The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
ECONOMIC REVIEW FEDERAL RESERVE BANK of CLEVELAND QUARTER IV Economic Review Federal Reserve Bank of Cleveland Quarter IV 1985 Stochastic Interest Rates in the Aggregate Life-Cycle/ Permanent Income Cum Rational Expectations Model .................................. 2 Recent tests of the life cycle/permanent income cum rational expectations model have assumed either that real interest rates are constant, or that consumers know the future path of real rates. This article estimates a life cycle cum rational expectations model that allows for sto chastic real interest rates. The results show that the model is strongly rejected using postWorld War II U.S. data. New Classical and New Keynesian Models of Business Cycles.................................. 20 Both Keynesian and Classical economists have developed new models of the business cycle dur ing the 1970s and 1980s. Both have striven to establish firm microfoundations for their the ory of fluctuations, so that events may be un derstood in terms of the basic economic en vironment and the actions of individual agents. In this article, economic analyst Eric Kades pre sents bare-boned models of both schools, at tempting to lucidly illustrate sources of busi ness cycles. Simulations show that both models can mimic observed time series convincingly. Theoretical strengths and weaknesses are dis cussed, followed by a cursory examination of empirical evidence for and against each model. Economic Review is published quarterly by the Research Department of the Federal Reserve Bank of Cleveland, P.O. Box 6387, Cleveland, OH 44101. Telephone: 216/579-2000. Editor: William G. Murmann. Assistant editor: Meredith Holmes. Design: Jamie Feldman. Typesetting: Liz Hanna. Opinions stated in Economic Review are those of the authors and not necessarily those of the Federal Reserve Bank of Cleveland or of the Board of Gover nors of the Federal Reserve System. Material may be reprinted provided that the source is credited. Please send copies of reprinted materials to the editor. K.J. Kowalewski is an economist at the Federal Reserve Bank o f Cleveland. The author would like to thank James Hoehn, Eric Kades, and Alan Stock man, who provided useful comments on an earlier draft. 1. See Tobin (1980) and Tobin and Buiter (1980) fo r good discussions o f these points. Stochastic Interest Rates in the Aggregate Life - Cycle / Permanent Income Cum Rational Expectations Model by Kim J. Kowalewski There has been renewed interest in consump tion behavior in the past 10 years. The origin of this interest is not so much due to deteri oration in the ability of economists to predict future output and prices, although that is clearly important. The main impetus is the challenge of the “ New Classical” school. Barro (1974) argued that rational private agents do not view bondfinanced increases in government spending or decreases in taxes as increases in wealth, be cause they know that the new bonds must be retired by additional future taxes. Rational private agents therefore will increase current saving to pay for these future taxes, no mat ter how far into the future they come due. This additional saving is exactly enough to purchase all of the new debt; interest rates and aggregate wealth remain unchanged. This implies that bond-financed increases in government spending have a multiplier value of 1 , and that bond-financed tax cuts have a zero multiplier. Money-financed increases in government spending also have a zero multiplier, because rational private agents view the faster growth in money as leading to a higher inflation rate in the future. This higher inflation is another “ tax” that private agents will save for. That is, money-financed tax cuts have no effect on real variables, because one tax is just sub stituted for another. These “ New Classical” results are a direct challenge to the Keynesian and Monetarist schools, which assign higher values to these multipliers (at least in the short run), because the effects of fiscal policy actions are distin guished by how they are financed.1 Barro’s result depends, among other things, on the assumption that private agents have the opportunity to offset these government actions. This, in turn, assumes that capital markets are perfect—that there are no trans actions or other costs that drive a wedge between borrowing and lending interest rates, and that there are no informational asymme tries that are controlled with down payments, 2. Muellbauer (1983) and Wickens and Molana (1984) reject the model using U.K. con sumption and in come data. security interests, rationing the quantity of credit, and other non-price loan provisions. Thus, with perfect capital markets, the length of a consumer’s spending horizon (that is, the time span over which a permanent increase in life-cycle wealth/permanent income is con sumed) is as long as his remaining lifetime. It may be longer if, as Barro assumes, a consum er’s utility function includes the utility of his direct descendants. A consumer can borrow any amount up to the current value of his net nonhuman wealth, plus the present value of all his expected future after-tax labor income, all discounted at the common rate of interest. An increase in life-cycle wealth/permanent income will be consumed over the remainder of the horizon, making the amount consumed in the short run very small. If capital markets are imperfect, however, then the length of a consumer’s planning hori zon may be shortened. A consumer may not be able to borrow against all of his life-cycle wealth (or permanent income), or may do so only at a penalty rate of interest. Increases in life-cycle wealth/permanent income will be consumed over this shorter horizon, enlarging the (short-run) impact of bond-financed tax cuts or spending increases. Clearly, shorter horizons make it possible for stabilization pol icies to affect real variables, at least in the short run. Thus, the recent interest in consumption be havior centers on learning the length of con sumer spending horizons. The approach taken by most recent studies is to test some variant of the life-cycle/permanent income cum ration al expectations (RE-LC/PI) model assuming perfect capital markets. Rejection of the RELC/PI model, incorporating perfect capital markets, is taken to mean that horizon lengths may not be long enough to diminish the power of stabilization policies. Hall (1978), Flavin (1981,1985), Hayashi (1982), Muellbauer (1983), Wickens and Mol ana (1984), Bernanke (1982), Mankiw (1983), DeLong and Summers (1984), Boskin and Kot- likoff (1984), Kotlikoff and Pakes (1984), and Mankiw, Rotemberg, and Summers (1985) test the RE-LC/PI model with aggregate time ser ies data, while Hall and Mishkin (1982), Ber nanke (1984), and Hayashi (1985) use crosssection or panel data on individual households. Of the studies employing micro-data, only Bernanke (1984) can reject the model. Of the studies that employ aggregate time series data, Hall (1978), Hayashi (1982), Mankiw (1983), Bernanke (1984), and Delong and Sum mers (1984) cannot reject the model during the post-World War II period. Kotlikoff and Pakes (1984) can reject the model, but con clude that the differences from the model are not large enough to matter in practice.2 These studies are not the first to be con cerned with the length of consumer spending horizons. For example, Tobin (1951) argued that capital market imperfections may have accounted for the different savings behaviors of black and white Americans in the late 1940s. Houthakker (1958), in his review of Friedman’s (1957) permanent income hypoth esis, argued that the exclusion of capital mar ket imperfections was the main defect of Friedman’s work. Friedman (1963) argued that consumer horizon lengths were about three years. Before rational expectations came into vogue, there were numerous tests of the life cycle and permanent income models, begin ning with Modigliani and Brumberg (1954) and Friedman (1957). The debate about the ef ficacy of the 1968 temporary tax increase fo cused on the length of consumer spending hor izons see, for example, Okun (1971) and Blinder (1981). There has been considerable theoretical work done on the impact of capital market imperfections (see, for example, Tobin and Dolde [1971], Dolde [1973], Pissarides [1978], Heller and Starr [1979], Foley and Hellwig [1975], and Watkins [1975,1977]). What is new about these recent studies is their assumption of rational expectations. Un fortunately, richness of detail seems to have been sacrified for this assumption. For exam- pie, none of the recent models that are esti mated with U.S. aggregate time series data allows for uncertain real interest rates. All of the models, except Bernanke (1982) and Mankiw (1983) assume that the real interest rate is constant. Bernanke (1982) and Mankiw (1983) allow real interest rates to vary, but as sume that consumers know all future real interest rates. It is rather curious that stochastic real inter est rates have been ignored, because the real interest rate is a key variable in the life cycle/permanent income model (and in many New Classical models). The interest rate mea sures the exchange rate between consuming today and saving today to consume more to morrow. The life-cycle/permanent income model determines the utility-maximizing al location of life-cycle wealth (permanent in come) across time by balancing the marginal rate of transforming consumption today into consumption tomorrow (the interest rate) with the marginal rate of substitution (the dis counted marginal utility from consuming to morrow relative to that from consuming to day). Changes in interest rates, expected or unexpected, should lead to a reallocation of consumption spending across time. Thus, an allowance for stochastic real interest rates should provide a more powerful test of the RE-LC/PI model and indirectly of the (maxi mum) length of the representative consumer’s spending horizon. In this article, we estimate a RE-LC/PI mod el that allows for uncertain future interest rates. The model is developed by Muellbauer (1983), which he estimated with United King dom (U.K.) data. To put Muellbauer’s model into perspective, the Hall and Flavin (1981) models are also discussed and estimated. Up dating the Hall and Flavin results with the 1980s data also may reveal any structural instabilities and shifts in the distribution of horizon lengths across consumers, which is a possibility ignored by all of the recent RELC/PI tests. Section II reviews the RE-LC/PI models, section III briefly outlines the pro cedures followed in estimating the three mod els and explains the results, and the third sec tion concludes our study. I. The Life-Cycle/Permanent Income Model With Rational Expectations Tests of the RE-LC/PI model begin with Hall (1978). The consumer is assumed to maximize the expected present discounted value of current and future utility. Income is exogenous and is known in the current period, but unknown thereafter; the consumer’s choice variable is the level of consumption each period. The horizon begins with the current period and ends at the (known) last period of the consumer’s lifetime. There are no bequests and no capital market imperfections. Expectations are rational— functions of all information available in the current period. Real interest rates and rates of time preference are assumed to be constant. The model is: T- t (1) m a x B .S t a 'W C w ) ] C t i=o subject to Tt X 1=0 T- t ( R - C „ i) - X i— 0 (R'yiti) - A „ where 1 , plus the pure rate of time preference, assumed constant, R is the inverse of 1 plus the real, after-tax rate of interest r, also assumed constant, (8 > R ) , C is real life cycle consumption (not NIPA personal consumption expenditures), y is real labor income, A is current real nonhuman wealth, £/(•) is the instantaneous utility function, and E t is the expectations operator, conditioned on the information available at time t (variables dated M and earlier). 8 is the inverse of The first order conditions for this problem are: (2a) (2b) E tU ' ( C t+i) = ( R/ 8 ) E tU ' ( C t+il), for i = 1 to T - 1; in particular, for i - 1 E tU ' ( C t+i) = (R/ 8 ) U ' ( C ,). There are two things to note about (2b). First, C , can be thought of as a sufficient sta tistic for C t+x, that is, no variable except Ct helps predict future marginal utility of con sumption U' ( C, +1). Second, with the assump tion of rational expectations, marginal utility follows the regression relation: (3) U ' ( C t+l) = y U ' ( C t) + e,+1. The term e t+x represents the impact on marginal utility of all new information that becomes available in period t + 1 about the consumer’s lifetime well-being. Under rational expectations, E te t+i = 0 and e +i is orthogonal to U'(C t). Moreover, e should be white noise, that is, unpredictable using variables in the information set. If the utility function is quadratic or “ the change in marginal utility from one period to the next is small, both because the interest rate is close to the rate of time preference and because the stochastic change is small.” (See Hall [1978, p. 975].) Then equation (3) becomes: (4) C,= 7 Cm + e t. That is, life-cycle consumption follows an AR (1) process—no other variables dated M or earlier affect Ct. If y = 1, then consumption follows a random walk. It is important to notice that (4) is not a structural model of life cycle consumption behavior. Because it is only the first-order condition for utility maximiza tion, it is only an implication of the life-cycle model under rational expectations. Indeed, it is only a necessary condition for this RE-LC model to be true. Hall also shows that lifetime resources evolve as a random walk with trend. First, nonhuman wealth follows the relation: (5) A , = R \A M + y m - C M). Second, human wealth, H ,, is the sum of current labor income and the expected present discount value of future labor income: (6) % ( R ‘E , y i . , ) , 1=0 where E , y t = y ,, from which it follows that: (7a) H, = R l( H t.