Saturday 19 February 2022

Forecasts That Affect the Forecast

I've seen so many people on Twitter berating SAGE forecasts because they don't come true. The one thing they fail to grasp is that a forecast is valid only at the time it's made, based on the available information and if circumstances don't change.


Forecasting the effects of a virus is very complex and a small change in behaviour can affect the forecast massively. Knowledge of the forecast itself can change behaviour, rendering the forecast redundant the minute behaviour is changed.

Take, for example, a company forecast that says the company will fail if it continues with a certain course of action. If, on reading the forecast, the company changes direction and avoids collapse, does that make the original forecast useless? Of course not - it was entirely valid at the time it was made. The forecast initiated the behaviour change.

Take a Waze ETA that changes while on a journey and says there's a traffic jam ahead that's going to result in an delay of an hour. It's rather stupid to say, when you get to your destination an, hour late, that the original forecast ETA of an hour earlier was rubbish. It was accurate at the time it was made, given the available information.

The same applies to the forecast of the number of cases, hospitalisations and deaths from a virus outbreak, On the basis of that forecast, people will change their behaviour and government may enforce that by mandating mitigating actions, such as distancing rules and enforcing mask wearing in public spaces. That itself will alter the trajectory of the forecast and in no way invalidates the initial forecast, especially if there are many unknowns at the time the forecast is made.

Forecasts that predict an impact that itself changes behaviour to avoid the impact are every bit valid. Climate change forecasts are similar, yet there will be people who deride such forecasts, despite them (and governments) having changed their behaviour to avoid the impending disaster.


No comments: