9. Post-orthodox econometrics

Presenting a paper to any conference with the name ‘Keynes’ in the title, particularly to one sponsored by the Post Keynesian Study Group, is fraught with danger. For the exegitical debate about ‘what Keynes really meant’ is often carried out with as much fervour as the arguments between the Fathers of the early Church. Indeed, some of the latter material has a familiar sound. Origen, for example, proposed a theory of Redemption in which the soul fluctuated between Heaven and Hell, a suggestion sternly denounced as heresy by Augustine, who specifically denied the existence of endogenous cycles. It is as well to make my own position clear from the outset. Keynes is obviously a very important figure in economics. But it is now sixty years since the General Theory was written, and the world has moved on. Particularly outside the discipline of economics, scientific methodology has changed. We should be concerned to use advances in knowledge to increase our understanding of the dynamics of capitalism, regardless of whether Keynes – or Smith or Ricardo or Marx, for that matter – really meant the same thing. So I do not intend to enter into definitional arguments about Keynesianism. The aim instead is to review how we should approach applied macroeconomic analysis. Section II of the chapter considers briefly the post-war research programme in conventional macroeconomic modelling. Both in terms of forecasting accuracy and of structural understanding of the economy, very little progress has been made. Section III points out that the failures of forecasting are inherent in the nature of macroeconomic time series. It is simply not possible with most macro data series to generate forecasts over time which have an error variance significantly less than the variance of the data themselves. An important implication of this result is that conventional short-run stabilization policies cannot work in any meaningful sense, even if the structure of the economy were well understood. In Section IV, I argue that most existing applied macroeconomic relationships are overfitted. In local time neighbourhoods very simple linear models offer a good explanation of key