Persistent wage differential and its implications on the Balassa–Samuelson hypothesis

Abstract

The objective of the study is to empirically examine the wage equalization assumption in the Balassa–Samuelson (BS) hypothesis. The wage equalization between the traded and the non-traded sectors is tested primarily based on resampling methods, permutation tests. The results show that the assumption does not hold uniformly. This study argues that a more general condition, which allows for wage differential between the sectors, can be used in the BS hypothesis as long as the wage differential is persistent. The persistent wage differential condition is empirically supported in this study.

Acknowledgements

I would like to thank Michael Boskin, Lawrence Lau, and Henry Rowen for their support. I am also grateful to Kangoh Lee for his invaluable suggestions. In addition, I have benefited from discussions with Trade & Development workshop participants at Stanford University for an earlier draft of the paper. I am solely responsible for all remaining errors.

Notes

 The sample includes Canada, Denmark, Finland, France, Japan, Korea, Norway, Sweden, Taiwan, and the USA.

 In fact, the original BS hypothesis assumes that labour is the only factor of production and each good requires a fixed amount of labour. Although the model in this study has both labour and capital, the assumptions of small open economy and the perfect capital mobility generate essentially the same results as in the original model. In this study, demand does not play a role since capital can be freely imported and exported. The relative price of non-traded goods is determined solely based on supply side.

 In Equation 14, because

(by the assumption) and

(due to the law of one price for the traded goods), inflation arises in the non-traded sector,

.

 Due to availability of data, not all the countries in the ISDB are considered. Also, the sample period of each country varies.

 In their study, Gregorio et al. (1994) provided a way to divide the traded and the non-traded sectors. They employed a cut-off point of trade, 10% of the total commodity expenditure in GDP.

 Examples include Strauss (1996, 1999), Chinn (1997, 2000) and Canzoneri et al. (1999) where they tested the assumptions of the BS hypothesis based techniques like Johansen cointegration tests, panel unit root tests, cointegration approaches, etc.

 The permutation tests used in this study are randomization tests, which generate the permutation distribution of a test statistic satisfying the null hypothesis. For details on the implementation of the randomization tests, see Kennedy (1995).

 Under the wage equalization, the mean value of the distribution of the relative wages should be one. In this study, the mean value is 1.019.

 The Kolmogrov–Smirnov test is often used for distribution comparisons. It compares the cumulative distribution functions by measuring the maximum vertical deviation between the two distributions.

 For each country, the generated and the observed samples are pooled and reshuffled. Thus, the pooled sample of each country includes both the generated and the observed samples of the relative wages.

 In terms of number of replications, this study chooses 1000, which is Efron and Tibshirani's recommended minimum number of replications.

 The ASL is approximated by Monte Carlo method.

 λ can be positive or negative. In this study, λ appears positive for Canada, France, Norway, and the USA and negative for Denmark, Japan, Korea, Taiwan.

 The persistent wage differential condition implies that λ is constant.