How not to deal with radical uncertainty

We have no doubt that the pollsters will try to find a way to tell us that they got it right after all, or at least that the results were not as they looked. Economic forecasters did the same thing after the global financial crisis. Macroeconomics has since gone into a decline that will be hard to reverse. We are confident to predict that the same will happen to the pollsters. For the most part, we have found that economic forecasting models and polling are worse than useless. Reliance on dumb statistical analysis is how we get elections wrong and misunderstand the dynamics of a pandemic.

All statistical methods are based on past patterns. Pollsters choose their samples on the basis of past voting distributions. The much-overrated Nate Silver attaches weights to polls on the basis of past forecasting errors. Behind all these approaches lies the notion that the truth about the future is hidden in some pattern of the past. We think this is not so.

John Maynard Keynes and Frank Knight talked about radical uncertainty, uncertainty that defies attempts to measure it. Modern probability theory and statistics are not based on their thinking, but on the Russian school that developed the axioms of probability in the 1930s. The world never looked back. People take statistics 101 courses and learn forecasting tools that have no space for doubt. It has many uses in life and in the sciences, but we find it utterly useless in analysing the politics and the economic developments that matter to us.

In political analysis, we apply radical uncertainty to processes that have no precedent, like Brexit negotiations, where it is simply not possible to attach probabilities. The best course of action is to admit this, and prepare for both outcomes. Elections generally do not fall into that category, but some do. Traditional polling, based on samples of usually around 1000 people, cannot capture dynamic effects. A promising future development is narrative analysis in social media. We ourselves tend to look at soft factors. The enthusiasm of Trump supporters and his massive rallies told us a lot more than hundreds of opinion polls. If you want to know how the second wave of the pandemic affects public opinion in Germany, for example, don’t listen to the polls but to local radio. The coverage is now very different compared to what it was in the spring. Radio hosts are much more frazzled. This tells you more than the poll.

An interesting way to look at election results and polling is offered by Nassim Nichalas Taleb and his co-authors, who have been using option pricing theory to make the observation of an inverse relationship between the volatility of an underlying asset, in this case the election result, and the volatility of an option, in this case the forecast. Taleb noted that the massive shifts in forecasts during the 2016 US elections violated this relationship, something with which we fully agree.

Our fundamental problem with polling is that it tries to take the uncertainty out of uncertainty. Policy makers, investors, and others have to take views on political outcomes. So do we. In a political sphere where the media have lost their control over the discourse, but have yet to realise it, this is becoming fraught with risk. Those who understand risk at a deeper level and who don’t suppress doubt will do better than those armed with a statistics 101 knowledge and a belief in the infallibility of their method.