How to predict economic growth
Forget fancy regressions on past growth numbers. They don’t work. You need to look at the real factors that let countries create new wealth
What makes one nation grow rapidly and another sink into economic stagnation? Ask any economist and they’ll probably rattle off a list of factors. Does a country have effective governance and property rights? What about its labor force and level of education? How about the nation’s outstanding debt? Is this likely to be a drag on future investment? It seems sensible that all these factors should matter, yet economists’ predictions of growth have never been too successful.
Another approach is to carry out sophisticated statistical analyses of past records of growth, and then see if these project a continuation into the future. This doesn’t work so well either, and seems to be more or less akin to reading tea leaves. As economists Lant Pritchett and Larry Summers argued two years ago, what the data shows is that periods of really impressive growth don’t generally predict anything but future reversion back to the global average growth rate. As they noted,
… the single most robust empirical finding about economic growth is low persistence of growth rates. Extrapolation of current growth rates into the future is at odds with all empirical evidence about the strength of regression to the mean in growth rates. Second, especially outside of the OECD countries, the process of economic growth is not at all well characterized by business cycle variations around a (relatively) stable mean. Developing country growth rates are strongly episodic and large (and apparently discrete) shifts in medium-term growth rates of 4 ppa or more common—and particularly prominent are large slowdowns. Episodes of super-rapid growth tend to be of short duration and end in decelerations back to the world average growth rate.
Applying this to the recent super performers, China and India, Pritchett and Summers suggested that we should expect their growth to wane. They’ll simply fall back into the realm of the ordinary.
I expect this is all justified by the numbers on past growth, taken on their own. Another way to put it is that time-series on growth year by year just don’t seem to contain information that is useful in making future predictions—except in the negative sense of saying that trends don’t tend to predict anything. Of course, this leaves aside the question of whether other kinds of information might be useful in making growth predictions, perhaps in combination with data on past growth. Perhaps there are hidden measures of economic health and growth potential which economists just haven’t yet identified? Surely something makes one nation grow faster than another. Why can’t we identify and measure that thing?
Well, according to some recent research of a rather unconventional kind, these measures do exist, and, if used appropriately, can be used to make better predictions of future economic growth. The key, it turns out, is Big Data—and a desire to drill down through it to identify or at least estimate just how diverse and sophisticated is the set of production skills that any nation possesses. Faster growth, it’s not crazy to think, follows on from improvements in productive skills. This new work shows how to detect recent changes in such skills — doing so means handling data on millions of products of varied kinds — and how to turn this knowledge into better predictions.
OK, a little detail and background. It’s natural to think that wealthier nations, in comparison with poorer ones, should be economically more complex, and more able to produce a vast array of different products. This intuition turns out to be right, although this was only discovered five years ago, when Harvard physicist Cesar Hidalgo and economist Ricardo Hausmann noted a strong correlation between national wealth, as reflected in GDP, and measures of the variety of products nations produce. They also found that wealthier nations not only produce a wider range of products than poorer nations, they also produce more highly specialized products — think cars powered by electricity, drones or advanced rocket boosters — which can be made only by a small number of other nations.
So GDP and diversification of productive skills go together. It’s only a short further step to wonder if near-future GDP growth might be predicted by looking at what countries have recently done in developing their productive capacities. Growth ought to be more likely in nations which have recently expanded their repertoire of export products, but which haven’t yet reaped the economic benefits.
The new work shows that this is true, although it takes some tricky analysis to show it. Physicist Matthieu Cristelli and colleagues used global data on exports over the past 15 years to estimate the economic “fitness” of various nations — this being a measure of the range of products they currently produce, as well as the sophistication of those products. For each year, plotting this number against current GDP, the data falls roughly around a line sloping upward. This is essentially the same thing Hidalgo and Hausmann found — that productive breadth and GDP go together (although Cristelli et al. use a slightly different measure for their “fitness” than the earlier work did). Some nations fall slightly above the line, with more GDP than you would expect for their capabilities, while other nations fall slightly below the line. From year to year, as economic output fluctuates, nations move around in this plane.
Cristelli and colleagues then looked to see if there were predictable patterns of movement. There are — but they aren’t quite as simple as some regression back to the mean. To find the patterns, they had to use some methods drawn from the analysis of weather dynamics. Technical details aside — see this short article in Nature, if you like— the figure below shows the core of what they found. What you see are arrows showing the directions of past economic evolution for nations, over the past 15 years, located at various places in the fitness-GDP plane. The analysis shows that there are both predictable and unpredictable regions—or “laminar” and “chaotic” in the terminology of fluid dynamics, which Cristelli and colleagues employ. Specifically, growth appears to be much more predictable for nations of intermediate to high fitness, yet very irregular and chaotic for nations of low fitness.
The authors cannot really say precisely why some zones show regular and predictable patterns, while others do not. This just comes out of the mathematics. But they speculate that, since the predictable regime exists for intermediate and higher levels of fitness, it may be that economic fitness only becomes able to dominate future growth once it becomes sufficiently developed. For nations of low fitness, differences in the diversity of productive skills may be insufficient to drive measurable growth differences, as they are swamped by other more powerful factors — civil wars, natural disasters, poor governance and so on. Only when fitness becomes sufficiently large, do the productive forces come to dominate future growth.
So what predictions come out of this analysis? First, that in contrast to prevailing doubt, China and India will probably keep growing steadily and reach a combined GDP of around 26 trillion dollars by 2022. That’s roughly a tripling of their combined GDP in the next 7 years. The analysis also suggests that a handful of other lower-fitness nations will finally escape the trap of chronic poverty. These include Senegal, Madagascar, Tanzania, Kenya and Uganda. What sets these nations apart, this analysis suggests, is their storing up, over the past fifteen years, of a wealth of new capabilities enabling them to make more complex and valuable products and to thereby enter new markets.
Considering all countries together, the study suggests something else— that future growth prospects really do appear to be shifting from Western developed nations to countries to Asia.
More important than this specific prediction, of course, is this idea as a new way to predict future growth. It suggests that finding predictable patterns means going beyond regressions on aggregate data — it means wading into the messy details of actual skills and capabilities. And today that means using Big Data. The analyses involved in this work were highly computational, and based on building and analyzing huge mathematical networks of nations linked together with the vast range of products they are capable of producing. It’s rather like judging the capabilities of a cell by looking in detail at its entire metabolic system and listing out all the biochemical reactions it is capable of carrying out, and also ranking them by their level of complexity.
This may well be the future of studies on economic growth. If so, it will be a lot more interesting than the past, mired as it has been in uninformative statistical regressions.
