For this blog post, let’s assume Keynesian economics.
For all the talk of a “large stimulus,” you don’t hear much about a “longer stimulus.”
The problem with a “too small” stimulus is that you get an initial economic boost, but when the stimulus expires the economy slumps back down, as indeed happened in mid 2011. Ideally a stimulus employs some idle labor, stops it from depreciating, and tides those workers over until they can look for other jobs in fundamentally better economic conditions. Those last few words are important. If conditions are not improving soon, the ability of the stimulus to “buy time” for those workers isn’t worth much. The workers get laid off from the government projects and their reemployment prospects are no better than to begin with. We end up having spent a lot of money to postpone our adjustment problems, rather than achieving takeoff.
Deleveraging recessions last a long time, as shown by Rogoff and Reinhart. The need for continuing deleveraging implies that even a stimulus twice the size of ARRA won’t turn the tide.
In those cases a well-designed stimulus program should not be so “timely.” For a given presented expected value sum spent on stimulus, it is better to spread it out across the years. It is better to help a smaller set of workers for five years (or however many years it takes for most of the deleveraging to end), after which they are reemployable , than to temporarily boost a larger number of workers for two years, and then leave them back in the dust because deleveraging is still going on.
The effectiveness of a stimulus will be measured by how many workers it bridges over until most of the deleveraging is over. For ARRA, that number is close to zero.
Length may be one reason why WWII was effective stimulus (again, we are operating within the Keynesian worldview here, no need to argue this point in the comments). The war lasted a while, and in the meantime a lot of balance sheet repair went on.
Oddly, there is not much discussion about the length of fiscal stimulus. But there should be.