Does offshoring pay? Firm-level evidence from Japan.

Hijzen, Alexander ; Inui, Tomohiko ; Todo, Yasuyuki 等

I. INTRODUCTION

The rise in offshoring--as reflected by the rising importance of imported intermediate inputs in domestic production--has been an important factor behind the growth in world trade (Yeats 1998; Yi 2003). East Asia is not an exception to the rise in offshoring: The growing geographical specialization along the value chain has given rise to the development of sophisticated production sharing arrangements within East Asia (Fukao, Ishido, and Ito 2003; Ng and Yeats 1999). In particular, Japanese firms have increasingly taken advantage of the business opportunities provided through offshoring of production activities to other East Asian countries (Kimura and Ando 2005).

Given the importance of these developments, understanding implications of offshoring should be of significant interest to academics and the policy-making community. However, most research so far has concentrated on the potentially adverse labor market aspects of offshoring in developed countries (Feenstra and Hanson 1996, 1999; Head and Ries 2002; Hijzen, Gorg, and Hine 2005), and much less attention has been directed toward understanding the benefits of the offshoring phenomenon. However, firms may benefit from offshoring through the improvement of the productivity of primary factors of domestic production by allowing firms to specialize in activities they perform relatively well.

For our analysis of the impact of offshoring on productivity, we make use of firm-level data for the Japanese manufacturing sector during the period 1994-2000. One great advantage of our data set is that it comprises information on the value of purchases of products, parts, and components from foreign providers so that we can construct a direct measure of materials offshoring. (1) This measure includes both sourcing to the firm's foreign affiliates and sourcing to other unaffiliated foreign firms. We refer to the combination of the two types of sourcing of intermediate inputs as "overall offshoring." In addition, we have data on the amount of purchases from the firm's foreign affiliates, which provides us with a proxy for the extent of international sourcing within the firm, or "intrafirm offshoring." (2) In contrast, we refer to sourcing from unaffiliated foreign firms as "arm's-length offshoring." (3) By including both measures simultaneously, we can infer to what extent the organizational form of offshoring, intrafirm or arm's length, influences productivity of the sourcing firm. Finally, we also consider the effects of sourcing to domestic providers, which we refer to as "domestic sourcing." Figure 1 provides a schematic overview of the various types of sourcing mentioned above.

Several previous studies have analyzed the relation between offshoring and productivity using industry-level data. For the measurement of offshoring, such studies typically rely on input-output data. Egger and Egger (2006) analyze how offshoring affects the productivity of low-skilled workers employed in the European Union manufacturing sector. They find that the rise in offshoring accounted for 6% of the increase in value added per worker during the period 1992-1997. Amiti and Wei (2006) analyze the productivity effects of materials and services offshoring on the productivity of U.S. industries. They find that both materials and services off shoring have a positive effect on productivity but that the positive effect of services offshoring is considerably larger, accounting for about 11% of productivity growth during the sample period compared to 5% for materials offshoring.

[FIGURE 1 OMITTED]

Gorg and Hanley (2005) and Gorg, Hanley, and Strobl (2008) were the first to analyze the impact of offshoring on productivity using firm-level data. The main advantage of using firm-level data is, no doubt, that they allow one to control for firm heterogeneity. Using data for Ireland, they find that both materials and services offshoring benefit firm productivity but that the benefits only accrue to multinationals and exporters. Criscuolo and Leaver (2005) who focus exclusively on services offshoring also find a positive impact on productivity using data for the United Kingdom. (4)

This study contributes to the existing literature in the following three ways. First, thanks to the richness of our data set, we can distinguish between the impact of intrafirm and arm's-length offshoring by incorporating a measure for overall offshoring and another one specifically for intrafirm offshoring. In addition, our data set allows us to examine possible differences in the impact on productivity between offshoring and domestic sourcing. (5) Finally, to the best of our knowledge, this study is the first to explore the impact of offshoring on productivity in the context of Japan. (6)

To preview our results, we find that intrafirm offshoring has generally a positive effect on total factor productivity (TFP) at the firm level. We control for the possible endogeneity of offshoring due to reverse causality by employing the system generalized method of moments (GMM) estimation developed by Blundell and Bond (1998). The results suggest that the median firm that engages in intrafirm offshoring gains an annual TFP growth rate that is 0.6 percentage point higher than that of nonoffshoring firms. In contrast to intrafirm offshoring, either arm's-length offshoring or domestic sourcing does not generally affect firm-level productivity. Interestingly, arm's-length offshoring has a negative impact on the productivity of nonmultinationals and nonexporters, while its impact on the productivity of multinationals and exporters is nonnegative. This may indicate that sourcing of intermediate inputs to unaffiliated foreign firms tends to be inefficient particularly due to presence of nonnegligible costs of searching foreign firms that meet the needs of the offshoring firm. This seems particularly important for firms with limited international experience. This is consistent with previous findings by Gorg and Hanley (2005) and Gorg, Hanley, and Strobl (2008).

The remainder of this article is structured as follows. Section II discusses the theoretical framework, whereas Section III explains empirical methodology. Section IV describes the data and provides some descriptive statistics on offshoring. In Section V, we discuss the estimation results. Finally, Section VI concludes.

II. THEORETICAL FRAMEWORK

Offshoring may affect firm-level productivity mainly because it allows firms to benefit from static and dynamic gains from specialization. Consider a developed country firm whose production process is characterized by multiple stages of various skill intensity. Offshoring labor-intensive (or less skill intensive) stages allows the firm to make a more efficient use of production factors that remain in employment and thus increase the firm's productivity. Moreover, the gains from specializing in skill-intensive stages of production process may be dynamic rather than static. Young (1991), for example, suggests that productivity growth in more sophisticated skill-intensive production stages may be higher than in less skill-intensive stages because of the potential of productivity growth through learning by doing in such activities. Since the more standardized, less skill-intensive activities are most suitable to offshoring, one would expect that specializing in skill-intensive production through offshoring generates higher growth in productivity thanks to the increasing importance of learning-by-doing effects. Therefore, offshoring may be expected to improve both the level and the growth of productivity within firms.

The size of these benefits from offshoring, however, is likely to depend on the way supplier networks are organized. For example, a firm that offshores production to its own affiliate abroad may find it easier to control product quality than firms that offshore to unaffiliated firms abroad. If this is indeed the case, intrafirm offshoring should be more productivity enhancing than arm's-length offshoring. Similarly, the productivity effects associated with international production networks may be more pronounced than those associated with domestic supplier networks as skill and factor differentials tend to be larger internationally.

In addition to the way supplier networks are organized, the benefits from offshoring may also depend on characteristics of sourcing firms. In particular, as Gorg and Hanley (2005) and Gorg, Hanley, and Strobl (2008) argue, the benefits from offshoring depend on a firm's experience in foreign markets since this may help to reduce the search costs associated with identifying suitable suppliers abroad. Hence, they posit that exporting and multinational firms benefit more from offshoring than firms that have no overseas experience, as they are likely to find suitable foreign firms more easily.

