Does rising import competition harm local firm productivity in less advanced economies? Evidence from the Vietnam’s manufacturing sector

s (fixed effect). However, one potential problem of differencing is that it maymagnifymeasurement error;therefore, longer differences should be used to reduce the impact of the measurement error.4 D ow nl oa de d by [D r T uy en T ran ] a t 0 6:2 9 2 1 M ay 20 15 8 T. Doan et al. In empirical analysis, we start with a baseline estimation of Equation (1) with the ordinary least squares (OLS) and FE model for level data, and then with time-differenced specifications (Equation 2).  lnVAijt = α + β1imp penj t + β2 lnKijt + β3 lnLijt +β4ownershipij t + λt + eijt . (2) This is a common parsimonious alternative (e.g. Edwards and Jenkins 2013; Bloom, Draca, and Van Reenen, 2011; Bugamelli, Fabiani, and Sette 2010) to estimate separate capital and labour coefficients for each of 128 separate industry groups at four-digit level over the study period, or using a two-step approach of regressing residuals from industry-specific production functions on import penetrationmeasures. Due to the inclusion of industry fixed effects, the coefficients on the import penetration measures are identified by within-industry (four-digit level) variation over time for time-difference models. In this construction, there is no within-industry variation in import penetration in any given year. Direct estimation of Equation 2 may give biased estimates of the key parame- ters of interest (β1) due to omitted variables that are correlated with both import penetration and value added, or due to the endogeneity of factor inputs or import penetration, where these covariates may respond to value added. Such biases may result in upward estimates of positive (or negative) effects even no effects taking place. The endogenous choice of factor inputs is another potential source of en- dogeneity bias. Firms may choose variable factor inputs in response to new in- formation on their (possibly time varying) firm-specific productivity (λi). This introduces an upward bias in the coefficients on variable inputs such as labour, and a consequent downward bias on the capital coefficient (Griliches and Mairesse 1998). The degree of bias that this form of endogeneity causes to estimates of import penetration effects is an empirical question. There are several solutions to the endogeneity of inputs. The first is the IV ap- proach. This method needs valid instruments that are correlated with endogenous variables (input level choice, e.g. labour) but not correlated with firm outcome or its residual (error terms). It is typically hard to find good instruments that satisfy the conditions. Input prices (interest rate and/or wage rate) can be potential in- struments, but input prices are often unavailable in data-sets or do not vary or do not vary enough across firms. Even if there is a variation in input price, it may account for market power in input markets or heterogeneity in quality of inputs, e.g. worker quality that may invalidate the use of input price as an instrument (Ackerberg, Caves, and Frazer 2006). Recently, some suggest using the approach of Levinsohn and Petrin (2003), an extension of Olley and Pakes (1996),that uses firm’s intermediate consumption to control for the endogeneity. However, there are some disadvantages in these approaches such as identification and estima- tion issues (see more detailed discussion in Ackerberg, Caves, and Frazer 2006, D ow nl oa de d by [D r T uy en T ran ] a t 0 6:2 9 2 1 M ay 20 15 The Journal of International Trade & Economic Development 9 and Wooldridge 2009). Due to the lack of valid instruments to address the endo- geneity and lack of outperforming methods, we therefore apply the conventional approach, the OLS and the FE to a Cobb–Douglas production function in this paper to estimate TFP. We rely on estimates that do not explicitly control for the endogeneity of factor inputs, focusing instead on controlling for the potential endogeneity of import penetration. Whereas firms are expected to endogenously adjust factor inputs in response to annual changes in firm-specific productivity (λit), import penetration is likely to respond to changes in overall productivity performance within industries. The inclusion of