I have sporadically posted on this blog some forecasts from a neural network model that I have devised to predict the time path of the industrial production series in the UK. (For geeks, this has 24 lags of the seasonally differenced, logged variable as inputs, two hidden layers, and is fitted using a hyperbolic tangent.) Since the beginning of this year, however, the Office for National Statistics has published monthly data on Gross Value Added – a series which covers all parts of the UK economy, not just the comparatively small production sector. I’ve found it irresistible to use this series (data for which go back to 1997) to check out how, at June 2016, my neural network forecaster predicted output would evolve in the period after the 2016 Brexit referendum – and how that compares with the outturn. The graph below shows the forecast (in orange) and the outturn (in blue). The figures on the vertical axis are the annual change in logged output.
Beyond the first year after the referendum – when the depreciation of the exchange rate boosted exports – output growth in the UK has consistently fallen about 1.5 percentage points below the pre-referendum forecast. The cumulative effect of the shortfall in growth since the referendum amounts to the equivalent of around £900m per week. This finding is remarkably consistent with, and reinforces, the results from other (much less back-of-the-envelope) studies, including those by Gertjan Vliehge of the Bank of England, by Benjamin Born and co-authors, and by Josh De Lyon and Swati Dinghra at the LSE.
The fact that Brexit has already caused considerable economic damage seems inescapable.
