The role of economic theory

I'm nearing the end of three conferences in various parts of Europe (it's a hard life being an academic at times), and the mix of papers at these conferences has meant that I've listened to a lot of papers that are purely theoretic, and a lot of papers that are purely empirical, and some others that are somewhere in between.

Being an applied person myself, my initial reaction to purely theoretical papers is: what's the point? For a paper to be of any use, my first thought goes, it must be empirically validated.

Of course, the problem here is, can a theory be empirically validated? And the answer is, a lot cannot. I listened to a paper this morning by Gerhard Illing on 'dancing banks', asking about the effect of bailing banks out that are failing in times of financial turmoil. This kind of people, theoretical in nature, cannot really be empirically validated, unless it so happened that some country already operated the kind of regime that Illing proposes, and another didn't.

Perhaps he could get hold of data for different countries, and try to measure how likely each country was to bail out its failing banks, and then get data on the amount of liquidity at various points in inter-bank financial markets, and try to come to some conclusions. However, measuring how likely a country is to bail out its banks is very hard indeed: it doesn't happen too often, and the actions of countries usually differs from words, as no government will say it's prepared to bail out failing banks, as this would encourage reckless behaviour.

So, there's clearly a number of papers for which empirical validation is impossible, particularly papers proposing reforms in governance, and other prescriptive papers.

This doesn't excuse papers like this one on the effect of financial markets being imperfect. These Dynamic Stochastic General Equilibrium (DSGE) models make great predictions about the economy, about policy, and wider, yet rely on very flimsy relevance for the real world.

Such models claim to have "structure", and this is their great claim in the light of the Lucas critique of 1976, which pointed out that simple regression models may be undone in structural change because underlying parameters, such as those of personal preferences of individuals, change.

But what if these models have the wrong structure? Then they are no better than what came before, and no more illuminating than a simple regression model. And do they? For one thing, they assume a representative agent: i.e. everyone is the same and has the same preferences.

Yet in microeconomics, regression models are perfectly acceptable if they claim to explain less than 30% of the observed variance in a given data series; the reason? Unobserved heterogeneity between individuals. So if so much heterogeneity exists, precisely in the kinds of studies that the above DSGE models use to choose the important parameters for their models, how can these same DSGE models assume a representative agent?

The simple fact is, these models make far too great claims on far too important things given their total lack of empirical validation. Theory is by no means useless, it is vitally important, but when empirical validation is possible, it should be carried out.