ALFRED MAUSSNER Otto-Friedrich UniversityBamberg Bamberg, Germany MARKUS KUEPPERS University of Cologne Cologne, Germany
Partisan Profiles in Presidential Policies: Methodological Issues and Empirical Evidence Reconsidered* In two recent articles Zalesld (1992) and Siebrand and Swank (1994) look for partisan profiles in U.S. economic policy. Za]eski finds no significant difference between Republican and Democratic policies to fight unemployment and inflation. Siebrand and Swank's estimates indicate a marked difference between Republican and Democratic administrations. We show that both papers suffer from methodological shortcomings and that their estimates are not comparable. Consistent and comparable estimates of both models reveal no notable partisan profile. (JEL E24, E31, E61, E65, H10)
1. Introduction The view that governments act as if they maximized a social welfare function has long been criticized, and public choice theory provides alternative models of government behavior. According to the vote-maximizing model, economic policy serves as a means to ensure reelection. Thus, policy outcomes reflect the preferences of the median voter. The partisan model, instead, holds that governments pursue policies that reflect the preferences of the party whose members are in office. More specifically, and with respect to the U.S. two-party system, it states that Republican led administrations will favor the fight against inflation whereas Democratic presidents will put more emphasis on tile reduction of unemployment. In a recent study of quarterly U.S. data Zaleski (1992) finds no support for the partisan model. His approach is based on a simple model: The policy maker minimizes a quadratic loss function in unemployment and inflation subject to a given short-run Phillips curve. The model predicts that the *The authors are grateful to four referees of this journal.
Journal of Macroeconomics, Spring 1998, Vol. 20, No. 2, pp. 387-395 Copyright © •998 by Louisiana State University Press 0164-0704/98/$1.50
387
Alfred Maussner and Markus Kueppers
inflation/unemployment ratio reflects the weight presidents assign to the arguments of the loss function. Siebrand and Swank (i994) extend the Zaleski model. They add instrument costs to the loss function and recover the weights from a regression of unemployment on lagged unemployment and inflation. Their results, in contrast, reveal a significant partisan profile in Presidential economic policies. We argue that both papers suffer from serious methodological shortcomings. Their estimates are biased and depend upon a priori restrictions that the data do not satisfy. Furthermore, Siebrand and Swank's estimator of the president's preferences differs markedly from that employed by Zaleski. We estimate both models consistently. To furnish comparison, we recover the weights of the loss function from Zaleski's (1992) model in two different ways, based on his own approach and that taken by Siebrand and Swank (1994), respectively. The Zaleski estimator reveals no significant difference between Republican and Democratic economic policy to fight unemployment and inflation. The Siebrand and Swank estimator of both models indicates a marked inflationary bias of the Carter administration. Except that this estimator, too, uncovers no significant partisan profile. The next section shortly presents the model of government behavior. In Section 3 we highlight the deficiencies of both the Zaleski and the Sicbrand and Swank estimates. Section 4 presents our estimates and Section 5 concludes.
2. The Model of Government Behavior Siebrand and Swank (1994) consider a policy maker who chooses the rate of unemployment U, in each period in order to minimize the loss function a(U, - kU*) e + bp} + c(U, - Ut_l) 2 w i t h : a , b , c > O, 1 = a + b + c , k E [0, 1],
(1)
subject to tile expectations augmented Phillips curve
p,=p~+a(u~-u*)+~s,,
a<0,
~>--0.