industry fixed effects and industry-specific time controls the influence of import firms targeting particular industries on the basis of average industry productivity or relative productivity growth of industries over the sample period. Imports may gravitate towards more profitable or higher mark-up sectors. Imports may also be attracted to less competitive sectors to exploit their greater competitive advantages. Therefore, there is potential reverse causality. The import penetration (or import penetration growth as we use time- differencing) variable is potentially endogenous and may be influenced by in- ternational trade shocks in the last few decades such as China’s accession to the WTO that resulted in Chinese export boom, and FTAs that Vietnam signed with its key trade partners. Therefore, estimates from (1) and (2) may suffer from bias due to endogeneity of import penetration. Our identification strategy to deal with the endogeneity is to exploit the ex- ogenous shocks to Vietnam’s imports. In the last decade, Vietnam signed many FTAs with its key trade partners including an FTA agreement between China and ASEAN of which Vietnam is a member.5 China’s accession to the WTO in 2001 may also have been a big shock to Vietnam’s imports. We make use of China’s exports (to the world) as a potentially suitable instrument candidate for our iden- tification strategy because China’s exports may meet two conditions for a good instrument: (1) Vietnam’s imports and China’s exports may be correlated, and (2) Vietnamese firm productivity is not directly correlated with China’s exports. China’s export penetrationmay not be directly correlated with Vietnamese firm productivity. Hence, we may model Vietnam’s import penetration as a function of China’s exports, either in levels or changes as seen in Equations (3) and (4). importj,t−m = f (CNexportj,t−m,Xj,t−m) (3) importj,t−m = f (CNexportj,t−m,Xj,t−m) (4) We also use lag length (m) to allow for reverse causality and firm dynamic response to import competition in the production function estimation. We shall exploit Chinese export growth as a source for instruments because they capture the impact of China’s accession to theWTO on China’s export boom.6 In particular, quotas on the Multi-Fibre Agreement were eliminated in two waves in 2002 and 2005; according to Brambilla, Khandelwal, and Schott (2010), the D ow nl oa de d by [D r T uy en T ran ] a t 0 6:2 9 2 1 M ay 20 15 10 T. Doan et al. removal of the quotas have led to a huge increase (270%) in Chinese textile and apparel exports. Bloom,Draca, andVanReenen (2011, p.52) suggests that ‘China’s [export] increase was substantially larger than other countries not just because it joined the WTO but also because the existing quotas seemed to bite more heavily on China as indicated by the higher “fill rates – the proportion of actual imports divided by the quota” of Chinese quotas and Chinese quotas were increased more slowly over time than those in other countries.’ Vietnam’s increasing import share from China seems to be exogenous to Vietnamese firm productivity and determined by the fast-growing export of China to the world in 2000s (see Appendices 1 and 2). To capture the shocks to Vietnam’s imports by China’s accession to the WTO and the impact of the quota removal for China in 2002 and 2005 and the China–ASEAN FTA in 2005, we can use the change in China’s exports to instrument change in Vietnam’s import penetration. As a further means of limiting the potential bias from endogeneity of import penetration, we use lagged rather than current values of import penetration and IV methods. Longer period lags of the import penetration variablesmay be appropriate as import penetration would take time to have an effect on local firm productivity. However, using long changes restricts the sample as a result of dropping initial periods, and also excluding short-lived firms. The latter may lead to survivor bias – by estimating impact on only surviving firms, we would miss out possible negative impacts of import penetration on short-lived firms. We estimate Equation 5 in time-differenced specifications – regressing changes in the logarithm of firm value added against changes in log of factor inputs and (lagged) import penetration. The differencing removes industry and