Join our community of smart investors

The sell in May dilemma

There are powerful rules telling us both to sell in May and not to do so. There is no rational way of resolving this dilemma
May 1, 2019

I have a dilemma. On the one hand, I usually sell a lot of my equities in May and buy back around Halloween time. But on the other hand, there are strong signals telling me to stay in the market.

One of these is that foreign investors have recently sold more US equities than ever before, dumping $205.1bn (£158.83bn) of them in the year to February 2019. This suggests that investors’ sentiment is unusually depressed. In the past, this has pointed to rising prices as sentiment recovers.

Another signal is the dividend yield on the All-Share index. At 4.1 per cent, this is well above its 30-year average of 3.5 per cent. Above-average yields have in the past predicted above-average returns.

On the other hand, though, 'sell in May' is also a strong rule. Ben Jacobsen at Tilburg University and Cherry Zhang of the University of Nottingham have shown that the rule has worked in almost all stock markets since they began trading. Since 1997, for example, the All-Share index has fallen by 1.1 per cent on average from May Day to Halloween, but risen by an average of 5.1 per cent from Halloween to May Day. And the rule has worked nicely recently. The market fell 5.4 per cent between May Day and Halloween last year but has risen by almost 5 per cent since Halloween.

 

 

How do I adjudicate between these competing signals? I cannot use judgment. The 'sell in May' rule works precisely because our judgments go wrong at this time of year. Lighter nights in the spring make us too optimistic, which drives prices up too far just as darker evenings in the autumn make us too pessimistic. Yes, there are reasons to be bullish right now – we should see signs of economic recovery in the euro zone and China in the next few months, and perhaps an easing of US-China trade tensions. But history warns us that we tend to overweight bullish signals at this time of year.

With judgment unreliable, you might think pure statistics should guide us. I should ask: how have foreign buying of US equities and the dividend yield, when used alongside the time of year, predicted returns in the six months to May Day and Halloween in the past?

If I do this, I find that the time of year has been the only statistically significant predictor of returns since 1997. Adding foreign buying and the dividend yield shows these to be statistically insignificant.

This is a case for selling. Or is it? Here, we run into a debate in the natural and social sciences: how much store should we set by statistical significance?

Many researchers, such as Columbia University’s Andrew Gelman and the University of Chicago’s Deirdre McCloskey say: not much. Statistical significance, in effect, tells us to pay attention to weak but precise signals but to ignore strong but noisy ones. This might be unwise. Foreign selling of US equities and the high dividend yield are telling us there’s a chance of shares rising a lot in the next six months. Should we really ignore this simply because an arbitrary threshold of statistical significance has not been reached?

If we ignore statistical significance completely, post-1997 relationships between six-monthly returns, the time of year, foreign selling of US shares and the dividend yield predict the All-Share rising almost 10 per cent by November, with just a one-in-six chance of it falling.

These, however, are not the only lead indicators of returns. Another strong predictor has been the ratio of the developed world’s money stock to the MSCI world index. Thanks to the rise in the latter so far this year, this is now below average, which points to prices falling. This is simply because if global investors own lots of equities and little cash, any rebalancing of their portfolios is likely to be away from equities and towards cash, which would cause the market to fall.

If I combine this indicator with the time of year, post-1997 relationships suggest there is a 70 per cent chance of the All-Share index falling in the next six months.

My dilemma therefore remains. And I can’t resolve it by abandoning classical statistics in favour of Bayesianism, in which we update our prior beliefs in light of new evidence. My problem is that my prior beliefs conflict: I believe both in the 'sell in May' rule and in the predictive power of the dividend yield and other lead indicators.

Luckily, though, in the real world I do not need to make all-or-nothing decisions. I don’t need to sell all my equities. I might, however, sell a few. But this decision reflects my personal aversion to risk as much as the hard evidence – which, as we’ve seen, is wonderfully ambiguous.