Friday, 3 June 2016

The Treasury Isn't As Wrong On Brexit Disaster As You Might Think

The Treasury's report on Brexit comes to some pretty stark conclusions.

GDP will be down by 3.6% in the short run and 6.2% in the long run, relative to remaining in the EU. There will be an immediate (though shallow) recession, with unemployment up by 1.6 percentage points, and the prices of stocks, houses and the pound will all fall through the floor.

Of course, the Treasury's analysis is a deeply political document. So what should we make of all these numbers? How sceptical should we be about their weirdly precise estimates? Here I'm going to try to ballpark their estimates, and (spoiler alert!) they're actually quite compatible with reality.

I don't think the task of trying to forecast the impact of Brexit is a totally futile one. In a world with true uncertainty, it would be impossible to judge the likelihood of Brexit causing Mervyn King to mutate into a giant velociraptor and destroying London, but the best evidence in the literature suggests that this is quite unlikely. An estimate is better than no estimate – at least we've got a model that makes some assumptions, and we can evaluate the reasonableness of those assumptions. That's quite a bit better than no model.

One thing that's interesting about the Treasury's model is it's quite ad-hoc. For the long-run model they do a few things:
  • Assume trade benefits productivity.
  • look at the trade literature to get a number for how much a given change in trade affects productivity.
  • Come up with an estimate how much leaving the EU will impact trade.
  • Therefore, work out the long-run impact on productivity of leaving the EU.

There's a similar process for the short-run model.
  • Assume that in the long run the economy will converge to your long-run forecast.
  • Plug data for 1989-2011 into a regression to try and estimate the effect of increased uncertainty on various different variables.
  • Assume leaving the EU causes uncertainty to increase by 1 standard deviation.
  • Plug that 1 SD increase into the model, and see how that affects convergence to the long-run steady-state predicted by the first model.

This sort of ad-hoc modelling is sort of inevitable. It's hard to build a model that takes into account the full complexity of the situation. But the shortcomings of this kind of approach are somewhat obvious. It's not clear that all fluctuations in trade should have the same impact on productivity, for instance – it depends on the size of the gap in relative prices and the degree to which particular industries experience economies of scale. Similarly, 'uncertainty' is quite a nebulous concept, and it's a bit of a stretch to say that the kind of uncertainty caused by bank failures around the world, as in 2008, has the same economic impacts as the kind of uncertainty caused by firms having less information about what the tariff regime they face will look like over the next decade.

One of the things these sorts of models sacrifice is internal consistency. If your model can't fully encapsulate the relationships between different variables, if you're reliant on inputting parameters for things that should really be endogenous, you run the risk that the stories your model is telling about what's happening to GDP, productivity, asset prices and so on don't quite add up.

So how well does the Treasury's model do by this score?

The Promised Recession

Let's break out the outputs of the short-run model a bit more.
  • Real GDP goes down by 3.6%
  • Of this, 2.7% is a fall in potential GDP.
  • That means the remaining 0.9% of the decline is due to a demand-side recession. In other words, the negative output gap increases by 0.9ppts.

This fall in output is supposed to lead to a 1.6pp increase in the unemployment rate. That seems like quite a big swing in unemployment compared to, for instance, the Great Recession. How much spare capacity there was in the economy before and after the financial crisis is hotly debated, but this OBR working paper estimates a 6pp swing, from 2% above potential to 4% below. And that swing resulted in a trough-to-peak rise in the unemployment rate of about 3%. That was surprisingly low, and no-one is quite sure why, but it seems like a 0.9% rise in the output gap would be unlikely to lead to a proportionally larger rise in the unemployment rate.

The Stock Market Crash

Another weird discrepancy is the huge wedge between GDP and asset prices in the short term model. Stock prices are, after all, the discounted stream of corporate future profits. Assuming that there is not a significant shift in corporate profits as a % of GDP and the risk premium remains the same, we might expect stock prices to move by the same amount as GDP in the long run. But the Treasury model's central scenario predicts a fall in GDP

It makes sense for increased uncertainty to increase the risk premium on stocks. The costs of Brexit are probably not evenly distributed across all possible Brexit-worlds. Most of the time, the status quo is likely to be approximately maintained through the EEA, and it's the unlikely event of a much larger reduction in the UK's openness to trade that is really worrying. But the sheer size of the discrepancy suggests that the rise in the risk premium predicted by the Treasury is unrealistically large.

