Physicists have a reputation for being a bunch of gooey beaks — they’ll jump into unrelated fields and tell everyone they’re doing it wrong. This reputation is so deserved that there is even a relevant XKCD. Sometimes, though, it all works out — usually because the physicists stick to their area of expertise, which just happens to be relevant to the problem.
In this case we are talking about economics. Coincidentally, economics falls into the category of a complex system, with which various physicists play a lot.
The paper in question seems to bring together some slowly evolving concepts in economics. All things considered, and with a dash of dynamic modelling, their merger leads to better forecasts for gross domestic product (GDP) – and I expect other economic indicators could be attacked by a similar procedure.
Complexity is complex
Now, before you go “well, of course the economy is complex,” the word complex has a specific meaning in this case. Complex systems are not necessarily that complex: certain waterwheels, for example, are complex.
Maybe it’s easier to think about it that way. Imagine you are on mountainous terrain. Your movement through the terrain depends on your current direction of movement, the slope of the terrain and your current position. If I could look at this from above, I could measure it and predict your route through the terrain.
If everything was linear and not complicated, then it doesn’t really matter if I’m inaccurate in my measurement. My predictions will never be quite correct, but they will be close enough for me to find you. In a complex system, a small initial error will amplify, and soon I’ll be looking for you on a hilltop as you trudge through a stream two valleys down. A whole host of techniques have been developed to allow us to extract some predictability from these types of systems.
Applying physics to economics
However, the economy is a bit different from many physical models: there is no complete model of an economy. There is not even a good approximation model. After all, simple models that provide insight do not provide predictions. Instead, predictive models are statistical in nature. These use historical economic data to predict future economic data – essentially the model looks for correlations between historical and recent data. These correlations are then used to take current economic data to predict future economic data. The model is then constructed based on our understanding of how the measured data relates to economic activity.
The problem with this approach is that if you don’t have enough data, predictions quickly become inaccurate. To resolve this issue, we are collecting more information. With that information, new processes can be incorporated into the model in the hope that this will yield greater accuracy over longer time intervals. And this certainly works: current models are better than older models.
To improve predictive models about the economy, researchers have taken a counterintuitive approach. They reduced the number of parameters in their model to just two: economic fitness and gross domestic product, with the idea that if economic fitness and GDP are measured at a given point in time, the change in GDP can be predicted.
So what is the economic suitability? In short, it is a measure of the complexity of a country’s exports. The idea is that exports represent the products from a country that compete with similar products from the rest of the world. The greater the variety of products exported, the fitter an economy is. An advantage of it as a measure is that exports and imports are measured very accurately, because companies rely on that data to survive. And that data is collected and reported in a relatively standardized manner. In fact, the researchers created a matrix with which to sum the variety of exports.
This number is then iteratively normalized with data from all other countries to arrive at a self-consistent scale of economic fitness. Economic fitness drives changes in economic growth, which is accounted for in GDP.
Now it’s important to realize that no one really has a model (in a physical sense) of the link between economic fitness and GDP. But we do have statistical data that can be used to infer how the two are related. We can estimate from the averages in the data set how high the economic fitness of an economy must be to support a given GDP and use that to determine whether GDP will rise or fall.
The rate of increase or decrease is estimated using some kind of force-response model. In other words, if GDP is far from the expected GDP of the current economic condition, there is a strong hidden economic drive to change GDP. Therefore, we can expect rapid economic growth (or contraction).
predicting the past
This case is illustrated by China in 1995. At that time, China had a low GDP but a high economic condition. As predicted by the model, China experienced 20 years of strong economic growth, with GDP increasing remarkably. In a standard economic analysis, this seems extraordinary. But the researchers argue that this is actually expected behavior: much like a stretched spring released to spring back into position.
The model also lets economic momentum play a role. The rate at which economic fitness changes also affects GDP change. The researchers found that using the economic fitness trajectory to predict GDP leads to even more accurate results.
The researchers tested their model on historical data from 169 countries over three different five-year periods. They compared their GDP forecasts with those of the International Monetary Fund (IMF) model and with actual GDP data.
They found that their model was better than the IMF model, especially when they also took into account the trajectory of economic fitness. Furthermore, a close analysis of how the IMF model and their dynamic model predictions differed revealed that the sources of inaccuracy were different. That meant that combining the two models led to predictions that were even more accurate.
Another important factor is that there is a kind of self-similar behavior in trajectories. While the overall size of the economy may be different, countries with similar ratios (I’m simplifying it here) of economic fitness to GDP have similar trajectories. And one last point: the model also shows where predictability fails. Countries with very low economic fitness are incredibly difficult to predict. This applies to both the IMF model and their model, but it emphasizes that the poorer you are, the more subject you are to the random rumble of economic noise.
I would like to draw a succinct conclusion here, but I am not an economist, nor do I like economic forecasting. The model looks interesting, but I have no idea if it will be used or not. I don’t even know if GDP is the right thing to measure. But at least another tool is available.
Nature physics2018, DOI: 10.1038/s41567-018-0204-y