\ - _yM) + jut, where m, represents the present value of the changes in expectations of future income that occur between period M and t: Tt (7b) = £ [ * ' ( £ , JV, i=0 Again, under rational expectations, E t ln t - 0, and n t should be white noise. Under cer tainty equivalence, 8 , = a tn t, where a , is an annuity factor modified to take account of the fact that the consumer plans to make con sumption grow at a proportional rate y over his remaining lifetime. Then the equation for total wealth is: (8) Al +tf,= fl-1(l-at.1)(AM +ffM) +M/. Flavin (1981) estimates a different version of the permanent income model using the in sight from (7) to eliminate the unobserved H She starts by defining current consumption as the sum of permanent and transitory con sumption. By equating permanent consump tion with permanent income (jyf), she has: (9) C , = y f + e2t, where e 2/ is transitory consumption. Thus, permanent income is defined to be the annuity value of the expected present dis counted value of human and nonhuman wealth ( A t+ H t), assuming the real, after-tax rate of interest, r, is constant: (10) y pt = r ( A t + ^ [ R ,+1 E ty ,+I]). i=0 3. This assumption is not unreasonable, given that her model explains short-run changes in con sumption. However, in her later paper, Flavin (1985) uses annual data where it seems less likely that changes in the rate o f return to cap ital dominate endog enous changes in wealth accumu lation. Flavin shows that E y f+1 = y fusing the insight implicit in equation (7b). Substituting (10) into (9) and using the nonhuman wealth constraint: (11) A /+1 = R 1 A i + y t - C t. Unlike equation (5), current period saving does not earn interest in equation (11). Equa tion (9) can be used to solve for C /+i in terms of Ct: 00 (12) C , . r C , + r S [ r l( £ , . , - £ , ) j w .,)] 1=0 - R t 2t + Flavin notes that because the coefficient of t 21 is not -1, C t will not evolve as a random walk unless the transitory consumption term 621 is zero for all t. Equation (12) contains revisions in expecta tions of future real labor income. Flavin notes that “ [a]s an empirical matter however, unan ticipated capital gains and losses on non human wealth probably constitute a signifi cant fraction of the revisions in permanent income this model is trying to capture.” (See Flavin [1981, p. 988].) She defines unantici pated capital gains as the present value of the revision in the expected earnings associated with the current nonhuman wealth position. By then assuming “ ... that changes in the rate of return to capital... are quantitatively more important than the endogenous changes (in nonhuman wealth) in determining the timeseries properties of the observed path of non labor income ...” , unanticipated capital gains can be approximated as the present value of the revision in expected future nonlabor in come. (See Flavin [1981, p. 988].) This permits her to use disposable personal income (YD) in place of labor income (y) in equation (12).3 Flavin next derives an expression for the revision in expectations of future YD by assuming that YD follows an ARMA process. She shows that the revision in the expectation of YD,+S(s> 0) between periods t and M is the product of the moving average error of YD in period t (u, and the sth coefficient from the corresponding moving average representation for YD (Bs). Then the present discounted value of the set of revisions is: (13) (£ [# * £ * ])« /. Thus, she demonstrates that the revision in income expectations is white noise. The ARMA model for YD plus the equation formed by substituting (13) into (12) is Flav in’s permanent income consumption model. Note that (13) still contains an unobserved var iable u t. This term is included with the other error terms in estimation, making her con sumption equation very similar to Hall’s. The difference is that Hall’s model can be viewed as a reduced form of Flavin’s structural model. Flavin argues that the error terms in the two equations are correlated because her model is incomplete. The income equation error will contain additional terms because the information set probably contains varia bles other than past income. These omitted in formation set variables will also appear in the consumption equation error through (13), thus producing the correlation between the two equation errors. She dismisses this ap parent specification bias by assuming that these omitted information set variables are serially uncorrelated and uncorrelated with the lagged income terms. Hayashi (1982) also uses equation (7) to elim inate the unobserved H t. He starts with the permanent income model in level form: (14) C t = a ( A , + H t) + e t, where et is defined as “ transitory con sumption” —a shock to preferences or meas urement error in Ct and A t. He notes that a , the propensity to consume, is a function of the expected real rates of return from nonhuman wealth and the subjective rate of time prefer ence: but, like Hall and Flavin, assumes that these factors are constant over time and indi viduals. Using (7a) with an “ overall” discount rate 1 +d in place of R, Hayashi eliminates H , from (14): (15) C l= ( l + d ) C t.I + a [ A r ( l + d ) ( A tl + yt-\)] + v „ where vt = u t- (\ + d)u M + a yut. Like Flavin, Hayashi also uses a two-equation model, com posed of equation (15) and a stochastic version of equation (5). He adds an error term to Hall’s nonhuman wealth identity to capture unanticipated movements in asset prices and measurement errors in A t, A t.\,yM, and CM. Note that Hayashi’s model uses labor in come instead of YD and is slightly more gen eral than either Hall’s or Flavin’s, because it does not assume that 1 + d = R 1. Hall, Flavin, and Hayashi test their models by adding other variables to the right-hand side of (4), the modified version of (12), and (14). It is clear that by doing so they test the joint hypothesis that both the life-cycle/ permanent income model and the rational ex pectations assumption are correct. If they were interested in testing only the assump tion of rational expectations, conditional upon the LC/PI model, for example, they would have compared their models with suitable transformations based on different hypoth eses about expectations formation. If the joint hypothesis is correct, then no other variable in the information set except CM will help forecast Ct. Although any set of variables could be used to test these models, income is an obvious choice, because a direct relation ship between consumption and current income in these models would be strong evidence against the simple life-cycle/permanent in come model assuming perfect capital markets and against Barro’s neutrality hypothesis. Recall that there is no direct structural relationship between consumption and income in these models. Current income may be cor related with current consumption, but the correlation arises only indirectly, because cur rent income represents new information about human wealth/permanent income. Unlike Friedman (1957) and Modigliani and Brumberg (1954), who allowed for the possi bility that some unexpected changes in income would not alter a consumer’s estimate of his permanent income or life-cycle wealth, all unexpected income changes in the Hall, Flavin, and Hayashi models lead to revisions in permanent income or life-cycle wealth and, hence, consumption. The models are estimated and tested with post-World War II U.S. aggregate time series data. Unfortunately, it is difficult to compare their results because they use different data and sample periods. This is partly due to the lack of reliable data on life-cycle/permanent consumption. Hall uses real, per capita PCEnondurables and services as the consumption variable, ignoring the service flow from con sumer durables because of the lack of reliable data. Flavin uses only real per capita PCEnondurables as the consumption variable. She notes that the consumption of durable services should exhibit a lagged response to changes in permanent income due to the transactions costs of adjusting durable good stocks. The same is true of housing services, which form a large part of PCE-services. By using only PCE-nondurables, she says that she gives the benefit of the doubt to the random walk hypothesis of one-quarter adjustment. However, this point is probably irrelevant, because Flavin detrends the consumption and income data before estimation. The strong trend in PCE-services most likely would be eliminated with detrending, allow ing her to use PCE-nondurables and services as the de pendent variable. Indeed, as shown below, Flavin’s model rejects the RE-LC/PI model, using PCE-nondurables and services as the de pendent variable. Hayashi uses real, per cap ita annual data constructed by Christensen and Jorgenson (1973 and updates) for the con sumption variable and a modification of their labor income variable for y. The consumption data contain imputations for the service flow of consumer durables. Flavin uses real per capita YD for the income variable, and all three use this variable (or its lagged value) for testing their models. Hall’s first test consists of adding three addi tional lagged C terms to the right-hand side of (4) and finds them to be statistically insignif icant individually and taken together. He finds the same result when one, four, and 12 lagged YD terms are added. In all cases, the coefficient on C M is not significantly different from 1, which leads Hall to conclude that aggregate consumption is a random walk process. However, when Hall adds four lagged stock price variables (Standard and Poor’s compre hensive index of stock prices deflated by the implicit deflator for PCE-nondurables and ser vices and divided by population), he finds that they are individually and collectively statisti cally significant. Hall argues that this evi dence does not contradict the joint hypothesis, if it is assumed that “ some part of consump tion takes time to adjust to a change in per manent income. Then any variable that is cor related with permanent income in period t -1 will help in predicting the change in con sumption in period t, since part of that change is the lagged response to the previous change in permanent income.” (See Hall [1978, p. 985].) He also says that “ the discovery that consumption moves in a way similar to stock prices actually supports this modification of the random walk hypothesis, since stock prices are well known to obey a random walk them selves.” (See Hall [1981, p. 973].) In all tests, the Durbin-Watson statistic, which is biased downwards in these models when the auto correlation of the errors is positive, cannot reject the hypothesis of no first-order auto correlation. Hall thus concludes that the model cannot be rejected. This is a rather curious inference. Hall finds a variable that contradicts the null hypothesis, and he subjectively rationalizes it! Moreover, it seems highly improbable that two truly random walks will be strongly cor related with each other. Since the two series are correlated, does this mean that the two series are not random walks, that they are random walks around a common trend, that there is a structural relationship between the two series, that the correlation is simply spur ious, or that they are an artifact of aggregate time series data? Unfortunately, Hall does not report any tests of these possibilities. Flavin adds the current and first seven lagged changes in real per capita YD to equa tion ( 12) with A C;as the dependent variable. By adding these eight terms, she obtains a just-identified system. The reduced form of her model thus becomes: (12a) YD t = m + a i YD M + a 2 YD t.2 + ••• + as YD ,-8+ 7711 A C /= H2 + /3o( U\ + ( a i- l ) YDt.i + (x2 YD +... + otsYD t.s) + fii AYD t i + ($2 A YD ,.2+ ... + AYD t 7+ r]2 U where «2 / contains e 2 1 and (13). The (3’s are “ measures of the ‘excess sensitivity’ of con sumption to current income, that is, sensitiv ity in excess of the response attributable to the new information contained in current income.” (See Flavin [1981, p. 990].) Thus, a test of the joint statistical significance of the (3’s is a test of the RE-PI model. Over the 1949:IIIQ to 1979:IQ sample, Flavin can reject the model at a 0.5 percent significance level. The coefficient /3o on the A YD t term allows her to test for a direct effect of current income on C, although her estimate of (3o is quite large relative to those of the other A YD terms, its /-statistic is only 1.3, suggesting that the test “ falls short of providing conclu sive evidence that the permanent incomerational expectations hypothesis fails in a quantitatively significant way.” (See Flavin [1981 p. 1002].) Hayashi adds YD tto equation (14) and finds its coefficient to be of the same order of mag nitude as the estimate of the discount factor, but statistically insignificant in his twoequation model. He also finds that the dis count rate is statistically different from the constant real rate of return, contrary to Hall’s and Flavin’s assumptions. Although this is 4. It is not clear how Bernanke lets the real interest rate vary over time. evidence in favor of the permanent income cum rational expectations hypothesis, Haya shi argues that the relevant measure of consumption for the liquidity-constrained households is personal consumption expendi tures as defined in the National Income and Product Accounts (NIPA), which excludes ser vice flows from consumer durables and in cludes expenditures on consumer durables. The foregoing test of the permanent income hypothesis seems to be in some sense unfair to the alternative hypothesis of liquidity con straints.” (See Hayashi [1978, p. 908].) When he uses PCE as the dependent variable and estimates only the consumption equation (be cause the asset equation includes consumer durables), he finds the coefficient on current YD to be fairly large (0.892) with a /-statistic of about 20. On the basis of this result, he is persuaded to reject the permanent income cum rational expectations model. Here again is a rather curious inference. In effect, Haya shi is saying that only PCE-durables pur chases can be liquidity-constrained. Other authors have tried to relax some of the assumptions made by these writers. Ber nanke (1982) and Mankiw (1983) focus on the separability issue by adding consumer dura bles to the life-cycle cum rational expectations model. They argue, like Flavin, that lagged stock adjustment and accelerator effects may lead to an incorrect rejection of the model. This is even true when durables are excluded from the analysis, if nondurables and dura bles are not separable in consumer utility functions. Moreover, as Hayashi points out, imperfections in capital markets are likely to show up in the pattern of durables purchases. Bernanke derives a two-equation system in current period PCE-nondurables and services and next period’s stock of consumer durables as the solution to the utility maximization problem. A quadratic utility function contain ing quadratic costs of adjusting consumer dur able stocks is used. Mankiw also obtains a two-equation model, only based on the firstorder conditions for utility maximization. Both show that consumption is not a random walk. In Bernanke’s model, this is due to the adjustment costs, which supports Hall’s asser tion that adjustment costs can be consistent with the life-cycle cum rational expectations model. In Mankiw’s model, consumption is not a random walk, because the real rate of interest and the relative price of durables are non-constant. Both economists test their models with postWorld War II U.S. aggregate time series data. Under the assumption of constant real inter est rates, Bernanke finds that the response of consumers to an income innovation is signifi cantly greater than predicted by the theoreti cal model and thus rejects the life-cycle cum rational expectations model. He claims, but unfortunately does not prove the evidence, that a similar result obtains if the real inter est rate is allowed to vary. Mankiw adds disposable income growth terms to both equations in his model and finds them statistically insignificant. He thus finds no evidence against the life-cycle cum rational expectations model and argues that his model “ ...is a useful framework for examining the linkage between interest rates, prices, and consumer demand.” (See Mankiw [1983, p. 23].) As in many past studies, he also finds that consumer durables are quite sensitive to the real rate of interest. Depending on the parameter values chosen, the short-run elas ticity of the stock of consumer durables with respect to the real interest rate varies between -1.7 and -4.3. Mankiw’s results also suggest that the assumption of rational expectations is unimportant because he obtains results similar to those studies that do not assume rational expectations. Real interest rates are not handled very sat isfactorily by Mankiw .4Consumers are as sumed not to know future income, but are assumed to know future interest rates (and the relative price of durables). Thus, interest rates are allowed to vary over time in a very uninteresting way. Muellbauer (1983) and 5. In general, when real interest-rate ex pectations are proba bilistic the coeffi cient on i depends on the joint distribution o f ex pected real incomes and real interest rates. In both cases, the optimal forecast o f current consump tion requires more information than provided by C t.