For short-term economic policy, judgement matters more than data

How did economies grow before the economists came along and started offering advice? I have asked this question to many people who have helped run economic policy. No one has had a satisfactory answer.

So I tried refining the question: how did economies grow before the econometricians and the data boys came along? Again there was no satisfactory answer.

So I asked the logical follow-up: do these fellows make any difference to growth? This time there was only an embarrassed silence. There lies the problem. Yes, data matters, but it’s not the only thing that matters for economic policy.

So any government that allows itself either to be stampeded by the data herds — or beguiled by them about its own beauty — is asking for trouble.

The problem I am pointing to is a well-known fallacy in economic theory. It’s called ‘post hoc, ergo propter hoc’,  which is Latin for ‘after this, therefore because of this.’ It used to be one of the first things to be taught to students to warn them, so that they wouldn’t arrive at hasty conclusions about causality. But this warning can be inconvenient to both politicians and data obsessed economists.

There is another major problem with those obsessed with data,  which I call the AI 855 problem. AI 855 crashed into the sea on January 1, 1978, 101 seconds after takeoff from Mumbai. 225 people died.

According to the investigation that was done, it went down “due to the irrational control inputs by the captain following complete unawareness of the altitude as his altitude indicator had malfunctioned. The crew failed to gain control based on other flight instruments."

In simple language, three different instruments measuring the same thing, (namely, the direction of the turn) were showing three different things. The captain could not see the horizon as it was dark. He trusted the data from the instruments and went down.

There is another related problem with data obsession. Just as it happened with AI 855, it’s hard to untangle causality in real time. We don’t know which of the several variables that affect an outcome, is the dominant one. Econometricians have worked on this problem for a long time but with very limited success.

In any case economy-wide data comes after such a long time that it’s pretty useless for policy. Even if it came quite quickly, it would be next to impossible to identify the dominant factor. You can see this in the case of demonetisation and GST. The first affected the informal sector, the second the formal one. How do you tell which caused the greater decline in output, spending and sentiment?

Then of course there all those things that can’t be measured, like the effect of income tax notices, bureaucratic recalcitrance, corruption, and now the new Bajaj variable, fear. Econometrics tends to dump all these in a single basket called the dummy variable whose value is either 0 or 1.

From this arises something called the dummy variable trap which is ok for solving the technical problem but useless for policy.

What I am saying is this: don’t over analyse a problem if you want to influence policy. By definition, short term policy is the equivalent of flying by the seat of your pants. You take your chances. In other words, when confronted with confusing inputs, rely on your judgment.

Had the Captain of AI 855 done so, that crash wouldn’t have happened.

Twitter: @tca_tca