As noted earlier, Stephen Moore has a dearth of peer-reviewed publications; however, he has coauthored work purporting to be analytical. One of these is these is the Economic Outlook ranking published by the American Legislative Exchange Council (ALEC). In Rich States, Poor States, 2015:
The Economic Outlook Ranking is a forecast based on a state’s current standing in 15 state policy variables. Each of these factors is influenced directly by state lawmakers through the legislative process. Generally speaking, states that spend less—especially on income transfer programs, and states that tax less—particularly on productive activities such as working or investing—experience higher growth rates than states that tax and spend more.
Hence I think it is reasonable to ask whether the ALEC-Arthur Laffer-Stephen Moore-John Williams “economic outlook” ranking (hereafter “ALEC ranking”) actually has any predictive power, above and beyond other demographic and geographic indicators. This analysis follows up on a more ad hoc analysis in this post, and confirms conclusions arrived at in Fisher w/LeRoy and Mattera (2012). In addition, this analysis serves as a rejoinder to ALEC’s rebuttal asserting that critiques did not incorporate sufficient controls and allow for differing time horizons, to wit:
Moreover, rigorous statistical methods show that a higher economic outlook ranking in Rich States, Poor States does indeed correlate with a stronger state economy.
Figure 1 depicts the one year growth in real Gross State Product and the ALEC ranking, lagged one year, for growth over the 2008-2014 period. Should the ALEC thesis be correct, one should see a negative sloped relationship.
Figure 1: Real Gross State Product growth (log first difference) and lagged ALEC ranking for Economic Outlook. Nearest neighbor fit line (red), window = 0.3. Source: BEA and ALEC, Rich States, Poor States, 2015, and author’s calculations.
There is no obvious correlation between annual growth rates and the ALEC rankings. Note that lagging the ranking so that, for instance, the growth rate between 2013 and 2014 is related to the ALEC ranking in 2012 (instead of 2013) does not change the results much. (The ALEC 2013 ranking pertains to data reported in 2012, as far as I can tell). This outcome is unsurprising because the ALEC rankings are highly persistent. If one estimates a panel autoregressive model for the ALEC rankings, the AR coefficient is 0.95, and the adjusted R2 is 0.90.
Defenders of the ALEC rankings have argued that the rankings are aimed at predicting growth at a longer time horizon than annual. Given that the perspective of the RSPS methodology is supply-side, this argument is prima facie sensible. In order to accommodate that argument, I display in Figure 2 the data for three year (nonoverlapping) horizons (for 2014, and 2011).
Figure 2: Nonoverlapping average three year real Gross State Product growth (log difference) and three year lagged ALEC ranking for Economic Outlook. Nearest neighbor fit line (red), window = 0.3. Source: BEA and ALEC, Rich States, Poor States, 2015, and author’s calculations.
Figure 3 provides analogous information, at the six year horizon (the maximum possible given the span of ALEC rankings).
Figure 3: Average six year real Gross State Product growth (log difference) and six year lagged ALEC ranking for Economic Outlook. Nearest neighbor fit line (red), window = 0.3. Source: BEA and ALEC, Rich States, Poor States, 2015, and author’s calculations.
The nonparametric fitted lines indicate a slightly negative slope overall, suggesting some content to the argument that lower ranked states grow slower. In both Figures 2 and 3, North Dakota and Texas constitute outliers along the y-dimension. In order to discern how much these two observations drive the results, I omit them and replot in Figure 4.
Figure 4: Average six year real Gross State Product growth (log difference) and six year lagged ALEC ranking for Economic Outlook, excluding North Dakota and Texas. Nearest neighbor fit line (red), window = 0.3. Source: BEA and ALEC, Rich States, Poor States, 2015, and author’s calculations.
No obvious pattern emerges from this last plot. In order to move beyond simple graphics, I now implement a series of regressions. First to a simple examination of the a bivariate relationship, encompassing the lower 48 states, one obtains at the annual frequency:
Δyi,t = -0.00003ALECi,t-1
Adj-R2 = -0.00, SER = 0.027, N=288. bold indicates significance at 10% msl, using heteroscedasticity and serial correlation corrected standard errors.
(I use the lower 48 as I do not have demographic and geographic data for the Alaska and Hawaii.)
