The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Bconomic Review Federal Reserve BaIllt of San Francisco Winter 1990 Brian Motley Carolyn Sherwood-Call Ramon Moreno Elizabeth S. Laderman Num.ber 1 Has There Been a Change in the Natural Rate of Unemployment? Assessing Regional Economic Stability: A Portfolio Approach External Shocks and Adjustment in Four Asian Economies-1978-87 The Public Policy Implications of State Laws Pertaining to Automated Teller Machines Table of Contents Has There Been a Change in the Natural Rate of Unemployment? . . . . 0. . . . . . . . . . . . . . . . „ . . . » . » „ . . . . . . . . . . . . „ 3 Brian Motley Assessing Regional Economic Stability: A Portfoio Approach *. 00., 0„»0*„„ .«»„»„ 0*. 0„„„s »0<»000*„00„„0»0„0. *9 * 17 Carolyn Sherwood-Call External Shocks and Adjustment in Four Aslan Economies— 1978-87 0. . . . . . „. . „ . . . . . <>.... 00. . „ . . „ . . „. . . . . . „ 0„ 27 Ramon Moreno The Pubic Policy Implications of State Laws Ftertalnlng to Automated Teller Machines 0„„008000. <>0 . 0<>00„0. 00„*0000*„ 43 Elizabeth S. Laderman F ederal Reserve Bank o f San Francisco 1 Opinions expressed in the Economic Review do not neces sarily reflect the views of the management of the Federal Reserve Bank of San Francisco, or of the Board of Governors of the Federal Reserve System. The Federal Reserve Bank of San Francisco’s Economic Review is published quarterly by the Bank’s Research Department under the supervision of Jack H. Beebe, Senior Vice President and Director of Research. The publication is edited by Barbara A. Bennett. Design, production, and distribution are handled by the Public Information Department, with the assistance of Karen Rusk and William Rosenthal. For free copies of this and other Federal Reserve publicatons, write or phone the Public Information Department, Federal Reserve Bank of San Francisco, P.O. Box 7702, San Francisco, California 94120. Phone (415) 974-2163. 2 E conom ic R eview / W inter 1990 Has There Been a Change in the Natural Rate of Unemployment? Brian Motley Senior Economist, Federal Reserve Bank of San Francisco. Editorial committee members were Bharat Trehan, John Judd, and Adrian Throop. Many economists argue that inflation begins to pick up when the unemployment rate falls below a critical level known as the natural rate. Those who worry that there is a risk offaster inflation argue that unemployment has fallen below this level. Others suggest that the natural rate has declined and hence that the inflation risk is less. The results of this paper do not support the argument that the natural rate has declined. The slow response ofinflation to changes in unemployment can explain why the low rate of unemployment since 1987 has not led to faster inflation. However, this slow response of inflation to employment suggests that if inflation is allowed to rise, it would take a long period of slow growth to bring it down again. Federal Reserve Bankof San Francisco In the last two years, the rate of unemployment in the U.S. has declined significantly. During 1989, unemployment averaged 5Y4 percent of the civilian labor forGe, ITlQre than one percentage point below its level in the first half of 1987 and lower than at any time since 1974. As the unemployment rate fell, concern increased that the associated high level of economic activity would lead to a pick-up in inflation. Many economists argue that the unemployment rate is an indicator of the strength of aggregate demand, and that inflation tends to increase when unemployment is low. These economists hypothesize that there is a critical level of the unemployment rate below which wages and prices tend to accelerate and above which they tend to decelerate. This critical level is called the "natural rate of unemployment." Those who worry that there is a risk that inflation will pick up argue that unemployment has fallen below this natural rate. Until recently, most estimates of the natural rate of unemployment placed it around six percent. I With a natural rate in this range, inflation pressures should have begun developing in the second half of 1987, when the unemployment rate fell to 5.9 percent from 6.4 percent in the first half. Concerned that the risk of faster inflation was increasing, the Federal Reserve began to tighten policy in mid-1987. In 1988, after a brief pause following the stock market crash, the Federal Reserve continued the process of policy tightening that it had begun in 1987. Despite the low level of unemployment, however, the increase in inflation since 1987 has been relatively modest. This raises the possibility that the natural rate of unemployment may be less than six percent.? or even that there is no critical level of unemployment below which inflation necessarily worsens. This article seeks to throw light on this issue by estimating the natural rate and examining whether it has changed in recent years. The first section reviews the theory of the link between inflation and unemployment. Particular attention is paid to the role of "supply-side" or "relative-price" shocks in the inflation process. Section II estimates four alternative empirical equations linking inflation to the unemployment rate and to relative-price shocks. The equations are estimated over a series of different sample periods to test 3 whether the natural rate of unemployment has changed through time. Section III summarizes the empirical results and discusses their implications for policy. The results of this paper do not support the argument that the natural rate has declined. The slow response of inflation to changes in unemployment can explain why the low rate of unemployment since 1987 has not led to faster inflation. However, this slow response of inflation to unemployment suggests that if inflation is allowed to rise, it would take a long period of slow growth to bring it down again. I. Inflation and Unemployment: Theory The notion that low levels of unemployment are associInflation Expectations ated with high inflation is generally known as the "Phillips 3 curve." Although there is widespread agreement that unemployment and inflation are related, at least in the short run, the direction of causation is disputed. Most economists argue that if unemployment falls below its natural level, this causes inflation to increase. However, some theorists maintain that causation runs from inflation 4 to unemployment. The empirical models in this paper are based on the first hypothesis. The theory that low unemployment causes faster inflation views the unemployment rate as an indicator of the demand for labor relative to the supply. When the demand for labor is strong, wages tend to be bid up. Because the prices of goods and services are set as a mark-up over their costs of production, most of which comprise wage costs, wage inflation leads to overall price inflation.5 In the labor market, there always is a number of persons who are searching for employment even when there are jobs vacant. This is because individual jobs and workers are unique, and matching unemployed persons to vacant .jobs requires a process of search on both sides." In a changing economy, persons continually are entering and leaving the unemployment pool and firms are creating and filling job vacancies. This observation means that even when the supply of and demand for labor are in overall equilibrium at going wages, there will be a certain remaining level of unemployment. The "natural rate" of unemployment reflects this "equilibrium" quantity of joblessness. When unemployment is below (or above) its natural rate, wages tend to rise more (or less) rapidly. If the mark-up of prices over production costs remains constant, (or, afortiori, if it increases when aggregate demand is strong7), the rate of overall price inflation also will be inversely related to the unemployment rate. This inverse relation between inflation and unemployment may be represented algebraically as: = RUt - UNAT) (1) where f' () < O. In this equation, 'ITt represents the inflation 'ITt rate, and U, and UNAT represent the actual and the natural rates of unemployment, respectively. 4 In negotiating wages, workers and employers pay attention not only to the state of labor supply and demand in their own markets, but also to the rate of overall inflation they expect in the future. If prices are expected to rise, workers demand wage increases to maintain their real incomes. Employers are willing to meet these demands, because they expect to be able to pass their higher costs on to their customers, and fear that, if they do not, they will lose their best workers. Hence, an expectation that prices and wages will rise tends to be self-fulfilling, even when unemployment is at its natural rate. This argument leads to the "expectations-augmented" Phillips curve, which may be written as: = Et- 1'ITt + RUt - UNAT) (2) where E, _ 1'ITt represents the rate of inflation in period tthat was expected in period t-l. When unemployment is below (above) the natural rate, actual inflation will exceed (fall short of) the rate that was expected. Under simple assumptions about how price expectations are formed, this model implies that inflation will increase continually if unemployment remains below its natural level. This will be the case, for example, under the common assumption that people raise their expectations of future inflation when current inflation is higher than they had expected. This assumption about expectations may be written as 'ITt E, 1'ITt = E, 2'ITt 1 + g( 'ITt - l - E, 2'ITt- l ) (3) where s'() > O. This assumption implies that the rate of inflation expected for the current period is a function only of past inflation. If this is a linear function, it may be written: Et-1'IT t S = S~lWS' 'ITt- s (4) The rate of inflation expected in the current period is a weighted average of the rates of inflation experienced in the past. 8 This is known as the "adaptive expectations hypothesis." Substituting equation (4) into equation (2) and assuming that the functionfO in equation (2) is linear yields: Economic Review / Winter 1990 1Tt S = sk,W s' 1Tt- s + I i~oai' (Ut - i - UNAT) (5) It seems plausible that if inflation has been constant i~ the past and unemployment is at its naturalle~el,. the~ 11 will remain constant in the future. This assumptionimplies that the weights (wsJ in equation (5) sum to one. In this case a level of unemployment below the natural rate not only'adds to current inflation, but also causes inflation to continue rising in the future. Even if unemploymentremains at its natural level, this does not guarantee a low inflation rate. Any rate of inflation, if it is anticipated, is compatiblewith unemployment at its natural level. Thus, although a decline in the unemployment rate below its natural rate will lead t~ fast~r than-expected inflation, there is no long-run relationship between the level of unemployment and the rate of inflation. The assumption that agents form inflation expectations adaptively implies that the only informa.tion th~y us~ in forming expectations is the past behavior of inflation. However, the theory of rational expectations suggeststhat, since agents know inflationis related to the unemployment rate, they should base their inflation expectations on t?eir forecasts of unemployment rather than on past inflation, Despite this reservation, models that assume adaptive expectations appear to fit the data reasonably well.? An alternativeexplanationwhy past inflationappearsto affect current inflation is that wage contracts are made for several years and are not all negotiated simultaneously!" When a firm and its employees are negotiating a new contract, they recognize that contracts made in the past in related industries will be in effect for at least part of their contract period. If those other contracts contained wage increases, new contracts likely will call for similar increases even if economic conditions have changed. As a result, inflation will tend to persist once it has begun. Recently, some economistshavesuggested that chang~s in the actual level of unemployment may affect the equilibrium rate at which inflation remains constant. According to this "hysteresis" theory, II although an increase in unemploymentmaylowerinflationinitially, it later leadsto a corresponding increase in the natural rate, as employers and workers become accustomed to higher rates of unemployment. As a result, its effect on inflation is only temporary. This hysteresis hypothesis may be tested by examining whether the coefficients on the unemployment rate sumto zero in equation (5). Supply Shocks Theaverageprice levelalso maybe influencedby shocks that affectthe prices of particular commodities. Forexam- Federal Reserve Bank of San Francisco ple, a rise in the prices of imported raw materials adds directly to costs of production and hence to the average price level. As prices adjust to the higher level, there will be an increase in the measuredinflationrate. This suggests that equation (5) should be extended to include the effects of such "relative-price" or "supply" shocks. On several occasionsin the last two decades, changes in the overall inflation rate have been attributed to changes either in the real exchange valueof the dollar or in the real price of oil. The real exchange rate measures the price of U.S.-produced goods relative to foreign-produced goods. A decline in the real exchange rate could add to U.S. inflation, by raising the cost of imports and reducing the . Iow.12 pressures on domesticproducersto keep thei eir pnces Thereal price of oil representsthe price of oil relativeto the general level of prices in the U.S. A rise in the price of oil adds directly to costs of production in the u.s. Thus, both the real exchange rate and the real price of oil are prime candidates for inclusion in equation (5). This discussion suggests an estimating equation of the form: 1Tt S = S;lws'1Tt-s G I + i~oai • (Ut - g~iog . SHKOILt g i - UNAT) + H + h~ixh . SHKEXt _ h (6) where SHKOILt _ g represents the change in the real price of oil and SHKEXt _ h the change in the real exchange rate.P The distributed lags on these variables capture the idea that the effects of supply shocks do not occur instantaneously. Several previous researchers'< have estimatedequationssimilar to equation (6) and havefoundthat relative-price shocks have a significant impact on the measured inflation rate. Supply Shocks and Expectations Equation (6) implicitly assumes that the impact of past inflationon current inflationis the sameregardlesswhether that past inflation was the result of excess demand or of relative-price shocks. In particular, equation (6) implies that a single relative-priceshock, unless offsetby a change in the unemploymentrate, will lead to a permanent change in the rate of inflation. This is because such a shock not only has a direct impact on prices but also has an indirect effect via expectations.15 This implication does not seem plausible. Economic agents generally would recognize that the rise in pri~es following an increase in the price of oil, for example, IS a "one-time" effect, and so would not change their longerrun inflation expectations. Hence, although such a shock would have a permanent effect on the level of prices, the 5 associated speed-up in inflation should be only temporary. This suggests that equation (6) may not be a fully satisfactory model of inflation in periods (such as the 1970s and 1980s) in which there were significant relative-price shocks. To incorporate the restriction that relative price shocks do not affect the inflation rate in the long run, two alternative modifications of equation (6) will be estimated. The first approach maintains the assumption that expected inflation is a weighted average of past inflation, but imposes the restriction that the coefficients on the shock variables in equation (6) sum to zero: 'iTt S = S~IWs'1Tt s I + i~oai . (U, N n'focon . SHKOILt n + VNAT) i - + M ~oCxm . SHKEXt _ m (7) where'v ICon = IC xm = O. This restriction implies that although relative-price shocks affect the average level of prices and so change the inflation rate temporarily, they have no long-run effect on inflation. The second approach modifies equation (4) so that changes in inflation resulting from relative-price shocks do not affect inflation expectations and so do not pass through into future inflation: J S E t - l1Tt = S~IWS [1T t s - (j~ioj . SHKOILt_ s_ j K + k'fixk . SHKEXt s-k)) (4') where dOj and d X k represent the direct effects of relative price shocks on inflation. This equation says that expected inflation is a weighted average of past inflation, excluding that part of past inflation that was due to relative-price shocks. This model of expectations yields an estimating equation of the form: 'iTt S J = sIlws [1T t - s - (j~Odoj K + k'fodXk . SHKEXt s k)] I i~oai . (V t- i - VNAT) + . SHKOILt s- j + J + j~OdOj SHKOIL t_ j K k'fixk' SHKEXt_ k (8) where no restrictions are placed on the dOj and dX k coefficients. Equation (8) also may be written as follows: 6 S I 1Tt = I Ws'1T t - s s=1 . + 1=0 .I ai J ' (U, i - VNAT) + S j~Odof(SHKOILt-j - s~ows'SHKOILt_s) K + k'flxk'(SHKEXt- k - S s~ows'SHKEXt-s-k) (9) In this form, the sums of the coefficients on the shock variables are Idoj . (1 - Iw s) and Id xk . (1 - Iw s)' respectively. Since Iws = 1, these coefficient sums are equal to zero. Thus, equations (7) and (9) are similar since both specify that current inflation is influenced by a distributed lag of past relative-price shocks with coefficients summing to zero. Equation (7) may be viewed as a generalization of equation (9) that puts fewer restrictions on the shapes of the distributed lags on the shock variables. Thus, the COn and CXm coefficients in equation (7) are combinations l7 of the underlying parameters (w sand doj , and w s and dX k ' respectively), representing both the direct and indirect effects of relative-price shocks. One objection that may be raised to this approach is that it is inconsistent to assume, as in equation (4'), that agents can distinguish between price increases resulting from supply shocks and those due to other factors, but cannot see that future inflation also will be affected by the unemployment rate. This objection may be less serious in practice, however, because relative-price shocks often have been sufficiently large that the public probably was able to recognize them as one-time events. This is particularly true of the oil-price shocks in 1974 and 1979. Demographic Shifts The natural rate of unemployment may vary as conditions in the labor market change. Faster technological change, for example, may lead to more job-changing and hence a higher natural unemployment rate. Demographic changes have similar effects. Because young persons have fewer skills and less work experience than adults, and also move in and out of the work force more often, they have a higher unemployment rate. If the proportion of young workers in the labor force declines, the measured unemployment rate will fall, but this does not imply an increase in inflationary pressure, but rather a decline in the natural rate. This argument implies that the unemployment measure used in estimating the Phillips relation should be adjusted for demographic changes. 18 If lit represents the proportion of the labor force that is in the ith population group at date t, and U it is the unemployment rate for that group, the total unemployment rate, VI' may be decomposed as follows: Economic Review I Winter 1990 unemployment rate. Conversely, the third component, - iii)' shows how overallunemployment would have changedas a resultof changesin thegroupunemployment rates, if the demographic structure of the labor force had remained unchanged. The final term represents ambiguous "cross" effects, some of which also may be demographic.'? Since their net contribution to overall unemployment is small, these cross effects are grouped with the demographic effects. The "demographically-adjusted" unemployment rate is constructed by subtracting the demographic and "cross" effects from the measured unemployment rate. Thus: I Li . (Uit where Li is the average proportion of the labor force in the ith groupovera periodof years, and iii is theaverage group unemployment rate. The first term on the right side of equation (10) is the average unemployment rateovertheperiod. Theremaining terms decompose the difference between the.actual rate at date t and the average rate into three components. The second term, I iii . (lit - LJ, measures how the overall unemployment rate would have changed if the group unemployment rates had remained constant and only the structure of the population had changed. This component measures the size of purely demographic effects on the V;'= I u, .I iii . (lit 1=1 I - Li ) - .I (Uit 1=1 - iii) . (lit - LJ I = I j;.• U· i= 1 I It (11) Chart 1A Unemployment Rate: Actual vs Adjusted Percent 11 10 9 8 7 6 / 5 Demographically-Adjusted Rate 4 3-h".--...-r-r-,...-...,....,,,.--...-r-r-,...-...,....,,,.--...-r-r-,...-...,....,,, 1963 1968 1973 1978 1983 1988 Chart 1B Percent Components of the Unemployment Rate 0.3 0.2 "Cross" effects 0.1 0.0 " +:::::'::"_ _-:::::::::~~-T---":::::o.....=::.;:;.......:~..-::;::~~ -0.1 -0.2 ....... Demographic Effects -0.3 -0.4 -0.5 -h".--...-r....,-,...-,-".....,...-r....,-....,..,-".....,...-r-,-....,..,-" 1968 1963 1973 1978 1983 1988 Federal Reserve Bank of San Francisco 7 In constructing the data series of U ~ the population was divided into eight age-sex groups: teens, young adults, prime-age adults and seniors.20 Chart 1 shows the components of the unemployment rate. The upper panel compares the actual unemployment rate with the adjusted rate definedin equation(11). Mostof the variation in the unemployment rate since1960has been due to factors that affected all population groups rather than to demographic changes and cross effects. However, the lower panel shows that purely demographic factors did have a significanteffect, raising the overalljobless rate by almost 0.7 percentage point between 1960 and 1978 and reducing it by 0.5 percentage point since then. II. Empirical Analysis The coefficients of the equations developed in the previous section were estimated using the demographicallyadjusted series (ut) constructedin equation(11). Thus, the equations estimated were: S 1ft = S~lWs'11't I s + i~oai . (Ut i- UNAT*) G + g~Obog . SHKOILt_ g H + h~obxh . SHKEXt_ h S 1ft (6a) I = S~lWs'11't-s + i~oai . (Ut-i - UNAF*) N + n~ocon . SHKOILt_ n M + m~oCXm • SHKEXt_ m S 1f = s~ lW s [11' t- s (7a) J (j~ioj . SHKOILt_ s _ j - K + k~odxk . SHKEXt_ s Full Sample Estimates k)] I + i=O 2 a· . (u*t . - UNAT*) I J I K + j~Odoj'SHKOILt_j + k~odxk'SHKEXt-k (8a) Thecoefficients on the relativepricevariables (SHKOIL and SHKEX) are unrestricted in equation (6a), are constrained to sum to zero in equation (7a), and are subjectto nonlinear restrictionsin equation (8a). Toassessthe role of the shock variables, an equation that excludes these variables also was estimated. The equations were estimated over a sample period from the first quarter of 1963 to the fourth quarter of 1988, a total of 104 quarterly observations. Since the equations are nonlinear in UNAT*, and equation (8a) includes restrictions on the relative-price shock coefficients, nonlinear regression estimation was employed.21 8 The measure of inflation used is the annualized quarterly growth rate of the fixed-weight GNP price index. Unemployment is measuredby the civilian unemployment rate. 22 The real price of oil is the ratio of the producers' price of crude petroleum to the producer price index for finished goods, and the real exchange rate is the Federal Reserve's multilateral trade-weighted value of the dollar deflatedby the ratio of a trade-weighted average of foreign consumer price indexes to the U.S. fixed-weight GNP price index. The equations also include dummy variables (NIXON and NIXOFF) to capture the effects of the imposition and removal of price controls by the Nixon Administration in the early 1970s. 23 The estimates of UNAT* represent the demographically-adjusted natural rate. These are converted to estimates of the actual natural rate by adding the difference betweenU,and U:, Fora givenestimate of UNAT;Chart 1 implies that the natural rate has fallen since 1978, as the proportion of the population in groups that have higherthan-average unemployment rates has declined. Earlier estimates of inflation equations have assumed that the distributed lag on past inflation is long. Gordon and King.>' for example, used a twenty-four quarter lag distributionconstrainedto lie on a fourth-order polynomial with a zero end-pointconstraint. In a later paper, Gordon-> used four-quarter averages of the inflation rate extending back six years (24 quarters). To determine the appropriate lag lengths, equation (6a) wasestimated using a series of alternative lag distributions on past inflationfromfour to twenty-four quarters. In these regressions, no restrictions were placed on any coefficients.w Table I shows the results of these regressions. The estimates reported in this table represent the cumulative sums of the coefficients on lagged inflation out to each indicated lag length.?" Regardless of the lag length chosen, the estimated sum of all the coefficients (the last figure reported in each column) is not significantly different from one. However, in each case, the sum of the Economic Review / Winter1990 coefficients reaches unity by the fifth or sixth lag, suggesting that current inflation is not affected by inflation more than five or six quarters in the past. Although some of the coefficients for more distant lags are statistically significant when lags longer than twelve quarters are introduced,28 several of the later coefficients in these equations take negative values-? which seems implausible. Column I of Table 1 shows the result of estimating the equation assuming a 24-quarter lag, but imposing the restriction that this lag follows a polynomial. 