Sunday, 24 September 2017

Non-linearity in economics

Gertjan Vlieghe is characterised as one of the more dovish members of the Bank of England MPC, so his recent speech in which he suggested that “the evolution of the data is increasingly suggesting that we are approaching the moment when Bank Rate may need to rise” was indeed noteworthy. It is thus a pity that the rest of the speech was overlooked for it was a fine exposition of the factors driving real interest rates. But it was his dismissal of the “fairly deeply rooted, but wrong, notion in modern macroeconomics, namely that real interest rates are primarily driven by the growth rate of the economy” that really got me thinking.

Vlieghe pointed out that “the idea persists, because of commonly adopted – but misleading – practices in solving macro models.” Modern macroeconomics is based on highly non-linear models but in order to make them more tractable for solution purposes, we use logarithmic transformations to linearise them. Vlieghe uses the example of how the linear transformation of the standard method of discounting future consumer utility results in “a tight relationship between the real interest rates and growth, and nothing else ... it kills off, mechanically, anything that might have been interesting about risk … In the linearised world, there is no risk-free real interest rate.” For those interested in the detail, the relevant part of Vlieghe’s speech is reproduced in a footnote[1].

As it happens, this is perhaps an overly-rigorous theoretical interpretation of the relationship between growth and interest rates. Admittedly, the fact that there is a strong correlation between (real) short-term interest rates and (real) GDP growth does not necessarily imply a causal relationship. But the work of the late-nineteenth century economist Knut Wicksell postulated that there is a ‘natural’ interest rate which is determined by the real disturbances affecting the economy. To the extent that these disturbances are manifest in output growth and inflation, it is clear that nominal GDP growth and interest rates ought to be closely related. It is not a 1:1 relationship, but over the long-run the rate of return on real assets ought to be similar to that on financial assets in order to satisfy equilibrium conditions. Not for nothing have many economists argued that there is a strong case for using nominal GDP growth as an anchor for monetary policy.

That said, Vlieghe’s point on how linear approximations can result in specious outcomes was well made. All students of econometrics spend a lot of time learning about the properties of linear regression models which partly explains why grubby practitioners like me are comfortable applying linear transformations in order to easily apply linear estimation techniques. Another reason for preferring linearity in econometrics is that non-linear solutions can be indeterminate because we do not know whether they are the universally right answer, or whether they apply only under certain conditions. The reason for this is that many of the common solution techniques rely on grid searches conducted over a range of values. In the jargon, we do not know whether we have found local maxima only within the range in which the search is conducted or whether the “true” answer lies outside it. Our models may thus be biased – in other words, deliver the wrong answers under certain conditions – and as a result many economists stick to what they know in the form of linearity.

But this bias towards linearity, tempting though it is, can be applied to situations in which it is not appropriate. The authors of a paper in experimental psychology[2] assessed the accuracy of long-term growth estimates by panels of “experts” and laypeople. Whilst both groups tended to underestimate growth at rates above 1%, the degree of underestimation was greater for “experts” because they ignored exponential effects more often than the group of laypeople. Another paper by DeBock et al (2013) is interesting because it provides a literature review of the reliance on linearity and reports the findings of an experiment amongst business economics students who were confronted with correct and incorrect statements on linearity in economic situations. The authors concluded that many of the students showed over-reliance on linearity in their analysis.

An interesting paper on nonlinearity by Doyne Farmer (here) looks at various aspects of nonlinearity and complexity in economics. He makes the point that DSGE models are too highly stylised to say anything useful about the behaviour of economies in the real world. Instead, economics might start to take lessons from areas such as meteorology which builds very complex data-based nonlinear models. This has been enabled by the significant increase in computing power which allows simulation analysis to be conducted much more cheaply and effectively than in the past.

For policymakers, the fear is that linear approximations in a nonlinear world lead to distorted policy conclusions. One problem is that economics tends to focus on equilibrium solutions. But as noted above, there may not be a single equilibrium. Indeed, the impact of the financial crash of 2008 was an object lesson in how nonlinear feedbacks can produce outcomes far beyond our expectations.

The mathematician Stan Ulam used to give lectures on nonlinear mathematics and apologise that the title was a misnomer, for all interesting maths involves nonlinearity. As Farmer put it, “just as almost all mathematics is nonlinear, almost all economic phenomena are complex … A more tractable topic would be whether there are any problems it does not illuminate, or should not illuminate.”



[2] Christandl, F., and D. Fetchenhauer (2009) ‘How laypeople and experts misperceive the effect of economic growth’, Journal of Economic Psychology (30) pp 381–92

No comments:

Post a Comment