Apparently we’re terrible at predicting earthquakes.
That’s what Chapter 5 is about, and it makes sense. Predicting rare events (Black Swans as Taleb would call them) is terribly difficult because you may only be working with a theoretical possibility and a limited data set. Even though we can get a general sense of where earthquakes may hit, we still don’t get much data on the major ones. This map from Wired shows some interesting regional information:
So with limited data points, the tendency for predictions is going to be to take every data point seriously and risk overfitting the model. The other problem is not going far enough back with the data. In Japan prior to the Fukashima disaster, evidence that major earthquakes had hit thousands of years ago was left off the risk assessment.
My most memorable earthquake experience was actually a few weeks after my son was born. I was feeding him, and I thought a large truck had gone by. Something felt off though, and he seemed surprisingly confused by it. When I went downstairs again, I checked the news and realized that “truck” had been an earthquake.