firm- level variation in the level of productivity, and the bias associated with import penetration. We present estimates using changes over two or three years. Estimates based on longer changes better capture the impact of more persistent changes, and are less affected by noise that biases the coefficients towards zero (Griliches and Hausman 1986).7 The estimating equation is shown as Equation 5, with all variables included as k-period changes, and import penetration variables lagged by m periods. Lagged changes are used to ensure that import penetration changes are predetermined relative to current plan productivity changes, and to allow for the possibility that the effect on productivity may take time to have an effect. We then combine level and time-differences with the IV method to further consolidate our findings. k lnVAijt = α + β1kimp penj,t−m + β2k lnKijt +β3k lnLijt + β4ownershipij t + λt + eijt (5) Equation (5) is estimated with standard errors clustered by industry and year to allow for the fact that measured import penetration does not vary within industry and year (Moulton 1990).8 We also want to test the hypothesis that import competition may affect various groups of firms differently, e.g. comparedwith high-tech firms, but the competition D ow nl oa de d by [D r T uy en T ran ] a t 0 6:2 9 2 1 M ay 20 15 The Journal of International Trade & Economic Development 11 from Chinese imports may adversely affect the outcomes of low-tech industries (firms), such that low-tech firms shrink or even exit, with consequent reduction in employment. Firm size could be a proxy for firm competitiveness so that we will run separate regressions for different firm size groups. We expect that larger firms are more likely to compete with imports while smaller firms may be crowded out by cheaper and better (to some extent) quality products from imports.We therefore estimate the effects of import penetration for different firm sizes and for various technology level industries. 4. Empirical results 4.1. Baseline effect estimates The first and second columns of Table 2 present the results of OLS and FE estimation of Equation (1), which models the log of value added as a function of the log of contemporaneous factor inputs and import penetration. The estimates of Equation (2) are shown in columns 3, 4 and 5 of Table 2, modelling changes in the log of value added on contemporaneous (k = 0) changes in the log of inputs and penetration, for changes over two and three years. The coefficients for both labour input and the cost of capital services are highly significant across the models, and provide credible production function estimates. The OLS estimates are likely to be upward biased due to the correlation of factor inputs with firm fixed effects. The implied increasing returns to scale coefficient (the sum of labour and capital coefficients) is 1.08. This reduces to 1.02 when fixed effects are controlled for in column 2, and lies between 0.86 and 1.02 in the differenced specification. The lowest estimate of 0.86 is for the second differenced specification, in which coefficients, especially the coefficient on labour, is lower due to transitory fluctuations. All the models are controlled for industry, year and ownership (foreign, state and private) dummies. The inclusion of concurrent import penetration variables in Table 2 may result in endogeneity bias, but they provide baseline estimates and also some evidence of self-selection of import penetration. For example, industries with lower pro- ductivity tend to import more. All regressions include industry and year effects, so that the estimated impact of import penetration reflects the association between changes in productivity and changes in penetration over time within an industry. The FE estimate of import penetration in the second column of Table 2 is smaller (in absolute terms) than that of the OLS estimation. The estimates across differenced specifications are also lower than the OLS estimate. The variation in the coefficient across these specifications reflects a combination of different samples and the potential impact of endogeneity associated with using concurrent changes of import penetration. Sample sizes are smaller when using longer time- differences because the data excludes short-lived