Multinationals' productivity advantage: scale or technology?

Girma, Sourafel ; Gorg, Holger

I. INTRODUCTION

There is now a substantial body of empirical work that documents a robust and positive correlation between foreign ownership and firm or plant productivity growth across a number of countries. The productivity advantage of foreign-owned firms is usually seen as reflecting multinationals' technological advantage vis-a-vis domestic firms. Multinationals are assumed to have a firm-specific asset, such as know-how, technology etc., which may be transferred easily across borders from the parent to subsidiaries abroad, as discussed by Markusen (2002). This allows them to be more productive than domestic firms. However, high productivity growth is not exclusively derived from technical progress, at least in theory where the direct link between technology and productivity is only valid in a neoclassical production framework with perfect competition, long run equilibrium, and constant returns to scale. Specifically, the productivity analysis literature, such as Balk (2001), highlights the role of changes in scale economies for productivity growth. This is consistent with the notion of learning-by-doing effects as described by Lucas (1988). Intuitively, as output expands, workers and firms gain proficiency at producing particular products. Thus, changes in scale efficiency can also provide an explanation for the observed productivity advantage of foreign firms.

One aim of this study was to decompose the productivity advantage of foreign multinationals into two components: the technology and the scale effect. Apart from being of academic interest, this issue is highly policy relevant. Many governments around the world actively promote inward foreign direct investment (FDI) under the assumption that it may lead to an influx of new technology that will ultimately spill over into the domestic economy. Hence, these policies are predicated on technical efficiency in multinationals. If, however, scale efficiency is the dominant component of foreign firms productivity premium in a particular sector, such policies may be misguided. Unfortunately, FDI theory gives little guidance as to the relative importance of technological progress and scale efficiency in the productivity premium due to foreign ownership. The empirical literature also appears to have neglected the issue of decomposing the productivity effects of multinationality. Hence, this study aimed to uncover the sources of productivity growth in a panel of domestic and foreign plants.

The second objective of this study was to contribute to the ongoing debate about the causal relationship between foreign ownership and productivity growth. While a number of studies, as cited above, have established that foreign-owned firms may have higher productivity growth than their domestic counterparts, there remains a fundamental problem in identifying the performance difference that is attributable to multinationality per se. As Tybout (2000), for example, points out, multinationals may be attracted to more technology-intensive industries, which are also more productive and pay higher wages. Hence, there would be an endogeneity problem in the regressions, and the wage differential between foreign and domestic firms would be difficult to interpret. The inclusion of some observable industry and firm characteristics, as well as unobservable time invariant effects, might go some way toward reducing this bias, though the inclusion of all possible relevant control variables is a difficult if not impossible task.

In this study, we tried to overcome this problem by analyzing the effects of an acquisition of a domestic establishment by a foreign multinational enterprise on productivity growth, decomposed into technology and scale effects. Assuming that an acquisition does not change any of the main characteristics of the takeover target (at least in the short run), a possible effect of the foreign acquisition on productivity growth in the domestic target can be attributed to the change in own ership from domestic to foreign. We attempt to identify the causal effect of a foreign acquisition using a combined propensity score matching and difference-in-differences methodology, as, for example, discussed in a recent study by Blundell and Costa Dias (2000). To our knowledge, this is the first study to provide a decomposition of the causal effects of foreign acquisition on productivity growth. (2)

The empirical setting of the study was the UK manufacturing industry, where FDI is seen as an important device of technology transfer. Some half a billion pounds was paid in grants for internationally owned companies by the UK government between 1991 and 1995 under the Regional Selective Assistance scheme. (3) We used plant-level data, covering the period 1980-1994. We focused our analysis on establishments in the UK electronics and food industries. (4) We decided to examine in detail two sectors separately rather than pooling data for the whole manufacturing since recent empirical studies of firm-level productivity dynamics, such as Bartelsman and Doms (2000), have established that there is large and persistent heterogeneity across firms even within sectors, let alone across heterogeneous sectors. Concentrating on two fairly narrowly defined sectors should allow us to alleviate the problems associated with aggregating over heterogeneous units. The choice of these two sectors is based on two reasons: first, foreign-owned firms are important players in both sectors, accounting for about 19% of employment in electronics and 10% in the food industry in 1996 as shown by Griffith and Simpson (2004; Table 4); second, we may expect those sectors to be different in their technology usage; hence, there may be differences in the determinants of productivity for establishments in the two sectors. (5)

The study proceeds with section II where the productivity growth decomposition and the propensity score matching methodologies are discussed. Section III presents the data. Section IV compares the sources of productivity growth between foreign and domestic plants and decomposes the productivity growth effects of foreign acquisitions. Section V concludes.

II. EMPIRICAL STRATEGY

A. Decomposing Productivity Changes: The Analytical Framework

The productivity analysis literature abounds with productivity growth decomposition methodologies, ranging from nonparametric to fully parametric techniques. This study uses a parametric method of decomposing an a Divisa index of total factor productivity (TFP) growth based on the estimation of a translog production function. (6)

Let y, [x.sub.1], [x.sub.2], [x.sub.3], and [x.sub.4] denote output, skilled labor, unskilled labor, capital, and material inputs, respectively, (7) and let t index a time trend variable. For plant i at time t, the translog production function is expressed as

(1) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where v is a disturbance term that is potentially correlated with the inputs. Compared to a simple Cobb-Douglas production function, this approach has the major advantage of allowing for varying returns to scales across plants and time periods.

Defining an index of TFP growth as the difference between the rate of output growth and the rate of input growth, [DELTA]TFP= [DELTA]ln y - [DELTA]ln x, and totally differentiating equation (1), TFP growth can be decomposed as

(2) [DELTA]TFP = [DELTA]T + ([epsilon] - 1) [n.summation over (k = 1)]([[epsilon].sub.k]/[epsilon])[DELTA]ln [x.sub.k]

where [DELTA]T = [partial derivative]ln y/[partial derivative]t, [[epsilon].sub.k] = [partial derivative]ln y/ [partial derivative][x.sub.k]. [epsilon] = [4.summation over (k = 1)] [[epsilon].sub.k] denotes returns to scale.

The first term on the right-hand side of equation (2) is the rate of technical change (the derivative of log output with respect to the time trend), while the second term captures the contribution of changes in scale efficiency. Note that scale efficiency at a point in time can be thought of as productivity relative to what is attainable under constant returns to scale.

