The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
ESTIMATINGINTERTEMPORALELASTICITY OF SUBSTITUTION:THE CASE OF LOG-LINEARRESTRICTIONS C/&g-Sheng Mao * 1. Introduction The modern theory of consumer behavior is concerned with how consumption adjusts to changing prices over time. When time is not involved, the demand for a normal consumer good declines as its relative price rises. Similarly, consumption at different points in time can be regarded as different goods, in which case the price that determines consumer behavior is the cost of today’s consumption in terms of tomorrow’s, or, equivalently, the cost of borrowing against the future. This price is called the real interest rate. When the expected real interest rate rises, consumers will attempt to defer current consumption by saving. Economists refer to the substitution between consumption at different points in time in response to changes in the real interest rate as intertemporal substitution in consumption. The mechanism of intertemporal substitution plays an important role in the theory of consumption and macroeconomics in general. For instance, it implies that consumers will smooth their consumption given the expected time profile of real interest rates and lifetime wealth. Thus, consumers respond to an increase in current income by raising both current and future consumption. This effect has been widely used in analyzing a number of important issues. These include the behavior of aggregate consumption over time, the volatility of stock prices, and the burden of government deficits and social security. Because the smoothing of consumption tends to propagate current shocks into the future, this mechanism also helps explain persistence of business cycles. Furthermore, the willingness of consumers to substitute intertemporally is a key determinant of the effectiveness of many government policies. Consider the recent debate over the reduction of capital gains tax rates. Proponents of the tax cut argue that it would * The author received helpful comments from Michael Dotsey, Marvin Goodfriend, Robert Hetzel, Thomas Humphrey, and Yash Mehra. FEDERAL RESERVE encourage saving by making current consumption more expensive relative to future consumption, i.e., by raising the after-tax real return to saving. In fact, however, the influence of the tax cut on saving and investment depends crucially on the response of consumption to the corresponding changes in the intertemporal terms of trade. Thus, to evaluate the empirical effect of the tax cut, or in fact any policy that is meant to promote saving and economic growth, one must know the intertemporal elasticity of substitution. While many authors have attempted to use actual data to estimate the intertemporal elasticity of substitution, their results are widely different. For example, using time series data in the United States, Hall (1988) concluded that there is no strong evidence that the elasticity is positive. By contrast, other studies have suggested a much stronger tendency of intertemporal substitution. The estimate obtained by Hansen and Singleton (1982, 1983), for instance, lies between 0.5 and 2, while the estimate obtained by Eichenbaum, Hansen, and Singleton (1986) can be as high as 10 depending on the data set used. The estimation by Hansen and Singleton (1988) even produces a negative elasticity estimate. At the very least, this wide range of figures raises questions regarding the reliability of the elasticity estimates. This paper explores the reliability of estimates of the intertemporal substitution effect using Monte Carlo simulation. A model economy is specified in which the modeler himself selects the intertemporal elasticity of substitution. Then, using conventional statistical techniques, data generated from model simulations are used to estimate the elasticity. Since the elasticity’s true value is known, one can check how closely the estimates conform to the value that was chosen in constructing the data. This technique allows one to evaluate the performance of the conventional strategies for estimating the intertemporal elasticity of substitution. Since many of the empirical BANK OF RICHMOND 3 Figure 1 studies on intertemporal substitution ignore the potential wage effect on consumption, this paper also examines the consequence of misspecification error for a simulated model in which changes in the real wage have effects on consumption behavior. It is shown that ignoring the wage effect can cause a substantial bias in the estimation of the elasticity of substitution in consumption. The next section outlines the notion of intertemporal substitution using a simple two-period model. Section 3 introduces a formal maximization problem, derives its first-order condition and discusses the estimation method. Section 4 lays out a model economy which serves a laboratory to generate simulation data. Section 5 summarizes the estimation results and Section 6 discusses the misspecification bias. 2. Intertemporal Substitution: A Two-Period Model To clarify the notion of intertemporal substitution, consider a simple two-period consumer’s problem. The consumer is assumed to be endowed with a fured income yr in the first period and yz in the second period. In period 1, there is a capital market where the consumer may borrow or lend at a competitive real interest rate rr. Let cl and c2 denote consumption in period 1 and period 2, respectively. Then the budget constraint, expressed in present-value form, is CI + cz/(l +rr) = yr + yz/(l +rr). That is, the present value of current and future consumption must exhaust but not exceed the present value of the consumer’s income stream. The consumer’s problem is to choose cl and c2 in order to maximize his utility, u(cr, cz), subject to the budget constraint. This is a standard textbook problem. The consumer will adjust his borrowing or lending so as to equate the marginal rate of substitution of cl for c2 with one plus the real interest rate. l In equilibrium, the consumer may be a net borrower or lender depending on his initial endowment position. Figure 1 depicts the consumer’s equilibrium in which the horizontal and vertical axes measure cl and cz, respectively. In equilibrium, the consumer will choose to consume at point E at which the indifference curve is tangent to the budget line, which has slope -(1 +ri). As depicted, this consumer is a net lender and saving is equal to (yr -cl). Now, suppose the real interest rate rises from rr to rr ‘, so that the budget line rotates clockwise around the endowment r In mathematical notation, this condition can be expressed as ur/ua = (1 +rr), where ui (i = 1, 2) is the marginal utility of consumption in period i. 4 ECONOMIC REVIEW, point (yr, ~2) and has a steeper slope. A key question is how the consumption ratio cz/ci will respond to such a change. First, because consumption becomes relatively more expensive in period 1, there is a substitution effect that induces the consumer to substitute cz for cl by making more loans in the bond market. Because the consumer is lending, however, there is also an income effect that tends to raise consumption in both periods. Whether or not the consumption ratio cz/cr will rise depends upon the relative magnitude of these effects. For the purpose seems of this paper, the standard assumption reasonable, namely, that on balance cz/cr increases or that the income effect on cl is not strong enough to outweigh the substitution effect and the income effect on 122.2As a result, the new equilibrium will be reached at point E ’ where the consumption ratio cz/ci is higher. Because of the assumption of constant elasticity, the increase in cz/cr is proportional to the increase in the real interest rate. The ratio of the percentage change in the rate of growth of consumption to the percentage change in the real 2 To be precise, the consumer’s utility function is taken to be homothetic and constant elastic. This assumption implies that the consumption good in each period is normal and that the slope of the indifference curve is constant along a given ray from the origin. Note that a utility function is called homothetic if the marginal rate of substitution depends only on the consumption ratio, and it is called constant elastic if the marginal rate of substitution is proportional to the consumption ratio. An explicit utility function will be specified in the next section. NOVEMBER/DECEMBER 1989 interest rate is called the intertemporal elasticity of substitution. It is clear that the curvature (or the elasticity) of the indifference curve will determine the extent to which the consumer responds to changes in the real interest rate. The more elastic or less curved is the indifference curve, the greater the response will be. Figure 2 depicts the difference in the intertemporal substitution effect of two utility functions with different curvatures. For simplicity, assume that the initial equilibrium is the same so that both indifference curves UI and uz are tangent at the same point E to the budget line. Note that the curve ur has flatter curvature and is therefore more elastic. Suppose the real interest rate rises from rr to rr ‘. Then the new equilibrium will move from point E to point F in the case of ur, and to point G in the case of u2. Comparing the consumption ratio CZ/CI at point F and G reveals that consumption grows faster when the indifference curve is more elastic. Thus, there is a positive relationship between the intertemporal elasticity of substitution and the elasticity of the indifference curve. Now, suppose an econometrician who observes data on consumption and real interest rates over time wishes to estimate the intertemporal elasticity of substitution. How would he go about doing this? The preceding analysis suggests that a natural approach is to think of each observation in time as represented by the tangent point between the indifference curve and the budget line. As one traces out these equilibrium points over time, one essentially looks at the change in these tangent points which are determined by the curvature of the indifference curve. Thus, to estimate the elasticity one could simply regress the rate of growth of consumption on the real interest rate. This approach has been widely used by many authors to study the dynamic behavior of consumption [e.g., Hansen and Singleton (1983) and Hall (1988)]. The foregoing discussion illustrates how equilibrium conditions can be used to interpret economic data. Its implementation, however, requires more rigorous elaboration. For example, because of the stochastic nature of the data one must consider individual behavior under uncertainty. Also, in order to account for the evolution of consumption over time a fully dynamic model needs to be developed. Accordingly, the next section presents a formal maximization problem in which the equilibrium conditions are explicitly used to construct the regression equation to be estimated. 3. The Optimization Framework To start with, the consumer is assumed to have a time-separable utility function of the following form:3 1 Uh) Figure = 2 I 1 -l/a [Ctl-l’o-l], I Ma), if (T > 0 and Of1 ifa= 1 This utility function, which has been widely used in the literature, has the property that the elasticity of substitution in consumption4 is constant and is equal 3 A utility function is called time-separable when the marginal utility of consumption in a given period is independent of the level of consumption in other periods. This assumption simplifies the analysis. 4 The elasticity of substitution in consumption is defined as the partial derivative of the rate of change in consumption with respect to the marginal rate of substitution holding the level of utility fixed. In notation, this can be expressed as: a Met + h) Ci In[u’(ct)/u’(ct+ r)] I u =; ’ where u ‘(.) denotes the marginal utility of consumption and ; a constant utility level. Note that this quantity measures an income-compensated substitution of consumption along a given indifference curve which is different from the uncompensated notion of intertemooral substitution. The two notions. however, turn out to be equivalent for two reasons. (1) The income effect is proportional to changes in wealth due to the homotheticity of the utility function. (2) The real interest rate will pin down the marginal rate of substitution in equilibrium. FEDERAL RESERVE BANK OF RICHMOND 5 to the parameter (T. As will be seen shortly, this parameter will control the interest rate effect on consumption. Now, let us consider the budget constraint. At the beginning of time t, the consumer carries kt units of capital from the last period. The capital is traded in a competitive market and yields a stochzs~icrate of return rt in units of consumption goods. At the end of period t, the consumer collects interest income rtkt and principal kt. This sum is the only income that the consumer allocates between consumption ct and new capital kt + 1 to be carried into the next period. Thus, the consumer’s budget constraint for period t is ct + kt + 1 = (1 +rt)kt. The consumer’s problem is to choose a path of consumption and capital, contingent on the realization of capital returns, that satisfies the budget constraint each period and maximizes the expected present value of lifetime utility over an infinite horizon.5 That is, given the initial capital stock ko, the consumer solves max Eo[ F @u(ct)] t=O subject to ct + kt + 1 = (1 +rt)kt for all t where /3 is the time preference discount factor that lies between 0 and 1, and Eo is the expectation operator conditional on information at time 0. The first-order condition (or Euler equation) of this problem is u’(ct) = P Eb’(ct+l) (l+rt+dl It1 (1) where It denotes the information set at time t.6 This equation is precisely a stochastic version of the equilibrium condition that the budget line must be tangent to the indifference curve as depicted in Figure 1.7 This equilibrium condition states that the marginal cost of investing an extra unit of consumption good at time t (i.e., the foregone marginal utility of consumption) should equal the marginal benefit from investing - this return being com5 The assumption that the consumer lives forever is here employed for analyticalconvenience only. The specificationof a finite horizon problem will not alter the results of this paper. 6 The informationstructure is unspecifiedhere. Note, however, that its specificationis necessary for computing the conditional expectation. ’ Ignoringthe expectationoperator, equation(1) simply says that the ratio of the marginal utilities (expressed in units at time t) is equal to one plus ;he real interest r&e, which is the first-order condition for the two-period model in Section 2. 6 ECONOMIC REVIEW. posed of the expected present value of the marginal utility of consumption times the investment proceeds at time t + 1 (principal plus interest). This condition implies that a small deviation from the optimal consumption plan will leave lifetime utility unchanged. From an empirical standpoint, the above first-order condition is all that is needed to estimate the intertemporal elasticity of substitution. Obtaining the estimate involves use of a simple procedure to derive a regression equation from (1). First, given the constant-elastic utility function specified at the beginning of this section, (1) takes the form EN (ct + l/cd - 1’0 (1 +rt+l) -l(It] = 0. (2) This equation says that the residual (i.e., the term defined in the bracket) has a zero mean conditional on information available at time t. It implies that any variable included in the information set should be uncorrelated with the residual. These restrictions, referred to as orthogonality conditions, admit a class of instrumental variables procedures for estimating the parameters p and n [e.g., Hansen (1982) and Hansen and Singleton (1982)]. As can be seen, equation (2) is highly nonlinear and difficult to work with. A common procedure is to make distributional assumptions on certain variables at hand, and to transform the equation into a linear representation. This transformation renders the equation easy to estimate but its tractability is obtained at the cost of an extra assumption which may not be true.8 Specifically, assume that the measured growth of consumption ct + l/c* as well as the real interest rate (1 +rt + 1) has a lognormal distribution.9 This assumption implies that ln(xt+ I), where xt + 1 = P(ct + lh) - l’? 