firms. To remove endogeneity bias and reduce the influence of volatile short-term fluctuations, our preferred specification for time differences relies on lagged values of import penetration changes, as in Equation (5), with changes measured over two D ow nl oa de d by [D r T uy en T ran ] a t 0 6:2 9 2 1 M ay 20 15 12 T. Doan et al. Ta bl e 2. E ff ec ts of im po rt pe ne tr at io n on fi rm pr od uc tiv it y, 20 00 –2 00 9. Tw o- ye ar Tw o- ye ar T hr ee -y ea r C on tr ol le d O L S Fi xe d ef fe ct di ff er en ce di ff er en ce di ff er en ce fo r te ch H ig h te ch V ar ia bl es (k = 0) (1 ) (k = 0) (2 ) (k = 2) (3 ) (k = 2) (4 ) (k = 3) (5 ) le ve l( 6) on ly (7 )  kl nL t 0. 69 8 0. 68 0 0. 65 3 0. 55 5 0. 68 0 0. 55 5 0. 64 9 (0 .0 06 )∗ (0 .0 23 )∗ (0 .0 11 )∗ (0 .0 15 )∗ (0 .0 12 )∗ (0 .0 15 )∗ (0 .0 38 )∗  kl nK t 0. 37 8 0. 33 6 0. 33 3 0. 30 5 0. 33 6 0. 31 0 0. 28 5 (0 .0 05 )∗ (0 .0 12 )∗ (0 .0 06 )∗ (0 .0 08 )∗ (0 .0 07 )∗ (0 .0 08 )∗ (0 .0 18 )∗  kI m p_ pe n t –0 .0 61 –0 .0 20 –0 .0 35 –0 .0 16 (0 .0 05 )∗ (0 .0 02 )∗ (0 .0 02 )∗ (0 .0 02 )∗  kI m p_ pe n t -3 –0 .0 16 –0 .0 16 1 –0 .0 14 7 (0 .0 02 )∗ (0 .0 02 )∗ (0 .0 02 )∗ C on st an t 1. 81 6 1. 97 2 0. 10 8 0. 13 5 0. 14 8 0. 13 5 0. 04 87 (0 .0 47 )∗ (0 .0 74 )∗ (0 .0 22 )∗ (0 .0 29 )∗ (0 .0 27 )∗ (0 .0 29 )∗ (0 .1 15 9) O bs er va ti on s 19 1, 90 7 19 1, 90 7 88 ,7 97 32 ,2 14 64 ,9 56 32 ,2 14 32 27 R -s qu ar ed 0. 87 (w )0 .5 7 0. 39 0. 29 0. 45 0. 29 0. 32 Y ea r ef fe ct Y es Y es Y es Y es Y es Y es Y es In d2 co nt ro ll ed Y es N o Y es Y es Y es Y es Y es N ot e: T he de pe nd en t va ri ab le is th e na tu ra l lo g of va lu e ad de d. C lu st er ed (b y ye ar an d in d4 ) st an da rd er ro rs ar e in pa re nt he se s. ∗ s ig ni fi ca nt at 1% .k re pr es en ts ti m e le ng th -d if fe re nc es ;‘ k = 0’ in di ca te s le ve ls .A ll m od el s co nt ro lf or ye ar ,t w o- di gi ti nd us tr y du m m ie s an d ow ne rs hi p (p ri va te ,S O E s an d F D I) .  k is k- ye ar di ff er en ce s. S w it ch in g in du st ry w as co rr ec te d to al lo w fo r fi xe d ef fe ct w it hi n in du st ry (i nd 4) an d al so co rr ec tly cl us te ri ng . Te ch le ve ls ar e lo w ,m ed iu m an d hi gh te ch no lo gy . D ow nl oa de d by [D r T uy en T ran ] a t 0 6:2 9 2 1 M ay 20 15 The Journal of International Trade & Economic Development 13 years (column 4 of Table 2). Lag lengths are used to ensure that lagged changes are measured prior to output changes. Estimates of our preferred specification are shown in column 4 of Table 2. As we use two-year differences, the shortest pre-determined lag of import penetration variables that we can use is a three-period lag. One of the costs of ensuring that import penetration variables are pre-determined is that the estimation sample is greatly reduced due to the absence of lagged values for early years. The estimation sample is reduced to 32,214, which is about one-third of the sample available for the two-year differenced specification in column 3 of Table 2. The first column of Table 2 presents our baseline estimates. Changes in import penetration are estimated to have a statistically significant and negative impact on domestic firm productivity. The coefficient of negative 0.061 implies that across all industries an increase of one per cent in import penetration lowers productivity by 0.061%. The estimated impact of penetration remains negative and significant when we move to FE and time-differenced specifications. Although the time-differenced specifications and distributed lag of import penetration remove a bulk of short- lived firms, the effect is still negative and statistically significant but the magnitude (in absolute terms) declines. This may imply that the effect of import penetration is either overestimated by the OLS or stronger for short-lived firms. The last two columns of Table 2 offer estimates from a model that controlled for technology level dummies (column 6) and a model for high tech industries only (column 7, see Appendix 3 for the definition of industry technological level). The results in columns 6 and 7 are comparable with those in column 4. The results from a model controlled for technology level are almost unchanged compared with those in column 4. The results in column 7 for high tech industries only show