Imported capital input, absorptive capacity, and firm performance: evidence from firm-level data.

Yasar, Mahmut

I. INTRODUCTION

The importance of productivity growth as a primary determinant of the underlying difference in income across countries is now a well-established empirical proposition (Hall and Jones 1999). Endogenous growth theory suggests that innovation is the main source of productivity growth (Grossman and Helpman 1991). However, the creation of new products and technologies is concentrated mostly in developed countries (Eaton and Kortum 2001). Developing countries rely much more on technology and knowledge produced by high-income countries than on direct investment in research and development. Therefore, a crucial question for developing economies is whether, how, and to what extent importing foreign technology can enhance the productivity of their firms to narrow the income gap between themselves and more developed countries.

In this article, we examine the productive impact of imported capital input by emphasizing its interaction with the absorptive capacity of manufacturing firms. The importing of foreign capital input, embodied by technology and knowledge, has been recognized in the economics literature as a significant conduit for technology transfer across countries (Coe and Helpman 1995; Eaton and Kortum 1996, 1999, 2001; Grossman and Helpman 1991; Xu and Wang 1999). (1) Coe and Helpman (1995), for example, find a significant impact on productivity at the country level from importing intermediate products and capital inputs, and Keller (2002a, 2002b) supports this finding with industry-level data. In studies at the firm level, it was observed that the productive impact of the imported capital input varies across countries in terms of both statistical significance and magnitude; in some countries it is strong, whereas in others it is weak or statistically insignificant. Kraay, Soloaga, and Tybout (2001), Keller and Yeaple (2003), and Vogel and Wagner (2008) find weak evidence of productivity effects of importing at the firm level, but Amiti and Konings (2007), Fernandes (2007), Gorg and Hanley (2005), Gorg, Hanley, and Strobl (2008), Halpern, Koren, and Szeidl (2006), Jabbour (2007), Kasahara and Rodrigue (2008), Lopez (2006), Lopez and Yadav (2009), and Yasar and Paul (2007) find statistically significant but highly varying effects. (2) These differences may stern in part from the fact that at different stages of development, countries have differences in some firm characteristics that help create mechanisms through which productivity gains are realized. This calls for further study of the productive contribution of imported capital input in various countries, at different stages of development, by emphasizing its linkages with firm characteristics such as input composition, skill intensity, or absorptive capacity.

The productive effects of importing can be examined using standard least-squares techniques by including a dummy variable to indicate the import of intermediate inputs in a production function or a productivity equation. Such a method, however, assumes homogeneity in the productive impact of the imported capital across firms by imposing a linearity restriction on the outcome equation. It is expected that importing capital from countries with higher levels of accumulated technical knowledge will improve the productivity of firms in the developing host country through R&D embodied in the capital and associated learning. (3) However, a shortage of absorptive capacity could potentially limit firms' ability to utilize these new technologies effectively. (4) The relationship may not be linear as the degree to which knowledge spillovers can be effectively utilized in the host country will depend on the skill intensity or absorptive capability (5) of firms in the host country, as well as some other firm, industry, or country characteristics. (6) Nelson and Phelps (1966) emphasize the importance of absorptive capability: "We suggest that, in a technologically progressive or dynamic economy, production management is a function requiring adaptation to change and that the more educated a manager is, the quicker will he be to introduce new techniques of production. To put the hypothesis simply, educated people make good innovators, so that education speeds the process of technological diffusion." They argue that the adoption of new technologies requires, in particular, a threshold stock of human capital in the host country, and the extent to which the gap between the technology frontier and current productivity can be closed depends on the level of this stock. (7) Spillover productivity effects from importing technology would thus be expected to occur when human capital in a host-country firm is above a certain threshold; and if labor skill levels fall short of this threshold, host-country firms are unlikely to be able to exploit the productive power of the foreign technology. (8)

In this article, our goal is to examine the productive impact of imported capital for a sample of Chinese manufacturing firms, with a focus on the interaction between imported capital and absorptive capacity. More specifically, we empirically assess the theoretical predictions of the studies highlighted above, by investigating whether the successful adoption of new technologies requires a threshold level of skilled labor in the host country and whether a higher level of skilled labor enhances productivity improvements from importing capital. We recognize that the parameter estimates may fail to distinguish between productivity differences from imported technology and those emerging from unobserved firm-specific characteristics that may have caused firms to choose to import capital, and that the productive impact may be heterogeneous across firms. Thus, to control for this, we estimate our model using two alternative econometric methods to ordinary least squares (OLS): an instrumental variable (IV) estimator to control for potential endogeneity and newly developed threshold regression techniques (Caner and Hansen 2004; Hansen 2000) that employ an endogenous search algorithm to determine the threshold level of the absorptive capacity and its statistical significance with and without endogenous explanatory variables. We use the share of skilled workers (technicians, engineers, and managers) as a proxy for the absorptive capacity of the manufacturing firms and examine the effect of the interaction of the human capital variable with imported technology on firm productivity.

After controlling for a number of firm characteristics that affect productivity and endogeneity, we find that importing capital is associated with significantly higher productivity in the Chinese manufacturing firms and that those firms with higher absorptive capacity gain significantly more from importing foreign technology, implying that the productive contribution of imported capital is significantly skilled-laborusing. These productivity gains also depend on a threshold level of human capital; the productivity differential for importing capital becomes statistically significant when the share of managers and engineers in total employment is about 38%, with a confidence interval of 28-44%. Furthermore, our results indicate that the productive contribution of skilled labor is significantly higher in firms that import foreign capital.

II. EMPIRICAL MODEL

To examine whether and to what extent the productive impact of imported capital varies with the ability of Chinese manufacturing firms to absorb new technologies, we use OLS, IV, and a recently developed endogenous threshold regression estimate of a production function specification that represents the most output, technologically producible from the given input vector, firm characteristics, and external conditions. (9) We assume that the production process is represented by a Cobb-Douglas production function, (10) Y = f (K, L, IMP, SH, IMP x SH, R), written as:

(1) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII],

where i is a firm subscript; ln Y is the log of value added; In L and In K are the log values of labor and capital input, respectively; IMP = 1 if the firm imported any machinery and 0 otherwise; SH represents the share of technicians, engineers, and managers in total employment; and u is a stochastic error term. (11) [R.sub.m] includes the variables for internal firm characteristics, which are age of the firm (AGE); whether or not the firm's products have been granted ISO900 certification (ISO); capacity utilization (CU, defined as the amount of output actually produced relative to the maximum amount that could be produced with existing machinery and regular shifts); and size (SD), (12) industry (ID), (13) and region (CD) dummies. (14)

In this model, our main variable of interest is the interaction of the imported machinery variable (IMP) with the variable representing the share of technicians, engineers, and managers in total employment (SH). (15) Based on the insights from theoretical studies, we expect highly skilled workers to use imported capital more effectively. The coefficient on the interaction term [[beta].sub.5] thus represents the extent to which human capital augments the productive benefits of imported machinery. As this coefficient represents the impact of imported machinery on the productive contribution of skilled labor or on the productive contribution of imported machinery of more skilled labor, a significantly positive (negative) coefficient estimate implies complementarity (substitutability) between imported machinery and skilled labor.