\. Wickens and Molana (1984) allow for random and unknown future real interest rates. Wickens and Molana show that when the in terest rate in the life-cycle cum rational expec tations model is random, the first order con dition for utility maximization becomes: (16) E t.xU'{ C t+Il) = 8 E t l [ a / R t+i)U'(Ct+l)] ( i > 0). This expression is obtained by substituting C, out of the utility function with the period-toperiod budget constraint ( 11 ) and maximizing the present discounted value of expected future utility with respect to A t. Expectations are formed with the information set available at the end of period M , which includes varia bles dated M and earlier. With the necessary assumptions, (16) can be written as: (17) E t.xCt+j = E t.\y t+i ( E t.iCt+i-i), (i > 0 ). where 2 is a function of the interest rate and the rate of time preference. Thus, as in Hall’s equation (2a), the coefficient on the lagged con sumption term varies with the real interest rate.5With the appropriate assumptions, Muellbauer obtains an expression in poten tially observable variables: (18) A l n C t - no + fis^Mr/.i + <5 1 ( 7 1 / + 8 2 0 2 1 + e f+ I t where o 1 and 02 are the innovations in period t real disposable income and the real interest rate based on information available at the end of period M, which includes variables dated t1 and earlier. The Wickens and Molana model differs only slightly from this, using r t+x in stead of r t_h because of a minor difference in the dating of the interest rate in the cash flow constraint. Both papers use post-World War II U.K. aggregate time series data. Also note that apart from the logarithms and the dating difference on r, Flavin’s model is nested in (18). However, Muellbauer and Wickens and Molana estimate their models dif ferently than Flavin, because the variables they use to test their consumption equations are all lagged at least one period. Recall that the Flavin model is simultaneous, because she uses A Y D , as one of her test variables. When deriving the reduced form of her two-equation system, the equation for YD is used to substi tute out the current YD term in A YD t. The revision to permanent income due to new in formation provided by current YD (13) cannot be identified and thus is thrown into the error term. Because Muellbauer and Wickens and Molana only use lagged variables to test their models, the income and interest-rate innova tions remain identified by the income and interest-rate equations. Thus, unlike Flavin, they can estimate the coefficients on the innovation terms. Ignoring the interest-rate terms in Muellbauer’s and Wickens and Molana’s model, it is not clear that their test is more powerful than Flavin’s. The presence of AYDtin the con sumption equation gives Flavin a direct test of the impact of current income on current con sumption. If the RE-LC/PI model is rejected, there is some knowledge about what the cor rect alternative may be, or at least in what direction the search for the correct alternative might go, but she cannot test for the impact of the income innovation, an important variable of the null hypothesis. By not adding any cur rent income terms, Muellbauer and Wickens and Molana cannot test for a direct effect of current income on current consumption, but they do have a direct test of the impact of innovations in income. The estimation procedure used by Muell bauer and Wickens and Molana requires two steps. The first step estimates with ordinary least squares (OLS) the simple reduced forms for disposable income and the real interest rate to generate the income and interest-rate innovations and expected values. Muellbauer’s In YD equation uses the first two lags of InYD and lnCt xas the information set. For his real interest-rate equation, Muellbauer argues that apart from seasonal factors, the U.K. real interest rate varies randomly about 6. It was decided not to update Hayashi’s model, because it is not so easily com pared with the Hall and Flavin models. The Wickens and Molana model was not updated either, because it is similar to Muellbauer’s, apart from some ad ditional terms that complicate the esti mation procedure. a constant from the 1950s until the pound ster ling began to float in 1972:IIQ; it follows a random walk thereafter. Wickens and Molana say that a broader information set than one that includes only lagged values of income and real interest rates, should be used with their more general model. They use the first four lags of InYD, InC, r, InA, the latter being the log of real consumer liquid assets, as the information set for both real disposable income and the real interest rate. The second step uses the residuals for the innovation terms and fitted values for the expected value terms in OLS regressions of the consumption equations. Both papers find that their models appear to fit the U.K. data very well. Wickens and Molana do not test the joint life-cycle rational expectations hypothe sis; Muellbauer does by adding the informa tion set variables to the right-hand side of (18) and tests for their joint statistical signifi cance. He finds the additional lagged terms to be significantly different from zero. He con cludes that allowing for stochastic interest rates does not seem to be a major cause for the failure of the simple Hall model to explain U.K. consumption found earlier by Daly and Hadjimatheou (1981). II. Updates of the Aggregate Life Cycle Cum Rational Expectations Model We update the estimates, test the Hall (1978) and Flavin (1981) models, and present esti mates of the Muellbauer model using postWorld War II U.S. aggregate time series data.6 Updating the Hall and Flavin models serves at least four purposes. First, the updates help put the results from Muellbauer’s model in perspective. The importance of allowing for stochastic interest rates is immediately clear. Second, by estimating the models through 1984, we can estimate their stability. Third, it is interesting to know how the 1980s data fit these models. Real output and prices varied over wide latitudes during the 1980s and, hence, offer macroeconometricians a rich set of high-influence data, which may help them estimate coefficients more precisely. It is likely that the 1980s data provide even stronger evidence against the RE-LC/PI model than found by Flavin. Finally, the different models are estimated with different information sets (reduced forms) and different sample periods. It is reasonable to wonder if either the content of the information set or the estimation period has a large influence on the estimates. Our interest in these models does not lie solely in determining whether the RE-LC/PI model is accepted or rejected, although that is a very important consideration. If these models are to be useful for policymaking and forecasting, however, they should be robust to different assumptions about the underlying structure used to derive the reduced forms. The Hall and Flavin models are updated with their original samples, specifications, and estimation techniques. To make the three models comparable, we had to make at least four decisions. The first concerns the specifi cation of the dependent and independent vari ables. Hall uses per capita PCE-nondurables and services, Flavin uses the change in per capita PCE-nondurables, and Muellbauer uses the change in the logarithm of per capita (U.K.) PCE-nondurables and services. The con sumption definition used in these tests is per capita PCE-nondurables and services. Although Flavin’s reasons for ignoring PCE-services may be valid, most of these problems should be eliminated once the data are detrended. The change in the logarithm of consumption and the logarithm of income are used here to facilitate comparison with the Muellbauer spec ification. This logarithmic specification should also minimize heteroskedasticity prob lems. The income definition is real disposable income per capita. The log real per capita in come and consumption data are detrended by their average growth trends over the 1947:IQ to 1984:IVQ period. When the same dependent 7. See Kowalewski (1985) fo r more detail on this point. variable is used, Flavin’s consumption equa tion is, for all practical purposes, the same as Muellbauer’s with constant interest rates. The second decision involves seasonal adjustment of the data. Muellbauer uses seasonally unadjusted data, while Hall and Flavin use seasonally adjusted data. We used seasonally adjusted data to maintain compar ability with other U.S. consumption results. A third choice concerns estimation tech niques. Hall uses OLS, Flavin uses maximumlikelihood to estimate her consumption equa tion jointly with her income forecasting equation, and Muellbauer uses a two-step OLS procedure. The original estimation tech niques used by Hall and Flavin are used to up date their models with the most recent data. Maximum-likelihood is used to estimate Muell bauer’s model, because the computer-generat ed coefficient standard errors produced by the two-step method are incorrect.7 A fourth choice is that of the definition of the real interest rate. Instead of using an ex post real interest rate, Muellbauer uses some thing like an ex ante rate—a nominal interest rate minus an expected inflation rate. He com putes this real rate by subtracting from the nominal rate a fitted value from an inflation equation. This choice of real rate is rather odd, for it means that instead of using an expected real interest rate as his theory requires, he is using an expected expected real interest rate in his consumption equation. It also means that he is using a three-step esti mation process, with the estimation of the in flation equation as the first step. Moreover, the inflation equation uses an information set different from that used for the income and interest-rate equations. A logical extension and correction of his model would be to spec ify separate forecasting equations for the nominal rate and the inflation rate, to use the same information set for all of the equations, and to use the fitted values and residuals from both equations to compute the expected real rate and its innovation. An equivalent strategy employed here is to use an ex post rate, as Wickens and Molana do. This re quires only one forecasting equation. The ex post real three-month U.S. Treasury bill rate, (nominal rate, minus current-quarter com pounded annual actual growth rate in the PCE-nondurables and services deflator) is used as the real interest rate in the estima tions of Muellbauer’s model shown below. Because there is no reason to think that U.S. real interest rates have behaved as random walks during the post-World War II period, the real interest-rate equation for Muell bauer’s model will have information set vari ables as regressors, and these will be the same as those used for the income equation—the first two lags of income, the first two lags of the real interest rate, and the first lag of con sumption. This is a simple extension of Muell bauer’s original information set, which con sisted of the first two lags of income and the first lag of consumption. The estimation results are shown in tables 1 to 5. The data used for the computations con tain revisions through the second revised esti mates for 1984:IVQ dated March 31, 1985. The models in tables 1 to 3 were estimated over their original samples and over 1949:IIIQ to 1984:IVQ. For the re-estimates of Hall’s mod el, the data were not detrended. For the reestimates of Flavin’s model, the consumption and income data were detrended using their average growth rates over the 1947:IQ to 1979:IQ period. When the two models are up dated with the data through 1984:IVQ, the consumption and income data are detrended us ing their average growth rates over the 1947:IQ to 1984:IVQ period, and a dummy vari able is added to control for the credit controls of 1980:IIQ. Detrending biases the test in favor of the random walk hypothesis, because it re moves the main source of correlation from these variables. Detrending may also remove structural correlation between C and YD, again favoring the random walk hypothesis. It unfortunately leaves the trend unexplained. The dummy variable is part of the maintained 8. Serially corre lated errors may not signal a breakdown o f the model, if as Hall argues when ra tionalizing the sta tistically significant stock price index terms, consumers take more than one quarter to assimilate new information and act upon a changed expectation o f life-cycle wealth. hypothesis and is not included among the varia bles included in the test of the RE-LC/PI model. The first table contains OLS estimates of Hall’s model. The first equation shows the reestimates of Hall’s model with only one lagged income term. The coefficients, though differ ent from Hall’s published numbers, yield the same apparent inference: the RE-LC/PI model cannot be rejected. The next equation shows the original Hall model updated through 1984:IVQ. Note that the addition of the 1980s data did not change the conclusion of the hy pothesis test—the coefficient on lagged per sonal income is small, has the wrong sign, and is statistically insignificant. However, the Durbin h-statistic rejects the hypothesis of positive serially uncorrelated errors at better than a 5 percent significance level using a one-tailed test. Because the theory predicts that the error should be white noise, the addi tion of the 1980s data may be signaling a breakdown of the model.8 The third equation contains the change in the detrended log of per capita PCE-nondura bles and services as the dependent variable and the detrended logarithm of real per capita disposable personal income as the income var iable. The estimation period is 1948:IQ to 1977:IQ. Neither coefficient is large, the /-statistics are very low, and the adjusted R 2 is negative. The results change very little when the estimation period is extended through 1984:IVQ; all of the explanatory power of the right-hand side variables comes from the Table 1 Hall Estimates C/ = ao + Cm + ot2 Yt.