(2)
The parameters a, b, and c reflect the weights attributed to the deviation of unemployment from its target rate kU*, to the rate of inflation Pt, and to the adjustment of the rate of unemployment (Ut - Ut-1), respectively. The 388
Partisan Profiles in Presidential Policies
model of the Phillips curve assumes that inflation depends upon expected inflation p~, the deviation of unemployment from the natural rate of unemployment U*, and tile rate of change of energy prices s t. Adjustment costs represent the policy maker's imperfect control over unemployment. Instead of modeling the uncertainty associated therewith explicitly, instrument costs can be included in a quadratic objective function (Gordon 1976). Zaleski (1992) ignores these costs, thus, he considers a special version of the present model with c = 0. Due to this simplification the first-order condition for a minimum of (1) can be rearranged to read a
-
b
a
Pt
(3)
G - ku*"
Hence, given an estimate of the slope of the Phillips curve, of the natural rate of unemployment and an assumption about k, the preference ratio can be inferred from observed rates of inflation and unemployment. Zaleski sets k = 0, adopts the estimate of a = -1.288 from Smyth, Washburn, and Dua (1989), and computes (a/b) as the average of the r.h.s, of (3) for each presidential period taking into account a two-quarter operational lag. If e > 0, the first-order condition implies a
G-
l~
kU* +
1 -a -b 1- b
G-1
b 1- b aPt
(4)
Siebrand and Swank also restrict k to zero, set e~ = - 1.288, and estimate Equation (4) with nonlinear least squares.
3. Methodological Issues Our fundamental reservations with the two papers rest on four main points. First, there is virtually no evidence to assume that all administrations of the three decades spanning from Kennedy to Bush faced the same unemployment-inflation trade-off that was estimated by Smyth, Washburn and Dua (1989) for the relatively short period from 1978 to 1988. Second, Smyth, Washburn and Dua (1989) estimate a Phillips curve slope based on annual data. Zaleski as well as Siebrand and Swank implement this coefficient in their estimations established on quarterly data. By way of doing so both papers link a government that maximizes its objective function on a quarter-to-quarter base to an electorate that revises its expectations only once a year. We argue that it is logically inconsistent to operate with a 389
Alfred Maussner and Markus Kueppers disparate time-index in the objective function and the constraint set, since in that case the variables in the objective function differ from those in the Phillips curve constraint. These problems take us to our third point. The model of government behavior outlined above consists of two equations, (2) and (4) (or [3] in the case of c = 0). Both inflation and unemployment are endogenously determined as linear functions of U*, Ut- 1, Pt- 1, and st. Hence, it makes no sense to estimate a traditional Phillips curve that considers unemployment as exogenous. By the same token, the estimates of a and b from Equation (4) are biased, since the rate of inflation is not exogenous. If we assume that both Equations (2) and (4) contain error terms that are contemporaneously correlated we should apply a systems of equations approach to estimate the parameters of interest. In the system of equations (2) and (4), however, Equation (2) is not identified if instrument costs are excluded, c = 0.1 In order to recover the slope of the Phillips curve additional restrictions need to be imposed. In the following we shall assume that the error term in the Phillips curve is not correlated with lagged unemployment. This gives rise to an instrumental variables estimator of Equation (2). Finally, Siebrand and Swank employ an estimator whose probability distribution differs markedly from the distribution of the estimator employed by Zaleski; standard errors pertinent to Siebrand and Swank's model do not justify a rejection of Zaleski's approach. The estimation of (2) and (4) provides the preference parameters and their estimated standard errors. Zaleski's procedure to obtain standard errors is based on the implicit assumption that the r,h.s, of Equation (3) equals (a/b) plus a serially uncorrelated error term with finite variance ~2. According to the central limit theorem this assumption implies that the mean of n observations of (ap]Ut) is asymptotically normally distributed with mean (a/b) and variance c52/n (see Greene 1993, 104). Thus, estimated standard errors and the t-distribution may be used for hypotheses testing. Our econometric model introduces a unifying framework which encompasses both the Zaleski and the Siebrand and Swank approach and subsequently it presents an estimator that has the desirable properties.
4. Empirical Evidence Reconsidered The cornerstone of the Zaleski model is the Phillips curve Equation (2). Like Smyth, Washburn and Dua (1989) we set p~ = Pt-1 but estimate 1In the case of c > 0 Equation (2) contains one endogenous variable, Ut, whereas the exogenous variable U~-I is excluded. Thus, Equation (2) is exactly identified. More than one exogenous variable is excluded from Equation (4). Hence, this equation is over identified.