30 This restriction smoothes the estimated lag distribution and has the effect of making it appear longer. However, the hypothesis that the data satisfy this restriction may be rejected. The results in Table 1 suggest that the finding of long distributed lags in earlier research" may be due to the use of overly restrictive polynomial specifications. Hence, the equations in Tables 2 and 3 were estimated using a five- Federal Reserve Bankof San Francisco quarter lag on past inflation, with no polynomial restriction. Table 2 shows the results of estimating the alternative models over the 1963 to 1988 sample period. Model I refersto an equation that excludes the relative-price shock variables. Models II and III refer to equations (6a) and (7a), which are identical except that Model III constrains the coefficients on SHKOIL and SHKEX to sum to zero. Model IV refers to equation (8a), which imposes additional restrictions on the relative-price coefficients. F) is the F-statistic.that tests the restriction that the coefficients on lagged inflation sum to one. In all four equations, this restriction may be accepted. 32, 33 F2 is an F-statistic that tests for the presence of first- through fourth-order autocorrelation in the residuals.>' In every case, the hypothesis that the residuals are autocorrelated may be rejected. The coefficients on SHKOIL and SHKEX in Table 2 9 represent the estimated sums of the distributed lag coefficients on these variables. 35 The figures in parentheses below these coefficients are the standard errors of these sums and those in brackets are F-statistics that test the joint hypothesis that the individual lag coefficients are all zero. The alternative estimates of the demographicallyadjusted natural rate (UNAT*) range from 5.99 percent in Model II to 6.20 percent in Model IV. In the fourth quarter of 1988, the demographically-adjusted unemployment rate (U*) exceeded the measured rate by 0.26 percentage point, so these estimates of UNAT* imply actual values of the natural rate between 5.73 and 5.94 percent. 36 These estimates are similar to estimates developed in earlier research by Gordon."? The finding that the results are not sensitive to the model used to capture the effects of relative price shocks adds to one's confidence in the estimate. On the other hand, in all cases, the one-standard-error confidence interval on the estimated natural rate is nearly one percentage point wide, which may be too wide for the unemployment rate to be a useful signal to policy-makers. The response of inflation to divergences between the actual and natural unemployment rates depends on the coefficients on lagged inflation and on the current and lagged unemployment rate. Dynamic simulations of the equations were used to estimate the response of inflation to temporary and permanent divergences between the natural unemployment rate and the actual rate. These simulations showed that, in Model IV, for example, a one-percentagepoint decline in the unemployment rate that lasts for a single quarter will raise the inflation rate by 0.11 percentage point. This increase in inflation will be permanent unless it is later offset by an equal single-quarter rise in unemployment. If the unemployment rate remains one percentage point below the natural rate permanently, the inflation rate will increase by 0.11 percentage point every quarter. The other estimated models yielded similar results. The coefficients on current and lagged unemployment are of opposite signs, implying that the initial impact on inflation of a higher unemployment rate is less than its long-run impact. The hysteresis hypothesis referred to earlier suggests that the inflation rate changes only in response to a change in the unemployment rate and that inflation may remain constant at any level of unemployment. Although the coefficients on current and lagged unemployment have different signs, they are not equal. The estimated sum of these coefficients ranges from - 0.23 to - 0.27 and is significantly different from zero. This means that the level of unemployment does have a significant impact on the inflation rate. Contrary to the hysteresis hypothesis, an unemployment rate that remains constant 10 below the natural level does cause inflation to increase continually. In Model II, the hypotheses that the sums of the coefficients on SHKOIL and SHKEX are both zero may be accepted, implying that relative-price shocks have only temporary effects on inflation. Models III and IV impose this restriction. In Model III, the hypothesis that all the coefficients on SHKOIL are individually zero can be rejected at the six percent level of significance (F = 2.88). Moreover, the coefficient on the current value of SHKOIL is positive and significant (t-statistic = 2.18), indicating that changes in the real price of oil have a significant temporary effect on the inflation rate. This conclusion is confirmed by Model IV, which also shows a positive and significant temporary impact of oil prices on inflation. Economic Review / Winter 1990 Although the effects of changes in the exchange rate are less significant, they go in the direction predicted by theory. In Model III, the sum of the coefficients on the current and twelve lagged values of SHKEX is constrained to zero, but the estimated coefficient-sum out to the eighth lag. is negative and significant at the five percent level, implying that a real appreciation of the dollar reduces inflation temporarily. However, the hypothesis that all the coefficients on SHKEX are individually zero cannot be rejected (F = 0.72). In the more restrictive Model IV, the sUrtlofthecoefficients on SHKEX (which in this model represent only the direct effect of exchange-rate changes) is negative and significant at the 13-percent level (onetailed test) and the hypothesis that these coefficients are all individually zero may be rejected at the 20-percent level of significance (F = 1.41). Thus, there is only weak evidence that changes in the real exchange rate have effects on inflation. Rolling Regressions To examine whether the natural rate has changed over time, the equations were estimated over fifteen overlapping twelve-year sample periods, beginning with 1963.1 to 1974A and ending with 1977.1 to 1988A. The twelve-year length of these sample periods was chosen to provide a reasonable number of degrees of freedom and ensure that even the earliest sample periods included a number of observations after the shift to flexible exchange rates in 1971 and the first oil-price shock in 1974. The lag lengths determined from the full 1963-88 sample period wereused for these "rolling regressions." Table 3 shows the estimates of Model IV.38 This table suggests that the estimated natural rate tends to be higher for samples that include the 1980s. The estimates of UNAT* range around six percent for samples ending before 1980,39 but are closer to seven percent for samples ending between 1980 and 1988. The estimate of UNAT* over the final period implies a natural rate in the fourth quarter of 1988 of 6.8 percent, more than IV2 percentage points above the actual rate in the quarter and % percentage point above the estimate from the full sample period. The estimates of the impact of relative-price shocks on inflation are mixed. In all sample periods, the impact of these shocks has the sign predicted by theory. On a onetailed test, the hypothesis that oil-price shocks initially have a positive impact on inflation (that is, the coefficient on the contemporaneous value of SHKOIL is positive) may be accepted at the lO-percent level in nine sample periods. The corresponding hypothesis that the sum of the coefficients on SHKEX is negative also may be accepted in nine Federal Reserve Bankof San Francisco sample periods. 40 Although relative-price shocks probably have affected overall inflation in the directions expected from theory, these effects have not been consistent. 11 Simulations To provide a further test of the alternative models, outof-sample simulations of each of the four models were conducted. For purposes of this simulation exercise, the models were re-estimated" over the period from 1963.1 to 1984.4, and simulated to 1988.4. Table 4 compares the error statistics for forecasts of the quarterly change in inflation obtained from static and dynamic simulations of the four models.V The static simulations are out-of-sample fitted values of the estimated equations. The dynamic simulations were begun in the first quarter of 1963. In the early years of the simulations, the simulated quarterly changes in inflation depend on actual inflation before 1963 as well as on current and past values of the unemployment rate and the shock variables. However, the effect of inflation before 1963 gradually dies out and by 1980 is effectively zero. The errors from the dynamic simulations are in most cases smaller than those from the static simulations. In the dynamic simulations, the simulated change in the inflation rate depends only on its underlying determinants (current and lagged unemployment and relative price shocks) and is not affected by lagged actual inflation. The results in Table 4 suggest that this may be a superior forecasting procedure. The errors from both the static and dynamic simulations of Model III and Model IV are lower than those from Model II. This result supports the argument that relativeprice changes should have no permanent effect on inflation. In addition, the errors from Model IV are lower than those from Model III, suggesting that the Model IV specification, in which changes in inflation associated with relative-price shocks are explicitly "purged" from the lagged dependent variable, is a better specification. 12 Economic Review / Winter 1990 However, Model I, which omits the relative-price variables entirely, appears to predict inflation as well as Model IV. This finding, together with the rolling regressionresults in Table 3, which showed that the effects of the shock variables were sporadic, suggests that even the direct impact of relative-price shocks on inflation is small, except in periods when these shocks are unusually large (as in the 1974 and 1979 oil shocks). Thus, Table 4 seems to suggest that the role of relative-price shocks in the inflation process probably has been over-emphasized in earlier research •and in media discussions of the sources of inflation. The dynamic simulations make it possible to decompose the change in the inflation rate into its underlying sources. These decompositions are shown in Table 5. For each year since 1985, the change in annual inflation is separated into the portions due to past and present divergences of unem- ployment from its natural level, to oil-price and exchangerate shocks, and to the cumulative simulation error. Inflation declined by about one percentage point between 1984 and 1986. All four models attribute a significantportion of this decline to the high level of unemployment relative to its natural rate. Conversely, the models agree in attributing much of the 1.8-percentagepoint increase in inflation in 1986-88 to the low rate of unemployment in those years. As inTable 4, the simulation errors are larger for Model Ilthan for the other three models, again suggesting that the former is an inappropriate specification. The decompositions suggest that the errors in Models II, III, and IV are largely due to the exchange rate variable. In every case, the annual error is opposite in sign and of a similar magnitude to the contribution of the exchange rate variable, implying that the error would be reduced by omitting that variable.43 III. Summary and Conclusions This paper has examined the link between inflation and the rate of unemployment. The feature of the inflation equations estimated in this paper that distinguishes them from other equations in the literature is that relative-price shocks are constrained to have only temporary effects on the inflation rate. In addition, inflation expectations are proxied by a distributed lag on past inflation that is much shorter than in earlier studies. As pointed out earlier, the failure of inflation to pick up significantly since 1987, despite the decline in unemployment, has led some economists to lower their estimates of the natural rate. However, apart from the effects of the change in the age-sex structure of the labor force, the results of this paper do not support the hypothesis that the natural rate has declined. If anything, the results of the rolling regressions in Table 3 suggest that the (demographically-adjusted) natural rate of unemployment has been higher in the 1980s. However, the estimates of the natural rate in these regressions are subject to an uncomfortably wide margin of uncertainty. This paper has not investigated the causes of any such change in the natural rate. One possible cause is the apparent increase in the rate of technological change during the last decade, as a result of advances in computer technology and the response of the economy to the oil-price shocks of the 1970s. More rapid change either in methods of production or in the types of goods and services being produced would be expected to add to job-changing and hence to the level of normal unemployment. Federal Reserve Bank of San Francisco A second important result of this paper concerns the role of relative-price shocks. Commentary on inflation in the media frequently focuses on the role of these shocks in influencing the inflation rate, and earlier research generally has found their effect on inflation to be statistically significant. The empirical estimates in this paper suggest that these shocks may raise or lower the level of prices and so cause temporary changes in measured inflation. However,these shocks do not influence the rate of inflation over longer periods. In all the equations estimated, we can accept the hypothesis that shocks have no long-run effect on inflation. Moreover, the estimates suggest that even the short-run effects of relative-price changes have been sporadic. In many sample periods, we cannot reject the hypothesis that the shocks had no impact on inflation, even in the short run. Also, in out-of-sample simulations, inclusion of these shocks did not improve forecasts of the inflation rate. These empirical results suggest that the role of these shocks in causing inflation has been over-emphasized in earlier research. The results of this paper suggest that since the Fall of 1988, the gap between the actual unemployment rate and the natural rate has ranged between 3f4 and 1Y2 percentage points. 44 In view of this gap, why has inflation remained relatively subdued? The primary reasorr" appears to be that the increase in inflation in response to low unemployment occurs quite slowly. Simulations of the estimated equations indicate that a permanent one-percentage point 13 gap between the natural and actual unemployment rates would cause inflation to increase by 0.11 percentage point per quarter or less than one-half percentage point per year. Unemployment has been below its natural rate only since mid-1987. Between the fourth quarters of 1986 and 1988, inflation increased by about lYz percentage points. The dynamic simulations reported in Table 5 indicate that the low rate of unemployment contributed between one and 1Y4 percentage points of this increase. Thus, the relatively modest pick-up in inflation may be explained without invoking a decline in the natural rate. At the same time, the estimated equations suggest that if the unemployment rate were to remain permanently at its present 5Y4 percentage level, inflation would continue to increase by about Y2 percentage point each year. The slow response of inflation to a rate of unemployment above or below the natural rate means that the costs of low levels of unemployment, in terms of rising inflation, are initially small-and policy-makers may be tempted to ignore them. However, once inflation has been allowed to increase to "unacceptable" levels, this slow response means that bringing it down again will require either a lengthy period in which unemployment is held above the natural rate or a shorter period of excessively high unemployment. The high costs associated with either course suggest that it is more prudent to move against rising inflation before it reaches unacceptable levels. NOTES 1. The estimate of "high-employment GNP" by the Commerce Department,for example, is computed as the level of output that the economy would produce at six percent unemployment, on the presumption that a lower jobless rate would be associated with increasing inflation. 2. Commenting on the report of an unemployment rate of 5.3 percent in June 1989, Michael Boskin, the Chairman of the President's Council of Economic Advisers, said, "I'm pleased that unemployment remains low, and I don't see the current level of unemployment as inflationary." The Wall Street Journal, Monday, July 10,1989, page 2. 3. A.w. Phillips, "The Relationship between the Unemployment Rate and the Rate of Change in Money Wage Rates in the United Kingdom, 1861-1957," Economica, November 1957, pp 283-299. 4. These theorists argue that higher-than-expected inflation leads to a decline in the unemployment rate. These alternative views of the theoretical underpinnings of the Phillips curve are illustrated intwo popular macroeconomics textbooks. Dornbusch and Fischer develop a model of unemployment and inflation in which causation runs from the unemployment rate, which serves as a proxy for the strength of demand in the economy, to the rateof inflation. See Rudiger Dornbusch and Stanley Fischer, Macroeconomics, Fourth Edition, McGraw-Hili, 1987, Chapters 13-14. The causation runs in the opposite direction in the "new classical" model presented by Hall and Taylor. In this model, unexpected changes in the inflation rate lead to changes in the unemployment rate. See Robert E. Hall and John B. Taylor, Macroeconomics: Theory, Performance and Policy, Norton, 1986, Chapter 13. 5. This argument assumes implicitly that the excess demand for labor by firms is the result of an increase in demand for their products. Clearly, if higher nominalwage rates reflect increases in labor productivity, they will not spill over into higher prices. 14 6. See Edmund S. Phelps, "Introduction: The New Microeconomics in Employment and Inflation Theory," in Phelps (Editor), Microeconomic Foundations of Employment and Inflation Theory, w.w. Norton, New York, 1970. 7. Seethe discussion of price determination in FlintBrayton and Eileen Mauskopf, "The Federal Reserve Board MPS Quarterly Econometric Model of the US Economy," Economic Modelling, July 1985, pp 202-203. 8. A special case arises when the function gO in equation (3) is linear. In this case, the weights Ws in equation (4) decline geometrically. 9. For a discussion of alternative models of inflation expectations, see Adrian Throop, "An Evaluation of Alternative Models of Expected Inflation," Economic Review, Federal Reserve Bank of San Francisco, Summer 1988. 10. John B. Taylor, "Staggered Wage Setting in a Macro Model," American Economic Review, Vol 69 (May 1979), pp 108-113. 11. For an example of this approach, see Olivier J. Blanchard and Lawrence H. Summers, "Hysteresis and the European Unemployment Problem," in Stanley Fischer (Editor), NBER Macroeconomics Annual 1986, National Bureau of Economic Research, 1986, and Robert J. Gordon, "Hysteresis in History: Was There Ever a Phillips Curve?," American Economic Review, May 1989. 12. A decline in the real exchange rate also tends to increase GNP growth by increasing exports and reducing imports. In the Phillips curve, these "aggregatedemand" effects of exchange rate changes will be captured through changes in the unemployment rate. 13. This specification implies that the direct effect of relative-price shocks is on the level of prices. An increase in the level of the price of imported oil, for example, adds to the average level of prices in the U.S. Hence, the average inflation rate, which is the dependent variable in Economic Review / Winter 1990 equation (6), is influenced by the growth rate of oil prices. 14. Robert Gordon has contributed extensively to this literature. See, for example, Robert J. Gordon, "Understanding Inflation in the 1980s," Brookings Papers on Economic Activity, 1:1985, pp 263-302. 15. An arithmetic examplemayclarifytheargument. Consider a simplified version of equation (6): 'ITt = 'ITt-1 + a . (Ut - UNAT) + bo' SHKOILt + bx . SHKEXt (6') Supposeunemployment is atthe natural rate (U, = UNAT) and initially therearenoshocks(SHKOIL t = SHKEXt = 0). This implies a constant inflation rate (TIt = TIt 1)' A onetime one percent increase in the price of oil means that SHKOIL t rises from 0 to 1 for a single quarter and then declinesback to O. The direct effectof thisshockisto raise the inflation rate by b o percentage point. However, because of the presence of the lagged inflation term (with a coefficient of one) in equation (6'), the inflation rate also will be b o higher in the next quarter and in all future quarters. Thus, a temporary shock leads to permanently higher inflation. Similarly, suppose the exchange rate begins to fall steadily at one percent per quarter, so that SHKEX t decreases permanently from0to -1. The inflation rate will rise by b x percent in the first quarter, by an additionalbx percentage point in the second quarter, and so on.Thus, a permanent shock leadsto continually rising inflation. 16. In equation (7) different symbols are used for the coefficients onthe shockvariables (con and CXm inplaceof bOg and b xh) to signify that these coefficients are constrained to sum to zero, whereas thosein equation (6) are not. 17. Inequation (9), the distributed lags onthe two relative price variables run from 0 to S + J and S + K, respectively. In equation (7) these lags run from 0 to Nand M, respectively. If equation (7) is interpreted as a generalization of (9), this implies that N = S + J and M = S + K. In the empirical estimations, this implication was used as a guide in choosing the lengths of the estimated distributed lags. 18. See Robert J. Gordon, "Inflation, Flexible Exchange Rates, and the Natural Rate of Unemployment," in Martin N. Baily, ed., Workers, Jobs and Inflation, Washington D.C., The Brookings Institution, 1982. 19. In cases where changes in the size of individual population groups are associated with unemploymentrate changes in the same direction, this term probably capturesdemographicfactors, as increases inthesupply of particular groups of workers lead to more unemployment. For both young adults and older workers, the correlation between unemployment and labor force share is positiveoverthe1963 to 1988 period, suggesting a demographic effectof this kind. However, for primeagewomen and for male teenagers, this correlation is strongly negative. In these cases, the causation may be reversed, with strong labor demand leading both to lower jobless rates and to greater labor force participation, Federal Reserve Bankof San Francisco 20. A similar approach is used by the Congressional Budget Office. See Appendix B, "Estimates of Potential Output," in Congressional Budget Office, The Economic and Budget Outlook: An Update, August 1987. However, CBO estimates the natural rate from the raw data rather tha.n fromthe adjusted data. 21. The restrictions that the coefficients on lagged inflation sum to unity and that the coefficients on each of the shock variables in equation (7a) sum to zero were imposed using a technique suggested by Scadding. See John L. Scadding, "Simple Technique for Imposing Restrlctions on Sums of POL Coefficients," Appendix 1 in RO$eMcl:lhattan,· "The:Response of Real Output and Inflation to Monetary Policy," Economic Review, Federal Reserve Bank of San Francisco, Summer 1981. 22, The demographic adjustments to the unemployment rate data were made using the same 1963-88 sample period as the regression estimates. 23. The definitions of these variables were adopted from Robert J. Gordonand Stephen R. King, "TheOutputCost of Disinflation in Traditional and Vector Autoregressive Models," Brookings Papers on Economic Activity, 1:1982. NIXON is defined as0.8 for the five quarters from 1971.3 to 1972.3 and NIXOFF is defined as 0.4 in 1974,2 and 1975.1 and as 1.6 in 1974.3 and 1974.4. 24. Gordon and King, cited in note 23. 25. Gordon, "Understanding Inflation in the 1980s." See note 14for full citation. 26. These preliminary regressions also include a constantterm, the currentand four lagged values of SHKOIL, the current and eight lagged values of SHKEX, and the current and one lagged value of the unemployment rate variable. 27. For estimation, the equations were transformed as follows: = b1(TIt-1 - TIt-2) + b2 ( TIt-2 - TIt-3) + ... bT- 1(TIt-T+1 - 'ITt-T) + bTTIt-T + ... In this form, b, s = k at. Thatis,theestimated coefficients t~1 (bs ) arethe cumulative sums of the underlying parameters (at). 28. Specifically, when more than 12 lagged values of inflation are included in the equation, one can reject the hypothesis that coefficients beyond the fifth lag are all zero. Forexample, for the regression that includes24 lags of the inflation rate, (column H) the F-statistic (with19and 63 degrees of freedom) for the hypothesis that the coefficientsfor lags beyond the fifth are all zero is 2.04, indicating that this hypothesis may be rejected with 97 percent confidence. On the other hand, in the regression that includesonly 12lags (column E), the f-statistic for the hypothesis that the coefficients on the sixth through twelfth 15 lags are all zero is only 0.83, implying that this hypothesis may not be rejected. 