\ + 0:3 DUM802 #1 + et #2 #3 #4 Chg in detrended log of Chg in detrended log of c NDS/POP NDS/POP NDS/POP NDS/POP Y Y D 72/P O P Y D 72/P O P Detrended log of YD72/POP Detrended log of YD72/POP 4 8:1 Q -7 7 :IQ 4 8 :1 Q -8 4 :4 Q 4 8 :1 Q -7 7 :1 Q 4 9 :3 Q -8 4 :4 Q -0.0376 (-2.2620) -0.0059 (-0.5835) 0.0007 (1.1492) 0.0005 (0.8562) 1.0811 (24.8721) 1.0081 (31.3779) -0.0480 (-1.6283) -0.0008 (-0.0341) -0.0063 (-0.4627) -0.0060 (-0.4902) -0.0068 -0.0131 (-2.3346) 0.0263 Sample ao <X2 C*3 -0.0441 (3.0626) R2 0.9989 0.9994 Durbin h 1.358 1.7327 SER 0.0136 0.0290 Variables: NDS = PCE-nondurables plus services, 1972 dollars. YD72 = disposable personal income, 1972 dollars. POP = non-institutionalized, civilian population. ad j a. Durbin-Watson statistic. NOTE: 1. The variables in equations #3 and #4 are detrended over the 1947:IQ to 1984:IVQ period. 2. The /-statistics are shown below the coefficient estimates. 1.77528 0.0058 1.7460* 0.0056 9. Flavin (1981), proves the equiva lence o f these two procedures in ap pendix II. dummy variable. Thus, Hall’s model can find no evidence to reject the RE-LC/PI model. The results for Flavin’s model (12a) are shown in tables 2 and 3. Only the coefficients of the A YD ( Ain YD) terms (the 0 coeffi cients in equation ( 12a) are shown because only they are relevant for the test of the RELC/PI model. Recall that these terms must be jointly statistically different from zero in order to reject the model. The first equation in table 2 shows the re-estimates of her original specification. Like the updates of Hall’s model, these coefficients are not quantitatively the same as the original estimates; qualitatively, however, they are very similar. The coeffi cient on A Y D t, fio, though fairly large, has a very low /-statistic; of the 0’s, only 0i is sig nificant at better than 5 percent using a one tailed test. The likelihood ratio statistic (LRS) tests the joint significance of the A YD terms. Surprisingly, the RE-LC/PI model cannot be rejected at the original significance level. 10. When the con sumption and in come variables are detrended with their average growth rates between 1947: IQ and 1984:IVQ, the LRS fo r the joint test o f the A In YD terms becomes 13.5, which implies the re jection o f the null hypothesis at about a 10 percent signifi cance level. Table 2 Flavin Re-estimates Var #i #2 0o A YDt 0i AYDt-i @2 A YDt-2 03 A YDt-s 04 A YDt 4 05 A Y D t -5 06 A YDt- e 07 A YDt-i 0.3194 (1.1164) 0.0605 (1.8388) 0.0079 (0.2493) -0.0662 (-1.2940) 0.0415 (0.8088) -0.0081 (-0.1410) 0.0068 (0.2163) 0.0074 (0.2381) 0.2712 (1.4596) 0.0650 (2.3574) -0.0099 (-0.3659) -0.0535 (-1.4499) 0.0136 (0.3915) -0.0082 (-0.1908) 0.0050 (0.1834) 0.0169 (0.6146) 0.0103 2.0101 0.0329 2.0008 11.754 0.0104 2.0521 0.0325 2.0009 17.148 4 9 :3 Q -7 9 :1 Q 4 9 :3 Q -8 4 :4 Q Coef C C Y Y SER D -W SER D -W LR Statistic Sample NOTE: Detrending occurs from 1947:1Q to 1979:1Q. Flavin’s original likelihood ratio statistic, which is asymptotically distributed as X 2( 8 ), is 27.0, significant at better than 0.5 percent. The LRS for the test of equation (1) is only 11.8, significant at slightly better than 25.0 percent. Identical test results are obtained by estimating only the consumption-reduced-form equation with OLS and by testing for the joint significance of the lagged income terms.9 Apparently, the results are sensitive to revi sions in the data and to the use of different trend values for PCE-nondurables and Y D .10 Equation (2) in table 2 updates Flavin’s orig inal model through 1984:IVQ. The 1947:IQ1979:IQ trend values are used to detrend the post-1979:IQ data. Interestingly, the model can now be rejected at better than a 5.0 per cent significance level; the LRS is 17.2, while the X 2 (8 ) cut-off value is 15.5 at 5.0 percent. The coefficient 0 Ois now smaller, but its tstatistic is larger; the coefficient and tstatistic on A in Y D t l are also larger. Moreover, the fit of the equation is improved over the longer period; the standard errors of the two equations are smaller in the longer sample. Thus, as was expected, the 1980s data appear to tighten up coefficient standard errors and help reject the RE-LC/PI model. Equations (3) and (4) in table 3 use the change in the logarithm of per capita real PCE-nondurables and services as the depend ent variable, and the log per capita consump tion and income data are detrended over the 1947:IQ to 1984:IVQ period. They compare to the Hall equations (3) and (4) in table 1. The third equation shows the unconstrained results over the 1949:IIIQ to 1979:IQ sample period. Notice that they are qualitatively similar to those of equation (1); 0 o is about 0.3 and is statis tically insignificant; 0i is large and is statisti cally significant. Testing the joint significance of the A In YD terms yields a LRS of 27.1, which is significant at better than 0.5 percent, Flavin’s original significance level. Note that this result is much stronger than Flavin’s original result, because the consumption vari able includes PCE-services, which Flavin argued would bias the results against the RELC/PI model. The fourth equation shows the estimation results over the 1949:IIIQ to 1984:IVQ sample. Qualitatively, these results are similar to those of equation (3). The LRS of the test of the lagged A In YD terms is now 29.4, greater than the LRS over the 1949:IIIQ to 1979:IQ sample; the standard errors of the equation al so are smaller in the longer sample. Again, it appears that the 1980s data provide additional stronger evidence against the RE-LC/PI model. Tables 4 and 5 contain the estimates of Muellbauer’s models. Only the coefficients on the information set, innovation, and expected interest-rate terms are shown. The dependent variable is the change in the logarithm of real per capita PCE-nondurables and services; detrending of the log real per capita consump- Table 3 Flavin Estimates Using Logs Coef Var #3 #4 0o &YDt 0i AYDn 02 A YD t 2 03 A YD 13 04 A YD a 05 A Y D tz 06 A Y D t6 07 AYDt-i 0.2794 (0.8282) 0.1208 (2.9091) 0.0709 (1.7267) -0.0977 (-1.7462) 0.0577 (0.6548) -0.1296 (-2.7135) 0.0444 (1.1380) 0.0162 (0.4045) 0.2652 (0.9903) 0.1280 (3.3398) 0.0597 (1.6026) -0.0762 (-1.5329) 0.0457 (0.6887) -0.1095 (-2.5909) 0.0459 ( 1.2202 ) 0.0423 (1.1439) 0.0051 1.8636 0.0050 1.9003 0.0098 C SER C D -W Y SER Y D -W LR Statistic Sample 0.0102 1.9942 27.068 4 9 :3 Q -7 9 :1 Q 2.0022 29.360 4 9 :3 Q -8 4 :4 Q NOTE: Detrending occurs over the 1947:1Q-1984:4Q. tion and income data occurs over the 1947:IQ to 1984:IVQ period. Table 4 shows the esti mates of equation (18) without the interestrate terms ii Mr Mand o 2/. The coefficient 61 on the income innovation should be positive, because positive innovations in current income should lead to upward revisions in life-cycle wealth/permanent income and, hence, in consumption. The first equation shows the results using the 1949:IIIQ to 1979:IQ sample. This equation compares to Flavin’s equation (3) in table 3. The coefficient is 61 positive and statistically significant. Surprisingly, the RE-LC/PI model cannot be rejected by this form of Muellbauer’s model, even though Flavin’s model could. The LRS is only 3.8, significant at slightly less than 30 percent. Again, the results appear to be sensi tive to the specification of the test. The second equation in table 4 updates Muellbauer’s model without the interest-rate terms over the 1949:IIIQ to 1984:IVQ sample. As was true of Flavin’s model, Muellbauer’s model without the interest-rate terms fits bet ter with the 1980s data. Moreover, the LRS is now 14.2, significant at better than 1 percent. Again, the 1980s data lead to a convincing re jection of the RE-LC/PI model. Note that the coefficients on the information set variables are the same order both of magnitude and sta tistical significance in equations ( 1 ) and (2); the difference is that the model fits better with the 1980s data. Table 5 contains the estimates of Muellbau er’s model including the interest-rate terms. Recall from equation (18) that S3, the coeffi cient on the expected interest-rate term, is a positive function of the ratio of one, plus the interest rate, to one, plus the rate of time preference; hence, it should be positive. Pre sumably, the coefficient <52 on the interest-rate innovation is negative, since a higher-thanexpected interest rate should cause consum- ers to save more in the current period. Equa tion (3) shows the results over the 1949:IIIQ to 1979:IQ sample. The two interest-rate coeffi- Table 4 Muellbauer Estimates Without the Interest Rate Coef Var <51 YRESID Pi InYDt-x P2 lnY D t -2 p3 lnCt-\ C SER Y SER LR Statistic Sample #1 #2 0.2185 (4.7533) 0.1207 (2.2775) -0.1683 (-3.4897) 0.0418 (1.0391) 0.2191 (5.3016) 0.1557 (3.4233) -0.1585 (-3.5539) -0.0112 (-0.4075) 0.0053 0.0104 3.800 0.0050 0.0104 14.200 4 9 :3 Q -7 9 :1 Q 4 9 :3 Q -8 4 :4 Q NOTE: YRESID is the current income innovation term. Table 5 Muellbauer Estimates with the Real Interest Rate Var #3 #4 61 YRESID 0.2318 (5.2208) 82 RRESID (0.2042) 0.0042 (1.8764) 0.0871 (1.6383) -0.0000 (-0.0005) 0.0006 (2.2494) -0.0024 (-2.1887) 0.2431 (5.8538) -0.0001 (-0.3660) 0.0026 (2.1257) 0.1433 (2.9831) -0.0419 (-0.6306) 0.0003 (1.0161) -0.0017 (-2.6248) -0.0738 (-1.0183) -0.1109 (-1.7685) 0.0049 0.0050 0.0108 Coef S3 E rt_i Pi InYD t_i P2 InYD ,_2 P3 rm Pi r t-2 Ps InC ;_i C SER Y SER r SER LR Statistic Sample 0.0001 0.0111 1.9737 27.800 4 9 :3 Q -7 9 :1 Q 2.0102 26.200 4 9 :3 Q -8 4 :4 Q NOTE: YRESID and RRESID are the current income and interestrate innovations. Ert.\ is the expectation of last period’s real interest rate based on information available last period. cients appear to be small in magnitude, but this is simply a scaling difference because, interest rates are measured in percentage points. The interest-rate innovation coeffi cient <52 is statistically insignificant, while 63 is significant at slightly better than 10 per cent. The LRS for the test of the RE-LC/PI model is 27.8, which is asymptotically distrib uted as X 2(5), and is significant at better than 1 percent. Compared with equation (1) in table 4, the allowance for stochastic interest rates now leads to the rejection of the RE-LC/PI model. Again, the specification of the test has an important effect on the results. Equation (4) in table 5 shows the estimates of Muellbauer’s model with the interest-rate terms over the 1949:IIIQ to 1984:IVQ period. All of the coefficients are estimated more pre cisely, but unlike the previous results, the equation fits the longer period less well. The coefficients 8 2 and 8 3 now have the correct signs and about the same statistical signifi cance as the earlier estimates. The LRS sta tistic for the test of the RE-LC/PI model is 26.2, rejecting the model at better than a 1 percent significance level, but it is a bit smaller than the LRS from the shorter sample period. Nevertheless, the results are qualita tively the same for both estimation periods, unlike the results of the Flavin tests. The worse fit using the 1980s data occurs because the interest-rate equation fits less well in the longer period. This is not surpris ing, given that interest rates behaved so dif ferently in the 1980s than in the earlier period.11 Does this mean that the test is invalid because the equation generating the interest-rate expectations is wrong? This does not seem likely. Although the /-statistics on 8 2 and 8 3 do not provide support for the model, the LRS of the joint significance of the two interest-rate terms in equation (4) is 46.1. Thus, the interest-rate terms are undoubtedly important, even if they are poorly computed. Moreover, it is not clear how quickly interest- 11. The standard error o f the con sumption equation also increased, but this is probably due to the poorer fit o f the interest-rate equation through the cross-equation constraints. rate forecasting models were adjusted in the 1980s. Given the lag in the learning process, the number of quarters for which the interestrate equation may be wrong is probably small er than 20. Even if the interest-rate equation is wrong, it is not necessarily irrational. Finally, the fit of the model did not worsen so much that this is likely to be the sole reason that the RE-LC/PI model is rejected. III. What Has Been Learned? The estimation results provide ample evi dence to reject this form of the RE-LC/PI model during the postwar period, especially when the 1980s data are included. Even though Hall’s specification cannot reject the model, minor generalizations of Flavin and Muellbauer can, and Muellbauer’s specifica tion including uncertain interest rates can reject the model with or without the 1980s data. It would appear that an important assumption for Barro’s neutrality hypothesis does not hold. Unfortunately, this rejection of the RELC/PI model does not offer an explicit alter native as a replacement. As mentioned earlier, these tests cannot distinguish the assumption of rational expectations from that of the lifecycle/permanent income model. All that can be inferred from these tests is that the joint hypothesis can be rejected. Flavin (1985) at tempts to determine whether the rejection of the RE-LC/PI model is due to the assumption of perfect capital markets or to that of the per manent income model. She uses her original model augmented with an equation for the unemployment rate, which is a proxy for the number of liquidity-constrained consumers. However, there are many problems using such a crude variable for such a complex hypoth esis; her tests undoubtedly have little power. Nor do these tests provide many clues about the exact length of consumer spending hori zons, or how the distribution of horizon lengths changes as interest rates, the distri bution of income, or the supply of consumer credit changes. That the distribution of consumer horizon lengths may vary over time is suggested by the increased significance of the likelihood ratio tests when the 1980s data are included. The early 1980s were apparently a time when the