390
Partisan Profiles in Presidential Policies T A B L E 1.
Equation (2), IV-estimation
Period 1961:iii-1993:ii 1961:iii-1973:iv 1974:i-1979:iv 1980:i-1993:ii
fl 0.017 (0.004) ~ 0.093 (0.021) -0.000 (0.008) 0.009 (0.006)
a
U*
-0.194 5.647 (0.036) c (0.316) ° - 0.075 3.843 (0.072) (1.422) --0.731 6.936 (0.126) (0.202) - 0.215 6.008 (0.077) (0.601)
~2
B pa
B Gb
0.294 8.398 37.337 -(0,015) d (0.000) a 0.332 0.747 7.335 -(0.688) (0.119) 0.713 0.674 6.774 -(0.714) (0.148) 0.187 2.299 14.611 -(0.317) (0.006)
NOTE: aBreuseh-Pagan-Test for heteroskedastieity; bBreuseh-Godfrey-Testfor serial correlation up to order 4; eestimated standard error, dp-value (probabilityof a type-I error)
this equation over the whole period 1961:iii-1993:ii by the instrumental variables method. 2 Based on a series of Chow tests 3 we decided to split our sample into three periods. Table 1 reports our estimates of (2). The estimated slope of the Phillips curve, c~, in the first row of Table 1 is well below that obtained by Smyth, Washburn, and D u a (1989). We find no significant inflation/unemployment trade-off in the sixties. Between 1974 and 1979 that trade-off played a major role: the Phillips curve explains more than two-thirds of the variance of inflation. Its importance declined in the eighties. Thus, U.S.-Presidents faced different menus of choice, and we now examine their influence on Zaleski's estimates. Table 2 presents the preference ratios a/b computed from (3) employing the estimates of (x and U* from Table 1. In column 2 we assume, like Zaleski, a target rate of unemployment equal to zero. We allow for non zero values of k in columns 3 to 5. 4 However, we do not consider values of "°Wemeasure p by the annual rate of change of the consumer price index (all items), taken from the OECD database, s by the annual rate of change of the energy price index, taken from the Citmorp database, and U by the rate of unemployment includingthe resident Armed Forces, also taken from the Citieorp database. We use Ut-I, Ut-,2, Ut-3, and Ut_4 as instruments. The standard error of U* is estimated by the Delta method, see, e.g., Greene (1993, 297). SThe Chow test strongly rejects the stability of Equation (2) throughout the full sample period 1961:iii-1993:ii. Assuming a break in 1974:i we find an f-value of 5.138 with a marginal level of significance of 0.002. It also rejects the stability of the estimates of (2) during the period 1974:i-1993:ii Assumingan additional break in 1980:iwe find anf-value of 3.353 (withp-value of 0.024). Therefore, we spht the penod of observation into the three intervals considered in Table 1. 4We computed preference ratios for k ~ [0.0, 0.7] in steps of 0.1. The results are well represented by those displayed in Table 2. 391
Alfred Maussner and Markus Kueppers
TABLE 2.
Average Preference Ratios (a/b) Computed from Equation (3)
Administration
k = 0
k = 0.1
k = 0.5
k = 0.7
1961:iii-1964:i Kennedy 1964:ii-1969:ii Johnson 1969:iii-1974:iv Nixon 1975:i-1977:ii Ford 1977:iii-1981:ii Carter 1981:iii-1989:ii Reagan 1989:iii-1983:ii Bush all Republicans
0.017 (0.001) a 0.058 (0.007) 0.329 (0.117) 0.677 (0.049) 0.823 (0.110) 0.131 (0.011) 0.147 (0.014) 0.257 (0.038) 0.304 (0.065) 0.725 b
0.018 (0.001) 0.065 (0.008) 0.373 (0.134) 0.743 (0.053) 0.927 (0.126) 0.144 (0.012) 0.163 (0.016) 0.288 (0.043) 0.342 (0.073) 0.761
0.026 (0.002) 0.0125 (0.018) 0.855 (0.340) 1.219 (0.078) 1.896 (0.307) 0.242 (0.023) 0.299 (0.037) 0.544 (0,102) 0,693 (0.160) 0.928
0.033 (0.003) 0.239 (0.041) 4.266 (2.302) 1.805 (0.111) 4.275 (0.872) 0.395 (0.050) 0.545 (0.090) 1.666 (0.651) 1.537 (0.401) -0.319 a
(0.472) C
(0.450)
(0.358)
(0.751)
all Democrats Ho: (a/b)u > (a/b)R
NOTE: aestimatedstandarderror, bt-valuefor H0: "Democratsdislikeunemploymentmore than Republicans",°p-value(probabilityof a type-I error), d = H0: (a/b)R> (a/b)o.
