Forecasting pine sawtimber stumpage prices: A comparison between a time series hybrid model and an artificial neural network

Abstract

We conducted a comparative analysis of the predictive ability of classical econometric models and artificial neural networks (ANNs) for pine sawtimber stumpage prices across 22 TimberMart-South regions in the US using quarterly prices from 1976 to 2022. We evaluated model accuracy via root mean square error and mean absolute percentage error metrics and employed the modified Diebold-Mariano test to determine if there was a significant difference in forecast accuracy between the two models. Our findings demonstrate that ANNs outperform classical models in predicting turning points, whereas classical models tend to smooth price trends and produce forecasts that are biased toward the average value. This study provides a basis for predicting timber prices in the southern timber market using ANN models and contributes to ongoing discussions on the effectiveness of machine learning algorithms in generating precise forecasts within the forest industry. The findings can help timberland investors to make informed business decisions in the timber market.