distribution of horizons lengths was skewed toward the shorter end, increasing the correlation of aggregate consumption to cur rent disposable income. Additional evidence about changes in the distribution of consumer spending horizons is provided by Kowalewski (1982), who studies the time series behavior of aggregate personal bankruptcy filings in the United States. Personal bankruptcy filings are countercyclical, increasing in recessions and falling in recoveries. For a variety of reasons discussed in the article, it is likely that just before they file for bankruptcy, personal bank rupts have about the shortest spending hori zons of all consumers. Thus, increases in the number of personal bankruptcy filings might indicate a shift in the distribution of consumer spending hori zons towards shorter lengths. In a regression explaining per capita personal bankruptcy fil ings, transitory income had a much larger impact than permanent income, suggesting that liquidity is very important for these financially distressed consumers. The compo sition of consumer portfolios was also signifi cantly related to the behavior of personal bankruptcy filings. Unfortunately, this evi dence is only about one tail of the distribution. It is clear that much work remains to be done before the time series behavior of aggregate consumption is understood. References Barro, Robert. “ Are Government Bonds Net Wealth?” Journal of Political Economy, vol. 82, no. 6 (November/December 1974), pp. 1,095-1,117. Bernanke, Ben S. “ Permanent Income, Liquid ity, and Expenditure on Automobiles: Evi dence from Panel Data,” The Quarterly Journal of Economics, August 1984, pp. 587-614. ______ “ Adjustment Costs, Durables, and Aggregate Consumption,” NBER Working Paper No. 1038, December 1982. Blinder, Alan S. “ Temporary Income Taxes and Consumer Spending,” Journal of Politi cal Economy, vol. 89, no. 1 (February 1981), pp. 26-53. Boskin, Michael J., and Laurence J. Kotlikoff. “ Public Debt and U.S. Saving: A New Test of the Neutrality Hypothesis,” unpublished manuscript, Carnegie-Rochester Confer ence, November 1984. Daly, Vince, and George Hadjimatheou. “ Sto chastic Implications of the Life CyclePermanent Income Hypothesis: Evidence for the U. K. Economy,” Journal of Political Economy, vol. 89, no. 3 (1981), pp. 596-99. Delong, Bradford J., and Lawrence H. Sum mers. “ The Changing Cyclical Variability of Economic Activity in the United States,” NBER Working Paper No. 1450, September 1984. Dolde, Walter. “ Capital Markets and the Relevant Horizon for Consumption Plan ning,” unpublished Ph.D. dissertation, Yale University, 1973. Flavin, Marjorie A. “ Excess Sensitivity of Con sumption to Current Income: Liquidity Constraints or Myopia?” The Canadian Journal of Economics, vol. 18, no. 1 (Febru ary 1985), pp. 117-36. ______ “ The Adjustment of Consumption to Changing Expectations about Future Income,” Journal of Political Economy, vol. 89, no. 5 (October 1981), pp. 974-1,009. Foley, Duncan K., and Martin F. Hell wig. “ Asset Management with Trading Uncer tainty,” Review of Economic Studies, vol. 42, no. 3 (July 1975), pp. 327-46. Friedman, Milton. “ Windfalls, the ‘Horizon,’ and Related Concepts in Permanent-Income Hypothesis,” in Carl F. Christ, et al., eds., Measurement in Economics. Stanford: Stan ford University Press, 1963. ______ A Theory of the Consumption Func tion. Princeton, NJ: Princeton University Press, 1957. (National Bureau of Economic Research General Series 63.) Hall, Robert E., and Frederic S. Mishkin. “ The Sensitivity of Consumption to Tran sitory Income: Estimates from Panel Data on Households,” Econometrica, vol. 50, no. 2 (March 1982), pp. 461-82. ______ “ Stochastic Implications of the Life Cycle-Permanent Income Hypothesis: The ory and Evidence,” Journal of Political Economy, vol. 86, no. 6 (1978), pp. 971-87. Hayashi, Fumio. “ The Effect of Liquidity Con straints on Consumption: A Cross-sectional Analysis,” The Quarterly Journal of Eco nomics, vol. 100, issue 1 (February 1985), pp. 183-206. ______ “ The Permanent Income Hypoth esis: Estimation and Testing by Instrumen tal Variables,” Journal of Political Economy, vol. 90, no. 5 (October 1982), pp. 895-916. Heller, Walter Perrin, and Ross M. Starr. “ Capital Market Imperfection, the Con sumption Function, and the Effectiveness of Fiscal Policy,” The Quarterly Journal of Economics, vol. 93, no. 3 (August 1979), pp. 455-63. Houthakker, H.S. “ The Permanent Income Hypothesis,” American Economic Review, vol. 48 no. 3 (June 1958), pp. 396-404. Kotlikoff, Laurence J., and Ariel Pakes. “ Look ing for the News in the Noise—Additional Stochastic Implications of Optimal Con sumption Choice,” NBER Working Paper No. 1492, November 1984. Kowalewski, KJ. “ Stochastic Interest Rates in the Aggregate Life-Cycle/Permanent Income Cum Rational Expectations Model,” Working Paper 8506, Federal Reserve Bank of Cleveland, October 1985. ______ “ Personal Bankruptcy: Theory and Evidence,” Economic Review, Federal Reserve Bank of Cleveland, Summer 1982, pp. 1-29. Mankiw, N. Gregory, Julio J. Rotemberg, and Lawrence H. Summers. “ Intertemporal Substitution in Macroeconomics,” The Quarterly Journal of Economics, vol. 100, issue 1 (February 1985), pp. 225-51. Mankiw, N. Gregory. “ Consumer Durables and the Real Interest Rate,” NBER Work ing Paper No. 1148, June 1983. Mishkin, Frederic S. A Rational Expectations Approach to Macroeconomics, Testing Policy Ineffectiveness and Efficient-Markets Mod els. NBER Monograph, Chicago: University of Chicago Press, 1983. Modigliani, Franco, and Richard Brumberg. “ Utility Analysis and the Consumption Function: An Interpretation of CrossSection Data,” in Kenneth K. Kurihara, ed., Post-Keynesian Economics. New Brunswick, NJ: Rutgers University Press, 1954. Muellbauer, John. “ Surprises in the Con sumption Function,” Supplement to The Economic Journal, Conference Papers, 1983, pp. 34-50. Okun, Arthur M. “ The Personal Tax Sur charge and Consumer Demand, 1968-70,” Brookings Papers on Economic Activity, 1, 1971. Pagan, Adrian. “ Econometric Issues in the Analysis of Regressions with Generated Regressors,” International Economic Review, vol. 25, no. 1 (February 1984), pp. 221-47. Pissarides, Christopher A. “ Liquidity Consid erations in the Theory of Consumption,” The Quarterly Journal of Economics, vol. 92, no. 2 (May 1978), pp. 279-96. Swamy, P.A.V.B., Roger K. Conway, and Peter von zur Muehlen. “ The Foundations of Econometrics: Are There Any?” Special Studies Paper No. 182, Division of Research and Statistics, Federal Reserve Board, Feb ruary 1984. Swamy, P.A.V.B., J.R. Barth, and P.A. Tins ley. “ The Rational Expectations Approach To Economic Modelling,” Journal of Eco nomic Dynamics and Control, vol. 4, no. 2 (May 1982), pp. 125-47. Tobin, James. “ Government Deficits and Cap ital Accumulation,” in James Tobin, ed., Asset Accumulation and Economic Activity. Chicago: The University of Chicago Press, 1980. ______ , and Willem Buiter. “ Fiscal and Mone tary Policies, Capital Formation, and Eco nomic Activity Vol. 11,” in George M. von Furstenberg, ed., The Government and Cap ital Formation. Cambridge, MA: Ballinger Publishing Co., 1980. ______ , and Walter Dolde. “ Wealth, Liquidity, and Consumption,” Consumer Spending and Monetary Policy: The Linkages, Federal Reserve Bank of Boston Monetary Confer ence Series No. 5, June 1971, pp. 99-146. Tobin, James. “ Relative Income, Absolute Income, and Saving,” in Money, Trade, and Economic Growth, Essays in Honor of John Henry Williams. New York, NY: The Mac millan Company, 1951. Watkins, Thayer H. “ Borrowing-Lending Interest Rate Differentials and the Time Allocation of Consumption,” Southern Eco nomic Journal, vol. 43, no. 3 (January 1977), pp. 1,233-42. ______ “ The Time Allocation of Consumption Under Debt Limitations,” Southern Econom ic Journal, vol. 42, no. 1 (July 1975), pp. 61-8. Wickens, M.R., and H. Molana. “ Stochastic Life Cycle Theory with Varying Interest Rates and Prices,” Supplement to The Eco nomic Journal, Conference Papers, vol. 94, 1984, pp. 133-47. Eric Kades is a research analyst at the Federal Reserve Bank o f Cleveland. 1. See fo r example, Lucas (1972), (1979); Prescott and Kydland (1980); or Long and Plosser (1983). New Classical and New Keynesian Models of Business Cycles by Eric Kades “Not the least misfortune in a prominent falsehood is the fact that tradition is apt to repeat it for truth. ” H osea B allou The alleged demise of classical economics was greatly exaggerated in the Keynesian era after World War II. The supposed death blow was the seeming inability of the purely competi tive model to explain the vagaries of the busi ness cycle. But in the last two decades, a number of articles have demonstrated that fluctuations with many of the central charac teristics of observed business cycles can arise in “ classical” market-clearing models.1 Market-clearing notions are among the strong est in economics, and the New Classical ability to explain business cycles has breathed new life into the equilibrium approach and many of its provocative conclusions. The existence of business cycles is no longer a reason to ring the death knell for classical models. Keynesian models have never had trouble ex plaining business cycles. Observed movements of output and prices have shaped Keynesian thinking first and foremost, and their models have always admitted these facts. Perhaps because of this preoccupation with empirical regularities, general equilibrium microfounda tions for Keynesian economics failed to arise quickly. Much of the New Classical rebellion against Keynesian orthodoxy in the late 1960s and 1970s was understandably inspired by this lack of a strong choice-theoretic basis for the neoclassic synthesis (the IS-LM and Phillips curve model). Economic theorists of all schools have become less and less willing to accept models not derived from explicit maximizing behavior in a general equilibrium setting. Such shortcomings led to premature eulo gies for Keynesian theories; New Classical economists found the inflation of the late 1960s and the stagflation of the 1970s evi dence of the failure of Keynesian ideas and policies. But the theoretical deficiencies have, in large part, been remedied, and indeed, the New Keynesian tradition employs more preci sion and adherence to general equilibrium rigor than the New Classical.2 Both New Keynesians and New Classical theorists either implicitly or explicitly are searching for the central cause or causes of business fluctuations. Economists of both 2. For a New Keynesian example, see Benassy (1976); Malinvaud (1977); Bohm (1978); or Grandmont (1982). For a representative New Classical exam ple, compare exist ence proofs in Dreze (1975) or in van den Heuvel with Lucas (1979) or in Long and Plosser (1983). Although the label “ New Key nesian” is not uncontroversial, I feel its use is warranted. First, Keynes states clearly in the Gen eral Theory that his model generalizes on the classical perspec tive. This is a cen tral point o f this paper, in reference to present-day theor ies. Second, market failures are at the root o f Keynes’s mod el. New Keynesian theory merely fo r malizes insights due in large part to the General Theory. F i nally, the modern authors who devel oped this approach (Benassy, Younes, Grandmont, and Dreze) refer to their models as “ Keyne sian, ” “ neo-Key nesian,, ’ ’ etc. Thus the use o f “ New Keynesian ” is historically accurate. schools agree that many factors are involved, but find the rough equivalence of cycles (in co variances; not in frequencies and amplitudes) striking and believe the essence of the issue can be illustrated in relatively simple models. When examining Classical and Keynesian models of the business cycle, one is weighing the evidence and deciding which fundamental insight best agrees with the data. For Keynesians, the central cause of the bus iness cycle has always been market failure. The formal definition of market clearing equi librium, that prices adjust to the attributes of agents so that trades balance under desired be havior, is employed by all theorists today. This rigorous definition of market clearing did not arise until the 1950s (Debreu and Arrow [1954]), and it was not for another decade that Clower (1965) clarified the Keynesian idea of market failure. This idea was then formalized and rigorously established as valid in a gen eral equilibrium framework (Benassy [1975] and Dreze [1975]), by the New Keynesians in the early 1970s. The basic notion of market failure is that quantities adjust faster than prices. Prices then do not clear markets, and the entire market-clearing house of cards col lapses. Such Keynesian models are compatible with rational expectations and full informa tion. And they do not rely on “ strange” utility or production functions; indeed we will see that, at present, disequilibrium models are more robust than equilibrium models as to the specification of these fundamentals. One way to highlight the difference between the Keynesian and Classical perspectives is to describe their view of the existing market mechanism. Keynesians view this market struc ture as an endowment that, at least over moderate horizons, agents must take as given, much like their endowments of various goods, such as labor, time, assets, etc. Conversely, Classical theorists view the market structure as much more fluid; any possibility for gains from trade between agents (taking into consid eration search and transactions costs) can and will be exercised. This is reflected in a price mechanism that works rapidly and effectively. In Keynesian models, the imperfect market structure causes business cycles. For Classi cal theory, fluctuations must arise from other sources. The essence of New Classical business cycles lies in agents’ intertemporal substitution of consumption and labor in response to technol ogy (supply) or other shocks. Agents desire to smooth their consumption paths and, to achieve this end, substitute between present and future consumption, present and future leisure and, intratemporally, between labor and leisure. Combined with very simple tech nology shocks, such a model can mimic ob served business cycles. Both schools of thought, then, have con structed models that “ explain” business cy cles in that they reproduce the basic empirical regularities of observed fluctuations. Trans acted quantities of all goods exhibit high posi tive correlation over time, and quantity move ments tend to persist in the same direction for many periods. Further, both generate pro-cycli cal real wages. These are the most basic fea tures of observed business cycles. How are economists to choose which model better explains economic fluctuations? Are the two theories observationally equivalent so that it is impossible to determine which truly describes the real economy? This question is important, since Keynesian models call for activist policy to smooth business cycles, while in Classical models these fluctuations are desired paths for the economy. We will demonstrate that the New Classical (NC) model is a special case of the New Keyne sian (NK) model. Thus, the NC model can be distinguished by the restrictions it places on the more general theory. In the decision-theo retic foundations of statistical scientific in quiry, we can state precisely that there will be less risk in working with the NK model, since it places less a priori restrictions on parame ters. And although testing hypotheses on these 3. Lucas (1972, 1979) requires mon etary policy mea sures along with asymmetric infor mation to constantly confound agents to produce cycles. Given the appear ance o f business cycles under an ex tremely wide range o f monetary policy regimes, in all mod ern economies, and fo r hundreds o f years, Lucas ’ model cannot be considered a general explana tion offluctuations. 