29. In columns G, H and I of Table 1, the sum of coefficients out to lag 12 is less than that out to lag 8. 30. Specifically, the curnutatixe sums.of the coefficients (that is, the figures reported in the table) are constrained to follow a second-order polynomial. This restrictive specification was adopted because third-, fourth- and fifthorder polynomials yielded estimated lag distributions in which the coefficient sums did not converge as the lag length was extended. 31. See Gordon and •King, cited above, and Adrian Throop, "A Macroeconomic Model of the U.S. Economy," Working Paper 88-06, Working Papers in Applied Economic Theory, Federal Reserve Bank of San Francisco, 1988. 32. F1 is less than onein all Cases, compared to its critical value at the 5 percent level of 2.5. 33. However, it should be mentioned that the estimate of UNAT* is sensitive to the impositionof this restriction. In Model III, for example, the relaxation of the restriction reduces the estimate of UNAT* by two standard errors from 6.18 to 5.33 percent. 34. A.C. Harvey, The Econometric Analysis of Time Series, New York: John Wiley and Sons, 1981, pp 276-77. 35. In Model III, the equations include the current and two lagged values of SHKOIL and the current and twelve lagged values of SHKEX; that is, N = 2 and M = 12. In Model IV, the direct effects of the shocks are represented by the current value of SHKOIL and the current and eight lagged values of SHKEX; that is J = 0 and K = 8. As implied in note 17, J < Nand K < M. 36. The differences between these alternative estimates are less than one standard error. 37. See, for example, the estimates by Gordon cited in Appendix B, "Estimates of Potential Output," in Con- 16 gressional Budget Office, The Economic and Budget Outlook: An Update, August 1987. 38. The estimates of Models I, II, and III are similar to those from Model IV and are available from the author. 39. The estimates of UNAT*are unexpectedly low and high in the 1963.,....74 and 1964-75 periods respectively. These estimates may be biased by the effects of the 1973-74 oil shock. 40. However, the hypothesis that all the coefficients on SHKEX are individually zero may be rejected with 90 percent confidence in only four sample periods. 41, The estimated coefficients over this sample period are not significantly different from those reported in Table 2. 42. In the static simulations, the errors in predicting the change in inflation are the same as those in predicting its level,because the equations include the lagged level of inflation. In the dynamic simulations, the simulated level of inflation depends on actual inflation in the quarters before the simulation begins. This means that the error in predicting the level of inflation depends on the starting date of the simulation. Hence, it is more appropriate to compare the errors in predicting the change in inflation, which do not depend on the starting date of the simulations. 43. Recall that the real exchange rate is only marginally significant in the equations reported in Table 2. 44. After adjusting for demographic change, Models I, III, and IV all imply a natural rate of six percent in the fourth quarter of 1988, when estimated over the full sample period. When estimated overthe 1977-1988 period, these models imply a rate of 6% percent. 45. The results in Tables 4 and 5 do not show a preponderance of negative errors that would suggest that the response of inflation to the unemployment rate in recent years has been atypical. Economic Review / Winter 1990 Assessing Regional Economic Stability: A Portfolio Approach Carolyn Sherwood-Call Economist, Federal Reserve Bank of San Francisco. The author wishes to thank Stephen Dean and Scott Gilbert for their diligent and capable research assistance. The editorial committee, Gary Zimmerman, Jonathan Neuberger, and Ronald Schmidt, provided many helpful insights. This paper examines regional economic stability using the analytical framework often used to study financial portfolios. The analysis shows that industrial diversification reduces economic volatility, just as portfolio diversification reduces financial risk. However, because the conditions that create a tradeoff between risk and return in financial markets do not exist for regional economies, regions do notface a tradeoffbetween stability and growth. Federal Reserve Bankof San Francisco State and local government officials often want to improve economic performance by changing their region's industry mix. For example, a state or local government might offer tax abatements to relocating firms in an industry that is expected to enhance the region's economy. However,it often is unclear just which industries improve a region's economy. Specializing in a small number of fastgrowing industries, or targeting fast-growing industries as promising sources of future growth, may make rapid growth possible, but the region's economy may become vulnerable to downturns in the industries in which it specializes. Thus, a specialized regional economy may be relatively volatile. If economic diversity reduces volatility, a region wishing to reduce volatility might see a diverse industrial mix as a desirable goal of economic development. Understanding the relationship between regional economic volatility and economic growth also provides useful insights regarding a region's optimal industry mix. If, for example, regional economies face a tradeoff between stability and growth, they may be willing to accept greater instability to achieve more rapid growth. However, if no such tradeoff exists, then stability would be a desirable goal regardless of the region's aspirations regarding economic growth. In a different context, the financial literature addresses the relationships between diversity and volatility. Portfolio theory suggests that diversification can reduce volatility, or risk. The logic of diversification is compelling for regional economies as well. Nevertheless, previous evidence regarding the relationship between regional economic diversity and regional economic instability is mixed. Conroy (1975) and Kort (1981) concluded that the extent of industrial diversity explains a significant proportion ofthe interregional differences in economic instability, while Jackson (1984), Steib and Rittenoure (1989), and Attaran (1986) found little evidence to suggest a relationship between diversity and instability. Others, including Brewer (1985), assumed that economic diversity explains regional differences in economic stability, and looked for the diversity measure that best captures this relationship. These studies 17 use a variety of measures to capture diversity and instability, but all suffer from a common conceptual problem: they examine the relationship between economic diversity and total instability. In contrast, the analogy with financial portfolios suggests that economic diversification should reduce only the amount of regionaleconomic volatility that is diversifiable, or nonsystematic. This result is derived from risk-spreading alone, and does not depend on restrictive assumptions about the economic or statistical characteristics of the region's industries. Since diversity is expected to be related to nonsystematic volatility, it is not surprising the previous studies of the relationship between diversity and total volatility have yielded conflicting results. Carrying the analogy with financial portfolios a step further also would suggest that the sensitivity of the region's economy to systematic, or nondiversifiable, factors could be associated with the regionalanalog to higher expected return, namely more rapid expected economic growth. If this were the case, regions might choose to accept more systematic sensitivity in exchange for higher growth. This hypothesis, however, relies on the marketclearing assumptions of the Capital Asset Pricing Model (CAPM), and those assumptions are quite tenuous for regional economies. This suggests that accepting higher systematic risk may not increase expected growth for a regional economy. . This paper discusses these relationships conceptually and tests them empirically. The analysis shows that there is, in fact, a strong correlation between diversity and nonsystematic volatility. However, systematic sensitivity is not compensated with higher economic growth. Thepaper is organizedas follows. SectionI presents the analogy between financial market portfolios and regional economies, alongwith its implications. SectionII explores the meaning of "diversity" in the regional economics context. SectionIII presentsthe data and variables usedfor the analysis. Section IV discusses the empirical evidence on the relationships among diversity, systematic and nonsystematic instability, and growth in regional economies. Conclusions and implications are drawn in Section V. I. Financial Portfolios and Regional Economies The finance literature distinguishes between two kinds of risk: "systematic" and "nonsystematic." Systematic risk is associated with broad economic and financial market conditions. As a result, it is common to all assets and cannot be diversified away. Nonsystematic risk, in contrast, is specific to a given asset and can be reduced through portfolio diversification. In a portfolio, diversification benefits investors by spreading risk among various assets, where each asset's "risk" is measured by the variance in its return. For example, assume that an investor starts off with a single asset with returnrI and variance VI' Adding a secondasset to the portfolio makes the portfolio's variance Vp> where: Vp = wyVI + W~V2 + 2W IW2C OV I,2 (1) In equation (1), wI and w2 reflect the weights of assets 1 and 2, respectively, in the portfolio. Thus, 0 '5 WI' o W2' and WI + w2 = 1. The relationship between VI and Vp depends on: (a) the magnitude of V2 relative to that of VI' (b) the relative proportions of the assets in the portfolio, WI and w2' and (c) the extent of covariance between the returns of the two assets, Covl,2' IfV 2is verylargerelative to VI' Vp maybe greater than VI' This is more likely if w2 is larger. Thus, adding an asset to the portfolio mayor may not reduce the portfolio's variance. However, as long as the covariance among individual assets (Covl,2) is less than one, the s: 18 variance of the portfolio is less than the weighted sum of the variances of the individual assets. This property is relatively easy to see in the case of uncorrelated returns, that is, when Cov, 2 = O. In this case: Vp = wyV I + W~V2 (2) Since WI and w2 are between zero and one, wy<wl and W~<W2' Thus, the variance of the portfoliois less than the weighted sumof the variances of the individual assets, and diversification reduces the risk associated with holding a portfolio that includes assets 1 and 2. The lower is the covariance between the returns of the two assets, the greater are the benefits of diversification, since the covariance term in equation (1) is smaller. Thus, under a wide range of circumstances, portfolio diversification reduces risk. Note that the benefitsof diversification are associated with the mathematical properties of variances, and do not depend on restrictive assumptions about the market characteristics oreconomic propertiesof the assetsthemselves. The benefits of diversification are even greater when the returns of the two assets are negatively correlated. In fact, the variance of the portfoliocan fall to zero in the case of perfect negative correlation. However, in real-world markets the returns to most assets are correlated with general economic and financial conditions, so the covariances between the returns for most pairs of assets are positive. Thus, investors cannot completely eliminate risk from Economic Review / Winter 1990 their portfolios. The risk that cannot be diversified away is referred to as systematic risk. Not all assets or portfolios have the same degree of systematic risk. According to the Capital Asset Pricing Model (CAPM), investors who take on greater systematic risk can expect to receive greater returns on their investments. Investors prefer the least possible risk at any given level of return, so prices for assets that face little systematic risk are bid up (thus reducing their returns) relative to prices of assets that offer the same yield with more systematic risk. Thus, the financial market bidding process results in a tradeoff between systematic risk and return. These principles suggest: (1) Nonsystematic risk should fall with greater portfolio diversification. (2) There should be a trade-off between systematic risk and return. These expectations have been verified in the financial literature (Fama and MacBeth, 1973; Black, Jensen, and Scholes, 1972; and Gibbons, 1982). The Analogy with Regional Economies In the analogy with regional economies, industries play the role of assets, and the region's industrial mix represents the portfolio. The "return" becomes the economy's growth rate, while its "risk" is the economy's volatility. In such an analogy, systematic volatility is associated with general economic conditions, such as fluctuations in the national economy, and nonsystematic volatility is the regional variation that is not associated with national influences. Since the relationship between portfolio diversity and nonsystematic risk depends on the mathematical properties of variances and not on specific assumptions about the assets' characteristics, the analogous relationship between a region's industrial diversity and its nonsystematic volatility is likely to hold.' In contrast, any relationship between systematic variability and growth would depend on a market-like mechanism. Under such a mechanism, risk-averse states would accept greater variability only if they were compensated in the form of stronger growth. However, several characteristics of regional economies make such a connection unlikely. First, although a financial asset earns the same return regardless of whose portfolio it is in, a given industry may perform differently depending on where it is located. For Federal Reserve Bank of San Francisco industries producing goods that are consumed in the same locale in which they are produced, the health of the region's economy affects the pace of activity, and this can differ across regions.? For example, auto repair services are by their nature provided in the same region where users of those services live, and interregional differences in the types of services provided are likely to be minor. Nevertheless, between 1980 and 1986, the real (inflation-adjusted) value of auto repair services grew 48 percent in fastgrowing Arizona and only 10 percent in slower-growing Oregon. Another weakness of the analogy is that different regions have different attributes that favor production of some goods over others. Natural resource endowments and transportation infrastructure are the most obvious sources of these regional differences in comparative advantage. (See, for example, North, 1955; and Schmidt, 1989.) Even when oil prices are high, residents of non-oil-producing regions generally cannot change their industrial structures to place more emphasis on oil production. Similarly, cities with limited access to overseas transportation are unlikely to become major transshipment points for international trade. These kinds of differences in comparative advantage limit the extent to which regions can (and should) diversify their economies. A final, and fundamental, problem with the analogy is that a region's officials cannot "trade" in a "market" for industries the way investors can trade in the market for financial assets. Although state and local governments often compete with each other to attract industries in order to improve their regions' economies, using such tools as tax incentives, infrastructure investments, and zoning variances, the "market" is thin and adjustments are slow. Since any local jurisdiction is unlikely to have its desired industry mix at a given point in time, equilibrium is not observed. Moreover, no individual has the power to change a region's industrial mix the wayan investor can alter a portfolio. 3 These differences between assets in a portfolio and industries in a regional economy suggest that a tradeoff between systematic variability and growth may not be observed for regional economies. These problems do not, however, affect the extent to which diversification should reduce nonsystematic volatility. This relationship is primarily a mathematical one, and does not rely on binding assumptions about the character of regional economies. 19 II. What Is "Diversity?" An investment portfolio that mimics the market portfolio in its composition (though not its size) is referred to as a fully "diversified" portfolio. Thus, a portfolio of ten different stocks would be somewhat diversified, but in a market in which hundreds of stocks are traded, it would not be completely diversified. Financial economists have agreed-upon standards by which to measure diversity.' Regional economists, in contrast, continue to debate what constitutes regional economic diversity. For the most part, this debate has been framed as a measurement issue, in which the "best" measure of diversity is the one that best explains regional differences in economic volatility. (See, for example, Conroy, 1975; Kort, 1981; and especially Brewer, 1985.) A "diversified" regional economy has been defined variously as one in which (1)all industries are of equal size, (2) the industry mix minimizes portfolio variance, or (3) the region's industry mix is the same as the nation's. Measures that define complete diversity as equal representation by all industries ("ogive" and "entropy" measures) are particularly arbitrary, since they depend critically on industry definitions. For example, an ogive or entropy measure that uses two-digit SIC data implies that tobacco manufacturing and health services would be equally important in a completely diversified regional economy. The portfolio variance concept currently is the most widely accepted measure of diversity, and it can be a valuable tool if used appropriately (Gruben and Phillips, 1989a and 1989b). However, it should not be used to test whether diversity reduces volatility (Conroy, 1975; and Brewer, 1985) because it does not measure diversity independent of volatility. Examining the formula for the portfolio variance measure reveals why: Vp = t ,: wiwjVij (3) where Vp denotes portfolio variance, Vij denotes the variance (i = j) or covariance (i :f::.j) for each industry or pair of industries, and Wi and wj are industry weights. Traditionally (Conroy), regional data are used to calculate the industry weights, w, but due to data and computing con- straints, or the particular task to which the measure is tailored (Gruben and Phillips, 1989b), the industry variances and covariances, V, are calculated using national data. As a general rule, if sufficient information and computing resources are available, the portfolio variance Vp should be calculated using regional variances and covariances. If all of the data on the right-hand side of equation (3) are consistent with each other, in terms of regional coverage as well as the economic concept they are measuring (employment, income, or gross product), the right hand side of equation (3) is simply the decomposition ofthe region's total variance. Thus, the portfolio variance measure of diversity, correctly calculated, is exactly the same as the region's total variance, which is a frequently-used measure of economic instability. Therefore, the portfolio variance measure does not measure diversity independent of volatility, and it is not surprising that the portfolio variance measure tends to "explain" differences in volatility better than other" diversity" measures do. If the analogy with portfolio theory holds, regional economic diversity should be defined in terms of the "market" industrial mix. Ideally, this "market" industry mix would reflect the comparative advantage of each region. However, it is impossible to calculate an ideal "diversified" industry mix that is different for each region and that distinguishes between ideal and actual industry structure. In view of these limitations, the national industry mix provides a standard with which to gauge a region's industry structure. Such a standard implies that regions seeking to diversify their economies should attempt to duplicate, to the extent possible, the industrial structure of the United States. Of course, no region could (or should) duplicate the U.S. industrial structure precisely, since geographical differences in comparative advantage will determine the region's optimal industry structure to a significant extent. Nevertheless, for most regions, the U.S. industrial structure provides a standard for diversity that is more reasonable than the available alternatives. III. Data and Variables The analogy between portfolio theory and regional economic stability suggests two testable hypotheses. First, regional economic diversity should reduce nonsystematic volatility. Second, growth should be positively correlated with systematic variations in the region's economy. Gross State Product (GSP) data, released by the Bureau of 20 Economic Analysis.> were used to test these hypotheses. These are annual data, adjusted for inflation, and disaggregated by state and by industry to the two-digit SIC level.6 They are available for the years 1963 through 1986. The variables used in this analysis are defined below. Economic Review / Winter 1990 systematic. Systematic volatility (SYSV), measured in standard deviation terms, is therefore: Variable Definitions Diversity Portfolio theory defines diversity as the extent to which a portfolio's composition approximates the "market" portfolio. Similarly, regional economic diversity is defined here as theextent to which a region's industrial structure approximates that of the nation. This measure (DIV) is derived using the following formula for each state and year. J D. :;;;: !, It )=1 (GSPj)t -GSPs)tF GSP US)t DIVj approaches infinity for states with economies that resemble the industrial structure of the U.S. very closely, and approaches zero for states with economies that deviate substantially from the U.S. industrial structure. Growth AVGRGSPj measures the long-term growth rate in real total GSP for state i. Annual percentage growth rates are calculated for each state and year (GROWTHit) , and averaged across time periods t for each state i. Total volatility Total volatility, TOTSTDj, is measured as the standard deviation over time in the state's annual percentage growth rate, GROWTHir In order to decompose the variance into its systematic and nonsystematic components, the variance (TOTVARj :;;;: TOTSTD?) also is calculated. Systematic and Nonsystematic Volatility A simple univariate regression of state growth on national growth is used to divide total volatility into its systematic and nonsystematic components: :;;;: IX + 13 GROWTH u s t + e jt (6) The (unadjusted) R2 from this regression measures the proportion of total variance in state i's growth rate that is associated with contemporaneous variations in national growth." This is the portion of the state's variance that is Federal Reserve Bank of San Francisco (7) Nonsystematic volatility is the total volatility that is not associated with variations in national economic growth. In standard deviation terms: NONSYSV j :;;;: vO - Ry) (TOTVARJ (8) Systematic Sensitivity (4) where GSP)t denotes the share of total GSP in industry j during period t, i subscripts denote states, and US subscripts denote national figures. 