So far, we have discussed causality from offshoring to productivity. However, firm-level productivity may also affect the decision to offshore as well as the way offshoring is organized. Antras and Helpman (2004) show theoretically in a model with incomplete contracts and firm heterogeneity building on the work by Antras (2003) and Melitz (2003) that a firm's organizational structure arises endogenously as a function of firm-level productivity and the (industry specific) intensity of intermediate inputs in production. They show that as a result of the larger fixed costs associated with offshoring, only the most productive firms tend to engage in offshoring, while less productive firms tend to outsource to domestic firms. In addition, in industries where intermediate inputs are less important, the most productive firms tend to engage in intrafirm offshoring rather than arm's-length offshoring in an effort to counteract hold-up problems. The finding of Tomiura (2007) using firm-level data from Japan that firms that engage in arm's-length offshoring tend to be less productive than firms that engage in intrafirm offshoring is consistent with these theoretical predictions.

Therefore, the relationship between productivity and offshoring may work in two ways, suggesting that we have to carefully distinguish between causality from off shoring to productivity, which is the focus of this article, and causality in the reverse direction, the focus in Tomiura (2007). The next section describes how we deal with this distinction in our estimation.

III. EMPIRICAL METHODOLOGY

In line with recent production function studies such as those by Aghion et al. (2004), we adopt a two-step estimation procedure in which we first derive a TFP measure and then estimate the effect of offshoring on TFP. In this section, we first describe how we construct our various TFP measures, then derive the estimation equation and finally discuss our estimation method.

A. Measures of TFP

To analyze the impact of offshoring on firm productivity, we define two different measures of TFP. We use two measures to ensure that our results do not rely on the choice of TFP measure. First, we employ the chained multilateral index of firm-level TFP based on the methodology in Caves, Christensen, and Diewert (1982) and Good, Nadiri, and Sickles (1996). This index is defined as follows:

(1) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where subscripts i and t represent firm i and year t, respectively. Y refers to value added, K to capital, L to labor, and [s.sub.ijt] is the cost share of factor J for firm i in year t. [bar.ln [Y.sub.t]], [bar.ln [J.sub.t]], and [bar.[s.sub.Jt]] are the arithmetic means of In [Y.sub.it], [ln.sub.Jit], and [s.sub.iJt], respectively, across all i in the same three-digit industry in year t. (7) Equation (1) implies that the multilateral TFP index, [TFP.sub.IN.sub.it], measures firm i's TFP level in year t relative to the TFP level of the hypothetical firm in year 0 whose input shares are equal to the arithmetic mean of input shares and whose output and input quantities are equal to the geometric mean of output and input quantities, respectively. This type of multilateral TFP index is used, for example, in Aghion et al. (2004).

Second, we derive a regression-based measure of firm-level TFP through the estimation of a standard Cobb-Douglas production function:

(2) In [TFP.sub.it.sup.OP] = ln [Y.sub.it] - [[??].sub.K]ln [K.sub.it] - [[??].sub.L]ln [L.sub.it],

where [[??].sub.K] and [[??].sub.t] represent the estimated capital and labor elasticity, respectively. To allow for differences in the production technology across industries, we divide manufacturing firms into six industries (8) and estimate [[??].sub.K] and [[??].sub.L] for each industry. More specifically, to estimate [[??].sub.K] and [[??].sub.L], we take account of the potential correlation between factor inputs and productivity generated from the fact that contemporaneous productivity affects the choice of factor inputs and the choice on firms' exit. Several procedures have been proposed that allow one to do this including that by Olley and Pakes (1996) and its extensions by Levinsohn and Petrin (2003) and Buettner (2003). This article uses Olley and Pakes (1996) method since this is more widely used, for example, in Javorcik (2004), than its extensions, probably due to requirements of strong assumptions in other methods. (9) Table 1 reports the results of estimating Equation (2) using both standard ordinary least squares (OLS) and the OlleyPakes procedure. In principle, one would expect the Olley-Pakes estimate of the elasticity for labor to be smaller than the OLS estimate due to the positive correlation between labor inputs and unobserved productivity shocks, whereas an opposite pattern is expected in the case of the elasticity of capital. Table 1 indicates that these theoretical predictions are generally supported by the evidence, justifying our use of the Olley-Pakes measure of TFP. (10)

It is not obvious which of our two measures of TFP should be preferred. An advantage of the multilateral TFP index given by Equation (1) is that we do not need to assume a specific functional form of the production function, while its drawback is that we have to assume perfect competition and constant returns to scale. In contrast, a major benefit of the regression-based TFP measure obtained from the Olley-Pakes method is that we do not need to assume constant returns to scale. Its main shortcoming lies in assuming a Cobb-Douglas production function. Therefore, these two measures of TFP can be viewed as complements.

Although these TFP measures are widely used for productivity analysis, we should note a shortcoming of these measures in our analysis, which stems from the fact that we deflate nominal values of outputs and inputs by industry-level average prices to compute their real values due to lack of information on prices for each firm. (11) Because of this, both of our TFP measures may capture markups and demand shocks in addition to true productivity. This issue is discussed in Klette and Griliches (1996), Katayama, Lu, and Tybout (2003), and De Loecker (2007) among many others. In the context of this article, if off shoring firms can purchase components from foreign firms at lower than average industry prices, the TFP measures can improve simply due to the price effect without any improvement in the true productivity. While a number of solutions to this problem have been suggested in the existing literature, we do not employ any of these in this article for two reasons: (a) because these solutions typically require strong assumptions on demand or (b) because additional information on product prices is needed, which is not available in our data set. However, as we will argue in Baseline Results section, biases due to this price effect are unlikely to be large in the present article.

B. Estimation Equation

Based on the argument in Section II, we hypothesize that the extent of offshoring has a positive effect on the growth of firm-level TFP. Since our data set includes data for both overall offshoring and intrafirm offshoring, we incorporate both in our estimation equation of TFP growth. We also include domestic sourcing and, following Griliches (1980), research and development (R&D) intensity as possible determinants of TFP growth. The lagged value of TFP is further included to allow for convergence in TFP levels. Thus, our estimation equation of TFP growth is given by:

(3) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where [TFP.sub.it] is one of the two measures of TFP discussed in the previous subsection. [OFFl.sub.i,t-1], [OFF2.sub.i,t-1], and [DOM.sub.i,t-1] represent the extent of overall offshoring, intrafirm offshoring, and domestic sourcing, respectively, for firm i in year t - 1. (12) R&[D.sub.i,t-1] is the R&D intensity for firm i defined as the ratio of R&D expenditure to value added. We use first lags of these variables since we assume a 1-yr time lag between sourcing and R&D activities and productivity growth. Subscript j denotes the industry affiliation of firm i so that [[alpha].sub.jt] represents industry-year-specific fixed effects and [[epsilon].sub.it], it is the error term.

Since our theoretical argument suggests that offshoring may affect both the level and the growth of TFP, we rearrange Equation (3) so that the dependent variable is the TFP level, In [TFP.sub.it]:

(4) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

This rearrangement enables us to interpret the estimated coefficients on offshoring and domestic sourcing as their effects on either level or growth of TFP. Moreover, this specification allows one to more readily compare the present results with those reported in previous studies such as those by Gorg and Hanley (2005), Gorg, Hanley, and Strobl (2008), and Amiti and Wei (2006), who also employ a level regression similar to that of Equation (4). Note that, in contrast to these existing studies, we explicitly control for the role of the R&D activities in TFP growth. Failing to do so may lead to omitted variable bias, when the decision to offshore and expenditure on R&D are correlated.

