Recession?

Forecasting business cycle turning points (i.e., peaks and troughs) is not very accurate.

Having said that, the OECD (look that up in your Handbook) maintains a set of global leading economic indicators, and it is starting to point towards recession.

BUT, recessions are rarely global. They can be transmitted broadly like contagion, and in fact this is what happened with the recession and financial crises of 2007-9. But that’s fresh in memory; it doesn’t always work that way.

FWIW: The U.S. has its own leading indicators for our country alone, and they are not indicating recession at this time.

************************************************************

The big problem in forecasting turning points is a familiar one from basic hypothesis testing (and from common experience with medical tests too).

Recall that in statistics you never know what “truth” is. All we do is state a null hypothesis, and the main goal in the choice of a null is that it can be tested, not that it’s true (although that would be nice).

Hypothesis tests give you an indication of this. We spend so much time designing experiments and tests because there are four possibilities.

• The null is true (even if we don’t know for sure), and the test can’t reject it.
• The null is true (even if we don’t know for sure), and the test can reject it.
• The null is false (even if we don’t know for sure), and the test can’t reject it.
• The null is false (even if we don’t know for sure), and the test can reject it.

We’d like our tests to always arrive at either the first or last one in that list.

In statistics, the second one is called a Type I error, and the the third one is called a Type II error.

(In medicine, the null is almost always that you don’t have what they’re testing for. A rejection is called a positive. So the second one is a false positive: you test positive, but you don’t really have the condition. That’s why you get second opinions, or re-run tests.† , The third one is then a false negative. These are a big problem in early detection of conditions).‡

In the case of forecasting turning points, the proportion of false positives and false negatives is startling. It’s often close to 50%, which is like saying the forecasts are no good at all.

So I worry a little about the fact provided in the article. They provide a false positive rate of 12.5%, and no rate at all for false negatives. Further, the OECD was around before 1970, but the article limits its sample of forecasts to after 1970, so I wonder if they’ve cherry-picked the data to make the forecasts look better than they are.

† BTW: re-running medical tests to look for false positive is not something that single payer healthcare plans take very seriously. That’s kind of a “victory” of accounting over good statistics practice.

‡ BTW: much of the high cost of healthcare in the U.S. is that we test earlier, and are paying for a lot of false negatives. On the other hand, the early detection leads to better long-term outcomes when the early test yields a true positive.