Davies et al. recently published an interesting paper.
Here are some key passages :
Theoretical models typically rely on the logic of Occam’s Razor and employ the fewest possible variables to explain outcomes.
… none assesses directly the information problem people in government face when trying to determine what is really going on. Nor do the models typically adjust for the lags between the recognition of a serious economic problem by, say, White House officials, and actions taken by Congress and then approved by the president.
Even more perplexing, the elegant models used to explain and predict stimulus effects do not consistently account for decisions by monetary authorities that may either support or confound stimulus policy actions developed by the executive and legislative branches of government. And finally, as good as the information obtained may be and as well coordinated as political decision making can be, most models cannot adjust for fiscal actions that may be taken across the 50 states.
Economists’ ability to measure unambiguously the effect on GDP growth or employment growth of increases in government spending requires a precise identification of how the nation’s economic engine is operating in the absence of the increased spending. At the same time, GDP growth may itself induce government spending, and this effect has to be accounted for somehow. In addition, all forces that might affect GDP growth must be held constant while assessing the effect of government spending.
But the economic engine is operated by millions of unrelated decision makers whose expectations regarding government spending may affect how they will react to increases or decreases, especially when those expenditure changes are well advertised in advance.
…
However, if Keynesian theory held, we should also expect to see a positive relationship between government spending now and economic growth in the future.
Reality does not reflect this theory. Figure 12 shows the relationship between changes in federal spending and real per capita economic growth one year in the future. The relationship appears to be negative, though it is statistically indistinguishable from a flat trend line.
It is possible that it takes more than a year for government spending to affect the economy. Comparing changes in federal spending to economic growth two years later reveals a slightly positive relationship, but again the relationship is statistically indistinguishable from a flat trend line (figure 13). Extending the time horizon out as far as 10 years reveals relationships between government spending and economic growth that are sometimes slightly positive (for 2-, 3-, and 9-year horizons) and sometimes slightly negative (for 1-, 4-, 5-, 6-, 7-, 8-, and 10-year horizons), but always statistically zero.
A counterargument is that what really matters is the relationship between stimulus spending and economic growth during recessions.
If we restrict our vision to recessions only (the red dots in figures 11–13), the same story emerges. There is no significant relationship between changes in government spending during recessions and economic growth at any point from one to 10 years later.
One could argue that because of a persistent baseline growth in per capita GDP, changes in federal spending should be compared to changes in per capita GDP growth. That comparison yields the same absence of results as do the previous comparisons. Figures 14 and 15 show the contemporaneous and one-year lagged relationships.
Figure 16 confirms the observation in figure 15 that stimulus spending is destabilizing because it continues to accelerate after the recession has begun to subside. It also confirms the suspicion among some economists that big government politicians do not believe in Keynes but judiciously quote him to justify expanding the government’s control over the economy.
And the nail in the coffin. John Maynard Keynes, toward the end of his life, wrote :
Organized public works, at home and abroad, may be the right cure for a chronic tendency to a deficiency of effective demand. But they are not capable of sufficiently rapid organisation (and above all cannot be reversed or undone at a later date), to be the most serviceable instrument for the prevention of the trade cycle.
And this one (september 2011) is of interest :
Did stimulus-funded projects hire the unemployed or the already employed? Our surveys indicate a near-tie on this question. Of the 277 respondents hired after January 31, 2009, 42.1 percent had been unemployed immediately beforehand and 47.3 percent had come directly from another job. Of the rest, 4.1 percent had been out of the labor force, and 6.5 percent had been in school. Thus, the weight of the evidence suggests that ARRA did an enormous amount of “job shifting” rather than “job creating.” There is evidence of the latter, but, under Keynesian reasoning, every worker hired away from another job reflects some weakening of the stimulus.