1 + rt + I), has a normal distribution with a constant variance v and a mean pt conditional on It. Using the lognormality assumption, we have E[xt + 1[It] = exp[pt + v/Z]. Comparing with equation (2) yields exp[pt + v/2] = 1, which in turn implies pt = -v/2. Since, by definition, pt = E[ln xt + II&], it follows that -v/2 = pt = In fi - l/a E[ln(ct+ I/ct)lIt] + EM1 +rt+ djL1. * It should be noted, however, that distributional-independent methods such as the generalized method of moments proposed by Hansen (1982) is available for dealing with nonlinear problems. The results pertaining to this procedure are beyond the scope of this paper, and are presented in Mao (1989). 9 A random variable X is lognormally distributed if the natural logarithm of X has a normal distribution. By definition, XY is lognormally distributed if both X and Y are lognormally distributed. If In(X) has a normal distribution with mean p and variance Y, then the mean of X is exp[p+v/Z]. NOVEMBER/DECEMBER 1989 Multiplying both sides by 0 and arranging terms yields EMct+dct)IItl = PO + u E[ln(l +rt+l)lItl, where /30 = a[ln P + v/21. Let Et+ 1 = ln(ct + l/et) - Ellnkt + ht) lItI, then Jn(ct+lW = PO + ~Elln(l+r~+d~Ll + et+l. (3) Note that the expectational error Et+ 1 is uncorrelated with the variables included in the information set, and is normally distributed with a zero mean and a constant variance. As can be seen, the parameter u identifies exactly the intertemporal elasticity of substitution. This equation is used later to estimate the parameter u. Equation (3) implies that the mean of the rate of growth of consumption is shifted only by the condotionai mean of the real interest rate. That is, information at time t is helpful in predicting the rate of growth of consumption only to the extent that it predicts the real interest rate. Since the expectedreal interest rate is determined endogenously within the model, an instrumental variables procedure will be used to estimate the parameter u. This procedure amounts to two-stage least squares in which the first stage estimates the expected real rate using variables (instruments) contained in the information set consisting of observations on past consumption growth and real interest rates. The projected real interest rates are then used in equation (3) to estimate u. This procedure yields, a consistent estimate of the intertemporal elasticity of substitution. As mentioned before, it has been difficult to pin down the parameter u. The point estimates vary widely, ranging from near 0 to 10. These results suggest that the linear regression equation (3) may not be a proper model for estimating the intertemporal elasticity of substitution. To examine this issue more closely, consider the following question. Given that the the true value of u is known, how accurately can that value be recovered by using (3) and the econometric procedure outlined above? A Monte Carlo experiment is carried out to answer this question. 4. The Data Generating Process The first step of the Monte Carlo experiment is to write down a model economy whose output will be used to simulate the data. In particular, the economy is represented by a general equilibrium model in which the underlying production process is explicitly specified. 10This approach allows quantities as well as prices to be endogenously determined within the model. The economy is similar to that described in Section 3 with the exception that the consumer now also plays the role of producer. In each period, the consumer carries from the previous period kt units of capital which are used to produce output. Due to the weather and other uncontrollable random factors, however, the volume of output is uncertain. To capture such uncertainty, the technology is represented by a production function of the form: yt = AIF = XtktU, 0 < a < 1, where yt is output produced at time t and Xt is a random shock with a known probability distribution. The output may be consumed or invested. If invested, the capital will depreciate at a constant rate 6 (0 < 6 < 1) so that the investment at time t is defined to be it = kt + 1 - (1 - 6)kt. The agent is assumed to have a constantelastic utility function as specified above. His problem is to choose a contingent plan for consumption and investment so as to maximize his expected lifetime utility. That is, the agent solves max Eo[ c” @u(ct)l t=O subject to ct + it = XtF(kt) for all t. The solution of the above maximization problem consists of a sequence of consumption and investment outcomes over time, contingent on the realization of the random shock Xt. In this way the model generates the consumption data for estimating the intertemporal elasticity of substitution u in (3) above. The model also generates an implied real interest rate time series, needed to estimate (3). To see this, consider the first-order condition: u’(ct) = P &(u ‘(ct + I) IA, + IF ‘(k + 1) + (1 - ml. (4) The intuition behind (4) goes as follows. Suppose at time t the agent decides to carry one extra unit of consumption good to the next period, which will cost him, in utility terms, the marginal utility of consumption. The gain that results is the expected present value of the marginal utility of consumption times the extra output that can be produced at time t + 1, which is equal to the sum of the marginal product lo Readers familiar with the literature on economic growth will recognize that the model specified is a standard optimal growth model as studied by Brock and Mirman (1972). FEDERAL RESERVE BANK OF RICHMOND 7 of capital and the amount of capital that is left over after depreciation. Equating the cost and benefit in equilibrium yields equation (4). As can be seen, equation (4) is identical to the first-order condition of the consumer’s problem [equation (l)] except that the real interest rate is replaced by the rate of return on investment, i.e., the marginal product of capital minus the depreciation rate. Because the optimization problem does not have a closed-form solution, a numerical method will be used to solve the problem. Specifically, a dynamic programming algorithm is employed to approximate the solution over a discrete state space.” It is assumed that the production shock Xt can take 5 distinct values over the set [0.9, 1.11, i.e., 0.9, 0.95, 1 .O, 1.05, 1.1, and that it evolves over time according to the following Markov transition probability: l* r !0.50 0.25 0 0.25 0.50 0.30 0 0.25 0.20 0.50 0.25 0.30 0.50 0 0.25 0.50 0 This transition matrix implies that the random shock will be, to some degree, persistent over time because the probability of staying in the same state is higher than that of switching to other states. The choice of this transition matrix is motivated in part by the fact that the actual production shocks in the United States, as measured by the Solow residual,13 are positively correlated over time. The estimation results reported below do not appear to be sensitive to the specification of this transition matrix. Other parameters that are held constant throughout the experiment are: 0 = 0.96, (Y = l/3 and 6 = 0.1. These numbers are also chosen to reflect data actually generated from the United States economy. For example, the value of /3 implies a real interest rate of about 3 percent a year, which is close to what is observed in the United States. The (Y value is *I The algcrithm, known as the value successive approximation, iterates on the problem’s value function over a discrete state space. Technical details can be found in Bertsekas (1976). r* The elements of this transition matrix assign the probability of moving from one state to another. For example, if the value of the production shock at time t is 1.0 (the third row), then there is 25 percent chance that it will move to 0.95 or to 1.05 in the next period and 50 percent chance that it will stay in the same state. I3 Whether the Solow residuals, i.e., the residuals arising from the regression of a production function, truly represent the underlying shocks of the economy is a controversial matter. This issue is ignored here. 8 ECONOMIC REVIEW, chosen to reflect the output elasticity of capital in the United States-that elasticity figure being roughly one-third and holding fairly steady over a long period of time. Given these parameters’ values, the model is solved for a set of four different values for u (0.1, 0.25, 1.0, and 2.5). Since no interest attaches to the numerical solution per se, it is not reported. It is crucial, nevertheless, to have some idea about the accuracy of the approximation procedure before the solution can be used to generate random samples. This accuracy can be assessed by checking whether the data generated from the model satisfy the first-order condition, i.e., equation (2). Let ht + r = fl(ct + i/et) - “O( 1 +rt + 1) - 1, then (2) can be rewritten as E[ht + rl~t] = 0. As mentioned before, this condition implies a set of orthogonality conditions which require that the residual ht + r be uncorrelated with any variable included in the information set. Let zt be a subset of It; then these conditions imply that the first sample moment of the cross product ht + rzt should be close to zero for a sufficiently large sample. The vector zt consists of a constant of ones plus the past observations on consumption growth ct + i/et and the real interest rate (1 + rt + 1). The constant term is included because the unconditional mean of ht + i must be zero. Reported in Table I are, for each u value, the sample means of the product ht+ rzt based on a realization of 2000 observations. The number of lags used for consumption growth and the real interest rate is 2, so in total there are 5 variables in the vector zt. The same set of variables will be used as instruments in the econometric procedure of the next section. As can be seen, the means are very small and insignificantly different from zero (standard deviations of the mean are reported in parentheses). This result also holds for smaller sample sizes which are not reported here. To conclude, the data generated from the solution procedure fulfill the Euler equation and have negligible approximation error. 5. Estimation Results This section pursues the second step of the Monte Carlo experiment. The intertemporal elasticity of substitution u is estimated using equation (3) and data generated from the simulated economy discussed in Section 4. The objective here is to see if this strategy produces a reliable estimate of u. A brief description of the simulation procedure follows. First, for each of the four u values considered in the experiment are generated a number of random samples from the artificial economy. These observations are then employed to estimate the parameter u. This process produces a sampling distribution of NOVEMBER/DECEMBER 1989 Table I ORTHOGONALITY CONDITIONS cl 0.10 0.25 constant (one) 2.50 Note: means of the cross product (ct + Jet) - 1 (ct+1’ct)-2 between h,,, and (l+r,+J-, (l+r,+,)-, 0.000048 0.000078 0.000052 0.000014 0.000026 (0.002415) (0.002417) (0.002416) (0.002508) (0.002507) -0.000017 (0.001073) 1.00 Sample - 0.000000 -0.000016 -0.000014 (0.001073) (0.001073) - 0.000000 -0.000001 -0.000025 (0.001117) - 0.000001 -0.000021 (0.001117) - 0.000000 (0.000218) (0.000218) (0.000218) (0.000227) (0.000227) 0.000003 0.000003 0.000003 0.000003 0.000003 (0.000004) (0.000004) (0.000004) . (0.000004) (0.000004) Calculation is based on 2000 random observations. Standard deviations of the mean are reported in parentheses. the point estimate a’ for a given sample size. To examine the convergence property of these estimates, the experiment is repeated using four different sample sizes, ranging from 50 to 500. As in Section 4, five variables are chosen as instruments, which include two lags of the the consumption growth ln(ct + r/c*) and two lags of the real interest rate ln( 1 + rt + 1). The estimation results reported below are not sensitive to the number of lags included in these instruments. Sampling Di.mhtion of the Point Estinzate a”. Consider Table II wherein are reported the means and the standard deviations of the elasticity estimate a, These statistics are calculated for each of the four u values and each of the four sample sizes considered in the experiment. At first glance, the sampling distribution of the point estimate a” appears to have a relatively small standard deviation and a mean that is close to the true value of cr. Although the means are slightly higher than the true value, the bias is not significant and is probably due to the approximation error of the solution procedure in Section 4. In fact, as the sample size increases, the bias as well as the standard deviation vanishes, a clear indication that the estimate 6 is asymptotically unbiased and consistent. Notice that, even for a relatively small sample, one cannot reject the hypothesis that the mean of the estimate a’ is equal to the true (Tvalue. Extensive simulations indicate that these results are robust to the specification of the stochastic process of the production shock Xt. For example, using an indeFEDERAL RESERVE pendently and identically distributed random shock the sampling distribution of the elasticity estimates is virtually identical to that reported in Table II. The implication is clear: Equation (3) as an empirical model of consumption is capable of producing a reliable estimate of the intertemporal elasticity of substitution, at least for the cases considered in this paper. This result is somewhat puzzling because the data used in the estimation procedure do not necessarily satisfy the lognormal restriction that renders the regression model linear. Violation of this distributional assumption tends to cause the estimate to be biased and inconsistent. This issue warrants closer examination. Figure 3a-3d plots, respectively for each of the u values, the frequency distribution of the random variable ln(xt+ I), where xt+ 1 = Pht + lh) - 7 1 + rt + 1). As mentioned in Section 3, this random variable should have a normal distribution if the lognormality assumption is correct. The figures indicate that while such a distribution appears to be the case when (T = 2.5, it is apparently violated when u = 0.1, 0.25, and 1.0. How can we reconcile this finding with the simulation results? In particular, how does one explain the unbiasedness of the estimates even if the distributional assumption is violated? It turns out that the answer is quite simple. What happens is that, under certain conditions, the Euler equation (2) can be approximated by a linear regression model without directly invoking the lognormality assumption. Recall the following approximation: ln(xt + 1) = ln( 1 + ht + 1) E ht + 1 BANK OF RICHMOND 9 Table SAMPLING DISTRIBUTION True (I 0.10 0.25 1.00 2.50 Number of observations 50 II OF THE POINT ESTIMATE 6 (a) (I Number of simulations Mean 780 0.257039 0.155508 150 520 0.172956 0.070608 300 480 0.142281 0.048254 500 400 0.129667 0.038071 50 780 0.414662 0.205668 150 520 0.321207 0.100773 300 480 0.286916 0.070803 500 400 0.273533 0.056699 50 780 1.126016 0.275207 150 520 1.044132 0.150668 300 480 1.017989 0.105218 500 400 1.009004 0.084706 50 780 2.504959 0.021614 150 520 2.503065 0.011713 300 480 2.502775 0.007199 500 400 2.502399 0.005670 ta) These results are based on assumed highly persistent shocks specified and identically distributed (iid) shocks yield similar results. in the text. Experiments for xt + 1 close to one or ht + 1 close to zero. Since the condition that ht + 1 be close to zero is approximately true for our data (see Table I and Figure 3), the linear regression equation (3) can be viewed as an approximation to the Euler equation (2). It is worth mentioning that in the United States the rate of growth of consumption is about 2 percent a year and the annual real rate of interest is about 3 percent, suggesting that xt + 1 is close to one. Hypothsk Testing Based on the regression model, a number of hypotheses can be tested. This subsection focuses on the simple hypothesis that the parameter u is equal to its true value. As usual, this hypothesis can be tested using a conventional t statistic. Since we know the true u value that is used to generate the data, we are interested in the Type I error for testing this hypothesis, that is, the proportion of time that the null hypothesis is rejected when it should have been accepted. The test results are summarized in Table III. As can be seen, the rejection frequency of the true model is higher than expected. This is particularly clear when ~7is small. 