4. In a discrete time model, prices are set at intervals frequent enough so that ex cess demands do not change within a period. highly abstract models is controversial (due to lack of desired statistics, problems of aggrega tion, and other problems with available data), existing empirical evidence casts doubt on the a priori restrictions of the NC model. We illustrate these points by presenting sim ple, but essentially complete, NC and NK mod els of the economy and business cycles that il lustrate the central forces behind fluctuations in each. We then discuss theoretical and statis tical arguments for and against each model. The models examined are intentionally bare boned; they assume perfect information, ration al expectations, and model only labor and goods markets. No assets exist; money is solely a unit of account. I. Equilibrium Model We choose Long and Plosser’s (1983) equilib rium model of business cycles for its simplic ity; it captures the essence of the New Classi cal explanation of economic fluctuations. Un like earlier models, such as Lucas’, this formu lation requires no monetary authority along with asymmetric information to fool agents and jolt the economy into fluctuations.3 For Fig. 1 Equilibrium model One-Period Solution Leisure 0 Consumption clarity, we assume perfect information and ra tional expectations. Business cycles arise from technology shocks and intertemporal labor, leisure, and consumption substitutions in response to these surprises. In equilibrium models, the market works in stantaneously at every date. Prices, although theoretically exogenous to households and firms, are actually precisely determined by the attributes of these agents. Imagine a rep resentative firm and household with very wellbehaved production and utility functions. For this Robinson Crusoe and Friday economy, we have equilibrium at the tangency of the indif ference curves and production frontier in leisure-commodity space. (See figure 1.) The key point is the equating of prices and marginal tradeoffs. Consumers equate the wage with the marginal utility of leisure; firms equalize wages and labor’s marginal product. Although there are technical compli cations in extending the equilibrium model to a world with many periods, economically this approach reduces to applying these marginal equalities over time, while correcting for interest rates and agents’ time preference. These marginal conditions are equivalent to the traditional notion of efficiency in econom ics (Pareto optimality); full markets insure that all gains from trade are achieved and that exogenous (government) policy measures cannot improve on this outcome. Equilibrium prices at time t, then, are determined precisely by the fundamental nature of agents: endowments and utility (or profit) functions. These basic parameters are completely summarized by excess demand (Z). Excess demands of the agents at time t (Z t) must determine prices continuously for market-clearing equilibrium to hold.4This may seem obscure, but it is important in understanding the nature of New Classical price adjustments. The idea can be lucidly illustrated by the basic functions involved. 5. In a representa tive agent model, with only one con sumer, the interest rate is determined by his or her rate o f time preference, i.e. p = l/(l+r). This would simplify the equations in (5), since 9 and (1+r) would cancel out in the denominators o f (b) and (c). They have been included in (5) to explicitly show the role o f r and p in intertem poral optimization. The excess demands are constructed by the hypothetical process of calculating excess demands (i.e., quantity desired minus endow ment level) at all possible price vectors. So we have the function: (1) u (C „L ,\ where L, represents labor and C, consumption in period t. Instead of a single-period maximi zation problem, the consumer in this model must solve the multi-period problem: (4) (5b) (5c) Wj_ du/dL, P, du/dCh P,.i (3(l + r,)P i du/dC,+1 w ,+1 P (l+ r,)W , du/dCh du/dL,+ du/dL,, p*t = Z,(p*,). Immediately we see that p* is defined by a function of itself. To avoid any time paradox in the determination of p* and the root of Z, these quantities must be determined simul taneously—instantaneous market clearing— at every date. Long and Plosser do not develop their price dynamics in a full general equilibrium model; they limit most of their study to a simple example. We will carry the analysis of the general case further, since it lucidly illus trates some of the central issues in equilib rium cycles. Consumers have an unchanging utility function: (3) (5a) Z, = Z,(p). This function is assumed to have a unique root, p*, which gives an equilibrium. But this equilibrium price vector p is determined by the excess demands. That is: (2) in the budget set at any date, then there can be no corner solution. In this case, the follow ing first-order conditions must hold: max subject to labor constraints in each period. (3 is the discount factor of the representative agent. Although solving this dynamic maxim ization problem in general is not possible, if the utility function is strictly concave and all markets are perfect so that there are no kinks where r is the interest rate, W is the nominal wage, and P is the nominal price of the con sumption good.5 These are the extensions of the marginal conditions to a dynamic setting. Equation (a) /w/ratemporally requires the real wage to equal the marginal utility of leisure; (b) equates trade-offs of consumption over suc cessive periods via the rate of time preference /3 times the price ratio across periods; and (c) requires that the labor/leisure decisions equal the interperiod wage ratio multiplied by the time preference rate. Even though no assets exist in the model, such intertemporal trades are feasible because of the rich market struc ture of the NC model. For example, there exists a contingent futures market for the consumption good in period t+s if a negative technology shock occurs; contracts on this market can be purchased with labor services in any period between t and t+s. Of course, the price on such a market varies over time according to the tastes and technology of the agents. The important point is that the mar kets do exist, and so all trades are possible. It is not yet clear that this model will produce business cycles. Indeed, since the economy is assumed to have a unique and stable equilib rium, w*, the phase diagram (see figure 2) for this model in the real wage w-W/P seems to indicate that cycles will not occur: As mentioned above, Long and Plosser do not attempt to show that cycling occurs in the unrestricted case. NC models have not, in gen eral, been shown to produce cycles. To derive concrete results, they specify a utility func tion that embodies the intertemporal substitu tions necessary for NC business cycles. Long and Plosser continue their argument with specific utility and production functions: (6) u(ChLt) = Q0lnL, + ^ Q ,ln C „ i =I Yi> t+1 = ^ i> t+\L ?<n X ai, t The standard logarithmic utility function has elasticities ©i which are constrained to be non-negative, ruling out inferior goods. Pre Fig. 2 Phase Diagram for Equilibrium Model sumably, if a ©i is zero, the good has some use in production. Otherwise, it is superfluous in the economy, since Long and Plosser assume free disposal. The Cobb-Douglas production function is unusual only in the appearance of Ai,£+i, the stochastic shock to the production of good i in period t. The subscript t + 1 refers to the date of completed production; that is, when it is ready for consumption. X is the vector of goods used as productive inputs. Then, in each period, the consumer maxi mizes expected utility according to: (7) E(U\S,) = E{ 2j8-'[e„l»A , +20,i«c„is,]}, where St = (F „ X ,,/+1). We require maximiza tion under the expectations operator E, since A is stochastic. A shock in the technology for producing a good will obviously change pre sent consumption. We expect co-movement of most goods, since they are all normal—chang es in income call for marginal increments or decrements of each in the equilibrium con sumption basket. Because leisure is also a normal good, some of the gain or loss due to the windfall will be taken in increased or decreased work hours. Co-movement is the first empirical regularity of the business cy cle captured by the NC model. Intuitively, persistence arises as consumers spread unexpected income changes over time as well as over all goods within a period. Sav ings or dissavings due to windfall gains or losses (from shocks) are used to increase or de crease income in many future periods. So, even without serially correlated shocks, we will have persistence in fluctuations. The utility function has been the source of co-movement and persistence in quantity fluc tuations discussed to this point. Production technology is another source of these business cycle characteristics. Since most goods are in puts in the production of some other goods, a technology windfall (disaster) not only in creases (decreases) present and future con sumption of that good, but also, since all goods are superior, some of the windfall (disaster) is used to produce more (less) of all other goods requiring it as an input. Again, it is all part of the smoothing over time, as well as among com modities of any unexpected change in income. This leads to cycles, even in response to serial ly uncorrelated technology shocks. The real wage can easily move pro-cyclically in this model. If tastes are fixed, then any increase in output (which requires more labor input) drives up the real wage required to induce workers to provide necessary labor services. As long as any decrease in labor’s marginal product doesn’t dominate this effect, the real wage will rise. Since increased output is associated with a windfall of some good that can serve as capital (increasing produc tivity), a pro-cyclical real wage seems likely. Although observed, it is beyond the scope of simple non-monetary models like this to pro duce co-movement among price for many or all goods. In an elaborate simulation for a six-sector model of their economy (Long and Plosser [1983], p. 65) observed the paths for sectoral Fig. 3 Simulated Output of Equilibrium Model Logged Output and aggregate output shown in figure 3. Co-movement of different goods appears clearly. The long swings in the time series in dicates higher degrees of autocorrelation than I, suggesting persistence. Long and Plosser’s equilibrium model, then, can generate three of the central aspects of ob served business cycles: quantities in almost all industries move roughly together, output variations tend to persist for many periods, and the real wage moves pro-cyclically. These characteristics arise from intertemporal trade offs to smooth consumption along with shocks to the production function. Long and Plosser emphasize that these are not the only factors in the business cycle, but claim their model is a “ useful benchmark” for evaluating other models. We take this to mean what we said at the outset: they are positing the central driving force of business cycles. They also point out that, in their model, busi ness cycles are preferred paths; any policy at tempting to smooth these fluctutations will be at best Pareto-equivalent with the free market outcome and may well be Pareto-dominated. II. Disequilibrium Model _ Manufacturing -------- --Mining . Services _____ _Construction Transportation/ trade _____ Agriculture Logged Output SOURCE: Long and Plosser (1983, p. 65). The only departure of disequilibrium models from the purely competitive Arrow-Debreu framework is the supposition that quantities may adjust faster than prices. Taken to the limit, this leads to fixprice models in each period with quantity rationing to balance trades. The market system that the economy is endowed with may not permit the perfectly fluid media for trade that exists in classical models. Some mutually desirable trades may simply not be possible under the constraint of an imperfect market structure. Dreze and Benassy (1975) laid the static foundations for this model; dynamic extensions have been numerous (see Bohm [1977] and Kades [1985]). Here, we outline a simple version of a dynam ic fixprice model (as outlined by Grandmont [1982], this is called a temporary equilibrium framework) and show how business cycles arise in it. In contrast to figure 1 above, a Robinson Crusoe and Friday disequilibrium economy can be illustrated, as in figure 4. The price vector p 0is exogenous within each period, so that the unique (under our same assumptions of very well-behaved utility and production functions) Pareto optimal Walra sian price vector p outcome almost never obtains. Under p 0, the consumer will wish to trade to point and the producer to point £, so it is not an equilibrium. Instead, we will have a new type of equilibrium, a fixprice equilibrium. We explicitly demonstrate this new type of equilibrium below. In this out come, in general, one or both agents fail to obtain desired quantities at given prices and so are rationed. Note that, if the exogenous price vector is p, we have an equilibrium model. Here we clearly see that in the static world disequilibrium models are more general than equilibrium models; they allow for both Fig. 4 Basic Structure of Disequilibrium equilibrium and rationing outcomes. NC models maintain that agents will always be able to find or create markets that will yield market-clearing prices. Since the marginal rates of substitution in consumption and the marginal rates of trans formation in production are not equated to prices, this economy lacks the Pareto optimal ity of the NC model. Within such a frame work, it is likely that government policies could improve the welfare of all agents. We will more fully specify a dynamic dis equilibrium model following Malinvaud (1977) and