An investigation of forecast horizon and observation fit’s influence on an econometric rate forecast model in the liner shipping industry

Abstract

This paper evaluates the influence of forecast horizon and observation fit on the robustness and performance of a specific freight rate forecast model used in the liner shipping industry. In the first stage of the research, a forecast model used to predict container freight rate development is presented by exploring the relationship between individual company’s rates and aggregated market rates, and thus assists in dealing with uncertainty and market volatility for a given business situation. In the second stage, a design of experiment approach is applied to highlight the influence of the forecast horizon and observation fit and their interactions on the forecast model’s performance. The results underline the complicated nature of creating a suitable forecast model by balancing business needs, a desire to fit a good model and achieve high accuracy. There is strong empirical evidence from this study; that a robust model is preferable, that overfitting is a true danger, and that a balance must be achieved between forecast horizon and the number of observations used to fit the model. In addition, methodological guidance has also been provided on how to test, design, and choose the superior model for business needs.

Notes

1. Transfer functions representing GRI includes: the rates of 2 weeks before GRI, 1 week before, the week of GRI, the week after GRI, 2 weeks after GRI.

2. See De Gooijer and Hyndman (2006) or Makridakis (1993) for an overview.

3. Normal forecast practices would suggest using the standard deviation of the residuals to calculate the confidence interval for predictions. However, there is a potential, especially as there are occasional jumps in the rates (due to GRIs), that the forecast errors may in fact be non-normal even though the residuals are.