10 s.d. ECONOMIC REVIEW, 6. with independently For example, at a 5 percent significance level, about 20 percent of the time one will reject u = 0.1 even though the sample size is relatively large (say, 500). At a 10 percent significance level, the proportion rises to above 30 percent. Although the rejection frequencies are somewhat moderate for other cases, it seems reasonable to conclude that the risk of committing the Type I error is still too high. Again, this result may appear puzzling because the point estimate is fairly close to the true parameter value. A moment’s reflection reveals that these errors stem from the standard error of the estimate’s being so small that the true parameter value lies outside of the confidence region. Misspecification Bias with Variable Labor Supply Many of the empirical studies on intertemporal substitution abstract from the interaction between consumption and labor supply decisions and thereby ignore the potential effect on consumption of changes in the wage rate [for example, Hansen and Singleton (1983) and Hall (1988)]. As noted before, such a simplification implies that the growth of consumption is determined only by the expected real interest rate. This section examines a more realistic model in which an individual chooses both consumption and labor supply at the same time. Such a model implies that changes in the real wage can have important effects on consumption behavior. It will be shown that failure to incorporate these effects can result in a sizable bias in estimating the intertemporal elasticity of substitution. As in the previous case, the starting point is a simple two-period model. For comparison, refer to Figure 1 in which the equilibrium moves from point E to E’ when the real interest rate rises. What would NOVEMBER/DECEMBER 1989 Figure FREQUENCY DISTRIBUTION 3 OF THE TRUE RESIDUALS (a): u = 0.10 141 1 10 8 F t! 2 E8 8 b, -0.2 -0.1 0.0 0.1 6 0.2 -0.016 (b): u = 0.25 -0.008 0.000 O.dO8 0.016 (d): u = 2.50 71 6c 6 5’ -0.06 -0.02 0.02 0.06 0.10 happen if the consumer is allowed to supply work effort in the labor market and earn wage income? In general, the point E ’ will no longer be an equilibrium because the labor supply decision, even if the wage rate remains unchanged, is likely to alter the rate of substitution in consumption. In this case, the equilibrium point can go in either direction depending upon the extent to which labor supply affects the marginal utility of consumption. In order to make a specific prediction, one needs an explicit model. The model considered below is similar to that described in Section 3. First, the consumer’s utility function is assumed to depend on consumption ct and leisure time It and has the following form: FEDERAL RESERVE - 0.0004 - 0.0002 0.0000 0.0002 0.0004 &$C’@ lt(l -~I~-:‘“,,~~~, UWt) z 1 = f3 In ct + (l-0) In It, ifa = 1 This utility function is similar to that specified before and is constant elastic with respect to a “composite good” defined as a Cobb-Douglas function of consumption and leisure. The parameter 8 lies between 0 and 1. As will be seen shortly, the parameter u can still be identified as the intertemporal elasticity of substitution. But, more importantly, the u parameter controls the effect of leisure on the marginal utility of consumption. Specifically, when BANK OF RICHMOND 11 Table III (19SS)l. Since there is no direct evidence on whether the utility function is separable, it is useful to check how serious the misspecification bias could be. To proceed, suppose the consumer solves the following maximization problem: REJECTION FREQUENCY OF THE NULL HYPOTHESIS: u true da) = (Type I Error) True 0 0.10 0.25 1.00 Number of observations Significance L level 5 Percent 10 Percent 50 26% 39% 150 21% 32% 300 18% 29% 500 19% 33% 50 23% 35% 150 16% 24% 300 12% 19% 500 11% 20% 50 19% 29% 150 13% 19% 300 7% 14% 500 9% 14% max I%] c” P’u(ctJ41 t=O s.t. ct + kt + r = (1 +rJkt + wtnt for all t where wt is the wage in terms of consumption goods and nt = 1 - It is work effort. Following the same derivation procedure as in Section 3 and assuming lognormality, it can be shown that consumption now obeys the following equation: ln(ct+dct) = PO + u EM1 +rt+ d\Itl + ,&EMwt+dw)~Itl + Et+I (5) r4 That is, uCr > 0 if u > 1, where uCris the partial derivative of the marginal utility of consumption with respect to leisure time. where fir = (1 - t9)(1 - a). Except for the additional term that captures the effect of wage growth on consumption, this equation is similar to equation (3) which abstracts from the labor supply decision. As can be seen, the parameter u still measures the interest rate effect on consumption. However, the wage will have a positive effect (pr > 0) on consumption growth if u < 1, and negative effect (/3r < 0) if u > 1. This is so because u < 1 implies ucr < 0, so that when the real wage rate rises, leisure will decline and the marginal utility of consumption will rise. As a result, consumption must rise to restore the equilibrium. Note that when u = 1, a change in the real wage has no effect on consumption because the utility function is additively separable in this case. What would happen if the true data were generated from the above model, and yet the econometrician erroneously ignored the wage effect and instead used (3) to estimate a? This is a typical specification error in which an important variable is omitted from the regression. Apparently, the estimate for u will be biased, with the magnitude of the bias measured by the true value of /I1 times the auxiliary regression coefficient of the wage growth on the real interest rate.r6 Thus, if the real interest rate and the growth of real wages are positively (negatively) correlated, then ignoring the wage effect leads to a downward (upward) bias if u > 1, and an upward (downward) bias if u < 1. Notice that, if the real interest rate and the growth of real wages are un- I5 A utility function u(x,y) is additively separable if it has the form: m(x) + n(y). This class of utility functions is not limited to the logarithmic case specified in the text. I6 This is a standard result on specification (1977). 2.50 50 11% 150 9% 19% 300 10% 20% 500 12% 20% 19% (a) These results are based on assumed highly persistent shocks specified in the text. Experiments with iid shocks yield much higher rejection frequencies (more than 50 percent). (I > 1, consumption and leisure are gross complements because an increase in leisure will raise the marginal utility of consumption.14 The opposite is true when u < 1. The value of u will dictate the effect of the real wage on consumption. It is important to note that the wage effect on consumption will depend on the form of the utility function. In particular, if the utility function is additively separable,15 then the marginal utility of consumption will be independent of the choice of leisure. In this case, changes in the real wage have no effect on consumption. Consequently, equation (3) will still be the correct specification for consumption. This assumption has been maintained by most authors [e.g., Hall 12 ECONOMIC REVIEW, NOVEMBER/DECEMBER 1989 bias. See Maddala .: d correlated, then the elasticity estimate using (3) will be unbiased. One way to evaluate the extent of the above misspecification bias is to conduct a Monte Carlo simulation. As in Section 4, the data are generated from a model economy in which the production function is assumed to be yt = Xtkt%t(’ - a), 0 < CY< 1.” The production shock is generated in the same way as before. Other parameters fixed in the experiment are fl = 0.96, 6 = 0.1, cx = l/3, and 0 = 0.3. Following the same procedure, u is estimated using (3) as well as (5). Because of the difference in the specification, the instruments used in estimating equation (5) include lags of ln(ct + I/et), ln( 1 +rt + 1) and ln(wt + l/wt). These instruments are used to project the expected real interest rate as well as expected wage growth. Table IV summarizes the means and the standard deviations of the estimated bias. It is clear that when the model is correctly specified, i.e., equation (S), the estimated bias is small and insignificant. However, the bias associated with equacion (3) is sizable. In particular, when (T = 0.25, the point estimates are scattered around the value of 2, and when u = 2.5, the point estimates are less than one and in some cases close to zero. These results show that ignoring a potential wage effect on consumption can introduce a substantial bias in the estimation of the elasticity of substitution. 7. Concluding Remarks The results of this paper can be summarized succinctly. First, for a moderate sample size (perhaps in the range of 100 to 150), the point estimate of the intertemporal elasticity of substitution produced by the linear model tends to be unbiased with small standard errors. This result implies that the loglinear model, despite its simplicity, is a useful and convenient framework for estimating the intertemporal elasticity of substitution. Second, the conventional t test tends to over-reject the true model. Therefore, one must be careful in drawing conclusions from this test. Third, if the estimated equation is erroneously specified and omits the effect of the real wage on consumption, then the bias of the elasticity estimate is sizable. One should not conclude, however, that it is always necessary to use the I7 Specifically, the data are generated from a real business cycle model: extended s.t. ct + kt+l = XtFhnt) model to estimate + (1 - 6)kt where F(. , .) is the production function which depends on capital and labor. As in Section 4, the equilibrium prices can be computed directly from the solution of the optimization problem. In particular, the real interest rate is the marginal product of capital minus the depreciation rate while the real wage is just the marginal product of labor. Table 0.25 2.50 Number of observations Number of simulations similar BIAS Bias: 0 - True o elasticity; IV MISSPECIFICATION Correct: the biases could arise in the extended model if it is also misspmified. In general, any econometric method founded on an intertemporal maximization problem and its resulting Euler equation is bound to be sensitive to measurement errors. Such errors are particularly characteristic of consumption data, especially data on durable goods consumption. They are perhaps max &[ c” /3’u(ct, 1 -nt)] t=O o Eq. (5) Mean Incorrect: sd. Mean Eq. (3) s.d. 50 600 0.119739 0.066889 1.958582 0.667838 150 400 0.053412 0.049080 1.732927 0.453833 300 400 0.030032 0.033670 1.692648 0.326624 500 300 0.022194 0.027314 1.670278 0.267501 50 600 0.433372 0.522541 - 1.770626 0.310914 150 400 0.174026 0.330437 - 1.657668 0.189137 300 400 0.080718 0.220140 - 1.607193 0.129013 500 300 0.057523 0.184815 - 1.596351 0.108533 FEDERAL RESERVE BANK OF RICHMOND 13 the most important reason why empirical studies have not been able to pinpoint the intertemporal elasticity of substitution. As shown above, however, even if the data are properly measured, the econometrician still must choose a correct specification. Ironically, the data themselves are supposed to aid in this task. There is no easy solution to this identification problem. There are at present more sophisticated test procedures, such as tests of overidentifying restrictions, that may be used to discriminate among different models. However, the properties of such test statistics under misspecification are not clear. References Bertsekas, Dimitri P. Dynamic Pmgramming and Stochastic Con&. New York: Academic Press, 1976. Brock, W. J., and L. J. Mirman. “Optimal Economic Growth and Uncertainty: The Discounted Case.” Jownaf of Lhomic Theory 4 (June 1972): 479-513. Eichenbaum, Martin S., Lars Peter Hansen, and Kenneth J. Singleton. “A Time Series Analysis of Representative Agent Models of Consumption and Leisure Choice Under Uncertainty.” Working Paper No. 1981. National Bureau of Economic Research, July 1986. Hansen, Lars Peter. “Large Sample Properties of Generalized Method of Moments Estimators.” Economettica 50 (july 198.2): 1029-54. ECONOMIC REVIEW, . “Stochastic Consumption, Risk Aversion, and the Temporal Behavior of Asset Returns.“Jounalof Pofirical Economy 91 (April 1983): 249-65. . “Efficient Estimation of Linear Asset Pricing Models with Moving Average Errors.” Manuscript. University of Chicago, April 1988. Maddala, G. S. Economerris. New York: McGraw-Hill, Hall, Robert E. “Intertemporal Substitution in Consumption.” humaf of Po.kicd Economic 96 (Apd 1988): 339-57. 14 Hansen, Lars Peter, and Kenneth J. Singleton. “Generalized Instrumental Variables Estimation of Nonlinear Rational Expectations Models.” Econotnetn~a 50 (September 1982): 1269-86. 