The coefficient on IMP alone, [[beta].sub.3], represents the direct impact of imported machinery on the productivity of firms; if [SH.sub.i] = 0 the productivity differential between the importers and nonimporters is [[beta].sub.3]. The coefficient on [SH.sub.i], [[beta].sub.4], similarly represents the productivity impact of skilled labor share for non-importers. The productivity impact of importing capital for firm i including the impact of human capital is thus [partial derivative] ln[Y.sub.i]/[partial derivative] IMP = [[beta].sub.3]+ [[beta].sub.3] x [SH.sub.i], and the productivity impact of human capital including the contribution of importing capital is [partial derivative] ln [Y.sub.i]/[partial derivative] [SH.sub.i] = [[beta].sub.4] + [[beta].sub.5] x IMP. We expect the impact of the simultaneous increase in skilled labor share and imported capital input on the productivity of firms to be positive. This requires [[beta].sub.5] to be positive and statistically significant.

The other firm-specific characteristics accommodate heterogeneity across firms. AGE is included because more experience would be expected to enhance productivity. It also represents the reputation of a firm. Whether or not the firm's products have been granted ISO9000 certification (ISO) is included to capture the firm's managerial or product quality. CU is included to control for the average utilization of a firm's fixed inputs. The remaining explanatory variables are firm-specific controls for size, industry, and region.

III. ESTIMATION METHODS

A. Ordinary Least Squares

Equation (1) can be estimated by using OLS with interaction terms to empirically examine the productive impact of imported technology by emphasizing its interaction with skill intensity for firms in the Chinese manufacturing industry. If [[beta].sub.5] is positive and statistically significant, then we can conclude that there is a significant association between productivity and imported technology, conditional on skilled labor share; firms differ in their stock of skilled labor and the productive impact of imported machinery depends to a large extent on these stocks. However, this cannot be interpreted as a causal relationship for a few reasons. First, firms with better performance might choose to import foreign technology. It may be that only the most productive firms are able to enter the import market. In other words, importing firms may be more productive at the outset, not as a result of enjoying benefits associated with imported capital. In addition to this self-selection, there may be some omitted determinants of productivity that will be inherently correlated with the import variable. In these cases, the error term will be correlated with the imported capital variable and the OLS estimates will be biased and inconsistent. (16)

Thus, to evaluate potential endogeneity issues that may affect the robustness of our model, we also employ an IV estimator. (17)

B. IV Estimator

We assume that we are interested in estimating [y.sub.i] = [x'.sub.i][beta] + [u.sub.i]. When endogeneity is present, one must control for potential estimation biases by identifying J instruments in an associated vector [z.sub.i]. (18) For [beta] to be a consistent estimator, the population moment condition E[([y.sub.i] - [x'.sub.i][beta])[z.sub.i]] = 0 must hold. The IV estimator solves the corresponding sample moment condition 1/N [[summation].sup.n.sub.i] ([y.sub.i] - [x'.sub.i][beta])[z.sub.i] z = 0 by choosing values of [beta] that drive this condition as close to zero as possible by minimizing: [Q.sub.N] ([beta]) = [1/N [[summation].sup.N.sub.i] ([y.sub.i] - [x'.sub.i][beta])[z.sub.i]'. [[OMEGA].sub.N] [1/N [[summation].sup.N.sub.i]([y.sub.i] - [x'.sub.i][beta])[z.sub.i]], where [[OMEGA].sub.N] is a weighting matrix.

Valid instruments ([z.sub.i]) must not be correlated with the error term, that is, the instruments used must have an impact on productivity only through the endogenous import variable and its interaction with the skilled labor share. (19) We assume that the following variables meet this requirement and use them as instruments to control for resulting potential endogeneity (20): whether the general manager of the firm is from an industrialized country (MN); whether the firm is a subsidiary or a joint venture of a multinational firm (FO); whether the firm was an exporter in the previous year (EXP); and the percentage of firm owned by the state/provincial government (SO). (21) If the manager is from an industrialized foreign country, the firm is more likely to import. Firms with some state ownership are less likely to import. A firm that is an exporter or a subsidiary/joint venture of a multinational firm is more likely to import. (22) We assume that these variables do not have an independent direct impact on firm productivity; their effect is through import and its interaction with the skill level of the firm. (23)

C. Threshold Regression Models

To check the robustness of our results obtained by the interaction effects using both OLS and IV estimators, we use an alternative method that allows the data to endogenously select a sample split and simultaneously test the presence of the threshold level of skilled labor share. As explained earlier, the productive impact of imported technology may be heterogeneous in the sense that it may vary at different levels of skilled labor share, indicating that the relationship is not monotonic but nonlinear. The failure to account for these potential threshold effect non-linearities can potentially result in biased results. Standard threshold models assume that one knows the break points with certainty. However, prior knowledge of how the impact of imported capital on productivity varies with the threshold variable is not available with certainty. Hansen (2000) introduced a threshold regression technique that treats the threshold as an unknown parameter and estimates it endogenously using the data, instead of assuming it arbitrarily. It also allows us to test the existence of this threshold. (24)

We implement this approach to endogenously search for regime changes in the data by using the skilled labor share as a threshold variable.

This endogenous threshold analysis allows the impact of the imported technology on productivity to vary with skilled labor share. When there is a single threshold, the regression model can be rewritten as follows. (25)

(2) [Y.sub.i] = [beta]'[X.sub.i] + [[delta].sub.i] I([SH.sub.i] [less than or equal to] [gamma])IMP + [[delta].sub.2] I ([SH.sub.i] > [gamma])IMP + [u.sub.i],

where [Y.sub.i] is the dependent variable; [SH.sub.i] is a threshold variable, corresponding to the skilled labor share; [X.sub.i] is a vector of control variables explained in the previous section (26); I is an indicator function; and [gamma] is the estimated threshold value. This method allows the division of the impact of import into two regimes, conditional on whether the skilled labor share is smaller or greater than the estimated threshold level [gamma]. In other words, it allows the regression parameters to differ depending on the value of the threshold variable. For instance, [[delta.sub.1] is the impact of imported technology on productivity when [q.sub.i] [less than or equal to] [gamma] and [[delta].sub.2] is the impact of imported technology on productivity when [q.sub.i] > [gamma]. This method estimates [gamma] and the parameters using least-squares estimator through concentration and obtains reliable confidence intervals for [gamma].

