Astrologers Would Likely Beat the Fed at Inflation Forecasting – Mish Talk

When Theory Ends 

Eurointelligence has an interesting column on “When Theory Ends“. 

They refer to the ECB, but if you substitute the Fed for ECB the article is just as accurate.

One of most egregious cases of where theoretical models fail is inflation forecasting. As we keep writing, they have no hope in hell of capturing underlying trends because they exclude too many interlocking factors, and because they do not take account of structural societal shifts. The ECB’s inflation forecast has been so appalling that it would have been beaten not only by the proverbial dart-throwing monkey but probably even by astrologers. This is for the simple reason that the model is biased towards a return of inflation back towards the status quo. Any method without such a bias will outperform it. The rise of modern statistical tools has also started to impact economic policy making and economic research. We would expect central banks and governments of the future to hire more data analysts and statisticians and fewer paper-writing economists.

We have also witnessed the theory-versus-statistics debate in our own line of work. As observers of European media, we have been relying on translation services, without which we would not have been able to read Finnish or Greek newspapers. When we started 15 years ago, the world of translation software was dominated by services using structural linguistic models. The linguists scoffed at the statistical translation techniques developed by Google and others at the time. Statistical translation ended up winning the battle. Things in life that are too complex and too chaotic to be subjected to linear structural models. Language is one of them.

The issue is not only that computers are better at finding relationships between data than humans. Perhaps the bigger issue is their lack of cognitive biases. We have never met an economist who admits that his or her framework is wrong on the grounds that it is not supported by empirical. 

The more beliefs creep into your model, the more prone your profession is to meet the same fate as the first generation of translation modellers. If you believe that money and debt are the same thing, the central premise of modern monetary theory, no data will ever get you out of this belief system. Macroeconomics in particular is closer to language translation where error correction is important, as opposed to areas where scientists are constantly updating their understanding of reality.