David Aaronovitch has a nice piece in the Times on the stories that Trumpites and Brexiters can tell themselves to avoid admitting being wrong*. I'd add another problem - that the ratio of noise to signal is so high that firm evidence is hard to find.

Let's take GDP per head. Remainers claim that this will be lower under Brexit. NIESR estimates that Brexit will cut GDP by up to 7.8% by 2030, and CEP economists (pdf) by up to 9.5 per cent.

But trend growth is noisy, even over longish periods. For example, in the last ten years the standard deviation of annual growth has been 2.3 percentage points. This implies that even over ten years the standard error of annualized growth has been 0.7 percentage points.

The intuition here is simple. If we take any time period, it's possible that it contains a disproportionate number of good years. It would be wrong to infer from this that trend growth is high: what's really happened is that we got lucky. Conversely, if our sample contains some bad years, trend growth will look low when in fact we've been unlucky.

Even over periods as long as ten years, luck by no means cancels out.

My chart shows the point. It shows that it's possible that trend growth didn't change much from around 2% from the 70s through to the late 90s.

And here's the problem. Even the worst-case costs of Brexit aren't much bigger than one standard error. CEP's estimate of a 9.5% hit is equivalent to 0.9% lower average growth over the 11 years from 2019 to 2030. That's only 1.3 standard errors. Which is the sort of thing we could get through bad luck.

Or it might be cancelled out by good luck. It's possible that growth will be higher in the 2020s simply because Brexit is one of only many influences upon growth. Maybe the scarring effect on animal spirits of the financial crisis fade. Maybe we'll get more sensible macroeconomic policies. Maybe the slowdown in world trade will be reversed. Or perhaps some new innovations will boost investment. There are many possibilities.

If this happens then even if the CEP's worst case scenario is bang right, Brexiters will be able to say in (say) 2030: "you said Brexit would hurt growth, but in fact the economy's done as well in the last 10 years as it did in the previous 10. You were wrong." Remainers will reply that things would have been even better had we stayed in the EU. Ex hypothesi, they'd be wholly correct. But this won't be clear from the facts.

My point here is that it's quite possible that the economic impact of Brexit won't be proven, at least sufficiently clearly to persuade Brexiters. This isn't just because of our reluctance to admit we were wrong. It's also because a combination of the Duhem-Quine problem and a high noise-signal ratio means that the debate might, in practice. be intractable. I find this a depressing prospect.

* Scoffers have pointed out that Aaronovitch has himself used such epistemic defence mechanisms in his continued support of the Iraq war. I think that just shows we are all much better at spotting cognitive biases in others than we are in ourselves.