Figure 3  Figure 4  DEA Efficient Frontier  Efficiency Scores for Banks 0 and E  Output/Input,  Output/I nput,  6  6  y  A  A  s.-------tL..  s.---------l.....  4  3  3  2  2  .. .. ' .. .. '  • E  ,  E' ,  X  o  2  3  5  6  Output ;Input,  an efficiency score of one. The efficiency score for bank D is 0.92, a number between zero and one that is the ratio of the distance of line segment OD (5.00) to line segment OD' (5.45).11 Because D' (the hypothetical comparison hank for hank D) is on line segment AB, hanks A and 13 form the efficiency reference set for hank D. Clearly, only efficient banks can compose the efficiency reference set for an inefficient hank. Similarly, the efficiency score for hank E is 0.60 and is computed by taking the ratio of OE to OE'. Obviously, bank E is the most inefficient of the five banks. Epstein and Henderson (989) point out that "the concepts of data envelopment, efficient frontier, efficiency score, efficiency reference set, and hypothetical comparison unit are easily extended to higher dimensions and are thus applicahle in a multiple-input, multiple-output context."  2  3  5  6  Oulpu Input  feasible. In other words, any point along the piecewise linear curve is assumed feasible even though there may not be a bank at that particular point on the frontier. Sexton (986) argues that "this assumption is questionable when there are only a small numher of production technologies from which to choose and where such weighted averages have no counterpart in reality." In other words, it may not be reasonable to assume that hank D could change its input or output mix to move to point D' on the frontier. It may be easier for bank D to move toward point 13, because this is a known feasible point on the frontier. The most optimal path for an inefficient institution to hecome efficient is an area for future research. Second, in Figure 3 the efficient frontier was extended vertically from point C to point X and horizontally from point A to point Y. Other  Some limitations of DEA Several assumptions are made in evaluating hanks in this manner. First, hypothetical hanks, like point D' in Figure 4, are assumed to he Economic Review - January 1992  ., See Barr and Siems (1991a) for mathematical computations to calculate individual efficiency scores using the OEA linear programming formulation  35  Figure 5  efficient frontier to change. For example, as Figure 5 shows, if bank C were dropped from the set of banks, only banks A and B would form the efficient frontier. Furthermore, bank E's efficiency score would improve from 0.60 to 0.75 because the efficient frontier (and its extension to the horizontal axis) is closer to bank E with the removal of bank C. Similar comments apply if new banks are added to the analysis; if new banks become part of the frontier, existing banks may have their efficiency scores altered.  Efficient Frontier Change from Dropping Bank C from the Population Output/Input, 8  5i---------;a..  4  The value of DEA to banking 3  c  •  2  ..................  2  3 Quip  5  6  Input  methods of forming the outer edges of the frontier could result in much different efficiency scores for banks near the outer areasY Additionally, using a ray out of the origin to measure efficiency is just one way to compute efficiency.13 Third, the efficiency scores for each bank are computed relative to all other banks under evaluation. Changes in the number of banks in the population and changes in the set of input and output variables in the model can cause the  Bank managers and regulators can use DEA in several important ways. First, the DEA efficiency scores can identify the banks that need the most attention. The least efficient banks can be analyzed on-site more thoroughly to identify specific problems. Second, for all banks with less-than-perfect efficiency scores, a subset of efficient banks--the effiCiency reference set--exists. From this infonl1ation, managers and regulators can formulate strategies to improve the less-than-efficient banks. The DEA results allow an analyst to build a theoretical or hypothetical bank that uses fewer inputs than