7 After D it is calculated, its reciprocal is taken, so that greater diversity is associated with a higher value for the diversity measure," and the measured is averaged over time within each state: I 1986 I DIV.:;;;: ~ (5) I 24 t=1963 Djt GROWTH jt SYSVj :;;;: v(Rr) (TOTVARJ The coefficient beta from regression (6) is analogous to the beta coefficient often calculated for individual stocks, and measures the region's sensitivity to national economic conditions. This measure differs from that for systematic volatility, described above. The beta measures the magnitude, and hence the sensitivity, of the response of state to national changes. In contrast, systematic volatility measures the extent to which variations in the national economy explain local fluctuations, regardless of the size of their impact. A Look at the Variables Table 1 presents the value of each variable calculated for each state. DIV exhibits a wide range of values across states, suggesting that states differ significantly from each other in their degree of diversity. According to this measure, Washington, D.C. is the nation's least diverse economy, while Illinois is its most diverse. The rankings implied by these values are not surprising. The District of Columbia's economy is strongly oriented toward government, and Illinois has a large and diverse economy. Moreover, the measures for Alaska's economy, which is quite specialized, and for California's economy, which is very diverse, appear reasonable. However, a few DIV values are somewhat surprising. For example, DIV values for Missouri and Colorado are higher than one might expect. Nevertheless, the overall rankings appear to be plausible. Average GSP growth (AVGRGSP) also varies considerably from state to state. Between 1963 and 1986, Alaska was the fastest-growing state, at an 8.1 percent average annual rate. The District of Columbia experienced the slowest GSP growth, at only 1.5 percent per year. Other fast-growing states included Arizona and Florida, while West Virginia, Pennsylvania, and Illinois were among the nation's slowest growing states. Considerable variation also is apparent in the values for the coefficient beta from equation (6), which measures systematic sensitivity. The strongest measured responses to national changes occur in the industrial states of Michigan 21 Nonsystematic Volatility (NONSYSV) LIO 3.14 2.96 1.94 3.43 2.55 2.18 2.87 0.48 22 10.28 2.86 1.60 1.32 1.95 1.76 3.26 2.20 2.23 1.13 3.33 3.34 0.97 I.l2 2.70 I.l5 1.32 4.04 1.71 1.22 1.71 2.17 1.25 2.03 I.ll 3.17 2.41 3.25 2.27 1.49 2.29 1.59 I.l4 5.43 0.71 2.58 2.34 0.82 1.75 1.33 3.24 I.l7 1.81 2.50 1.97 I.l5 2.76 1.88 0.77 5.94 Economic Review / Winter 1990 and Ohio. In contrast, the weakest responses are found in the energy-dependent states of Wyoming and Oklahoma. A look at the standard deviation of the annual growth rates reveals that Alaska's was by far the most volatile state economy in the nation during this period. Other relatively volatile economies included Wyoming and North Dakota. At the other end of the spectrum, the nation's most stable economies during this period included Kansas, the District of Columbia, California, and Colorado. Changes in the national economy affect different states in different ways, as reflected in the R2S for equation (6), which are listed in column 6 of Table 1. National influences are relatively unimportant for Hawaii, Wyoming, and North Dakota, but they explain more than 90 percent of the total variations in the economies of Illinois, Indiana, Ohio, Pennsylvania, and Wisconsin. The remaining columns in Table 1 decompose the total volatility into that explained by national fluctuations (SYSV) and that which is nonsystematic (NONSYSV). Nonsystematic volatility is highest for states with a combination of a high standard deviation and relatively low R 2, such as Alaska and Wyoming. Nonsystematic volatility is low for states that exhibit only moderate variation, most of which is explained by national movements. Wisconsin and Pennsylvania. fall into this category. IV. Empirical Results This section presents the results of tests of the following two hypotheses: (1) Nonsystematic volatility should be lower in states with more diverse economies. (2) Growth should be positively correlated with systematic sensitivity, as measured by the beta coefficient calculated in equation (6). Note that the discussion of the analogy between portfolios and regional economies suggests that the first hypothesis is more likely to be corroborated than is the second. Diversity and Volatility Correlations between diversity and volatility are summarized in Table 2. 10 The correlation coefficient between diversity and nonsystematic volatility is significantly different from zero at the 99.8 percent level, with a magnitude of -0.425. The extremely high level of statistical significance is particularly noteworthy. Thus, as expected, states with more diverse economies tend to experience less nonsystematic volatility. This suggests that risk spreading is applicable to regional economies. To get a sense of how important the components of volatility are to this hypothesis test, Table 2 also presents correlations between diversity and both systematic and total volatility. Results suggest that no correlation between diversity and systematic volatility exists. The correlation coefficient is 0.087, and is significant only at the 45.4 percent level." The correlation coefficient between diversity and total volatility is -0.284, and is significant at the 95.6 percent level. This relationship is slightly weaker than that between diversity and nonsystematic volatility, although it is somewhat stronger than most other measured relationships between national average diversity and total volatility. 12 Federal Reserve Bank of San Francisco Systematic Sensitivity and Growth The .relationship between systematic volatility and growth is measured as the "security market line" relationship in the financial literature. (See, for example, Sharpe, 1985.) The equation estimating this relationship is: AVGRGSP = 4.00 - 0.85 BETA (15.00) (3.37) R2 = .172 Numbers in parentheses are t-statistics, Note that the coefficient, which the portfolio analogy predicts should be positive, is in fact negative and statistically significant. However, Alaska's summary statistics in Table 1 suggest that the state may be an outlier. If Alaska is omitted from the sample, the coefficient becomes positive, but statistically insignificant: AVGRGSP = 2.98 (7.72) + 0.20 BETA R2 =- .015 (0.51) 23 The lack of a significant positive relationship between beta and growth strongly suggests that there is no mechanism in regional economies that generates a tradeoff between systematic sensitivity and growth. In fact, a negative relationship between systematic sensitivity and growth is consistent with previous work by Sherwood-Call (1988) and with Schmidt's 1989 work on resource industries during the 1963-1986 period. Schmidt found that resource-dependent states tended to grow more rapidly during this period than did states that did not depend heavily on natural resource industries. Sherwood-Call found that resource dependence tended to be negatively correlated with the extent of linkage to the national economy. Taken together, these results suggest that resource-dependent states may have weaker associations with movements in the national economy than most states do, which could translate into smaller beta coefficients, while at the same time these states experienced relatively rapid growth during the period under study. SUmmary of Empirical Results These empirical results suggest that regions maybe able to improve the stability of theireconomies by diversifying them. 13 •Regional economic diversity is negatively correlated•• withthenonsystematic.component of volatility in an extremely significant way. However, regionsdonotseemto be compensated for accepting more systematic sensitivity through higher growth rates. v. Conclusions and Implications Previous studies of the relationship between regional economic diversity and economic volatility have yielded mixedresults. These studies focussed onmeasurement and econometric issues in seeking to explain the conflicting results. These measurement and econometric issues are serious ones, but this paper has focused on a fundamental conceptual problem with the previous studies. Most researchers have looked for a relationship between diversity and total volatility, whereas the portfolio analogy suggests that the relationship is between diversity and nonsystematic volatility. In this paper, simplestatistical tests have shown that the expectedrelationship between diversity andnonsystematic volatility does exist and is extremely strong. These observations, which parallel those in the portfolio literature, reflect the risk-spreading that occurs as regional economies diversify. However, there is no correlation between systematic sensitivity and growth, although the portfolio analogy seemsto suggest thatsucha relationship should exist. This result is not surprising, since the mechanism by whichthe tradeoff occurs in financial markets does not exist for regional economies. The financial market relationship between systematic risk and return in portfolios occurs becauserisk-averse investors willnotholdhigh-risk assets unless they expect to be rewarded with higher returns. 24 Regional economies, in contrast, lack a singleomnipotent decision-maker, and the "market" for industries is illiquid and slow to adjust. The implications for regional policy makers are relatively straightforward: greater economic diversity improves the stability of a region's economy. Thus, other thingsequal,regional development officials should be able to improve their region's economic stability by making their regional economies more diverse.v' However, the instability that is associated with fluctuations in the national economy remains significant source of instability formoststates,and it is notcompensated by highergrowth rates as the analogy with portfolio theory suggests it should be. While this study has focussed on issues of regional economic stability, it is important to notethat regions may pursueothereconomic goals, suchas rapidgrowth, instead of or in addition to seeking economic stability. Fora region that has a natural resource, or an agglomeration of activity that provides it with a comparative advantage in a particular industry, pursuing that advantage may be a more effective overall strategy than a diversification strategy would be. At the same time, a region that develops an industry mix that yields strong growthneed not "pay" for that rapid growth by accepting greater instability. a Economic Review / Winter 1990 NOTES 1. However, because an industry is made up of many firms, small states may have more volatile economies even if they have diversified industrial mixes. Since differentfirms in a particular industry mayexperience different fortunes, diversification across firms within an industry probably has benefits as well. These issues are not addressed in this paper. 2.:. The differences. among regions' industries are even greaterthanthe data usedin this study indicate, because industry detail. is available only to the two-digit SIC level. Thus, for example, the transportation equipment category does not distinguish between motor vehicle manufacturing, which is important in Michigan, and aerospace production, which is important in California. 3. Even if local officials had control over their region's industry mix, the community's residents and politicians are likely to disagree about what industry mixthe region should movetoward. While some may preferto maximize economic growth, others might prefer slower growth if it allows them to maintain the community's character. 4. The most commonly used measures include a representative "marketbasket"of securities, suchasthe stocks included in the Dow Jones or S&P 500 index. These measures do not. however, include bonds, real estate, or other non-security assets. 5. Most previous studies of the relationship between economic diversity and economic stability have used employment data. While the employment data have the advantage of being monthly, they provide a less comprehensive measure of economic activitythan GSP does, and also suffer from a large number of missing values. 6. Most industries are disaggregated to the 2-digit level. A few, including construction and retail trade, are disaggregated only to the 1-digit level. 7. U.S. production for each industry was calculated by summing GSP across states. 8. The reciprocal istaken onlyso thata highermeasure is associated with greater diversity, making results easier to interpret. It does not materially affect the results. 9. An alternative measure of the relative contribution of national changes to regional economic fluctuations was developed in Sherwood-Call (1988). That linkage measure accounted for lags in the transmission of economic changes from the national to the state level. However, the R4 measure parallels work done in the portfolio literature. 10. The data presented in Table 1 suggest that Alaska is anoutlier, whichmaybiasthe results presented inTable 2. To. determine whether this is the case, all of the empirical estimates wererecalculated usinga sample thatexcludes Alaska. The results indicate that the calculations presentedin Table 2 are not driven solely by Alaska. 11. The positive sign on the correlation coefficient may be due to a spurious correlation that results from the way the diversity variable is constructed. The most "diverse" economies are those with industrial structures that most closely resemble the national economy. If each industry exhibits similar fluctuations over time in various regions of the country, then the states that haveindustry mixes that mostclosely resemble the U.S. industry mixalsoare likely to experience economic fluctuations in concert with national economic fluctuations. 12. The differences between these results andthe results of other studies that used national average diversity measures may be due to differences in the geographical or industrial coverage. Most previous studies looked at metropolitan areas rather than states, and examined only manufacturing activity. 13. The empirical work presented hereexamines a static measure of diversity overa cross-section ofstates. Thus, it does not explicitly examine the benefits that a particular state would gain from diversifying its own economy. Gruben and Phillips (1989a) address that issue directly. 14. Gruben and Phillips (1989a) suggest that regions interested in reducing total volatility target industries that have small or negative covariances with existing industries. REFERENCES Attaran, Mohsen. "Industrial Diversity and Economic Performance in U.S. Areas," Annals of Regional Science 20:2, 1986, pp. 44-54. Barth, James, John Kraft, and Philip Wiest. "A Portfolio Theoretic Approach to Industrial Diversification and Regional Employment," Journal of Regional Science 15:1,1975, pp. 9-15. Black, F., M.C. Jensen, and M. Scholes. "The Capital Asset Pricing Model: Some Empirical Tests," in M.C. Jensen, ed., Studies in the Theory of Capital Markets. NewYork: Frederick A. Praeger, Inc., 1972. FederalReserve Bankof San Francisco Brealey, Richard and Stewart Myers. Principles of Corporate Finance, 2nd ed. New York: McGraw Hill, 1984. Brewer, H.L. "Measures of Diversification: Predictors of Regional Economic Instability," Journal of Regional Science 25:3,1985, pp. 463-470. Brown, Deborah J. and Jim Pheasant. "A Sharpe Portfolio Approachto Regional Economic Analysis," Journal of Regional Science 25:1,1985, pp. 51-63. Conroy, Michael E. "The Concept and Measurement of Regional Industrial Diversification," Southern EconomicJournal 41 , 1975, pp. 492-505. 25 Fama, E.F. and J.D. MacBeth. "Risk, Return and Equilibrium: Empirical Tests," Journal of Political Economy 81,1973, pp. 607-636. Gibbons, M.R. "Multivariate Tests of Financial Models," Journal of Financial Economics 10, 1982, pp.3-27. Gruben, William G..and •Keith R.Phillips. "Diversifying Texas; Recent Historyand Prospects, "Economic Review, Federal Reserve Bank of Dallas, July 1989(a), pp.1-12. Gruben, WiliiamC. andKeith R. Phillips. "Unionization and Unemployment Rates: A.reexamina.tionof Olson's LaborCartelization Hypothesis," mimeo,1989(bt Jackson, Randall W."An Evaluation of Alternative Measures of Regional Industrial Diversification," Regional Studies 18:2,1984, pp. 103-112.. Kort, John R. "Regional Economic Instability and IndustrialD.iversification intheU.S.,"Land Economics 57:4, 1981, pp. 596-608. North, Douglass C. "Location Theory and Regional Economic Growth," Journal of Political Economy (63), 1955, pp. 243-258. 26 Schmidt, Ronald H. "Natural Resources and Regional Growth," Economic Review, Federal Res.erve Bank of San Francisco, Fall 1989. Sharpe, William F. Investments, 3rd ed. Englewood Cliffs, NJ: Prentice-Hall, 1985. Sherwood-Call, • Carolyn.. 'lfxploring the Relationships Between National and Regional Economic Fluctuations," Economic Review, Federal Reserve Bank of San Francisco, Summer 1988. Steib, Steve B. and R. Lynn Rittenoure. "Oklahoma's Economic Growth and. Diverslflcatlon,": paper presented a.tthe annual meetingottbEl VVElstElrnRegiQoa.1 Science Association, San Diego, February 1989. St. Louis, Larry. "A Measure of Regional Diversification and Efficiency." Annals of Regional Science, March 1980, pp. 21-30. Wasylenko, Michae.1 J. and Rodney A. Erickson. "On Measuring Economic Diversification," LandEconomics 54:1,1978, pp. 106-109. Economic Review / Winter 1990 External Shocks and Adjustment in Four Asian Economies-1978-87 Ramon Moreno Economist, Federal Reserve Bank of San Francisco. Editorial committee members were Gary Zimmerman, Hang-Sheng Cheng, and Brian Motley. Research assistance by Judy Horowitz is acknowledged. This paper presents a small open-economy model that illustrates thepossible equilibrium real exchange rate and current accountresponses to changes in the world rate of interest and the terms of trade. The model provides a simple framework for assessing whether theadjustment to external shocks in four Asian economies in 1978-87 was roughly consistent with equilibrium, and the implications for economic performance. Overall, the ability of these four economies to prevent real exchange rate overvaluation in the face of adverse external shock appears to have contributed to their very successful economic performance in 1978-87. However, thediscussion provides somesupportfor theview thatreal exchange rates may have become undervalued in some Asian economies, particularly after 1985, and thatopportunities for increasing investment and consumption in these economies may have been missed. Federal Reserve Bank of San Francisco Since the second half of. the 1970s, the developing economies of easternAsia have experienced faster growth and lower inflation than other developing countries as a group have. This economic performance is remarkable in light of the significant disturbances to the world economy duringthis period-a runupin oil pricesandin world rates of intereststartingin the late 1970s, a slumpin commodity prices and a slowdown in the growth of industrial economies in the early 1980s, sharp changes in the value of the U.S. dollar, and the LDC debt crisis. The success of Asian economies generally is attributed to soundeconomic management. One oftencited theme is that these Asianeconomies responded more appropriately to external disturbances than did other developing economies,' thereby successfully preventing overvaluation in their real exchange rates. However, not everyone believes that the policy responses of Asian countries have been appropriate. Critics argue that Asian economies have carried their adjustment efforts too far, and that as a result, their exchange rates have become undervalued. This paper attempts to examine the appropriateness of economic policies in Asian economies in greaterdetail by addressing two questions. The first is whether adjustment in Asianeconomies in response to external shocks appears to be broadly consistent with long-run equilibrium. The second question is how shocks and the response to them may have influenced economic performance in the economies of the region. The four Asian economies examined here may be loosely divided into two groups following roughly comparable economic policies: Korea and Thailand, and Malaysia and Singapore. This paper is organized as follows. Section I develops a theoretical framework for assessing whether adjustmentin Asianeconomies wasconsistent withequilibrium. Section II describes the equilibrium response to external shocks and provides definitions of real exchange rate misalignment. Section III describes the experiences of Asianeconomies in theperiods from1978 to 1982 andfrom 1983 to 1987. External shocks and their impact on the region are described, and the possible role of policy responses in explaining variations in economic performance is discussed. Some conclusions on the appropriateness of policy responses in each of the four Asian economies are also offered. 27 I. External Shocks and Equilibrium Adjustment Two main types of external shocks affected Asian economies in the ten years from 1978to 1987:sharp fluctuations in world rates of interest, and changes in the terms of trade. External shocks during the 1978-82 period generally had adverse effects on the Asian economies. After declining sharply from 1974 to 1977, the world rate of interest, as represented by the 3-month London Interbank Offer Rate, rose between two and three percentage points every year between 1978and 1981, to peak at a level of over 18percent in 1981. The run-up in interest rates tended to reduce the disposable income of the debtor countries in the region, although the vulnerability of each of the four countries varied considerably. Over the period 1978-82, Korea and Thailand averaged higher ratios of external debt to GNP (respectively, 45 and 28 percent), and consequently were more vulnerable to increases in world interest rates than were Singapore and Malaysia (with external debt-to-GNP ratios, respectively, averaging 18 and 23 percent). Other major shocks affected the terms of trade of these Asian economies. Double-digit increases in oil prices each year between 1979 and 1981 adversely affected the terms of trade of three of the four Asian economies, with only Singapore the exception since it exports petroleum products. And although oil prices declined to still-high levels in 1982, the favorable impact on the terms of trade was offset by a slowdown in the growth of industrial countries. Among other things, the slowdown in growth contributed to a decline in non-oil commodity prices, which particularly affected Thailand and Malaysia. Finally, the dollar began a steady appreciation against the currencies of other industrial countries after 1979. This increased the competitiveness of industrial countries in U.S. markets, potentially at the expense of the four Asian economies, particularly those pegging to the U.S. dollar. The world economic environment changed significantly in the second period, 1983-1987. The nominal world interest rate declined an average of 1.3 percentage points a year, and by 1986 interest rates were close to their lows of ten years earlier.? Earlier shocks affecting the terms of trade of Asian economies also were reversed. Oil prices declined further, and the growth of industrial economies recovered, producing an erratic recovery in world commodity prices and a moderate and sustained expansion not seen since the 1960s. After 1985, the dollar's depreciation apparently increased the competitiveness of Asian economies in the U. S. market as the cost of Japanese and European goods rose. Against this largely favorable background was the interruption in voluntary lending to heavily indebted developing countries after 1982, which had a rela- 28 tively small direct impact on the economies discussed in this paper, but may have encouraged them to seek to limit the growth in their external debt. To examine how these interest-rate and terms-of-trade disturbances may have affected the equilibrium real exchange rates and external balances of Asian economies, a theoretical model of a small, open economy is developed below. Aggregate Supply and Demand Consider an economy with three goods, non-tradable, exportable, and importable, that are all produced and demanded by domestic residents. The full-employment aggregate supply of goods (S) in the economy is given exogenously by endowments. Domestic prices are assumed to be flexible, the nominal exchange rate is fixed by the government, and world prices of importable and exportable goods are determined in world markets. Total supply can be expressed as the sum of the supply of each of the three types of goods produced in the economy: S = PNS N + PMS M + PxS x = PNSNS + PMSMS + PxsxS (1) where PN, PM' and P x, respectively, refer to the domestic currency prices of non-tradable, importable, and exportable goods, S, (i = N, M, and X) is the supply of each good, and Sj is the share of each good in total supply. The Appendix provides a full list of variable definitions. Aggregate demand, D, is the sum of the demand for non-tradable, exportable, and importable goods: D = PNDN + PMDM + PxD x = PNdND + PMdMD + PxdxD (2) where D, is total demand for each good and d, is the share of each good in aggregate demand. Aggregate demand is the sum of equilibrium private demand, De, and government demand, Ds. Equilibrium private demand, in turn, is a function of consumer demand and investment demand. Equilibrium consumer demand is the result of intertemporal utility maximization subject to the economy's intertemporal budget constraint, while investment demand is determined by the optimizing condition that the world cost of funds (i*) should equal the marginal product of capital (MPK). The solution to these two conditions yields the equilibrium demand of the private sector.3 The equilibrium private demand De is increasing in the Economic Review / Winter 1990 private sector's disposable permanent income, PI. Permanent income, in turn, depends on world interest rates and the terms of trade. If the country is a net debtor, a permanent increase in world interest rates increases the volume of exports needed to service a given stock of debt outstanding, thus reducing the permanent income available for domestic consumption. A permanent decline in the terms of trade also reduces permanent income by increasing the volume of exports needed to purchase a given volume of imports. Aggregate demand in the economy also can be directly influenced by government policy (Ds), which can help or hinder the private sector in achieving an optimal spending path." These relationships may be summarized as follows: D = De[PI(i*, Px/P m), i* - MPK] +- + Dg + Equilibrium in: Non-tradable goods market e = PN I EP; (3) + N° Figure 1 Equilibrium in Non-Traded Goods It is assumed that the respective shares of non-traded goods in total supply and demand depend on the relative price of traded and non-traded goods, e, here defined as5 PN e = Non-tradable goods Importable goods market (4) E(PT *) where E is the nominal exchange rate (in units of domestic currency per unit of foreign currency), PT is the price of tradable goods, defined as a suitably weighted average of exportable and importable goods prices," and the * refers to prices expressed in foreign currency. The relative price, e, is often called the "real exchange rate" in the literature, since it reflects the country's profitability, or competitiveness, of production in the traded goods sector. (Moreover, when non-traded goods and world prices adjust sluggishly, changes in e reflect changes in the nominal exchange rate. 7) Thus, a rise in e increases the share, SN' of non-traded goods in total supply (SN is rising in e) because the increased profitability of non-traded goods production shifts resources to the non-traded goods sector. At the same time, a rise in e lowers the share, dN, of non-traded goods in total demand (dN is falling in e). Given the above assumptions, the equilibrium in the non-traded goods sector is determined by the requirement: PNsN(e)S = PNdN(e)D E/ P~ . .. .. M o- ---..... Importable goods Figure 2 Exportable goods market E/P N (5) + In equation (5), aggregate domestic supply S is determined exogenously by endowments, while aggregate domestic demand D is determined in equation (3) by ....x o... Exportable goods Figure 3 Federal Reserve Bank of San Francisco 29 variables that also are exogenous to the model. Since nominal exchange rates (E) and the world prices of exports and imports are exogenously determined, the price of nontraded goods will adjust to satisfy equation (5), resulting in an equilibrium real exchange rate that clears the nontraded goods market. The short-run equilibrium real exchange rate is represented as eO in Figure 1. excess demand for importable goods, resulting in total imports MO. The level of exports can be determined in a similar fashion, that is, PxS x = Pxsx(e, Px/PM)S (8) + and External Sector PxD x = Pxdx(e, Px/PM)D To see the relationship between the equilibrium real exchange rate and the external sector, consider the market for imports of goods and services. The shares of importable goods in total domestic supply and demand are functions of the real exchange rate as well as of the relative price of exportable and importable goods. Thus, a rise in the real exchange rate or in the price of exportable versus importable goods (the terms of trade) will reduce the domestic supply of importable goods, while increasing the domestic demand for such goods. That is, PMSM = PMsM(e, P/PM)S (6) and PMDM = PMdM(e, P/PM)D + (7) + Figure 2 illustrates the domestic supply and demand schedules for importable goods in the economy. Since the schedules are defined in value terms, changes in export or import prices are reflected in shifts in the supply and demand curves. On the other hand, changes in the nominal exchange rate, or in the price of non-traded goods (shown in the vertical axis) will be reflected in movements along the curves. Given a nominal exchange rate set by the government, the price of non-traded goods pg.that clears the non-traded goods market (which corresponds to the real exchange rate eO in Figure 1), is associated with an (9) + In this case, the price of non-traded goods, ~, is associated with an excess domestic supply of exportable goods, corresponding to the equilibrium level of exports XO (Figure 3). Since the non-traded goods market always clears (SN= D N), the difference between the aggregate supply and demand in the economy equals the current account balance (CA), or the excess supply of goods and services in the traded-goods sector: S - D = CA = Px(Sx-Dx) - PM(DM-S M) = PxX - PMM (10) Suppose there is no government, so D in equation (10) equals the equilibrium demand of the private sector, De, in equation (3). Given eO, equation (10) then defines an equilibrium current account, or level of external lending or borrowing, that is consistent with private intertemporal utility maximization. In equilibrium, current account deficits (and borrowing) are expected in a developing country for two reasons. First, per capita incomes are lower than they are expected to be in the future, so it is welfare-enhancing to borrow in order to smooth consumption. Second, due to the relative scarcity of installed capital, the marginal product of capital is likely to be higher than that of industrial countries, making borrowing to finance investment profitable. 