The estimation strategy includes three main steps. (27) First, we estimate the threshold value [gamma] (a value that minimizes the concentrated sum of squared errors). Let [S.sub.n]([gamma]) represent the sum of squared errors for any given value of [gamma]. Then, the slope coefficients are estimated by OLS by obtaining a [??] that minimizes [S.sub.n]([gamma]). As the parameter [delta] depends on the threshold value [gamma] (i.e., [delta]([gamma])), the model is not linear in its parameters. Hansen (2000) suggested obtaining the OLS estimates through concentration. Conditional on a threshold value, say [[gamma].sub.0], the regression equation is linear in its parameters, and the sum of the square error function can be minimized to obtain the conditional OLS estimators [??]([[gamma].sub.0]) and [??]([[gamma].sub.0]). Suppose the concentrated sum of square errors function is [S.sub.n]([[gamma].sub.0]) = [S.sub.n]([??]([[gamma].sub.0]), [??]([[gamma].sub.0]), [[gamma].sub.0]). By trying all values of the threshold variable (SH), the estimator of the threshold variable will be a value of [gamma] that results in the smallest [S.sub.n]([[gamma].sub.0]), which is defined as follows: [??] = arg min [S.sub.n]([gamma]). This minimization problem is solved by a grid search over the values of the threshold variable (SH).

Second, we test the statistical significance of the threshold effects. More specifically, we test the null hypothesis of no threshold (i.e., [H.sub.0] : [[delta].sub.1] = [[delta].sub.2]) against the alternative hypothesis of a threshold regression model (i.e., [H.sub.A]:[[delta].sub.1] [not equal to] [[delta].sub.2]) by using a bootstrap procedure that allows us to obtain critical values for the test statistic by treating the explanatory variables and the threshold variable (SH) as given and holding their values fixed in each repeated bootstrap sample. The bootstrap dependent variable is generated by drawing a sample of residuals from N(0, [[??].sup.2.sub.i]), where [[??].sub.i] is the OLS residual from the estimated threshold model. Repeating this procedure a large number of times, the bootstrap estimate of the p-value under the null is given by the percentage of draws for which the simulated statistic exceeds the actual one. This approach produces asymptotically correct p-values. The null of no threshold effect is rejected if the p-value is smaller than the desired critical value.

Finally, if the null hypothesis is rejected, we construct confidence intervals for the threshold ([gamma]) variable. We first need to test for a threshold value (i.e., [H.sub.0] : [gamma] = [[gamma].sub.0]; [H.sub.1] : [gamma] [not equal to] [[gamma].sub.0]). Assuming normality, a standard method to test this is to use likelihood ratio statistics:

[LR.sub.n](y) = n([S.sub.n]([gamma]) - [S.sub.n]([??])/[S.sub.n]([??])).

The null hypothesis is rejected when [LR.sub.n]([[gamma].sub.0]) is too large. Hansen (2000) illustrated that [LR.sub.n]([gamma]) does not have a standard [chi square] distribution under the endogenous threshold specification. He derived the correct distribution function and tabulated the asymptotic critical values.

By using this methodology, we expect to obtain a more exact assessment of the productive impact of import than that obtained by traditional monotonic approaches.

The threshold model introduced by Hansen (2000) assumes that all right-hand-side variables are exogenous. Caner and Hansen (2004) introduced a technique that allows us to estimate the threshold models with endogenous variables but an exogenous threshold variable. First, a threshold reduced-form model is estimated by least squares, and then the fitted values of the endogenous variable are obtained based on this reduced-form estimate. Finally, these fitted values are used in place of the endogenous variables in a structural equation that is estimated by least squares. Although the threshold parameter is estimated using a two-stage least-squares estimator, the slope parameters are estimated by employing a generalized method of moment estimator. We have also estimated the threshold parameter by using this method to check the robustness of our results. (28)

IV. DATA

To estimate the productive impact of imported technology associated with skilled labor intensity, we use cross-section survey data from the Investment Climate Survey conducted by the World Bank in 2003 on a sample of Chinese manufacturing firms. The World Bank's Enterprise Surveys use either simple random sampling or random-stratified sampling to ensure the randomness of their sample. Face-to-face interviews were conducted with firms' managers and accountants, using a sampling method designed to ensure adequate representation of firms by industry, size, ownership, export orientation, and location. The firms in the survey are from 18 Chinese regions: Benxi, Changchun, Changsha, Chongqing, Dalian, Guiyang, Haerbin, Hangzhou, Jiangmen, Kunming, Lanzhou, Nanchang, Nanning, Shenzhen, Wenzhou, Wuhan, Xian, and Zhengzhou. The industries represented are Food Processing, Garment and Leather Products, Electronics Equipment, Electronics Parts Making, Household Electronics, Auto and Auto Parts, Chemical Products and Medicine, Biotech Products and Chinese Medicine, and Metallurgical Products. Summary statistics of the data are presented in Table I. Although the survey was conducted for firms in both the service and manufacturing industries, we dropped the firms in the service industries because information on input and sales was not available for many firms in this sector.

Furthermore, we dropped observations that were clearly erroneous, such as negative values of age, output, and labor. After dropping erroneous observations and the missing observations for the variables included in our model, 1,161 observations remained. Of these 1,161 firms, about 33% reported that they imported foreign capital (machinery) and the remaining 67% did not. The average skilled labor share in our sample is about 27%. As shown in Table 2, none of the variables of interest for our model exhibit a very high degree of correlation, putting aside inference problems because of multicollinearity. (29)

V. RESULTS AND DISCUSSION

A. OLS Results

The OLS parameter estimates for our model are presented in Table 3 for the Chinese manufacturing firms in our data. The statistically significant estimates are identified by asterisks. (30) The parameter estimates show a positive productive contribution of skilled labor associated with imported machinery and the productive contribution of imported machinery associated with more highly skilled labor.

More specifically, the productive impact of additional skilled labor for non-importers (IMP=0) is [[beta].sub.4] =0.918, whereas that for importers (IMP = 1) is [[beta].sub.4] + [[beta].sub.5] x IMP = 0.918 + 0.885 = 1.803; so [[beta].sub.5] = 0.885 indicates how much importing capital augments the productive contribution of more human capital (a greater share of managerial, engineering, and technical labor). Conversely, the productive contribution of importing capital for a firm with no skilled labor is insignificantly different than zero ([[beta].sub.3] = -0.006), but skilled labor augments the productivity of importing as [[beta].sub.5] = 0.885, which is significantly different than zero. For example, for a firm with a mean skilled labor share of 0.271, the productivity enhancement associated with importing capital is [[beta].sub.3] + [[beta].sub.5] x SH = -0.006 + 0.885 x (0.271) = 0.235--about 23.5%. The productive effect of importing capital is thus driven by a strongly positive skilled labor bias; importing capital is skilled labor intensive in the sense that the productive differential for importers increases with additional human capital.

