Modelling the Great Lakes freeze: Forecasting and seasonality in the market for ferrous scrap

The paper offers a methodology for modelling seasonality in a volatile commodity market. It gives a practical example of the way seasonal factors can be incorporated into industrial forecasts. Recycled ferrous scrap is a widely traded commodity used in the steel and foundry industries. This paper considers the problems of forecasting scrap prices in the US market. Scrap prices display seasonal behaviour as a result of weather and patterns of industrial production. The authors consider various ways of modelling this seasonality, use of seasonal vector autoregression, the concept of seasonal integration and the use of dummy variables. A seasonal vector autoregression is developed. Here the quarterly series is decomposed into four annual series, one for each quarter. The authors regress each of these resultant series on its own lags and lags of other series, so developing a periodic autoregressive model. A series of tests enables them to determine the type of seasonality exhibited by the data. The simplest form of seasonal adjustment using seasonal dummy variables turns out to be the best for forecasting US scrap prices. Use of the test procedure suggests that employing seasonal dummies is the correct specification in this case. Inclusion of seasonal effects usually improves the estimation and forecasting performance of time series models. Comparison of a range of alternative forecasting models suggests a periodic autoregression only forecasts satisfactorily in the short run. ARIMA models with seasonal dummies show the best performance. A long lag length is necessary to capture long run cyclical effects.