8 II. External Shocks and Misalignment This model can be used to determine how an economy would adjust to two kinds of external shocks. First, an interest-rate shock is considered, and then a terms-of-trade shock is evaluated. Assume that a country is initially at an equilibrium e" that clears the non-traded goods market and that corresponds to an equilibrium current account deficit. A permanent rise in the cost of funds (world rate of interest) reduces permanent disposable income for countries that are net debtors and raises the cost of funds above the existing marginal product of capital. The reduction in 30 permanent income leads to a fall in consumer demand, and the rise in the cost of funds reduces investment demand. The result is a fall in aggregate demand D in equation (3), shifting the demand schedule for non-tradable goods to the left, as illustrated in Figure 4. In order to clear the nontraded goods market, the real exchange rate must depreciate to e'. If nominal exchange rates are fixed and prices are flexible, the depreciation will be accomplished through a reduction in the price of non-traded goods. As can be seen in Figure 5, the reduction in permanent Economic Review I Winter 1990 An adverse external shock reduces aggregate demand and leads to: A reduction in the demand for non-tradables and real exchange depreciation e = PN / EP; Non-tradable goods A fall In the equilibrium level of imports E/ P~ E/ P~ ""'..-r; •••• c .; ,,""' ~M/ - MoFigure 5 ~ Importable goods and an increase in the equilibrium level of exports E/P N E/P~ .. E/p~I··· ·;....: income and aggregate demand shift the domestic demand schedule for importables to the left, which, in combination with the decline in the equilibrium price of non-traded goods, results in a reduction in the equilibrium amount of imports. The corresponding fall in the domestic demand for exportable goods leads to an increase in the level of exports (Figure 6), and a decline in the equilibrium current account deficit. 9 The analysis of terms-of-trade shocks is very similar to that of interest-rate shocks. Suppose a supply-side disturbance, .such as a rise in imported oil prices, leads to a worsening in the terms of trade. The resultant decline in permanent income lowers the demand for non-tradable goods,thereal exchange rate depreciates, imports fall, and exports and the current account balance tend to rise, as illustrated in Figures 4 to 6. 10 In the present discussion, shifts in supply have been ignored to simplify the illustration,but they can be incorporated by assuming that factor productivity depends on imported inputs. In this case, a rise in import prices may lead to a reduction in aggregate supply which is associated with a real exchange rate appreciation, and a deterioration in the current account balance in the short-run." These effects will tend to be reversed as domestic demand contracts in response to the reduction in permanent income. In addition to being subject to supply-side disturbances, the terms of trade may be affected by changes in foreign demand and in third country exchange rates. The terms of trade may tend to improve if foreign trading partners grow faster, as demand abroad rises.P On the other hand, the impact of changes in third country exchange rates on the terms of trade is ambiguous. For example, if Korea exports mainly to the US. and imports from Japan, a dollar appreciation against the yen can lower the price of imports relative to exports, thus improving the terms of trade.P However, the appreciation also can induce a substitution in US. demand awayfrom Korean exports in favorof cheaper Japanese exports, which may adversely affect Korea's terms of trade. The appropriate policy response will depend on which effect appears to be dominant. In particular, if the substitution effect is stronger, it is appropriate for Korea to weaken its currency against the US. dollar, in order to offset the incipient substitution toward Japanese goods. Misalignment and Nominal Exchange Rates ...xo...... - - - X1 Exportable goods • Figure 6 Federal Reserve Bankof San Francisco An increase in world interest rates or adverse movements in the terms of trade may produce two different kinds of real exchange rate misalignment, which may be illustrated by reference to Figure 4. First, price rigidity in 31 non-traded goods may prevent an adjustment in the real exchange rate from e" to e' in Figure 4. The real exchange rate will be overvalued in the sense that there is an excess supply of non-traded goods in the short-run. Equilibrium in the non-traded goods sector can only be achieved by contracting output, resulting in unemployment, which over time deflates the price of non-traded goods and eliminates the misalignment. One way of avoiding unemployment under these conditions is to depreciate the nominal exchange rate, speeding the adjustment to e'. 14 A second kind of misalignment results when the government seeks to offset the contractionary impact of external shocks by adopting expansionary fiscal and monetary policies to prevent unemployment, thus maintaining the short run equilibrium at eO. In this case, the demand schedule will not shift to the left (demand will still be represented by the schedule D~) in Figure 4, and the levels of imports and exports will remain at MO and XO in Figures 5 and 6, respectively. The current account balance will also remain unchanged. Since a domestic demand contraction is the equilibrium response, the real exchange rate at eO will now be overvalued in the sense that a country is borrowing more than is optimal, or perhaps sustainable, in the long-run. 15 In this model, nominal exchange rate policy cannot prevent the real exchange rate overvaluation that results from expansionary domestic demand policies. Ifthe nominal exchange rate depreciates, e may temporarily fall below the equilibrium eO in Figure 1. Over time, however, the resulting excess demand for non-traded goods leads to inflation, and a real appreciation back to e". By the same token, an increase in permanent income due to a decline in world interest rates or an improvement in the terms of trade can lead to misalignment if prices adjust slowly and the real exchange rate fails to appreciate, resulting in excess demand in the non-traded goods sector and inflationary pressures. Alternatively, misalignment can arise if domestic demand fails to expand in response to the increase in permanent income. In this case, the country is borrowing less than is optimal, at the cost of missed opportunities for increasing present consumption or investment in a manner that will maximize intertemporal utility. In both of these examples, the real exchange rate is undervalued. Qualifications Adjustment to external shocks in developing economies frequently differs from the preceding description of the equilibrium, at least initially. Three factors may account for the discrepancy. 32 First, adjustment takes time. A rise in the world rate of interest, or adverse movements in the terms of trade, initially may be associated with an increase (rather than a decrease) in the current account deficit, because interest payments to foreigners rise or export revenues fall. In addition, if adverse terms-of-trade shocks lead to a contraction in domestic supply, domestic prices may rise more than the prices of traded goods, leading to real exchange rate appreciation, rather than depreciation. However, initial increases in the current account deficit and real exchange rate appreciation may not be inconsistent with adjustment to the long-run equilibrium suggested by the model if they are reversed over time. Second, the equilibrium pattern of adjustment depends on whether external shocks are permanent or temporary. In the preceding discussion, and in what follows, it is assumed that shocks are perceived as permanent at the time that they occur. (This is plausible, as the shocks were large and generally were not reversed for several years.) However, if an adverse terms-of-trade shock is perceived as temporary, an increase in external borrowing to smooth consumption will enhance welfare, and expansionary policies (or government guarantees for private external borrowing) may be consistent with equilibrium adjustment if domestic residents are unable to borrow abroad. Third, the equilibrium pattern of adjustment is influenced by a country's previous borrowing. A country that has borrowed little in the past may adopt expansionary policies that offset the contractionary impact of external shocks without producing disequilibrium. On the other hand, a country whose exchange rate has been systematically overvalued in the past, and has been borrowing more than the optimal amount probably will have to respond more vigorously to depreciate the exchange rate and reduce current account deficits. In cases where past borrowing has been excessive, countries may even seek to generate current account surpluses to reduce external debt to manageable levels. This may be a deliberate choice on the part of policy makers, or may be compelled by the interruption of voluntary lending by creditors. Equilibrium Responses To sum up, in equilibrium, a permanent increase in the world rate of interest or a permanent decline in the terms of trade is likely to be associated with a depreciation in the equilibrium real exchange rate (or an initial appreciation in the real exchange rate that is later reversed) and a decline in domestic demand and in the equilibrium current account deficit. Policies to facilitate this type of adjustment (a currency devaluation if domestic prices are rigid, or pal- Economic Review / Winter 1990 icies to reduce domestic demand and external borrowing) would tend to prevent real exchange rate misalignment. On the other hand, policies that hinder adjustment to the external shock may lead to an overvalued real exchange rate, resulting in declining growth or in unemployment, or alternatively in higher than optimal external borrowing. Conversely, when a country experiences favorableexternal shocks, policies that facilitate the equilibrium real appreciation and increases in external borrowing would prevent misalignment. A number of factors, notably adjustment lags and debt management considerations, may influence the observed pattern of adjustment. III. External Shocks and Their Impact The responses of the four east Asian economies to external shocks in the ten years from 1978 to 1987 can be evaluated against the equilibrium suggested by the model presented in the preceding section. Since the external shocks in the first half of this period largely were adverse and the shocks in the second half generally were favorable, the analysis proceeds by evaluating the four economies' adjustment first in the 1978-82 period and then in the 1983-87 period. Table 1illustrates the impact of changes in world interest rates and in the terms of trade in the four Asian economies during the 1978-82 period. To represent the impact of changes in world interest rates on income, the product of the change in the U.S. dollar London Interbank Offer Rate and the ratio of gross external debt to nominal GNP was multiplied by -1for each country. The impact of changes in the terms of trade was calculated as the product of the percentage change in the commodity terms of trade in each country and the corresponding ratios of real imports to real GNP. Korea was most adversely affected by external shocks. The rise in Korea's estimated interest burden, as measured by the ratio of estimated interest liabilities to GNP,16 reduced disposable income by an average of 0.5 percentage points a year in the period 1978-82, while adverse movements in the terms of trade had an average annual impact of 1.2 percent of GNP over the period.'? Due to their status as commodity exporters, both Thailand and Malaysia experienced declines in the terms of trade somewhat later than did Korea, although the average impact of external shocks appeared to be smaller, especially in Malaysia. In contrast, Singapore appears to have benefited from external shocks over the period, as improvements in the terms of trade, magnified by the openness of Singapore's economy, had a very strong impact, 6.6 percentage points a year. Moreover, the increase in Singapore's estimated interest burden was relatively small (about 0.2 percentage points a year). Although the reasons Singapore's terms of trade improved over this period are not clear from the aggregate data, 18 it appears that Singapore was less vulnerable to external interest rate shocks because historically Federal Reserve Bank of San Francisco very large current account deficits were financed by private direct investment rather than external borrowing. This meant that any capital losses from a rise in world interest rates would be borne by foreign investors, rather than by Singapore residents. 19 Given the estimated effects of the external shocks during the period 1978-82, the earlier discussion suggests that Korea, Thailand, and to a lesser degree, Malaysia should have adjusted by reducing current account deficits and allowing their real exchange rates to depreciate. Moreover, the adjustment should have been more pronounced in Korea, which suffered the largest external shocks, and relatively moderate in Malaysia. In contrast, Singapore appears to have benefited from the external environment 33 over this period, so real exchange rate appreciation and domestic demand expansion would be consistent with equilibrium. Details of the actual pattern of adjustment and economic performance in these Asian economies are provided in Table 2. The table reviews trends in the current account (as an indicator of relative domestic demand stimulus), the trade-weighted real and nominal exchange rates, 20 inflation, and real GNP. To evaluate the possible role of debt-management considerations in influencing policy responses, the debt-to-GNP ratio also is included. In the case of Korea, current account deficits quadrupled to nearly nine percent of GNP in 1980, and the real exchange rate also appreciated over this period. Korea's 34 expansionary policies contributed to rising inflation, while failing to prevent a slowdown in GNP growth. Real exchange rate appreciation exacerbated adverse external and domestic shocks, culminating in a recession in 1980. Efforts to correct the economic difficulties of Korea began in earnest in 1980-82, when Korea adjusted in textbook fashion. Domestic demand restraint produced a more than 50 percent drop in current account deficits as a proportion of GNP, and real exchange rate appreciation virtually ceased as inflation dropped sharply. As the rate of real exchange rate appreciation slowed, GNP growth rebounded (from a rate of decline of three percent in 1980 to increasesof7.4and 5.7 percent in 1981 and 1982, respectively), in spite of slower growth in industrial economies anclcontractionary domestic demand policies in Korea.21 The rate of increase in the debt-to-GNP ratio slowed after 1980,although at 54 percent, it was still above the average for the four Asian economies. It is worth highlighting that contractionary demand management policies in Korea were consistent with rapid growth and, as suggested by the discussion in the preceding section, appeared to be crucial in ensuring the effectiveness of nominal exchange rate depreciation. Korea's nominal exchange rate depreciated sharply in 1980 and 1981,but this was not fully reflected in real exchange rate depreciation, due to high inflation associated partly with past expansionary domestic demand policies. The appreciation of Korea's real exchange rate was interrupted only after contractionary demand management policies took effect and inflation fell. 22 In Thailand, as in Korea, sharp increases in the current account deficit were reversed, and real exchange rate appreciation slowed late in the 1978-82 period. In contrast to Korea, however, Thailand's nominal exchange rate drifted upward over the period due to its link to an appreciating U. S. dollar.P While nominal appreciation apparently led to more moderate inflation in Thailand, it may also have contributed to the declining (and somewhat erratic) trend in Thailand's GNP growth over the 1978-82 period. In particular, Thailand's GNP growth fell significantly below the average for the three other Asian economies in 1982, because of the contractionary combination of declining terms of trade, domestic demand restraint (as reflected in the decline in the current account deficit) and nominal exchange rate appreciation. Over the period, Thailand's debt-to-GNP ratio rose 14 points to 35 percent. In Malaysia. and Singapore, current account deficits increased and the real exchange rate appreciated strongly, and, in contrast to Korea and Thailand, there was no significant reversal in these trends over the period. To an even greater degree than in Thailand, the appreciation in Economic Review / Winter 1990 Malaysia's and Singapore's real exchange rate was largely attributable to rapid nominal exchange rate appreciation. Given the relatively benignexternal environment facing bothMalaysia and Singapore, the combination of expansionarydomestic demand and nominal exchange rate appreciation appears to have been quite effective; inflation was relatively moderate in botheconomies over the entire period,while real GNP growth rates were consistently high. •Malaysia's expansionary policies were associated with largeincreases in its external debt over the period, from 21percent to 32 percent of GNP. In contrast, because Singapore's largecurrentaccount deficits were toa significant extentfinanced by foreign direct investment, Singapore's debt-to-GNP ratio did not increase over the period. Our review indicates that Korea and Thailand, which were more adversely affected by external shocks, experienced slower real exchange rate appreciation anda reductionin currentaccount deficits latein theperiod 1978-82. On the other hand, Malaysia and Singapore, which were lessadversely affected or benefited from external developments, andwere less indebted, experienced sustained real exchange rate appreciation and increases in current account deficits. The pattern of adjustment is thus roughly consistent withequilibrium as described by the model and may partlyexplain the impressive economic performance of the fourAsian economies in 1978-82.24 Nevertheless, the adjustment over this period in some cases had adverse effects. Thailand's policy of linking its currency to an appreciating U.S. dollar was probably too deflationary, while in Malaysia and Singapore, the steep realexchange rateappreciation posedtheriskofa contraction when domestic demand stimulus ended.P Furthermore, in Korea, Thailand, and Malaysia, external debt grew significantly over the period (in spite of eventual reductions in current account deficits in the first two economies), increasing the importance of debt managementconsiderations in future adjustment. tively, thedecline in the interest burden averaged 0.5 and 0.6 percent annually, while in Singapore, the annual decline averaged only 0.2 percent. Table 3 also shows thatin this period, Korea, Thailand, and Malaysia on average benefited from improvements in theterms of trade, although only in Korea was there a consistentimprovement. In Thailand and Malaysia, the terms oftrade declined up to about 1985, before rising sharply in1985-87. In the case of Singapore, the terms of tradedecIined overmost.of the period. 'rhemodelsuggests thatin1983-87, declines in interest rates andimprovements in the terms of trade would be consistent with larger current account deficits and real exchange rate appreciation in Korea, and somewhat later, in Thailand and Malaysia. On the other hand, the equilit>riumresponsefor Singapore would be a reduction in currentaccount deficits and real exchange rate depreciation.Table A suggests .that.in all Asian economies, the actualpathsofreal exchange rates and of current account balances over this period differed from the equilibrium paths predicted by the model. In the cases of Korea and Thailand, current account deficits fell sharply over the 1983-87 period, turning to 1983-87 Theperiod 1983-87reversed manyoftheshocks experienced in 1978-82. As a result, Korea, Thailand, and Malaysia, which had been adversely affected by external shocks in the earlierperiod, benefited from changes in the external environment in 1983-87. In contrast, Singapore, which had been favored in 1978-82, was adversely affected by external shocks. Table 3 shows that the decline in world rates of interest was associated with an average annual decline in the estimated interest burdenof Korea of about 0.7 percent of GNPfrom 1983 to 1987. InThailand andMalaysia, respec- Federal Reserve Bankof San Francisco 35 large surpluses in Korea by the end of the period. At the same time, there was largely uninterrupted, and accelerating, real exchange rate depreciation in both economies, reinforced by sharp nominal exchange rate depreciation starting in 1984-85. Partly as a result of the domestic demand restraint reflected in rising current account balances, inflation rates were moderate. However, inflation rose after 1985 and recent reports of accelerating wage demands in Korea and of supply-side bottlenecks in Thailand, as well as the sharp acceleration in growth in both economies, suggest that the inflation rates in Table 4 may understate underlying inflationary pressures. Rising current account balances also contributed to a decline in the external debt-to-GNP ratio from 53 percent to 34 percent 36 inKorea, and moderated the rise in the debt-to-GNP ratio in Thailand {the latter rose from 35 to 44 percent). Adjustment in Malaysia and Singapore in 1983-87 may be divided into two distinct phases. In the first phase, from 1983to early 1985, current account deficits fell sharply as a result of reductions in investment spending to apparently more sustainable levels. 