I'm now going to do a bit of back-of-the-envelope discontinuity analysis to see how realistic the 20% drop in equities is. Shortly before 4pm on Tuesday 31st May, two Guardian/ICM polls dropped suggesting that the nation was perhaps a bit more Brexit-leaning than markets had previously been assuming. On betting markets, the subjective probability of Brexit dropped by around 3-5%.

If you squint, you might be able to make out the impact of the poll on the FTSE 100:

So let's conservatively estimate a .5% drop in the stock market caused by a 5 percentage point increase in the probability of Brexit. That suggests that a 100 percentage-point swing (ie the difference between Brexit and no Brexit) would be about 10%. Quite a bit short of the treasury's esitimate, in other words, but in the same ballpark.

There are some other interesting implications of this sort of estimate. The market capitalisation of HSBC, the second-largest firm traded on the FTSE, is about £88bn. Let's assume that HSBC is about as exposed to Brexit risk as the average FTSE 100 company (this is the reason that I didn't pick Shell, the largest company on the FTSE by market cap). A 1% swing in the probability of Brexit would then cost HSBC's shareholders about 0.1% of its value. Therefore, if spending £88m could shift the probability of Brexit by 1%, it would be a smart thing to do on the part of HSBC's management.

This is a very interesting illustration of the Tullock Paradox and makes the £5m budget of the official Remain campaign look quite low.

This has been quite a long post but here are the two main takeaways:

  •   We should be very careful to check the calibration of our models, especially ad-hoc ones. Reality, surprisingly, actually has some quite interesting theoretical implications.
  • The Treasury analysis perhaps exaggerates the short-term costs of Brexit, but it isn't wildly implausible.

Tuesday, 31 May 2016

Yes, The World Scales

I think it's a bit absurd to suggest, as Conor Sen does, that economies of scale are falling in the modern economy.

(Sidebar: this post is unashamedly a reaction to his – but it's too big and full of interesting ideas to quote properly, so you should go and read it! Let's bring back the golden age of back-and-forth econ blogging.)

On-demand startups, the Ubers and AirBnBs and Deliveroos of the world, can be viewed as a setback for scale. But the alternative view is that, as returns to scale rise, attempts to bring some of the benefits and efficiencies of scale to smaller businesses become more worthwhile.

Maybe people are driving Teslas instead of Model Ts. But that's because, in a world where scale is driving down the cost of standardised products, it's uniqueness and quality that stand out. Everyone and their mother has a cheap, high-quality Mercedes – that's why you need to spend your excess wealth on a gleaming electric status signal.

Sen cites Uber's regulatory problems as an unexpected problem caused by their scale. But it's not just Uber's core technology and business model but its legal strategy that gets advantages from scale. Part of the reason Uber has been expanding so aggressively is that, without scale, it wouldn't have the clout to fight regulators. The incentive for regulatory arbitrage in the ludicrously restrictive taxi industry has always been there. It's only with modern technology that an arbitrage operation like Uber has been able to justify the sort of scale needed to fight the regulators and win.

It's also important to distinguish actual observed scale from the costs of scale. Sure, the most-viewed TV episodes ever (at least for the US) are M*A*S*H, Cheers and The Fugitive. You have to go down the list to #11 before you get to a finale that aired this millenium - in this case, the bleeding edge of modern culture that is Frasier. But that owes less to the benefits of scale and more to the fact that they were the only game in town - M*A*S*H is on top for the same reason that there will never be another Bradman. But if you're trying to scale a TV series today, peer-to-peer networks will get it into the hands of millions of people for free - which is why content producers are now having a harder time getting people not to access their content than the reverse.

Sure, Chipotle isn't as big as McDonalds and Whole Foods will never be Walmart. But Buzzfeed is bigger than the New York Times. Facebook is busy monetising 1.6 billion pairs of eyeballs. (That's almost 3.2 billion individual eyes!)

The future may not deliver economies of scale large enough to justify 1999's stock market evaluations. But no one person has ever been able to impact the lives of so many people as the engineer who tweaks Facebook's code.

EDIT: Well, maybe this isn't such a real-time debate, since it took three weeks for Tyler Cowen to notice Conor's post, and another day for me to notice his. Costs and benefits of relying on mavens, I suppose.