The Econophysics Blog

This blog is dedicated to exploring the application of quantiative tools from mathematics, physics, and other natural sciences to issues in finance, economics, and the social sciences. The focus of this blog will be on tools, methodology, and logic. This blog will also occasionally delve into philosophical issues surrounding quantitative finance and quantitative social science.

Sunday, November 25, 2007

Forecaster as Entrepreneur (or How Do You 'Know' Without Knowing)

I came across a very interesting article in the Economics Focus feature of The Economist: A new fashion in modelling: What to do when you don't know everything (22 Nov. 2007). The article describes the work of NYU economist, Roman Frydman. In his new book, Imperfect Knowledge Economics, Prof. Frydman makes the case for using predictive models that are more intuitive and qualitative than those traditionally favored by economists.

As the article notes, Prof. Frydman starts from a very interesting proposition: "The forecaster is like an entrepreneur. He uses quantitative methods, but he also studies history, and relies on intuition and judgment." From this brilliantly crafted idea (and I am very sympathetic to this viewpoint), he crititques traditional methods of economic forecasting -- using exchange rates as a case study (of sorts). The method that comes under the most scrutiny is the dominant approach of the 'rational expectations' school. Much to -- what I can safely forecast -- the chagrin of the adherents of rational expectations, Prof. Frydman compares them to communist idealogues who insisted, to the bitter end, that communism could be made to work.

The problems with rational expectations -- which is so dominant in economics that even its critics essentially fall under its spell -- are legion and isn't worth detailing here. But there is one fact that best encapsulates its failings: Predictions of currency movements using rational expectations type approaches are usually worse than flipping a coin!

Someone who follows the bloodsport of the endless debates within economics and finance faculties between different schools of thought might think that Prof. Frydman might favor 'behavioural economics.' Surprisingly, although he does give behaviourialists credit when they critique rational expectations, Prof. Frydman is critical when behaviouralists turn around and try to 'precisely' forecast human behaviour (which, if you will recall, is what I call falling into the 'rational expectations trap.')

Prof. Frydman doesn't just cast stones; he offers solutions. What Roman Frydman proposes will be familiar to intellectually oriented traders -- mixing (or 'mashing,' as kids now a days would call it) quantitative methods with the spotting of qualitative regularities (or irregularities) that only a 'forecasting entrepreneur' (e.g., thinking traders and investors like George Soros, Warren Buffets, Nassim Nicholas Taleb, etc.) can do. Prof. Frydman's approach to uncertainty is essentially Bayesian as opposed to 'frequentist' -- i.e., it is inherently subjective. (See the latest book by financial theorist, Riccardo Rebonato: Plight of the Fortune Tellers: Why We Need to Manage Financial Risk Differently. In that book, Rebonato also advocates a Bayesian approach.)

The 'problem' with Prof. Frydman's -- and, I suppose, Riccardo Rebonato's or Nassim Nicholas Taleb's -- approaches are that the forecasting models that they are advocating will not be as 'precise' as the models advocated by the mainstream. My response to that: Good! What is the point of having precisely wrong models of reality (other than for falsification)? If we can get more accurate models of reality, we should be glad to sacrifice false precision.

In light of the recent credit crises, the closing comments in the article seem especially important:
Messrs Frydman and Goldberg are now turning their attention to the troubled subprime-mortgage markets, and the performance of the rating agencies. The rating agencies, argues Mr Frydman, have generally been better at rating corporate bonds than rating asset-backed collateralised-debt obligations. Why? One reason is that the rating agencies used both a mathematical model and the judgment of their in-house specialists when forecasting the default probabilities of corporate bonds; on subprime-related securities, they could only use mathematical models, not least because the instruments were so new. “They had no experience, no intuition, no entrepreneur,” he says. That is “empirical proof that relying on models alone is not wise.”

Labels: , , ,


Post a Comment

<< Home