To implement this decomposition empirically, we estimated equation (1) separately for each two-digit industry via instrumental variables methods to account for the potential endogeneity of inputs. (8) Furthermore, we allow for different factor elasticity coefficients for domestic- and foreign-owned plants in the estimation, by interacting all production function coefficient with a foreign ownership dummy. Thus, a foreign-acquired plant is allowed to have different coefficients in the production function before and after the acquisition year. (9)

B. Identifying the Causal Effects of Foreign Acquisition on Productivity Growth

Having decomposed productivity growth, the next step was to analyze whether there is a causal effect from an acquisition of a domestic establishment by a foreign owner on either or both of the components (technical change or scale efficiency). In other words, the empirical modeling problem is the evaluation of the causal effect of foreign acquisition on y, where y represents total productivity growth or one of its components in the domestic target.

Let [ACQ.sub.it] [member of] {0,1} be an indicator of whether a domestic plant i is acquired by a foreign establishment at time period t, and let [y.sup.1.sub.it + s] be the productivity growth at time t + s, s [greater than or equal to] 0, following acquisition. Also, denote by [y.sup.0.sub.it + s] the productivity growth of the plant had it not been acquired. The causal effect of foreign ownership for plant i at time period t + s is defined as:

(3) [y.sup.1.sub.it + s] - [y.sup.0.sub.it - s]

The fundamental problem of causal inference is that the quantity [y.sup.0.sub.it + s] is unobservable for plants that have been acquired (i.e., for which we observed t [y.sup.1.sub.it + s]). Thus, the analysis can be viewed as confronting a missing data problem. Following the microeconometric evaluation literature such as that by Heckman et al. (1997), we defined the average effect of acquisition on the acquired plants as

(4) E{[y.sup.1.sub.t + s] - [y.sup.0.sub.t + s]|[ACQ.sub.it] = 1} = E{[y.sup.1.sub.t + s]|[ACQ.sub.it] = 1} - E{[y.sup.0.sub.t + s]|[ACQ.sub.it]= 1}

Causal inference relies on the construction of the counterfactual for the last term in equation (4), which is the outcome the acquired plants would have experienced, on average, had they not been acquired. This is estimated by the average productivity growth (or its components) of the plants that remained in domestic hands, E{[y.sup.0.sub.it + s] | [ACQ.sub.it] = 0}.

An important feature in this exercise is the selection of a valid control group. One way of doing so is by using matching techniques. The purpose of matching is to pair each foreign-acquired plant with a domestic plant that has not undergone any ownership change on the basis of some observable variables, in such a way that the control plants' productivity growth trajectories can be studied to generate the counterfactual for the acquired plant.

Since matching involves comparing acquired and nonacquired plants across a number of observable preacquisition characteristics (e.g., preacquisition productivity, size, and age), it would be difficult to determine along which dimension to match the plants or what type of weighing scheme to use. It is therefore desirable to perform the matching on the basis of a single index that captures all the information from those variables. In this study, we adopted the method of propensity score matching due to Rosenbaum and Rubin (1983), who suggest the use of the probability of receiving treatment (foreign acquisition in the present context) conditional on those characteristics, to reduce the dimensionality problem. Accordingly, we first identified the probability of being acquired (or "propensity score") using the following logit model

(5) P([ACQ.sub.it] = 1)= F([X.sub.it - 1], [D.sub.it])

where D is the full set of industry and time dummies, and the vector X consists of the preacquisition level and growth of TFP, returns to scale (which is calculated from the instrumental variable translog production function estimates), plant size (proxied by log of fixed capital), age, and an indicator of whether the plant is located in an officially designated assisted area. The choice of these variables is motivated by the existing literature on ownership change, for example, Lichtenberg and Siegel (1987), McGuckin and Nguyen (1995), Conyon et al. (2002), Harris and Robinson (2002).

Now, let Pit denote the predicted probability of being acquired at time t for plant i (which is an actual takeover target). A nonacquired plant j, which is "closest" in terms of its propensity score (or probability of being a foreign takeover target) to an acquired plant, is then selected as a match for the latter using the "caliper" matching method. (10) The caliper method uses the nearest control plant whose propensity score falls within a prespecified radius as a match for an acquired plant. More formally, at time period t and for each newly acquired plant i, a domestic plant j is selected, such that (11)

(6) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where [lambda] is a prespecified scalar, which is set at 0.01 in our analysis. Furthermore, we impose the so-called common support condition in the matching algorithm. This involves dropping acquired plant observations whose propensity score is higher than the maximum or less than the minimum propensity score of the control group plants.

Having constructed the comparison group (C) of plants that are similar to the acquired plants (A), we exploit the panel nature of our data and use a difference-in-differences estimator. (12) This is motivated by recent studies that argue that standard matching estimators are usually unsatisfactory but in combination with difference-in-differences, methodology can have the potential to "... improve the quality of non-experimental evaluation results significantly" (Blundell and Costa Dias 2000, 438).

At its simplest, the combined matching and difference-in-differences estimator we use can be described as follows. First, the difference between the average productivity before and after the change of ownership, say [[DELTA].sup.a.sub.y], is calculated. Then, this difference is further differenced with respect to the before and after difference for the comparison control group, say [[DELTA].sup.c.sub.y], to obtain the difference-in-differences estimator [delta] = [[DELTA].sup.a.sub.y] [[DELTA].sup.c.sub.y]. Defining ACQ as a vector of dummy variables for the postacquisition period, the regression

(7) [DELTA][y.sub.it] = [phi] + [delta][ACQ.sub.it] + [u.sub.it]

on the matched sample of plants should produce a coefficient [delta] that can be interpreted as the average change in y that can be attributed to foreign acquisitions. In order to allow for differential acquisition effects across time, we considered two lags of the foreign acquisition dummy in addition to the contemporaneous effect. Furthermore, to control for possible observable factors that may be correlated with changes in TFP growth, we extended this basic framework by including a vector of regressors that consists of plant age and size and full sets of time and four-digit sectoral dummies.