the inefficient bank but produces the same outputs, thereby increasing the bank's efficiency. Third, DEA can help identify efficient banks that are managed differently from most other banks because they use an unconventional mixture of inputs or produce a different mixture of outputs. Additional in-depth analyses of these banks may help identify some underlying managerial techniques that could improve the performance of other banks. Management quality in surviving and failed banks  12  Banks near the outer edges emphasize certain inputs and outputs to a greater degree than others Because of distortions that this emphasis can create in computing efficiency scores. a minimum level of a given input or output may need to be specified before including a bank in the population  13  I have presented here the original approach in measuring overall efficiency (see Charnes, Cooper, and Rhodes 1978) Seiford and Thrall (1990) describe many recent developments in DEA. contrasltng the original model with newer proposed DEA models  6  To test the usefulness of DEA in measuring management quality, I compared average DEA scores for failed banks with scores for those that survived. Surviving banks are those institutions in operation from 1984 through 1989. Failed banks are those institutions declared insolvent by a regulatory agency sometime between 1986 and 1988. The sample population had 611 survivors Federal Reserve Bank of Dallas  Table 2  Average DEA Scores for Survivors and Failure Groups Date  December 1984 June 1985 December 1985 June 1986 December 1986 June 1987  87:2  87:1  86:2  86:1  .80  .77  .76  .77 .78 .73 .73 .67  .76 .75 .69 .70 .63  .72 .72 .67 .64  .75 72 .72 .65  .74 68 .68  SURV  88:2  88:1  .82 .83 .83 .82 .84 .81  .74 .72 .74 .72 .72 66  NOTE: SURV = Survivors 88:2 = Banks failing 88:1 = Banks failing 87:2 = Banks failing 87:1 = Banks failing 86:2 = Banks failing 86:1 = Banks failing  between between between between between between  JUly 1, 1988, and December 31, 1988 January 1, 1988, and June 30, 1988 July 1, 1987, and December 31, 1987 January 1, 1987, and June 30, 1987 July 1, 1986, and December 31, 1986 January 1, 1986, and June 30,1986  SOURCE OF PRIMARY DATA: Consolidated Report of Condition and Income statements  :lI1d .1 I\) fa iImes, \\'it h tol:i I asscts between S20 million :l11d S500 million, :l11d c:lch hank ',Yas at least threc \'ears old. I ' T r:,ndoml) 'eil'cted the (111 sun i\ ing institutions from the tot:t1 popu!:ltiun of morL' th:ln 12.000 national l'<lIl1ml'1"cial banks. :111c1 thL' Ltilcd institutions \\"LTl' :dso n:ltional IXlllks. To an:t1yl.l' the managcmcnt qu:t1ity Illetric, I di\ idnl till' h:lnks into the following Sl'\'cn groups: SI'I~\ = Suni\ors HH: 2 = Banks failing hu\\ Cl'n Juh- I, 19HH. :Ind Dl'ceml1l'r :) I. 19H!) HH: 1= Banks failing hct\\ Cl'n .!:lnUalY I. 19HH, and June 50. 19HH Banks failing betwccn July I, 19H7. ami Dcce1l1hl'1" ,11, 19H7 H'7:1 = B:lTlks failing het\\ C'L'n January 1. 19W', and ,lune 50. 19,<1-'H(1:2 = Banks failing het\\ l'en ,lulY 1. 19H(J. and Decemhcr :) I. I91i() H6:1 = Banks failing hct\\"l'cn janu:llY I. 1986. and june .--10. I \)x6  I .sing :1 DEA computl'r code described ami documl'nted in Kennington ( 1()HO) and Ali and others ( 19H I J. T computed rL'sulls to c\'aluate the relati\ L' ellil'ienc\ of all h:lnks in the sample Thc 1l1:ltri:-; in T:lhlc 2 sho\\s till' :I\t'ragc DEA scores Ecollomk' Review - January 1992  for each group for each six-month rcriod fr m Deccmher 19K'f to JUlll' 19H7. One call dr:1\\ t\\ 0 general conclusions from this anah-sis Firs!. the closer a bank is to its LliJure dale. the lo\yer its DEA score is. on a\ l'rage For L'x:lI11ple, the :l\'('r:lgl' DEA score for HH: I hanks rebtive to all otlllT h:lTlks in the sampll' at the l'nd of 19H4 ,,,as O.HO. As thl' group's railurl' date approaches (ml)\ ing down the column). thl' :l\'erage DEA scorL' generally declines, In jlllll' 19W:i. the group's :I\'l'rage DEA scorl' \\':IS 0 7 7 By thl' end of 19H5. thL' :l\"t'l"age score had risL'1l slightly to (PH, hut it quickly tdl to 0,75 by,lulll' 19H(1. The gl'lll'ral