On reading my recent post on Kansas economic performance in the reign of Brownback, which included this graph:
Figure 1: Private nonfarm payroll employment in Kansas (red), in US (blue), in logs normalized to 2011M01=0. Dashed line at 2011M01, Brownback term begins. Source: BLS, author’s calculations.
A&M Professor/Extension Economist Levi Russell writes “Your analysis is highly flawed”.
He then refers me to a post a year ago, critiquing both Paul Krugman’s post “This Age of Derp, Kansas Edition”, as well as my post.
Below the childish title of Paul Krugman’s latest post is a simplistic analysis of the state of the Kansas economy. Krugman references another post by Menzie Chinn that is slightly more sophisticated. I’ll focus on Krugman’s post but will refer to Chinn’s post as well.
Since Chinn and Krugman both mention California, let’s have a look at this “boom” in California. For whatever reason (he doesn’t say why he chooses these states), Chinn plots a business indicator from the Philly Fed for California, the US as a whole, Minnesota, Wisconsin, and Kansas. The graph clearly shows faster growth for CA, MN, and the US as a whole than for WI and KS. This is supposed to be a result of fiscal policy and evidence that KS and WI governments (with Scott Walker as its governor) are wrong for implementing right-wing policies.
That doesn’t imply, as Krugman states, that we should abandon our skepticism of the effectiveness of gov’t intervention. As I’ve shown here, a more nuanced analysis doesn’t tell the same neat little story about the Wheat State as those told by Krugman and Chinn.
Let me preface my comments by noting it is an honor to be lumped in with Paul Krugman.
One might excuse Professor Russell for not knowing how bad things would get over the year following his post, but in fact he has doubled down on his view that fiscal policy does not explain the evolution of the Kansas economy. In February, he wrote:
… the constant drumbeat from the [Kansas Center for Economic Growth] that the Kansas economy is weak due to the Brownback tax cuts is rather silly. Kansas has had strong private employment growth when compared with neighboring states since the tax cuts have gone into effect. As the recent “Rich States, Poor States” report indicates, two of Kansas’ biggest industries, oil and aircraft, have been under significant stress recently. This is, of course, not caused by the recent tax policy change in Kansas. Agriculture can certainly be added to that list.
Time to look at the data.
First, one cannot appeal to weather, cattle prices, wheat prices, oil prices, or even the aircraft industry’s woes, to explain the Kansas disaster. Consider the following out of sample forecasting exercise. I first estimate an error correction model with Kansas and US GDP over the 2005Q2-2010Q4 period:
(1) ΔyKSt = -8.49 – 0.38yKSt-1 + 0.78 yUSt-1 + 1.46ΔyUSt + 0.002droughtt + ut
Adj-R2 = 0.60, SER = 0.0085, N = 23, DW = 1.77, Breusch-Godfrey Serial Correlation LM Test = 2.07 [p-value = 0.16]. Bold face denotes statistical significance at 10% msl, using HAC robust standard errors. y denotes log real GDP, and drought is the Palmer Drought Severity Index for Kansas (PDSI, lower is more severe).
The data used in this analysis (and for calculations in Figures 3 and 4) are here. Log Kansas and US GDP appear I(1) (fail to reject Elliott-Rothenberg-Stock unit root test) and Kansas PDSI (borderline) rejects a unit root.
Using this ECM to dynamically forecast out of sample in an ex post historical simulation (i.e., using realized values of US and Kansas GDP, and the drought variable), I find that (1) actual Kansas GDP is far below predicted (5.3 billion Ch.2009$ SAAR, or 3.9%, as of 2015Q3), and (2) the difference is statistically significant. This is shown in Figure 2.
Figure 2: Kansas GDP, in millions Ch.2009$ SAAR (blue), ex post historical simulation (red), 90% band (gray lines). Forecast uses equation (1). NBER defined recession dates shaded gray. Source: BEA, NBER, and author’s calculations.
Note that historical simulation incorporates the effect of drought, despite the fact that the coefficient enters with significance only at the 21% level. In other words, based on historical correlations of Kansas GDP with national, and weather, Kansas should have performed measurably better.
Now, the last refuge the defenders of the Kansas experiment is to say it’s the agricultural or oil sectors, and/or it’s the aircraft industry’s woes, or maybe the plotting of the Illuminati. In Professor Russell’s critique of a Kansas Center for Economic Growth (KCEG) analysis, he uses all three (but omitting the role of the Illuminati).
As the recent “Rich States, Poor States” report indicates, two of Kansas’ biggest industries, oil and aircraft, have been under significant stress recently. This is, of course, not caused by the recent tax policy change in Kansas. Agriculture can certainly be added to that list.
