I ran a regression of ΔY on ΔX and ΔZ, over the 1967Q1-2011Q3 period. I found that the coefficient on ΔX was 0.007, and on ΔZ was -0.080. Neither coefficient was statistically significant at conventional levels, so I concluded that neither affected ΔY.
It turns out ΔY is the GDP deflator inflation rate, ΔX is the growth rate of M1, and ΔZ is the growth rate of real GDP. In other words, I concluded money had no significant impact on inflation.
Well, you might say, that was a silly regression to run. It happens to be exactly analogous to the regressions run by the Phoenix centre for Advanced Legal and Economic Public Policy Studies, in its assessment of the effectiveness of government spending. As the Mises blog published today: “a recent study published by the Phoenix centre looked at the empirical evidence for the US over the last 50 years and found that government spending/stimulus had zero positive impact on private sector job creation.”
To see how the Phoenix centre came to its conclusion, now let ΔY be employment growth, ΔX be investment growth, and ΔZ be government consumption and investment growth, and let the run the regression over the 1960-2011 period et voilà! Proof positive that multipliers for government spending are zero while those for investment very large. Some additional observations:
- The authors assert these multipliers are different over high growth and low growth periods (i.e., there are two regimes governing these relationships). Of course footnote 30 indicates that one can’t reject the null that the government spending coefficients are the same.
- The regressand selection of private employment is not quite standard. The usual interpretation of the multiplier is to relate total employment growth to growth in government consumption and investment, not just private employment.
- For the life of me, I could not replicate their basic linear model results, even using private employment. If some one can, please tell me.
This is what I get running a regression of log difference of private payroll employment (average of monthly data) on log first difference of real investment and real government consumption and investment (growth rates annualized).
Δn = 0.013 + 0.088 × Δinv + 0.045 × Δgov
Adj.R2 = 0.38, DW = 0.76. Bold Face denotes significance at 10% msl.
Man, oh, man, and I thought the St. Louis money equations were bad; not only does this regression have an extremely restrictive lag structure, it is so restrictive that there are only contemporaneous effects (i.e., no lags — perhaps problematic, given the excruciatingly low DW statistic, but this did not seem to trouble the authors, so I won’t dwell on it besides noting the inappropriate use of conventional standard errors to make statistical inference). Anyway, for those of you wondering why I wasted brain cells on this, apparently some people take these results (and the Phoenix centre) seriously, including these periodicals: , reprinted in , , and .