An Industrial Application of Time Series Forecasting of Lumber Demand.

Abstract

Forecasting lumber demand is vital for operational purposes in the Distribution Centers of Home Improvement retail chains. This paper describes econometric time series analyses applied to specific lumber skus from the largest Home Improvement chain in the United States. We propose simple univariate smoothing models and examine the causal relationship between temperature, housing starts, price and lumber demand. We find that complicated ARIMA models are not necessary; simple smoothing models are more appropriate. The results indicate that monthly seasonal models fit better that weekly moving average models and that even though the Point-of-Sale time series and Housing Starts time series show similar trends, the relationship is not strong enough for housing starts to be used as a short-term predictor. Also, the local maxima of the Point-of-Sale time series trends in the Spring, Summer and Fall result in low correlations between that series and the average monthly temperature or price series. So, temperature and price cannot be used as short-term predictors either.