26 At the same time real exchange rates continued to appreciate or at the very least, did not. fall, partly because nominal exchange rates were still appreciating. These contractionary policies, during a period when Singapore in particular experienced declines in terms of trade, produced strong reductions in inflation and growth, culminating in recessions in both economies in 1985. In the second phase, from 1985to 1987, current account deficits in Singapore and Malaysia turned to surpluses, and real. exchange rates depreciated sharply, due to strong nominal exchange rate depreciation. This was associated with a recovery in growth and a gradual increase in inflation in both economies. In spite of the turnaround in current account balances, Malaysia's external debt-toGNP ratio rose from 42 to 65 percent over the period 1983-87. The rapid growth in external debt in Malaysia was a major source of concern for domestic policy makers and external creditors, and increases in current balances reflected efforts to limit this growth. To sum up, in spite of recessions in Malaysia and Singapore that were quickly reversed, economic performance in all four Asian economies in 1983-87 was again better than the average for all developing countries as a whole. Nevertheless, the earlier discussion suggests that the real exchange rate depreciation and the associated tendency toward current account surpluses may have been carried too far, given the decline in world interest rates and the improvement in the terms of trade in most of these economies. In terms of the model, exchange rate undervaluation is suggested by large current account surpluses in Korea and Malaysia, and by the indications of excess demand in Korea and Thailand cited earlier. It could also be argued that current account surpluses indicate that Singapore exceeded the depreciation required to respond to adverse movements in terms of trade and to correct an apparent currency overvaluation. In addition to fostering resentment abroad, current account surpluses suggest that opportunities for consumption and investment in these rapidly growing economies are not being fully exploited. One possible justification for current account surpluses inAsian economies is external debt management. Current surpluses or declining current deficits have significantly EconomicReview / Winter 1990 reduced the external debt of Korea, while slowing the growth of external debt in Thailand, and to a lesser degree, Malaysia, Prudent debt management permitted the Asian region to escape largely unscathed from the debt crisis of the early 1980s and reduces the vulnerability of these economies to external shocks in the future, Nevertheless, debt-to-GNP ratios are low in the four Asian economies in comparison to other developing countries, and debtmanagement does not appear to be a consideration in the case of Singapore. Furthermore, given the decline in world rates of interest, it can be argued that Korea in particular might have earned a higher return by stepping up investment spending rather than reducing its gross external debt tnrouen current account surpluses in 1985-87, Even if it were desirable to prevent further increases in debt-to-GNP ratios in Asian economies, it can be argued that current account surpluses (and the sacrifice of present investment and consumption opportunities) may not be required to accomplish this, The experience of Singapore suggests thatforeign private direct investment can finance increased consumption and investment with little or no accumulation of external debt.s? Given the outstanding economic performance of the region, debt reduction could also be achieved easily by converting foreign debt into foreign equity, In conclusion, the ability of the four economies of the region to prevent real exchange rate overvaluation in the face of adverse external shocks appears to have contributed to their very successful economic performal1CeiI11978--87, However, the discussion provides some support for the view that real exchange rates may have become undervalued in some Asian economies, particularly after 1985, and that opportunities for increasing inve!;tmentand consumption in some of these economies may have been missed, NOTES 1. Notably Latin America and Africa. For discussions of economic performance and policy responses to external shocks, see Dornbusch (1985), Balassa (1986), Balassa and Williamson (1987), Edwards (1988), Khan (1986), and Sachs and Sundberg (1988). 2. Because of disinflation, however, real rates in the 1980s still were higher than they were in the mid-1970s. 3. For a rigorous discussion of this type of optimization problem with a focus on private consumption demand, see Ostry (1988). For an example of this type of analysis applied to Asian economies, with a discussion of investment demand, see Alesina (1987). Sen and Turnovsky (1989) also analyze investment demand in an intertemporal framework. 4. If capital markets are imperfect, so that the private sector cannot borrow abroad while the government can, government policy can be designed to create a spending and borrowing path that is consistent with intertemporal utility maximization. For a discussion, see Alesina (1987). On the other hand, government policy often ignores such intertemporal concerns, leading to suboptimal spending paths. 5. For convenience, the definition of the real exchange rate that follows is the reciprocal of the standard definition. 6. A proxy for the traded goods price is described in note 20. See Dornbusch (1980). 8. To correspond to an equilibrium, equation (10) must satisfy certain constraints. In particular, in the absence of government intervention, we know that De is given in equation (3), S is given by endowments and that the Federal Reserve Bank of San Francisco elasticities of supply of each good are given by technology. The price elasticities of demand between nontradable, importable, and exportable goods must then be such that, given the overall demand and supply, the equilibrium real exchange rate that clears the non-traded goods market is also consistent with achieving an equilibrium current account and rate of external borrowing. 9. It is implicitly assumed here that the reduction in permanent income in the long-run reduces the current account deficit by more than the increased interest payments initially increase the current account deficit. The reason is that the reduction in permanent income, as well as the higher cost of current consumption reflected in the increase in world interest rates, reduces the equilibrium level of external borrowing. 10. It is implicitly assumed that the contraction in demand more than fully offsets the tendency for the higher import price to decrease the current account balance. The prediction that a worsening of the terms of trade tends to increase the equilibrium current account balance would contradict the so-called Laursen-Metzler effect. In Laursen and Metzler's 1950 framework, adverse movements in the terms of trade cause a rise in the export value of expenditure, a decline in saving, and a current account deficit. Obstfeld (1982) notes that this prediction results from Laursen and Metzler's assumption that individuals reduce their saving when experiencing a decline in real income, and shows that this assumption may be invalid if individuals maximize intertemporal utility, and the rate of discount is increasing in utility. (The latter condition is required for stability in the stationary state. See Svensson and Razin, 1983.) In an expanded optimizing framework 37 which includes investment, Sen and Turnovsky (1989) cite conditions under which an adverse terms of trade shock, leads to a decline in the steady-state equilibrium capital stock, so that current investment spending and the current account deficit fall, contradicting the Laursen-Metzler hypothesis, A related literature focuses on the proposition adopted in the text that adverse movements in the terms of trade, due, say, to a rise in import prices, are associated with a depreciation of the equilibrium real exchange rate, Edwards and Van Wijnbergen (1988) note that this standard proposition may be questioned, because it implies that income effects (which lead to a reduction in the demand for non-traded goods and a tendency toward real exchange rate depreciation) dominate substitution effects (which lead to an increase in the demand for non-traded goods due to the rise in import prices, and a corresponding tendency toward real exchange rate appreciation), Although they concede that it is quite possible for income effects to dominate substitution effects, they note that such a result is generally considered an anomaly (presumably because it contradicts traditional assumptions made in the literature), One way of addressing this objection is to note that in many small open economies imported goods are often not good substitutes for non-traded goods, so that income effects may dominate substitution effects, (In certain important cases, such as oil imports, imports and non-traded goods may be complements.) In the discussion in the text, the ambiguity between income and substitution effects does not arise because it is implicitly assumed that changes in relative prices that are not accompanied by shifts in the sectoral demand or supply schedules lead to excess demand or supply in the short-run but do not affect the long-run equilibrium real exchange rate, 11, A rise in (oil) import prices will tend to shift the supply schedule in Figure 4 to the left, producing a real exchange rate appreciation and an associated increase in the price of non-traded goods, In Figure 5, the import supply schedule also will shift to the left, increasing the total level of imports, Similarly, the level of exports will fall, as will the current account balance, 12, For the terms of trade to improve, the expansion in world economic activity must increase export prices more than import prices, This will occur, for example, if the world supply of exportable goods is less elastic than the supply of importable goods, Thus we might expect commodity exporters to be more affected by fluctuations in the growth of industrial countries, 13, See Lipschitz (1979). This effect does not depend on whether a country pegs its currency to the dollar or the yen. 14, Alternatively, a mix of commercial policies may be used, See Edwards (1988). Note that in theoretical discussions it is generally assumed that a nominal exchange rate depreciation has adverse effects on the terms of trade, which may further reduce demand. This assumes that importers and exporters pass-through the full impact 38 of exchange rate changes to their respective markets, so that import prices rise and export prices remain unchanged in domestic currency. However, the terms of trade may not worsen in response to a nominal depreciationifexporters price-to-market (which some argue is the case for Asian exporters), as export prices in domestic currency can then rise with a nominal depreciation, without adversely affecting export volume. Such nominal eXChange rate effects do not change the direction of equilibrium adjustment and should not prevent the attainment of a new equilibrium. t5..Fordefinitions of real exchange rate misalignment, see Edwards (1988, 1989). 16. Note that the estimate does not reflect actual interest payments, which in some cases were deferred into the future. Furthermore, the use of gross, rather than net, external debt statistics tends to overstate the debt burden ofthe Asian economies. This is likely to be true of Singapore for most of the period, and for Korea after 1985, Nevertheless, the figure presented in the table gives a fair picture of the burden faced by these economies during a period when world rates were rising. The extent to which a change in world interest rates affects demand and the real exchange rate also depends on the gap between the marginal product of capital and the cost of funds produced by the change in interest rates. Due to lack of available data, no attempt was made to estimate this effect. 17. The estimates show the impact of terms-of-trade changes in all Asian economies in 1978-82, except for Singapore, where the data correspond to 1979-1982, For a more precise specification of the effect of changes in the terms of trade on disposable income, as a function of the share of imports, see Ostry (1988). 18. One possibility is that the sharp rise in crude oil prices starting in 1979 contributed to the improvement in the terms of trade, since Singapore was a net exporter of petroleum products. The assumption is that Singapore's profit margins from processing petroleum products increased with the rise in oil prices. However, the relationship between oil prices and profit margins is quite unstable, Furthermore, a case can be made that profit margins may rise more strongly when crude oil prices fall, if refiners do not fully pass on the savings to consumers. 19. The analysis in Section II assumed that external deficits are financed through borrowing, so that changes in interest rates affect the permanent income of domestic residents. Since Singapore financed its current account deficits by foreign direct investment, the rise in world interest rates did not automatically lead to an increase in the flow of payments to owners of foreign capital (as it would to foreign creditors). Instead, through arbitrage, the market value of the Singapore capital owned by foreigners may have fallen. In contrast to the case where current account deficits are financed by external borrowing, the rise in world interest rates meant that foreigners, rather than domestic residents, took a loss. Although the rise in interest rates still would tend to lower equilibrium Economic Review / Winter 1990 investment spending in Singapore, theeffectondomestic permanent income and consumption would be smaller. 20. The weighted average CPls of major industrial countries, adjusted for bilateral exchange rates, were taken to represent traded goods prices (or factors that would heavily influence such prices), while the domestic CPI was taken to represent non-traded goods prices for each Asian economy. In an effort to focus on those economies whose prices and currencies are most likely to influence theworld prices of traded goods,theU.S., Japanese, and EC CPols, adjustedfor bilateral exchange rates, were used to denve a measure of traded goods prices for the Asian economies (the German deutschemark served asa proxy forthe European currencies). Weights were based on the 1980 bilateral trade of each Asian economy with the U.S., Japan, and the EC. . ~ith the exception of Malaysia, these indices give ~Imllar results to broader real exchange rate indices that Include the currencies of a larger number of industrial countrvtradino partners as well as newly-industrializing economies. In the case of Malaysia, the fact that Singapore is one of its larger trading partners affects the r~sUlt~. Both the Singapore dollar and the Malaysian Ringgit strongly appreciated on a trade-weighted basis against the currencies of the major industrial countries in 1978-80. However, since the Singapore dollar appreciated by more, Malaysia's real exchange rate appears to appreciate more strongly in the three-currency basket, because the Singapore dollar is excluded. 21. Two points are worth making here. First, the sharp contraction in output and the sharp rebound thatfollowed were p~rtly the result ofthe1980 contraction in agricultural output In Korea, and are thus not entirely attributable to economic policies. Second, the decline in oil prices after 1980 contributed to disinflation in the more indebted economies. Nevertheless, this was notthe onlyfactor. The rapid increases in inflation in othereconomies after 1980 (notably Latin America) suggests that declining oil prices are not sufficientto guarantee a decline in inflation. ~2. Inthe absence of more detailed empirical analysis, it IS difficult to determine the "equilibrium" real exchange Federal Reserve Bank of San Francisco rate (for a description of an attempt, see Edwards, 1988) and to make precise statements about whether real exchange rate adjustment was sufficient to correct overvaluation. One reason is that, aside from the magnitude of the shocks, the.extent of equilibrium real exchange rate adjustment to contractionary external shocks dependson additional factors that are not easily measured, such as the price elasticities of demand and supply between traded and non-traded goods sectors. The smaller these elasticities, the larger is the required depreciation. 23.. Thepeg to a rising dollar nullified the impact of a 10 percent depreciation of the Thai Baht against the U.S. dollar in 1981. 24. Although Malaysia responded to adverse shocks with expansionary policies, its relatively low external debt at the time suggests that there was scope for increasing domestic spending. See discussion in Section II. 25. The composition of domestic demand also posed potential problems. InbothSingapore and Malaysia, stimulusto domesticdemand was reflected in increases ofthe investment to GNP ratio of up to 10 percentage points over the. period. It is not clear whether it was appropriate to stimulate investment demand sharply at a time when w~)fld rates of interest rates were rising. In particular, such stimulus may have prevented these two economies from increasing investment to exploitthe decline in world interest rates in 1983-87. 26. In Malaysia, a desire to limit increases in external debt appears to have been an additional motivation. See below. 27. There is some evidence thatthisstrategy isnowbeing p.ursued by Thailand and, to a lesser degree, by Malaysia. In 1988 Thailand's current account deficit increased sharply, while in Malaysia the current surplus declined. A five-fold increase in foreign direct investment partly financed Thailand's growing current account deficit, while foreign direct investment in Malaysia increased 54 percent. In the case of Korea, foreign direct investment increased 72 percent in 1988, but this was accompanied by an increase in the current account surplus. 39 Appendix Variable Definitions and Data Sources Variables i* M MPK N PI P*M P*x P*T PM PN Px PXPM S totaldemand equilibrium demand by private sector in the absence of govemmentintervention, basedon intertemporalutilitymaximization (exogenous}f.lggregate demand resultingfromgovernment intervention thetotal demand for eachgood,i""N,M, and X the shareof each good in aggregate demand real exchange rate nominal exchange rate (unitsof domestic currency per unit of foreign currency), set by the government world rate of interest, set abroad importable goods marginal product of capital non-tradablegoods pennarient income = the discounted presentvalue of disposable income (exogenous) foreign currency priceof importable goods (exogenous) foreign currency priceof exportable goods (exogenous) weighted average foreign currency price of importable and exportable goods domestic currency priceof importable goods = EP*M the price of non-tradable goods, set endogenously domestic currency price of exportable goods = EP*x commodity termsof trade the (exogenous) total supply of goods in the economy theshare of each good in aggregate supply, i = N, M, and X the total supply of each good exportable goods Data sources. Three-month LIBOR, exportand importunit values, nominal exchange rates, current account, GNP and CPI series are from the IMF, International Financial Statistics, various issues. Trade weights were constructed using IMF Direction of Trade statistics. Debt/GNP ratios are from World Bank, World Debt Tables 1988-89. 40 Economic Review / Winter 1990 REFERENCES Aghevli, Bijan B. "Experiences of Asian Countries with Various ExchangeRatePolicies," in JohnWilliamson, ed., Exchange Rate Rules: The Theory, Performance and Prospects of the Crawling Peg. New York: St. Martin's. 1981, pp 298-318. Alesina, Alberto. "Optimal Borrowing Policies for Developing Countries: the Cases of Korea, the Philippines and Thailand, 1965-1983," Working Papers WP/87/41 , International Monetary Fund. June 10, 1987. Balassa, Bela. "Policy Responses to Exogenous Shocks in Developing Countries," American Economic Review. Vol 76, NO.2. May 1986. Balassa, Bela and JohnWilliamson. Adjusting toSuccess: Balance of Payments in the Far East Asian NICs. Washington, D.C.: Institute for International Economics.1987. Dornbusch, Rudiger. Open Economy Macroeconomics. New York: Basic Books. 1980. ____ . "Policy and Performance Links between LDC Debtors and Industrial Nations," Brookings Papers on Economic Activity, 2:1985. Edwards, Sebastian. "Real Exchange Rates in the Developing Countries: Concepts and Measurement," NBER Working Paper, No. 2950, April 1989. ____ . "Exchange Rate Misalignment in Developing Countries," Occasional Paper Number2/New Series. Baltimore and London: Johns Hopkins, 1988. Edwards, Sebastian and Sweder Van Wijnbergen. "Tariffs, the Real Exchange Rate and the Terms of Trade: On Two Popular Propositions in International Economics," Oxford Economic Papers. Vol. 39, 1987, pp 458-464. Khan, Mohsin, S. "Developing Country Exchange Rate Policy Responses to Exogenous Shocks," American Economic Review. Vol 76, No.2, May1986, pp 85-87. Federal Reserve Bankof San Francisco Khan, Mohsin S. and Malcolm D. Knight. "Determinants of Current Account Balances of Non-Oil Developing Countries in the 1970s: An Empirical Analysis," Staff Papers, 30, International Monetary Fund. Washington, December 1983, pp. 819-42. Laursen, S. and L. A. Metzler(1950). "Flexible Exchange Rates and the Theory of Employment," Review of Economics and Statistics. 32, 281-99. Lipschitz, Leslie. "Exchange Rate Policy for a Small Developing Country and the Selectionof an Appropriate Standard, StaffPapers, International Monetary Fund. Washington, September 1979, pp. 423-449. Obstfeld, Maurice. "Aggregate Spending and the Terms of Trade: Is There a Laursen-Metzler Effect?," Quarterly Journal of Economics. 97, 1982. pp. 251-70. Ostry, Jonathan D. "The BalanceofTrade, Terms ofTrade, and Real Exchange Rate: An Intertemporal Optimizing Framework," StaffPapers, International Monetary Fund. Washington, 1988, pp 541-573. Sachs, Jeffrey D. and Mark W. Sundberg. "International Payments Imbalances of the East Asian Developing Economies," in Norman S. Fieleke, ed., International Payments Imbalances in the 1980s. Proceedings of a Conference Sponsored by the Federal Reserve Bank of Boston, October 1988. Sen, Partha and Stephen Turnovsky. "Deterioration of the Terms of Trade and Capital Accumulation: A Reexamination of the Laursen-Metzler Effect," Journal of International Economics. 26, No. 3/4, 1989. pp. 227-250. Svensson, Lars E.O. and Assaf Razim. "The Terms of Trade and the Current Account: The HarbergerLaursen-Metzler Effect," Journalof Political Economy. Vol. 91, No.1, 1983. pp. 97-125. 41 42 Economic Review / Winter 1990 The Public Policy Implications of State Laws Pertaining to Automated Teller Machines Elizabeth S. Laderman Economist, Federal Reserve Bank of San Francisco. I would like to thank Rachel Long and Deborah Martin for research assistance, and the members of the editorial committee, Randall Pozdena, Gary Zimmerman, and Fred Furlong, for many helpful comments. In the early 1970s, as automated teller machines (ATMs) were beginning to grow in popularity, some states instituted mandatory sharing laws, whereby ATM-owning banks were required to share their ATMs with any other bank that wished to do so. It was perceived that ATM technology was subject to significant economies of scale, and it was thought that these laws would increase small bank customers' access to ATM services. Empirical tests in this paper reject the hypothesis that mandatory sharing increases the level of ATM services for small bank customers and show that mandatory sharing may in some cases decrease the level of ATM services for all bank customers. It also is shown that, under certain conditions, branching restrictions may have negative effects on the supply ofATM services. Federal Reserve Bank of San Francisco In the early 1970s the automated teller machine (ATM) was introduced, enabling people to perform banking transactions such as cash withdrawals, deposits, balance inquiries, and interaccount transfers without the aid of a human teller. By the mid-1970s, banks had started sharing ATMs, allowing other banks' customers access to their machines.' Beginning at this time, too, certain states instituted mandatory sharing laws, which required that any ATM-owning institution share its off-premise machines for a "reasonable fee" with any other financial institution in the state that wanted to share. The intent of such laws was to ensure that customers of small banks would have access to ATMs, despite ATM systems being subject to significant economies of scale. In this paper, I will investigate whether there is any empirical evidence that mandatory sharing laws have been successful in this regard. I will begin with a general discussion of the market for ATM services, economies of scale in ATM networks, and the legal environment surrounding ATMs before proceeding to the empirical analysis. 43 I. The Demand for and Supply of ATM Services The Demand for ATM Services According to one estimate, 137.7 million ATM cards were outstanding in July 1988. 2 Estimates of the percentage of households thatownat least one ATM card run from 45 percent" to 54 percent. 