More specifically, the measure of overall offshoring, OFF1, is represented by the ratio of real purchases of products, parts, and components to foreign providers to real value added of the firm. Real purchases of products, parts, and components to foreign providers are defined as their nominal value, which can be directly obtained from each firm's survey response, deflated by the price level of intermediate goods at the three-digit industry level. Similarly, real value added is constructed from nominal sales deflated by the three-digit industry price level of outputs less nominal intermediate inputs deflated by the industry price level of intermediate goods. Since the survey question on offshoring does not distinguish between purchases from unaffiliated and affiliated firms, this measure embodies both intrafirm and arm's-length offshoring.

In addition, OFF2, the measure of intrafirm offshoring, in particular, is defined as the ratio of nominal purchases from the firm's own foreign affiliates deflated by the industry price level of intermediate goods to real value added. Unfortunately, the survey question refers to all purchases from foreign affiliates including purchases of products, parts, and components, which are included in OFF1 and related to offshoring, as well as raw materials, which are not included in OFF 1 and less likely to be related to offshoring. Therefore, OFF2 may be considered a "broad" measure of intrafirm offshoring, whereas OFF1 constitutes necessarily a "narrow" measure of overall offshoring. Nevertheless, by including both measures simultaneously, we can get some idea to what extent the organizational form of offshoring, intrafirm, or arm's length influences the firm's productivity. (13)

The intensity of domestic sourcing, DOM, is represented by the ratio of the nominal purchases of products, parts, and components to domestic providers, taken directly from the data set, deflated by the industry price level of intermediate goods to real value added. (14) Table 2 summarizes the definition of the key variables used in the present analysis.

C Estimation Method

An econometric concern that needs to be addressed when estimating Equation (4) is the endogeneity of regressors. Specifically, estimation will be biased if firms decide to self-select into off shoring on the basis of productivity differences across firms, as suggested by Antras and Helpman (2004) (see Section II for a more detailed discussion). Since we start from the growth Equation (3), which can be obtained from first differencing a TFP-level equation, the error term in Equation (4), [[epsilon].sub.it], is in fact a first-differenced error term in the TFP-level equation. Since the offshoring measures are first lagged, we would expect the coefficients on offshoring to be downward biased when more productive firms are more likely to engage in offshoring, that is, when there is positive contemporaneous correlation between the level of productivity and offshoring. The same applies to our domestic sourcing and R&D variables.

Therefore, we employ the system GMM estimation developed by Blundell and Bond (1998) to correct for the possible endogeneity of any of our right-hand side variables in Equation (4) and to eliminate firm-specific fixed effects. In the system GMM estimation, we apply GMM estimation to the system of Equation (4) and its first difference in which firm-specific fixed effects are eliminated, using the lagged first-differenced regressors as instruments for Equation (4) and the lagged regressors as instruments for the first-differenced equation. More specifically, we use as instruments the second and third lags of endogenous regressors (15) for the first-differenced equation and their second 1 st-differenced lags for the level equation. We employ one-step GMM using robust standard errors.

IV. DATA DESCRIPTION AND SUMMARY STATISTICS

The data employed in this article are drawn from Kigyo Katsudo Kihon Chosa (Basic Survey of Enterprise Activities), which is conducted annually by the Ministry of Economy, Trade and Industry. This data set covers all firms with more than 50 employees and 30 million yen of assets in manufacturing, mining, and commerce industries. Participation in the survey for those firms is compulsory. The survey was first conducted in 1991 and then annually from 1994 onward. We restrict ourselves to manufacturing firms during the period 1994-2000 since for more recent years, no information on domestic or international sourcing is available. (16)

Figure 2 provides time trends in the measures of offshoring and domestic sourcing. During the period 1994-1999, the average overall offshoring intensity (OFF1) rises from 1.2% to 1.8% in terms of value added, whereas the average intrafirm offshoring intensity (OFF2) rises from 3.4% to 5.3%. In contrast to the increasing trend in the overall and intrafirm offshoring intensity, the trend in the domestic sourcing intensity (DOM) is negative. (17)

[FIGURE 2 OMITTED]

It is worth noting that the intrafirm offshoring intensity, OFF2, in our data set is greater on average than the overall offshoring intensity, OFF1. This may be surprising as one would expect intrafirm of f shoring to be a subset of overall offshoring. The fact that this is not the case in practice reflects differences in the definition between the two variables mentioned earlier. OFF2 consists of all purchases from the firm's foreign affiliates, including purchases of raw material that may not be related to offshoring, whereas OFFI includes only purchases of products, parts, and components from foreign providers, which is directly related to offshoring. In other words, OFF2 is based on a substantially broader definition of offshoring than OFF1.

Figure 3 represents the offshoring intensity across different industries showing significant industry variation. According to the overall offshoring measure, the apparel and leather industries appear to be the most active offshoring industries in Japan. Both industries are relatively intensive users of unskilled labor and well-known examples of offshoring industries. The presence of large foreign-home wage differentials are likely to play an important role in explaining the offshoring decisions in these two industries. The electrical machinery and electronics industry and the medical, precision, and optical instruments industry are also industries where offshoring is important. While production in these industries may be characterized as fairly high tech overall, the production process also consists of less skill-intensive activities.

[FIGURE 3 OMITTED]

Figure 3 also shows that industries with large overall offshoring intensity tend to exhibit large intrafirm offshoring intensity as well. There are, however, some exceptions. Most notably, the coke and petroleum products industry shows an extremely large value of OFF2, while showing a value of OFF1 close to zero. This of course reflects the fact that OFF2 includes raw materials, whereas OFF 1 does not.

We present summary statistics for the regressand and regressors as well as TFP growth in Table 3. Panel A provides summary statistics for the full sample, while Panel B provides a comparison of the means for the same variables between firms that offshore and those that do not offshore. We are particularly interested in the relationship between the offshoring and the level and growth of TFP. The descriptive statistics indicate that firms that engage in overall offshoring or intrafirm offshoring exhibit a higher productivity level and faster productivity growth than nonoffshoring firms. (18) Offshoring firms also are more active on average in domestic sourcing and R&D than other firms. In our formal econometric analysis, we will now examine whether off shoring in fact leads to higher productivity, controlling for other possible factors and endogeneity due to reverse causality.

V. RESULTS

A. Baseline Results

Table 4 presents the baseline results on the impact of off shoring on TFP based on estimations using the multilateral index of TFP. Note that in these regressions, we can interpret the coefficient of the regressors as their impact either on the TFP level or on the TFP growth (not just the level) because the independent variables include the first lag of the dependent variable.