It has been noted in the literature by, for example, Lapan and Bardhan (1973) and Wang and Blomstrom (1992) that the degree of technology transfer from parent company to new subsidiary is likely to be a function of the acquired plant's existing technological capability or absorptive capacity. Some threshold level of absorptive capacity or technological congruity' might be needed for the acquired plants to benefit fully from their new association with multinationals. But it can also be argued that a domestic plant that operates near the technological frontier might have less to learn from their association with multinationals than otherwise equivalent plants. To explore the above conjectures, we also interacted the acquisition dummy variables with the preacquisition or initial level of TFP (ITFP). Our final estimating equation can then be expressed as

(8) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

C. Testing the Reliability of the Propensity Score Matching Method

The propensity score matching method will provide a reliable and robust method for estimating the effects of foreign acquisitions if, conditional on the propensity score, the distribution of the preacquisition covariates is independent of the incidence of being acquired. This can be achieved by choosing a specification of the propensity score model (equation 5) that "balances" the preacquisition variables between the treatment and the control groups conditional on the propensity score. As emphasized by Rosenbaum and Rubin (1983), Dehejia and Wahba (2002), and Smith and Todd (2005a), among others, it is important to verify that this balancing condition is satisfied by the data.

In this study, we performed two tests to satisfy ourselves that the balancing conditions are not violated. The first is to conduct two sample t-tests of equality of the covariates between the treatment and the control groups. The second balancing test we explored is suggested by Smith and Todd (2005b) and it is cast within a regression framework. Let [??](X) denote the estimated propensity score and let ACQ be a dummy variable assuming a value of 1 if a firm is foreign acquired. Then, for each variable included in the matching algorithm, the following regression function that is quartic in [??](X) is estimated (using the TFP variable as an example)

(9) TFP = [[beta].sub.0] + [4.summation over (k = 1)] [[beta].sub.k][??][(X).sup.k] + [4.summation over (k = 1)] [[gamma].sub.k] ACQ[??][(X).sup.k] + [epsilon]

and the joint significance of the coefficients on the terms involving the program participation dummy (that is the [gamma]s) is tested. As explained by Smith and Todd (2005b), if the propensity score satisfies the balancing condition, the treatment dummy (acquisition dummy in our case) should not provide any additional information and we should expect the [gamma]s to be jointly statistically insignificant.

Ill. DATABASE DESCRIPTION AND SAMPLE CHARACTERISTICS

We used confidential microdata from the Annual Respondents Database (ARD) provided by the Office for National Statistics (ONS) in the United Kingdom under controlled conditions. The data set consists of individual establishments' records underlying the annual census of production. As Barnes and Martin (2002) provide a very useful introduction to the data set, we only included a brief discussion of some of the features of the data that are relevant to the present work.

For each year, the ARD consists of two files. What is known as the "selected file" contains detailed information on a sample of establishments that are sent inquiry forms. The second file comprises the "nonselected" (nonsampled) establishments, and only basic information such as employment, location, industry grouping, and foreign ownership status is recorded. Some 14,000-19,000 establishments are selected each year, based on a stratified sampling scheme. The scheme tends to vary from year to year, but for the period under consideration, establishments with more than 100 employees were always sampled.

In the data, an establishment is defined as the smallest unit that is deemed capable of providing information on the Census questionnaire. Thus, a "parent" establishment reports for more than one plant (or "local unit" in the parlance of ARD). For selected multiplant establishments, we only have aggregate values for the constituent plants. Indicative information on the "children" is available in the nonselected file. In our sample period, about 95% of the establishments in these industries are single-plant firms. In the actual sample we used for the econometric estimation, this figure is around 80%. Hence, most of the data used are actually plant-level data, and we therefore tend to use the terms plant and establishment interchangeably.

This study uses data for two broad industries, electronics and food and drinks, spanning 49 four-digit SIC80 industries (13) over the period 1980-1994. (14) A consistently defined nationality indicator identifies whether an establishment is domestic or foreign owned. Table 1 gives the frequency distribution of plants by year and ownership in the two industries under consideration. We define the incidence of foreign acquisition in year t as an establishment that has been in domestic hands up to year t-1 and becomes a subsidiary of a foreign-based multinational, as identified by a change in its nationality indicator. (15) Since the matching process described in the previous section requires data on the preacquisition period, we considered foreign takeovers that took place between 1981 and 1994. Table 2 provides the frequency distribution of foreign acquisitions. It can be seen that most of these occurred in the electronics industry.

Table 3 reports some summary statistics of the main variables of interest, namely, TFP growth and its two components, technical change and scale effects. These variables are calculated from the instrumental variables translog production function estimates, as described in section II. Economically significant average productivity growth is observed in the office machinery and data processing equipment sector (SIC2 33) where technical progress accounts for most of this growth. Also, foreign plants enjoy a higher productivity growth and technical progress than domestic plants in this sector. By contrast, we found an average negative effect of technical change in both food sectors. This may at least be partly due to low innovation activity in that sector; for example, Morgan, Blake, and Poyago-Theotoky (2003) discuss the low R&D intensity in the food industry compared to other manufacturing sectors in the United Kingdom. (16)

IV. EMPIRICAL RESULTS

To investigate more closely whether there are productivity growth differentials between foreign and domestic plants, we started our econometric analysis by decomposing the productivity growth differentials for the two types of plants. These results are shown in Table 4. For each industry, we ran three sets of regressions separately: explaining TFP growth and its components technical change and scale effects. Our results show that, controlling for plant age, size, and four-digit industry and time effects, foreign-owned plants exhibit higher TFP growth in the electronics industry. However, there are some noteworthy differences between the two 2-digit sectors of this industry. In the office machinery and data processing equipment sector (SIC2 33), the TFP of foreign plants grows by 1.2% points faster than domestic plants, and this advantage is entirely due to their higher rates of technical change. By contrast, foreign plants in the electrical and electronic engineering sector (SIC2 34) enjoy a more modest productivity growth advantage over their domestic counterparts, at just above half a percentage point. It is worth noting that technical progress accounts for just a third of this TFP growth differential, while scale effects appear to be a much more important component in this sector.

The picture that emerges from the food industry is more mixed. In the food manufacturing sector (SIC2 41), foreign plants have a 1.1% points productivity growth advantage, and this is entirely explained by technical efficiency. In sharp contrast, the TFP of domestic plants in the confectionary and drink manufacturing sector (SIC2 42) grew by more than 1.24% points faster than that of foreign plants. Interestingly, the contribution of scale efficiency to this TFP growth advantage is quite small, and faster technical progress appears to be the major factor responsible for this finding.