The significantly positive estimated coefficient on the interaction between imported machinery and skilled labor share means that they are complements--sometimes referred to as technology-skill complementarity. This finding is consistent with theoretical studies that predict the productivity of importing machinery to be positively related to a more highly skilled workforce because effective adoption or absorption of new technology requires human capital. Thus, firms wishing to obtain new technology by importing machinery need to hire skilled workers to utilize this technology effectively. Moreover, the magnitude of the technology spillover effect from importing is larger for firms with more human capital.

In fact, the productive impact of importing capital with no skilled labor is essentially zero but increases significantly with more human capital. It can thus be surmised that a threshold of skilled labor is required to exploit the imported technology. To test this hypothesis, we estimated the productivity impact of imported machinery at various percentiles of skilled labor share, as reported in Table 4. This table presents the percentage difference in productivity between importers and non-importers at different percentiles (31) of skilled labor share. The first column of Table 4 shows the skilled labor share; the second column shows the differential impact of a higher skill share on the productivity effect of importing capital based on the OLS results; and the third column presents the differential based on the IV estimator, which will be highlighted in the next section. The corresponding standard errors obtained from the Delta Method are presented in parentheses. The productive differential for importers is positive at all percentiles of skilled labor share. However, the differential does not become statistically significant at the conventional significance level of 5% until the share reaches about 23%, which corresponds to the 50th percentile of the skilled labor share. At this percentile, the productivity of importing capital becomes -0.06 + 0.885 x (0.231) = 0.199--a productivity premium of about 19.9% for importers. An increase in the skilled labor share to 50% implies a productivity differential of nearly 43%. These results are consistent with the suggestions in the literature that absorptive capability of domestic firms is the main factor affecting the degree to which knowledge spillovers can be utilized in the host economy, and that a threshold level of human capital must be attained to exploit such spillovers (Kokko, Tansini, and Zejan 1996).

Further, our other coefficient estimates suggest that total factor productivity is significantly related to the control variables. (32) The statistically significant positive coefficient on ISO9000 indicates that certified firms are about 28% more productive than those that are not, and the statistically significant positive coefficient on CU indicates that productivity increases with average CU. Finally, age seems to have a negative linear impact on firm productivity.

B. IV Results

The parameter estimates using the IV estimator are presented in the third column of Table 3, with robust standard errors in parentheses. The results, denoted IV, are based on the instrument set explained above. Using the Sargan test of overidentifying restrictions to test for the correlation between the instruments and the error term gives the p-value of .406, which validates the use of these instruments. We also use an F-test to examine whether these instruments are sufficiently related to the endogenous variables by estimating their joint significance in reduced-form equations. F-tests show that these instruments are valid instruments.

As shown in the third column of Table 3, (30) controlling for potential endogeneity by IV estimation increases the productive impact of importing. The parameter estimates show a positive productive contribution of skilled labor associated with imported machinery and the productive contribution of imported machinery

associated with more highly skilled labor. The productive impact of additional skilled labor for non-importers (IMP=0) is [[beta].sub.4] =0.721, whereas that for importers (IMP = 1) is 2.170; so [[beta].sub.5] = 1.449 indicates how much importing capital augments the productive contribution of a greater share of managerial, engineering, and technical labor. Conversely, the productive contribution of importing capital for a firm with no skilled labor is insignificantly different than zero ([[beta].sub.3] = 0.051), but skilled labor enhances the productivity of importing as [[beta].sub.5] = 1.449 is significantly different than zero. For a firm with a median skilled labor share of 0.231, the productivity boost associated with importing capital is about 38.7%. This finding is consistent with the OLS results, but the parameter estimates obtained by the IV estimator are higher. These results indicate that OLS biases the parameter estimates downward, assuming that the identified IVs used are strong enough. Overall, these results also verify the previous results and the theoretical predictions that the productivity of importing machinery is positively related to a more highly skilled workforce.

Furthermore, as illustrated in the third column of Table 4, although there is no significant productivity differential between the importers and non-importers when the skilled labor is essentially zero, the gap increases significantly with more human capital. The coefficients for the remaining independent variables are similar in terms of both their magnitude and significance for the IV estimation.

C. Threshold Regression Model Results

The results from the threshold regression model are reported in the first column of Table 5. Using skilled labor share as the potential threshold variable, Hansen's (2000) endogenous search technique identifies a statistically significant threshold, which corresponds to a skilled labor share of [??] = 0.427 with 95% confidence interval [0.289, 0.466]. (33) The corresponding bootstrapped p-value, which is obtained by using a Lagrange Multiplier test for a threshold with 1,000 replications (as explained in Hansen 1996), is 0.041. Thus, the null hypothesis is rejected at the conventional 5% significance level, indicating that the productive impact of imported capital varies across the two regimes, that is, there is a sample split based on our threshold variable of skilled labor share. (34) Although import is significantly associated with the firm productivity in the first regime only at the conventional 10% level, it becomes statistically significant at the 1% level when the skilled

labor share is above the threshold level of 0.427, which is the threshold endogenously determined by the data. In the first regime, where the skilled labor share is less than the threshold of 42.7%, the importing firms are only 16.2% more productive than the non-importing firms. This productive difference is not significant at either 0.05 or 0.01 significance levels. However, in the second regime, the productivity premium for the importing firms is 49.9% and significant at the 1% significance level. The threshold level divides the sample of 1,161 into two subsamples of 996 and 165 firms; 996 firms in the sample have a skilled labor share that is less than [gamma], whereas 165 firms have a skilled labor share that is higher than [??].

The normalized likelihood ratio sequence [LR.sub.n]([gamma]) statistic as a function of our threshold variable is also illustrated in Figure I. The least-squares estimate of [gamma] is the value that minimizes the function [LR.sub.n]([gamma]) which occurs at [??] = 0.427. The asymptotic 95% critical value for the threshold estimates is shown by the portion of the curve in the graph that lies below the dotted line drawn at the critical value. As illustrated in the graph, the 95% confidence set is [0.289, 0.466].

We also estimated the threshold parameter using a method introduced by Caner and Hansen (2004) that allows us to estimate the threshold models with endogenous variables. As illustrated in the third column of Table 3 and in Figure 2, this method identifies a threshold that corresponds to a skilled labor share of 0.380, with 95% confidence intervals of [0.289, 0.443]. These results are consistent with those obtained by using the threshold regression technique that treats all the explanatory variables as exogenous.

