Forecasting models for Quebec’s lumber demand and exports using multivariate regression technique

Abstract

The business environment of the forest products industry is impacted by a variety of factors that makes it hard to predict the market’s behavior. Moreover, companies operating in this industry are continuously seeking to improve their understanding of the market by transforming available data into valuable knowledge and meaningful forecasts. This paper proposes a methodology to extract and use open data for Quebec’s lumber demand and exports forecasts using multivariate regression techniques. A number of methods were applied to estimate the models’ coefficients using a training data set, namely the Ordinary Least Squares method with a “backward” variable selection approach, LASSO and RIDGE regressions, and the Two-Step Least Squares method. Then their forecast accuracy was tested on an out-of-sample data set. The best selected models in terms of forecast accuracy succeeded in predicting Quebec lumber demand and exports on the testing data set, with a Root Mean Square Error of 0.12 and 0.08 respectively, and a Mean Absolute Error of 0.1 and 0.06 respectively. Furthermore, the developed data visualization tool appeared as a powerful tool to highlight the reliable forecasts generated by the models, while deducing relevant information through interactive graphics. Such a visualization tool could therefore help in better understanding the market when making decisions related to the evolution of lumber demand.