Corn Cash Price Forecasting
研究了七个州的玉米现货价格数据，评估了30个时间序列模型和10种基于修剪策略的组合预测，发现无截距的普通最小二乘法组合预测误差最小，且模型重校准频率不低于一个月。
We examine the forecasting problem in a data set of daily corn cash prices from seven states: Iowa, Illinois, Indiana, Ohio, Minnesota, Nebraska, and Kansas. We assess thirty individual time series models and ten combined forecasts based on six trimming strategies across three out‐of‐time evaluation periods, seven horizons, and two systems (bi‐ and multivariate). Using the unrestricted least squares without an intercept to estimate combination weights of individual models without trimming arrives at the lowest root mean squared errors across all evaluation dimensions. Incorporating local cash prices in a model could benefit accuracy, especially for relatively longer out‐of‐time evaluation periods and forecast horizons. Our results suggest model recalibration frequency no lower than one month. Discussions of empirical findings at a more granular level also are presented, including comparisons of individual time series models and those of combined forecasts based on different trimming strategies. The forecasting framework shown here is not difficult to implement and has potential of generalizing to other commodities. This article thus contributes to fulfilling different forecasting users' information needs for decision making under various circumstances.