k > 0.7, since in that case the target rate of unemployment, kU*, exceeds the actual rate of unemployment, which the model predicts not to happen. The figures in column 2 are consistent estimates of Zaleski's model. Due to the much smaller value of the slope parameter 5, none of the estimates exceeds one. Except for the value of the Carter administration all preference ratios are even significantly smaller than one. Thus, in the interpretation of Zaleski, almost all administrations were biased toward unemployment. This conclusion, however, is not robust. As can be seen from column 5, the average preference ratios of both Republican and Democratic administrations differ not significantly from one in the case of k = 0.7. Nevertheless, Table 2 confirms Zalestd's conclusion: there is no evidence of a marked difference in the revealed preferences of Republican and Democratic presidents. We turn to the estimation of a and b from (2) and (4). We used 3SLS 399.
Partisan Profiles in Presidential Policies TABLE 3.
3SLS-estimation of(2) and (4)
Restriction Administration
k = 0, c ¢ 0
Weight Estimate
1961:iii-1969:ii Kennedy/ Johnson 1969:iii-1977:ii Nixon/Ford
a b a/b a b ado 1977:iii-1981:ii a Carter b a/b 1981:iii-1989:ii a Reagan b a/b 1981:iii-1993:ii a Reagan/ b Bush a/b
0.027 0.256 0.105 0.067 0.229 0.293 0.129 0.063 2.057 0.035 0.332 0.105 0.029 0.346 0.085
k#0, c#0
k = 0, c = 0
a
Estimate
rr
Estimate
0.011 0.079 0.178 0.016 0.079 0.069 0.028 0.079 0.237 0.009 0.079 0.046 0.007 0,079 0.037
0.122 - 0.332 - 0.367 0.071 0.228 0.311 0.065 0.069 0.930 0.006 0.290 -0.019 0.009 0.324 0.027
0.201 1.432 1.005 0.031 0.058 0.152 0.081 0.016 1.233 0.028 0.082 0.098 0.020 0.095 0.061
0.002 0.998 0.002 0.289 0.710 0.408 0.685 0.316 2.169 0.140 0.859 0.163 0.112 0.888 0.126
0.079 0.079 0.080 0.046 0.046 0.092 0.026 0.026 0.259 0.043 0.043 0.058 0.039 0.039 0.050
and derived standard errors employing the Delta method (see Greene 1993, 297). Like Siebrand and Swank we maximized the degrees of freedom by pooling the data where it was possible. As a first exercise, we set k = 0, included two (multiplicative) dummy variables in Equation (4), and estimated both equations over the full sample period 1961:iii-1993:ii. 5 This approach pretends that all administrations faced the same inflation/unemployment trade-off. It reveals a difference between the Democratic and the Republican (a/b) ratio of 0.13, which is significant at the 5% level. Yet, as we show next, this result depends crucially upon the hypothesized constant slope of the Phillips curve. Table 3 presents estimates for the same subsamples as those eonsidered by Siebrand and Swank. The system of equations approach delivers the 5Let d t = 0 ( = 1) in quarters of a Republican (Democratic) administration. The equation estimated together with the Phillips curve, then, reads Ut = 7~lUt_1 +
1~2dtUt_ 1 + n d p t
1-a-b rtl
1 -
b
+ r~4dtp t ,
b , n3
1 -
1 -~-F) b a' nl
+ ~r2 =
1 -
b
b ' zcz + n4 =
-1
-
b Ct "
(The tilde denotes Democratic preference parameters.)