[FIGURE 1 OMITTED]

VI. CONCLUSIONS

Recent studies have suggested that importing capital is an important channel for firms in developing countries to transfer new technologies from developed countries. However, the literature does not clearly establish the importance of particular firm characteristics (such as the firm's absorptive capacity) on the productive impact of this capital input, which involves the mechanisms through which these productivity gains are realized. In particular, firms' internal conditions, such as a lack of skilled labor, might limit a firm's ability to exploit the potential productive benefits of imported technology. This study focuses on the role played by firms' absorptive capacity in the link between imported technology and firm productivity. To capture this capacity, the share of skilled workers for a sample of Chinese manufacturing firms is interacted with a variable representing the imported machinery in a Cobb-Douglas production function specification. In addition to a standard least-squares (OLS) estimator with interaction effects, two alternative econometric methods are used to estimate the parameters of production function: an IV estimator and the endogenous threshold regression technique.

[FIGURE 2 OMITTED]

After controlling for observable variables that represent heterogeneity across firms and potential endogeneity, our empirical results suggest that importing machinery plays an important role in enhancing firm productivity, but that the level of absorptive capacity determines the level of positive effects attained. Thus, in keeping with the theoretical predictions of existing literature, this study finds that the productive impact of the import does not increase monotonically and that the impact is more profound when the level of absorptive capacity is above a certain threshold. Hence, a higher absorptive capacity can allow firms to maximize benefits associated with new technologies and manufacturing techniques transferred from high-income countries. Conversely, importing machinery more significantly increases the productivity of firms with higher skill shares. This gap widens as the skill share increases. This finding implies capital-skill or technology-skill complementarity. Thus, enhancing firm productivity will involve simultaneously encouraging deepening capital and augmenting the skill level of the labor force.

The productive impact of the imported technology may vary with the country's stage of development, over time, as well as with the capacity of the firms to absorb new technologies. Thus, further research is needed to analyze the effects of temporal changes on firms' performance across various countries by utilizing firm-level panel data. The estimated productive impact of importing is higher in this study than in some of the related studies in the literature. One apparent explanation is that we allowed for differences in the absorptive capacity across firms in this article. In addition, previous studies used the intermediate material, whereas in this article we used an import variable that represents the imported machinery. It might also be related to the identified IVs set used and further endogeneity bias. In subsequent research, it can be helpful to identify different instruments and add a time series dimension to the cross-section analysis to investigate this further. Furthermore, we used the firm-level skill intensity as a measure for differences in absorptive capacity and estimated the productive impact of imported machinery by allowing for differences in absorptive capacity across firms. It can be valuable to use other proxies for the absorptive capacity such as research and development expenditures, the number of patents granted, or relative initial productivity.

ABBREVIATIONS

IV: Instrumental Variable

IMP: Imported Machinery Variable

OLS: Ordinary Least Squares

CU: Capacity Utilization

VIF: Variance Inflation Factors

FDI: Foreign Direct Investment

doi: 10.1111/j.1465-7295.2010.00352.x

REFERENCES

Amiti, M., and J, Konings. "Trade Liberalization, Intermediate Inputs. and Productivity: Evidence from Indonesia." American Economic Review, 97(5), 2007, 1611-38.

Aw, B. Y., M. Roberts, and T. Winston. "'Export Market Participation, Investments in R&D and Worker Training, and the Evolution of Firm Productivity.'" World Economy, 3(1(1), 2007, 83-104.

Aw, B. Y., M. Roberts, and D. Y. Xu. "R&D Investment, Export, and the Evolution of Firm Productivity.'" American Economic Review, Papers and Proceedings, 98(2), 2008, 451-56.

--. "R&D Investment, Exporting. and Productivity Dynamics.'" NBER Working Paper No. W 14670, 2009.

Benhabib, J.. and M. M. Spiegel. "The Role of Human Capital in Economic Development: Evidence from Aggregate Cross-Country Data." Journal of Monetary Economics, 34, 1994, 143-73.

Bernard, A., S. Redding, and P. Schott. "'Multi-product Firms and Trade Liberalization." NBER Working Paper No. 12782, 2006.

Bustos, P. "The Impact of Trade on Technology and Skill Upgrading: Evidence from Argentina." Department of Economics and Business, Universitat Pompeu Fabra, Working Paper No. 1189, 2007.

Cameron, A.C., and P. K. Trivedi. Microeconometrics: Methods and Applications. Cambridge: Cambridge University Press, 2005.

Caner, M., and B. Hansen. "Instrumental Variable Estimation of a Threshold Model." Econometric Theory. 20, 2004, 813-43.

Clerides, S., S. Lach, and J. Tybout. "Is Learning-by-Exporting Important? Micro Dynamic Evidence from Colombia, Mexico, and Morocco." Quarterly Journal of Economics, 53, 1998, 903-47.

Coe, D. T., and E. Helpman. "International R&D Spillovers." European Economic Review, 39, 1995, 859-87.

Cohen, W. M., and D. A. Levinthal. "Absorptive Capacity: A New Perspective on Learning and Innovation." Administrative Science Quarterly, 35, 1990, 128-52.

DeLoecker, J. "Product Differentiation, Multi-Product Firms and Estimating the Impact of Trade Liberalization on Productivity." NBER Working Paper No. 13155, 2006.

Eaton, J., and S. Kortum. "Trade in Ideas: Patenting and Productivity in the OECD.'" Journal of International Economics. 40, 1996, 251 78.

--. "International Patenting and Technology Diffusion: Theory and Measurement." International Economic Review, 40, 1999, 537-70.

--. "Trade in Capital Goods." European Economic Review, 45(7), 2001, 1195-235.

Feenstra, R. C. "Integration of Trade and Disintegration of Production in the Global Economy." Journal of Economic Perspectives, 12, 1998, 31-50.

Fernandes. A. "Trade Policy, Trade Volumes and Plant-Level Productivity in Colombian Manufacturing Industries." Journal of International Economics, 71. 2007, 52-71.

Goldberg, P., A. Khandetwat, N. Pavcnik, and P. Topalova. "Multi-Product Firms and Product Turnover in the Developing World: Evidence from India." NBER Working Paper No. 14127, 2009.

Gorg, H., and A. Hanley. "'International Outsourcing and Productivity: Evidence from the Irish Electronics Industry." North American Journal of Economics and Finance, 16(2), 2005, 255-69.

Gorg, H., A. Hanley, and E. Strobl. "Productivity Effects of International Outsourcing: Evidence from Plant Level Data." Canadian Journal of Economics, 41(2), 2(108, 670-88.

Grossman, G., and E. Helpman. Innovation and Growth in the Global Econonomy Cambridge, MA: MIT Press, 1991.

Hall, R., and C. Jones. "'Why Do Some Countries Produce So Much More Output per Worker than Others'?" Quarterly Journal of Economics, 114(1), 1999. 83-116.

Halpern, L.. M. Koren, and A. Szeidl. "Imports and Productivity." CEPR Discussion Paper No. 5139, 2006.

