Saturday, February 4, 2012

Why economic modelling is so tricky

When I was studying economics at university in the mid-1960s, I wasn't quite sure what a "model" was. These days, politicians and business people are always using the word, and most of us think we know what they mean.

In case you're not sure, here's an explanation I would love to have seen at uni, provided by Dr Richard Denniss, executive director of the Australia Institute, in his new paper, The Use and Abuse of Economic Modelling.

"A model, be it a model car or an economic model, is a simplified representation of a more complex mechanism. A model is typically smaller, simpler and easier to build than a full-scale replica. A model sheds light on the main features of the reality it seeks to represent," Denniss writes.

"An economic 'model' is not a physical thing, like a model car. Rather, it is a mathematical representation of the linkages between selected elements of the economy."

Thus an economic model is, unavoidably, a simplified version of the economy, or an aspect of the economy. It includes those aspects of reality the model-builder regards as most important in explaining what happens, and leaves out all those aspects that don't seem to make a big difference.

So the results you get from a model are only as good as the modeller's choice of what to include and what to leave out. In practice, the model's predictions will often prove astray because some factor the modeller assumed wouldn't be important turned out to be.

Different types of economic models are used for different purposes. Earlier this week I used Denniss's paper to discuss the input-output model that industry lobbies use to make their industry sound bigger than it is.

Today let's discuss the most sophisticated models economists have developed; "computable general equilibrium" models. These are often used to shed light on the effect of a major policy change on the economy over the next 10 or 20 years.

Economists often analyse only a part of the economy, called a "partial equilibrium analysis", while assuming ceteris paribus (all other things remain equal) in the rest of the economy. This is unrealistic because, in the economy, everything is connected to everything else. Changes in one bit lead to changes in other bits, which then feed back on the first bit. So general equilibrium models attempt to capture all these interactions between different industries and markets. (In practice they can't capture all of them, so they still rely on the ceteris paribus assumption to cover those they leave out.)

If you remember nothing else about models, remember this: their weakness is they are built on a host of assumptions, and therefore are only as good as the assumptions on which they are built. Some assumptions are obviously unrealistic and couldn't possibly hold; some happen to be overtaken by events.

Models are sets of equations with dependent and independent variables. The modeller decides the values of the independent (or "exogenous") variables and the model calculates the values of the dependent ("endogenous") variables. So, if the modeller puts in the wrong independent variables (usually assumptions or guesses about the future), the dependent variables will be wrong, too.

The model-builder also specifies the "elasticity" (sensitivity) of the relationships between the variables. For instance, when the exchange rate rises will the reduction in exports be big or small? Elasticities are partly based on empirical evidence, but they also reflect the model-builder's beliefs about how the economy works. Should that belief be wrong, the model's results will be wrong.

It's common for general equilibrium models to be Keynesian in the short run (up to 10 years) but neoclassical in the long run (20 years or more). That is, key variables such as inflation, unemployment and economic growth are determined by the strength of aggregate (total) demand in the short run, but by the strength of aggregate supply in the long run.

This means the economy is assumed to be at full employment in the long run, and economic growth over the period is assumed to be determined solely by the growth in the labour force (the population of working age and its rate of participation in the labour force) plus the rate of improvement in the productivity of labour.

How do you know what the average rates of growth in population and productivity will be over the next 20 years? You take an educated guess, then plug them in. But here's the trick: once you've done that, you've predetermined where the model's results will end up, regardless of whatever policy changes you simulate happening to the economy in the meantime.

No matter how much some change knocks the economy off its assumed long-run course, the model's specifications assume it will not only get back on course but also catch up to where it would have been.

The economy can take up to 10 years to return to its "steady state".

So, the bigger the initial departure from the long-run trend, the bigger the ultimate bounce back - by design. And, by design, nothing can ever happen that changes our destiny.

Thus the model assumes away "path dependency" - the idea that where we end up is determined by what happens to us on the way; that some developments leave us permanently better off, while some leave us permanently worse off. This is clearly unrealistic.

But see what it means? It means the policy change you're purporting to be testing doesn't stand a chance of making much lasting difference, for good or ill. And that means your test is a sham. You give the appearance of testing some proposition, but the outcome is essentially predetermined.

I think such models should be used only in private by consenting economists. They have a good understanding of the assumptions on which the model's results are built and they know whether they share the modeller's faith that the economy works the way her model assumes it does.

When the results of these models are paraded before the public - by governments and treasuries, as well as interest groups - they can't help but mislead. They appear to be proving some policy change would be good or bad but, in truth, they're coming to predetermined conclusions.

The sign that the sponsors and modellers are out to mislead is shown by their failure to highlight their model's key assumptions in some sort of comprehensible product disclosure statement.