The previous post on state employment trends sparked some debate regarding the generality of the negative correlation between the ALEC-Laffer “Economic Outlook” ranking and economic growth, as measured by the Philadelphia Fed’s coincident index. One reader argued four observations were not sufficient to make a conclusion, and I concur. Here, without further ado, is the correlation for all fifty states.
Figure 1: Ranking by annualized growth rate in log coincident index 2013M01-2014M03 versus 2013 ALEC-Laffer “Economic Outlook” ranking. Nearest neighbor nonparametric smoother line in red (window = 0.7). Source: Philadelphia Fed, ALEC, and author’s calculations.
If a higher ALEC-Laffer ranking resulted in faster growth, then the points should line up along an upward sloping 45 degree line. This is not what I see.
Some of the fastest growing states are oil exporters, so in order to partly control for this factor, I omit the top five oil producer states.
Figure 2: Ranking by annualized growth rate in log coincident index 2013M01-2014M03 versus 2013 ALEC-Laffer “Economic Outlook” ranking, excluding top five oil producing states. Nearest neighbor nonparametric smoother line in red (window = 0.7). Source: Philadelphia Fed, ALEC, and author’s calculations.
An econometric point. Some commentators have repeatedly concentrated on unemployment rates as an indicator of inter-state relative performance, over time. As I’ve pointed out, there are state fixed effects apparent in unemployment data, so it makes sense to look at differences, or growth rates. Hence, I am looking at growth rates over time (in the previous post, I examined cumulated growth rates from 2011M01 onward). (For individual fixed effects, consult a standard econometrics textbook — I am using Stock and Watson Econometrics for teaching this semester; people who keep on ignoring fixed effects should consult).
Notice that whatever relationship there is, it doesn’t seem particularly positive in either sample. A linear regression delivers a negative (but statistically insignificant) coefficient. Of course, since the dependent variable is a ranking, an ordinal regression is more appropriate. In this case the coefficients are not interpretable as slope coefficients, but rather changes in the z statistic, distributed by assumption normally.
In the full sample, the ordered probit regression with 50 ranks yields a coefficient of -0.001, z-statistic of 0.01. Since the ranking is likely to include many cases where the gap in growth is very small, I place the growth ranking and ALEC index rankings into 10 bins, and re-estimate the ordered probit.
The full sample coefficient on ALEC2013 is -0.028, z-statistic of -0.56, so that the p-value is 0.58. Examining the same data, excluding oil producers yields a coefficient of -0.046, z-statistic of 0.85. The p-value for rejecting the null hypothesis of zero coefficient is 0.40.
While the proportion of correct predictions is quite low (the pseudo-R2 is 0.003), the coefficient on ALEC2013 is always negative regardless of specification. The interpretation of the impact of a higher ALEC-Laffer ranking on growth rank is ambiguous in general (and has to be calculated out numerically). However, for the top decile (using the “binned” data), it indicates a higher ALEC-Laffer ranking reduces the probability of being in the top decile. For the bottom decile, a higher score implies increases the probability moving into the lower decile.
Bottom line: If there is any evidence, it suggests that a higher ALEC-Laffer Economic Outlook score is associated with a worse economic performance, as measured by 2013M01-2014M03 growth using the Philadelphia Fed’s coincident indices. However, a more definitive conclusion must await a more comprehensive analysis. (Here‘s a start.)
Addendum: For the data reported by BLS, over the 1976M01-2014M03, the Wisconsin fixed effect (relative to the US), is 0.9 ppts. That is, on average, Wisconsin’s unemployment rate is 0.9 percentage points below that of the United States. The t-statistic (using Newey-West standard errors) for the null hypothesis of 0 difference is 9.5. If you do not understand this paragraph, you should not be comparing inter-state unemployment at a point in time, without referring to a prior period.
Update, 11:10AM Pacific 4/24: Here is a link to Kolko, Neumark and Cuellar Mejia, J.Reg.Stud. (2013), which assesses the relationship between growth and various business environment indices (but not the ALEC-Laffer indices).
Update, 9AM Pacific 5/8: CBPP finds little impact of tax rates on migration, here.