Hansen, B. E. "Inference When a Nuisance Parameter is Not Identified under the Null Hypothesis." Econometrica, 64(2). 1996. 413-30.

--. "Sample Splitting and Threshold Estimation." Econometrica, 68, 2000, 575-603.

Head, K., and J. Ries. "Offshore Production and Skill Upgrading by Japanese Manufacturing Firms." Journal of International Economics, 58, 2002. 81-105.

Jabbour, L. "Outsourcing, Offshoring and Firm's Performance: Evidence from the French Manufacturing Industry." Working Paper, University of Paris, 2007.

Kasahara, H., and J. Rodrigue. "Does the Use of Imported Intermediates Increase Productivity'? Plant-level Evidence." Journal of Development Economics, 87(1), 2008, 106-18.

Katayama, H., S. Lu, and J. Tybout. "Firm-Level Productivity Studies: Illusions and a Solution." International Journal of Industrial Organization, 27(3), 2009, 403-13.

Keller, W. "Geographic Localization of International Technology Diffusion." American Economic Review, 92, 2002a, 120-42.

--. "Knowledge Spillovers at the World's Technology Frontier." Working Paper, University of Texas, Austin, 2002b.

Keller, W., and S. R. Yeaple. "Multinational Enterprises, International Trade, and Productivity Growth: A Firm Level Evidence from the United States." Brown University, Department of Economics, Working Paper No. 2003-06, 2003.

El Khoury, A. C., and A. Savvides. "Openness in Services Trade and Economic Growth." Economic Letters, 92(2), 2006, 277-83.

Kokko, A., R. Tansini, and M. Zejan. "Local Technological Capability and Spillovers from FDI in the Uruguayan Manufacturing Sector." Journal of Development Studies, 34, 1996, 602-11.

Kraay, A., I. Soloaga, and J. Tybout. "Product Quality, Productive Efficiency, and International Technology Diffusion: Evidence from Plant-Level Panel Data." World Bank Research Working Paper Series No. WPS 2759, 2001.

Lopez, R. "Imports of Intermediate Inputs and Plant Survival." Economics Letters, 92(1), 2006, 58-62.

Lopez, R., and N. Yadav. "Imports of Intermediate Inputs and Spillover Effects: Evidence from Chilean Plants." Journal of Development Studies, 46(8), 2009, 1385-1403.

Nelson, R., and E. Phelps. "Investment in Humans, Technological Diffusion and Economic Growth." American Economic Review, 56, 1966, 69-75.

Olley, G. S., and A. Pakes. "The Dynamics of Productivity in the Telecommunications Equipment Industry." Econometrica, 64(6), 1996, 1263-97.

Papageorgion, C. "'Trade as a Threshold Variable for Multiple Regimes." Economies Letters, 77, 2002, 85-91.

Pavcnik, N. "What Explains Skill Upgrading in Less Developed Countries?" Journal of Development Economics, 71, 2003, 311-28.

Rodrik, D. "Institutions for High-Quality Growth: What They Are and How to Acquire Them." Comparative International Development, 35(3), 2000, 3-31.

--. "Industrial Development: Stylized Facts and Policies." Working Paper, 2006. Available at http://www.hks.harvard.edu/fs/drodrik/Research%20 Papers/industrial%20development.pdf.

Tan, H. W., and G. Batra. "Enterprise Training in Developing Countries: Incidence, Productivity Effects, and Policy Implications." Washington, DC, World Bank, 1995.

Van Biesebroeck, J. "'Exporting Raises Productivity in SubSabaran African Manufacturing Firms." Journal of International Economics, 67(2), 2005, 373-91.

Vogel, A., and J. Wagner. "Higher Productivity in Importing German Manufacturing Firms: Self-Selection, Learning From Importing, or Both?" Working Paper, University of Luneberg, Institute of Economics, Working Paper No. 106, 2008.

Xu, B., and J. Wang. "Capital Goods Trade and R&D Spillovers in the OECD." Canadian Journal of Economics, 32, 1999, 1258-74.

Yasar, M., and C. J. M. Paul. "International Linkages and Productivity at the Plant Level: Foreign Direct Investment, Exports, Imports and Licensing." Journal of International Economics, 72(1), 2007, 373-88.

Yasar, M., R. Raciborski, and B, Poi. "Production Function Estimation in Stata Using the Olley and Pakes Method." The Stata Journal, 8(2), 2008, 1-11.

(1.) See Keller (2002b) for a detailed discussion of international technology diffusion.

(2.) Trade liberalization may also increase the firm's productivity through the reallocation of resources across products within firms (Bernard, Redding, and Schott 2006; Goldberg et al. 2009). Our data do not allow us to study the relationship between importing and changes in firms' product mix over time.

(3.) See Feenstra (1998) and Head and Ries (2002).

(4.) By developing a dynamic structural model, for instance, Aw, Roberts, and Xu (2008, 2009) found significant interactions between R&D, exporting, and productivity for exporting firms in the Taiwanese electronics industry.

(5.) Absorptive capacity in this context means that firms in the host country must have a certain proficiency level to use the imported technology to enhance productivity.

(6.) In a more general context, it is now well established that higher quality institutional arrangements are necessary for trade liberalization to enhance countries' productivity and thus income (Rodrik 2000, 2006).

(7.) Benhabib and Spiegel (1994) use country-level data to test the Nelson-Phelps hypothesis and conclude that the flow rate of technology spillovers from leaders to followers depends on human capital levels. Cohen and Levinthal (1990) further show that the skills required to benefit from knowledge spillovers are usually the same as those required to create knowledge; benefiting from spillovers and creating knowledge are closely related processes.

(8.) The literature on the productivity impact of foreign direct investment (FDI) and exporting on domestic firms provides support for these predictions. For example, using data on manufacturing firms in Uruguay, Kokko, Tansini, and Zejan (1996) find that spillover productivity effects are largest for host-country firms that have sufficient human capital to effectively utilize the foreign technology. Tan and Batra (1995) find that joint ventures with foreign companies can also facilitate the transfer of technology because they are implemented by foreign management and accompanied by training. Aw, Roberts, and Winston (2007) find that a firm's investments in R&D and worker training can bolster its ability to internalize the benefits associated with the export market. See also Pavcnik (2003) and Bustos (2007) for the significant relationship between trade and skill upgrading.

(9.) Although we use cross-section data, and thus most questions in the survey obtain information from 2002, the input and sales variables were available for the previous 2 years for some firms in the sample. Thus, we also used a two-step approach, where we first estimate the total factor productivity using both OLS and a semi-parametric technique introduced by Olley and Pakes (1996), which allows us to control for simultaneity bias when estimating a production function and thus obtain consistent parameter estimates (Yasar, Raciborski, and Poi 2008). The two-step approach produces similar results to those reported here.

