Estimating the impact of tax cuts is a tricky business. You can fairly easily calculate how the revenue from current income and spending will change, but that’s just the beginning. The problem is that people don’t stand still: they change their earning and spending habits in response to your tax changes, which changes the revenues from the taxes. The UK government is pretty good at estimating that but economists have long known that there are a couple more stages before you have a full picture of what’s going on. That’s why HM Treasury has begun to use a dynamic, computable, general-equilibrium (CGE) model to estimate the effect of tax changes.
CGE models bring us closer to reality…
The CGE model accounts for the long-term effect on the economy of changing behaviour. In the case of cuts in the fuel duty it accounts for the growth in production caused by a reduction in transport costs. Increasing production generates more road traffic, which yields more fuel duty revenues and partially offsets the cost of the cut. Using the CGE model to ‘dynamically score’ (as the jargon goes) the cost of the tax cut incorporates effects these effects that are not a part of the traditional approach.
…but it doesn’t account for everything
Even the CGE model doesn’t include all known effects: commentators have been quick to point out that externalities, such as pollution and congestion, are not included in the model. Nonetheless, it is better than the previous estimates and HM Treasury should be applauded for their efforts. The current, static scoring suffers from the same problems so dismissing the CGE estimates on those grounds would allow the good to be the enemy of the perfect.
John McDermott, in the FT, also criticises the model for ignoring the effects of monetary policy and the current slump, saying that “their absence means we should be sceptical about any attempt to simulate GDP two decades down the track”. This is a tricky question. The absence of money from a long-run model shouldn’t matter because, over a twenty year horizon, monetary policy rarely has much effect. However, a tax can have different effects in a slump than it would in a boom and CGE analysts commonly ignore these subtleties because it is difficult to know their magnitude. But these are minor quibbles since the current, static scoring method also suffers from the same problems.
The counterfactual is crucial
The most telling critique is made by Chris Giles and Simon Wren-Lewis, who claim that the Treasury has modelled the tax cut in manner designed to make it look good. The issue is that when you ask, ‘how much will it save?’ the answer is ‘compared to what?’ A reduction in the fuel duty could be matched by either a cut in expenditure or a rise in taxes elsewhere. The Treasury chose to compensate with extra taxes that change growth by the least amount possible. Chris and Simon’s point is that the overall impact would be far less rosy had the Treasury chosen to hike income taxes instead. That is true but, most likely, the Treasury analysts don’t know what the Government would do to compensate and tried to be as neutral as possible. Ideally, they would model various scenarios to give an idea of the possible range of impacts, but it is quite possible that they did not have the time to do that. Certainly, it would be good to see more scenarios in future work but it is telling that this question is never asked of static scoring. The reason is that static analysis tends to hide these assumptions through omission, rather than making them explicit as CGE analysis requires. The discipline of having to think about, and debate, these questions is a good one and certainly not a reason to favour the current, static techniques.
It’s great to see the Treasury conducting more sophisticated analyses of tax policy and doing so publicly. Publishing the results allows for these discussions and can only improve work they do in future. The analysis of fuel duties has obviously hit a political nerve and commentators have been quick to jump on the difficulties inherent in complex estimation tasks. In the hubbub we shouldn’t lose sight of the fact that the CGE model is still a great advance on what came before.