The above discussion focused on the importance of the various sources of productivity growth to the average domestic and foreign plants TFP differential. As discussed in the Introduction, it is inappropriate to draw any conclusions about causal relationships from this type of analysis. In what follows, we hence examine whether there is a more direct relationship between ownership change and the sources of productivity growth. To this end, we used the difference-in-differences estimator based on propensity score-matched plants described in section II.

Appendix 1 reports the marginal effects from logit regressions of the determinants of foreign acquisitions to illustrate the procedure used to calculate the propensity scores. Foreign multinationals appear to target older plants and plants with either lower level of productivity (in the electronics industry) or lower productivity growth (in the food industry). Furthermore, in the electronics sector, domestic plants with higher levels of returns to scale are more likely to be acquired.

Also, Appendices 2 and 3 summarize the results from the balancing tests for the propensity score matching method, and reassuringly, we found that the balancing conditions are met in both sectors. Thus, the tests give robust support for the soundness of the matching approach adopted in this study.

Table 5 provides a decomposition of the productivity growth effects of foreign acquisitions, and three broad patterns emerge: (i) any positive impact of ownership change is predominantly due to change in technical efficiency, (ii) the preacquisition TFP level of the erstwhile domestic plants play a role--positive or negative--in mediating the rate of technology transfer from the multinational parent companies, and (iii) the productivity growth effects are not confined to the year of acquisition and tend to persist through time. We provide a more detailed discussion of the results.

Keeping the preacquisition level of productivity constant, ownership change brings a statistically and economically significant benefit to the acquired firms in the office machinery and data processing equipment sector (SIC2 33). Within 2 years of acquisition, technical efficiency is growing by about (0.0095 + 0.007 + 0.0085=) 2.5% points more than would otherwise be the case. Furthermore, we detected a negative and statistically significant initial TFP-acquisition interaction term 1 year after ownership change. This indicates that domestic plants that were further behind the technology frontier appear to benefit more from their new associations with multinationals. On the other hand, the impact of foreign acquisition in the electrical and electronic engineering sector (SIC2 34) is less pronounced. Within 2 years of ownership change, the average effect (holding initial TFP constant) on technical change is about 0.65% points. Also, domestic plants with lower or higher levels of initial TFP do not seem to derive any differential benefits from their new status as subsidiaries of multinationals. In both sectors, scale effects appear to play little or no role for overall productivity growth.

Focusing on the food sector, the unconditional TFP growth effect of foreign acquisitions in the food manufacturing sector (SIC2 41) appears to take place within a year of acquisition. On average, new foreign ownership is associated with a 1.88% points premium in the rate of technical change. However, conditioning on the preacquisition level of TFP, the higher this level of TFP, the less marked the rate of technology transfer appears to be. In sharp contrast, the results from the confectionery and drink manufacturing sector (SIC2 42) are quite different from the ones considered thus far. Ownership change is initially detrimental to the TFP growth trajectory of the acquired plants. Holding initial TFP constant, the rate of technical change is slower by 1.27% points within a year of acquisition, than would otherwise be the case. We also detected a decline in TFP growth due to a loss in scale efficiency a year into ownership change, and this loss is higher, the higher the level of initial TFP of the plant. However, we observed an unconditional positive effect of 1.02% points increase in the technical progress 2 years after acquisition. This suggests that while there may be losses in the short run, these seem to be significantly counteracted by positive effects 2 years after the acquisition.

V. CONCLUDING REMARKS

This study had two objectives: first, to decompose the productivity advantage of foreign multinationals into two components: the technology and the scale effects--arguably a highly policy relevant issue that, however, has been neglected in academic research thus far; second, the study contributes to the ongoing debate about the causal relationship between foreign ownership and productivity growth. We do so by analyzing the effects of an acquisition of a domestic establishment by a foreign multinational enterprise on productivity growth, decomposed into technology and scale effects. In order to identify a causal effect, we used a combined propensity score matching and difference-in-differences estimation approach.

We analyzed separately plant-level data for the UK data for the electronics and food industries, which unearths substantial sectoral heterogeneity that would be lost if pooling data for the whole manufacturing industry. From our econometric investigation, we draw three major conclusions: (i) any positive impact of ownership change is predominantly due to change in technical efficiency and not in scale effects, (ii) the preacquisition TFP level of the erstwhile domestic plants play a role--positive or negative--in mediating the rate of technology transfer from the multinational parent companies, and (iii) the productivity growth effects are not confined to the year of acquisition and tend to persist through time.