4 The group that reports the 54 percent figure states that, as a comparison, 76 percent of households have at least one credit card." ATM use has been increasing over time, and now, forthe first time, more than 50 percent of all cardholders use their cards at least once a month.6 Whether a customerof a particular bank will choose to obtain transactions services from an ATM or a teller partially will depend on the direct charges the customer faces when using the two procedures. For example, a customer may face a choice between paying 25 cents to use the bank's ATM or paying 10 cents to cash a check through a teller.7 In addition to the levels of direct ATM and teller use charges, the convenience of using an ATM versus a teller will influence a bank customer's decision. Whether a particular customer will choose an ATM or a teller will depend on such factors as the value of the time and the effort that the customerneeds to contributein order to get to and use an ATM or a teller. The decision also will depend on the customer's attitudes and tastes regarding, for example, computers versus human interaction. Apparently, age and income are determining factors in the choice between ATMs and human tellers. A typical ATMuser is under 40 and uses an ATM three to fourtimes a month, on average. Very heavy ATM users, those who use the machines as often as three times a week or more are apt to be between 18 and 24 years of age." A 1986 survey revealed that the percentage of families with less than $10,000 in yearly income that owned ATM access cards was 32 percent. This percentage increased with income,up to 60 percentforthoseearning$50,000or more a year. However, the survey also revealed that those families in the lowest and highest income categories were the most frequent users of ATMs, at least for the purpose of withdrawing cash.? The Supply of ATM Services ATM industryobservers have cited at leasttworeasonsa bank might chooseto offerATM servicesto its customers. First, banks mayintroduce ATMs to increasemarketshare. In a market where ATMs are not prevalent, or ATM networks are not extensive, a bank may be able to differentiate its services from those of other banks and thereby 44 attract new customers. For example, one East Coast bank attributes the increase in its statewide share of checking andNOW accounts from16percentin 1984 to 19 percentin 1988 to its extensive ATM network. An executive vice president of the bank claims that the ATM network is one of the two factors peoplemention most often as reasons for banking with that bank. 10 This bank's experience raises the possibility that some banks may adopt ATMs even if the per transaction cost is higherwith ATMs than with tellers. ATMs raise the value of transactions servicesby, forinstance, lowering the" time tax" that customers face whentheycarry out bank transactions. If this attracts new customers, it can lead to economies of scale in some other aspect of bank operations besidestransactions services. However, giventhe ubiquity of ATMs, it seems unlikely that this is the primary means by which ATMs increase profitability for most banks. A second, and more important, reason banks might choose to install ATMs is that, above a certain level of operations, the cost of a single transaction performed at an ATM potentially is less than the cost of a transaction conductedat a teller window. II This is because ATMs are capableof handlingmoretransactions per unitof time than are tellers. However, Allen Bergerhas found that the costper dollar withdrawn is significantly higher for ATM withdrawals than for withdrawals conducted at a teller window. 12 This is because ATMs are sufficiently more convenient than te~lers that customers tend to make more frequent trips and withdraw smaller amounts each time than they would if they had to use tellers. Despite this, ATMs still are attractive to banks as long as the price per transaction is lowerand banks are able to cover the cost of transactions by charging transactions fees. The per-transaction cost of ATMs apparently is subject to significant economies of scale, due to the relatively high fixed cost of installing and operating an ATM system. Purchasing and installing an ATM costs about $25,000 to $30,000. Moreover, armored car services and data processing can add $200,000 a yearto theoperating cost of an ATM system.P In light of these high fixed costs, Walker (1980) estimated that economies of scale associated with ATM transactions in a network of ATMs are realized up to at least 43,600 transactions per month per ATM.14 This number should be interpreted with some caution, however, because Walker's cost data were from early 1974, a time when sharing of ATMs was not very prevalent. Therefore, this may be a better measure of Economic Review / Winter 1990 economies of scale for proprietary ATM networks that are used by only one bank's customers than for shared ATM systems. Shared ATM Networks Some have claimed that the economies of scale in ATM systems help to explain the rise of shared ATM networks. A shared network is a collection of ATMs that are owned by different banks but can be used by any customer of any bank in the network. IS By spreading the fixed cost associated with ATMs over transactions initiated by customers of many different banks, a shared network can take advantage of economies of scale. Shared networks also may be attractive because they increase the convenience of ATMs by enabling a given bank's customers to carry out banking transactions over a wider geographic area than would be possible with a proprietary network." This factor may be particularly attractive to banks in states that place geographic restrictions on branching and the placement of ATMs. Statistics show that ATMs more often than not are shared, and that customers take advantage of shared machines. In 1987, 75 percent of the banks that operated ATMs shared them with other institutions,'? and in 1988, 90 percent of the ATM terminals in the U. S. were shared with at least one other institution. IS A 1986 survey found that about 28 percent of families with ATM cards used another institution's ATMs.19 ATM sharing also has been growing over time, as shown in Table 1.20 It is important to note, moreover, that banks form shared Federal Reserve Bank of San Francisco networks even in states that do not require sharing. In 1983, 23 states had mandatory sharing laws, yet every state had banks or other financial institutions that belonged to shared networks. Banks that participate in a shared network pay fees to the network owners to cover the various costs of the network's operation. These costs include the costs of transferring "foreign" transactions, those transactions that are carried out by one bank's customers on another bank's ATMs. Such transactions are commonly sent through a central "switch," in which case the ATM-owning institution pays a "switch fee" to the network. In addition, "interchange fees" are paid by the card-issuing bank to the ATMowning bank. The fees charged by a given network depend on two countervailing factors. On the one hand, economies of scale associated with high transactions volume should help to keep network fees low. There appears to be some evidence that networks do, in fact, pass on cost savings resulting from economies of scale. For instance, the American Banker recently reported that increased transaction volume at many of the nation's largest regional networks has enabled them to reduce fees to network members. Last year, interchange fees averaged 15 to 20 cents per transaction, but are now running around five to 10 cents." On the other hand, network costs may rise as the number of network members rises, and this could partially offset cost savings from economies of scale in transaction volume. As the number of network members rises, costly telecommunications technology is needed, and negotiat- 45 ing costs and the costs associated with settling accounts among institutions also rise. Given that networks with high transaction volume also may have many members, it is not surprising that some fees apparently do not vary much with transactions volume. As Table 2 shows, the average switch fee across five different volume classes is very close to 20 cents a transaction. These observations suggest that the marginal cost of adding a small bank to a shared network could outweigh the marginal benefit this bank would contribute by way of increased transaction volume. As a result, even with the existence of sharing, small banks are less likely than are large banks to own and operate ATMs. Table 3 shows that, indeed, relatively few small banks own and operate ATMs, and the small banks that do own ATMs own fewer terminals, on average, than do larger banks. 46 1,000 or More 99.0% .8 Economic Review / Winter 1990 II. Mandatory Sharing Laws and ATM Branching Laws The discussion in the preceding section suggests that the economics of shared ATM networks discourages the participation of smaller banks. The perception that smaller banks have more limited access to shared ATM networks may explain why a number of states have adopted mandatory sharing laws. Table 4 shows that as of September 1983, 23 states had instituted some sort of mandatory sharing, whereby banks must share their ATMs with any other in-state financial institution that wishes to do so and is willing to pay a reasonable fee. 22 Many of the mandatory sharing statutes do not specify the level of payments which may "reasonably" be required of banks wishing to join a network. This is an important issue because it has implications not only for whether small Federal Reserve Bankof San Francisco banks join networks, but also for whether the incentives to form networks diminish with mandatory sharing. Where legally imposed sharing requirements exist in other institutional contexts, courts have ruled that new members had to be admitted on the same terms applicable to the preexisting members, and that "open admissions" and "equal treatment" are called for.> Since many mandatory sharing laws predate the widespread formation of shared networks, the open admissions and equal treatment provisions are the most relevant of the three principles. Unfortunately, these provisions have been defined only vaguely by the courts. Nevertheless, they seem to imply that sharing requirements preclude discriminatory fees and fee schedules. 47 Thus, the "reasonable fee" clause in mandatory sharing statutes may prohibit shared ATM networks from imposing surcharges on banks that contribute too few transactions to the network. Coupled with the open admission provision, this pricing approach likely would increase the number of small banks with access to ATM machines. However, the addition of these banks likely would decrease efficiency and increase network fees for all members, large and small, since the marginal cost of these banks' membership would outweigh the marginal benefits they contribute. There is another problem with the "reasonable fee" provision; that is, it is difficult to determine what rate of return on risk taking in shared networks ought to be incorporated into the reasonable fee. Baxter, Cootner, and Scott (1977) argue that regulators are likely to underestimate the degree of risk faced ex ante by network founders, and are thus likely to underestimate the appropriate rate of retum.>' Ex post, the successful networks to which new members will wish to gain access will appear to the regulator not to have faced extraordinary risk, these critics maintain. Therefore, according to this argument, rates of return and compensating fees will be set too low. In expectation of this outcome, banks in mandatory sharing states will be discouraged from forming shared ATM networks. The Justice Department's view on mandatory sharing is consistent with this line of reasoning. The Department argues that mandatory sharing "undercuts in advance any incentive to innovate, creating a 'free rider' problem with respect to initial risk-taking."25 Other observers note that mandatory sharing may introduce an additional free rider problem simply by allowing banks to join in after the initial capital costs have been borne by the original ATM installing bank or network members. 26 Other state laws pertaining to ATMs include those that set the geographic limits for off-premise ATMs within states. A list of these statutes can be found in Table 4. Note that all of the states that have constraints on ATM placement also constrain the geographic expansion of traditional branches. However, not all unit banking or limited branching states limit ATM placement. III. A Model of the Market for ATM Transactions As noted in the preceding section, mandatory sharing may increase small bank customers' access to ATM services, but also may make sharing more costly for all network participants, thereby decreasing the level of ATM services for all customers. To empirically test whether mandatory sharing laws have increased the supply of ATM services to small bank customers, I develop the following model of the supply of ATM transactions, which includes mandatory sharing as an explanatory variable. The supply of ATM transactions will depend on the cost of ATM transactions and the price of ATM transactions. It may also depend on the banking market structure in the sense that a less competitive banking market will yield a lower supply of ATM transactions. The aggregate supply of ATM transactions thus is given by: SUPPLY = s(BANKS, COST OF ATM TRANSACTIONS, STRUCTURE, PRICE OF ATM TRANSACTIONS), (1) where s is a continuous function, BANKS is the number of banks, and STRUCTURE indicates the bank market structure. The aggregate supply of ATM transactions depends positively on the number of banks and negatively on the cost to each of those banks of providing ATM transactions. It depends positively on the price of ATM transactions. 48 The cost of ATM transactions, in turn, is given by: COST = c(BANK SIZE, ATM LAWS, BRANCHING LAWS), (2) where BANK SIZE is the size of the bank in terms of number of depositors, ATM LAWS are laws governing ATMs, including mandatory sharing laws, and BRANCHING LAWS are laws governing traditional branching. 27 As bank size decreases, the cost of ATM transactions rises. For any given bank, however, this may be modified by the existence of mandatory sharing laws, other ATM laws or branching laws. I will test the hypothesis that mandatory sharing mitigates the negative effects of a decrease in bank size. The possible effects of other ATM laws and of branching laws will be discussed in more detail in the next section. The aggregate demand for ATM transactions should depend on the population, its income, and its age. It should also depend on the price of ATM transactions and on the number of traditional branches and main bank offices available. The aggregate demand for ATM transactions is given by: DEMAND = d(POP, PER CAPITA INCOME, AGE, OFFICES, PRICE OF ATM TRANSACTIONS), (3) Economic Review / Winter 1990 where POP is population, AGE is the mean age of the population and OFFICES is the number of bank offices (main offices plus branches). Aggregate demand will depend positively on population and per capita income and negatively on the mean age. It also will depend negatively on the number of bank offices, since these are a substitute forATMs, and negatively on the price of using an ATM. This model was given a log-linear specification, and the resulting reduced form, derived in the Appendix, is: ATM transactions = Bl + B2*POP + B3*PCINC + B4*BANKS + B5*BRANCHES + B6*MAND + B7*ATMLIM + B8*(MAND)(BANKS) + B9*(ATMLIM)(BANKS) + BlO*(UNIT)(BANKS) + Bll*(LIM)(BANKS) + B12*CONC + B13*UNIT + BI4*LIM + Z (4) where, POP = population PCINC = per capita income, BANKS = number of banks, BRANCHES = number of bank branches, MAND is a binary variable indicating the presence or absence of mandatory sharing, ATMLIM is a binary variable indicating the presence or absence of limitations on "branching by ATM," UNIT is a binary variable indicating the presence or absence of unit banking, LIM is a binary variable indicating the presence or absence of limited branching, CONC = the degree of concentration of the statewide banking market, Z is a normally distributed error term with mean zero, and Bl - B14 are coefficients to be estimated. Data and Regression Specification To determine the effect of mandatory sharing laws, I estimate the reduced form given in equation (4) for a crosssection of 50 states, using two different proxies for the number of ATM transactions. Data on monthly transaction volumes by state are not available. Although data on monthly transaction volumes are available for each network, these data are not useful for measuring the effects of mandatory sharing laws. Sharing laws affect ATM transactions initiated by customers of only the banks within a given state, while transactions on shared networks frequently involve banks that are located outside the state in Federal Reserve Bank of San Francisco question. As a result, I tried two different proxies for transaction volume, the total number of ATM debit cards in each state in 1987,28and the number of ATMs in each state in 1987. 29 The first regression uses the number of ATM cards as the dependent variable. Population, per capita income, the number of banks , and the number of bank branches in 1987 are all included in the regression as explanatory variables.P? Increases in population and per capita income should increase the aggregate demand for ATM cards. Increases in the number of bank branches should decrease demand for ATM cards, since traditional branches are to some extent substitutes for ATMs. Variations in the number of banks, holding population and per capita income constant, should be negatively related to variations in the average size of banks, in terms of number of depositors. It is expected that states with banks that are larger on average, in terms of number of depositors, will have more ATM cards because larger banks are more likely to have ATM programs. Therefore, states with fewer banks, holding all other factors constant, should have more ATM cards. However, a decrease in the number of banks also may decrease the aggregate supply of ATM transactions by decreasing the number of suppliers. (See the Appendix for more detail.) I also include a statewide concentration ratio on the right-hand side of the regression.:" This controls for the competitive effects of bank market structure. It is possible that a less competitive banking market would lead to a lower supply of ATM cards. However, since bank services are a multi-dimensional "good," with many different characteristics, it is not obvious a priori that a decrease in competition would decrease the supply of ATM services or ATM cards in particular. A binary variable for mandatory sharing enters the regression by itself and in an interaction term with the number of banks. The mandatory sharing binary takes a value of one if a state has mandatory sharing between like institutions and takes a value of zero otherwise. 32 Whether there is mandatory sharing between unlike institutions is not considered. The interaction term is the product of the mandatory sharing binary and the natural logarithm of the number of banks. As such, it allows the effects of an increase in the number of banks to be modified by the binary, and it allows the effects of a change in the binary variable from zero to one to be modified by the number of banks. The interaction term is included on theoretical grounds. If mandatory sharing is working, it may modify the depressing effect that an increase in the number of banks, and thus a decrease in 49 their average size in terms of depositors, would haveon the number of ATM cards. A binary variable indicating the presence of ATM branching limitations is also included in the regression, by itself and in an interaction term with the number of banks.P ATM branching limitations should decrease the profitability of an ATM program, and may also exacerbate the effect of a decrease in bank size. For example, if large banks from metropolitan areas are prohibited from placing ATMs in communities with small banks that find ATMs too costly, those communities may have no access to ATMs at all. Unit banking and limited branching laws also may have some negative effects on the number of ATM cards. Studies have shown that barriers to entry in the form of branching restrictions decrease competition in local banking markets. 34 Therefore, unit banking and limited branching binaries are included as indicators of the level of competition in the local market, in addition to the statewide concentration measure. The unit banking and limited branching binaries also appear in interaction terms with the number of banks. There are two reasons for including these interaction terms. First, there is likely to be more dispersion in bank size in a unit or limited branching state than in a statewide branching state, all other factors equal. This is because under statewide branching, banks are freer to seek the most efficient scale of operations, unconstrained by geographic limitations. The greater size dispersion in states with narrower branching provisions may mean that average bank size is a less useful measure of the overall scale of banking operations in the state. Second, if unit banks are relatively small, on average, then they may want to take advantage of the "branching" opportunities ATMs can provide. The data for all of the state law binary variables are as of 1983 and are reported in Table 4. 35 Regression Results with ATM Cards The regression results are presented in Table 5. The population coefficient has the expected positive sign and is highly significant. The coefficient on the number of branches has the expected negative sign, and it is highly significant. The per capita income coefficient is positive, but significant only at the 10 percent level. The coefficient on the number of banks is insignificant. The coefficient on mandatory sharing is positive and significant at the 10 percent level, while the coefficient on the mandatory sharing-bank interaction term is negative and significant at the five percent level. This means that the 50 overall effect of mandatory sharing will be negative whenever the number of banks is sufficiently large to cause the interaction term to outweigh the constant positive effect. To aid interpretation of the mandatory sharing coefficients, Chart 1compares the predicted effect of an increase in the number of banks in a mandatory sharing state with that in a state that does not have mandatory sharing, holding constant the other explanatory variables at their sample means and the other legal variables at zero. The point estimates in Chart 1 show that mandatory sharing is associated with a decrease in the number of ATM cards when there are many banks. 36 This decrease is significant beyond about 270 banks.I? There are nine mandatory sharing states in the sample with at least 270 banks. At 342 banks, the mean number of banks in mandatory sharing states, mandatory sharing reduces the number of ATM cards by about 36 percent. Three possible explanations can be given for the negative effect of mandatory sharing. First, for the reasons Economic Review / Winter 1990 discussed above, mandatory sharing may increase the cost of ATM services. Thus, the growing strength of the negative effect as the number of banks increases may be because of member-related network costs, which would be higherthe smaller are the banks in the state. Second, mandatory sharing may mostly encourage twowaysharing, thereby eliminating the need for a customer to hold more than one institution's card in order to use more than one institution's ATMs. Thus, mandatory sharing simply may discourage customers from establishing secondarytransaction accounts and obtaining multiple cards, which they otherwise would do. This argument assumes that the major reason bank customers hold more than one transaction account is to obtain relatively small amounts of cash-at multiple locations. Available evidence indicates, however, the secondary checking accounts are typically used for large expenditures that constitute a significant proportionofa family's spending-" This argument also implies that an increase in ATM sharing would significantly reduce the use of secondary checking accounts. However, between 1984 and 1986 the percent of ATMs shared increased from 46 percent to 76 percent, and the proportion of families with secondary checking accounts increased, from 20 to 22 percent. 39 Third, the existence of mandatory sharing laws may not cause a reduction in the number of ATM cards, but may instead be indicative of the existence of other factors, not included in the regression, that inhibit the establishment and growth of ATM systems. States with mandatory sharing laws may have passed them because they knew their banks would have difficulty supplying ATM services. The negative coefficient on the mandatory sharing-bank interaction term may merely be an indication that mandatory sharing did not succeed in overcoming whatever other forces were depressing the level of ATM services. A likely left-out factor is some aspect of bank size that has not been considered. Table 6 shows the results of a regression of mandatory sharing on a constant and the number of banks and population. The significant positive coefficient on the number of banks and the significant negative coefficient on population indicate a positive correlation between mandatory sharing laws and small banks.v' Chart 1 Cards (in thousands) The Effect of Mandatory Sharing on the Number of ATM Cards 12,000 10,000 Mandatory Sharing 8,000 6,000 4,000 2,000 0'---"---"-------..1-_ _...1..