Estimating Manufacturing Activity via Machine Learning Analysis of High-frequency Electricity Demand Patterns

Abstract

In order to forecast the economic trend, it important to ascertain what is actually going on in the economy in a timely manner. In this research we measure production activity on the basis of the data of electricity used in manufacturing industry production processes. Major Japanese power companies publish actual electricity consumption data for every hour or every five-minute period. In this research, we set out a method of assessing economic activity in real time by focusing on this kind of high-frequency electricity demand data. Concretely, we estimate factors which means the pattern of the electric demand based on principal component analysis (PCA) for the electricity demand data, and build the regularized regression models in order to estimate the economic activity by using the PCA factors. In Japan, the official statistics on the production activities of the manufacturing industry is Industrial Production Index released by the Ministry of Economy, Trade and Industry. Based on the proposed method, it is possible to estimate the manufacturing activity about one month earlier than the publication day of the official statistic. Furthermore, we confirmed that the estimation of the Industrial Production Index based on our method can achieve higher forecast accuracy than the market forecast average.