My view: Whenever someone quotes Rich States, Poor States as if it were a source of reliable analysis, watch out!! But let’s take at face value this assertion – it’s anything but fiscal policy.
Let me systematically appraise each of these alternative explanations.
First, re-estimate equation (1) over the entire 2005Q1-2015Q3 period. The drought variable (the Kansas PDSI) does not enter with statistical significance at even the 50% msl. So maybe drought is it, but it doesn’t show up statistically.
Second, substitute in for the PDSI either the real wheat price (CPI deflated), real oil price (core CPI deflated) or the real cattle price (CPI deflated) for the drought variable, and once again the respective coefficients are not significant at the 50% msl, or for cattle is significant at the 50% msl, but with the wrong sign. (Note: log real wheat and cattle prices appear I(1), so I enter in first differences, while PDSI and log real oil prices appear I(0), so I enter in levels).
Hence, one is quickly exhausting the set of possible excuses for the Kansas dropoff.
What about aircraft production? I’ve addressed this in a previous post (Professor Russell shares certain views with Ironman in this regard), but why not repeat? Figure 3 shows a decomposition of growth on a quarter-on-quarter basis (not annualized).
Figure 3: Contributions to real GDP growth, from agriculture (red), durable manufacturing (green), and rest-of-economy (blue). Source: BEA and author’s calculations.
At certain junctures, durable manufacturing and agriculture do subtract from growth; at other times add. But it has been quite some time since these factors have exerted a drag on output growth. Consequently, recent lagging performance cannot be attributed to the factors that Professor Levi highlights.
Finally, what about fiscal policy? Figure 4 shows the correlation between two variables: the gap between actual and predicted Kansas GDP, and the gap between the actual government spending on goods and services and trend (from 2005-2010). Well, as expected, the bigger the shortfall in government spending, the bigger the shortfall in growth.
Figure 4: Gap between actual and predicted Kansas GDP (blue), and gap between actual and trend Kansas government spending on goods and services (red), both in millions in Ch.2009$ SAAR. Predicted GDP is from ex post historical simulation using equation (1). Trend government spending is exponential trend over 2005-2010 period. Source: BEA and author’s calculations.
The positive correlation is statistically significant. A regression yields the following result:
(2) Δy_gapt = 5.76 + 2.39Δg_gapt-1 + ut
Adj-R2 = 0.09, SER = 1193.7, N = 18, DW = 0.94. Bold face denotes statistical significance at 10% msl, using HAC robust standard errors. y_gap denotes gap between actual and predicted Kansas GDP, g_gap denotes gap between actual and 2005-2010 exponential trend in Kansas government spending on goods and services.
The slope coefficient is statistically significant at the 6% msl. The confidence interval would encompass the conventional regional multiplier of approximately 1.5 . What’s omitted is tax revenues –- revenues are now down about 2% relative to 2008Q1 peak . That would have exerted an expansionary effect, but since I’ve also omitted transfers, it’s hard to say what the bias would be on the point estimate.
I conclude with my standard graph of GDP growth of the ALEC darlings and betes noire, Kansas and Wisconsin, and Minnesota and California, respectively, because Professor Russell could not discern the basis for my choice of states.
Figure 5: Log Gross State Product for Minnesota (blue), Wisconsin (bold red), Kansas (green), California (teal) and US (black), all normalized to 2011Q1, seasonally adjusted at annual rates, in Chained 2009$. Source: BEA, and author’s calculations.
Notice that pre-recession Kansas was growing faster than the other states and the nation; in the era of Brownback, well, the graph says it all, if you didn’t believe the econometrics.
Digression on fixed effects: At least back in June 2015, Professor Russell made a point of highlighting the difference in extent of slack in the Californian economy vs. the Kansas, arguing that when there is lots of slack, employment growth will tend to be fast. Hence, in his view, the relatively slow Kansas employment growth in Figure 1 is due to the smaller amount of slack, as measured by unemployment. However, apparently, Professor Russell has forgotten about fixed effects. Over the 1976-2007 period, California unemployment is typically about 0.77 percentage points higher than the national average; so the March 2016 California rate of 5.4% is only 0.4 percentage points higher than the national average. By that criterion, California’s labor market is tight. The Kansas situation is the opposite. On average, Kansas unemployment is 1.67 percentage points lower than the national average, and yet, as of March, it is only 1.1 percentage points lower. Hence, there would seem to be more ecnomic slack in Kansas than in California, by this criterion. Given this, the argument that California employment growth is faster because of more slack seems dubious to me. (Additional graphs of CA vs. KS in employment and coincident indices here).