(10.) We also estimated our model using a translog production function, which allows for a full range of substitution among inputs and thus captures differential productivity patterns for firms with heterogeneous input composition. The coefficients for the variables of interest were similar to those reported here.

(11.) We also used gross output formulation. The results were consistent. For instance, the coefficient on the interaction term was 0.673 in the IV model, which was statistically significant at the 0.01 significance level.

(12.) We use two size dummies (representing small and medium firms with less than 100 employees and large firms with 100 or more employees) to capture differences in production technology across different size firms.

(13.) One may reasonably argue that firms belonging to different industries in the sample will use different imported capital that embodies a different "kind of technology. To capture this heterogeneity across the industries, we also interacted the industry dummies with the import variable, but the coefficients on the interacted terms were neither individually nor jointly significant. The results are available upon request.

(14.) As shown by two recent influential papers, Katayama, Lu, and Tybout (2009) and DeLoecker (2006), measured plant-level productivity may confound true plant-level productivity and differences in mark-ups across plants. The data on unit prices are not available, making it impossible to tackle this issue at this point. Using concentration ratios does not change the results (Amiti and Konings 2007).

(15.) We also interacted the import variable with the capital input, the foreign ownership, and export variables, but the magnitude and the significance of the variables of interest were similar to those obtained without these interactions.

(16.) One may also expect the firms with high productivity to hire more skilled labor. An omitted variable may result in increases in skilled labor and productivity among importing firms. Thus, we also used two sets of instruments for the skilled labor share to check the robustness of the results. First, we used whether or not the firms have a contractual relationship with universities or research institutions as an instrument for the skilled labor share. We alternatively used the average levels of the skilled labor share for other firms (j [not equal to] i) in a firm's location category for each industry. The results are consistent with those presented in this article, but the first instrument is not strong enough.

(17.) Another way of controlling for the endogeneity is to jointly estimate an import status equation with the production function (Clerides, Lach, and Tybout 1998: Van Biesebroeck 2005). We tried this also and obtained consistent results. However, as we are also interested in the interaction of the import variable with the skilled labor share, we do not report the results from this approach.

(18.) See Cameron and Trivedi (2005) for more detailed information.

(19.) To evaluate the correlation between the instruments and the error term, we use the Sargan test of overidentifying restrictions. The null hypothesis is that the instruments and the error terms are independent, so failure to reject this hypothesis validates the use of the instruments.

(20.) Similar instruments were used by Van Biesebroeck (2005) to examine the relationship between exports and productivity for the sub Saharan African manufacturing firms.

(21.) It is possible that any of these instruments may affect firm productivity not only through import and its interaction with the skill level of the firm but also through some other channels. Thus, to check the sensitivity of the results to the specification of the instruments, we alternatively used different sets of these four instruments, which produce similar results. We alternatively added the averages of the import variable for other firms (j [not equal to] i) in a firm's location category for each industry and its interaction with the skilled labor share to the instrument sets. The results are consistent with those presented in this article.

(22.) To assess whether our instruments are sufficiently related to the endogenous variable, we use an F-test of the joint significance of the instruments in reduced form.

(23.) We estimate two reduced-form equations: one for the imported capital input variable and one for the interaction of the imported capital input with the skill share variable. We then obtain the fitted values for imported capital input and its interaction with the skilled labor share to form estimates of these two reduced-form equations, replace the imported capital input and its interaction with skilled labor share in the original model by its fitted values, and estimate the resulting model by least squares.

(24.) See Hansen (2000) and Caner and Hansen (2004) for details.

(25.) Papageorgiou (2002) and Khoury and Savvides (2006) use this method to examine the relationship between openness to trade and growth.

(26.) Because of convergence issues the region and industry dummies are not included in the base model. We, however, included the region and industry dummies that were not highly collinear with other variables in the threshold model that allows an endogenous regressor. The resulting threshold parameter estimate was 0.385, with a confidence interval of [0.380, 0.413]. Thus, including these dummy variables does not change the threshold parameter, but results in a narrower confidence interval.

(27.) We estimated the threshold models using the modified Gauss programs written by Bruce Hansen, which are available at his website http://www.ssc.wisc.edu/ ~bhansen/progs/progs.htm.

(28.) We also implemented a recently developed "heterogeneous treatment effects" model, which allows for the treatment to vary with the values of explanatory variables (see Wooldridge 2002, chapter 18). We estimated our model by interacting the import variable with the demeaned skilled labor share to allow for the treatment effect to vary with the skilled labor share. We used the predicted values from a first stage logit model and the interaction of these predicted values with the demeaned skilled labor share as instruments in the second stage. The heterogenous effects are captured by the coefficient on the demeaned interaction term. The results are consistent with those presented in this article; the coefficient on the interaction term, the heterogenous effects, was statistically significant at a 0.05 significance level, but the coefficient on the import variable was insignificant.

(29.) None of the variables had VIFs greater than 10 and tolerance levels less than 0.1, which are generally accepted thresholds to identify multicollinearity problems. Thus, multicollinearity tests based on variance inflation factors (VIFs) and tolerance levels (presented in Table 2) illustrate that relying on the variables in Equation (1)is justified.

(30.) The statistically significant estimates are identified by asterisks: three asterisks mean statistical significance at the I% level, two at the 5% level, and one at the 10% level.

(31.) For instance, 50th percentile indicates that 50% of the firms had a skilled labor share of 23. 1% or less and 50% of the firms in the sample had a skilled labor share equal to or higher than 23.1% of the total employment.

(32.) Originally, we also included the exports and foreign direct investment as explanatory variables in our model but it did not change the results discussed here. The parameter estimate for exporting was not statistically significant, but the coefficient on the foreign direct investment variable indicated that the firms with foreign shares are more productive.

(33.) One would expect that there are multiple thresholds for the productive impact of importing across the skilled labor share distribution. Thus, we searched for another sample split in the sample of firms with skilled labor share above the first threshold. The second sample split produced an insignificant p-value of 0.194 using 1,000 replications, indicating that there is no significant sample split. The statistically insignificant threshold corresponds to a skilled labor share of [??] = 0.540 with 95% confidence interval [0.444, 0.559].

(34.) The firms below and above the threshold value have similar observable characteristics, indicating that the difference in the outcome variable can be considered attributable to importing and different level of skilled labor.

MAHMUT YASAR *

* I am indebted to the co-editor and an anonymous referee for their constructive and useful suggestions, I wish to acknowledge helpful comments by Gary Ferrier, Takao Kato, Fabio Mendel Kaz Miyagiwa, Catherine J. Morrison Paul, Jigna Sampat, Zhizhong Shan, seminar participants at the University of Arkansas, and participants at the International Industrial Organization conference, North American Productivity Workshop, and Southern Economic Association meetings on the earlier version of the article. I also would like to thank the World Bank Enterprise Surveys Staff for helpful discussions about the data.

