Join our community of smart investors

Implementation failures

Even when we have good theories, we can lose money by not implementing them well
October 25, 2017

Investing is applied social science. Everything you do is a test of particular hypothesis. Holding equities tests the theory that the equity risk premium should be positive. Buying tracker funds tests the idea that the efficient market hypothesis isn’t grievously wrong. Buying housebuilding stocks tests the hypothesis that cyclical risk pays off in normal times. And so on.

Theories, however, are general, whereas tests of them are specific. This creates a problem: when we lose money, is the theory falsified, or merely our particular implementation of it? Philosophers call this the Duhem-Quine problem: scientific hypotheses can never be tested singly, but only jointly with others which means we cannot always be sure which ones are right and which not.

It’s this problem I’ve run into since selling in May. Since I did so, the All-Share index has given a total return in real terms of just over 4 per cent. That means I was wrong.

But is the problem with the theory, or my specific implementation of it? The theory behind selling in May is that investors’ appetite for risk is seasonal; it increases in the spring, causing shares to become overpriced, and declines in the autumn causing them to become underpriced. And the market’s behaviour this year is consistent with this theory: between late May and mid-September UK share prices did indeed slip back.

Exactly the same problem applies to Woodford’s equity income fund. This tests a general theory – that on average income stocks do well. And as with 'sell in May' we’ve good historic evidence to support the theory: in the last 20 years the total return on the FTSE 350 high-yield index has averaged 7 per cent per year compared with 4.9 per cent per year on the low-yield index.

Neil Woodford’s particular implementation of the theory has, however, gone awry. Losses on stocks such as AA, Provident Financial and AstraZeneca mean his fund has underperformed.

Is this underperformance a refutation of the general theory that income investing pays, or just a sign that Mr Woodford’s precise test of the theory has failed?

Two things complicate the question. One is that what we have here is one failure after years of (on average) good performance. Mr Woodford has beaten the market for much of his career, just as 'sell on May Day, buy on Halloween' has worked on average most of the time: since 1966, total real returns on the All-Share index have averaged minus 0.5 per cent between those days.

Secondly, it’s easy to set the bar low so that our general theories seem to pass. Of course, we can point to many income stocks that have done well this year as evidence that income investing still works, just as the market’s fall between late May and mid-September corroborates the theory that appetite for risk is seasonal. But these are easy tests to pass.

In both cases, then, we have the same problem: we can’t be sure that our previously good theories have broken down, or just that they’ve been badly implemented this year.

Personally, my response to this Duhem-Quine problem is to ask two questions.

One is: which mistakes would I rather make? Every investor has the occasional failure. The question is: which failures do we want to be exposed to?

When I sold in May, I consciously chose the danger of missing out on a small gain. I thought that was a price worth paying for avoiding the risk of a big loss: 11 of the past 53 May Day-Halloween periods have seen real losses on the All-Share index of over 10 per cent. By contrast, the Halloween-May Day period has seen only two such losses and more big gains.

The same question applies to income stocks. When you buy a high-yielding defensive stock (such as AstraZeneca), you’re taking on idiosyncratic stock risk. On average, you are well-rewarded for doing so as defensives tend to beat the market. The risk of buying the wrong defensive might well therefore be worth taking. The same, though, can’t be said for some other income stocks. Some recovery plays (such as Provident Financial) expose you to adverse momentum effects that can be powerful. For me, these mean that the small chance of a big bounce in price isn’t worth taking.

My second question is: which should we trust – well-established statistical tendencies or unaided judgment? History tells us that there have been a few patterns in stock markets: seasonality; good performance by defensives and momentum stocks; and some forms of cyclical risk have paid off. And we’ve reasonable theories to explain why these work.

Of course, these patterns are only statistical tendencies which don’t work all the time. But I would rather back them than rely upon the pure judgment of ad-hoc stock-picking which we know can go awry because of countless cognitive biases. (When you buy an income fund, you’re both backing a reliable pattern – that income investing pays on average – and backing a fund manager’s ad hoc judgment. Which is the Duhem-Quine problem again.)

For me, sticking to proven rules is the least-bad solution we have to the problem that our judgment and knowledge of a complex world is inherently limited. It’s not perfect. But nothing is.