Models Used for Policy Should Reflect Recent Experience, But Do They?

Data from the Department of Commerce show that short-term stimulus funds did not go to increase federal purchases, or state and local purchases, or even consumption purchases by much over the past few years. Thus the packages did not materially stimulate GDP or employment. In a recent Commentary piece “Where Did the Stimulus Go?" John Cogan and I review and explain our empirical research using these data.

Unfortunately, most Keynesian models have been not adjusted to incorporate these facts, so they keep making the same predictions. To cite one example, the multipliers in Mark Zandi’s model as of July 2008 are found here (page 52 of the full document, 5 of the paper) while the multipliers as of December 2010 are found here. They are virtually the same. The model assumes a multiplier from a temporary tax rebate or refund which is greater than one, even though the actual data show it was much less than one in the case of the 2008 and 2009 stimulus packages. Note that the model also assumes a multiplier from a permanent tax cut which is only about 1/3. The relative sizes of permanent and temporary effects are exactly the opposite of what basic economics implies.

Using such results people write stories like this one “Zandi Analyses Show ‘Democratic’ Measures in Tax Cut-UI Deal Boost Economy, ‘Republican’ Measures Add to Deficit Risks” from the Center of Budget and Policy Priorities, supposedly because Democrats favor temporary actions and Republicans favor permanent actions. But the “analyses” are simply old simulations from models which appear to be ignoring the facts. More consistent with the facts and the theory is that the recent tax deal will be more beneficial to the economy than the past stimulus packages because it extends the Bush tax cuts and thus makes them more likely to be permanent.

Another example of the problem with the modeling assumptions is the multiplier from “general aid to state governments,” which is assumed to be 1.36 in the Zandi model. Yet the Commerce Department data are very clear that virtually none of this aid to state governments in the 2009 stimulus (ARRA) went into government purchases; most went to reduce borrowing. Here is a diagram from the Commentary article which shows this. You cannot get a multiplier of 1.36, or even much greater than zero, when none of the funds went to government purchases and more than half went to reduced borrowing.