Superforecasting 2017

Between the beginning of December and around about this time each year, we are assailed with forecasts for the year ahead. Sometimes the forecasts turn out to be right, other times they are badly wide of the mark. Years of bitter experience have taught me that making a point forecast for any economic quantity one year ahead is often an exercise in futility. But any forecast based on a reasonably well-thought out story is better than taking no view at all and trusting to luck. I was thus intrigued by the findings of the recent book by the political scientist Philip Tetlock and journalist Dan Gardner entitled “Superforecasting: The Art and Science of Prediction” (here).

Essentially, Tetlock and Gardner conclude that forecasting is a skill that can be learned although some people are better at it than others. The so-called superforecasters generally manage to outperform experts in a wide variety of fields because they adopt an eclectic approach to analysis, preferring to process information from a wide variety of sources. Tetlock assumes that forecasters can be divided into two categories – hedgehogs and foxes. Hedgehogs tend to have in-depth understanding of a small number of areas, whereas foxes believe the world is a very complex place and tend to avoid shoe-horning their ideas into a limited number of boxes. Perhaps not surprisingly, foxes make the best superforecasters.

Although they tend to be smart people, Tetlock finds that superforecasters are in no way geniuses. They tend to look at a wide range of information in making their judgements and are happy to revise their assessment if new information becomes available (in the same way as Bayesian statisticians, as I noted here). Whilst my record disqualifies me as a superforecaster, I was struck by one of the lessons which came out of the analysis, which is that they think in fine gradations. Thus, rather than offering an outcome with a probability of 60-40, a superforecaster might carefully weigh up the evidence and instead offer odds of 63-37.

This brought to mind my own deliberations in the immediate wake of the Brexit vote when I was prevailed upon to offer an unambiguous view of what would happen next, but the more I thought about the issues the less clear they seemed. I recall my assessment on 27 June was that the UK was likely to leave the EU with a probability of only 59% whereas in the wake of the Conservative Party conference in October, I raised the likelihood to 90%. As new information comes in, that figure may well change again. This highlights a view which is gaining common credence – and one which I have long been convinced by – that the central case forecast is by itself not much use unless we attach some form of weight to show the degree of conviction with which we hold to the view. Otherwise the forecast becomes a binary decision which is either going to be right or – more often – wrong, which is when forecasters open themselves up to the charge that they have no idea what they are doing.

So in the face of all these caveats, what are the key issues we should be looking out for in 2017? The biggest local risks are: (i) the UK triggers Article 50 in March without making any contingency plans in the event that discussions with the EU prove more difficult than expected; (ii) Marine Le Pen wins the French presidential election; (iii) Angela Merkel fails to win the German election.

As regards (i), I genuinely do not know how this will pan out. I am working on the assumption that the Supreme Court will uphold the High Court judgment and that parliament will be allowed a say on the triggering of Article 50 which will delay its implementation. As a result, I currently assign a probability of 45% to this outcome. On (ii), the polling evidence suggests (for what it is worth) that Ms Le Pen has no chance of winning the second round of voting, and consequently I would give this a probability of 25%. And I see no alternative to Angela Merkel continuing as German Chancellor, so this is assigned a probability of 10%. The joint subjective probability of all these outcomes occurring is just over 1% - negligible but not impossible (which is how in early 2016 I characterised the joint likelihood of the UK voting for Brexit and the US for President Trump).

On the other side of the Atlantic, I would be surprised if Donald Trump can do much damage to the US economy in 2017, although further out it is likely that greater difficulties will become evident. He is unlikely to build his wall on the Mexican border; jail Hillary Clinton; deport illegal immigrants or completely dismantle Obamacare. But I suspect markets will not get the benefit of the hoped-for fiscal stimulus and as a result I would be surprised if US equities continue rising much beyond the spring (I won’t even put a probability on this one).

We should be in no doubt (as if anyone needs reminding) that the year ahead is more plagued by uncertainty than at any time since 2009. But as I say to journalists who ask whether I expect any surprises, it is the unexpected surprises which tend to do the most damage, and since by definition they are unknowable, time has a habit of making fools of us all.