degradation in till' HH: 1 :lverage DEA scores cOlltillues as time progrl'ssc's. By June 19H7, only six to t\velve months bdore r~lilurl', the group's <In:ragl' DEA score was 0.67. Oyer thesl' [\\"0 ~1l1c1 one-h:dr years (from Decl'J11hl'r 19H+ to  The age and size limitations will allow lor an analysis 01 a key segment 01 the banking industry that has tile greatest need of problem identification Most mId-sized banks (those with $20 million to $300 million in total assets) operate according to the ImanClal intermediary model presented In this article and are roughly 71 percent 01 all banks operating in the United States  37  Figure 6  Average DEA Scores for Survivors and First-Half 1988 Failures Average DEA score  85  .80  ............................................... .•••.•••• 988  75  Fpllures  .....................................  ".'.  70  .........••....  '. 65+----.-----r------r---~---...,  Dec 1984  June 1985  Dec 1985  June 1986  Dec 1986  June 1987  SOURCE OF PRIMARY DATA: Consolidated Report of Condition and Income statements.  June 1987), this group's average DEA score declined by 0.13, while the average score of the survivors declined by less than 0.01. Figure 6 displays the average DEA scores for the survivors and the 88: 1 failure group from December 1984 to June 1987. A similar deterioration in average DEA scores occurs for all the failure groups, Only a few anomalies exist when the average DEA score rises instead of falls for each successive period closer to a group's failure date. In all cases, however, there is a statistically significant drop in average DEA scores from the first date analyzed (December 1984) umil the group's actual failure. Second, the average DEA score for the survivors is higher than that for the failed groups. In Table 2, examine the row that corresponds to June 1986. The average DEA score for the survivors was 0.82. For the failure groups, the scores generally decline as the individual groups' failure  " Evanoff and Israilevich (1991) state that "firms whose management does an inadequate job of utilizing factor inputs may soon find it difficult to survive in the more competitive market. "  3  dates get nearer (reading left to right across the row). The average score for the group failing in the second half of 1988 was 0.72. For banks failing in the first half of 1988, the DEA score was slightly higher at 0.73, but for failures occurring in the second half of 1987 the average score was lower at 0.69. Moving left to right across the row, the average DEA scores continue to decline. To further examine the significance of this result, I performed t tests for statistical differences in the mean scores of the failure groups as compared with those of the survivors. Table 3 shows the t values for the average DEA scores of the failed bank groups compared with those of the survivors for each six-month period. All the comparisons are significant at the 0.01 level for the December 1984 data except for banks failing in the first half of 1988. Here, the t value is 1.72, which corresponds to a significance level of roughly 0.10. The results in Table 3 show that the average DEA scores for failed banks are significantly different from the scores of surviving banks at the 0.01 level for every six-month period up to two and one-half years before the bank's failure. As a group's actual failure date nears, the differences in means generally become more significant.  Conclusion In this article, I have shown that the quality of bank management can be quantified just like capital, asset quality, liquidity, and earnings. I computed this new management quality metric by using data envelopment analysis to proxy the multiple-input, multiple-output transformational efficiency of a bank. The important contribution from this analysis is that by utilizing DEA with the six inputs and three outputs identified herein, the surviving and failing groups can be statistically differentiated on the basis of the resulting efficiency scores. Long before failure occurs, there appear to be significant statistical differences between the quality of management for banks that fail and those that survive. This result, that management is important to the success or failure of a bank, is intuitively appealing and validated statistically in this study. Banks whose managers poorly allocate resources and disregard the needs of their customers and markets have a greater chance of failing. 