Reflexivity in Time Series

Why COVID-19 forecasting has failed, and a key ingredient to success in financial markets

Photo by Elias Kauerhof on Unsplash

Reflexivity occurs when a forecast affects the thing being predicted. Some say you cannot make reliable predictions under it. Others acknowledge reflexivity as key to their multi-billion bank account. How can we cope with it?

In a previous post, I identified reflexivity as an important thing about time series. Here, I will dive a bit deeper into this topic.

Forecasting systems are usually regarded as decoupled from the thing being predicted. This is generally true in domains where human behaviour has no impact on the outcome. A forecast that says it will rain will not affect whether it actually rains.

But, sometimes the human reaction to a forecast may change the outcome of the situation.

Photo by Héctor J. Rivas on Unsplash

Here’s a story told by George Smith about a textile company. The company’s analyst predicted the already typical sales drop in January. Attempting to avoid it, the sales manager launched a campaign to improve sales. Sure enough, instead of decreasing, the sales actually soared. The analyst end up…