ABBREVIATIONS

ARD: Annual Respondents Database

FDI: foreign direct investment

OECD: Organisation for Economic Co-operation and Development

ONS: Office for National Statistics

TFP: total factor productivity

doi: 10.1111/j.1465-7295.2006.00008.x APPENDIX 1 The Determinants of Foreign Acquisitions: Marginal Effects from the Logit Regressions Electronics Food Age 0.008 0.001 *** 0.003 0.001 * Size -0.001 0.001 0.000 0.000 TFP -0.025 0.014 * 0.002 0.002 TFP growth -0.197 0.229 -0.336 0.130 *** Return to scale 0.271 0.098 * 0.021 0.014 Assisted area 0.002 0.008 -0.006 0.001 Observations 1,900 1,810 Pseudo [R.sup.2] 0.322 0.569 Notes: Heteroscedasticity robust standard errors are given in parentheses. * Significant at 10%, ** significant at 5%, and *** significant at 1%. All regressions include the full set of time dummies. APPENDIX 2 Balancing Test Results: Electronics Sector Paired t-test Regression-Based Test p > p > [absolute [absolute value value t value of t] F value of F] Variable Age 0.84 0.404 0.36 0.843 Size 1.28 0.203 0.35 0.843 TFP -1.88 0.062 0.19 0.941 TFP growth -0.26 0.797 0.58 0.674 Return to scale -1.06 0.289 0.33 0.857 APPENDIX 3 Balancing Test Results: Food Sector Variable Paired t-test Regression-Based Test p > p > [absolute [absolute value value t value of t] F value of F] Age -0.03 0.973 1.08 0.377 Size 0.62 0.54 1.79 0.142 TFP 1.25 0.213 0.38 0.823 TFP growth -1.53 0.13 0.95 0.441 Return to scale -0.03 0.973 0.79 0.538 APPENDIX 4 Instrumental Variables Translog Production Function Estimates by Two-Digit Industry SIC80 Two-Digit Industry 33 34 x1 19.729 (1.17) -12.637 (4.96) x2 20.203 (1.66) * 7.136 (2.55) ** x3 -25.105 (1.55) 10.703 (3.20) *** x4 -3.613 (0.76) 0.924 (1.49) x1x1 0.027 (0.36) 0.159 (9.96) *** x2x2 0.084 (2.15) ** 0.052 (6.41) *** x3x3 0.167 (1.93) * 0.121 (5.68) *** x4x4 -0.005 (0.46) -0.013 (6.21) *** xlx2 -0.033 (0.76) -0.035 (4.69) *** xlx3 -0.054 (0.80) -0.106 (5.99) *** xlx4 0.021 (1.56) 0.004 (1.67) * x1time -0.010 (1.12) 0.007 (5.49) *** x2x3 -0.048 (0.91) -0.029 (3.38) *** x2x4 0.017 (1.65) * -0.002 (1.10) x2time -0.010 (1.60) -0.003 (2.38) ** x3x4 -0.022 (1.49) 0.006 (2.71) *** x3time 0.012 (1.47) -0.006 (3.51) *** x4time 0.002 (0.82) -0.000 (1.32) F * x1 0.000 (.) 0.000 (.) F * x2 0.000 (.) -21.512 (3.75) *** F * x3 1.106 (0.37) 5.494 (2.19) ** F * x4 0.963 (0.24) -2.208 (1.50) F * x1x1 -0.048 (0.41) 0.020 (0.42) F * x2x2 -0.069 (0.74) 0.020 (0.89) F * x3x3 -0.090 (0.72) 0.165 (1.64) F * x4x4 0.039 (1.16) -0.001 (0.l2) F * x1x2 0.081 (0.99) 0.045 (1.52) F * x1x3 0.034 (0.42) -0.081 (1.22) F * x1x4 -0.014 (0.34) -0.006 (0.51) F * x1time -0.000 (0.39) 0.001 (1.13) F * x2x3 0.001 (0.01) -0.045 (1.18) F * x2x4 0.026 (0.80) 0.000 (0.04) F * x2time -0.000 (0.68) 0.011 (3.87) *** F * x3x4 -0.046 (1.73) * 0.003 (0.20) F * x3time 0.001 (0.30) -0.004 (3.19) *** F * x4time -0.000 (0.23) 0.001 (1.54) Time 1.587 (0.32) -0.875 (1.01) F * time -0.029 (1.35) 0.026 (2.03) ** Timetime -0.001 (0.33) -0.000 (1.09) F * timetime 0.000 (1.03) -0.000 (1.62) Observations 538 8,834 Sargan test 0.134 0.213 (p value) SIC80 Two-Digit Industry 41 42 x1 -5.078 (1.96) ** -3.313 (0.59) x2 8.234 (4.13) *** 10.030 (2.28) ** x3 -3.219 (2.00) ** 4.292 (0.89) x4 1.244 (1.89) * -1.935 (0.98) x1x1 0.093 (7.50) *** 0.057 (2.18) ** x2x2 0.081 (11.77) *** -0.001 (0.03) x3x3 0.114 (19.61) *** 0.090 (4.51) *** x4x4 -0.007 (8.99) *** -0.014 (5.18) *** x1x2 -0.015 (1.89) * 0.017 (1.11) x1x3 -0.092 (14.02) *** -0.078 (4.23) *** x1x4 0.000 (0.11) 0.017 (2.33) ** x1time 0.003 (2.44) ** 0.002 (0.74) x2x3 -0.036 (8.38) *** -0.040 (3.15) *** x2x4 0.003 (1.67) * 0.005 (1.03) x2time -0.004 (3.98) *** -0.005 (2.14) ** x3x4 0.005 (3.57) *** -0.005 (0.95) x3time 0.001 (1.68) * -0.002 (0.91) x4time -0.001 (1.85) * 0.001 (1.10) F * x1 -20.927 (3.18) *** 0.000 (.) F * x2 -6.327 (1.00) 0.000 (.) F * x3 0.000 (.) 0.000 (.) F * x4 -0.201 (0.15) 0.000 (.) F * x1x1 0.136 (1.31) 0.014 (0.12) F * x2x2 0.145 (3.81) *** 0.100 (1.22) F * x3x3 0.046 (0.77) -0.062 (0.43) F * x4x4 -0.044 (4.63) *** -0.077 (2.98) *** F * x1x2 -0.081 (1.52) -0.021 (0.l9) F * x1x3 -0.011 (0.17) 0.078 (0.61) F * x1x4 0.060 (2.39) ** -0.051 (0.69) F * x1time 0.010 (2.95) *** -0.000 (0.34) F * x2x3 -0.083 (2.23) ** -0.135 (1.66) * F * x2x4 0.013 (1.27) 0.088 (1.55) F * x2time 0.004 (1.11) 0.000 (0.55) F * x3x4 -0.056 (2.52) ** 0.073 (1.26) F * x3time 0.000 (0.37) 0.000 (0.18) F * x4time 0.001 (1.12) -0.000 (0.29) Time 1.800 (2.44) ** -2.153 (1.18) F * time 0.054 (3.16) *** 0.008 (1.17) Timetime -0.001 (2.45) ** 0.001 (1.19) F * timetime -0.000 (3.70) *** -0.000 (0.73) Observations 6,288 3.462 Sargan test 0.121 0.176. (p value) Notes: Dependent variable is output and factor inputs are considered. These are skilled labor (x1),unskilled labor (x2), intermediate inputs (x3). and capital (x4). The prefix F * is used to denote interaction terms with the foreign ownership dummy. * Significant at 10%, ** significant at 5%, and *** significant at 1%. Standard errors are given in parentheses.

REFERENCES

Arnold, J., and B. S. Javorcik. "Gifted Kids or Pushy Parents? Foreign Acquisitions and Plant Performance in Indonesia." CEPR Discussion Paper 5065, 2005.

Balk, B. M. "Scale Efficiency and Productivity Change." Journal of Productivity Analysis, 15, 2001, 159-83.