15 Federal Reserve Bank of Dallas  Table 3  t Ratios for Survivors Versus Failure Groups Date December 1984 June 1985 December 1985 June 1986 December 1986 June 1987  88:2  88:1  87:2  87:1  86:2  86:1  710 10.57 10.04 912 10.62 17.25  1.72 3.95 4.34 6.91 8.62 8.51  3.47 6.35 6.33 9.58 8.69 8.51  4.70 6.81 8.19 9.58 12.63  5.48 7.57 8.13 10.91  5.74 6.85 6.99  NOTE: 88:2 = Banks failing between JUly 1, 1988, and December 31, 1988 88:1 = Banks failing between January 1,1988, and June 30,1988 87:2 = Banks failing between July 1, 1987, and December 31, 1987 87:1 = Banks failing between January 1, 1987, and June 30, 1987 86:2 = Banks failing between July 1, 1986, and December 31, 1986 86:1 = Banks failing between January 1, 1986, and June 30,1986 SOURCE OF PRIMARY DATA: Consolidated Report of Condition and Income statements  The muiliple-input, muilipil.'-oulput efficiency Ille;lsure. or l1unagement Cju:t1ity IlK'lric, developed in this article is L'asier to undersland and :ll1alvze Ihan single in put-out put r;ttios I)F:\ i.s superior 10 .single Lilio ;\Il;lh sis hecause the model :l!lcm s one lo compulL' L'fTiciency h\ e";llllining manageI11L'nt's role in 1ll:lking resource ;t1location and product decisions. Ikcause ha n k m:ll1:tgers ma ke ;t plethora of dL'cisions. a muiliple-inpul. multiple(lutput mock·1 is more suitahle ;\Ild understanclahle lh:ln the 1ll0r:lSS 01 Ilumhers presL'nled in single inpu\-outpU\ r;ltios. in \\'hich one r:ltio mav shcm ;1 highly elriL iL'nl hank :lnd anollllT m:ty sh(m :\Il ind'ficient O!x'I';ltion. [ mock'lnl 1ll;\Ilagement <jU:t1II\ in a t\\·o.suge reprL'.senl:lli'lI1. in \vhich ciL'p< l.sits \',erL> produced in tlw I-Irst .stage ;\Ilcl .suhseCjuenth' usnl in the second .sUge \0 makL' IO:ll1s and il1\est-  Economic Review - January 1992  ments The scleC'lion of vari:thles for the model is of critical importance, and the resulting efficiency measure is highl) sensitin' to till' \'ariahles selected, \"hiil.' economists, h:mkers. and polic\'l11:lkers \\ ill certainly argue ()\ er the appropriate \'ariahles for an efficiency model. the \':triahles ident if iec! here ditlerenti:lte effectively hetween surviving and failed hanks There ;lre manv potential applications for this ne,v model. The hank managcmcnt quality metric could he used as a \ :tri;thle in an earl} \\arning Illodel (sec Barr ami Siems 1991h) ]kgulat()l"s could use the metric to identify the most inellicienl hanks that require the greatest :lllention. For these institutions, regulators could the results or DEA to construct :t hvpothetic:t! L'flicient h:tnk to help the institution focus on prohlem :lre;I.S ..such as ()\'Crulilizcd inputs.  39  ---  -  -------  References Ali, I., A. Bessent, W. Bessent, and J. Kennington (981), "Data Envelopment Analysis of the Efficiency to Decision-Making Units with the DEA3 Code (Version 3.0)," Research Report CCS 410 (Austin, Texas: University of Texas, Center for Cybernetic Studies, July). Barr, R. S., and T. F. Siems 0991a), "A New Paradigm for Assessing the Management Quality of Banking Institutions," Federal Reserve Bank of Dallas Research Paper, forthcoming. - - - , and - - - 0991b), "Predicting Bank Failure Using DEA to Quantify Management Quality," Federal Reserve Bank of Dallas Research Paper, forthcoming. Berg, S. A., F. R. Forsund, and E. S. Jansen (989), "Bank Output Measurement and the Construction of Best Practice Frontiers," Research Report, Bank of Norway, July. Bessent, A., W. Bessent, J. Kennington, and B. Reagan (982), "An Application of Mathematical Programming to Assess Productivity in the Houston Independent School District," Management Science 28 (2): 1355-67. Byrnes, P., R. Eire, and S. Grosskopf (984), "Measuring Productive Efficiency: An Application to Illinois Strip Mines," Management Science 30 (6): 671-81. Cates, D. C. (985), "Bank Risk and Predicting Bank Failure," Issues in Bank Regulation 9 (2): 16-20. Charnes, A., W. W. Cooper, Z. M. Huang, and D. B. Sun (990), "Polyhedral Cone-Ratio DEA Models with an Illustrative Application to Large Commercial Banks," Journal ofEconometrics 46 0/2, October/November): 73-91. - - - , - - - , and E. Rhodes (978), "Measuring the Efficiency of Decision-Making Units,"  European Journal of Operational Research 2 (6): 429-44.  