Using artificial intelligence in economic policy forecasting

Abstract

We hypothesize that machine learning algorithms are better equipped at forecasting policy rates. To test this hypothesis, we gathered several machine learning algorithms and compared their forecasts of the Norwegian policy rate against Norges Bank's own forecasts. The hypothesis builds upon the idea of machine learning as a general tool. Therefore, we tested a broad set of machine learning algorithms instead of developing a hyper speci c model. The machine learning algorithms we tested were the elastic net algorithm, the decision tree algorithm, the long short-term memory neural network, the convolutional neural network, and an ensemble learner. Consistent with our hypothesis, the algorithms did indeed exhibit lower prediction errors than the benchmark. A deeper analysis of the results indicated that this is due to their ability to better adjust to drastic changes in the economy and that Norges Bank's model performs better during stable economic periods.