Table shows the liberalization measure ( L2, or state output as a percent of all industrial output) for the beginning and ending of the time period in question, 1986 and 1999, and the percentage change between those two periods. Provinces are rank-ordered, from smallest absolute percentage change in L2 to largest. In order to get a sense of whether liberalization captures policy variables that are distinct from those that prescribe openness, we have also included average FDI as a percent of total investment (average FDI I ), as well as average trade as a percent of GDP (average T GDP ) over the same time period.
If openness is equivalent to liberalization for the Chinese case, one would expect the liberalization variable to simply mirror FDI and trade. We also tried using the ratio of state sector workers to all employees, which performed in a similar manner to the state output measure, but the latter was more robust overall and made more sense in the employment regressions. Eventually we hope to improve the liberalization measure by exercising a principal components strategy that combines a number of different measures of liberalization policies, such as the proportion of output subject to the Plan, or the proportion of free prices.
China is actually divided into 22 provinces, five autonomous regions (Guangxi, Tibet, Xinjiang, Inner Mongolia, and Ningxia), and three municipalities (Beijing, Tianjin and Shanghai). Autonomous regions and municipalities have the same administrative rights as provinces; the entire group will be referred to as “provinces. ” Tibet has been left out of the dataset. Two ways in which FDI can directly help workers is by raising wages and employment. In this section we analyze the impact on wages and in the next section we consider employment. While wages are often used as an independent variable to explain FDI, it is rare to find
the causality running the other way in the empirical literature on FDI in developing countries. 6 But there is a clear causal link. First, FDI may affect labor demand (depending on whether it is greenfield investment or mergers and acquisitions, and on what competitive impact it has on domestic investment), thereby affecting wages.
Secondly, spillover effects from potentially higher productivity (and paying) foreign enterprises could raise wages throughout the country. And lastly, because capital is internationally mobile and labor is not, FDI may enhance capital’s bargaining power relative to labor, thereby lowering wages (Paus and Robinson 1998). Turning to the Chinese case, the core model we used is based on the notion that shortterm (annual) changes in wages depend on labor demand and supply. Data are panel data for China’s 29 provinces between 1986 and 1999 (please see the data appendix for a fuller explanation of the data used). 7 In regression equation (1) below the average provincial wage, measured as the average annual wage for a particular province, is a function of: total investment
( I ); foreign direct investment ( FDI ); total foreign trade (imports+ exports = T ); the total available labor force ( LF , defined as the population fifteen and over); productivity, where GDP employment = q ; and finally the liberalization variable, L2, the ratio of state sector output to all industrial output. Provincial fixed effects are ? ’s, a time trend has been added to control for uniform shocks, and ? is a serially uncorrelated random error. We use two measures of investment: gross investment ( I ) and “adjusted investment” ( adjI ), which subtracts FDI from
There are of course important exceptions. Using panel data that included both developing and developed countries, Paus and Robinson (1998) find that: FDI has a direct positive impact on real wages; that that impact is especially true in developing countries (but not in developed countries); and finally that this positive impact is true only for the period 1968-87, after which there is some evidence that the threat effect of relocating has had a negative effect on wage growth in industrialized countries. In a comparative study of Mexico, Venezuela, and the United
States, Aitken, Harrison and Lipsey (1995) find that higher levels of FDI are associated with higher wages in all three countries, but in Mexico and Venezuela, this association was limited to foreign-owned firms. This lack of evidence of wage spillovers to domestic firms is consistent with the large wage differentials between foreign and domestically-owned firms in these countries. Unless otherwise discussed, all units are in nominal yuan. This choice was taken because we could not get reliable deflators for the full set of variables, especially trade.
It should be noted, though, that when we tried the regressions using the consumer price index on wages and the GDP deflator on everything else, the results were consistent with the nominal numbers. In general, our results with imperfect deflators were similar to the nominal results reported here. gross investment to get a clearer sense of the effects of FDI on wages. The regression is run in logs to get elasticities, and first differences were used to address non-stationary in the variables. The results are detailed in Table 4. A Levin-Lin panel unit root test, used to determine nonstationarity in panel data, was applied.
The test may be viewed as an Augmented Dickey-Fuller test for panel data (Levin and Lin 1992). First differencing the variables addresses the fact that the variable means change over time, which could result in spurious correlations if left as is. Table 4 Wage equations for 29 provinces, using data for 1986-1999 (Estimation with first differences and fixed effects; dependent variable: ln average wage) Significant at the 90 percent level. 19 Table 4 above presents a series of regressions that build up to the full model in regressions V and VA.
Regressions I-V use the gross investment measure; regressions IA-VA use the adjusted investment measure. The potential for simultaneity – FDI, investment or trade could just as reasonably be argued to be the result of wage changes as the source of them –induced us to do some Granger causality testing. The results, which indicated mutual causation between these variables and wages, were not really satisfactory owing to the short time series and limited degrees of freedom. But the exercise did prompt us to use some lagged values on the right-hand side to partially address this problem.