Epstein, M. K., and J. c. Henderson (989), "Data Envelopment Analysis for Managerial Control and Diagnosis," Decision Sciences 20 (1): 90-119. Evanoff, D. D., and P. R. Israilevich (991), "Productive Efficiency in Banking," Federal Reserve Bank of Chicago Economic Perspectives 15 (4): 11-32. Farrell, M. J. (957), "The Measurement of Productive Efficiency," Journal ofthe Royal Statistical Society, Series A, 253-81. Graham, F. c., and J. E. Horner (988), "Bank Failure: An Evaluation of Factors Contributing to the Failure of National Banks," Issues in Bank Regulation 12 (2): 8-12. Gunther, J. W. (989), "Texas Banking Conditions: Managerial Versus Economic Factors," Federal Reserve Bank of Dallas Financial Industry Studies, October. Humphrey, D. B. (991), "Flow Versus Stock Indicators of Banking Output: Effects on Productivity and Scale Economy Measurement," paper presented at the Western Economic Association meetings in Seattle, Washington, June 29-July 3. Kaufman, G. G. (986), "Banking Risk in Historical Perspective," Federal Reserve Bank of Chicago Bank Structure and Competition, May, 231-49. Kennington, J. (980), "A Primal Simplex Code for Computing the Efficiency of Decision-Making Units," Technical Report OREM 80001 (Dallas: Southern Methodist University, Department of Operations Research). Kolari,]., and A. Zardkoohi (987), Bank Costs, Structure, and Peiformance (Lexington, Mass.: D. C. Heath and Company). Lewin, A. Y, R. . Morey, and T. J. Cook (982), "Evaluating lh ' Administrative Efficiency of Federal Reserve Bank of Dallas  Courts," Omega: Internationaljournal of Management Science 10 (4): 401-11. Nunamaker, T. R. (985), "Using Data Envelopment Analysis to Measure the Efficiency of Non-Profit Organizations: A Critical Evaluation,"  Managerial and Decision Economics 6 (1): 50-58. Pantalone, C. c., and M. B. Platt (987), "Predicting Commercial Bank Failure Since Deregulation," Federal Reserve Bank of Boston New England Economic ReView, July/August, 37-47. Parkan, C. (987), "Measuring the Efficiency of Service Operations: An Application to Bank Branches," Engineering Costs and Production Economics 12, 237-42. Rangan, N., R. Grabowski, H. Y. Aly, and C. Pasurka (988), "The Technical Efficiency of U.S. Banks," Economics Letters 28, 169-75. Seballos, L. D., and J. B. Thomson (990), "Underlying Causes of Commercial Bank Failures in the 1980s," Federal Reserve Bank of Cleveland Economic Commentary, September. Seiford, L. M. (990), "A Bibliography of Data Envelopment Analysis 0978-1990)," Version 5.0, Technical Report (Amherst, Mass.: University of Massachusetts, Department of Industrial Engineering, April).  Economic Review - January 1992  - - - , and R. M. Thrall (990), "Recent Developments in DEA: The Mathematical Programming Approach to Frontier Analysis," journal of Econometrics 46 0/2, October/November} 7-38. Sexton, T. R. (986), "The Methodology of Data Envelopment Analysis," in Measuring Effi-  ciency: An Assessment ofData Envelopment Analysis, ed. R. H. Silkman (San Francisco: Jossey-Bass). Sherman, H. D. (984), "Improving the Productivity of Service Businesses," Sloan Management Review, Spring, 11-23. - - - , and F. Gold (985), "Bank Branch Operating Efficiency: Evaluation with Data Envelopment Analysis," journal ofBanking and Finance 9 (2) 297-315. Siems, T. F. (991), "An Envelopment-Analysis Approach to Measuring Management Quality and Predicting Failure of Banks," Ph.D. dissertation, Southern Methodist University. Triplett, J. E. (991), "Measuring the Output of Banks: What Do Banks Do?" paper presented at the Western Economic Association meetings in Seattle, Washington, June 29-July 3.  41