While it’s tempting to dismiss big data as an over-hyped buzzword, a number of projects have already shown its potential. The past year or so have seen a range of fascinating and diverse projects emerge that utilize big data to predict the future.
• A predictive model by UCLA researchers to predict where crimes will be committed (and therefore where police should be placed)
• A neural network that can predict the price of oil developed by a team from the Middle East
• A platform to use predictive analytics to provide businesses with what it calls ‘event intelligence,’ which is basically insight into how upcoming events might influence them
• An algorithm that aims to predict the probability of success for a startup
• A horizon scanning platform to spot trends in the academic literature
Those projects are undoubtedly impressive indicators of the potential, but none have taken on a topic as vast and complex as predicting GDP growth. That was the challenge undertaken by Harry Raymond Joseph, a reputed quantitative researcher at an investment bank in London, in a recent paper. In it, Harry describes a small sub-set of his methods that use machine learning and data algorithms to provide more accurate GDP growth forecasting using just import seaport data.
Predicting GDP Growth
The study, which has been supported by the Department of Foreign Affairs and International Trade in Canada, is believed to be the first to solely use high granularity import data as a means of forecasting GDP, with the algorithm successfully forecasting GDP changes in both India and the U.K.
It’s also the first successful study to use deep learning to underpin GDP forecasting. The deep learning models that form the central algorithm allow for a range of macroeconomic factors to be analyzed and therefore provide a rounded approach to GDP forecasting.
By using seaport import data, Harry believes the approach is extremely scalable, as the data is readily available in most countries. Not only is the data available in high levels of granularity, in some cases availability of seaport import data is ensured by national and international agreements, thus making it a good source of data for the algorithm.
We’ve living in an era where the rejection of expert opinions is common, no doubt supported by the challenges faced when trying to make accurate forecasts. It brings to mind the famous Deming quote, that “in God we trust, everyone else must bring data.”
This, and the examples given at the start of the article, highlight the potential to make decisions based upon the best available data, and therefore the best available evidence. Harry’s model has already delivered accurate forecasts of GDP growth in both India and the U.K., in both cases surpassing the accuracy of analysts.
Interestingly, the model was especially accurate in economies with a high proportion of sea bound trade. For instance, GDP forecasts in Japan, Australia and the U.K. were made within 0.05% of actual GDP growth. This compares to forecasts for the U.S., France, India and Brazil that were within 0.15% of actual GDP growth. It’s a distinction that Harry believes is due to the differing proportion of sea trade in these economies.
Managing Macroeconomic Uncertainty
Harry’s method is particularly interesting as it applies very successfully to the kind of uncertain macroeconomic situations that are increasingly common. For instance, in India, the recent demonetization program implemented by the Indian government prompted many analysts to issue bearish growth forecasts for the last quarter of 2016. The Brexit vote prompted similarly pessimistic forecasts by economists. However, in both instances Harry’s algorithm beat analyst forecasts and came closer to the actual GDP.
The model is currently being refined, with a number of interested parties looking to support its development. There is certainly no shortage of demand for better tools for economic forecasting, especially with the number of extreme political events seeming to be on the rise. Such models have significant potential therefore, not just for banks and hedge funds, but also within think tanks and policy making circles.
“Whilst some GDP forecasters consider import data, it is often given low weighting in the overall forecasting mix. Deep learning algorithms are effective when input data is most precise, and so import data is ideal, as high levels of precision can be guaranteed. I hope this helps governments and policy think-tanks in effective macroeconomic policy-making,” Harry said.