To provide a benchmark, we first discuss the results based on standard OLS reported in columns (1)-(3) of Table 4 before going to the results based on system GMM. The results in columns (1) and (2) show that both overall offshoring and intrafirm off shoring have a positive and significant effect on TFP when included separately, but the effect of overall offshoring is only significant at the 10% level when the two variables are included jointly (column (3)). (19) Domestic sourcing and R&D have a significant and positive effect on TFP at least at the 10% level in all the three specifications. However, these results may be biased when offshoring decisions are positively correlated with the productivity level, for example, when firms self-select into offshoring. Indeed, the finding that overall offshoring has a positive effect on productivity and its coefficient is also larger than that on domestic sourcing is consistent with the theoretical predictions of Antras and Helpman (2004) and the empirical findings of Tomiura (2007): the most productive firms engage in offshoring, less productive firms tend to source domestically and the least productive firms do not make use of supplier networks of any kind.

To examine whether the correlation between offshoring variables and productivity found in the OLS regressions reflects the effect of offshoring on productivity or the other way around, we reestimate our model employing the system GMM. The results are reported in columns (4)-(6) of Table 4. In all specifications, the Hansen J statistic and the ArrellanoBond statistic presented in the last two rows suggest that the instruments are orthogonal to the error term and that there is no serial correlation in the error term.

The system GMM estimations point at statistically significant effects of intrafirm offshoring, OFF2, on TFP (columns (5) and (6)). These effects are larger than the results from the OLS estimations, suggesting that the OLS results are downward biased due to a positive contemporaneous correlation between offshoring and unobserved productivity shocks and accordingly a negative correlation between offshoring and first-differenced productivity shocks. (20) More specifically, the GMM results suggest that a 1 percentage point increase in the intrafirm offshoring intensity raises TFP by 0.12%. This suggests that for the median offshoring firm, for which OFF2 equals 0.052, (21) TFP growth rate is 0.6 percentage point higher than when it had not engaged in off shoring, everything else equal. Thus, we conclude that the effect of intrafirm offshoring on TFP growth is quantitatively large.

In contrast to the positive impact of intrafirm offshoring, the measure of overall offshoring, OFF1, has no significant impact in any GMM estimation (column (4) or (6) of Table 4). Thus, we find that while intrafirm off shoring within a multinational raises the productivity, arm's-length offshoring does not have such impact in general, suggesting that sourcing of intermediate inputs to affiliated foreign firms is more effective than sourcing to unaffiliated foreign firms. This finding is consistent with previous findings by Gorg and Hanley (2005) and Gorg, Hanley, and Strobl (2008) that suggest that offshoring is productivity improving only for firms with experience in international markets through either exporting or foreign direct investment.

In addition, we find that the domestic sourcing intensity has a positive but insignificant impact on TFP in all GMM estimations. This result confirms our presumption that small differentials in skills and factor prices among domestic firms may lead to small, or even negligible, benefits from domestic sourcing.

In summary, we find that the multilateral TFP index is positively correlated with the extent of intrafirm offshoring but not with the extent of arm's-length offshoring or domestic sourcing. However, the results should be interpreted with caution since inputs are deflated by industry-level average prices, rather than firm-level prices, so that this TFP measure may overstate the true productivity level for offshoring firms (see the discussion in Measures of TFP section). The measurement error in TFP, given by the difference between the TFP measure and the true TFP level, is captured by the error term in Equation (4). Since offshoring is the source of and thus is correlated with the measurement error, the price effect will induce endogeneity bias in the off shoring variables. This certainly is a problem in the OLS estimations but may also affect the GMM results when the instruments are weak. However, the result that the effect of overall offshoring is positive and significant in the OLS estimations while it is insignificant in GMM suggests that our GMM estimations are quite effective in correcting for biases due to the price effect. Since the price effect should influence the OLS and GMM estimates on the two measures of offshoring, overall and intrafirm, to a similar extent, a positive and significant effect of intrafirm off shoring in both OLS and GMM estimations indicate that intrafirm offshoring has a positive effect on productivity after controlling for the price effect.

B. Robustness Checks

To check the robustness of the baseline results in the previous section, we employ the following three alternative specifications. First, we use Olley and Pakes (1996) regression-based measure of TFP instead of the multilateral TFP index used in the baseline regressions. (22) The OLS and GMM results are represented in columns (1) and (2) of Table 5. The p value of the Hansen J statistics for the GMM estimation is close to zero in the GMM estimation using the Olley-Pakes measure, implying that instruments are not orthogonal to the error term. (23) Although the Hansen test suggests that the GMM results using the Olley-Pakes measure are biased, the results on offshoring and domestic sourcing are qualitatively the same as and quantitatively very similar to the baseline results using the multilateral TFP index.

Second, we use ratios to sales, rather than value added, when we construct the offshoring, domestic sourcing, and R&D intensity variables as well as our multilateral TFP index. The OLS and GMM results from these modifications are presented in columns (3) and (4) of Table 5. These results are virtually the same as the baseline results except that thep value of the impact of intrafirm offshoring is barely above 5%.

Finally, we use the second-lagged regressors, rather than the first lagged in the baseline regressions, to account for possible time lags between sourcing activities and productivity improvement. The results shown in columns (5) and (6) of Table 5 indicate that intrafirm offshoring 2 yr before also has a positive and significant impact on the current productivity, although the impact is smaller in size than the impact of intrafirm offshoring using the first lag. This evidence suggests that the impact of off shoring diminishes over time. As in the baseline regressions, neither overall offshoring nor domestic sourcing has an impact on productivity.

In summary, all the three alternative estimations lead to results similar to those in our baseline specification, supporting our conclusion that intrafirm offshoring is productivity improving, while arm's-length offshoring and domestic sourcing are not.

C. Differences in the Size of the Effect of Offshoring across Firms

So far, we have estimated the average effect of offshoring on TFP, ignoring any possible variation in the size of effect across firms. However, the size of the offshoring effect may be expected to depend on the degree of firms' experience in international markets since the benefits from off shoring may vary in the level of the search costs associated with selecting foreign suppliers (Section II). Gorg, Hanley, and Strobl (2008) therefore split the sample between multinational enterprises (MNEs) and domestic firms and exporters and nonexporters and find that experience in foreign markets matters.

Following this idea, we split the sample into two subsamples in two different ways: MNEs and local firms (24) and exporting and nonexporting firms. (25) Table 6 reports the means of the key variables for the various subsamples. It can be seen from the table that there exist no major differences in terms of their offshoring profiles between exporters and nonexporters. However, we do observe, perhaps not surprisingly, that MNEs are more important offshorers than purely domestic firms. MNEs, after all, have access to an international production works, which have been established with the express purpose of offshoring or at the very least may be expected to facilitate offshoring arrangements.

To formally examine how the impact of offshoring on productivity depends on such firm characteristics, we augment Equation (4) with interaction terms between the respective offshoring measures and the certain firm characteristics. First, we estimate whether the effect of the overall off shoring measure differs between multinational and domestic firms using the interaction term based on a dummy variable for MNEs and the overall offshoring measure. (26) The GMM results presented in column (1) of Table 7 show that the coefficient on the interaction term is positive and significant at the 10% level, whereas the coefficient on the overall off shoring measure itself is negative and significant. In addition, the hypothesis that the sum of the coefficients on the overall offshoring variable and its interaction term with the MNE dummy is zero cannot be rejected. These results imply that the impact of arm's-length offshoring on TFP is zero for MNEs, (27) while it is negative for local firms probably due to large search costs. This is consistent with Gorg, Hanley, and Strobl (2008).

