Join our community of smart investors

Outlook: normal

I say this not as a matter of opinion but because it is what is implied by some indicators that have had some success in predicting returns in the past. Three particular indicators have done quite well over the past 30 years:

■ The global money-price ratio, measured by the ratio of the broad money stock in the OECD to MSCI's world equity index. The idea here is simple. When investors' cash holdings are high relative to their equity holdings, they are likely to rebalance their portfolios away from cash and towards equities. Their buying will then raise prices. And when cash holdings are low, they are likely to sell shares to raise cash, thus depressing their prices. There has been, however, a tendency for the money-price ratio to rise over time (despite falling real interest rates), so I’ve detrended it for forecasting purposes.

■ The dividend yield. As you might expect, a higher yield predicts higher returns.

■ Consumption-wealth ratios. High ratios of consumer spending, relative to shares or house prices, lead to high equity returns. This is because consumer spending is partially forward-looking - we spend more if we think our economic future is bright - and there is wisdom in crowds: for everyone who is wrongly pessimistic, someone else is wrongly optimistic and the errors often cancel out.

My table summarises the ability of these indicators to forecast three-year total inflation-adjusted returns on the All-Share index since March 1985; I'm using quarterly data as we only have consumer spending numbers at this frequency, and three-year returns because shorter-term returns have a bigger random element and are harder to predict. (Also, I am, I hope, three years away from retirement so this time horizon is important for me at least.)

Predicting three-year returns
ForecastProb of loss (%)Prob of 10% lossR-squared
Money-price ratio13.427.915.30.38
Dividend yield15.124.312.40.45
Based on quarterly data since March 1985

All three of these indicators explain a big chunk of the variance of returns; R-squareds are between around two-fifths and three-quarters. However, because there has been a lot of variance in returns, this leaves a big chunk unexplained: the standard errors are high.

My table also shows what these indicators predict for the next three years. And there is, happily, rough agreement here. The money-price ratio and dividend yield agree that real total returns will be 4-5 per cent per year. This is slightly less than the average over the past 30 years (of 6.4 per cent) because both these indicators are below average. This implies that most of the returns we get in the next three years will come from dividends rather than capital appreciation.

Consumption-wealth ratios are slightly more optimistic, predicting returns of almost 7 per cent a year. This is because the consumption-All-share ratio is slightly above average, offsetting the slightly negative signal of a below-average ratio of spending to house prices.

By applying historic standard errors to these forecasts we can estimate the chances of a loss. I wouldn't set much store by the precise numbers here because I've assumed a normal distribution of returns and - much more problematically - that past relationships will continue to hold in future. The message is that there is a significant chance of zero returns or worse even over a three-year horizon.

We can combine these indicators to produce a forecast that has in the past been more accurate. Doing this tells us that the money-price ratio becomes statistically insignificant - it adds nothing to the dividend yield or consumption-wealth ratios - so I've dropped it. The new combined indicator can explain 70 per cent of the variance in three-year returns since 1985. And it shows no sign of breaking down; our indicator predicted real returns of 28 per cent in the three years to March against an out-turn of 26.3 per cent - which is as accurate as it gets in economics.

This indicator under-predicted returns during the tech bubble, under-predicted the extent of the tech crash and over-predicted the recovery from the 2008 slump but has otherwise done OK. In particular, it successfully warned us of impending negative returns in 2000 and 2006.

It now predicts a total real return of 16.7 per cent in the next three years, or 5.3 per cent per year: this is based on Q4's consumption-wealth ratios, which are the latest we have.

This might sound reasonable but unexciting. But it is consistent with the market actually falling between mid-2016 and mid-2017, as the dividend yield and consumption-wealth ratios were low at the end of 2013. So the path might not be a smooth one.

You might find all this surprising. This forecasting procedure told us nothing about the state of banks in 2008 and it systematically ignores a ton of evidence (or is it noise?) about the economy. And yet it has done reasonably well. This perhaps vindicates the words of Columbia University's Jon Elster: "Sometimes we can explain without being able to predict, and sometimes predict without being able to explain."

This might explain why this procedure has worked well. Because these indicators are so thin - in the sense of not telling us any detailed plausible story - that investors have underrated their power with the result that they have sometimes driven prices too far up or down. Perhaps the difficult thing is not so much generating forecasts for returns, but having the discipline to stick to them and ignoring all other evidence and our own instincts.