Over-reaction to new data. What have you done for me lately ?

An explanation of over-reaction to new data starts with a principal agent model. The idea is that the forecasts are made by agents who are trying to impress people who pay for the forecasts.  Another, and more important, case is that of professional money managers who are investing other people’s money and trying to convince their clients that they are making valuable choices.  In these cases, it is useful to claim one has learned something new and especially useful to argue that the conventional wisdom is not satisfactory. In fact, in the data, forecasts can be improved by moving them towards lagged forecasts by the same professional forecasters, and can be vastly improved by moving them towards the average of lagged forecasts.

Something similar might be happening in the much more general case of people who are not agents working for a principal. We want to believe we have learned something new and also that we know useful things which are not generally known.’

In the experiments which show under-reaction to new data, that is anchoring, the experimental subjects can not be proud of the new information which is provided by the experimenter. So the motivation for over-reaction discussed above does not exist. A different explanation of the difference focuses on anchoring.  It might be that people are reluctant to admit that something they just said was a bad forecast.

In both the case of under-reaction and over-reaction the story focuses on self flattery – either the conviction that we were right to begin with or the conviction that we have learned something new and important. It is definitely a problem that there are similar and roughly equally plausible explanations for both over-reaction and under-reaction. That doesn’t mean that either explanation is invalid.