Kades (1985a). Cycles occur in a more general model here than with Long and Plosser; there is no need to adopt specific util ity and production functions. We use L, C, P, W, and w, as in the equili brium model. Consumers are described by a utility function U that is constrained only to be quasi-concave. Our representative consu mer’s sole endowment consists of time that may be “ spent” on either labor or leisure. A simple concave stochastic production func tion, F (L t) +£„ describes the activity of the firm. Consumers maximize utility, and firms maximize profits. Instead of assuming that the very special Walrasian price vector is found, the fixprice approach imagines that the price vector is truly parametric at a given trading date and will be Walrasian only by accident. Between dates, the price vector moves according to the so-called law of supply and demand; excess demand for a good in period t (and possibly in previous periods) tends to pull prices up, while excess supply causes prices to fall. This does not restore the auctioneer and the instantaneous achievement of the equili brium price vector. It more modestly posits that market forces work in the right direction and possibly with lags. Thus, there are other forces beyond current excess demands Z(p) (and specifically, its root) that may enter into the function determining prices. It is already easy to illustrate that, dynam ically, NC models are a special case of NK models. The most general form of the price equation is: (8) p, = i4(x), where x is the vector of all conceivable state variables in the world. The object of theory is to pare down the size of the vector x as far as Fig. 5 Firm’s Demand (Production Function) Fig. 6 Household’s Demand (Consumption Expansion Line) possible without ignoring anything of impor tance. NC models reduce the dimension of x to only contemporaneous excess demands Z t\ (9) P '= f(Z ,). They further require that f(0) obtain at every date. NK models allow for a broader range of variables to enter, such as lagged excess demands or even lagged prices. Here are some examples: (10) p t = g[\ Z t + (l-A )Z / J p t = h [\ f(Z ) + (1 -\)p M] Further, the Z t’s are allowed to take non-zero values. If the New Classical special case held in real ity, proper econometric estimation of equation (12) would find that k was statistically indis tinguishable from 1. And only this singular result could yield direct evidence that New Keynesian theories were over-parametrized. Returning to the outline of the model, there is no reason to believe that Walrasian supplies and demands will balance at an arbitrary price vector in a disequilibrium world. More struc ture must be imposed here to define demands and to determine actual transactions. The most basic requirement imposed in fixprice models is voluntary trade: no agent is ever forced to trade (supply or demand) more of a good than he desires—what his preferences dictate. Since markets do not clear and we disallow forced transactions, agents will have to be rationed in quantities at the given price vector to balance trades. This model requires a new definition of “ equilibrium.” Fixprice equilibrium means the maximiza tion of quantity-constrained utility and profit functions with trades balancing. Disequilib rium Benassy (1975) demands, which we will refer to (following the ideas of Clower) as effec tive demands, are derived from considering all constraints except the constraint in the indi vidual market where demand is being formed. We denote them with a + superscript; they are defined from the maximization problems: (11) Households: L h+ = M AX u(L,C,w) subject to WL < pC Ch+= M AX u(L, C,w) subject to WL < pC Firms: L,+ = M AX r(L, C, w) subject to C < F(L) Cf+ = M AX r(L, C, w) subject to C < F(L), where C and L are perceived constraints on other markets, and r is the profits function. Benassy showed that, when solved, these de mands yield balanced trades while simultane ously determining perceived constraints. The perceived constraints are the minimum of the effective demands when the system of simul taneous demands is solved. Thus agents’ max imizing decisions under these constraints bal ance in the aggregate, yielding a fixprice equi librium with rationing. The rationing mecha nism is usually assumed to be stochastic. Formally, however, this point needn’t be ad dressed in representative agent models. We develop some graphs to represent this model. (See figures 5 and 6.) We will be using graphs to show the behavior of the household and firm in the trade space (L,C). The firm simply obeys “ efficient production” in this Fig. 7 Fixprice Static Equilibrium model and always produces somewhere along the production function C = F(L). However, the firm will never produce beyond its Walra sian point (Lf*, CJ*) under the given wage and price (the exogenous parameter x) since, beyond this point, the exogenous wage ex ceeds labor’s marginal product. The shape stems from our assumptions on the produc tion function. The household obeys “ efficient consump tion” ; it consumes along a line going through the origin (no work, no pay) whose slope is dictated by the real wage rate. The household will never work beyond its notional quantities (L h*, Ch*) since, beyond this point, the marginal utility of the good falls below the marginal utility of leisure. To determine the fixprice equilibrium, we combine the two curves. (See figure 7.) Beyond the possibility of a Walrasian equil ibrium (WE) when notional points coincide, there are two possible outcomes to this model. If consumers are rationed in selling labor and firms in selling the consumption good, then we have general excess supply. This has been labeled a Keynesian equilibrium (KE). If gen eral excess demand prevails, we have an infla tionary equilibrium (IE). Thus, disequilibrium Benassy demands give rise to a much broader range of market out comes than Walrasian models, where Z t=0 in all markets. Even at an arbitrary price vector, Walras’ Law holds for New Classical de mands: excess demand in one market is, by definition of budget constraints, balanced by excess supply in another market. General ex cess supply or demand cannot arise even hypo thetically in an equilibrium model. Clower correctly stressed that the key to disequili brium models must be to establish a rigorous framework within which Walras’ Law did not hold. This is one way to describe the main accomplishment of New Keynesian theorists. The dynamics of our disequilibrium model are very simple, since there is only one state variable, the real wage w. In the state space R +, we have a unique value of w, w*, that gives a Walrasian equilibrium. But the movement of the real wage in KE and IE regions (on either side of the Walrasian equil ibrium) is, at first inspection, undetermined. In the case of KE, labor is in excess supply in terms of effective demands, so the nominal wage should fall. But the commodity is also in excess supply, and so its price also should drop. Qualitatively, it seems difficult to determine the direction of real wage move ments. The same holds for IE, where we have general excess (effective) demand. Elsewhere (Kades 1985b), it has been shown that it is likely that steady states exist in both Fig. 8 Phase Diagram for Disequilibrium Model IE KE “Uw* Fig. 9 Simulation of Disequilibrium Model Logged output Time the KE region and the IE region. How does this occur? In Keynesian steady states, the nominal price of both labor and the good fall at the same rate in the price (vector-valued) function. Then the real wage rate is unchang ing, and since it is the only state variable in this simple model, a steady state obtains. A symmetric case explains a steady state in the IE region. Further, all Keynesian steady states of the model are stable (Kades 1985b); in a one dimensional model, this implies uniqueness. Since lagged demands are generally included, the WE will almost never be an equilibrium (i.e., it is a measure zero event). Inflationary steady states may be either stable or unstable. Figure 8 presents a typical phase diagram for this system. This system can easily give rise to cycles in the presence of exogenous shocks to the pro duction function. The system can move further and further into either the KE region (a recession) or the IE region (boom). It can move either towards a stable or away from an unstable node until any type of shock moves the system to the other side, changing the cycle. The unstable IE effectively marks the border between the two regimes. White noise shocks can produce outcomes much like those observed in real economies. Figure 9 shows a simulation of this model similar to Long and Plosser’s for aggregate output only. Further, this model fully captures the observed co movement of prices and quantities. By mak ing C a vector, it is easy to show that different quantities move together in the model. So the fixprice/disequilibrium paradigm ex plains the most fundamental aspects of ob served business cycles, and does so without re course to special utility and production func tions. The only reason for such fluctuations in the model is the general inability of the market mechanism to always find the marketclearing price vector. This economy is en dowed with a cumbersome market structure that may or may not accurately reflect reality. III. The Evidence It is difficult to directly test hypotheses on whether or not all markets clear. But we can heuristically and formally examine evidence and arguments on a number of issues and measure the degree to which equilibrium and disequilibrium business cycle models agree with observation and rigorous thought. The Great Depression stands as perhaps the most memorable single twentieth-century cyclical swing. The ability of a business cycle theory to plausibly explain this experience is important in establishing its credibility. Therefore, we first discuss the extent to which both models can explain this event. Pigou and other Classical theorists in the 1930s blamed the Great Depression on an ex cessive reservation wage rate demanded by laborers. Thus for them, recessions were caused by a market imperfection in labor mar kets. In a sense, this view stands closer to disequilibrium paradigms, although the classi cal notion of market failure differs substan tially from the New Keynesian view discussed above. For many early Keynesians, this was also seen as the cause of the Great Depression; they disagreed with Classical theorists only on the effectiveness of expansionary policies. Today’s New Classicals must argue that recessions occur when low wages are ex pected; workers then find leisure less costly in terms of wages foregone and bide their time until renumeration rates improve. But can the Great Depression best be explained as a multi year withdrawal from labor markets by most Americans because they expected an eventual wage rise? The other explanations that New Classical theory can offer seem no more cred ible. One is that the utility function of most laborers called for a “ . . . spontaneous out burst of demand for leisure . . . ” from 19291939. Another possibility is that a large nega tive shock to production technology was responsible, but then the problem becomes specifying the source of this shock. New Keynesian explanations of the Great Depression are likewise unconvincing. Iron ically, the most prominent possibility is due to Milton Friedman, a theorist not usually asso ciated with Keynesian ideas. Friedman and Schwartz (1961) argued that a major cause of the Great Depression was the decline in the money supply from 1929-1933. In a slightly modified version of our New Keynesian model with money (Malinvaud 1977), it can be shown that low money-growth rates (or a for tiori money stock declines) are associated with Keynesian recessionary outcomes. But farreaching questions have been raised about this evidence (Temin 1975) and it is not clear which way causation runs between money and output. Further, as argued in footnote 3, cycle theories based on monetary phenomenon are less robust than real theories since cycles have occurred under a wide range of mon etary systems. Perhaps monetary factors con tributed to the severity of the Great De pression, but their role must be explicitly tied into a general model of cycles to provide a satisfactory story. Like New Classical theor ies, New Keynesian explanations may point to some particularly violent shock as the root cause of the Great Depression, but then the difficulty becomes uncovering and explaining the shock. No convincing explanation has been presented. Although some economists find merit in these heuristic arguments, they are based on vague notions and “ stylized facts,” and lack precision. In a formal econometric study, Man kiw, Rotemberg, and Summers (1985) test the first order conditions in equation (5) for a util ity function more general than Long and Plosser’s. That is, they test the first-order condi tions of consumers’ maximization in the NC model. Although not sufficient, the first-order conditions are still necessary for any interior solution; if they are rejected, then the model can be rejected. There are, of course, difficult questions of aggregation in treating national data as if it is created by a representative con sumer. No consensus on a solution to this issue exists, and this methodology is, at pre sent, the de-facto standard for empirical work. Mankiw, Rotemberg, and Summers find that the data (NIPA) reject the hypotheses, that these maximizations are carried out by consumers. None of the three over-identifying restrictions in equation (5) placed by equilib rium models is supported by the data. Further, the rejections occur for almost all permuta tions of the specifications of the hypothesis tests: separable or non-separable utility, annual, or quarterly data. Indeed, many of the restrictions actually force the shape of the utility function to be convex, in which case a maxima would occur at a corner and the Clas sical tangency conditions illustrated in figure 1 could not hold. When the utility function is concave, either leisure or “ consumption” (NIPA) becomes an inferior good—which like convexity casts serious doubt on the model. Simultaneous estimation of all three restric tions in (5) is similarly rejected and produces either a convex utility function or inferiority of either leisure or consumption. This rejection can be interpreted in two ways. Mankiw, Rotemberg, and Summers ar gue that the data show that markets (both la bor and capital) fail to clear. There is another Fig. 10 Corner Solution for LiquidityConstrained Consumer possibility: the structure of the utility func tion may be such that intertemporal substitu tion effects are very weak. In this case, a radi cally different utility function must be speci fied to dovetail with observation. At any rate, either explanation leads us to question Long and Plosser’s equilibrium paradigm of busi ness cycles. It seems that either markets fail to clear, or that substantial intertemporal elas ticities of substitution do not exist; both inter pretations of the evidence reject this NC ex planation of business cycles. The disequilibrium model cannot be reject ed by any such hypotheses concerning the structure of the utility function; it requires only that the utility function be quasi-con cave. Beyond this, the disequilibrium model is robust to the form of the utility function. Apart from rejecting the restricted form of the utility function needed to generate equilib rium business cycles, there is also strong eco nomic evidence that key markets do not clear. Specifically, we shall discuss evidence that capital (lending) markets fail to clear. Recall from the first-order conditions in the equilibrium model (5) that the interest rate ap pears in consumers’ decisions just as in any other price. Equilibrium models require that agents can buy or sell as much of a good as they want at a uniform price, subject only to their endowment constraint. This constraint prevents any kinks from existing in the agents’ budget sets so that, with a concave utility function, no corner solutions to maximization problems exist. Keynesians (Old and New) have long argued that consumers, in reality, face liquidity con straints: either they cannot borrow at all against future income or they must pay an in terest rate greater than the rate they receive for lending funds (even accounting for risk premia). Figure 10 shows that if agents lend at one price, but borrow at another, they are likely to solve maximization problems at cor ners of their budget set. Here, the equality of prices and intrapersonal utility trade-offs breaks down, and the economy may no longer be efficient. 