Second, we employ the interaction term between a dummy variable for exporting firms and each of the two measures of offshoring. Column (2) of Table 7 shows that the interaction term between the export dummy and the overall offshoring has a positive but insignificant effect, while the measure of overall offshoring is negatively correlated with TFP. The sum of the two variables is not statistically different from zero. These results provide further evidence that for firms without international experiences such as exporting, offshoring has a negative impact on productivity presumably due to the presence of nonnegligible search costs.

VI. CONCLUDING REMARKS

In the present article, we explore the impact of offshoring on firm productivity using firm-level data for the Japanese manufacturing industries during the period 1994-2000. We find that intrafirm offshoring, that is, sourcing of intermediate inputs to foreign affiliates within a particular multinational firm, has generally a positive effect on productivity of the offshoring firm, while arm's-length offshoring, that is, sourcing to unaffiliated foreign firms, does not have such an effect. These effects are robust to controlling for the possible endogeneity of offshoring due to reverse causality from productivity to off shoring. In addition, arm's-length offshoring has a negative impact on the productivity of non-MNEs and nonexporters, while its impact on the productivity of MNEs and exporters is nonnegative. These results suggest that the costs of searching foreign firms suitable for offshoring are nonnegligible.

Although our findings shed some light on the offshoring literature, we should note that our results need to be interpreted with care. First, offshoring may induce measurement error in the dependent variable through its impact on product prices thereby introducing a bias in our results. However, we believe that this bias does not have major impact on our results, as we discussed in Baseline Results section.

Second, we only allow the offshoring intensity (as well as the domestic sourcing and the R&D intensity) to shift the isoproduct curve and we do not allow for an effect of offshoring that leads to .the rotation of the isoproduct curve. In other words, we only focus on Hicks neutral productivity effects and disregard the role of offshoring as a channel for skill-biased technological changes as, for example, Feenstra and Hanson (1996) argue. The present empirical model may thus be considered as a short-run model in which factor shares are constant. (28)

Finally, our empirical specification only captures partial equilibrium effects and disregards general equilibrium effects. In the long run, however, general equilibrium effects are also likely to affect productivity, if, for example, individual offshoring decisions at the firm level are concentrated in certain sectors so as to induce sector-wide technological change. (29) Therefore, the results of the present article should be interpreted at the level of the individual firm and cannot straightforwardly be used to make inferences about the total effect of offshoring on the Japanese economy.

ABBREVIATIONS

GMM: Generalized Method of Moments

MNE: Multinational Enterprises

OLS: Ordinary Least Squares

R&D: Research and Development

TFP: Total Factor Productivity

doi: 10.1111/j.1465-7295.2008.00175.x

APPENDIX: CONSTRUCTION OF VARIABLES

This appendix provides supplementary information on the construction of our data set. (30) To construct data employed in the present analysis, we use firm-level data from Kigyo Katsudo Kihon Chosa (KKKC, Basic Survey of Enterprise Activities) and industry-level data from the Japan Industry Productivity (JIP) Database 2006. The JIP Database 2006 is constructed by the Firm- and Industry-Level Productivity Research Group organized in the Research Institute of Economy, Trade and Industry (RIETI) of Japan and headed by Kyoji Fukao and Tsutomu Miyagawa. The JIP Database 2006 includes various data during the period 1970-2002 at the three-digit industry level, including price deflators of output, intermediate inputs, and capital goods and input-output matrices. The complete database is available at the Web site of RIETI (http://www.rieti.go.jp).

Real sales is defined as nominal total sales reported in KKKC deflated by the output deflator at the three-digit level taken from the JIP Database. The nominal value of intermediate inputs is defined as the sum of costs of goods sold and general and administrative expense minus labor costs and the value of depreciation. The nominal value of intermediate inputs is deflated by the intermediate goods deflator also taken from the JIP Database to obtain the real value of intermediate inputs. Real value added is defined as real sales less the real value of intermediate inputs.

Firms' real capital stock represents the real value of the stock of tangible fixed assets excluding land since the book value of land may not reflect the true value of the land, in particular, if the land was purchased long time ago. However, the value of land owned by each firm is available only in the KKKC data for 1995 and 1996, although information on the total value of tangible fixed assets including land is available for all years. Therefore, we estimate the nominal value of tangible fixed assets excluding land of firm i in industry j in year t, Nom[K.sub.ijt], by multiplying the firm's total tangible assets including land by one minus industry j's average share of the land value in the total tangible fixed assets in 1995 and 1996. Then, we derive the real capital stock of firm i in industry j in year t, [K.sub.ijt], from Nom[K.sub.ijt], using the industry total of nominal tangible fixed assets excluding land, Nom[K.sub.jt] = [[SIGMA].sub.i[epsilon]j] Nom[K.sub.itt], and the estimated real value of the corresponding variable, [K.sub.jt], taken from the JIP Database: [K.sub.ijt] = Nom[K.sub.ijt] x [K.sub.jt]/ Nom[K.sub.jt]. [K.sub.jt], is obtained by the perpetual inventory method, using industry-level data on fixed capital formation during the period 1975-2000 and industry-level data on fixed assets in 1975.

Labor inputs are measured in the man-hour base. Since information on working hours for each firm is not available in KKKC, we use the industry average of working hours taken from the JIP Database. R&D expenditure of each parent firm is deflated by the industry price deflator of intermediate inputs.

We limit our sample to firms whose TFP level, R&D expenditure, the measure of offshoring, and the measure of domestic sourcing are available for at least five consecutive years during the 7-yr period 1994-2000. Then, to alleviate biases due to outliers, we drop firms whose R&D, offshoring, or domestic sourcing intensity is among the top 1%.

REFERENCES

Aghion, P., R. Blundell, R. Griffith, P. Howitt, and S. Prantl. "Entry and Productivity Growth: Evidence from Microlevel Panel Data." Journal of European Economic Association, 2, 2004, 265-376.

Amiti, M., and S. Wei. "Services Offshoring and Productivity: Evidence from the United States." NBER Working Paper No. 11926, 2006.

Antras, P. "Firms, Contracts, and Trade Structure." Quarterly Journal of Economics, 118, 2003, 1375-418.

Antras, P., and E. Helpman. "Global Sourcing." Journal of Political Economy, 112, 2004, 552 80.

Blundell, R., and S. Bond. "Initial Conditions and Moment Restrictions in Dynamic Panel Data Models." Journal of Econometrics, 87, 1998, 115-43.

Buettner, T. "R&D and the Dynamics of Productivity." Unpublished, London School of Economics, 2003.

Caves, D. W., L. R. Christensen, and E. W. Diewert. "Multilateral Comparisons of Output, Input, and Productivity Using Superlative Index Numbers." Economic Journal, 92, 1982, 73-86.

Criscuolo, C., and M. Leaver. "Offshore Outsourcing and Productivity," Presented at OECD Conference on Globalisation of Production, November 14-15, 2005.

De Loecker, J. "Product Differentiation, Multi-Product Firms and Estimating the Impact of Trade Liberalization on Productivity." NBER Working Paper No. 13155, 2007.

