I think the answer to the first question is "yes, but not easily". We appear to be witnessing a structural change in the makeup and behaviour of the electorate; let’s call it the anti-establishment trend. Models often fail when there is structural change. But the response should be to improve the model by incorporating the new information, rather than not to use models at all.

Pollsters will be spending many sleepless nights to improve their methods

The same applies to the current situation. Reverting to gut feeling is not the answer. I’m no expert on what you would change, but I’m sure the polling industry will be spending many sleepless nights trying to improve their methods. And I would expect, over time, for the errors to shrink…until the next time there is a structural change.


I have a bigger problem with the failure of commentators and pundits to understand the changing world. They are less bound by models. But it seems the echo chamber of the Washington and media elites prevented many from understanding the changing mood of the population in the rest of the country; as much as it prevented many analysts based in London from understanding the dynamics that led to Brexit.

I remember a political analyst who gave me a '99% guarantee' that Trump won’t be the next President

Working in Finance in London, it was difficult to find anyone who wanted to vote ‘Leave’, but it was clearly a mistake to extrapolate from the London-centric experience to the rest of the country. On the US election, I remember a political analyst from a major US broker, who in April gave me a '99% guarantee' that Trump won't be the next President.


It’s true that there were some that predicted a Trump victory, but there is always someone willing to provide a prediction for anything. There were some that had money on Leicester City winning the Premier League last season as well. However, it’s one thing to make an extreme prediction with only reputational upside from getting it right and another to invest other people’s money on that basis.

 There is always someone willing to provide a prediction for anything

Instead, during the night of the election, our team were glued to our screens. We believe in multi-asset diversification as the first line of defence against these kind of idiosyncratic shocks. However, where appropriate, we also rapidly tilted our portfolios from a marginal Clinton win to a Trump victory before dawn on Wednesday.


So what about us? Could we have predicted a Trump victory? Or more appropriately, could we have used a Trump victory as our base case? Realistically and unfortunately, I don’t think so. We do not have to stick slavishly to poll-based forecasts or bookmaker’s odds on an event like the US election. We can assign different odds, and we did. But we have to admit that we are no experts in polling, modelling polling data or predicting political events; this is not where we have an edge. So our conviction in deviating from consensus must be limited and going from bookmakers’ odds of 25% to more than 50% would have been irresponsible.

We can assign different odds, and we did

Our judgement throughout the election campaign was to pitch the odds of a Trump victory as a bit higher than the poll-based models and bookmakers’ odds, by around 5-10%. This was based on several factors:


  • the experience from the EU referendum in the UK
  • the view that there was a general shift towards anti-establishment populism in Western Democracies
  • the view that there was a risk of a structural change that would make models less reliable
  • precedents from regional elections in Germany where the AfD succeeded in changing the electorate

So our hope and expectation is that polling methods will improve in future elections, but until we see evidence of this we will maintain our scepticism of poll-based forecasts and keep making adjustments. But what we will not do is follow our gut feeling. Instead we ensure that we're prepared for such events and how they could impact our portfolio - we'll look to address our approach in a separate post.