Why you can seldom say that a forecast is wrong

The publicity surrounding a set of depressing growth forecasts for the UK economy last week has, inevitably, been accompanied by a chorus of commentators  reminding us that forecasts are almost always wrong. In fact, you can seldom say that a ‘proper’ forecast is wrong. By a proper forecast I mean one that acknowledges the unavoidable uncertainty we face when trying to estimate what may prevail in the future.

The forecasts scattered across the headlines are invariably single numbers: 1.3% growth in 2020, 2.2% inflation in 2019, a 4.3% rate of unemployment in quarter 3 of 2018. But, underlying these figures, and less attractive to newspaper editors, there is usually a more detailed assessment of a range of possible futures and their associated probabilities. A forecaster’s model may suggest, for example, that there is a 10% chance that growth will be less than one percent, a 15% chance that it will be between one and two percent and so on. The single number that surfaces –a so called point forecast – simply represents an average of all the possible outcomes that can be foreseen –after taking into account their chances of occurrence.

In theory, a point forecast of 1.3% growth for 2020 means that, if we could re-run the 2020 economy a large number of times allowing different combinations of chance events to occur each time then, on average, we would expect to have growth of 1.3%. Of course, we will only experience the 2020 economy once so we will never know what this true average would have been. To claim that a point forecast is wrong amounts to saying that an average is wrong when you’ve only seen one outcome. Imagine someone concluding that an estimate that the average height of American men is 69.3 inches must be wrong because they have just been speaking to an American man who is 73 inches tall. Condemning a single forecast as being wrong is no different.

The same applies to forecast of events. If I forecast that you won’t win the jackpot in the National Lottery next week, I must mean that I think this is the most likely outcome of your gamble. Thinking otherwise would suggest that I have delusions that I can see the future with certainty – a trait usually reserved for astrologers, necromancers and their like. If you win the jackpot, you can’t say my forecast was wrong. Not winning was still the most likely outcome even though things didn’t turn out that way. Similarly, if I forecast that Manchester United will beat Arsenal when they next play soccer at Old Trafford and Arsenal win, this does not prove that a Manchester United win was not the most likely result. If the game was replayed a hundred times, Manchester United might win 75% of the time.

In a single result we just don’t have the luxury of being sure that the most likely event has revealed itself.

Paul Goodwin

 

When can you trust a forecast?

We all consume forecasts -weather forecasts,  political forecasts, forecasts of sporting events, economic forecasts. We also all make forecasts -how will my savings fare in the future?  What will my new job be like? Will I enjoy the holiday I’ve booked in Spain?

Recent events have caused many to question our ability to make accurate forecasts. Few people saw Donald Trump’s election victory in the USA. Most economists didn’t foresee the credit crunch of a decade ago.  The last two UK election results and the Brexit vote shocked  the majority of political experts.

This blog  will look at when we can trust a forecast -whether it’s our own or one made by someone else -and how we can make more reliable predictions.  It will also look at how forecasts influence our decisions and the risks we are prepared to take.

You can read more about this in my book Forewarned: A Sceptic’s Guide to Prediction (Biteback Publications), but the blog will highlight  the latest developments and thinking in the world of prediction and forecasting .

Paul GoodwinForewarned Final cover