1977. Mao, Ching-Sheng. “Euler Equation Estimation: A Simulation Study.” Manuscript, Federal Reserve Bank of Richmond, August 1989. NOVEMBER/DECEMBER 1989 LABORMARKETDATA Roy H. Webb and WiL4’am Whe@‘q * This art&e is part of a series that will be pubdished by this Bank under the title Data: A User’s Guide. The book wil’l contain introductions to important series of mac?veconomic data, including p&es, employment, pmduction, and monq. The articles in the book are designed to he& the reader accurately inte?pret economic data and thereby allow the numberx to be use&i analytical tools. Macroeconomic Aggregate data on jobs, unemployment and earnings are closely watched by millions of Americans. The unemployment rate is probably the single most widely followed economic indicator. Among financial market participants, the number of people employed is perhaps the most closely followed macroeconomic statistic that appears monthly. These and other selected labor market indicators are described in this article. HISTORICAL DEVELOPMENT Statistics describing the labor market were estimated as early as 1820, based on questions from the decennial Population Census. In the last decade of the nineteenth century, the newly formed Bureau of Labor-the predecessor of the Bureau of Labor Statistics (BLS)-began to collect detailed data on wages and earnings. In 19 1.5, the Bureau began a monthly survey of employers to collect wage and employment data. This survey is still conducted, and data from it are reported on a monthly basis; it is often referred to as the establishment survey, or also as the pay& survq. After a century of collecting data on labor markets, there was surprisingly little systematic information on the extent of unemployment. When national attention focused on unemployment during the Great Depression, it was not immediately obvious how to define or to gather relevant information. In 1940 a monthly survey was designed, which is now known as the Curt Population Surwq. Information from the survey allowed an unemployment rate to be calcuWebb is a vice president and economist at the Federal Reserve Bank of Richmond; Whelpley is a principal of Whelpley Associates Inc., and was an -as&tant economist at the Federal Reserve Bank of Richmond when he contributed to this article. The authors gratefully acknowledge helpful comments from Dan M. Bechter, Timothy Q. Cook, William E. Cullison, Thomas M. Humphrey, Janice Shack-Marquez, and employees of the Bureau of Labor Statistics. lated. By 1945 the questions were developed which form the basis of the Survey used today, which is usually referred to as the household sumq. d MAJOR DATA SERIES Data From the Household Survey Each month over fifty thousand households are interviewed by the Census Bureau for the BLS as part of the household survey. The BLS then analyzes the survey results and reports its findings near the beginning of the next month, usually on the first Friday. Many statistics from this survey could be discussed; the key concepts in this section are the unemployment rate, the number of people employed, and the labor force participation rate. Unemployment rates are calculated for the entire nation and also for more narrowly defined demographic groups and geographic areas. l An unemployment rate is defined as the number of people unemployed as a percentage of the daborforce. The size of the labor force, in turn, is defined as the number of people empbyed plus those unempbyed, that is, people without jobs who are willing and able to work. All three terms, employed, unemployed, and labor force, have very specific definitions. A person is counted as unemployed if he or she did not work during the survey week and: (a) made a specific effort (which can be anything from talking to friends to interviewing for a specific opening) to find a job within the previous four weeks, and was available for work during the survey week; or (b) was waiting to be called back to a job after being laid off; or l FEDERAL RESERVE r Press reports often mention two unemployment rates. One is calculated by removing military personnel from the calculations and is slightly smaller than the overall rate. BANK OF RICHMOND 1.5 those who did not actually contact potential employers as being out of the labor force. (c) was waiting to report to a new job within 30 days of the survey. A person is defined to have been employed or she: if he (a) did any work at all as a paid employee, as a proprietor or farmer, or worked 15 hours or more as an unpaid worker in an enterprise operated by a member of the family; or (b) had a job but was not working during the survey week due to a temporary absence resulting from illness, bad weather, vacation, labormanagement disputes, or personal reasons. Employment status is not affected by whether or not pay is received during the absence, nor by whether or not another job is being sought. Finally, the labor force is simply the sum of persons who are employed plus those who are unemployed. The overallpa&$ation rate is defined as the labor force as a percentage of the population at least sixteen years of age. Participation rates are also calculated for smaller segments of the population, again defined as the labor force as a percentage of the relevant population segment. There are many reasons why a person may not be in the labor force, such as age, health, home responsibilities, being in school, not wanting to be employed, or not believing that job search would be fruitful. The latter category is referred to as discouraged WOK&X; they are counted as those who would like a job but are not looking for work for one of the following reasons listed in the household survey: “thought no jobs were available in their line of work or area.” “previously tried unsuccessfully to find work.” “lacked the necessary schooling, training, experience, or skills.” “felt employers considered the person too young or too old.” “had some other personal handicap in finding work.” One’s intuitive definitions of employment or unemployment may be somewhat different from the specific definitions given above. In particular, people who are not working vary tremendously in the amount of thought and effort spent on finding work; it is inherently arbitrary to divide people without jobs into only two categories, unemployed or not in the labor force. Some analysts would add discouraged workers to the unemployed, thereby boosting the reported unemployment rate. Others would lower the unemployment rate by defining 16 ECONOMIC REVIEW, Behuvior Over Time Chart 1 shows the unemployment rate over the post-World War II period. One notable feature is that sharp swings are associated with the business cycle, the alternating periods of expansion and recession in the whole economy. Another feature is the general upward drift for much of the chart after abstracting from business cycles. Chart 2 shows the participation rate. Especially notable is the substantial increase over the past 2.5 years. The major factor behind that increase can be seen in the table, which contains the current demographic composition of the labor force and contrasts it with the labor force in 1948 and 1969. The rapidly growing fraction of adult women in the labor force more than counteracts a decline in the fraction of men in the labor force, resulting in a growing participation rate for the whole population. The table also reveals relatively high unemployment rates for blacks and teenagers. DATA FROM THE ESTABLISHMENT SURVEY The establishment survey covers the industry, hours, and earnings of most employed members of the labor force. State agencies send survey forms to over 300,000 establishments, who then record the requested information and return the forms to the state agencies for processing. These agencies then forward the tabulated information to the BLS in Washington, D.C. Th e information is sent back and forth between the collecting agencies and participating establishments for one year; a written record of the numbers can therefore be reviewed by both the providers and collector of the information. Employment and earnings figures are classified by each worker’s characteristics, such as sex, industry, and job category. A person is counted as empkyed if he or she is on the payroll of an establishment for the pay period which includes the 12th of the month.2 This measurement excludes proprietors, unpaid volunteers, family workers, farmers and farm workers, and domestic household workers. Salaried officers of corporations, civilian government employees, and part-time workers are included, however.3 Industry hours and earningsjgures also originate in the establishment survey. Figures are presented in 2 Employees of the federal government are counted if they occupy a position as of the last day of the calendar month. 3 Employees of the Central Intelligence Agency and the National Security Agency are explicitly excluded from the survey. NOVEMBER/DECEMBER 1989 Chart UNEMPLOYMENT RATE January 1948 - September 1989 Percent 8 1 i r 6 1950 52 54 56 58 60 62 64 66 detail for Production and Related Workers in manufacturing and mining, Construction Workers, and Nonsupervisory Employees in service industries. The hours statistic reports the number of hours paid for by the employer in the current reporting period, not the number of hours actually worked. This figure therefore includes items like holidays, vacations, and sick leave. Overtime /wun includes that time for which a premium is paid. Weekend and holiday hours are included separately only if overtime premiums are paid. Hours which have only incentive premiums attached, such as shift differential and hazard premiums, are excluded from the overtime hours measurement. Average hourly and weekly earnings for nonsupervisory workers are estimated from data reported in the establishment survey. Three features have led some observers to question the relevance of that concept for studying certain problems. First, the data do not include fringe benefits, which play a major role in the compensation of most workers. Second, the data do not cover executive, administrative, and FEDERAL RESERVE 70 68 72 74 76 78 80 82 84 86 88 managerial workers in private industry, nor do they cover state and local government workers. And finally, the data are affected by changes in the composition of employment. To address those problems, the BLS also publishes a quarterly employment cost index (ECI),4 which is based on a special survey of employers. It is designed to cover all workers in private industry plus state and local government. The EC1 adds the cost of providing a wide range of fringe benefits to wage and salary payments; some of the most expensive benefits are social security and unemployment insurance taxes, paid vacation and sick leave, health and disability insurance, and retirement plans. The EC1 is also based on a fixed industry and occupational structure. Shifts between industries or occupations do not directly affect the index. 4 A more accurate title might be employee compensation index, however. Significant elements of labor cost that are not included are the costs of hiring, training, and strike activity. BANK OF RICHMOND 17 Chart 2 PARTICIPATION RATE January 1948 - September 1989 Percent 70 64 62 60 58 56 54 1950 52 54 56 58 60 62 64 66 68 Chart 3 compares the EC1 and average hourly earnings statistics. Both show a substantial decline in the growth rate of compensation since the early 198Os, as general price inflation also declined substantially. The EC1 has grown faster than average hourly earnings for much of the period, however, reflecting the growing relative importance of fringe benefits. CAUTIONS The data series described above provide a wealth of timely, relevant information. The data can be misinterpreted, however. The following cautions are designed to help place data series in perspective. The first two concern the exact meaning of widely used terms. Meaning of Terms Unemplgyment Some observers tend to equate the level of unemployment with an unambiguous measure of economic hardship. The unemployment 18 ECONOMIC REVIEW. 70 72 74 76 78 80 82 84 86 88 rate, however, is a much more complex statistic. It does not refer to an unchanging group totally composed of desperate individuals. It instead is a snapshot-a view at an instant of time-of people who are entering and leaving the labor force, and of those who are starting and ending particular jobs. Some unemployed persons find jobs quickly, others more slowly, and some people move directly from outside the labor force to employment. Some job changes are voluntary, others are involuntary.5 To help put unemployment rates in perspective, note that it is often not in the best interest of an unemployed person to take the first available job. It may take time to achieve a good match between a person’s interests, skills, and abilities on the one hand, and a job’s skill requirements, working conditions, and promotion possibilities on the other. 5 In June 1989, for example, 42 percent of the unemployed had lost their last job, 15.5 percent had quit their last job, and 42.5 percent were new entrants or reentrants into the labor force. Half were unemployed less than six weeks, while 9.1 percent were unemployed more than a half year. NOVEMBER/DECEMBER 1989 DEMOGRAPHIC COMPOSITION OF THE LABOR FORCE IN THE UNITED STATES (Thousands of persons unless otherwise Characteristic indicated) 1948 1969 1989 60,621 80,733 123,291 TOTAL Civilian Labor Force Percent of total population Employed Unemployed Unemployment MEN, 58.8 60.1 66.4 58,344 77,902 116,900 2,276 2,831 6,391 rate 3.8 3.5 5.2 AGE 20 & OVER Civilian Labor Force Percent 40,687 of adult male population Employed 39,382 Unemployed Unemployment WOMEN, 46,351 86.61 83.0 45,398 1,305 963 3.2 2.1 rate 63,468 78.1 60,642 2,827 4.5 AGE 20 & OVER Civilian Labor Force Percent 15,500 of adult female population Employed Unemployment Civilian 41.5 51,890 57.6 26,397 49,514 564 1,016 2,376 3.6 3.7 4.6 14,936 Unemployed TEENAGERS 27,413 31.3” rate (16-19) Labor Force Percent of teenage population Employed Unemployed Unemployment 4,435 6,969 52.5 49.4 55.2 4,026 6,117 6,745 409 852 1,188 9.2 12.2 15.0 rate 7,933 WHITE Civilian Labor Force Percent of white 71,778 population 58.2b Employed Unemployed Unemployment rate 3.5 of black population 64.0b 59.9 105,964 66.7 69,518 101,338 2,260 4,626 3.1 4.5 13LACKc Civilian Labor Force Percent 8,959 Employed Unemployed Unemployment Note: rate 5.9 13,444 62.1 64.4 8,384 11,898 570 1,561 6.4 11.2 Data represent the first quarter of 1989 and the full years of 1948 and 1969, and are taken from the Month/y Labor Review and the Economic Report of the President, various issues. Unless otherwise indicated, all population figures exclude military and institutionalized personnel, and young persons less than sixteen years old. a Age 14 and over. Recognizing the inevitability of such search IMtemp/oymentimplies a positive unemployment rate. In short, a normally functioning economy will have some unemployment, and every unemployed person does not experience substantial hardship.6 To provide a perspective for business cycle analysis, some economists refer to a naturalrate of unemployment, defined in one textbook7 as “that rate of unemployment at which flows in and out of unemployment just balance, and at which expectations of firms and workers as to the behavior of prices and wages are correct.” The natural rate is neither constant nor precisely known; at the present time many economists believe that it is between five and six percent in the United States. If actual unemployment were much higher, that would be evidence of cyclical slack in the economy: and if the actual rate were much lower, that would signal an overheated economy. The term “natural” is widely used but may be misleading, since there should be no presumption that the current natural rate is either optimal or immutable. The natural rate is affected by the incentives and constraints facing persons and firms; anything that affects the average frequency or duration of unemployment will also affect the natural rate. Some important factors affecting the natural rate 6 An individual’s hardship is also affected by household wealth and by whether transfer payments, such as severance pay or unemployment insurance, are received. In addition, some unemployed persons are on temporary layoff and will almost certainly be recalled; others may have accepted a job that begins in more than a month. 7 Rudiger Dornbusch and Stanley Fischer, Macroeconomics, 3rd ed. (New York: McGraw-Hill) 1984, p. 466. b Data are for 1954, not 1948. c Nonwhite before 1972. FEDERAL RESERVE BANK OF RICHMOND 19 Chart 3 CHANGES IN EMPLOYMENT COSTS 3Q 1976-2Q 1989 Percent Change fonm Year Ago 8 1977 78 79 80 81 82 83 are the unemployment insurance system, household wealth, minimum wage legislation, the demographic composition of the labor force, the mobility of labor, and the dispersion of skill levels in the labor force. The next caution involves one concept, employment, that is estimated from both the household and 20 ECONOMIC REVIEW, 85 86 87 88 89 establishment surveys. The two should move together closely in the long run; however, in any month they can diverge substantially. Compensation of EmpZoyees Many forms of compensation are ignored in the wage figures reported each month, including some that are growing especially rapidly. Fringe benefits are excluded, as are contingent payments such as lump sum payments in lieu of wage increases, bonuses, profit-sharing payments, and stock options. In addition, some benefits are not even included in the ECI. For example, medical benefits for retirees have been promised by many employers with no provision having been made for funding those costly benefits. They are thus not included in the ECI. Two Definitions of Employment 84 To see why employment totals can differ, note the slightly different definitions of employment for each survey. The establishment survey counts jobs, not people; dual job holders are therefore doublecounted. The household survey only covers the number of people employed, so that a person is never double-counted. The household survey also counts self-employed persons, agricultural workers, and household workers, all of whom are omitted from the establishment survey. Many observers may prefer to ignore monthly changes and focus on the longer run; for them it probably does not matter which series they focus on. But those with a short-run perspective often have to choose one or the other when the two series give conflicting signals. Many choose the establishment series, since its growth is more closely correlated NOVEMBER/DECEMBER 1989 with real GNP growth than is the other estimate.