393
Alfred Maussner and Markus Kueppers Phillips curve pertinent to each administration. Except for the figure obtained for the Carter administration, our estimates of a and b in column 3 differ considerably from those of Siebrand and Swank (i994, 556). A glance at the standard errors in column 4 immediately show that the revealed preferences of the Democratic presidents Kennedy and Johnson are essentially the same as those of the Republican presidents Nixon, Ford, Reagan, and Bush. The Carter administration, however, appears to be very different. As predicted by the partisan model, it exhibits a considerable inflationary bias. Column 5 presents estimates of the unrestricted Equations (2) and (4). None of the estimated preference ratios is significant at the 5% level, thus, all presidencies appear to be similar. However, from these estimates, it was not possible to reject the hypothesis of k = 0. Hence, the restricted estimates of column 4 are acceptable. Despite the fact that our estimates strongly reject the restriction c = 0, we estimated (2) and (4) under that constraint in order to produce comparable figures for the Zaleski model, As can be seen from column 7 we find significant differences between all administrations, except the Reagan and the Bush period. In the ease of Kennedy/Johnson and Nixon/Ford, however, these are opposite to the prediction of the partisan model: Nixon and Ford put more weight on the reduction of unemployment than Kennedy and Johnson did. As it appears, the partisan profile found by Siebrand and Swank is the outcome of both the inappropriate method of estimation and the unjustified hypothesis of an unchanged unemployment/inflation trade-off.
5. Conclusion This paper attempts to clarify the discussion between Zaleski (1992) and Siebrand and Swank (1994). Applying consistent estimation techniques to their models and relaxing some of the restrictions imposed by these authors we found no significant partisan profile in U.S. Presidential economic policies to fight inflation and unemployment. Thus, our results confirm Zaleskfs conclusion. Nevertheless, the debate may not be entirely resolved by the rather simple model employed. Indeed, we discovered some signs of misspeeification. The restriction of a unitary coefficient on expected inflation is not justified by the data and the estimated errors of most equations are not entirely flee of autoeorrelation. Therefore, in future studies, it may be worthwhile to expand the model of the Phillips curve introducing variables that explain the wedge between producer and consumer real wage rates and that account for hysteresis effects in the wage setting process. Furthermore, assuming that policy makers discount the future at an infinite rate may be 394
Partisan Profiles in Presidential Policies unwarranted. Resolving the partisan debate may crucially depend on a sound approach to address these issues.
Received:October1995 FinalVersion:January1997
References Greene, William H. Econometric Analysis', 2d ed. New York: Macmillan, 1993. Gordon, Robert. "An Interpretation of the Costs on the Instruments in Deterrninistic Linear-Quadratic Control." International Economic Review 17 (1976): 779-81 Siebrand, Jan C., and Otto H. Swank. "Partisan Profiles in Presidential Policies: An Extension of Presidential Preferences for Inflation Versus Unemployment." Journal of Macroeconomics 16 (1994): 553-58 Smyth, David J., Susan K. Washburn, and Pami Dua. "Social Preferences, Inflation, Unemployment and Political Business Cycles: Econometric Evidence for the Reagan Presidency." Southern Economic Journal 56 (1989): 336-48 Zaleski, Peter A. "Presidential Preferences for Inflation Versus Unemployment." Journal of Macroeconomics 14 (1992): 555-61
Appendix Variable Definitions Ut kU* Pt st a b c
= = = = = = =
unemployment rate. target rate of unemployment. inflation rate. rate of change of energy prices. weight attributed to the deviation of Ut from its target rate. weight attributed to the inflation rate. weight attributed to the adjustment of the rate of unemployment. ot = regression coefficient (slope phillips curve). 13 = regression coefficient. dt = Dummy-Variable, 0(1) for a Republican (Democratic) administration.
395