Yasar: Assistant Professor of Economics, Department of Economics, University of Texas, Arlington, 701 S. West Street, Arlington, TX 76019. Phone 817-272-3290, Fax 817-272-3145, E-mail myasar@uta.edu;

Adjunct Assistant Professor of Economics, Emory University, Department of Economics. Atlanta GA 30322.

TABLE 1
Descriptive Statistics (N = 1,161)

 Standard
Variable Mean Deviation

Log value added of firm (In Y) 8.760 2.239
Log capital input (In K) 9.106 2.249
Log labor input (In L) 5.185 1.360
Whether or not firm imported any machinery
 (IMP) (a) 0.334 0.472
Share of managers, engineers, and technical
 workers in total employment (SH) (b) 0.270 0.171
Age of the firm (AGE) (c) 16.310 13.753
Capacity utilization (CU) (d) 72.112 24.146
Whether or not firm's products have been
 certified by IS0900 certification (ISO) (e) 0.508 0.500
Whether or not the firm is a subsidiary or a
 joint venture of a multinational firm (FO)
 (f) 0.115 0.320
Whether or not the firm is an exporter in the
 previous year (EXP) (g) 0.242 0.429
Whether or not the general manager is from an
 industrialized country (MN) (h) 0.396 0.195
Percentage of firms owned by the state/
 provincial government (SO) (i) 0.043 0.196

(a) In the survey, managers were asked whether they imported any
machinery. We use a dummy variable that is equal to I if firms
imported any machinery, and 0 otherwise.

(b) Share of managerial, engineering, and technical workers in total
employment.

(c) Managers were asked in what year their firm began operations.

(d) Capacity utilization is the amount of output actually produced
relative to the maximum amount that could be produced with existing
machinery and equipment and regular shifts.

(e) Managers were asked whether the firm's products have been
certified by IS0900 certification. We use a dummy variable that is
equal to 1 if the firm has a certified product, and 0 otherwise.

(f) Managers were asked whether the firm is a subsidiary or ajoint
venture of a multinational firm. We use a dummy variable that is
equal to 1 if firms have a foreign partner, and 0 otherwise.

(g) Mnnagers were asked what percent of sales are exported. We use a
dummy variable that is equal to I if firms reported positive shares,
and 0 otherwise.

(h) In the survey, managers were asked about the nationality of the
general manager. We use a dummy variable that is equal to I if firms
have a general manager who is from an industrialized country, and 0
otherwise.

(i) Managers were asked what percent of the firm is owned by the state
or provincial government.

TABLE 2
Tests of Multicollinearity: Variance Inflation
Factors (VIF) and Tolerance

Variable VIF Tolerance

In K 3.510 0.285
In L 4.900 0.204
IMP 4.390 0.228
SH 1.860 0.538
IMP x SH 4.250 0.235
AGE 4.000 0.250
[AGE.sup.2] 3.740 0.268
CU 1.180 0.849
ISO 1.470 0.682
Mean VIF 3.420

Note: The "rule of thumb" in the econometric literature
is that a VIF > 10 or a tolerance level < 0.1 is a sign of
a severe multicollinearity problem. Age variable is mean
adjusted.

TABLE 3
Parameter Estimates: Dependent
Variable = Natural Log of Value Added

Independent
Variables OLS Estimates IV Estimates

IMP -0.006 (0.146) 0.051 (0.283)
SH 0.918 (0.262) *** 0.721 (0.294) **
IMP x SH 0.885 (0.415) ** 1.449 (0.577) **
In K 0.355 (0.027) *** 0.3336 (0.033) ***
In L 0.706 (0.053) *** 0.708 (0.054) ***
AGE -0.022 (0.011) * -0.021 (0.011) *
[AGE.sup.2] 0.000 (0.000) 0.000 (0.000)
CU 0.015 (0.001) *** 0.015 (0.001) ***
ISO 0.278 (0.079) *** 0.258 (0.083) ***
Number of 1,161 1,161
 observations
[R.sup.2] 0.759 0.758
Sargan test (p-value) 0.406

Notes: Robust standard errors in parentheses.

* Significant at the 10% level, ** significant at the 5%
level, *** significant at the 1% level. The standard errors
clustered by industry and region are much smaller.
The regression run in this table includes dummy variables
that control for size, industry, and region characteristics.
However, they are not reported here in the interest of
space. They are available from the author upon request.

TABLE 4
Skilled Labor Share and Productive Impact Differential between
Importers and Non-importers

 Productive Impact Productive Impact
Percentiles of Share of Managers, Differential: Differential:
Engineers and Technicians in Between Importers Between Importers
Total Employment and Non- and Non-
 importers (Based importers (Based
 on OLS Results) on IV Results)

1th percentile of SH = 0.036 0.026 (0.134) 0.103 (0.271)
10th percentile of SH = 0.092 0.076 (0.117) 0.184 (0.256)
20th percentile of SH = 0.132 0.111 (0.107) 0.243 (0.248)
30th percentile of SH = 0.169 0.144 (0.098) 0.296 (0.241)
40th percentile of SH = 0.200 0.171 (0.093) * 0.341 (0.237)
50th percentile of SH = 0.231 0.199 (0.089) *** 0.387 (0.235) *
60th percentile of SH = 0.271 0.234 (0.086) *** 0.444 (0.233) *
70th percentile of SH = 0.310 0.269 (0.087) *** 0.501 (0.234) **
80th percentile of SH = 0.378 0.329 (0.095) *** 0.599 (0.241) ***
90th percentile of SH = 0.500 0.437 (0.125) *** 0.776 (0.267) ***

Note: Robust standard errors in parentheses, which are obtained by
using the Delta Method.

* Significant at the 10% level, ** significant at the 5% level, ***
significant at the 1% level.

TABLE 5
IV Threshold Model Statistics

Statistics TR Model IV TR Model 2

Threshold value (y) 0.427 0.380
95% confidence interval--White [0.289, 0.466]
 correction for
 heteroskedasticity
Confidence interval-- [0.289, 0.443]
 heteroskedasticity is
 corrected by using quadratic
 variance estimate
Confidence interval-- [0.289, 0.444]
 heteroskedasticity is
 corrected by using
 nonparametric kernel
LM-test for no threshold 24.512
LM-test for no threshold 0.041
 (p-value)
[[??].sub.1] 0.162 (0.093) *
[[??].sub.2] 0.499 (0.190) ***
Number of observations 1,161 1,161
Number of observations with a 996 936
 SH < y
Number of observations with a 165 225
 SH > y

* Significant at the 10% level, ** significant at the 5% level, ***
significant at the I%n level.