* Also, the number of firms surveyed is much larger than the number of households surveyed, which could in principle result in a more accurate estimate from the establishment survey. And finally, it is noted below that some analysts question the accuracy of survey responses from households. Volatile Monthly Observations Sampling Ermr- A final set of cautions warns a user not to overemphasize a single month’s data. A basic reason is sampling error-that is, statisticians are attempting to esiima~e a statistic for a large population from a relatively small survey. It is especially important as smaller segments of the labor force or smaller geographic areas are studied. As Geoffrey Moore put it: A rise, say, from 5.0 to 5.3 percent in the unemployment rate is statistically significant, whereas a rise from 9.7 to 10.4 percent in the unemployment rate for blacks is not. The reason is that the population of whites is about ten times that of blacks, so that the sample of whites is also about ten times as large. Coupled with the fact that the unemployment rate for blacks is about twice that for whites, this means that the sampling error of the unemployment rate for blacks is about four times as large as for whites.9 The key concept is that of statist;cacsign%~cance,that is, whether a result is likely to have resulted simply from chance; a statistically significant result is not likely to be due to sampling error. Moore uses a 0.2 percent change for the total unemployment rate, and a 0.8 percent change for the black unemployment rate, as thresholds for statistical significance. One should therefore be cautious in attaching much importance to a single month’s changes without having some idea of how large a change must be to be statistically significant. This caution applies more forcefully as the size of the relevant population becomes smaller. On the other hand, consistent movements for several months considerably reduce the likelihood of the fluctuations being due to chance. Also, one’s confidence in a single month’s change can be bolstered or reduced by movements in related statistics. For example, suppose that employment growth is reported to have been relatively strong but also that average weekly hours were relatively soft. In that case one could reasonably question the economic importance of the employment figure. Responses to Swwy Data Individuals responding to the household survey may respond for themselves and any other adults in the household without checking written records. Some observers have questioned the reliability of that information. It is, of course, difficult to know the exact relevance of answers to questions from any survey. One piece of evidence is a test in 1977 that compared individual responses with employer records.1° Relative to employers’ records, household respondents overstated the number of hours worked and understated both average hourly and weekly earnings. Iregular Events All the monthly data series described in this article are adjusted to remove predictable seasonal fluctuations such as the swell in Christmas employment, or the effects of summer vacations for students. Events that occur on an irregular basis can be more difficult to take into account. Strikes, for example, lower employment estimates from the establishment survey but do not directly lower employment (or raise unemployment) estimates from the household survey. And while the BLS may note an estimate for the direct effect of a strike, the indirect effects may be substantial but not estimated; an example of an indirect effect would be layoffs of railway and port workers after a coal strike reduced coal shipments. Extreme weather conditions can also affect the data, even after routine seasonal adjustment. SUGGESTIONS FOR FURTHER READING a To check the validity of that common assertion, we regressed real GNP growth on four own lags plus four lags of quarterly employment growth, from 1948 to 1989. For the household series, the R statistic was 0.36; for the payroll series it was 0.56. Since both employment statistics are subject to sampling error, it is possible that the average of the two might be better than either one individually. We therefore substituted the average of the two for the employment variable in the regression equation; the R* statistic was 0.5 1. For monitoring the overall economy, it therefore looks like the payroll series is the better choice, and that averaging the two does not improve matters. 9 Geoffery H. Moore, Business Cycles, Inflation, and Forecasting (Cambridge: Ballinger Publishing Co. for the National Bureau of Economic Research, 1980) p. 111. FEDERAL RESERVE Many books, professional journals and government reports have been written about labor market data. For an overview of labor markets and how they fit into the larger economy, readers may wish to look at a macroeconomics textbook such as Robert Barro, Macmeconomics, John Wiley and Sons; or Dornbusch 10Accounts of this test are taken from Joseph R. Antos, “Analysis of Labor Cost,” in Jack E. Triplett ed., Tfie Measurement of Labor Cost, (University of Chicago Press for the National Bureau of Economic Research, 1983) p. 162. BANK OF RICHMOND 21 &zmings summarizes current and historical statistics collected from both the household and establishment surveys. The Monthly Labor Review also summarizes labor market statistics. It also contains articles that discuss many aspects of labor markets, data concepts, data collection procedures, and the series themselves; several of the articles were helpful in preparing this paper, such as an article contrasting the payroll and household estimates of employment in the August 1989 issue. Finally, the BLS Handbook of Methods, revised and published periodically, presents a discussion of the technical aspects of how the BLS collects, transforms, estimates, and presents labor market data. and Fisher, op cit. For a more detailed analysis of labor supply and demand and market institutions, see a text on labor economics, such as Ronald G. Ehrenberg and Robert S. Smith, Modern Labor Economics, Scott Foresman and Co. A good discussion of problems in the data can be found in the report of the 2979 NahnaG Commission on Employment and Unemp.bymment Statztics. The report contains a number of background papers in addition to the summary of recommendations. The data series described in this article only hint at the large quantity of statistics that describe the labor market; many more series can be found in two monthly publications of the BLS. Employment 63 Economic Review Index Volume Federal January/February Determinants Monetary March/April Aggregates: Banking Lender of the Federal of Last Resort: America’s An Examination May/June The Future Market July/August Precursors The U.S. September/October Some David 1970 Results Fedwire Michael Data in the 1980s An Analysis Highest Daylight David Overdrafts Say B. Humphrey Thomas M. Humphrey William E. Cullison Roy H. Webb and Rob Willemse of Shift in Ml Demand in the 1980s Yash P. Mehra Robert Policy Some Provocative Employment William Findings in North Carolina Counties, 1980-85 L. Hetzel E. Cullison Christine Chmura and Jane lhrig Ching-Sheng Mao Roy H. Webb and William Whelpley Data Small Dotsey John R. Walter and Donald L. Welker in Decade Estimating Intertemporal Elasticity of Substitution: The Case of Log-Linear Restrictions Top Performing Heller Anatoli Kuprianov and David L. Mengle of the Alternatives What the Experts on the Source Labor Force: M. Humphrey H. Robert on Assets: Slowdown: L. Mengle Thomas in History Model in Manufacturing Labor Market 22 since Price Indexes M2 and Monetary November/December Return to Pricing Productivity The Changing The Fifth District Trade Insurance: of the P-Star Further Changes Rules: Cook John R. Walter Competitiveness Banks’ Macroeconomic Timothy 1979-1982 Guide of International Responses 75 Bank of Richmond Rate: The Concept of Deposit Fifth District Funds A User’s under Changing Improving Reserve 1989 Banks: ECONOMIC Making REVIEW, Money the Old-Fashioned NOVEMBER/DECEMBER 1989 Way Benton E. Gup and John R. Walter TOP PERFORMING SMALL BANKS: MAKING MONEY THE OLD-FASHIONED WAY Benton E. Gup and John R. Walter’ Introduction Average profit rates of small banks (assets less than $100 million) declined in the 1980s but about 2 percent had persistently high returns. Some have attributed persistent profits to collusion, risktaking, or chance. In contrast, this study finds that consistently profitable small banks were those that stressed basic banking, in other words, acquiring lowcost funds and making high-quality investments. Small bank average profitability declined in the 1980s for several reasons. Losses at many small banks, especially at those located in regions of the country beset with problems in the agricultural or oil industries, accounted for much of the decline. Some of the decline may have resulted from the increased competition in the retail loan and deposits markets. Federal legislation expanded the number of retail deposit products banks and thrifts could offer and deregulated interest rates on existing deposits while allowing thrifts to compete more effectively with banks for both deposits and loans. The specific acts were the Depository Institutions Deregulation and Monetary Control Act of 1980 (DIDMCA) and the Garn-St. Germain Depository Institutions Act of 1982. In this study we compare small banks having persistently high profits to all small banks over the period 1982 through 1987. We identify differences in portfolio structure, income, and expense between the two groups of banks located throughout the country. Moreover, to determine how the factors associated with high performance may have differed from regionto-region, high performers and all small banks are grouped by region and compared on a regional basis. * Table I summarizes the significant differences Gup holds the Chair of Banking at The University of Alabama; Walter is an associate economist at the Federal Reserve Bank of Richmond. The authors wish to acknowledge the unflagging efforts of Richard K. Ko in the construction of the data base for this article. l 1 The regions are shown in Table II and are the same as those used by the Federal Deposit Insurance Corporation (FDIC) in its “Quarterly Banking Profile” (1989). FEDERAL RESERVE between the average high-performance and the average small bank. small bank Theories of Persistent Profits Mueller (1986) observed that in the long run, above- and below-average profits tend to converge toward the industry norm. Competition should eliminate abnormally high profits over time. Where persistent high profits occur, as they did at the 206 high-performance banks in our study, economists offer a variety of explanations, including the following four: Co&&on It has been argued that firms can maintain high profits by agreeing explicitly or tacitly to limit their competitive behavior. Collusion becomes more difficult as the number of competitors in a market increases; that is, as market concentration declines. We would expect the number of competitors in banking markets to be larger in more populated areas. Thus, if collusion is important to profitability, high-profit banks should be found more frequently in less populated areas. In our study, we defined a populated area as any metropolitan statistical area (MSA). While our data did show that non-MSA small banks were likelier to be persistently profitable than were MSA small banks, the difference was not significant. Therefore we find no evidence that collusion may have been responsible for the strong performance of the high-profit small banks. Using different proxies for market concentration, Kwast and Rose (1982) and Wall (1985) reached the same conclusion. The consistently aboveGreater Risk- Taking normal profits produced by the 206 high-performance small banks identified in our study cannot be explained by greater risk-taking since these banks operated in a less risky manner than average for all small banks. They had fewer loan losses than their peers, indicating that they were taking less credit risk. They were less dependent on debt financing because of stronger equity-to-assets ratios. Finally, they limited their credit and liquidity risks by holding more securities than did their peer group. BANK OF RICHMOND 23 Table I SUMMARY OF MAJOR FINDINGS OF STUDY SIGNIFICANT DIFFERENCES BETWEEN HIGH-PERFORMANCE SMALL BANKS AND ALL SMALL BANKS: High-Performance Small Banks vs. All Small Banks Area of Difference I Interest Higher Income/Total Assets High-performance small banks produced significantly more interest than the average for small banks while bearing less credit risk income relative to assets Loans/Total Assets The high-performance small banks had a significantly lower ratio of loans to total assets than the average small bank, meaning that they bore less credit risk since loans generally are more risky than the other major category of assets held by banks-securities Lower Securities/Total Assets Higher ratio at high-performance Higher banks indicating Municipal Securities/Total Securities High-performance banks had more income advantage of municipals Earning Interest Assets/Total lower credit risk Higher to shelter so they made greater Higher Assets Expense/Total Assets High-performance banks funded themselves at lower cost by emphasizing structure and a conservative capital structure Demand Deposit/Total High-performance Noninterest Expense/Total High-performance resources Assets/Employees High-performance Salaries/Employees High-performance use of the most traditional liability of funding sources Lower retail deposits to gather funds Higher banks had a stronger or more conservative Assets banks held these expenses capital structure Lower to a lower level indicating a more efficient use of Higher banks required fewer employees per million dollars in assets Higher banks’ employees Loan Loss Provisions/Total Assets High-performance banks limited restraining their credit risk Loan Charge-Offs/Total Loans Lending to high-quality borrowers Nonperforming Loans/Total Loans Lending to high-quality borrowers their books FACTORS NOT SHOWING AND ALL SMALL BANKS: Lower a traditional Higher Liabilities banks made greater Interest Expense/Interest-Bearing Liabilities High-performance banks made greater use of low-cost Capital/Total Assets High-performance use of the tax SIGNIFICANT were better paid Lower their lending and only lent to high-quality borrowersLower meant fewer loan charge-offs meant high-performance at high-performance banks Lower DIFFERENCES BETWEEN banks carried fewer bad loans on HIGH-PERFORMANCE SMALL BANKS Location in a Metropolitan Area Bank Holding Company Affiliation Loan Income/Total Loans Securities Income/Total Securities Loan Portfolio Composition Loan Maturity Noninterest Income/Total Assets High-performance small banks placed income than the average small bank Fee Income/Total Assets Gains or Losses on Securities/Total 24 no more emphasis on these less traditional Assets ECONOMIC REVIEW, NOVEMBER/DECEMBER 1989 sources of Table II SMALL BANKS BY GEOGRAPHIC REGION, 1987” All Banks Regionc Northeast Number 377 High-Performance Number As a Percent of All Small Banks Bank@ As a Percent of All HighPerformance Banks 25 6.6 12.1 Southeast 1,196 54 4.5 26.2 Central 2,290 44 1.9 21.4 Midwest 2,841 34 1.2 16.5 Southwest 1,909 33 1.7 16.0 880 16 1.8 7.8 West Total Average 9,493 206 100.0 random.