Adaptive Trees: A Novel Approach to Macroeconomic Forecasting – Economic and Policy Implications of AI, Blog Post #5

Adaptive Trees: A Novel Approach to Macroeconomic Forecasting – Economic and Policy Implications of AI, Blog Post #5

The following is a summary of Adaptive Trees: A Novel Approach to Macroeconomic Forecasting by Nicolas Woloszko. This paper was presented at the Technology Policy Institute Conference on The Economics and Policy Implications of Artificial Intelligence, February 22, 2018.

Machine learning is all about making better predictions. In Adaptive Trees, Nicolas Woloszko explores how we might improve our national accounts data by applying machine learning techniques to macroeconomic forecasting.

Woloszko notes that the econometric techniques that are the basis of most forecasting models focus on in-sample accuracy, fitting a model to observed data to explain them after the fact. In contrast, machine learning focuses on out-of-sample accuracy, using observed data to predict what will come next. The modern economy is constantly evolving and changing, as is the context in which it operates. Machine learning can adapt to these changes more quickly than econometric models can, potentially yielding more predictive power.

Woloszko introduces a machine learning algorithm based on an adaptive trees approach that is intended to complement existing economic forecasting techniques, particularly in situations characterized by uncertainty or immediately preceding or following economic shocks. For , econometric forecasts for housing prices in 2006, 2007, and 2008 using observed data would have looked similar, discounting a significant structural change in the housing market in 2008. The model that he proposes, which can incorporate discontinuities like the sudden decline in housing prices in 2008 and 2009 is particularly useful around economic turning points and in predicting recessions.

Woloszko tests his theory by comparing economic forecasts generated using the Adaptive Trees algorithm to forecasts generated using traditional economic forecasting models in G7 countries three, six, nine, and twelve months ahead. Across all time horizons, the Adaptive Trees forecasts were as accurate or more accurate than predictions by traditional indicator models. Importantly, the Adaptive Trees forecasts also produced more accurate forecasts around recessions. In discussing the implications of his findings, Woloszko notes that Adaptive Trees algorithms are complements, rather than substitutes, for traditional forecasting methods. Using both will provide a more complete economic forecast than using either alone.

Share This Article

View More Publications by Wallis G. Romzek

Recommended Reads

Related Articles

Sign Up for Updates

This field is for validation purposes and should be left unchanged.