a slight change with the impact of import penetration reduced (in absolute terms) marginally, from negative 0.016 to negative 0.015. The contribution of labour input increases while that of capital declines in comparison with results in columns 4 and 6. This can be explained by the fact that the high technological industries employ higher skilled workers and then labour input plays a more important role in the firms’ output. However, the difference between the estimated coefficients of the import penetration variable from high technological industries is very small (0.001)) suggesting that the technology level does not matter in this case. Because of similar or even lower level of Vietnam’s technological advancement in relation to countries exporting to Vietnam, rising import penetration may lead to fiercer competition among firms in Vietnam. Therefore, firms in Vietnam do not have an advantage to cut costs or to innovate in order to improve efficiency or to escape competition. 4.2. Instrumental variable approach The other approach to correct for endogeneity of import penetration is to use an IV method. As discussed earlier, China’s exports (or export growth) are a potential instrument for the prediction of the level of Vietnam’s imports, but may not D ow nl oa de d by [D r T uy en T ran ] a t 0 6:2 9 2 1 M ay 20 15 14 T. Doan et al. directly affect firms’ productivity in Vietnam. We thus make use of China’s export penetration, the proportion of China’s exports over total sales of an industry, as an instrument for import penetration. We apply the IV approach to both level variable specifications (the left panel of Table 3) and time-differenced specifications (the right panel of Table 3). We report some statistics for the relevance, validity and weak IV tests on the bottom rows of Table 3. All the test results in Table 3 show that the instrument is valid and strong, implying that the instrument is good. The result from the IV approach is consistent with the existing evidence in the literature which finds that China’s exports were a good instrument to predict imports (Edwards and Jenkins 2013; Bloom, Draca, and Van Reenen 2011; Bugamelli, Fabiani, and Sette 2010) and in line with the theory of trade shocks from China’s export boom. The left panel of Table 3 presents the estimated effects of import penetration on firm productivity using level data. Three model specifications are investigated: contemporaneous, the first lag and second lag of import penetration. The endoge- nous variable, import penetration were instrumented by the corresponding Chinese export penetration. The Anderson and Robin (AR) test for weak instrument rejects the hypothesis of the weak instrument in all model specifications at the one per cent level. The results in columns 2 and 3 are more statistically significant, in other words, the estimated results in columns 2 and 3 are more precisely estimated (see the last row of Table 3 for the test statistics). The right panel of Table 3 combines three approaches, IV, time-difference and lagged level of the difference. This combination can simultaneously address omitted variable and endogeneity bias. The AR test statistics show even more precise estimates especially in columns 4 and 6. The estimates are also smaller (in absolute terms) than those from the left panel. However, it must be noted that the sample size in the right panel is smaller than the ones in the left panel. Consequently, we infer that the decline in the estimated effect is due to the removal of short-lived firms from the sample. Columns 7 and 8 of Table 3 offer estimates from a model that controlled for technology level dummies (column 7) and a model for high technological industries only (column 8). The results on columns 7 and 8 are comparable with those in column 6. The results in both columns 7 and 8 are again almost unchanged compared to that in column 6. In addition, the results are similar to those of time-differenced specification models. This suggests that the impact of import penetration on productivity does not vary across different technological levels. Overall, the estimated results show that import penetration has a statistically significant and negative effect on firm productivity. However, the magnitude of the estimated coefficient of import