6. It is interesting to note that asset mar kets are almost al ways assumed to more closely approx imate the competi tive ideal than other markets. I f the data show that these markets fail to clear, then it seems du bious to assume that labor and goods markets clear. Agents are endowed with e = (e 1 , e2) of a good in periods one and two respectively. The interest rate for borrowing in period one is r 1; while the lending rate is less, r 2. With a con cave utility map, it is then immediately apparent that a corner solution can occur. Strong evidence exists that such liquidity constraints have been binding for significant numbers of American consumers. Fumio Hay ashi (1985), modifying an idea originally ap pearing in Kowalewski and Smith (1979), uses cross-sectional data and divides consumers in to high-and low-savings groups. He assumes that high-savings households are unlikely to be liquidity-constrained, so they may be used as a control group to be compared to other (po tentially liquidity-constrained) households. By estimating consumption behavior for each group separately, and then by comparing the two parameter sets, Hayashi finds a signifi cant difference that can be explained by the existence of liquidity constraints. Although there are other explanations for the result, they require the rejection of either the perma nent income hypothesis or of market clearing. Since both market clearing and the permanent income hypothesis embody the New Classical idea of the markets’ abilities to smooth con sumption over time, this interpretation too, casts doubt on the equilibrium business cycle model. Flavin (1981) and Kowalewski (1985) provide time series evidence that liquidity constraints have persistently shaped agents’ budget sets in the postwar American economy. On the other hand, the disequilibrium model is robust to either interpretation of Hayashi’s results. If liquidity constraints do exist, they are an instance of the imperfect markets of New Keynesian theory.6 If we view the results as a rejection of all utility functions that give rise to permanent-income consumption paths, we already know that the NK model is not subject to this criticism. In discussing the compatibility of both mod els with observed business cycles, we have examined only three central patterns: the co movement of different quantities, the persis tence of trends, and the positive correlation between quantities and the real wage. But there are other empirical regularities in busi ness cycles that both models should similarly mimic if they are to be adequate representa tions of the central force in business cycles. Although they were raised by Arthur Okun (1980) in objection to Lucas’s equilibrium mod el (Lucas 1972), they also point to shortcom ings in Long and Plosser’s model and in NC models in general. First, many secondary aspects of labor mar kets (beyond pro-cyclical wages) are at odds with the NC model. Productivity may or may not be pro-cyclical in the NC model. It depends on the size of the technology shocks and on the intensity of the disutility of labor. But observed productivity is strongly pro-cyclical. In non-market clearing models, this pheno menon is explained by implicit contract the ory, where workers are insured against unem ployment by their employers in return for a lower wage. When demand slackens, there are no layoffs; with the same amount of labor and less production, productivity must decline. As demand improves, the same work force is called on to produce more; hence, productivity increases. Implicit contract theory comprises one market imperfection that could be the fundamental source of fixed prices (wages) in the short run. The market clears by a non price mechanism. No institutional factor (that is, exogenous parameter) explaining marketclearing could produce pro-cyclical productiv ity in Long and Plosser’s model. Similarly, quits induced by pro-cyclical factors, counter cyclical layoffs (moreover, the existence of layoffs, which involve rationing the sale of labor), and wage increases in recessions seem inexplicable in the present NC model. Although Long and Plosser examine and dis cuss only the consumer side of their model, firms as well as the household may seek to smooth over time their objective—profits. One rationale for such behavior is that since house holds own the firms, smoothing profits is simply one part of smoothing income. This mo tivation is superfluous in an NC model, since in market-clearing models economic profits have, by definition, a zero expected value in each period (under the usual assumption of constant returns to scale). However, we ob serve very large pro-cyclical fluctuations in profits. New Classical theorists must explain Fig. 11 Continuous Market Clearing with a Unique Stable Equilibria Shock response <C=) Wf -- u_-- u--- W2 W3 Wo = W0 c> Shock Fig. 12 Failure of Continuous Market Clearing with Multiple Equilibria c> o <=> u W i UW2 o * = > W Move in opposite directions to initial shock 1 parameter correlation why the value of entrepreneurial talent and risking capital fluctuate so sharply with the business cycle to lend credibility to their para digm. Conversely, pro-cyclical profits exist under implicit contracts in a New Keynesian framework, since wage costs are constant while productivity varies with business cy cles. However, this criticism must be tem pered by remembering the substantial contro versies in defining and, moreover, in measur ing economic profits. Finally, Fisher (1984) has raised a methodo logical objection to the equilibrium paradigm. In response to any shock, these models re quire that prices adjust so rapidly—almost in stantly—that agents never face disequilib rium prices. However, even in the world of physics, adjustment to a new shock takes time, and a mechanical system must move out of one equilibrium before a new rest state is attained. Even in the case of a unique equilib rium, New Classical dynamic behavior vio lates the usual properties of differential equa tions in avoiding disequilibrium, but we can imagine the shock (£t) and the real wage (wt) move in tandem precisely to produce continu ous market clearing. (See figure 11.) If NC models contain a dynamic structure as rich as the New Keynesian (to avoid the ne cessity of specifying a restricted class of util ity functions to produce cycles), then the hy pothesis of continuous market clearing cannot be maintained. (See figure 12.) The system must move through the unsta ble disequilibrium y and cannot give continu ous market clearing. It is not clear why New Classical theorists feel that economic adjust ments can be approximated by instantaneous movements from one equilibrium to another. They must provide an explicit, testable mech anism for this behavior before it can be used convincingly. Disequilibrium dynamics call for the economy to adjust along paths more in line with established notions about change over time. IV. Conclusion References Arrow, Kenneth, and Gerard Debreu. “ Exist Both equilibrium and disequilibrium theories can construct model economies that mimic the ence of Equilibrium for a Competitive Econ basic behavior of real business cycles: strong omy,” Econometrica, vol. 22, no. 3, (1954), co-movement among quantities of different pp. 265-90. goods, persistence of quantity movements in the same direction for many periods, and pro Benassy, Jean-Pascal. “ The Disequilibrium cyclical real wages. But existing NC models Approach to Monopolistic Price Setting and cannot explain other aspects of observed busi General Monopolistic Equilibrium,” Review ness cycles, such as pro-cyclical productivity of Economic Studies, vol. 43, no. 1 (Febru or other observed characteristics of the labor ary 1976), pp. 69-81. market. Further, evidence exists that capital markets do not clear. Finally, the data reject ________ “ Neo-Keynesian Disequilibrium the New Classical utility function exhibiting Theory in a Monetary Economy,” Review of strong intertemporal substitutions. Without Economic Studies, vol. 42, no. 4, (October such a utility function, the model has not been 1975), pp. 502-23. shown to produce persistent output cycles. The NK model is robust to most of these Bohm, Volker. “ Disequilibrium Dynamics in a criticisms. It requires no specific utility func Simple Macroeconomic Model,” Journal of tion to generate cycles, it fits the observed Economic Theory, vol. 17, no. 2, (April regularities of business cycles more fully, and 1978), pp. 179-99. it employs a more general model of price move ments over time. However, like the NC model, Clower, Robert. “ The Keynesian Counterrevo it provides no convincing explanation of the lution: A Theoretical Appraisal,” in F.H. Great Depression. Since equilibrium models: Hahn and F.P.R. Brechling, eds., The The (1) Comprise a subset of disequilibrium ory of Interest Rates. London: Macmillan, models; 1965. (2) employ identifying restrictions that are not empirically validated; and Dreze, Jacques. “ Existence of an Exchange (3) require nonstandard dynamical adjust Equilibrium under Price Rigidities,” Inter ments; it appears that, at present, despite this national Economic Review, vol. 16, no. 2 shortcoming, New Keynesian theories provide (June 1975), pp. 301-20. a better paradigm of the business cycle. Fisher, Franklin M. Disequilibrium Founda tions of Equilibrium Theory, Econometric Society Monograph in Pure Theory #6, New York: Cambridge University Press, 1984. Flavin, Marjorie. “ The Adjustment of Con sumption to Changing Expectations about Future Incomt," Journal of Political Econ omy, vol. 89, no. 5 (October 1981), pp. 974-1009. Friedman, Milton, and Anna Schwartz. A Monetary History of the United States. Prin ceton, NJ: Princeton University Press, 1963. Long, John B., and Charles I. Plosser. “ Real Business Cycles,” Journal of Political Econ omy, vol. 91, no. 1 (1983), pp. 39-69. Grandmont, Jean-Michel. “ Temporary Gen eral Equilibrium Theory,” in Kenneth J. Arrow and Michael D. Intriligator, eds., Handbook of Mathmatical Economics: Volume II. Amsterdam: North-Holland Pub lishing Company, 1982, pp. 879-922. Lucas, Robert E., “ An Equilibrium Model of the Business Cycle," Journal of Political Economy, vol. 83, no. 6 (December 1975), pp. 1,113-44. Hayashi, Fumio. “ The Effect of Liquidity Con straints on Consumption: A CrossSectional Analysis,” Quarterly Journal of Economics, vol. 100, issue 1 (February 1985), pp. 183-206. Kades, Eric. Dynamics of Fixprice Models, Working Paper 8505, Federal Reserve Bank of Cleveland, September, 1985a. ________ Fixprice Models for Dynamic Studies, Working Paper 8504, Federal Reserve Bank of Cleveland, August, 1985b. Keynes, John Maynard. The General Theory of Money, Interest, and Employment. London: MacMillan, 1936. Kowalewski, Kim. The Stochastic Interest Rates in the Aggregate Life Cycle Permanent Income Cum Rational Expectations Model, Working Paper 8506, Federal Reserve Bank of Cleveland, October 1985. ________ , and Gary Smith. “ The Spending Behavior of Wealth- and Liquidity-Con strained Consumers,” Cowles Foundation Discussion Paper #56, New Haven: Yale University, 1979. ________ “ Expectations and the Neutrality of Money,” Journal of Economic Theory, vol. 4, no. 2 (April 1972), pp. 103-124. Malinvaud, Edmond. The Theory of Unem ployment Reconsidered. Blackwell: Oxford University Press, 1977. Mankiw, N. Gregory, Julio Rotemberg, and Lawrence H. Summers. “ Intertemporal Substitution in Macroeconomics,” Quar terly Journal of Economics, vol. 100, issue 1 (February 1985), pp. 225-51. Okun, Arthur. “ Rational-Expectations-withMisperceptions as a Theory of the Business Cycle,” Journal of Money, Credit, and Bank ing, vol. 12, no. 4 (November 1980, Part 2), pp. 817-25. Prescott, Edward, and Finn Kydland. “ Time to Build and Aggregate Fluctuations,” Econometrica, vol. 50, no. 6 (November 1982), pp. 1,345-70. Temin, Peter. Did Monetary Forces Cause the Great Depression? New York, NY: W.W. Norton, 1976. The working paper series is published by the Research Depart ment o f the Federal Reserve Bank o f Cleveland to stim u late discussion and critical comment. Copies o f working papers, either future or past issues, are available through our Public Inform a tion Department, Federal Reserve Bank o f Cleveland, P.O. Box 6387, Cleveland, OH 44101: (216) 579-2047. Working Paper Series 8501 A Bureaucratic Theory of Flypaper Effects Gary Wyckoff 8502 Federal Reserve Credibility and the Market’s Response to the Weekly M l Announcements William Gavin and Nicholas Karamouzis 8503 Forecasting GNP Using Monthly M l Michael Bagshaw 8504 Fixprice Models for Dynamic Studies Eric Kades 8505 Dynamics of Fixprice Models Eric Kades 8506 Stochastic Interest Rates in the Aggre gate Life Cycle/Permanent Income Cum Rational Expectations Model Kim J. Kowalewski 85 0 7 Forecasting and Seasonal Adjustment Michael Bagshaw 85 0 8 The Ohio Economy: Using Time Series Characteristics in Forecasting James Hoehn and James Balazsy 85 0 9 Total Factor Productivity and Electric Utilities Regulation Philip Israilevich