Every year I post an update of the ‘average return doesn’t exist’ blog to show that the average return on equities is mostly a statistic, and is seldomly seen in reality. And last year is, again, no exception. The Dow Jones Industrial Average realized a negative return of -5.5% in 2018, far away from the average return of 7.3% since 1900. It also underpins that using this long-term average for forecasting annual returns, as many professional forecasters tend to do, will do little good for your ‘guru’ status. Those who dear to put in ‘obviously’ far too bearish or far too bullish forecasts have (much) better odds in getting it right.

The graph below shows the calendar year returns on the Dow Jones Industrial Average Index since 1900, ranked from lowest (left) to highest (right). During this period, the return on the Dow Jones has averaged 7.3%. From this perspective, a return forecast of let’s say 5 to 10% doesn’t sound that crazy. You take some safety margin around the long-term average and you’re probably good, right?

Wrong! A forecast like the one above is pretty naive. If we zoom in on the graph, it reveals that, since 1900, the return on the Dow Jones has been between 5-10% in just ten occasions (the black bars). That’s not that often given the time span of 119 years. The Dow Jones return comes in between the 5-10% zone roughly 8% of the time, or just in one out of every 12 years! Conclusion: ‘Gurus’ forecasting a return of between 5-10% would have got it wrong most of the time.

Eternal optimists and ultra bears

Instead, you would have been much more successful by predicting a return of 20% or more for every single calendar year. In no less than 31 out of the 119 years since 1900, the Dow Jones generated a 20%+ return. This, ‘obviously’ too optimistic view on the stock market, has a success rate of 26%(!), more than three times bigger than the success rate of the conservative 5 to 10% return forecast.

And what about the ultra bears? They too had beaten the ‘stick with the average’ forecasters. Pessimists, who dared to predict a negative return of 20% or less for every single calendar year, got it right in eleven out of 119 years since 1900. Still one more than forecasters predicting a return between 5-10%.

If anything, the statistics above show that the number of ‘doomsayers’ and ‘eternal optimists’ in the market is probably too low, instead of too high. Too few forecasters take the historical return distribution into account and bet on massive equity rallies or heavy losses. Perhaps there is some logic behind it. Because who has the ability to forecast the Dow Jones for 119 years? But more importantly, predicting a stock market bubble in a year when stock prices collapse is probably a bit too damaging for your guru status.

One response to “The average return doesn’t exist (2019 edition)”

Interesting blog and post!
I think you make a good point, but nonetheless seems to me you’re measuring with two sticks. You’re comparing giving off a “two-sided estimate” (5-10%) with a one-sided one (> 20%). Clearly the two-sided one is harder in general. If a one-sided estimate is allowed, the best one, according to your metric (success rate), would be to say > -100%. Or if that’s considered cheating, I’d estimate > 0%.
On top of this, it depends on your metric. If you evaluate by e.g. mean-squared error instead of success rate, a 5-10% estimate would beat a (say) 20-25% estimate. It’s debatable what the better metric is, but as you point out, a forecaster would probably incur a large cost if the forecast was far off.
(I am not a professional forecaster, btw 😉 )

Interesting blog and post!

I think you make a good point, but nonetheless seems to me you’re measuring with two sticks. You’re comparing giving off a “two-sided estimate” (5-10%) with a one-sided one (> 20%). Clearly the two-sided one is harder in general. If a one-sided estimate is allowed, the best one, according to your metric (success rate), would be to say > -100%. Or if that’s considered cheating, I’d estimate > 0%.

On top of this, it depends on your metric. If you evaluate by e.g. mean-squared error instead of success rate, a 5-10% estimate would beat a (say) 20-25% estimate. It’s debatable what the better metric is, but as you point out, a forecaster would probably incur a large cost if the forecast was far off.

(I am not a professional forecaster, btw 😉 )