Essentially, there is no explanatory power for the ALEC-Laffer indices in terms of year on year GSP real growth. Allowing for individual state-fixed effects, one obtains:
Δyi,t = 0.0004ALECi,t-1
Adj-R2 = -0.00, SER = 0.027, N=288. bold indicates significance at 11% msl, using heteroscedasticity and serial correlation corrected standard errors.
The coefficient is positive, indicating that lower ranked states, and hence states that have a less business friendly environment according the ALEC criterion, exhibit higher economic growth. The coefficient is borderline statistically significant. I would say that the use of state-level fixed effects is a fairly blunt way to account for state-specific factors. A better approach controls for state factors that economic theory suggests might be important for growth, such as geographic and demographic factors. Here we include log population density (LDENSITY, to account for urbanization), dryness (measured as inverse, WET), mildness of weather (MILD) and proximity to navigable waterways (DISTANCE); these four variables are defined such that positive coefficients are expected. These variables are time-invariant; hence one cannot estimate a fixed effects regression incorporating these variables. I also include log real price of oil (LRPOIL) to control for oil producing states, allowing the coefficient to vary across states.
Δyi,t = 0.0002ALECi,t-1-0.01LDENSITYi + 0.019WETi + 0.001MILDi + 0.0001DISTANCEi + LRPOILi,t
Adj-R2 = 0.36, SER = 0.021, N= 288. bold indicates significance at 10% msl, using heteroscedasticity and serial correlation corrected standard errors.
The ALEC ranking has no statistical significance, while a drier and more mild climate is associated with faster growth, with statistical significance. Proximity to navigable water is also a positive factor.
What if we estimate a comparable regression, looking at 6 year growth rates (but omitting oil prices which are the same for all states)? Then one obtains:
Δyi,t = -0.018 -0.0001ALECi,t-6 +0.001LDENSITYi – 0.0007WETi – 0.0003MILDi -0.00001DISTANCEi
Adj-R2 = -0.02, SER = 0.007, N= 48. bold indicates significance at 10% msl, using heteroscedasticity and serial correlation corrected standard errors.
Notice that the ALEC coefficient is not statistically significant. Thus far, the one case it has shown up as borderline significant, it goes the opposite of the ALEC-Laffer-Moore-Williams thesis.
Finally,Professor Ed (“no recession”) Lazear recently asserted that states with fast growing employment have low tax rates and right-to-work laws. He also argues that one needs to incorporate the depth of the drop in employment in 2008-09, in order to explain the growth in employment (so, a sort of version of the bounceback thesis he forwarded in the Economic Report of the President, 2009). I can’t find the working paper that provides the basis for his assertion, but I can estimate a comparable regression. In order to make the proposition testable, I examine 5 year growth rates in output, and refer to the ALEC rankings in 2009. I add the change in output from 2008 to 2009 (so TROUGHDEPTH takes on a value of -0.05 for observation i if output dropped 5% going into 2009).
Δyi,t = 0.065 -0.0010ALECi,t-5 +0.0046LDENSITYi + 0.0042WETi – 0.0014MILDi +0.00005DISTANCEi – 0.3413TROUGHDEPTHi
Adj-R2 = -0.04, SER = 0.037, N= 48. bold indicates significance at 10% msl, using heteroscedasticity corrected standard errors.
The ALEC variable does not exhibit statistical significance. Omitting ND and TX would not change this basic results. Moreover, the trough variable coefficient, while exhibiting the right sign, is not statistically significant.
Bottom line: The ALEC ranking, which purports to measure business-friendly policies, is not correlated with real GSP growth, either short term or medium term. This is true if one controls for additional variables.
I don’t think this necessarily means that policies such as right to work, or tax rates, and so forth do not have an impact on growth. For instance, Kolko, Neumark, and Cuellar Mejia, J.Reg.Stud. (2013) conclude that what’s important is “less spending on welfare and transfer payments; and more uniform and simpler corporate tax structures.” Hence, the ALEC economic outlook ranking is, in my assessment, a manifestation of faith based economics.
[This is a revision of a post from 2015]
Note: Thanks to Professor Neumark, who kindly provided the data set used in his J.Reg.Stud. paper for use by my students in my PA819 statistics course (all students had to use his data set to analyze whether business conditions as measured by various indices (but not the ALEC ranking) influenced state level growth). The demographic and regional variables are drawn from that data set.