Barnes, M., and R. Martin. "Business Data Linking: An Introduction." Economic Trends, 581, 2002, 34-41.

Bartelsman, E. J., and M. Doms. "Understanding Productivity: Lessons from Longitudinal Microdata." Journal of Economic Literature, 38, 2000, 569-95.

Blundell, R., and M. Costa Dias. "'Evaluation Methods for Non-Experimental Data." Fiscal Studies, 21, 2000, 427-68.

Conyon, M. J., S. Girma, S. Thompson, and P. W. Wright. "The Productivity and Wage Effects of Foreign Acquisition in the United Kingdom." Journal of Industrial Economics, 50, 2002, 85-102.

Dehejia, R., and S. Wahba. "Propensity Score Matching Methods for Non-Experimental Causal Studies." Review of Economics and Statistics, 84, 2002, 151-61.

Diewert, W. "Exact and Superlative Index Numbers." Journal of Econometrics, 4, 1976, 115-45.

Doms, M. E., and J. B. Jensen. "Comparing Wages, Skills, and Productivity between Domestically and Foreign-Owned Manufacturing Establishments in the United States," in Geography and Ownership as Bases for Economic Accounting, edited by R. Baldwin, R. Lipsey, and J. D. Richardson. Chicago: Chicago University Press, 1998, 235-55.

Fare, R., S. Grosskopt, M. Norris, and Z. Zhang. "Productivity Growth, Technical Progress and Efficiency Change in Industrialized Countries." American Economic Review, 84, 1994, 66-83.

Girma, S., and H. Gorg. "Evaluating the Foreign Ownership Wage Premium Using a Difference-in-Differences Matching Approach." Journal of International Economies, (in press).

Girma, S., D. Greenaway, and K. Wakelin. "Who Benefits from Foreign Direct Investment in the UK?" Scottish Journal of Political Economy, 48, 2001, 119-33.

Globerman, S., J. C. Ries, and I. Vertinsky. "The Economic Performance of Foreign Affiliates in Canada." Canadian Journal of Economies, 27, 1994, 143-56.

Grifell, E., and C. A. K. Lovell. "The Sources of Productivity Change in Spanish Banking." European Journal of Operational Research, 98, 1997, 364-80.

Griffith, R. "Using the ARD Establishment Level Data to Look at Foreign Ownership and Productivity in the United Kingdom." Economic Journal, 109, 1999, F416-42.

Griffith, R., and H. Simpson. "Characteristics of Foreign-Owned Firms in British Manufacturing," in Creating a Premier League Economy, edited by R. Blundell, D. Card, and R. Freeman. Chicago: Chicago University Press, 2004.

Harris, R., and C. Robinson. "The Impact of Foreign Acquisitions on Total Factor Productivity: Plant level Evidence from UK manufacturing, 19871992." Review of Economics and Statistics, 84, 2002, 562-68.

Heckman, J., H. Ichimura, J. Smith, and P. Todd. "Matching as an Econometric Evaluation Estimator: Evidence from Evaluating a Job Training Programme." Review of Economic Studies, 64, 1997, 605-54.

Heshmati, A. "Productivity Growth, Efficiency and Outsourcing in Manufacturing and Service Industries." Journal of Econamic Surveys, 7, 2003, 79-112.

Koop, G., J. Osiewalski, and M. F. J. Steel. "The Components of Output Growth: A Stochastic Frontier Analysis." Oxford Bulletin of Economic and Statistics, 61, 1999, 455-87.

Lapan, H., and P. Bardhan. "Localised Technical Progress and the Transfer of Technology and Economic Development." Journal of Economic Theory, 6, 1973, 585-95.

Leuven, E., and B. Sianesi. "PSMATCH2: Stata Module to Perform Full Mahalanobis and Propensity Score Matching, Common Support Graphing, and Covariate Imbalance Testing." 2003, http://ideas.repec. org/c/boc/bocode/s432001.html.

Lichtenberg, F., and D. Siegel. "Productivity and Changes in Ownership of Manufacturing Plants." Brookings Papers on Economic Activity: Microeconomics, 3, 1987, 643-73.

Lucas, R. E. "On the Mechanics of Economic Development." Journal of Monetary Economics, 22, 1988, 3-42.

Markusen, J. R. Multinational Firms and the Theory of International Trade. Cambridge, MA: MIT Press, 2002.

McGuckin, R. H., and S. V. Nguyen. "On Productivity and Plant Ownership Change: New Evidence from the Longitudinal Research Database." Rand Journal of Economics, 26, 1995, 257-76.

Morgan, C. W., A. Blake, and J. Poyago-Theotoky. "Managing Technological Change in the Food Industry: Some Case Studies." International Journal of Biotechnology, 5, 2003, 334-53.

Orea, L. "Parametric Decomposition of a Generalised Malmquist Productivity Index." Journal of Productivity, Analysis, 18, 2002, 5-22.

Rosenbaum, P., and D. B. Rubin. "The Central Role of the Propensity Score in Observational Studies for Causal Effects." Biometrika, 70, 1983, 41-55.

Smith, J., and P. Todd. "Does Matching Overcome Ladonde's Critique of Nonexperimental Estimators?" Journal of Econometrics, 125, 2005a, 305-63.

--. "Rejoinder." Journal of Econometrics, 125, 2005b, 365-75.

Tybout, J. "Manufacturing Firms in Developing Countries: How Well Do They Do, and Why?" Journal of Economic Literature, 38, 2000, 11-44.

Wang, J.-Y., and M. Blomstrom. "Foreign Investment and Technology Transfer: A Simple Model." European Economic Review, 36, 1992, 137-55.

(1.) See, for example, Globerman, Ries, and Vertinsky (1994), Doms and Jensen (1998), and Girma. Greenaway, and Wakelin (2001).

(2.) In a related study, we used a similar approach to investigate the effect of foreign acquisitions on wages, distinguishing the nationality of foreign acquirers: see Girma and Gorg (2006). Arnold and Javorcik (2005) also applied a similar approach to investigate the effect of foreign acquisitions on plant performance in Indonesia.

(3.) For more detail, see the official report at http://www.dti.gov.uk/regional/evaluationRSA91-95.pdf.

(4.) More precisely, using SIC 1980 classification, SIC 33 (manufacture of office machinery and data processing equipment), SIC 34 (electrical and electronic engineering), and SIC 41142 (food, drink, and tobacco).