penetration is relatively small compared to the coefficients of the capital and labour variables. The estimated coefficient of minus 0.016 implies that an increase in import penetration equivalent to 1% of industry output is estimated to decrease domestic firm productivity by 0.016%. This is small. It therefore can be concluded that the rising import penetration does not affect firm productivity to any significant economic extent. D ow nl oa de d by [D r T uy en T ran ] a t 0 6:2 9 2 1 M ay 20 15 The Journal of International Trade & Economic Development 15 Ta bl e 3. E ff ec ts of im po rt pe ne tr at io n on fi rm pr od uc tiv it y, 20 00 –2 00 9 (I V es ti m at io n– G M M m et ho d) . K = 0 (l ev el ) K = 2 (t w o- ye ar di ff er en ce ) V ar ia bl es (1 ) (2 ) (3 ) (4 ) (5 ) (6 ) (7 ) (8 )  kl nL t 0. 69 8 0. 68 4 0. 67 7 0. 65 3 0. 57 8 0. 55 5 0. 55 5 0. 64 9 (0 .0 06 )∗ ∗ (0 .0 06 )∗ ∗ (0 .0 07 )∗ ∗ (0 .0 11 )∗ ∗ (0 .0 12 )∗ ∗ (0 .0 15 )∗ ∗ (0 .0 15 )∗ ∗ (0 .0 38 )∗ ∗  kl nK t 0. 37 8 0. 39 0 0. 40 1 0. 33 3 0. 30 1 0. 30 5 0. 30 5 0. 28 5 (0 .0 05 )∗ ∗ (0 .0 06 )∗ ∗ (0 .0 06 )∗ ∗ (0 .0 06 )∗ ∗ (0 .0 07 )∗ ∗ (0 .0 08 )∗ ∗ (0 .0 08 1) ∗∗ (0 .0 18 )∗ ∗  ki m p_ pe n –0 .0 91 –0 .0 35 (0 .0 41 )∗ (0 .0 03 )∗ ∗  ki m p_ pe n_ L 1 –0 .0 67 –0 .0 14 (0 .0 18 )∗ ∗ (0 .0 04 )∗ ∗  ki m p_ pe n_ L 2 –0 .0 25 (0 .0 07 )∗ ∗  ki m p_ pe n_ L 3 –0 .0 16 –0 .0 16 –0 .0 15 (0 .0 03 )∗ ∗ (0 .0 03 )∗ ∗ (0 .0 02 5) ∗∗ co ns ta nt 1. 81 6 1. 88 5 1. 95 2 0. 07 1 0. 11 5 0. 13 5 0. 27 2 0. 06 0 (0 .0 47 )∗ ∗ (0 .0 38 )∗ ∗ (0 .0 36 )∗ ∗ (0 .0 24 )∗ ∗ (0 .0 19 )∗ ∗ (0 .0 29 )∗ ∗ (0 .0 24 )∗ ∗ (0 .0 12 )∗ ∗ O bs er va ti on s 19 1, 90 7 12 9, 27 4 95 ,3 82 88 ,7 97 62 ,6 89 32 ,2 14 32 ,2 14 32 27 C en tr ed R 2 0. 87 0. 89 0. 89 0. 39 0. 30 0. 29 0. 29 0. 32 In st ru m en te d va ri ab le im p_ pe n im p_ pe n_ L 1 im p_ pe n_ L 2  2 im p_ pe n  2 im p_ pe n_ L 1  2 im p_ pe n_ L 3  2 im p_ pe n_ L 3  2 im p_ pe n_ L 3 E xc lu de d ex pC N _p en ex p ex p  2 ex p  2 ex p  2 ex p  2 ex p  2 ex p in st ru m en t C N _p en _L 1 C N _p en _L 2 C N _p en C N _p en _L 1 C N _p en _L 3 C N _p en _L 3 C N _p en _L 3 1s ts ta ge P ro b > F 0. 00 00 0. 00 00 0. 00 00 0. 00 00 0. 00 00 0. 00 00 0. 00 00 0. 00 00 Pa rt ia lR 2 0. 93 0. 98 0. 98 0. 99 0. 99 0. 99 0. 99 0. 98 Te st fo r in st ru m en t 2. 6e + 06 6. 7e + 06 9. 2e + 06 1. 2e + 07 7. 5e + 06 1. 2e + 07 1. 2e + 07 3. 4e + 07 eq ua lz er o in th e (0 .0 00 0) (0 .0 00 0) (0 .0 00 0) (0 .0 00 0) (0 .0 00 0) (0 .0 00 0) (0 .0 00 0) (0 .0 00 0) 1s ts ta ge ,F -v al (P -v al ) A R te st fo r w ea k IV , 7. 96 ∗∗ 15 .5 6∗ ∗ 11 .9 9∗ ∗ 13 9. 43 ∗∗ 8. 86 ∗∗ 39 .9 ∗∗ 39 .8 ∗∗ 36 .5 ∗∗ C hi 2( 1) (P -v al ue (0 .0 04 8) (0 .0 00 1) (0 .0 00 5) (0 .0 00 0) (0 .0 02 9) (0 .0 00 0) (0 .0 00 0) (0 .0 00 0) in br ac ke t) N ot e: T he de pe nd en tv ar ia bl e is th e na tu ra ll og of va lu e ad de d. C lu st er ed (b y ye ar an d in du st ry ) st an da rd er ro rs ar e in pa re nt he se s. ∗ s ig ni fi ca nt at 5% ;a nd ∗∗ at 1% . k re pr es en ts ti m e le ng th -d if fe re nc es ;‘ k = 0’ in di ca te s k- ye ar -d if fe re nc e. L is la g le ve l. A ll m od el s co nt ro ll ed fo r ye ar ,t w o- di gi ti nd us tr y fi xe d ef fe ct an d ow ne rs hi p (p ri va te ,S O E s an d F D I) .C ol um n 7 co nt ro ll ed fo r te ch le ve ld um m ie s, an d co lu m n 8 is fo r hi gh te ch in du st ri es on ly . D ow nl oa de d by [D r T uy en T ran ] a t 0 6:2 9 2 1 M ay 20 15 16 T. Doan et al. Table 4. Decomposing effects by firm size (employees), 2000–2009. Firm size fewer than Firm size greater than or equal 20 100 200 200 Variable (1) (2) (3) (4) 2lnLt 0.588 0.552 0.553 0.560 (0.027)∗∗ (0.017)∗∗ (0.015)∗∗ (0.030)∗∗ 2lnKt 0.325 0.311 0.310 0.285 (0.014)∗∗ (0.009)∗∗ (0.009)∗∗ (0.016)∗∗ 2Imp_pent-3 –0.018 –0.013 –0.013 46.691 (0.004)∗∗ (0.002)∗∗ (0.002)∗∗ (25.086) Constant 0.371 0.081 0.084 0.183 (0.279) (0.041)∗ (0.034)∗ (0.056)∗∗ Observations 8032 19693 23947 8267 R-squared 0.30 0.30 0.30 0.26 Note: The dependent variable is the two-year difference in the log of value added. Clustered standard errors are in parentheses; ∗significant at 5%; ∗∗significant at 1%. Dependant variable 2lnva, 2 is the two-year difference. All models controlled for year, two-digit industry fixed effect and ownership (private, SOEs and FDI). 