Saturday, February 27, 2016

Why economic modelling results are dodgy

Modelling is hugely fashionable in Canberra (joke intended). Any lobby group seeking to persuade the government to do something – or not do something – produces allegedly independent modelling that supposedly backs up their case.

Government isn't above using modelling to make its case, either. Had Malcolm Turnbull decided to go ahead with raising the goods and services tax, he would have produced modelling to show how wonderful that would be.

When he decided not to increase the GST, he naturally produced modelling to show that the gains to the economy would have been minor.

Have you detected my note of scepticism? The mathematical models of the economy that economists produce are supposed to be an aid to thinking. In the public debate, however, they're used as a substitute for thinking.

When it suited the government to make public a version of Treasury's modelling of a "tax mix switch" – raising the GST and using the proceeds to cut income tax – it came with a product warning: "This modelling is indicative at best and care should be taken with its use".

That was inserted to protect the modellers' professional reputation from criticism by people like me.

It's a gross understatement. What it should say is: we've been asked to answer a question our model is incapable of answering with any reliability but, because we enjoy modelling – and are quite well paid to do it – we've fudged up something that looks like an answer.

People are impressed by economic modelling because it's done on a computer. It has to be because it involves solving so many equations and so many calculations.

So, at one level, modelling is highly sophisticated. At another level, however – and this is the bit the punters never get – it's amazingly primitive. Why? Because the economy is so hugely complicated that no simplified model of it is capable of doing justice to its many possible reactions to some development.

Be clear on this. I was pleased when Turnbull decided not to increase the GST. It wouldn't have been a good change. And I happen to agree that a tax mix switch reliant on the GST would do little to increase economic efficiency or foster "growth and jobs".

So I'm not attacking Treasury's modelling for the usual reason people do: because its findings don't accord with their prejudices. No, what I object to is the way economists (who just want to be paid to keep playing with their models) collude with politicians and vested interests in using modelling to hoodwink the public.

Try to con my readers and I'm like the mother bear in The Revenant.

What if I told you that rather than a tax with an $18,200 annual tax-free threshold and rates of tax rising from 21 per cent to 47 per cent as incomes rise, the personal income tax modelled by Treasury was a flat-rate tax of 16.7 per cent, applying from the first dollar of income?

It's true. You can read all about it in a Treasury working paper published last April.

But we've never had such a tax and are never likely to. It's radically different to the progressive tax we do have. So why on earth would you use it for the modelling?

Because it's the best the model can do. It's simply not capable of modelling our real-world income tax. Why not? Because the model has just one household in it - the "representative" household.

But it does have 111 businesses – one for each of the 111 industry sectors it identifies. To this it adds one government (not eight) and one "foreign sector".

The basic model Treasury – and many other modellers – uses is called a "static, representative household, computable general equilibrium (CGE) model".

The CGE bit means the model attempts to cover the whole economy, not just one bit of it. This allows it to capture some of "the main second-round effects of taxes on households, firms and investors". It's also able to capture interactions between different taxes.

But that advantage comes with the huge disadvantage of the single, representative household. We know Australia's 9.2 million households differ greatly in many respects – size, age, income, spending patterns, saving rates, ability and desire to work.

But for "analytical tractability" (to stop the model becoming impossible to understand, even by the modeller) all these dimensions are reduced to averages. And, as we know, the statistical average household, being a mixture of everything, is often quite un-representative.

Apart from being unable to cope with a progressive income tax, the model is also unable to cope with means-tested welfare benefits. It was able to model only a fraction of the compensation that would have had to accompany an increase in the GST (the huge cost of which is the real reason Turnbull abandoned the idea, not the happy modelling about benefits to the economy).

The model assumes that an increase in the rate of tax discourages people from doing as many hours of paid work. But it uses a single, average "elasticity of labour supply" (the degree of workers' responsiveness to changes in their after-tax wage) for the representative household.

Really? Professor Patricia Apps, of Sydney University, points to the empirical evidence showing that "primary" earners (mainly husbands working full time) have quite low elasticity – have you tried telling your boss that from now on you'll be working only 37 hours a week, not 38? – whereas "secondary" earners (mainly mothers working part time) have quite high elasticity.

Averaging the two together makes the exercise meaningless.

The modellers concede that "general equilibrium models are necessarily a simplification of the economy and, as such, they can only incorporate a stylised representation of the tax system".

I'd say models are a cartoon caricature of the economy, quite incapable of answering the intricate questions we ask of them.