Join our community of smart investors
OPINION

Forecasts and forecasting

Forecasts and forecasting
February 27, 2020
Forecasts and forecasting

For the average door-stepping hack, Mr Cummings’ response may have been utterly delphic. Not, however – or so one hopes – to readers of Investors Chronicle. For us, Professor Tetlock’s manual, the product of research into forecasting going back to the early 1980s, is a way of life that Bearbull has written about (13 March 2015) and which was a magazine cover (13 January 2017).

Okay, ‘way of life’ might be an exaggeration. Even so, Mr Cummings’ intervention has given a publicity boost to Professor Tetlock’s methods and it provides an opportunity to stress how vital it is to apply discipline to the most important thing that investors do – make forecasts.

Sadly, too many investors practice their craft much as the Whitehall machine practices theirs, according to a review of Superforecasting that Mr Cummings wrote for The Spectator in 2015, they behave more like astrologers than scientists.

Astrologer-cum-investors – like their political equivalents – make the same misjudgments time and again. They fall for ‘cognitive biases’, the phoney bits of intuitive logic that prompt them to jump to false conclusions; they confuse forecasts with ego building, thinking that making smart and confident forecasts is the same as making good ones; they frame their forecasts in terms so vague that they can rarely be proved wrong. Most of all, they lack a rigorous framework on which to hang their forecasts.

A big step to building that framework should be taking the trouble to understand those cognitive biases; nowadays, there is no excuse not to. After all, behavioural economics, which studies these things, has become a fashionable branch of the dismal science since its best-known advocate, Daniel Kahneman, wrote a best-seller, Thinking Fast and Slow, in 2011. The ability to spot mental short-cuts such as ‘availability’, ‘anchoring’, ‘representativeness’ plus a host of cognitive fallacies should be basic.

Another basic requirement is numeracy. That doesn’t need to extend to fluency in differential calculus, but it does include understanding probability theory and concepts such as mean regression and Bayesian revision (updating estimates in the light of new data).

But arguably the most important component is the right frame of mind. When Professor Tetlock formed a team to compete in – and easily win – a forecasting tournament run by the US intelligence agencies he found that the best at their craft were cautious, modest (they knew their limits), open-minded (they did not confuse beliefs with opinions), self-critical, pragmatic, hard-working and tenacious; as well – of course – as being numerate and focused on probabilities. Also, even if not all were team players, they made better forecasts in teams, where their range of skills minimised the chances of group-think.

True, many retail investors will be condemned to go it alone. Yet they can help themselves by being rigorous in their approach to decision-making. Consider, for example, whether or not to buy shares in GlaxoSmithKline (GSK), the pharma giant that plans to split itself into two groups in 2022. That exercise would begin by listing the key identifiable factors that will drive Glaxo’s share price.

First, it would be good to have an initial target return in mind – say, 10 per cent over the coming six months (which would then be rolled forward). To achieve that, Glaxo’s shares might need help from the market. So how likely is it that London’s equity market rises 10 per cent in any six-month period? Churn that data and we can find the probability. Quite likely, actually – based on the past 20 years, there is about a one-in-five chance; a probability factor of 0.2.

Related to that, it would be useful to know how sensitive Glaxo’s share price is likely to be to movements in the market. This, too, can be quantified via the beta co-efficient of Glaxo’s price to, say, the FTSE All-Share index.

Then questions will focus more on Glaxo, starting, perhaps, with its share rating. In particular, Glaxo’s shares come with a fat 4.9 per cent dividend yield. How likely is that to help the share price bounce? Yet for the yield to be fat, the dividend must be paid, so how likely is it that Glaxo will generate the free cash to cover its cost?

Focusing on Glaxo’s operations, we can ask, based on comparable companies, what rating its consumer healthcare side may command when it is separated out and what that implies for Glaxo’s value now? On the pharmaceuticals side, there are questions about the likely sales performance of key drugs, both in production and in the development pipeline.

Against this proliferation of questions there are probability factors that each desired result will come through. Connected to that, each question gets a weighting to indicate its likely importance in the mix. Multiply each probability factor by its weighting and you get the weighted probabilities that, when added together, give a score where the higher the number, the better.

Sure, the questions don’t stop and nor does the revision. That’s how it is with probabilistic reckoning. No one said it is easy; nor is it a substitute for conventional analysis of a company’s finances and business prospects. And if the forecasts sometimes turn out to be rubbish, no matter because disciplined forecasting remains indispensible.