Egger, P., and H. Egger. "International Outsourcing and the Productivity of Low-skilled Labor in the EU." Economic Inquiry, 44, 2006, 98-108.

Feenstra, R. C., and G. H. Hanson. "Globalization, Outsourcing, and Wage Inequality." American Economic Review, 86, 1996, 240-45.

Feenstra, R. C., and G. H. Hanson. "The Impact of Outsourcing and High-Technology Capital on Wages: Estimates for the United States, 1979-1990." Quarterly Journal of Economics, 114, 1999, 907-41.

Fukao, K., H. Ishido, and K. Ito. "Vertical Intra-Industry Trade and Foreign Direct Investment in East Asia." Journal of the Japanese and International Economies, 17, 2003, 468-506.

Good, D. H., M. I. Nadiri, and R. C. Sickles. "Index Number and Factor Demand Approaches to the Estimation of Productivity." NBER Working Paper No. 5790, 1996.

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

Gorg, H., A. Hanley, and E. Strobl. Forthcoming. "Productivity effects of international outsourcing: Evidence from plant level data." Canadian Journal of Economics, 41, 2008, 670-88.

Griliches, Z. "Returns to Research and Development Expenditures in the Private Sector," in New Developments in Productivity Measurement and Analysis, edited by J. W. Kendrick and B. N. Vaccara. Chicago: University of Chicago Press, 1980, 419-61.

Head, K., and J. Ries. "Offshore Production and Skill Upgrading by Japanese Manufacturing Firms." Journal of International Economics, 58, 2002, 81-105. Hijzen, A. "International Outsourcing, Technological Change and Wage Inequality." Review of International Economics, 15, 2007, 188-205.

Hijzen, A., H. Gorg, and R. C. Hine. "International Outsourcing and the Skill Structure of Labor Demand in the United Kingdom." Economic Journal, 115, 2005, 860-78.

Javorcik, B. S. "Does Foreign Direct Investment Increase the Productivity of Domestic Firms? In Search of Spillovers Through Backward Linkages." American Economic Review, 94, 2004, 605-27.

Katayama, H., S. Lu, and J. R. Tybout. "Why Plant-Level Productivity Studies Are Often Misleading, and an Alternative Approach to Inference." NBER Working Paper No. 9617, 2003.

Kimura, F. "Subcontracting and the Performance of Small and Medium Firms in Japan." Small Business Economics, 18, 2002, 163-75.

Kimura, F., and M. Ando. "Two-Dimensional Fragmentation in East Asia: Conceptual Framework and Empirics." International Review of Economics and Finance, 14, 2005, 317-48.

Klette, T. J., and Z. Griliches. "The Inconsistency of Common Scale Estimators When Output Prices Are Unobserved and Endogenous." Journal of Applied Econometrics, 11, 1996, 343-61.

Kohler, W. "International Outsourcing and Factor Prices with Multistage Production." Economic Journal, 114, 2004, C166-85.

Levinsohn, J., and A. Petrin. "Estimating Production Functions Using Inputs to Control for Unobservables." Review of Economic Studies, 70, 2003, 317-41.

Matsuura, T. "Estimation of economic activity of Japanese multinationals." Keizai Toukei Kenkyu, 32, 2004, 1-16.

Melitz, M. J. "The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity." Econometrica, 71, 2003, 1695-725.

Ng, F., and A. Yeats. "Production Sharing in East Asia: Who Does What for Whom and Why?" Working Paper No. 2197, World Bank, 1999.

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

Olsen, K. B. "Productivity Impacts of Offshoring and Outsourcing: A Review." Technology and Industry Working Papers No. 2006/1, OECD Science, 2006.

Organisation for Economic Co-operation and Development. Employment Outlook. Paris: OECD, 2007.

Tomiura, E. "Foreign Outsourcing and Firm-Level Characteristics: Evidence from Japanese Manufacturers." Journal of the Japanese and International Economies, 19, 2005, 255-71.

Tomiura, E. "Foreign Outsourcing, Exporting, and FDI: A Productivity Comparison at the Firm Level." Journal of International Economics, 72, 2007, 113-27.

Yeats, A. J. "Just How Big Is Global Production Sharing?" Policy Research Working Paper No. 1871, World Bank, 1998.

Yi, K. -M. "Can Vertical Specialization Explain the Growth of World Trade?" Journal of Political Economy, 111, 2003, 52-102.

Young, A. "Learning by Doing and the Dynamic Effects of International Trade." Quarterly Journal of Economics, 106, 1991, 369-405.

(1.) Due to limitations of our data set, we focus on materials offshoring and neglect services offshoring. Services offshoring has recently become the center of the offshoring debate but does not come close, as of yet, to the importance of materials off shoring (Amiti and Wei 2006). Particularly, in Japan, materials off shoring has been more prevalent than services offshoring. In 2000, the share of imported material inputs in total output amounted to 2.7%, while the share of imported service inputs in total output amounted to 0.7% (Organisation for Economic Co-operation and Development 2007, chapter 3).

(2.) This corresponds to "international insourcing" used in Olsen (2006).

(3.) One may also refer to this as international outsourcing.

(4.) See Olsen (2006) for an excellent survey on the productivity impact of offshoring.

(5.) Domestic sourcing includes both sourcing to unaffiliated domestic firms (domestic outsourcing) and sourcing to affiliated domestic firms (domestic insourcing).

(6.) Kimura (2002) analyzes the relationship between sourcing to domestic firms and productivity but does not consider international sourcing. Tomiura (2005) analyzes the determinants of offshoring decisions. He finds that firms that engage in offshoring tend to be larger and more productive than firms that do not offshore, suggesting that there may be sizable fixed costs associated with offshoring.

(7.) See the Appendix for more details on the construction of data.

(8.) When we use more detailed industry classification such as the International Standard Industry Classification (ISIC) two-digit classification, we find that for some industries, the elasticity of labor or capital is above one. We obtain these clearly unfavorable results probably because the number of firms is small in some two-digit industries and hence outliers would affect estimation results. Therefore, we classify firms into only six industries.

(9.) We did, however, also implement the procedure proposed by Buettner (2003) resulting in very similar findings as in the case of Olley and Pakes's TFP measure.

(10.) The only exception is the machinery sector for which the estimated elasticity of capital using Olley and Pakes exceeds that using OLS.

(11.) We thank an anonymous referee for pointing this out.

(12.) A more precise definition of these variables will be presented below.

(13.) The inclusion of raw materials in OFF2 and its exclusion in OFF1 also explains why we cannot simply subtract OFF2 from OFF 1 to explicitly measure intrafirm and arm's-length offshoring.

(14.) Thus, DOM is consistent with the narrow definition of offshoring used in OFF1.

(15.) Since regressors are first lagged, instruments for the first-differenced equation in which the dependent variable is In [TFP.sub.it] - ln[TFP.sub.i,t-1] are the offshoring and other variables in years t - 3 and t - 4. When we instead use variables in years t - 2 and t - 3 as instruments, the Hansen J statistic indicates that the instruments are not orthogonal to the error term as may be expected.