“2 According to this theory the highperformance banks in this study may have selected, by chance, the management, investment, and lending policies that turned out to be very profitable during the 1980s. To test if this was so, the average ROA for the 206 highperformance small banks and all small banks were calculated for each year between 1970 and 198 1. The average for the high-performers was considerably above the average for all small banks for each of the twelve years, indicating that the high performers of the 1980s produced supernormal profits during the 1970s as well. Chance alone is an unlikely explanation of almost two decades of persistently high profits. Prior Empirical Research 2.2 Several other analysts have attempted to pinpoint factors associated with bank profitability. a Small banks are those with end-of-year assets of $100 million or less that were A study of bank profitability in the 1970s by opened on or before December 31, 1982. Kwast and Rose (1982) included large banks b High-performance small banks have ROAs of 1.5 percent or more for all years, 1982-87. from throughout the nation. The authors determined that neither pricing, operating costs, c For regions, see map below. market concentration, or macroeconomic effects were responsible for the higher earnings of some banks. They hypothesized, instead, that differences in regional factors, portfolio make-up, or managerial abilities must explain the better earnings of high-performance banks. Wall (1985) examined small and mid-sized banks over the period 1972 to 1981 to identify factors important to bank profits. Wall found that consistently profitable banks had lower interest and noninterest expenses than did their less profitable counterparts because of more capital, more demand deposits, slightly lower rates paid on liabilities overall, greater holdings of securities, and more efficient management. Wall concluded that interest and noninterest income at consistently profitable banks was no Unique Quah2ie.c These include leadership in the higher than at less profitable banks, and that asset market, provision of services other firms cannot size, number of branches, and market concentration duplicate, having the dominant market share, or did not explain higher earnings. Wall’s findings on being first to arrive in the market. Perhaps one or the factors associated with small and mid-sized bank more of these apply to the high-performance banks. profits in the 1972 through 1981 period differ little from our findings for small banks in the 1980s. S@&z.s~ Pmcess Persistent profits may result from historical chance. The basic idea of the stochastic process, as explained by Alchian, is that “where there is uncertainty, people’s judgments and opinions, even when based on the best available evidence, will differ; no one of them may be making his choice by tossing coins; yet the aggregate set of actions of the entire group of participants may be indistinguishable from a set of individual actions, each selected at Methodology Data for our study came from the Reports of Condition and Income (call report), a detailed financial * Alchian (1950), p. 216. Alchian is an excellent background source for understanding the issues involved in stochastic growth. Also see Nelson and Winter (1982) and Steindl (1965). FEDERAL RESERVE BANK OF RICHMOND 25 year and had been established in 1982 or before.3 The number of banks in this category declined each year, from 12,353 in 1982 to 9,493 in 1987 as the banks grew in asset size, merged, or failed. To be included in the high-performance subset a bank must have had no more than $100 million in assets and must have produced a return on assets (ROA) greater than 1.5 percent for each of the six years from 1982 through 1987. Banks with ROAs greater than 1.5 percent have very strong profits. Banks established after 1982 could not have had high ROA in that year, so are excluded from the high-performance group by our convention that requires high ROA in every year. There are 206 high-performance banks. They are listed in Table IA in the appendix. The period 1982-87 is used in this study for two reasons. First, it offers the most recent extended period since the passage of DIDMCA and the GarnSt. Germain Act. Second, it provides an interval long enough to be sure that luck or accounting choices alone did not influence the selection of the highperformance small banks. statement filed quarterly by banks with their regulators. A set of income, expense, and portfolio ratios were calculated for all small U.S. banks established in 1982 or before. Ratios were then averaged across all small banks and all high-performance small banks throughout the nation for each year from 1982 through 1987. Because economic conditions varied from region to region, ratios for both groups of banks were also computed on a regional basis. For each of the six years, the average ratios, regional and national, for high-performance small banks and all small banks were compared using a standard t test to determine statistically significant differences (see Table III). A difference between the ratios of high-performance small banks and all small banks is considered to be due to factors other than chance if the t statistic is significant at the 5 percent level. Regional patterns in the ratios are identified and discussed. The same banks are included in the highperformance group for each year of the study while the number of banks in the all-small-banks category varies. The all-small-banks category, for any given year, includes all banks throughout the nation that had assets less than $100 million at the end of that 3 Unless otherwise stated, the phrase al’lstnallbanks or average smab’ bank should be assumed to include only those banks meeting these two requirements. Table III COMPARISON OF SELECTED RATIOS: HIGH-PERFORMANCE 1982 NE 1983 SE CN MW SW P N P N 1 2N W U.S. NE P PPPPP NNNNNNN P N P N 1984 CN MW SW W U.S. P N NNN N N NN SE CN MW SW W U.S. na na na na na na na NNNNNNN N NNNNNNN 5NNNNNNN N N N N N N N N N N N N 6 7PPPPPPP 8 NNNNNN 9PPPPPPP 1OPPPPP P 11NNNNN N 1 2 Interest Interest 3 4 Noninterest Noninterest Loan Loss Provision/Assets Securities 7 Return P P P P P P PPPPPPP PPPPP P N P N P N P N P N P P P P P P N N 8 P 9 P 10 P N N Income/Assets Expense/Assets 5 P NN Income/Assets Expense/Assets 6 PPPPPPP NNNNNN 11 Gains/Assets on Assets Loans/Assets Securities/Assets Equity/Assets Total Assets that data were not available. P indicates that the mean for the ratio for the high-performance different at the 1 percent level. P indicates NE N 3N 4N na indicates SE BANKS VERSUS ALL SMALL BANKS that the mean for the ratio for the h.p.s.b. Blank space indicates that there was no significant small banks (h.p.s.b.) exceeded difference exceeded that for all small banks and was statistically that for all small banks and was statistically between h.p.s.b. significantly significantly different at the 5 percent level. and all small banks for the ratio. N indicates that the mean for the ratio for all small banks exceeded that for the h.p.s.b. and was statistically significantly different at the 1 percent level. N indicates that the mean for the ratio for all small banks exceeded that for the h.p.s.b. and was statistically significantly different at the 5 percent level. SEE TABLE IIA IN APPENDIX FOR RATIO AND T STATISTIC VALUES. 26 ECONOMIC REVIEW, NOVEMBER/DECEMBER 1989 Characteristics of High-Performance Small Banks compared for the nation. When tested by region and across years, only in the Southwest were highperformance small banks significantly less likely to be located in MSAs. The asset size of the average high-performance small bank was $40.8 million in 1987 compared with $37.5 million for all small banks. Asset size of the average high-performance small bank increased by 56 percent from 1982 through 1987, while the asset size of the average small bank increased by only 20 percent. The percentage of high-performance and all small banks that were subsidiaries of bank holding companies (BHCs) increased through the period. In 1987, 46 percent of high-performance and 66 percent of all small banks were subsidiaries of BHCs. A test was performed to determine if the difference in BHC affiliation between the two groups of banks was statistically significant across the years. For the nation as a whole the difference was significant, but statistically significant regional differences were not found except in the Northeast and Southwest regions. Firm conclusions about the relationship between BHC ownership and profits based on these data are difficult to draw. Table II shows that high-performance small banks were not distributed proportionately throughout the country. The Northeast had the highest, and the Midwest the lowest, proportion of high-performance small banks relative to all small banks. During the 1982 through 1987 period, there were substantial differences in regional economic performance which likely caused some of the corresponding regional differences in the proportion of high-performance small banks. Slumping prices for energy, real estate, and farm commodities had adverse effects on the Southwest, Midwest, and Central regions, while strong economic growth was occurring in the Northeast and Southeast through the period. Although not shown in Table II, approximately 30 percent of high-performance small banks were headquartered in or near large population centers, represented here by metropolitan statistical areas (MSAs), while the figure averaged a slightly higher 33 percent for all small banks. Only in 1982 and 1983 were the differences statistically significant when small banks, high-performance versus total, were 1985 NE SE 1986 CN MW SW W U.S. P P 1PPP 2NNNNNNN NE SE 1987 CN MW SW W U.S. NE SE CN MW SW PPPPPPP PPPPP NNNNNNN NNNNNNN W U.S. Pl 2 3 4N 3 N 5NNNNNNN 6 N 7PPPPPPP NNN N 8 NNNNNN 9PPPPPPP 1OPPPPP 11 P N N NNNNNNN N NNN N N N N NNNNNNN NN 4 5 6 N PPPPPPP PPPPPPP NNNNN PPPPPPP PPPPPPP NNNNN PPPPPPP PPPPPPPlO N 7 N8 9 11 FEDERAL RESERVE BANK OF RICHMOND 27 How The High Performers Did It ROA OF SMALL BANKS The high-performance small banks identified in this study differed from the average small bank in several ways. They depended more on low-cost demand deposits, invested more in securities (especially long-term and municipal securities), made more highquality loans, and were more highly capitalized. As a result, the high-performance small banks produced higher interest income, lower interest expense, lower noninterest expenses, and lower provision for loan losses than did the average small bank. The highperformance small banks did not differ significantly from the average small bank in interest income from loans and securities, in loan portfolio makeup, in noninterest income, or in income from securities gains. There was little variation among regions in how the high-performance small banks operated. As shown in the chart, average ROA for the 206 high performers exceeded 2 percent in every year and was fairly stable, while average ROA for all small banks declined in every year except 1987 and ended the period at .51 percent. In~emstIncome Except for one or two years’ observations for three regions, high-performance small banks produced significantly more tax-equivalent interest income relative to assets than the average for all small banks (see Table III, line l).4 Among the major categories of income and expense, higher interest income was second only to lower interest expense as a contributor to the earnings differential of the high-performance banks across the years and regions of the study. Averaged for the six years of the study, high-performance small banks’ interest income relative to assets was 58 basis points higher than the average small banks. Wall (1985) found that higher interest income was not associated with higher profits for small and medium-sized banks between 1972 and 198 1. Greater pressure on interest expense resulting from deregulation in the early 1980s of rates paid on deposits may have made interest income more important to profitability for our study period. Interest income relative to assets depends on the earnings per dollar of the various types of interest- 4 The interest income on most securities issued by local and state governments is exempt from federal income taxes. These securities, therefore, pay lower rates of interest than taxable securities of equivalent risk and maturity. To put the tax-exempt income on a basis comparable to the pretax return on taxable securities, or on a tax-equivalent basis, an adjustment is made to income from state and local securities. For banks with positive profits before taxes, income from state and local securities is increased by t/( 1 -t) times the lesser of profits before taxes or interest earned on state and local securities, where t is the bank’s marginal federal tax rate. 28 ECONOMIC REVIEW. Net Income/Total Assets Percent High Performers 1.8 1.6 1.4 1.2 1982 83 84 8.5 86 87 earning assets, their proportions in the asset portfolio, and the proportion of nonearning assets to all assets. LOANS The difference between loan income relative to total loans at the high-performance small banks and at the average small bank was not significant for most regions across years or for the national average except in 1982 and 1983. As shown on line 8 of Table III, the ratio of total loans to total assets was significantly lower for high performers than for all small banks. In the Southwest and Midwest where agriculture and oil industry problems were prevalent, the high performers eschewed lending, especially in the later years of the study. While at the national level the high-performance small banks differed statistically from the average of all small banks in terms of loan composition, the regional data do not corroborate this finding. The high performers in the West and Midwest made fewer commercial and industrial loans than average for small banks in those regions and high-performance small banks in the Southeast made more loans to individuals than average for small banks in that region. Other regions show no consistent differences in portfolio makeup. There was no difference in the maturities of loans made by high performers and all small banks. SECURITIES High-performance small banks had a much higher ratio of securities to total assets than did all small banks (Table III, line 9). The difference was statistically significant across all regions and all years in the study. High-performance banks also had more municipal securities than their counterparts, accounting for most, but not all, of the higher NOVEMBER/DECEMBER 1989 securities-to-assets ratios of high-performance banks. Municipal securities are generally tax-exempt and pay tax-adjusted rates comparable to other securities only for those holders with high marginal tax rates. As a bank’s net income increases, its ability to make use of the tax-free income these securities generate increases. Accordingly, high-income banks would be expected to hold more municipal securities than less profitable banks. At the national level the ratio of taxable securities to total assets was higher at the high-performance small banks than at the average small bank for the years 1982 through 1984 only. On a regional basis, the difference was consistently significant only for the Southwest, probably because of the lack of good lending opportunities in depressed oil industry areas of the region. On average the high-performance banks generally had more securities with maturities greater than one year than did their counterparts. The difference was significant for the nation across all years but only consistently different for three of the regions in all the years. High-performance small banks did not consistently earn more on securities than did all small banks. Securities income relative to total securities was significantly greater at the high-performance small banks than at the average small bank in some years but not in others at the national level and varied from region to region across the years. In addition, there was no significant difference between securities gains and losses relative to assets between high-performance small banks and all small banks (Table III, line 6). Securities gains or losses are realized when a bank sells a security, prior to the maturity of the security, for a price different than that paid to purchase it5 EARNING ASSETS-TO-TOTAL ASSETS The national average proportion of earning assets-to-total assets at high-performance small banks was 9 1.4 percent in 1987 compared with 90.4 percent at the average small bank. High-performance small banks’ earning assets-to-total assets ratio exceeded the average small banks’ ratio significantly in every year from 1982 through 1987 at the national level and for most regions across the years. This accounts for some of the higher interest income relative to assets of the high performers. Examples of nonearning assets are cash, and foreclosed real buildings, equipment, estate. 