(5.) According to an Organisation for Economic Co-operation and Development (OECD) classification, "electronics and communication" are classified as high-tech industries, while "food and beverages" are low-tech industries.

(6.) See, Fare et al. (1994) and Grifell and Lovell (1997) for examples of nonparametric; and Koop, Osiewalski. and Steel (1999), Balk (2001), and Orea (2002) for examples of fully parametric approaches. See also Heshmati (2003), for a review.

(7.) Output is defined as gross output. Skilled and unskilled labors are the number of nonproduction and production workers, respectively. Capital is defined as capital stock obtained using investment data (as described in Griffith 1999) and material as cost of materials and fuel used. All variables are in logs.

(8.) Twice lagged values of the factor inputs are used as instruments.

(9.) The point estimates of the production function are reported in Appendix 4 along with p values from the Sargan tests for the validity of the instrumental variable candidates.

(10.) The matching is performed in Stata Version 9 using the PSMATCH2 software provided by Leuven and Sianesi (2003).

(11.) A nonacquired g plant can be matched to more than one acquired plants. By the same token, it can happen that an acquired plant may not have a match.

(12.) The simplest form of a matching estimator of the causal effect of foreign acquisition can be written as [delta] = [summation over (i [subset] A)]([y.sub.i]- [summation over (j [subset] C)]w([p.sub.i], [p.sub.j])[y.sub.i]) where the w([p.sub.i], [p.sub.j]) are the weights placed on the comparison plant L generated by the matching algorithm.

(13.) These are SIC80 two-digit industries 33 and 34 (electronics) and 41 and 42 (food and drinks). We sometime refer to the latter as simply "food" throughout the manuscript. In the SIC80, the tobacco industry is also classified in the food sector. However, in our analysis, we do not consider tobacco manufacturing plants.

(14.) The motivation for concentrating on these two industries is discussed in the Introduction.

(15.) Establishments that appear to have experienced more than one change of ownership between 1980 and 1994 are excluded from the analysis. This is partly to avoid conflating the effects of different events and partly because we suspect the presence of measurement error problems.

(16.) They report that the R&D intensity in the food industry was 0.3% compared to a manufacturing average of 2.1% in 1997.

(17.) Throughout, insignificant coefficients will be ignored when calculating the magnitudes of the acquisition effects.

SOURAFEL GIRMA and HOLGER GORG *

* This work contains statistical data from ONS, which is Crown copyright and reproduced with the permission of the controller of Her Majesty's Stationery Office and Queen's Printer for Scotland. The use of the ONS statistical data in this work does not imply the endorsement of the ONS in relation to the interpretation or analysis of the statistical data. The authors thank David Greenaway; Richard Harris: Jonathan Haskel; Pehr-Johan Norback; Steve Redding; Don Siegel: Eric Strobl; Dieter Urban; Frederic Warzynski; participants at seminars at LSE, IUI Stockholm, Berlin, Lausanne, Nottingham, Turin; and two anonymous referees for helpful comments and suggestions. All remaining errors are, of course, the authors'. Financial support from the European Commission Fifth Framework Programme (Grant HPSE-CT1999-00017 and SERD-2002-00077) and the Leverhulme Trust (Grant F114/BF) is gratefully acknowledged.

Girma: Associate Professor/Reader, Nottingham University Business School, University of Nottingham, Jubilee Campus, Nottingham NG8 IBB, UK. Phone 44115-8466656, Fax 44-115-8466667, E-mail sourafel. girma@nottingham.ac.uk

