Throw Out the Probability Models by Nassim Taleb
After the events that started in 2007 and the subsequent reactions by economists, anyone who takes the current economics establishment seriously needs to spend time in a sanatorium.
This does not mean we should write off the entire body of knowledge. By now, we can see what works and what does not work. Simply, a certain class of consequential rare events, what I’ve called “black swans,” are not predictable and their probabilities unmeasurable, so anything that relies on a computation of the probability of these events should go out of the window. Now. Such models induce fragilities and bring harm. We’re better off with no model than with a defective model, something people understand intuitively, but they tend to forget when they don’t have “skin in the game.” If you are a passenger on a plane and the pilot tells you he has a faulty map, you get off the plane; you don’t stay and say “well, there is nothing better.” But in economics, particularly finance, they keep teaching these models on grounds that “there is nothing better,” causing harmful risk-taking. Why? Because the professors don’t bear the harm of the models.
The good news is that those models that miss rare events also break down when one introduces a higher layer of uncertainty into them, called “parameter uncertainty”. This gives us a fault detection mechanism. What goes out of the window? The entire discipline of modern finance and portfolio theory (the theories named after Harry Markowitz, William Sharpe, Merton Miller), the model-based methods of Paul Samuelson, much of time series econometrics (which don’t appear to predict anything), along with papers and theories that are based on “optimization.” These bring fragility into the system. So, simply, we would have great jumps in knowledge if we avoided teaching these models, and replaced them with anything, even gardening classes.
But the broad principles of economics survive such expurgation. We should just ignore much of what has happened in the past half-century of trying to be too sophisticated with quantitative and probability-based models, ending up in dangerous pseudo-science.
Continue reading what other debaters have to say at the New York Times.