5 For additional information on the relationship between market rates of interest and securities prices see Gup, Fraser, and Kolari (1989), Chapters 2 and 5. FEDERAL RESERVE Interest lCq!mse Interest expense relative to assets in 1987 was 3.9 percent for the average of all highperformance small banks in the nation and 4.6 percent for the average of all small banks. The difference was significant across all regions and years with the exception of the Southwest and West regions in 1982 (Table III, line 2). Among the major income and expense categories, interest expense was the largest contributor to higher ROA at the high-performance banks. Interest expense relative to assets depends on the proportion of liabilities that are interestpaying, the rates paid on the interest-paying liabilities, and the level of the capital-to-assets ratio. DEMAND DEPOSITS TO TOTAL LIABILITIESThe major liability not paying interest is demand deposits. The high-performance small banks had a lower level of interest expense relative to assets than the average small bank, in part because they had more demand deposits. The difference between the ratio of demand deposits to total liabilities for high-performance small banks and that of the average small bank was significant in all years for the nation and for varying regions across the years. RATES PAID ON INTEREST-BEARINGLIABILITIES Interest expense relative to interest-paying liabilities was lower at the high-performance small banks than at the average small bank. The difference was significant across most regions and at the national level for all six years and accounted for one-third to one-fourth of the total difference in interest expense relative to assets. For the national average, the highperformance banks were able to gather a higher proportion of their liabilities from passbook and statement savings, normally the least costly of the interestbearing liabilities, and were less dependent on expensive large certificates of deposit (CDs) than average for all small banks throughout the nation. Again, the regional data are not consistent in their support of this finding. High performers made greater use of savings only in the Northeast and Central regions and lower use of large CDs in only the Southwest and West regions. Other regions show no consistent patterns. CAPITAL-TO-ASSETSRATIO The average highperformance small bank had a significantly greater equity-to-assets ratio than the average for all small banks (Table III, line 10). That is, the highperformance banks had more capital than did their counterparts. The difference was significant across all regions in all years except for the West and was significant at the national level for all years. Since equity funds do not pay interest, they do not add to interest expenses, so that higher ratios of equityBANK OF RICHMOND 29 to-assets tended to lower interest expense-to-assets ratios. Because one method of increasing equity is to retain earnings, banks that maintain consistently high-earnings can be expected to have more capital than the average bank. Nonihmst Income and Expetise With the exception of the Northeast region in 1982 and 1983, noninterest income from fees and other sources was never, in the period under study, significantly different at the high performers than at small banks in general (Table III, line 3). High-performance small banks apparently did not make fee income a priority. The high-performance banks had lower noninterest expense relative to assets than did their counterparts except in the Southeast and Midwest regions (Table III, line 4). Relative to assets, the difference averaged 37 basis points for the 1982-87 period. Noninterest expense includes salaries expense, bank premises and fixed asset expenses, and a category reported on the call report as “other noninterest expense, ” including legal fees, deposit insurance fees, advertising expenses, management fees paid to parent BHCs, and other expenses. Bank premises and fixed assets expenses and other noninterest expenses were significantly lower at high-performance small banks, though salaries expense was not. Assets per employee also were higher at high-performance banks. However, higher average salaries at those banks made salaries relative to assets about the same as at the typical small bank. A lower noninterest expense-to-assets ratio could indicate more efficient management. But it is difficult to tell simply from call report data what, if anything, was being managed more efficiently. As mentioned previously, a smaller percentage of high-performance small banks were BHC subsidiaries than was the case for all small banks. Since management fees paid to parent BHCs are an expense faced only by BHC subsidiaries, banks not owned by BHCs might tend to show up more frequently in the high-performance group. Management fees are included in other noninterest expenses on the call report. Small BHC subsidiary banks had only a five basis points higher other noninterest expense in 1987 than did small banks without a holding company affiliation. This difference is so small it is not likely to have biased the selection of high-performance small banks in favor of non-BHC banks. Ptiion&r Loan Losses For every region in every year and for the national averages for every year, provision for loan losses relative to assets was significantly lower at high-performance small banks than at the average small bank (Table III, line 5). Provision for loan losses relative to assets was, on average 30 ECONOMIC REVIEW, for the six years of the study, 49 basis points lower at the high-performance banks. By substituting investments in securities for lending, that is, by holding fewer loans relative to assets, the high-performance banks decreased the proportion of the asset portfolio subject to credit risk and therefore lowered their level of loan losses relative to assets. In addition, the highperformance banks made higher quality loans. They had significantly fewer charge-offs and nonperforming loans relative to total loans than other banks, suggesting that the high performers lent to low-risk borrowers. While many small banks in depressed regions were having serious problems with their loan portfolios, some banks in those same regions were able to prosper. For example, 20 of the 206 highperformance small banks were located in Texas, where many banks were having trouble producing profits. As of 1987, there were 1,066 small banks in Texas, so that 1.9 percent were high-performance, close to the national average. Conclusion While the average small bank’s profits were fairly low and falling for most of the 1982 through 1987 period, there were 206 banks, out of 9,493 small banks (assets of $100 million or less) operating in 1987, that had a return on assets of 1.5 percent or more in each of those six years. Although there were fewer high-performance small banks in geographic regions that had economic difficulties, highperformance banks were found in all regions. Highperformance small banks seemed to choose similar strategies in all regions. The high-performance banks did not engage in exotic financial activities. Instead, they did a very good job of basic banking-acquiring funds at low cost and making high-quality, profitable investments. Wall (1985) found much the same for the 1972 through 1981 period. Our study provides evidence that the deregulation of the early 1980s did not change the methods for producing profits at small banks. The high-performance small banks earned abnormally high returns for long periods. On the contrary, economic theory suggests that abnormally high profits should be short-lived. Other banks, seeking higher returns, will engage in similar activities and drive down returns to the industry norms. The highperformance banks we studied were able to maintain persistent profits in the face of competition. Importantly, the high-performance banks were able to acquire funds at lower cost than their competition through demand and other low-cost deposits. How they were able to attract these deposits in the face of competition is a subject that deserves further research. NOVEMBER/DECEMBER 1989 References Alchian, Armen A. “Uncertainty, Evolution, and Economic Theory.” J&anal of Political Economy 58 (June 1950): 2 1 l-2 1. Barry, Lynn M. “A Review of the Eighth District’s Banking Economy in 1986.” Federal Reserve Bank of St. Louis R~Y.&w 69 (April 1987): 16-21. Clair, Robert T. “Financial Strategies of Top-Performance Banks in the Eleventh District.” Federal Reserve Bank of Dallas Economic Revim (January 1987), pp. 1-13. . “Profitability Differences Among Large Commercial Banks During the 1970s.” Th Magazine of Bauk Administration (September 1983), pp. 54, 56, 58, 62. Mester, Loretta J. “Owners Versus Managers: Who Controls the Bank?” Federal Reserve Bank of Philadelphia Business Rev&~ (May/June 1989), pp. 13-22. Mueller, Dennis C. Pn$ts in tke Lang Run. New York: Cambridge University Press, 1986, Chap. 2. Cook, Timothy Q., and Timothy D. Rowe, eds. Zn.srruments of tke Money Market, 6th ed. Richmond: Federal Reserve Bank of Richmond, 1986. Nelson, Richard R., and Sidney G. Winter. An Evohtionary Tkeoty of Economic Chnge. Cambridge: Harvard University Press, 1982. Federal Deposit Insurance Corporation. “The FDIC Quarterly Banking Profile.” First Quarter, 1989. Scott, William L., and Gloria Shatto. “Social Efficiency in Banking: A Stochastic Model of Bank Growth.” Quaflw& Review of.&onomah andBusiness 14 (Autumn 1974): 85-93. Fraser, Donald R., and James W. Kolari. Th Future of Small Banks in a Deregulated Environment. Cambridge: Ballinger Publishing Company, 1985. Steindl, Josef. Random Pnxesses and The Ghwtk of Finns. New York: Hafner Publishing Company, 1965. Geroski, Paul A., and Alexis Jacquemin. “The Persistence of Profits: A European Comparison.” Tke Economic Journal 98 (June 1988): 375-89. Gup, Benton E., Donald R. Fraser, and James W. Kolari. Commercial Bank Management, New York: John Wiley & Sons, 1989, Chap. 2. Wall, Larry. “Why Are Some Banks More Profitable Than Others?” Journalof Bank Reseat& 15 (Winter 1985): 240-56. Walter, John R., and David L. Mengle. “A Review of Bank Performance in the Fifth District, 1986.” Federal Reserve Bank of Richmond Economic Review 73 (July/August 1987): 24-36. Gup, Benton E., and John R. Walter. “Profitable Large Banks: The Key to their Success.” MidZand Corporate Finance Journal 5 (Winter 1988): 24-29. Watro, Paul R. “Have the Characteristics of High-Earning Banks Changed? Evidence from Ohio.” Federal Reserve Bank of Cleveland Economic Commentary, September 1, 1989. Kwast, Myron L., and John T. Rose. “Pricing, Operating Efficiency, and Profitability Among Large Commercial Banks.” Journal of Banking and Finance 6 (June 1982): 233-54. Whalen, Gary. “Concentration and Profitability in Non-MSA Banking Markets.” Federal Reserve Bank of Cleveland Economic Rewiew (First Quarter 1987), pp. 2-9. FEDERAL RESERVE BANK OF RICHMOND 31 APPENDIX Table HIGH-PERFORMANCE City Bank Brunswick Bank & Trust Co. state Manalapan TWP IA SMALL BANKS Bank City state NJ First National Sylacauga AL Maywood NJ National Trust Co. of Ft. Myers Ft. Myers FL Carmel NY Peoples Bank of Graceville Graceville FL Coxsackie NY Peoples State Bank Groveland FL Dryden NY Springfield Springfield FL Florida NY Capital Hermon NY Wilcox County State Bank Bank of Millbrook Millbrook NY Braselton National Stamford NY Bank of Camilla Community Putnam Bank of Bergen City County National National Bank of Carmel Bank of Coxsackie First National National Bank of Dryden Bank of Florida First National Bank of Hermon Bank of Stamford Commercial City Second Banking First National Bank of Wyoming Wyoming DE First National First National Bank of Tuckahoe Tuckahoe NJ Merchants Ashland PA Commercial Citizens National East Prospect Citizens Bank of Ashland State National New Tripoli Union Bank Bank of Lansford National Bank Bank & Trust Co. Summit Hill Trust Co. Guaranty Harlan Deposit National Bank County Bank Jackson First State Bank Farmers Baltic & Trades State Bank Bank Custar State Bank Co. Corn City State Bank City Banking Farmers National Farmers Valley Co. Bank National Bank Peoples National National Capital Centreville Caroline Bank of Rural Valley Bank of Washington National County Bank of Southern New Windsor Bank of Maryland Bank Maryland State Bank Co. Bank of Polk County & Farmers FL Abbeville GA Braselton GA Camilla GA Cedar-town GA Comer GA Crawford GA Bank of Danielsville Danielsville GA Darien Darien GA New Tripoli PA Fairburn Banking Fairburn GA Pottsvi I le PA Citizens Bank Folkston GA PA Bank of Hazlehurst Hazlehurst GA Cumberland KY Hinesville Bank Hinesville GA Harlan KY Wilkinson County Irwinton GA La Fayette GA Hill Bank Co. Bank McKee KY Bank of La Fayette Manchester KY Farmers & Merchants Mt. Olivet KY Security State Bank Baltic OH Pembroke Custar OH First State Bank State Bank Bank OH Farmers & Merchants OH Bank of Thomson Plain City OH West Union Freeport Bank Lakeland GA McRae GA Pembroke GA Stockbridge GA Summerville GA Darby Bank & Trust Co. Thomson Vidalia GA GA OH First National Bank of West Point West Point GA PA First National Bank in Deridder Deridder LA Rural Valley PA Bank of Sunset Washington DC Citizens Centreville MD Abingdon Greensboro MD First Trust & Savings La Plata MD Algonquin City & Trust Co. Bank & Trust Co. of Grainger Bank & Trust Co. MD District MD Irving Bank Bank of Currituck Moyock NC National Avery County Newland NC First National Heath SC First Bank & Trust Co. Springs Bank State Bank New Windsor Springs Tallahassee PA Ocean City Bank of Heath Bank Bank Bank of Ocean City Bank Bank PA Junction Bank of Plain City National Lansford Deshler Junction Bank East Prospect Summit Bank Bank in Sylacauga National Bank of Chicago Bank of N. Evanston Bank of Fairmount Co. Sunset LA Rutledge TN Abingdon IL Albany IL Algonquin IL Chicago IL Chicago IL Evanston