Budget Rules and State Business Cycles: A Comment. Arik Levinson Georgetown University. September 4, 2006

Transcription

1 Budget Rules and State Business Cycles: A Comment Arik Levinson Georgetown University September 4, 2006 Economics Department Georgetown University 3700 O Street, NW Washington DC (202) aml6@georgetown.edu

2 Budget Rules and State Business Cycles: A Comment My 1998 paper provides evidence suggesting that states with strict balanced budget rules have more volatile business cycles. In that paper, I subjected the findings to numerous robustness checks. I used two measures of business cycle volatility, two measures of balanced budget rule stringency, and two measures of state size. I tried dropping unusual states, using only the fifteen largest and fifteen smallest states, and examining different periods of time separately. 1 In most (but not all) specifications, I found that large states with strict balanced budget rules had more volatile business cycles. In their clear and careful replication of that paper, Krol and Svorny (2006) present even more robustness checks. They add two new measures of cyclical volatility, an alternative measure of state size, and an alternative classification of balanced budget rule stringency. In some (but not all) specifications, they also find that large states with strict balanced budget rules had more volatile business cycles. First, Krol and Svorny use as the measure of volatility the standard deviation of annual first differences in incomes, instead of the standard deviation of the deviations (residuals) from predicted long-run income growth, which I used. Here they find the key coefficient statistically insignificant (column (2) of their table 2). This is not unexpected. Variation in annual fluctuations (first differences) will be smaller than variation in deviations from long-run growth (residuals). If strict budget rules exacerbate recessions by prolonging them past one year, the effect will not be seen in annual first differences. My 1998 paper found that 1 Levinson (1998), tables 6 and 7, pp

3 regressions based on differences yielded less statistically significant results. 2 Krol and Svorny's first check does not contradict those earlier findings. Second, Krol and Svorny detrend the state income data using the Hodrick-Prescott (HP) filter (column (3) of their table 2). Here their results also match mine: large states with strict balanced budget rules have more volatile business cycles. Third, Krol and Svorny substitute a continuous measure of state size (population) for my binary measure (large versus small). Again our results are consistent: the residuals and HP filter both suggest that large states with strict balanced budget rules have more volatile cycles (columns (7) and (9) of their table 3), while the difference measure is not statistically significant (column (8)). Finally, there is one specification change that Krol and Svorny make for which their results differ substantially from mine, where they use different data to describe whether states have strict balanced budget rules. Apparently (and this is an important contribution of theirs), classifying balanced budget stringency is not straightforward. Most of the papers in this literature use a simple distinction: if states are allowed to carry unexpected deficits into the subsequent year, then the balanced budget rule is classified as lenient; if states must make up budget deficits in the subsequent year, the rule is strict. In their table 1, Krol and Svorny present classifications of states according to this criteria performed by three different organizations. Surprisingly, the three classifications differ. One organization, the National Association of State Budget Officers (NASBO), surveyed states four separate times and often received conflicting answers, which Krol and Svorny believe to be an error. 2 I used the maximum peak-trough difference, which will necessarily be larger than the standard deviation of annual first differences. See Levinson (1998) tables 6 and 7, pp In 12 out of 13 specifications the t- statistics for the maximum difference in columns (1b) and (2b) are smaller than for the standard deviations of residuals in columns (1a) and (2a). 2

4 Krol and Svorny substitute the General Accounting Office (GAO) classification of balanced budget strictness for the Advisory Commission on Interstate Relations (ACIR) classification in the ten states where they differ (excluding Nevada which the GAO omits). The results reverse my 1998 findings using the ACIR classification. Large states with lenient balanced budget rules do not appear to have less volatile cycles (their table 2) and in some cases appear to have more volatile cycles (their table 3). Why believe the GAO classifications rather than the ACIR? Krol and Svorny cite a GAO describing "structured interviews with state budget officials," and argue that "the GAO's reputation and methods appear to favor the GAO results over those produced by NASBO and the ACIR, which relied on self-reporting by state budget officials." Perhaps. But in some cases the GAO classifies a state as "lenient" even though the state increased revenue, reduced spending, and took other "actions midyear to close budget gaps." 3 Krol and Svorny use the GAO classification even when it differs from that of the ACIR and all four of the NASBO reports. (See Delaware and Utah.) Suppose that instead of simply replacing the ACIR classification with the GAO, we let the NASBO break ties. Any time two or more of the four NASBO surveys support the ACIR, use the ACIR classification, otherwise use the GAO's. This procedure adds two states to Krol and Svorny's list of lenient states and subtracts five. 4 The results, reported in table 1 of this note, support my original 1998 findings. Large states with lenient balanced budget rules have less volatile business cycles. 5 3 See GAO table III.1 p.32. Arizona and Iowa, for example, are classified as lenient by GAO, even though they took all three actions midyear. The ACIR classifies both states as strict. 4 Alaska and Oregon switch from strict to lenient, while Delaware, Georgia, Iowa, Nebraska and Utah switch from lenient to strict. 5 Even if we use the ACIR classification in lieu of GAO only when three or more (rather than two or more) of the NASBO classifications support the ACIR, the key lenient*large coefficient remains about the same size ( ) and is only slightly less statistically significant (std. err. = ). 3

5 Out of four major specification changes in Krol and Svorny (first differences, the HP filter, a continuous population measure, and the GAO stringency classification), only the last yields results that are substantively different from my 1998 findings. And even that last point deserves a caveat, since when the NASBO classification is used to settle disparities between the GAO and ACIR, my 1998 findings remain intact. The Krol and Svorny paper teaches two valuable lessons. First, the key determinant of balanced budget rule stringency, whether deficits can be carried over into subsequent years, appears open to interpretation for many states. Second, studies like these depend on fifty-state cross-section regressions, and replacing just a few observations can change results. While states are often thought of as test venues for national policies, the sample size is small and inferences can be tenuous. Krol and Svorny have injected an appropriate note of caution into our ability to draw conclusions from the experience of the states. Nevertheless, the preponderance of the evidence here suggests that strict balanced budget rules do exacerbate business cycle fluctuations. References Krol, Robert and Shirley Svorny "Budget rules and state business cycles." Public Finance Review, forthcoming. Levinson, Arik "Balanced budgets and business cycles: Evidence from the States." National Tax Journal 51(4): U.S. General Accounting Office Balanced budget requirements: State experiences and implications for the federal government. Briefing Report to the Chairman, Committee on the Budget, House of Representatives, GAO/AFMD-93-58BR. 4

6 Table 1. The Effect of State Balanced Budget Rules on Business Cycle Volatility: Using the NASBO Classification to Resolve Discrepancies Between ACIR and GAO. Dependent variable: std. dev. of difference between log of personal income per capita, and that predicted by the mean log growth rate Covariates Coefficients Constant 0.145* (0.042) Pop Dummy (Large=1) (0.0051) Lenient (NASBO breaks * ACIR/GAO ties) (0.0059) Lenient * Pop Dummy * (0.0071) Income (0.0016) Avg. Growth Rate 5.94 (3.40) % Agriculture (0.069) % Mfg * (0.037) % Mining (0.131) % Service 0.432* (0.112) Northeast 0.020* (0.007) Midwest (0.006) West 0.019* (0.006) R 2 N * statistically significant at 5 percent. statistically significant at 10 percent Notes: Nevada is eliminated because it is missing from the GAO classification. Including Nevada using its ACIR classification changes the "lenient*pop dummy" coefficient to (std. err ). States classified as "lenient" using NASBO to resolve discrepancies between ACIR and GAO: AK AZ CA CT IL LA MA MD MI NH NY OR PA TX VA VT WA WI WY. 5