-_ _..1-_---' 20 54.6 148.4 403.4 1096.6 Note: Calculations assume mean values for demographic and structural variables. Federal Reserve Bank of San Francisco Banks (Log Scale) 51 I hadpresumably controlled forbanksize, but it may be thatthe numberof.banksandpopulation donotadequately control for therelevant aspects of bank size. Forinstance, although thenum\)erofiJanks andpopulation should pretty closelydetermineJhe .average .number of depositors per bank ina state, they do not determine the distribution of bank sizes within a state. If mandatory sharing is correlatedwith particular.sizedistributions in addition tobeing correlated with particular average sizes, and if size distributionsinfluenqetheJevel ofATM services, thenmandatorysharingmaysimplybereflecting this correlation and may have no causal effect on the supply of ATM transactions. $ome observers havesuggested thatmandatory sharing laws were passed under pressure from. small rural banks hoping to protect their. markets from larger metropolitan banks.41 If so,.andif small.rural banks supply lower levels ofATM services,.then mandatory sharing would be correlated with decreases in the number of ATM cards. There are several possible reasons small rural banks may be especially likely to supply lower levels of ATM services. One is that they are small and distant from large metropolitan banks, so sharing is more costly. Another is that they may have a protected monopoly market and may thus supplylower levels of services thanwould banks in a more competitive market. Athird reason may be associated with the low population density. Evenif a given rural bank has the same number of depositors as a metropolitan bank, it will be more costly for it to provide ATMs with the same level of locational convenience, since its depositors will be more geographically dispersed than the metropolitan bank's customers. Totestwhether mandatory sharing is associated withthe influences of small rural banks, I reestimated the regression reported in Table 6 with an additional explanatory variable, the percent of the population in metropolitan areas. The coefficient on this variable turned out to be insignificant, while population •and the number of banks remained significant. Although this evidence does not completely dismiss the rural bank argument, it does cast some doubt. The effects of unit banking are shown in Chart 2.·Unit banking ·has a significant negativeeffectup·toabout 284 banks.P This may be a consequence of reduced competition in local banking markets in unit banking states. All otherthings equal, local competition would be lowerin stateswith fewer banks. Thiswould help explain whythe negative effects become stronger as the number ofbanks decreases. Thepositive effects of unit banking, as seenin Chart2, become significant beyond about 1,540 banks.v' Texas, a unit banking state with 1,765 banks in 1987, is the only state in the sample with at least 1,540 banks. However, there areno statesin the sample withthis manybanks, unit banking, andnoATM placement constraints. Asexplained below and as seenin Chart 2, ATM placement constraints eliminate the positive effects of unit banking. Theeffects of ATM limitations in eitherunit or limited branching statesare negative and significant beyond about 395 banks.44 Theeffects of unit bankingand ATM placementconstraints together arenegative andsignificant upto about 395 banks. It is interesting to note that, once ATM constraints are added, the positive effects of unit banking disappear. This may be because ATM placement con- Chart 2 The Effect of Unit Banking and ATM Constraints on the Number of ATM Cards Cards (in thousands) 16,000 14,000 12,000 Unit Banking Alone " 10,000 8,000 6,000 Statewide Branching and no ATM Constraints ~ 4,000 2,000 o Unit Banking and ATM Constraints" 20 54.6 148.4 403.4 1096.6 Banks (Log Scale) Note: Calculations assume mean values for demographic and structural variables. 52 Economic Review / Winter 1990 straints foreclose any opportunity that unit banks would have to "branch by ATM." Several alternative specifications of the model were estimated. When regressions without the interaction terms were estimated, the coefficient estimate for the mandatory sharing dummy variable alone was insignificant. This indicates that bank size does play a role in helping to explain the effect of mandatory sharing. A regression using 1987 data for the legal variables also was estimated, and all ofthe legal variables were found to be insignificant. This suggests that the effects of regulation work with a lag. Regression Results with ATM Machines We have seen that mandatory sharing is associated with decreases in the number of ATM cards for states with relatively small banks. However, the number of ATMcards is only a proxy for the number of ATM transactions. Below, I have estimated a second regression, this time with the number of ATM machines as a proxy for ATM transactions. I have estimated a regression of roughly the same form as the ATM cards regression. The explanatory variables in the regression are defined as before. The results are presented in Table 7. As before, and as expected, population and per capita income have positive and highly significant coefficients. The number of branches has a significant negative coefficient. This time, though, the coefficients relating to mandatory sharing are insignificant. This result makes it difficult to draw inferences regarding the effect that mandatory sharing may have on the Chart 3 The Effect of Unit Banking on the Number of ATMs ATMs 7,000 6,000 5,000 Unit Banking 4,000 3,000 r---- Statewide Branching ~:::::::"~~-_oL-_...l--_==:::::L=-_..J.--_.....l-_---I 2,000 1,000 ~ 20 54.6 148.4 403.4 1096.6 Banks (Log Scale) Note: Calculations assume mean values for demographic and structural variables. Federal Reserve Bank of San Francisco 53 number of ATM transactions. We already know that mandatory sharing has a significant negative effect on the numberof ATM cards, and weassume thatATM cards transactions volume are. positively correlated. We also know that, across networks at least, ATM machines and monthly transactions are very strongly correlated." One possibility is that the numberof ATMs in a state is not a verygood measure of the numberof transactions in a state. There may be more uniformity in the relationship between growth in ATMs and growth in transactions within networks than withinstates because networks seek an efficient level of operations across state lines. Thus, the close relationship between machines and transactions within networks may not hold within states. Alternatively, changesin the numberof cardsmay more ana sensitively measure changes in the number of transactions than do changes in the number of ATMs. Any given percent change in transactions volume is likely to be represented by a percent change in cards that is greaterthantheconcomitant percentchangeinthe number of ATMs. The effects of unit banking are shown in Chart 3. Unit bankinghas a significantly negative effect on the number of ATMs up to about 254 banks and has a significantly positive effect beyond about1,236 banks. Again, Texas is theonly state in the sample withat least this manybanks. The positive effect may be due to ATMs serving as a substitute for traditional branches in states with small unit banks. IV. Conclusion As of September 1983, 23 states had instituted mandatory sharing statutes that required banks to share their ATMs with any other banks that wished to do so. The purposeof theselaws was to ensurethatsmallbankswould be able to offer their customers access to ATM systems. Since ATM systems were perceived to be subject to significant economies of scale, small banks feared that without mandatory sharing, only large banks would be able to participate in proprietary or shared ATM networks. There is some evidence that the cost structures of both sharedandproprietary ATM systems dopossess characteristics that make it difficult for small banks to gain access to ATMs. However, mandatory sharing does not appearto accomplish its goal. It either directly decreases the number of ATM access cards in the hands of depositors, or it simply does not sufficiently counteract negative independent forces that were left out of the regression and with which mandatory sharing is correlated. If the "reasonable fee" clause in mandatory sharing statutes does not in fact constrain the fee-setting behavior of shared network owners in mandatory sharing states, then there must be some such independent factor, correlated with mandatory sharing, that reduces the number of ATM cards. One such independent factor may be the presence of a large number of rural banks. However, there was no statistical supportforthis possibility. Although mandatory 54 sharing is correlated with the presence of banks with few depositors, it is notcorrelated withthe degreeof urbanization of the population. Moreover, bank size, as measured by number of depositors, was controlled for in the ATM cardsregression. Therefore, thenegative effects of mandatory sharing do not appear to be due to rural banks. Although mandatory sharing does not have a significant effecton the numberof ATMs, it does reduce the number of cards, suggesting that mandatory sharing may be increasing the costand priceof ATM transactions, or may be associated with such an increase. Thus, it is possible that mandatory sharing does give small bank customers access to ATM machines, but only at a significantly higherprice for all customers. Mandatory sharing does not appear to be able to legislate away the higher ATM costs faced by small banks. Unit banking has a significantly negative effect on the numberof ATM cardsin stateswith banksof largeaverage size, anda positive effectin thepresenceofrelatively small banks, butthe positive effectis reducedto insignificance if there are also ATM placement constraints. These results are consistent withthe view that unit banking is associated with reduced competition, higher prices, and lower service. The fact that unit banking has a significant negative effect on the number of ATM machines in states with banks of large average size also is consistent with this view. Economic Review / Winter 1990 NOTES 1. Commercial banks, savings and loans, and credit unions all have ATM programs. However, most ATMs are owned by banks. 2. Source: Kutler (July 22, 1988). The number of access cards is one measure of the scale of long-run demand for ATM services. The Electronic Funds Transfer Act, passed inNovember 1978 as an addition to the Consumer Credit Protection Act, states that a financial institution may issue a validated access card to a consumer only in response to an<oral or written request or application for the card. (Source: Regulation E, 12 C.F.R., Section 205.5(a)(1)). Invalidated cards may be distributed unsolicited, but the customer has to sign and return a form in order for the card to be validated for use. Therefore, the number of cards provides a better measure of the number of people that expect to use an ATM at least once than if validated cards could be distributed unsolicited. 3. American Banker (1988). 4. Kutler (September 30,1988). 5. The group that reports the 54 percent figure states that out of all ATM cardholders, 13 percent never use ATMs, 46 percent use them less than once a week and 41 percent use them at least once a week. In contrast, two thirds of cardholders over 54 years of age report either never using ATMs or using them less than once a month. Only 22 percent of the total population use ATMs at least once a week. 6. BankNetwork News (November 10, 1988). 7. Informal surveys indicate that consumer demand for ATM transaction services is fairly insensitive to direct fees. Although studies of actual ATM use are not available, in a 1988survey of customers who use ATM cards, 35 percent of those who pay fees said the fees caused them to cut back on their use of ATMs and 43 percent said fees did not do so, while 19 percent said they had always paid fees and 3 percent were unsure whether they paid fees. (Source: Kutler, September 30, 1988.) Another survey concluded that customers were not too price sensitive around a charge of about 30 cents. (Source: Herscher, 1988.) 8. American Banker (1988). 9. Avery, et a/. (1987). 10. Source: Kutler (September 30, 1988). 11. See, for example, Kantrow (1989), Herscher (1988), and ABA Banking Journal (1988). 12. Berger (1985). 13. American Banker (December 12, 1988). 14. "Electronic Funds Transfer Cost Models and Pricing Strategies," David A. Walker, Journal of Economics and Business, Fall 1980, pp. 61-65. 15. The network logos on the back of a customer's ATM card tell the customer that he has access to machines displaying those logos, as well as his own bank's machines. Different customers of the same bank may have Federal Reserve Bankof San Francisco different logos on their cards. It is nonetheless a fair generalization that there is universal access for all cardholding customers of all banks in the network. Also, it should be pointed out that a bank need not necessarily own any ATMs itself in order to belong to a shared network. 16. Many shared networks operate across state lines. 17. American Bankers Association (1987), p. 21. 18. Bank Network News (November 24, 1988). 19. Avery, et a/. (1987), p. 186. 20. TransData Corporation (1987). 21.. Cox (1989). 22-. Source: Conference of State Bank Supervisors (1984). More recent data are available, and some changes in state laws have occurred since 1983, but these are the data that were used for the regressions. Mandatory sharing laws are not completely uniform. Some states require sharing between like institutions, for example, banks with banks, but not between unlike institutions, for example, banks with savings and loans. All states that address the topic, however, at least permit sharing between like institutions. Furthermore, Nebraska is the only state that explicitly prohibits sharing between unlike institutions, though it allows third parties to own, operate, and maintain shared systems between unlike institutions. Most states that have mandatory sharing do not require sharing with out-of-state banks which request it. Those states that do not explicitly restrict mandatory sharing to in-state banks appear to be those which, under separate statutes, prohibit customers of out-of-state banks from using ATMs belonging to in-state state chartered banks. 23. The information on case law is from Baxter, et a/. (1977), pp. 138-140. 24. Baxter, eta/. (1977), pp. 141-143. 25. Einhorn (1988), p. 44. 26. However, as far as development costs go, initial owners of ATMs or participants in a network can expect the courts to uphold their right to demand some compensation for these expenses from any new members. See Baxter et a/. (1977), p. 141. 27. This measure of bank size differs from the traditional measures that use assets or deposits. 28. The figures for ATM cards do not include credit cards that may be able to access a line of credit for cash. The card data were obtained from a private consulting firm. 29. Network-level data on both the number of ATMs and transaction volume are available. These data show a strong positive relationship between the number of machines and transaction volume. In a regression of the log of ATM transactions on a constant and the log of the number of ATMs, the adjusted R2 was .8, and the coefficient on the number of ATMs was estimated to be 1.01, 55 with a t-statistic of 19.98. Assuming this relationship holds at the level of individual states, it appears that the number of machines is a good proxy for transaction volume. 30. The number of banks and number of branches were obtained from the year-end 1987 Reports of Condition and Income (Federal Financial Institutions Examination Council (1987». Population and percapita income in thousands in 1987 were obtained from the U.S. Bureau ofthe Census (1989). Neither the mean age of the population nor variables indicating the age distribution of the population were found to be significant in preliminary regressions. Therefore, age variables were excluded from the final reported regression. 31. The concentration ratio is the total share of deposits held by the four largest banking organizations in the state. It was obtained from the Board of Governors of the Federal Reserve System (1988). 32. I classified New Jersey as a mandatory sharing state, even though it was classified as a non-mandatory sharing state in the data source. I did so, because, as noted in a footnote in that source, sharing may be required by the New Jersey Banking Commissioner if the institution requesting to share maintains a principal, branch, or minibranch office within 5 miles of the proposed terminal location. 33. The limited ATM placement dummy variable takes a value of one if ATMs are not allowed to be placed statewide and zero otherwise. If the state has no statute or a silent statute regarding this topic, this dummy was given a value of zero. Louisiana, which allows statewide placement of ATMs only if they are shared, was assigned a value of one for this dummy. . 34. For a review of these types of studies, see McCall (1980). 35. Data for mandatory sharing laws and ATM branching laws were obtained from the Conference of State Bank Supervisors (1984). Data for traditional branching laws were obtained from the Board of Governors of the Federal Reserve System (1984). 56 36. Because of the log-linear specification of the regression, proportional changes in predicted values matter, not arithmetic differences in predicted values. This should be kept in mind when viewing Chart 1. Statistical tests reveal that the positive effects of mandatory sharing that appear in Chart 1 are insignificant. 37. The linear combination of coefficient estimates, B6 + B8*ln(banks),wastested for sign and significance at values ofln(banks) between 1 and 8 (banks between about 3 and 2980). All positive values of B6 + B8*ln(banks) were found to be insignificant. Negative values of B6 + B8*ln(banks) were found to be significant at a 5 percent level at and beyondln(banks) = 5.6 (banks = 270). The sample range for the number of banks is from 11 (Alaska) to 1,765(Texas). 38. Averyetal. (1986). 39. Avery et a/. (1987). 40. The high correlation between mandatory sharing and small banks may help to explain why the coefficient on BANKS is insignificant. There may not be enough states with both small banks and no mandatory sharing to obtain a good estimate of the coefficient on BANKS. 41. Baxter et a/. (1977), p. 139. 42. Montana, North Dakota, West Virginia, and Wyoming are the unit banking states in the sample with fewer than 284 banks. 43. At 602 banks, the mean number of banks in unit banking states, the effects of unit banking are insignificant. 44. The specification of the model assumes that the effects of ATM placement constraints in statewide branching states also would be negative and significant beyond about 395 banks. However, because there are no such states, this result is doubtful. 45. See note 29. Economic Review / Winter 1990 APPENDIX Derivation of Reduced Form of the ATM'ftansactions Model Assume that all variables arein logform. Theaggregate supply of ATM transactions in a state is given by: S = al + a2*BANKS + a3*SIZE + a4*MAND + a5*ATMLIM + a6*MANDSIZE + a7*ATMLSIZE + a8*UNITSIZE + a9*LIMSIZE + alO*CONC +all*UNIT aI2*LIM + a13*PRICE + e. (1) where BANKS is the number of banks, SIZE is the average size of a bank, in terms of number of depositors, MANDisa binary variable indicating thepresence or absenceof mandatory sharing, ATMLIM is a binary variable indicating the presence or absence of limitations on "branching by ATM," UNITis a binaryvariable indicating the presence or absence of unit banking, LIM is a binary variable indicating the presence or absence of limited branching, CONC = thedegree ofconcentration of thestatewide banking market MANDSIZE = (MAND)(SIZE), ATMLSIZE = (ATMLIM)(SIZE), UNITSIZE = (UNIT)(SIZE), LIMSIZE = (LIM)(SIZE), and PRICE = priceof an ATM transaction. The errorterm e is assumed to be normally distributed with mean zero. All the right-hand side variables except for PRICE are assumed to be exogenous. The variables UNIT and LIM are included as indicators of local market structure, in addition to the measure of statewide market structure, CONC. The bank size interaction terms are included because the effect of changes in bank size maydepend on whether or not laws regarding branching and ATM placement are in place. The signs of many of the coefficients are uncertain a priori. However, a2, a3 and a13 should all be positive. The aggregate demand for ATM transactions is given by: D = bl + b2*POP + b3*PCINC + b4*AGE + b5*OFFICES + b6*PRICE + n, (2) where POP = state population, PCINC = per capita income Federal Reserve Bank of San Francisco AGE = mean age, and OFFICES = total bank offices (main bank offices plus branches). Theerrorterm n is assumed tobe normally distributed with mean zero. All of the right-hand-side variables in (2) except for PRICE are assumed to be exogenous. The coefficients b2 and b3 should be positive, while b4, b5, and b6 should be negative. Setting S in (1) equalto D in (2) allows us to solve for PRICE. Substituting this solution back into (2), we eliminate PRICE from the. equation for ATM transactions.Two further assumptions are madein order to arrive at the final reduced form that-is estimated. First, the. SIZE variable should depend negatively on BANKS and positively on POP-. It is.assumed that SIZE is a non-stochastic function of BANKS and POP: SIZE = kl *BANKS + k2*POP, (3) where kl is negative and k2 is positive. Second, OFFICES = BANKS + BRANCHES, (4) by definition. Substituting from (3) and (4) into (2), and simplifying, we arrive at: ATM transactions = Al + A2*POP + A3*PCINC + A4*AGE + A5*BANKS + A6*BRANCHES + A7*MAND + A8*ATMLIM + A9*(MAND)(BANKS) + AI0*(MAND)(POP) + All *(ATMLIM)(BANKS) + AI2*(ATMLIM)(POP) + AI3*(UNIT)(BANKS) + AI4*(UNIT)(POP) + AI5*(LIM)(BANKS) + AI6*(LIM)(POP) + AI7*CONC + AI8*UNIT + A19* LIM +W, (5) where W is an error term. A regression of this form was estimated, and AGEand all of the population interaction terms were found to be insignificant. Eliminating them did not significantly change either the size or significance of the remaining variables' coefficients, so these variables were dropped from the final regressions. The final reduced form is then: ATM transactions = Bl + B2*POP +B3*PCINC B4*BANKS + B5*BRANCHES + B6*MAND .+ B7*ATMLIM + B8*(MAND)(BANKS) + B9*(ATMLIM)(BANKS) + BlO*(UNIT)(BANKS) Bll*(LIM)(BANKS) + BI2*CONC + B13*UNIT BI4*LIM + Z, + + + (6) where Z is an error term. 57 The coefficients BI through Bl4 are functions of al through a13, bl through b6 and kl and k2. Given the assumptions about the signs of a2, a3, alO, a13 and b2 through b6, B2 and B3 should be positive, and B5 should be negative. The sign of B4 is ambiguous because of the coexistence of a positive direct effect of an increase in the number of banks on the aggregate supply of ATM transac- tions and a negative effect of an increase in the number of banks on the size of banks. The coefficients B6, B7, B12, B13 and Bl4 will have the same signs as a4, as, alO, all and a12, respectively. The coefficients B8, B9, BlO and Bll will have signs opposite from those of a6, a7, a8 and a9, respectively. REFERENCES ABA Banking Journal. "What Roles for ATMs?," November 1988. American Banker. "Should Banks Charge User Fees?," December12, 1988. American Bankers Association. 1987 Retail DepositServices Report. Washington: American Bankers Association,1987. ____. Remote Electronic Facilities: An Analysis of Enabling Acts. Washington: American Bankers Association, 1976. Avery, Robert B., Gregory E. Elliehausen, and Arthur B. Kennickell. "Changes in the Use of Transaction Accounts and Cashfrom1984 to 1986," Federal Reserve Bulletin, March 1987. Avery, Robert B., Gregory E. Elliehausen, Arthur B. Kennickell, and Paul A. Spindt. "The Use of Cash and Transaction Accounts By American Families," Federal Reserve Bulletin, February 1986. Bank Network News. "1989 EFT Network Data Book," November 24,1988. ____. "EFT Growth Holds a Steady Course," November 10, 1988. ____. "1985 EFT Network Data Book," September 25,1984. Baxter, William F., Paul H. Cootner, and Kenneth E. Scott. Retail Banking in the Electronic Age; The Law and Economics of Electronic Funds Transfer. Montclair, New Jersey: Alanheld, Osmun and Co., 1977. Berger, Allen N. "The Economics of Electronic Funds Transfers," unpublished outline, 1985. Board of Governors of the Federal Reserve System. Annual Statistical Digest: 1987, Washington, 1988. ____ . Annual Statistical Digest: 1983, Washington, 1984. Cox, Rebecca. "ATM Networks Reducing Banks' Fees," American Banker, May 17,1989. Conference of State Bank Supervisors. A Profile of StateChartered Banking. Washington, 1984. 58 Einhorn, Theresa A. and Zimmer, Robert C. The Law of Electronic Funds Transfer. Washington: Card Services, lnc., April 1988. Federal Financial Institutions Examination Council. Reports of Condition and Income by All Insured Banks (FFIEC 031-034), 1987. Felgran, Steven D. "SharedATM Networks: Market Structure and Public Policy," New England Economic Review, Federal Reserve Bank of Boston, January/ February 1984. Herscher, Elaine. "Why Popular Bank ATMs Went From Free to Fee,"San Francisco Chronicle, December 26, 1988. Kantrow, Yvette D. "ATMs Called 'Only Hope' to Minimize Costs, Cope with Nation's Shrinking Labor Force," American Banker, January4,1989. Kutler, Jeffrey. "41% of ATM Cards Used Often; Need for More Marketing Cited," American Banker, September 30,1988. ____ . "Japanese Run 6 of Top 10 ATM Systems," American Banker, July 22, 1988. McCall, A. S. "The Impact of Bank Structure on Bank Service to Local Communities," Journal of Bank Research, Summer 1980. Regulation E, 12 C.F.R. Section 205.5(a)(1) (1987). Schmitzer, lana L. ATMFactBook. Washington: American Bankers Association, 1984. TransData Corporation. National Directory of Shared ATM/POS Networks, 1987 Edition. Salisbury, Maryland: TransData Corporation, 1987. U.S. Bureau of the Census. Statistical Abstractof the U. S.: 1989, (109th ed.). Washington, 1989. Walker, David A. "ElectronicFunds Transfer Cost Models and Pricing Strategies," Journal of Economics and Business, Fall 1980. Economic Review / Winter 1990