IL Fairmount IL Palatine IL Latta Bank & Trust Co. Latta SC Reynolds Dorn Banking McCormick SC First National Bank of Ridgeway Ridgeway SC Tiskilwa State Bank Tiskilwa Bank of York York SC Vermont State Bank Vermont IL Middleburg VA Auburn Auburn IN Middleburg Co. National First & Citizens Bank Bank State Bank Bank of Schiller State Bank Reynolds Park Schiller IL Park IL IL Monterey VA Rockville Rockville IN Tazewell VA Iowa State Bank Calmar IA Bank of Waverly Waverly VA Ossian State Bank Ossian IA Farmers Windsor VA Palmer Palmer IA Hamlin WV Home State Bank Royal IA Northfork WV Solon State Bank Solon IA Rainelle WV State Bank of Hesperia Hesperia Ml Cleveland WI Tazewell Lincoln National Bank Bank National Bank of Hamlin First Clark National Bank of Northfork First State Bank & Trust Co. Western Greenbrier National Bank National State Bank Bank Rainelle WV Cleveland Bank of War War WV Citizens Bank Delavan WI Citizens Fayette AL Kilbourn State Bank Milwaukee WI State Bank Bank First National Bank of Fayette State Bank Fayette AL Palmyra Peoples Bank of Greensboro Greensboro AL Sharon State Peoples Bank Red Level AL Bank of South Wayne 32 ECONOMIC REVIEW, NOVEMBER/DECEMBER Bank 1989 Palmyra WI Sharon WI South Wayne WI City Bank Stoughton State Bank First National Farmers Bank of Altheimer & Merchants Leachville Bank State Bank Smackover Egyptian State Bank state Stoughton WI Citizens Smithville MO Altheimer AR Ashton State Bank Ashton NE Des Arc AR State Bank of Du Bois Du Bois NE Leachville AR First National Bank of Friend Friend NE First National Bank of Hooper Hooper NE Randolph NE NE Smackover State Bank Carriers AR Mills IL Bank & Trust Co. First State Bank Bank of Christopher Christopher IL State Bank of Riverdale Riverdale State Bank of Farina Farina IL State Bank of Table Table First National Staunton IL Bank of Talmage Fort Knox KY First National Fredonia KY American Poole KY Bank of Locust Grove Locust Grove OK Sacramento KY Park State Bank Nicoma OK Shepherdsville KY First National Pryor OK luka MS Vian State Bank Vian OK Bank of Okolona Okolona MS Farmers State Bank First National Pontotoc MS Western Commerce Water Valley MS Citizens Bank Dexter MO First National Bank of Wellsville Wellsville MO Farmers First Bank of Coon Rapids Coon Rapids MN Farmers Lester Prairie Maplewood Bank of Staunton Fort Knox National Fredonia Valley Poole Deposit Sacramento Peoples Bank Bank Bank Deposit Bank Bank luka Guaranty Mechanics Citizens Bank Bank of Pontotoc Savings Bank Bank State Bank Town & Country Farmers Bank-Maplewood State Bank First WE Savings Northern Bank of St. Louis Park State Bank Bank of West Point Exchange Bank Bank of Pryor Bank Bank of Albany State Bank TX TX Rothsay MN Dilley State Bank Dilley TX St. Louis Park MN First National Falfurrias TX Thief MN First State Bank Frankston TX Hebronville TX Hidalgo TX Hillsboro TX TX River Falls Bank in Falfurrias Forman ND Citizens National Stock Growers Bank Napoleon ND Industry State Bank First Western Wall SD Muenster Durand WI First National WI co Omnibank Haxtun Southeast Community Bank State Bank of Wiley Fort Riley National Miners Gypsum Bank State Bank Valley Bank First National Bank of Howard Bank of Hebronville Industry Muenster TX TX First State Bank Premont TX Peoples State Bank Rocksprings TX co Citizens Bank Rusk TX Denver co First State Bank Rusk TX Denver co Eisenhower Haxtun co First State Bank Wiley co First National Fort Riley KS Bank of Montreal Frontenac KS Gypsum Boulder County Commerce City State Bank Bank of Odonnell San Antonio TX Three Rivers TX Coachella CA San Francisco CA First Bank of San Luis Obispo San Luis Obispo CA KS Torrance Torrance CA Howard KS First National KS Pioneer Trust Co. EIY Salem OR Kaysville UT Morgan UT National State Bank Moundridge Farmers State Bank Winona KS Barnes Banking Leeton MO First National Bank Bank in Coachella California National Citizens Bank of Leeton Bank of Hillsboro Odonnell Ladysmith Bank & Trust Co. TX Devine County Bank Century NM Albany Columbus Valley State Bank Sargent State Bank Tucumcari Medina First National Metropolitan NM First State Bank Border Bank NA WY Carlsbad MN MT of Gunbarrel Pine Bluffs MN MN State Bank Park TX Conrad Firstbank OK TX Warren Security NE Lindsay Big Sandy State Bank Bank of Durand West Point First State Bank State Bank Bank NE NE Bertram Farmers National Rock Talmage Peoples Security Rock Bank Bank of Ely Co. Bank of Morgan FEDERAL RESERVE BANK OF RICHMOND NV 33 Table IIA NORTHEAST Higha 1982 INTEREST INCOME/ASSETSd INTEREST EXPENSE/ASSETS NONINTEREST INCOME/ASSETS NONINTEREST EXPENSES/ASSETS LOAN LOSS PROVlASSETS SEC. GAINS/ASSETS RETURN ON ASSETS LOANS/ASSETS SECURITIES/ASSETS EQUITY/ASSET TOTAL ASSETS (000) 11.22 5.25 0.36 2.64 0.14 -0.09 1.94 48.01 36.45 12.96 $31,892 SOUTHEAST Allb 11.14 6.18 0.70 3.59 0.28 -0.01 1.00 50.58 27.92 9.11 $41,903 T St& C.29) c-2.92)**’ t-2.37)** (-4.69)**” y-;-a;;;*** (14.85)*** (- 1.04) (3.42)**’ (5.34)‘*’ (-2.79)“’ CENTRAL High All T Stat High All 12.20 5.92 1.52 3.63 0.18 -0.05 2.26 36.76 43.85 12.97 11.68 6.82 0.78 3.53 0.45 -0.02 0.93 47.48 31.00 9.48 (3.91)‘*’ ( -(;:4;; * * * 11.87 6.18 0.55 2.55 0.10 -0.07 2.06 37.22 45.91 12.50 11.39 7.13 0.50 2.95 0.35 0.00 0.85 48.17 32.17 8.79 $27,044 $33,149 L 18) t-7.93)*** (-1.30) (8.59)*** t-6.27)*** (6.99)“’ (5.00)‘*’ c-3.011*** $26,250 $33,173 T Stat (2.!7)*** C-4.05)“” l.36) C-2.16)** -8.55)*** (- - 1.36) (15.45)**’ -6.12)*** (6.99)“’ (6.27)*” -2.77)“’ 1983 INTEREST INCOME/ASSETS” INTEREST EXPENSE/ASSETS NONINTEREST INCOME/ASSETS NONINTEREST EXPENSES/ASSETS LOAN LOSS PROVlASSETS SEC. GAINS/ASSETS RETURN ON ASSETS LOANS/ASSETS SECURITIES/ASSETS EQUITY/ASSET TOTAL ASSETS (000) 10.71 4.63 0.39 2.63 0.09 0.00 2.06 46.63 38.02 13.26 $35,496 %80 0.51 3.25 0.23 0.01 1.04 49.78 31.08 8.85 $45,107 (2.02)** c-3.45)**’ (-2.44j** t-3.88)**’ t-4.66)*** (- .36) (lO.BO)*‘* I- 1.26) (2.66)*** (5.95)“’ (- 1.91) 11.38 5.05 1.57 3.70 0.24 0.02 2.22 36.01 ;S:t: $29,973 10.62 5.85 0.77 3.43 0.52 0.00 0.88 47.03 33.57 9.02 $35,578 (7.16)*** (-5.03j”* C.99) C.38) C-5.83)*** L99) uo.53j*** t-6.24)*** (6.26)*** (7.23)**’ 11.05 5.26 0.57 2.46 0.11 0.00 2.15 36.60 46.60 12.98 $29,298 (-2.50)” 10.45 6.19 0.51 2.91 0.40 0.01 0.84 48.02 34.66 8.69 $35,035 (3.56)*** (,-4.701*** t.48) (-2.85)“’ f-10.54)*** t-.16) (20.55)* * * t-6.16)*** (6.04)*** (6.73)*** f-2.04)** 1984 INTEREST INCOME/ASSETS INTEREST EXPENSE/ASSETS NONINTEREST INCOME/ASSETS NONINTEREST EXPENSES/ASSETS LOAN LOSS PROV/ASSETS SEC. GAINS/ASSETS RETURN ON ASSETS LOANS/ASSETS SECURITIES/ASSETS EQUITY/ASSET TOTAL ASSETS (000) 5N..Aol 0.42 2.59 0.13 0.03 2.06 48.34 36.49 13.60 $39,067 NA 5.87 0.87 3.54 0.22 -0.02 1.04 52.53 28.70 8.96 $47,037 (-3.59)**’ (- 1.73) (-3.14)“’ (-3.41)“’ L97) (10.76)*‘* (-1.53) (2.96)*** (6.83)*” (- 1.54) :.A37 1.56 3.54 0.24 -0.02 2.15 38.40 44.16 13.80 $33,599 sN.?B 1.09 3.70 0.48 - 0.01 0.89 48.87 32.10 9.60 $37,349 -5.31)“’ t.62) t-.26) -5.441*** C-.38) (11.79)*** -5.411*** (6.02)‘*’ (6.16)*+* sNp59 0.62 2.48 0.14 0.01 2.06 39.51 43.10 12.88 $32,231 - 1.47) NA 6.55 0.55 2.92 0.43 -0.01 0.80 50.05 32.42 8.68 $36,457 f-5.07)*** t.411 C-2.71)*** f-7.14)‘** f.72) (18.31)*** f-5.50)‘*’ (5.32)*** c7.11j*** (- 1.21) 1985 INTEREST INCOME/ASSETS” INTEREST EXPENSE/ASSETS NONINTEREST INCOME/ASSETS NONINTEREST EXPENSES/ASSETS LOAN LOSS PROV/ASSETS SEC. GAINS/ASSETS RETURN ON ASSETS LOANS/ASSETS SECURITIES/ASSETS EQUITY/ASSET TOTAL ASSETS (000) 10.84 4.66 0.40 2.46 0.12 0.05 2.19 47.15 38.23 13.98 $43,197 10.21 5.33 1.11 3.74 0.28 0.07 1.14 52.33 29.32 9.18 $49,477 (3.65)“’ c-3.00)*** (- 1.87) t-3.24)“* (-3.931’*’ (- .34) (9.43)*** (- 1.77) (3.19)*** (7.04)*** (- 1.22) 11.21 4.94 1.71 3.74 0.26 0.01 2.22 40.17 43.64 14.12 $36,820 10.61 5.64 1.18 3.86 0.54 0.06 1.02 49.88 31.36 9.89 $38,624 (5.59)*** (-5.18)**+ t.571 C-.15) c-6.61)‘** t-4.39)**’ (9.17)*** c-4.95)*** c5.941*** (4.98)’ l l C-.56) 10.72 5.07 0.63 2.41 0.15 0.07 2.14 40.34 41.91 13.34 $35,181 10.25 5.92 0.55 2.94 0.62 0.07 0.79 48.87 32.57 8.69 $38,171 (2.88)*** C-4.86)‘*’ (.51) f-3.49)*** (-13.25~*** Lll) (21.41)*** f-4.32)“’ (3.67j”+ (7&u*** (- ,831 1986 INTEREST INCOME/ASSETS INTEREST EXPENSE/ASSETS NONINTEREST INCOME/ASSETS NONINTEREST EXPENSES/ASSETS LOAN LOSS PROVlASSETS SEC. GAINS/ASSETS RETURN ON ASSETS LOANS/ASSETS SECURITIES/ASSETS EQUITY/ASSET TOTAL ASSETS (000) 10.03 4.10 0.38 2.35 0.10 0.12 2.10 45.77 35.84 13.77 $49,113 9.34 4.65 1.22 3.77 0.24 0.10 1.08 53.09 26.74 9.27 $50,730 (3.46)* * * c-2.79)+** (-1.37) t-2.28)*’ t-5.26)*** t.27) (9.48)*** t-2.48)‘* (3.15)*** (6.34)*** (6.32) 10.32 4.30 1.54 3.52 0.29 0.04 2.08 41.48 38.76 13.77 $41,093 9.69 4.91 1.32 3.97 0.50 0.11 0.99 50.00 30.04 9.92 $40,797 (4.43)*** c-5.09)*** (. 28) (- .68) c-5.40)*** (-6.02)**+ (9.63)*‘* c-4.351*** (4.25)*** (6.25)*” LO91 10.03 4.50 0.60 2.37 0.21 0.12 2.05 40.45 39.78 13.61 $37,820 9.42 5.23 0.54 2.93 0.54 0.12 0.77 48.06 32.38 8.68 $39,696 c4.311*** f-4.54)*** I.37) C-3.42)*** ‘-W2;*” (24.30)*** C-3.84)*** (3.43)* l * (7.92)‘** t-.52) 1987 INTEREST INCOME/ASSETS’ INTEREST EXPENSE/ASSETS NONINTEREST INCOME/ASSETS NONINTEREST EXPENSES/ASSETS LOAN LOSS PROV/ASSETS SEC. GAINS/ASSETSRETURN ON ASSETS LOANS/ASSETS SECURITIES/ASSETS EQUITY/ASSET TOTAL ASSETS (000) 9.43 3.81 0.39 2.43 0.09 0.07 2.02 50.41 35.33 14.37 $52,300 8.94 4.35 1.46 4.06 0.21 0.04 1.07 58.04 25.42 9.67 $53,223 (2.99)*** (-3.31)*** (- 1.32) (-2.02)*+ C-4.20)*** t.62) (10.84)*** C-2.72)** (3.32)*** (6.89)**’ t-.171 9.60 3.93 4.46 5.67 0.26 0.04 2.49 44.28 37.38 15.46 $43,519 9.07 4.47 1.33 3.86 0.46 0.02 0.96 52.18 29.89 10.00 $41,679 a Mean for all high performance banks, in percent terms unless otherwise stated. b Mean for all small banks, in percent terms unless otherwise stated. c *** indicates high performance and all banks are statistically significantly different at the 1 percent level. ‘* indicates high performance and all banks are statistically significantly different at the 5 percent level. d INTEREST INCOME/ASSETS 34 is stated on a taxable-equivalent basis. ECONOMIC REVIEW, NOVEMBER/DECEMBER 1989 (3.95)*** t-5.28)*** C.85) t.65) t-4.36)**’ (1.07) (2.97)*** C-3.87)*** (3.72)‘** (3.60)*** C.57) 9.16 4.06 0.60 2.41 0.18 0.05 1.89 42.25 39.52 14.00 $40,679 8.77 4.71 0.54 2.93 0.37 0.03 0.81 49.83 32.86 8.88 $40,631 (3.03)*** -4.26)*** C.41) -2.96)*** -5.60)“* C.66) (19.77)**’ C-3.24)*** (3.08)**’ (7.86)“’ LO11 MIDWEST High 12.36 5.74 0.48 2.62 0.19 -0.08 2.39 40.60 43.32 14.81 $18,851 11.61 5.02 0.53 2.69 0.22 -0.02 2.42 39.91 45.14 16.20 $20,759 !A33 0:79 2.69 0.24 -0.02 2.35 40.00 44.30 16.77 $22,585 10.72 4.66 0.75 2.63 0.30 0.09 2.27 37.54 44.27 16.94 $24.331 9.80 4.06 0.71 2.66 0.37 0.14 1.99 35.76 44.81 16.89 $26,345 9.03 3.65 0.70 2.63 0.28 0.00 1.95 36.61 44.34 17.61 $27,038 SOUTHWEST All T Stat High All 12.00 7.32 0.51 2.92 0.38 -0.02 1.10 50.57 32.48 9.06 (1.61) C-3.69)“’ C-.21) (- 1.22) t-3.85)*‘* (- 1.61) (8.93)’ t-4.80)** (4.92)*** t5.091*** 12.20 6.00 0.75 2.74 0.20 -0.03 2.22 38.45 42.94 12.68 11.64 6.54 0.78 3.41 0.49 -0.01 1.13 49.57 26.82 9.67 $25,193 10.92 6.42 0.52 2.91 0.54 0.01 0.91 50.84 34.36 9.03 $26,394 !..A81 0.60 2.91 0.91 0.00 0.62 51.64 32.75 8.99 $27,188 10.46 6.13 0.60 2.97 1.31 0.11 0.41 48.69 33.36 8.91 $27,804 9.33 5.31 0.60 3.01 1.23 0.19 0.25 45.11 35.35 8.57 $28,981 8.61 4.66 0.62 2.97 0.64 0.02 0.56 45.25 37.67 8.81 $29,767 l * C-2.59)** (2.76)*** t-3.84)*** f.08) (-‘.82j t-3.50)‘** (- 1.97) (14.17)*** t-5.16)*** (4.90)**’ (5.15)*** (- 1.61) (-3.93)“’ C.91) (- ,921 C-10.38)*** (- 1.02) (ll.ll)**’ C-5.33)*+* (4.981*** (5.20)*** - 1.30) (I.291 (.-4.4u*** f.67) - 1.42) (-- 18.82)“’ CF.811 (18.45)*** c-3.93)*** (4.38)“” (5.24)*** C-.981 (2.16)** c-4.45)‘** C.45) (- 1.50) (-12.11)*” C-.87) (23.75)*** C-3.27)*** (3.65b”’ t5.541*** (- ,721 (2.23)** t-4.17)*** C.40) (- 1.36) t-4.87)*** (- 1.80) (20.12)*** t-2.77)*** c2.1u** (5.691*** (- .73) $25,633 11.50 5.14 0.80 2.63 0.32 -0.01 2.45 37.44 44.59 13.38 $29,313 NA 5.57 0.79 2.68 0.29 -0.01 2.13 38.10 44.47 14.28 $32,190 11.34 5.13 0.78 2.64 0.48 0.02 2.17 38.36 43.30 14.28 $35,030 10.56 4.51 0.79 2.62 0.52 0.08 2.12 37.27 42.87 14.78 $36,847 9.47 4.04 0.72 2.52 0.41 0.04 1.96 35.52 45.45 14.61 $39,661 WEST T Stat $34,003 10.52 5.72 0.82 3.39 0.72 0.02 0.85 50.66 28.00 8.94 $36,836 NA 6.45 0.87 3.42 0.87 0.00 0.64 53.30 25.40 8.63 $38,749 10.57 5.92 0.92 3.59 1.18 0.09 0.40 53.43 24.28 8.53 $39,644 (2.77)*** ‘4;;; c-4.451*** c-5.17)*” (- ,921 i4.iij*** -4.a71*** (7.20)*** c4.941*** -3.38)*** (3.43) (--3.06)*** C-.13) C-3.82)+** (64.26j*** (- 1.06) (14.73)*” t-5.431*** (7.02)*** (8.67)*** l l l t-2.62)** C-5.02)*‘* (- .63) t-3.62)‘*’ t-9.61)‘** C-.19) (i9.3Oj*** C-6.14)*** (7.88)‘** (7.11)*** t-2.08)** (4.17)“’ C-4.78)*** (- 1.19) c-5.44)*** (-8.98)“’ (- 1.62) (23.77)*** c-5.87)*‘* (7.45)“’ (10.42)*** t-1.10) 9.50 t5.391*** 5.22 C-7.18)*** 0.90 (- 1.04) 3.68 C-6.53)*** 1.56 (- 12.23)*** 0.21 C-3.81)**’ -0.13 c25.2sj*** 50.95 C-5.24)*** 23.78 (7.211*** 8.11 (9.88)**’ $39,930 C-.74) 8.81 (4.23)*** 4.70 t-7.11)*** 0.90 (-1.61) 3.63 t-7.30)+*’ 1.29 t-12.05)*** 0.04 (- .03) -0.14 (26.94)“’ 49.56 C-5.22)*** 27.79 (6.21)**+ 8.04 (lo.ol)*** $39,823 (- .04) U.S. High All T Stat High All 12.09 4.80 0.91 3.55 0.15 -0.12 2.36 45.25 31.25 15.44 11.15 5.78 0.90 4.84 0.60 0.01 0.36 55.69 18.83 12.21 <2.84)** (--i.9ij C.03) t-2.41)** (-6.72j**’ (- 1.67) t10.741*** t-3.28)*** (3.12)*** (1.33) 12.03 5.79 0.83 2.97 0.16 -0.07 2.20 11.61 6.84 0.65 3.33 0.42 -0.01 0.95 49.83 29.56 9.47 $31,017 11.06 4.11 1.09 3.31 0.18 -0.04 2.50 45.43 32.73 16.22 $34,796 2.25 1.15 3.38 0.02 -0.03 2.68 46.52 31.89 18.64 337,433 11.49 3.91 1.03 3.34 0.31 0.09 2.80 44.84 34.52 18.24 $39,851 9.97 3.53 1.04 3.45 0.24 0.05 2.08 43.92 34.34 18.31 $42,840 9.35 3.25 1.20 3.51 0.30 0.03 2.05 $27,156 10.41 5.23 1.02 4.63 0.63 0.01 0.48 57.98 20.01 9.50 $31,218 !.tO 1.12 4.63 0.80 0.00 0.44 59.60 18.91 8.79 $33,669 10.47 5.40 1.20 4.75 1.08 0.08 0.08 58.56 18.64 8.39 $34,294 9.34 4.57 1.35 4.77 1.17 0.15 0.03 45.46 19.35 8.12 $36,337 8.87 4.09 1.21 4.72 0.91 0.03 0.11 55.49 22.78 8.35 43KE 19.24 $45,566 $36,664 FEDERAL RESERVE BANK OF RICHMOND C.71) (1.87) t-2.81)** L31) C-4.12)*** C-6.57)*** C-1.10) (5.42)*‘* t-3.11)*** (3.99)^” (2.01) t.63) t-3.42)*** C.17) C-2.78)**’ (-4.991”’ (- 1.09) (5.77)*** (--2.4oj** c3.05j*** (1.80) t.65) 11.77) (-‘3.soj*** (- ,961 f-4.67)*** C-9.551*** C.32) (5.39)*** f-2.45)** t3.591*** (2.04) C.981 (2.45)*‘ t--3.osj*** (-1.18) t-3.09)*** (-11.61j**’ c-3.29)*** (14.62)“’ (-2.09) t3.101*** (2.16)** (1.09) (1.82) c-.2.89)** (- .05) C-2.52)** t-6.30)*** t.291 (13.84)*** (- 1.70) c3.33j*** (2.13)** (1.48) 39.79 42.18 13.31 $26,193 11.26 4.98 0.88 2.94 0.20 0.00 2.28 39.03 43.60 13.99 $29,247 0.94 2.91 0.20 0.00 2.19 40.69 42.12 14.52 $32,224 11.02 4.84 0.96 2.91 0.26 0.05 2.24 40.69 41.95 14.75 $35,131 10.10 4.25 0.90 2.84 0.29 0.09 2.07 40.35 39.94 14.77 $38,388 9.35 3.86 1.67 3.41 0.25 0.04 2.10 42.02 39.92 15.53 $40,799 $31,131 10.61 5.98 0.66 3.26 0.53 0.01 0.84 50.20 31.55 8.97 $33,257 Et5 0.77 3.32 0.68 -0.01 0.71 52.01 29.61 8.91 $34,693 10.44 5.87 0.81 3.41 0.95 0.08 0.60 50.94 29.51 8.88 T Stat (5.24)*** t-8.50)*** L94) C-2.16)** (-14.78)“’ C-3.36)*** (22.72)*** c-11.44)*** c13.37j*** (10.18)**’ c-4.44)*** (8.35)‘** C-9.98)++* (i.03j (- 1.67) t-12.60)*** (- 1.42) (28.691*** t-12.22)*** u2.3ij*** (11.83)**’ C-3.21)*** c-11.15)*** L78) (-2.34)” t-19.64)“’ (.I51 (28.91)*” (-11.93)*** (12.50)*** (10.06)** * (- 1.78) (7.02)‘** (-11.19)“’ (.61j t-2.36)** c-27.991*** C-3.06)*** (29.361*** C-9.56)*** $35,715 9.44 (9.831*** 5.12 C-12.38)*** 0.84 C.31) 3.45 t-3.32)**’ 1.00 (-26.49)“’ 0.16 C-4.09)*** 0.40 (40.60)“’ 48.94 t-7.88)*** 29.73 (9.271”’ 8.66 u2.121*** $36,888 8.78 4.59 0.83 3.41 0.69 0.03 0.51 49.55 31.68 8.81 $37,482 c.901 (8.60)*** C-11.24)*** L86) (0.001 t-17.70)*** t.85) (11.56)*** C-6.27)*** t7.391*** uo.45j* * * (1.81) 35