(16.) Data for 2000 are only used to construct the TFP level, whereas data for the period 1994-1999 contain information on offshoring and domestic sourcing. Since we use the third lagged variables as instruments, our dependent variable is the TFP level from 1997 to 2000.

(17.) However, given the short nature of our panel any inferences regarding the time trend should be taken with caution.

(18.) The multilateral TFP index indicates firm-level TFP relative to the hypothetical average firm in the same industry at the initial period (see Measures of TFP section). Therefore, it is not meaningful to compare the multilateral TFP index between firms in different industries or compare it with the Olley-Pakes TFP measure of the same firm.

(19.) One may argue that the insignificance of the overall offshoring measure is due to multicollinearity between the overall offshoring measure and the infrafirm offshoring measure. However, the correlation coefficient between the two variables is only .12.

(20.) Amiti and Wei (2006) also observe that the effect of offshoring increases once they control for the correlation between offshoring and unobserved productivity shocks.

(21.) We use the median rather than the mean because the distribution of OFF2 is highly skewed.

(22.) The correlation coefficient of the growth rate of the two TFP measures is .99.

(23.) Another notable difference is that the size of the estimated coefficient on the R&D intensity is substantially smaller in the case of the Olley-Pakes measure than in the baseline regressions.

(24.) MNEs are defined as firms that report a positive balance of capital investment to foreign companies.

(25.) As noted in Note 19, the multilateral TFP index of a firm cannot be compared with that of another firm in a different industry. Therefore, we do not show average value of the TFP index for the subsamples.

(26.) We do not use the interaction term between the MNE dummy and the intrafirm offshoring measure since local firms should not be engaged in intrafirm offshoring by definition.

(27.) Note, however, that the impact of intrafirm offshoring is still positive as we found in the previous estimations.

(28.) This characterization is convenient for the present case as we are interested in the benefits of offshoring to the firm rather than the distributional issues, which have preoccupied the large share of the existing literature.

(29.) See Kohler (2004) and Hijzen (2007) for more a detailed analysis of such general equilibrium effects.

(30.) When importing raw data sets, we heavily relied on Stata programs written by Toshiyuki Matsuura for Matsuura (2004).

ALEXANDER HIJZEN, TOMOHIKO INUI and YASUYUKI TODO *

* This research was conducted as part of a project on industry- and firm-level productivity in Japan undertaken at the Research Institute of Economy, Trade and Industry. The authors would like to thank RIETI for providing us the opportunity of conducting this research and the Ministry of Economy, Trade and Industry for providing us valuable data sets. The authors are also grateful to Kyoji Fukao, Tsutomu Miyagawa, Jungsoo Park, Eiichi Tomiura, Masaru Yoshitomi, and seminar participants at the RIETI-21 st Century COE Hi-Stat Program Workshop and the RIETI DP seminar for helpful comments and suggestions and Young Gak Kim, Hyeog Ug Kwon, and Toshiyuki Matsuura for their help in constructing the data set. In addition, the authors are particularly grateful to three anonymous referees for their helpful comments. Inui thanks the Japan Society for the Promotion of Science (Grant-in-Aid for Scientific Research). Hijzen gratefully acknowledges financial support from the Leverhulme Trust (Grant No. F114/BF). The opinions expressed and arguments employed in this article are the sole responsibility of the authors and do not necessarily reflect those of RIETI or any of the institutions the authors belong to.

Hijzen: Economist, OECD and GEP, University of Nottingham, Paris, Cedex 16, France. Phone +33-01-4524-9261, Fax +33-0-1-4524-9098, E-mail alexander. hijzen@oecd.org

Inui: Professor, College of Economics, Nihon University, Chiyoda-ku, Tokyo 101-8360, Japan. Phone + 81-33219-3468, Fax + 81-3-3219-3468, E-mail inui@eco. nihon-u.ac.jp

Todo: Associate Professor, Graduate School of Frontier Sciences, University of Tokyo, Kashiwa, Chiba 2778563, Japan. Phone +81-4-7136-4863, Fax +81-71364842, E-mail yastodo@k.u-tokyo.ac.jp

TABLE 1
Estimated Elasticity of Labor and Capital

 Food, Textiles, Furniture,
 Beverages, Apparel, Paper,
Industries Tobacco Leather Publishing

Elasticity of labor
 OLS 0.472 0.807 0.927
 Olley-Pakes 0.399 0.706 0.829
Elasticity of capital
 OLS 0.535 0.260 0.219
 Olley-Pakes 0.616 0.434 0.327

 Machinery,
 Chemical, Electrical
 Petroleum, Machinery,
Industries Rubber Metal Transport Equipment

Elasticity of labor
 OLS 0.690 0.669 0.955
 Olley-Pakes 0.639 0.649 0.897
Elasticity of capital
 OLS 0.412 0.393 0.216
 Olley-Pakes 0.407 0.427 0.287

TABLE 2
List of Key Variables

Variable Name Description Definition

OFF1 Overall offshoring Ratio of purchases of
 (intrafirm and products, parts, and
 arm's-length components from foreign
 offshoring) providers to value added

OFF2 Intrafirm offshoring Ratio of purchases of any kind
 from the firm's own foreign
 affiliates to value added

DOM Domestic sourcing Ratio of purchases of
 products, parts, and
 components from domestic
 providers to value added

R&D R&D intensity Ratio of R&D expenditure to
 value added

TABLE 3
Summary Statistics

 Standard
Variable Mean Deviation Minimum Maximum

(A) Whole Sample

1nTFPIN 0.224 0.496 -4.118 2.774
InTFPoP -4.197 0.461 -8.841 -1.440
AlnTFPIN 0.011 0.344 -4.705 3.034
AInTFPoP 0.009 0.343 -4.705 3.028
OFF I 0.018 0.073 0.000 0.844
OFF2 0.052 0.238 0.000 11.974
DOM 0.468 0.673 0.000 5.218
R&D 0.076 0.094 0.000 0.631

Subsamples OFFI > 0 OFFI = 0 OFF2 > 0 OFF2 = 0

(B) Offshoring versus
Nonoffshoring Firms

Number of observations 1,669 8,461 2,967 7,163
1nTFPIN 0.321 0.205 0.318 0.185
1nTFPoP -4.163 -4.203 -4.126 -4.226
AlnTFPIN 0.018 0.010 0.020 0.007
AlnTFPoP 0.015 0.008 0.019 0.005
OFF I 0.110 0.000 0.035 0.011
OFF2 0.073 0.048 0.178 0.000
DOM 0.933 0.376 0.547 0.435
R&D 0.092 0.072 0.095 0.068

Note: See Table 2 for the definition of the variables used.