4.3. Effects by firm size Firms may have different competitive capacity to compete with imports. The liter- ature on import competition suggests that firms with less advanced technologies or small firms may find it hard to compete with imports (e.g. Bloom, Draca, and Van Reenen 2011; Bugamelli, Fabiani, and Sette 2010). In this section, we hypoth- esize that larger domestic firms may benefit more from the presence of increasing imports, due to their generally more sophisticated technologies and business pro- cesses, while smaller firms suffer a negative crowding out effect from intensifying import competition. The estimates are presented in the first two columns of Table 4. We use three cut-offs: 20, 100 and 200 employees to classify firms. Our estimates show that the coefficients are more negative for smaller groups. However, the estimated coefficients for subsample in columns 1, 2 and 3 are small so that the difference between coefficients in these columns is negligible. In the meantime, the effect is positive for firms with 200 or more employees, but the effect is not estimated precisely, i.e. not statistically significant at the five per cent level. 4.4. Impact of increased import penetration on the likelihood of firm death We now look at the effects of rising import penetration on firm survival that have not been captured by the main estimates earlier. Firms that cease operation within two years of a change in import penetration are automatically excluded from the estimating sample. The bias from excluding ceased firms will depend on whether the firm death is raised or lowered and on whether the affected firms have high D ow nl oa de d by [D r T uy en T ran ] a t 0 6:2 9 2 1 M ay 20 15 The Journal of International Trade & Economic Development 17 Ta bl e 5. Im pa ct of in cr ea se d im po rt pe ne tr at io n on th e li ke li ho od of fi rm de at h. Tw o- ye ar di ff er en ce (k = 2) T hr ee -y ea r di ff er en ce (k = 3) (1 ) (2 ) (3 ) (4 ) (5 ) (6 )  kl nL t –0 .0 41 6 –0 .0 61 9 –0 .0 90 7 –0 .0 48 8 –0 .0 84 2 –0 .0 81 9 (0 .0 03 0) ∗∗ (0 .0 04 1) ∗∗ (0 .0 10 7) ∗∗ (0 .0 03 4) ∗∗ (0 .0 06 3) ∗∗ (0 .0 09 0) ∗∗  kl nK t –0 .0 15 7 –0 .0 24 4 –0 .0 38 4 –0 .0 17 2 –0 .0 30 0 –0 .0 36 8 (0 .0 02 0) ∗∗ (0 .0 02 3) ∗∗ (0 .0 05 7) ∗∗ (0 .0 02 1) ∗∗ (0 .0 03 8) ∗∗ (0 .0 05 1) ∗∗ Fo re ig n –0 .1 49 8 –0 .1 19 4 –0 .0 64 2 –0 .1 12 8 –0 .1 11 1 –0 .0 66 8 (0 .0 13 2) ∗∗ (0 .0 13 5) ∗∗ (0 .0 28 0) ∗ (0 .0 13 4) ∗∗ (0 .0 21 8) ∗∗ (0 .0 28 1) ∗ P ri va te 0. 06 78 0. 09 27 0. 23 11 0. 10 21 0. 17 64 0. 23 32 (0 .0 13 2) ∗∗ (0 .0 13 7) ∗∗ (0 .0 26 6) ∗∗ (0 .0 13 5) ∗∗ (0 .0 21 7) ∗∗ (0 .0 26 2) ∗∗  ki m p_ pe n t 0. 03 86 0. 04 45 (0 .0 11 4) ∗∗ (0 .0 10 9) ∗∗  ki m p_ pe n_ L 1 0. 62 76 5. 10 52 (4 .4 29 0) (1 0. 93 12 )  ki m p_ pe n_ L 3 4. 79 24 1. 54 33 (1 6. 31 91 ) (2 1. 91 77 ) O bs er va ti on s 63 ,4 24 42 ,9 71 11 ,0 16 44 ,3 56 20 ,2 85 11 ,0 75 W al d ch i2 36 97 .5 7 33 03 .4 2 n/ a 35 06 .4 31 86 .1 6 46 07 6. 25 P ro b > ch i2 0. 00 00 0. 00 00 0. 00 00 0. 00 00 0. 00 00 0. 00 00 N ot e: T he re su lt s sh ow n ar e av er ag e m ar gi na l ef fe ct s es ti m at ed fr om P ro bi t m od el s. T he de pe nd en t va ri ab le is an in di ca to r of w he th er th e fi rm ce as es to op er at e w it hi n th e fo ll ow in g th re e ye ar s. It ta ke s th e va lu e of 1 fo r th re e ye ar s pr io r to th e de at h of a fi rm ,a nd 0 ot he rw is e. R ob us t cl us te re d st an da rd er ro rs ar e gi ve n in pa re nt he se s; ∗ s ig ni fi ca nt at 5% ; ∗∗ si gn ifi ca nt at 1% . T he sa m pl e is re st ri ct ed to 20 00 –2 00 7 so th at ea ch fi rm ’s su rv iv al ca n be ob se rv ed fo r th e fo ll ow in g th re e ye ar s. D ow nl oa de d by [D r T uy en T ran ] a t 0 6:2 9 2 1 M ay 20 15 18 T. Doan et al. or low productivity levels. If the rising import penetration raises the probability that local firms with slow productivity growth cease operation, this may lead to survivor bias – by estimating impact on only surviving firms, we would miss out possible negative impacts of rising import penetration on ceased firms. In order to gauge the possible impact of rising import penetration on firm survival, we estimate the probability of firms ceasing operation as a function of changes in import penetration. In order to focus on the subset of firms potentially