A new method for forecasting gross domestic product (GDP) growth that complements and outperforms International Monetary Fund forecasts is reported in a paper published online this week in Nature Physics. The new forecasting approach treats economic growth as a physical system whose dynamics can be predicted by analysing product-level export data using techniques derived from the physics of complex systems.
Modelling systems as complicated as a country’s economy is extremely challenging. Although economists can access increasingly large amounts of data, producing reliable and reproducible outcomes from such remains far from trivial. Help might come from the physics of complexity, where techniques exist to model systems that cannot be understood in terms of their individual components - such as the spread of epidemics, or traffic movements.
Andrea Tacchella and colleagues develop a GDP forecasting method based on the idea that the behaviour of an unknown, complex system can be predicted by finding in available historic data a close analogue, and looking at its time evolution. Such methods only work reliably when using a low-dimensionality model - where the dimension is the number of variables inputted. Indiscriminate addition of data, therefore, produces less reliable predictions. Instead, the authors demonstrate efficient GDP prediction using a model with only two dimensions, inputting a country’s GDP per capita and ‘fitness’. The latter is a single quantity that captures a country’s competitiveness, one the authors construct from export data by extensive mathematical process.
By comparing their predictions to those forecasts based on past data published by the International Monetary Fund, the authors demonstrate higher accuracy in 25% of cases. Additionally, the errors of the models are uncorrelated, suggesting that the two provide complementary insights on GDP growth, offering the potential for improved, combined forecasts.
Medical research: Robot-assisted supermicrosurgery demonstrated in humansNature Communications