TABLE 4
Baseline Results

 Dependent Variable: Log of the
 Multilateral TFP Index

 (1) (2)
Regressor OLS OLS

1n[TFP.sub.t-1] Lagged TFP 0.784 (0.007) ** 0.784 (0.007) **
OFF1 Overall 0.083 (0.035) *
 offshoring
OFF2 Intrafirm 0.066 (0.011) **
 offshoring
DOM Domestic 0.007 (0.004) *** 0.008 (0.004) *
 sourcing
R&D R&D 0.288 (0.029) ** 0.285 (0.029) **
Number of observations 10,130 10,130
[R.sup.2] .76 .76
Hansen J statistic
Arrellano-Bond statistic

 Dependent Variable: Log of the
 Multilateral TFP Index

 (3) (4)
Regressor OLS GMM

1n[TFP.sub.t-1] Lagged TFP 0.784 (0.007) ** 0.620 (0.047) **
OFF1 Overall 0.059 (0.035) *** 0.077 (0.115)
 offshoring
OFF2 Intrafirm 0.063 (0.011) **
 offshoring
DOM Domestic 0.007 (0.004) *** 0.025 (0.020)
 sourcing
R&D R&D 0.285 (0.029) ** 0.609 (0.154) **
Number of observations 10,130 10,130
[R.sup.2] .77
Hansen J statistic .69
Arrellano-Bond statistic .05

 Dependent Variable: Log of the
 Multilateral TFP Index

 (5) (6)
Regressor GMM GMM

1n[TFP.sub.t-1] Lagged TFP 0.616 (0.048) ** 0.612 (0.047) **
OFF1 Overall 0.081 (0.129)
 offshoring
OFF2 Intrafirm 0.124 (0.046) ** 0.123 (0.(46) **
 offshoring
DOM Domestic 0.020 (0.020) 0.022 (0.020)
 sourcing
R&D R&D 0.635 (0.156) ** 0.618 (0.160) **
Number of observations 10,130 10,130
[R.sup.2]
Hansen J statistic .80 .66
Arrellano-Bond statistic .06 .07

Notes: See Table 2 for the definition of the variables used. Standard
errors are in parentheses. All specifications include industry-year
dummies. P values are reported for Hansen J statistics and the
Arellano-Bond statistics for second-order serial correlation.

*** Significant at 10% level; * significant at 5% level; and **
significant at 1% level.

TABLE 5
Robustness Checks

 Dependent Variable: Log of TFP

 (1) (2)
 OLS GMM

 TFP Constructed by
 Olley and Pakes (1996) Method,
Regressor Modification Not Multilateral TFP Index

1n[TFP.sub.t-1] Lagged TFP 0.803 (0.007) ** 0.549 (0.050) **
OFF1 Overall 0.059 (0.036) *** 0.084 (0.133)
 offshoring
OFF2 Intrafirm 0.063 (0.011) ** 0.106 (0.043) *
 offshoring
DOM Domestic 0.005 (0.004) 0.006 (0.022)
 sourcing
R&D R&D 0.082 (0.030) ** 0.280 (0.170) ***
Number of observations 10,130 10,130
[R.sup.2] .730
Hansen J statistic .000
Arrellano-Bond statistic .094

 Dependent Variable: Log of TFP

 (3) (4)
 OLS GMM

 Variables Based on Sales,
Regressor Modification Not Value Added

1n[TFP.sub.t-1] Lagged TFP 0.813 (0.006) ** 0.574 (0.045) **
OFF1 Overall 0.048 (0.033) 0.033 (0.112)
 offshoring
OFF2 Intrafirm 0.055 (0.015) ** 0.083 (0.042) ***
 offshoring
DOM Domestic -0.004 (0.005) 0.013 (0.019)
 sourcing
R&D R&D 0.215 (0.025) ** 0.653 (0.172) **
Number of observations 10,130 10,130
[R.sup.2] .791
Hansen J statistic .172
Arrellano-Bond statistic .461

 Dependent Variable: Log of TFP

 (5) (6)
 OLS GMM

 Second Lags for Regressors,
Regressor Modification Not First Lags

1n[TFP.sub.t-1] Lagged TFP 0.781 (0.007) ** 0.599 (0.048) **
OFF1 Overall -0.005 (0.038) -0.027 (0.078)
 offshoring
OFF2 Intrafirm 0.053 (0.013) ** 0.058 (0.031) ***
 offshoring
DOM Domestic -0.001 (0.004) 0.015 (0.010)
 sourcing
R&D R&D 0.239 (0.030) ** 0.236 (0.067) **
Number of observations 10,130 10,130
[R.sup.2] .763
Hansen J statistic .083
Arrellano-Bond statistic .009

Notes: See Table 2 for the definition of the variables used. Standard
errors are in parentheses. All specifications include industry-year
dummies. P values are reported for Hansen J statistics and the
Arrelano-Bond statistics for second-order serial correlation.

*** Significant at 10% level; * significant at 5% level; and **
significant at 1% level.

TABLE 6
Means of Variables for Various Subsamples

Subsamples MNEs Local Firms Exporters

Number of observations 4,887 5,243 7,810
1n[TFP.sup.OP] -4.149 -4.242 -4.176
[DELTA]1n[TFP.sup.IN] 0.018 0.004 0.013
[DELTA]1n[TFP.sup.0P] 0.017 0.001 0.010
OFF1 0.027 0.010 0.019
OFF2 0.113 0.000 0.054
DOM 0.522 0.391 0.469
R&D 0.100 0.065 0.090

Subsamples Nonexporters

Number of observations 2,320
1n[TFP.sup.OP] -4.267
[DELTA]1n[TFP.sup.IN] 0.005
[DELTA]1n[TFP.sup.0P] 0.003
OFF1 0.015
OFF2 0.060
DOM 0.408
R&D 0.053

Notes: See Table 2 for the description of the variables used.
Multinational firms are defined as firms with any positive balance in
foreign investment.

TABLE 7
Effect of FDI and Export on Impacts of Offshoring

 Dependent Variable: Log
 of TFP Index

Regressor X (Interacted Variable)

In[TFP.sub.t-1] Lagged TFP
OFF1 Overall offshoring
OFF1 x X
OFF2 Intrafirm offshoring
OFF2 x X
DOM Domestic sourcing
R&D R&D
Number of observations
Hansen J statistic
Arrellano-Bond statistic

 Dependent Variable: Log
 of TFP Index

 (2)
 GMM
Regressor Dummy for MNEs

In[TFP.sub.t-1] 0.540 (0.055) **
OFF1 -0.496 (0.227) *
OFF1 x X 0.545 (0.321) ***
OFF2 0.159 (0.043) **
OFF2 x X
DOM 0.027 (0.021)
R&D 0.646 (0.180) **
Number of observations 10,130
Hansen J statistic .376
Arrellano-Bond statistic .178

 Dependent Variable: Log
 of TFP Index

 (3)
 GMM
Regressor Dummy for Exporting Firms

In[TFP.sub.t-1] 0.523 (0.055) **
OFF1 -0.440 (0.193) *
OFF1 x X 0.388 (0.258)
OFF2 0.139 (0.032) **
OFF2 x X 0.034 (0.097)
DOM 0.022 (0.021)
R&D 0.638 (0.180) **
Number of observations 10,130
Hansen J statistic .120
Arrellano-Bond statistic .199

Notes: See Table 2 for the definition of the variables used. Standard
errors are in parentheses. All specifications include the interacted
variable X and industry-year dummies. P values are reported for Hansen
J statistics and the Arrelano-Bond statistics for second-order serial
correlation. FDI, foreign direct investment.

*** Significant at 10% level; * significant at 5% level; and **
significant at 1% level.