My entry at the Romer Roundtable:
WHEN deciding between two alternatives such as whether the government should intervene in the economy with a fiscal stimulus or not, the choice of the null and alternative hypotheses influences the type of errors we are likely to make.
A standard example to illustrate this is the problem of deciding guilt and innocence. If we make guilt the null hypothesis, and only reject the null when there is overwhelming evidence to the contrary, then we are going to find people guilty unless there is enough evidence presented to convince jurors the person did not act as charged. The biggest risk here is that innocent people will be sent to jail since innocence must be established through overwhelming evidence, but guilt is presumed. It’s possible that a guilty person will be found innocent, but the more evidence that we require to find someone innocent—e.g. a strict standard like beyond a reasonable doubt—the smaller the chance that the guilty will be set free.
If we make innocence the null hypothesis, then things are reversed. In this case we will only send someone to jail if the evidence overwhelmingly points to guilt, so the biggest risk under this null is that guilty people will be found innocent. Again, it’s possible that an innocent person will be found guilty, but the system requires a high burden of proof so as to minimise the possibility of this happening.
In the US, we think sending innocent people to jail is a bigger mistake than setting guilty people free, so we make innocence the null hypothesis in court cases rather than guilt, and we require a high level of “proof” before rejecting the null. (When money rather than freedom is at stake as in civil cases, the standard for conviction is often lower. This causes more innocent people to have to pay fines, but fewer guilty people escape them).
What does this have to do with deficit spending and recessions? Our tendency is to assume that the economy is doing fine and doesn’t need help from fiscal authorities unless there is clear evidence the economy is crashing. Only then does the government intervene with deficit spending.
Thus, our null hypothesis is that the economy does not need any help, and we require a fairly high burden of proof to overcome that presumption. Because of this, the error we are most likely to make is to do nothing when action is called for, and that includes ending help too soon, particularly given the information lags we face in assessing the state of the economy. It’s possible that we could get fooled by the data into acting when it isn’t necessary, but since we require clear signals that the economy is in trouble before we act, and because data are slow to arrive, acting when it isn’t needed is less likely than doing nothing when, in fact, active intervention is called for.
Unlike in court cases, however, where a null of innocence allows us to minimise the costly error of jailing the innocent, the null that the economy is “innocent”, i.e. that intervention from authorities is not required, leads us to minimise the wrong outcome.
Which is the bigger error, to deficit spend when it’s not needed, or to fail to do so when it is? I think the bigger risk is doing nothing when it’s needed, particularly when the economy has the type of difficulties we are seeing now. The risk of doing nothing is a severe depression, while the risk of overreacting is inflation or, perhaps, slightly slower growth for a period of time in the future. I don’t see those risks as balanced at all, allowing a depression is—to me—the more severe error, the equivalent of sending the innocent to jail.
We saw the problem with having the wrong null hypothesis when the economy was slipping into the recession. The stimulus package should have been in place long before it was actually implemented in order to be maximally effective, but policymakers were reluctant to act until there could be no doubt that action was called for.
And we are seeing this again now. We may very well need another stimulus package, and we ought to be doing the work to get it ready, but there just isn’t enough evidence to convince policymakers that’s the case. They will have to see clear evidence of continuing troubles and also believe there’s no chance that recovery is just around the corner before they will act, and that’s a high standard to meet. And worse, as we’ve seen recently, as soon as the evidence that we are still in a recession becomes a bit foggy—at the first sign of green shoots—many people will be ready to end the intervention even though that may not be the right course of action.
I can understand the tendency to resist intervention at the first inkling of troubles. So on the front side of a recession, I can understand a null hypothesis that no action is needed, but it ought to be one that can be overturned with a fairly low burden of proof. We shouldn’t have a “beyond a reasonable doubt” standard, it’s not criminal to mistakenly intervene, it should be more like the “clear and convincing” standard sometimes used in civil procedures, or, even better, the weaker “preponderance of the evidence” standard used in civil cases.
But once we are in a recession, as now, I cannot understand why the null does not change to being one where the economy is presumed to need help until there is—beyond any reasonable doubt—evidence that the economy is on the path to recovery. Look at the 1937 experience on the graph in Mrs Romer’s article again and see how costly it is to pull back too soon, then compare that to what the cost would have been to continue the policies for a year, or several years, even though they weren’t needed. It seems pretty clear to me that the cost of pulling back too soon was the much bigger worry. We are facing the same choice now, or will soon. Do we pull back at the very first signs of green shoots, as we seem to want to do, or do we wait until we are much, much more certain that things have, in fact, improved to the point where recovery is all but certain before withdrawing stimulus measures?
As Brad DeLong notes, if it is the long-run budget you are worried about, ending the stimulus package, say, six months or a year earlier makes little difference to the long-term budget outlook. That being the case, and given the dangers of not doing enough and the dangers of ending the help too soon, why are we in such a hurry to end the stimulus package, and we are we so reluctant to consider doing more?