A reminder: My 10 rules of forecasting
It’s election day in the United States and I bet everyone is breathing a sigh of relief that it is finally over - or is it? There is a non-negligible possibility that we don’t know the winner of the election at the end of today or even within a relatively short time frame after the polls close. But there is also a non-negligible possibility for a landslide victory.
As always, pundits will go into overdrive to predict the outcome of the election and the longer we have to wait for a decisive result, the more extreme the forecasts will become and the more likely it is that conspiracy theories and falsehoods will take hold and influence the narrative. Thus, it is important for all of us to remind us how to make forecasts. The most egregious forecasts (“Landslide for Donald Trump”, “No one wins the electoral college”) are going to get way more airtime than they deserve - especially after the perceived failure of forecasts four years ago. But did they really fail? If a pollster predicts that Biden has a 80% chance of winning the election, can we really say the forecast failed if Trump wins? If the forecast is correct then Trump should win in one out of five cases anyway. And if you run the election only once, then that one case could materialise.
My advise is: pay no attention to extreme forecasts.
And today, more than normal, it is important to remind ourselves of my 10 rules of forecasting so we can watch TV on election night and not fall prey to unreliable forecasts:
Data matters: We humans are drawn to anecdotes and illustrations but looks can be deceiving. Always base your forecasts on data and not on qualitative arguments. Euclid’s Elements was the first book on geometry, yet it does not have a single drawing in it. Corollary 1: Torture the data until it confesses, but don’t fit the data to the story. It is easy to fall into the trap of data mining. Corollary 2: Start with base rates. The assumption that nothing changes, and an event is as likely in the future as it was in the past, is a good starting point, but not the end point. Adjust this base rate with the information you have at the moment.
Don’t make extreme forecasts. Predicting the next financial crisis will make you famous if you do it at the right time but will cost you money and reputation in any other instance. Remember that there are only two kinds of forecasts: lucky and wrong.
Reversion to the mean is a powerful force. In economics, as well as in politics, extremes cannot survive for long. People trend towards averages and competitive forces in business lead to mean reversion.
We are creatures of habit. If something has worked in the past, people keep doing it almost forever. This introduces long-lasting trends. Don’t expect these trends to change quickly even though there is mean reversion. It is incredible how long a broken system can survive. Just think of Japan.
We rarely fall off a cliff. People often change their habits at the last minute before a catastrophe happens, but for behavioural change to happen the catastrophe must be salient, the outcome must be certain, and the solution must be simple.
A full stomach does not riot. Revolutions and riots rarely happen when people have enough food and feel relatively safe. A lack of personal freedom is insufficient to create revolutions, but lack of food, medicine or injustice all are. The Tiananmen Square revolt in China was triggered by higher food prices that students couldn’t afford. The Arab spring was triggered by food inflation as well.
The first goal of political and business leaders is to stay in power. Viewed through that lens, many actions can easily be predicted.
The second goal of political and business leaders is to get rich. Combined with the previous rule this explains about 90% of all behaviour.
Remember Occam’s Razor. The simplest explanation is the most likely to be correct. Ignore conspiracy theories.
Don’t follow rules blindly. This applies to these rules as well.