Aggregate Import Demand Function for Japan: A Cointegration Re-investigation

Abstract

Abstract Applying a rolling windows approach, this study aims to re-visit the existing literature on cointegration of aggregate import demand function in Japan. In contrary to many studies, the results of cointegration tests suggest that the presence of cointegration relationship for Japan's aggregate import demand function is not stable over the sample period, 1973.Q1–2007.Q2. Again, the estimated elasticities of income and relative prices of imports are varying over the sample period—the income variable is elastic while, the price is found to be inelastic, on average.

Keywords:

• Aggregate import demand
• cointegration
• Japan
• rolling windows

Acknowledgements

I would like to acknowledge the comments and suggestions made by Dietrich K. Fausten from Monash University, Australia on the crude draft of this study.

Notes

1. Japan experienced trade deficits only in the period 1974, and 1979–1980 over the period 1973–1999 (World Tables, World Bank).

2. Typically, the relative import prices is the ratio of import price to domestic price level which is often used to ease estimation problems, i.e. reduce multicollinearity (Houthakker & Magee, 1969). There are several proxies for domestic price level, i.e. GDP deflator, consumer price index (CPI), wholesale price, etc.

3. As highlighted by Hong (1999), the factors behind relative prices include: relative endowments of resources and productive factors, taste, market structure, scale, exchange rate, trade barriers, etc. The impacts of changes in these factors on import demand will take place through a change in relative prices.

4. The sample period is started from 1973 is due to the consideration of the period of the flexible exchange rate regime as in Hamori and Matsubayashi (2001).

5. If a GDP related measure of income is to be used, the essential choice is annual data or quarterly data.

6. The conventional tests of CUSUM and CUSUM of squares are inappropriate when the right-hand side of the equation includes lagged dependent variable(s) as regressor(s) such as ARDL specification, see Otto (1994).

7. Of six estimation techniques considered by Abeysinghe and Tan (1999) i.e. OLS, unrestricted ECM (or ARDL), fully modify least squares, the three-step estimator, OLS regression augmented by leads and lags of the differenced regressors, and Johansen's multivariate estimator, the study concludes that “In small samples OLS may still be the best choice” (Abeysinghe & Tan, 1999, p. 645). Their study covers annual data from 1963 to 1992 with a total of 30 observations.