Since 1985 the All-Share index has, on average, delivered a real total return of just 0.1 per cent in the six months from May Day to Halloween. That compares with an average return of 8 per cent from Halloween to May Day. The UK is not unusual in this. Ben Jacobsen at the University of Edinburgh has shown that the same pattern exists in 81 of 108 national stock markets since records began.
In fact, May Day is a stronger signal than a conventional valuation measure. Since 1985, variations in the dividend yield have explained 10.6 per cent of the variation in six-monthly returns to the end of April and October. But a simple May Day dummy explains 13 per cent of the variation. If you think valuations matter you should therefore think that May Day does, because it's a stronger signal. (The May Day signal is also strong even controlling for the dividend yield).
My favourite explanation for this is that risk aversion is seasonal. As the nights get lighter in the spring we become more optimistic and more willing to take risks. This drives share prices up in April, to levels that are often unsustainable. And as the nights draw in, in the autumn, we become more anxious, which depresses prices: September has historically been the worst month for returns. Such swings in sentiment are reflected in our Pagan festivals: May Day is a celebration of fertility and hope, while Samhain (Halloween) is a time of fearfulness.
This explanation is consistent with an otherwise odd fact - that UK equities tend to outperform Australian ones in our winter and their summer but underperform them in our summer.
It also answers the obvious question: why don't investors sell before May and buy in the autumn and thus remove this anomaly? It's precisely because swings in risk aversion make them reluctant to do so: selling in the spring requires that we sell when we are optimistic and buying in the autumn requires us to do so when we are pessimistic. It's difficult to trade against our emotions.
But springtime optimism is even more unwarranted than the low average return between May Day and Halloween would suggest. This is because the distribution of returns in the summer is nasty.
My chart shows this. It shows returns in the summer (May to October inclusive) and winter. This shows that big gains are unlikely in the summer; since 1966 there has been only one year in which real returns exceeded 20 per cent; that was 1980. This is fewer than we'd expect if returns were normally distributed. By contrast, 10 winters have seen such gains.
Granted, small gains are quite likely in summer: 18 of the past 49 have seen returns of between zero and 10 per cent. But big gains are improbable.
Instead, what's odd is the increased likelihood of big losses in the summer. Eleven of the past 49 have seen losses of more than 10 per cent, compared with just two of 49 winters. The financial crises of 1998 and 2008 and most of the bursting of the tech bubble in the early 2000s all occurred in the summer.
I'm not sure this is just bad luck. For one thing, given the summer's low average returns, four falls of 20 per cent or more are just what we'd expect if returns were normally distributed. And for another, it could be that springtime optimism makes the market more vulnerable to bad news.
For me, this distribution strengthens the case for selling in May, insofar as one can do so without incurring high dealing costs or tax liabilities. Doing so protects us from the risk of a big loss while there's not much chance of us missing out on big profits. For investors who are loss-averse - that is, more scared of the small chance of big losses than of the large chance of small losses - or those who are approaching retirement and want to lock in their wealth, selling in May looks like a reasonable thing to do.