The objective of this paper is to suggest a new predictive system for international trade, based on an unobserved component model. We employ the predictive system developed by Pastor and Stambaugh (2009), which is unlike other conventional predictive regression models. This paper derives an equivalent linear predictive regression from the predictive system, and explains why the proposed predictive system is able to achieve superior out-of-sample predictive power. When predictors are imperfect in an estimated equation, the equation fails to utilize all information from the predictors’ past history, and unexplained variations are captured by residuals in the estimated equation. With the use of the predictive system, we can more effectively deal with the dynamics of imperfect predictors.
For empirical illustration, we show that, in the case of Korea’s export and import growth rates, the predictive system has better out-of-sample predictive powers than the conventional regressions based on Root Mean Squares Error(RMSE). Results from an out-of-sample analysis show that, compared to the benchmark model, the predictive system improves forecast precision by 18.90% for the export growth rate, and by 7.95% for the import growth rate.
 

1. Introduction

2. Overview of the Conventional Productive Regression

3. Model Specification

3-1 Comparison with a Conventional Predictive Regression
3-2 Why Does the Predictive System Produce Better Forecasting Performance?

4. Estimation Procedure

4-1 A State-Space Representation
4-2 Forecasting Evaluation

5. Empirical Application

5-1 Data
5-2 The Predictability of the Export Growth Rate
5-3 The Predictability of the Import Growth Rate

6. Conclusion and Policy Implications

References