Gorg: Associate Professor/Reader, Leverhulme Centre for Research on Globalisation and Economic Policy, and School of Economics, University of Nottingham, Nottingham NG7 2RD, UK. Phone 44-115-8466393, Fax 44-115-9514159, E-mail holger.gorg@nottingham.ac.uk TABLE 1 Frequency Distribution of Plants by Year and Ownership Electronics Food Year Foreign Domestic Total Foreign Domestic Total 1980 172 830 1,002 63 1,103 1,166 1981 183 827 1,010 67 1,068 1,135 1982 170 866 1,036 62 1,074 1,136 1983 170 859 1,029 63 1,041 1,104 1984 186 1,168 1,354 67 1,439 1,506 1985 173 935 1,108 61 1,098 1,159 1986 170 936 1,106 55 1,066 1,121 1987 180 957 1,137 50 1,010 1,060 1988 187 985 1,172 55 1,046 1,101 1989 221 1,360 1,581 73 1,387 1,460 1990 197 1,031 1,228 69 1,067 1,136 1991 238 979 1,217 84 1,038 1,122 1992 261 979 1,240 70 1,014 1,084 1993 242 952 1,194 62 1,042 1,104 1994 226 804 1,030 58 737 795 Total 3,098 15,274 18,372 1,032 17,585 18,617 TABLE 2 Frequency of Foreign Acquisitions Year Electronics Food 1981 11 3 1982 7 4 1983 6 9 1984 32 12 1985 10 I 1986 6 3 1987 15 5 1988 18 5 1989 30 14 1990 14 5 1991 41 15 1992 34 5 1993 24 10 1994 21 6 Total 269 97 TABLE 3 Summary Statistics for Productivity Growth SIC2 = 33 Foreign Domestic Standard Standard Variable Mean Deviation Mean Deviation TFP growth 0.070 0.033 0.059 0.034 Technical 0.071 0.012 0.055 0.011 change Scale effect -0.001 0.031 0.004 0.031 SIC2 = 34 Foreign Domestic Standard Standard Variable Mean Deviation Mean Deviation TFP growth 0.000 0.010 0.006 0.027 Technical 0.000 0.006 0.002 0.009 change Scale effect 0.000 0.008 0.004 0.026 SIC2 = 41 Foreign Domestic Standard Standard Variable Mean Deviation Mean Deviation TFP growth 0.000 0.034 -0.011 0.007 Technical -0.001 0.019 -0.011 0.006 change Scale effect 0.001 0.027 0.000 0.005 SIC2 = 42 Foreign Domestic Standard Standard Variable Mean Deviation Mean Deviation TFP growth -0.029 0.017 -0.016 0.021 Technical -0.028 0.006 -0.017 0.007 change Scale effect -0.001 0.016 0.001 0.020 TABLE 4 Decomposing Productivity Growth Differentials Between Foreign and Domestic Plants Electronics Sector TFP Growth Technical Change Foreign x SIC2 = 33 0.0121 (0.0038) *** 0.0166 (0.0021) *** Foreign x SIC2 = 34 0.0059 (0.0007) *** 0.0020 (0.0005) *** Size -0.0002 (0.0001) ** -0.0003 (0.0000) *** Age -0.0001 (0.0000) ** 0.0001 (0.0000) *** Constant 0.0560 (0.0026) *** 0.0538 (0.0018) *** Observations 12,038 12,038 [R.sup.2] 0.45 0.83 Electronics Sector Scale Effect Foreign x SIC2 = 33 -0.0045 (0.0031) Foreign x SIC2 = 34 0.0039 (0.0006) *** Size 0.0001 (0.0001) Age -0.0002 (0.0000) *** Constant 0.0021 (0.0020) Observations 12,038 [R.sup.2] 0.03 Food Sector TFP Growth Technical Change Foreign x SIC2 = 41 0.0109 (0.0031) *** 0.0100 (0.0025) *** Foreign * SIC2 = 42 -0.0124 (0.0011) *** -0.0100 (0.0006) *** Size -0.0002 (0.0001) * -0.0003 (0.0001) *** Age -0.0001 (0.0001) 0.0001 (0.0000) ** Constant -0.0036 (0.0014) ** 0.0001 (0.0009) Observations 12,178 12,178 [R.sup.2] 0.12 0.40 Food Sector Scale Effect Foreign x SIC2 = 41 0.0009 (0.0017) Foreign * SIC2 = 42 -0.0024 (0.0009) *** Size 0.0001 (0.0001) ** Age -0.0002 (0.0001) *** Constant -0.0037 (0.0011) *** Observations 12,178 [R.sup.2] 0.02 Notes: Heteroscedasticity and within-plant serial correlation robust standard errors are given in parentheses. * significant at 10%, ** significant at 5%, and *** significant at 1%. All regressions include the full set of time and four-digit industry dummies. TABLE 5 Decomposing the Productivity Growth Effects of Foreign Acquisitions Electronics Sector TFP Growth Technical Change SIC2 = 33 Acquisition year 0.0114 (0.0055) ** 0.0095 (0.0021) *** Acquisition year x initial TFP 0.0106 (0.0116) -0.0043 (0.0044) After 1 year 0.0056 (0.0066) 0.0070 (0.0025) *** After 1 year x initial TFP -0.0018 (0.0137) -0.0104 (0.0052) ** After 2 years 0.0024 (0.0067) 0.0085 (0.0026) *** After 2 years x initial TFP 0.0024 (0.0135) -0.0001 (0.0051) SIC2 = 34 Acquisition year 0.0031 (0.0015) ** 0.0024 (0.0006) *** Acquisition year x initial TFP -0.0068 (0.0056) 0.0029 (0.0022) After 1 year 0.0033 (0.0017) * 0.0025 (0.0006)*** After 1 year x initial TFP -0.0026 (0.0063) 0.0025 (0.0024) After 2 years 0.0049 (0.0018) *** 0.0016 (0.0007)** After 2 years x initial TFP 0.0035 (0.0069) 0.0038 (0.0027) Observations 2.226 2,226 [R.sup.2] 0.35 0.79 Electronics Sector Scale Effect SIC2 = 33 Acquisition year 0.0001 (0.0060) Acquisition year x initial TFP 0.0162 (0.0124) After 1 year -0.0029 (0.0070) After 1 year x initial TFP 0.0088 (0.0142) After 2 years -0.0076 (0.0074) After 2 years x initial TFP 0.0034 (0.0016) SIC2 = 34 Acquisition year 0.0004 (0.0016) Acquisition year x initial TFP -0.0102 (0.0059) * After 1 year 0.0005 (0.0017) After 1 year x initial TFP 0.0058 (0.0066) After 2 years 0.0036 (0.0020) * After 2 years x initial TFP -0.0010 (0.0077) Observations 2.226 [R.sup.2] 0.03 Food Sector TFP Growth Technical Change SIC2 = 41 Acquisition year 0.0113 (0.0031) *** 0.0077 (0.0019) *** Acquisition year x initial TFP -0.0199 (0.0177) -0.0398 (0.0109) *** After 1 year 0.0115 (0.0058) ** 0.0111 (0.0036) *** After 1 year * Initial TFP 0.0618 (0.0495) 0.0052 (0.0303) After 2 years 0.0965 (0.0685) 0.0671 (0.0419) After 2 years x Initial TFP -0.0025 (0.0035) -0.0052 (0.0021) ** SIC2 = 42 Acquisition year -0.0077 (0.0031)** 0.0067 (0.0019) *** Acquisition year x initial TFP -0.0148 (0.0138) -0.0025 (0.0085) After 1 year -0.0115 (0.0036) *** -0.0060 (0.0022) *** After 1 year x initial TFP -0.0282 (0.0164) * -0.0022 (0.0100) After 2 years 0.0116 (0.0071) 0.0102 (0.0043) ** After 2 years x initial TFP 0.0012 (0.0162) 0.0015 (0.0099) Observations 703 703 [R.sup.2] 0.31 0.55 Food Sector Scale Effect SIC2 = 41 Acquisition year 0.0034 (0.0029) Acquisition year x initial TFP 0.0185 (0.0162) After 1 year 0.0005 (0.0052) After 1 year * Initial TFP 0.0577 (0.0442) After 2 years 0.0465 (0.0698) After 2 years x Initial TFP 0.0029 (0.0032) SIC2 = 42 Acquisition year -0.0013 (0.0028) Acquisition year x initial TFP -0.0129 (0.0129) After 1 year -0.0058 (0.0032) * After 1 year x initial TFP -0.0271 (0.0145) * After 2 years 0.0037 (0.0078) After 2 years x initial TFP 0.0009 (0.0146) Observations 703 [R.sup.2] 0.09 Notes: Heteroscedasticity and within-plant serial correlation robust standard errors are given in parentheses. * Significant at 10%, ** significant at 5%, and *** significant at 1%. All regressions include the full set of time and four-digit industry dummies. The regressions also control for plant size and age, but the corresponding estimates are omitted to save space.