Invariant predictions
This is an updated version and translation of "De l'utilité des predictions invariantes" written back in August 2008.
To design any corporate strategy, reasonably predicting baseline elements like raw material prices, shipping costs, or tax increases is indispensable. In recent years, though, these baseline elements have been widely shifting away from what was considered normal. Blame the rise of the Chinese GDP, troves of instabilities in the currency market, the cost of energy, or anything else. The reality is that the building blocks of corporate strategy are becoming quite tricky to assemble, and any sort of prediction seems like magic thinking.
Back in 2008, Justin Wolfers playfully dismantled common ideas about this type of prediction on baseline elements. For example, he explained that for the price of oil, over a period of a month, a quarter, or even a year, it would be quite effortless to beat fossil fuel experts and traders at their predicting anything. How so? All expert methods for projecting the price of oil (based on econometric models, mathematical analysis of commodity financial markets, and geopolitical analysis) failed to do better over time than simply predicting that the price of oil tomorrow will be the same as it is today! This is referred to as a no-change forecast or an invariant prediction.
You may even recall some college maths class in statistics where your professor may have told you that to predict the weather tomorrow, nothing beats, on average, a simple description of today's weather. This works too for stock trends, epidemiological models, consumer behavior, energy consumption, demographic trends, or, yes, even technology adoption.
Now, if you have doubts about all this, keep in mind the two magic words I constantly use: on average.
For the price of oil in Wolfers' study, the invariant prediction was 34% better than the predictions of a dozen experts over a month and 18% better over a quarter. It doesn't mean that invariant predictions are, in any case, reliable at any given time. It just means that over a period of time, they mostly beat experts at their own game. Or, said differently: invariant predictions might be less wrong than experts on average.
Despite their widespread use as predictors of the spot price of oil, oil futures prices tend to be less accurate in the mean-squared prediction error sense than no-change forecasts. - What do we learn from the price of crude oil futures? (Ron Alquist, Lutz Kilian, May 2010)
But past the amusing nugget of statistics, the point I want to make is that any corporation large enough now has to write a big five or ten-year strategic plan periodically (and then update it every year as the initial plan goes sideways). And when working on your 2030 Corporate Strategic Plan, I'm in the camp of saying that you should probably spend the least amount of time possible on it. Such plans are now mostly exercises in reassuring shareholders and stakeholders about the fact that you do have a plan. And I do respect the importance of corporate communication, I just don't think you need to spend too much time finely modeling future fluctuations like the price of raw materials and elaborating on multiple complex scenarios to take them into account. Generally, an invariant prediction like "the price of the raw materials used will be the same in a year" allows you to build your baseline strategy, even when knowing it's not going to be the case. But that's also the part where you add an envelope of a pessimistic and optimistic scenario to navigate. I mean, this is Strategic Planning 101, and again, I don't want to go there.