excluded by our main specification, we estimate a firm’s likelihood of ceasing operation within three years, following a two-year change in import penetration.9 Table 5 presents estimates from a Probit regression of firm death, with estimates reported as (average) marginal effects. The positive coefficients on all model specifications of import penetration change imply that increases in import penetration raise the probability of firm death, though the estimated effect is statistically significant only for columns 1 and 4 (Table 5). The positive coefficient implies that rising import penetration raises the probability of firm death within three years. The columns 2, 3, 5 and 6 of Table 5 show whether changes in import pene- tration affect firm survival over a longer time period. As in Table 5, the inclusion of longer lags has the effect of restricting the sample to firms that have been operating for longer than three years. As the sample size reduces considerably, the estimates become less precise, but the signs of coefficients are still positive. The signs of other variables are as expected: larger firms (more employment and capi- tal) and foreign owned firms are more likely to survive, ceteris paribus. Overall, we observe that increasing import competition raises the likelihood of firm death. 5. Conclusion and discussion This paper examines how firm productivity is affected by rising import penetra- tion in the Vietnam manufacturing sector. The analysis employs a panel data-set of more than 200,000 firms during the period 2000–2009, thus making the sample comprehensive and representative of the Vietnam manufacturing business popu- lation. The source of the data was the VES from the Vietnam General Statistics Office. The study uses different approaches to deal with biases caused by omitted variables and the endogeneity of import penetration. Overall, as we set out in the beginning of the paper, Vietnamese manufacturing firms are typically less technologically advanced and have low competitiveness. We find that an increase of one per cent in import penetration results in a small decline (0.016 per cent) in productivity. There is no evidence of a positive effect on productivity from many robust estimates even our sample included only survival firms.10 Our finding is consistent with Federico (2014), who find no positive effect of import penetration from low-wage countries on firm employment, the wage bill and outputs in Italy’s manufacturing firms those with low tech advancement, and Alvarez and Claro (2008) who find no positive effect of rising import penetration on firm productivity in Chilean manufacturing firms. Unlike most research on D ow nl oa de d by [D r T uy en T ran ] a t 0 6:2 9 2 1 M ay 20 15 The Journal of International Trade & Economic Development 19 developed countries, in Vietnam as well as in Chile capital and skilled labours are scarce, which impedes or makes it costly for product upgrading. Moreover, imports displace or crowd out domestic production (Edwards and Jenkins 2013). This may cause local firms to lose market shares to importing firms, thus reducing their scale of production which impacts negatively on productivity. Firms lose their scale efficiency. Firms’ exposure to increasing import competition leads to lower productivity, but this is economically small. The small negative impact is mainly from smaller firm groups, who have 200 or fewer employees. We also find some evidence of positive effects for larger firm groups, but the effects are less precisely estimated. In addition, we do not find any significant evidence of variation in the impact of import penetration across different technological level industries, but we observe some evidence of positive effects of increased import penetration on the likelihood of firm death. In future research, we would refine our study by separating out imports from developing and developed countries. We also would separate various impacts of imported inputs by classifying imported intermediate inputs and imports of final competing goods to see how they affect firm performance in developing economies. Furthermore, as suggested by Yahmed and Dougherty (2012), one may look at the position of firm’s productivity in relation to the industry produc- tivity frontier to provide a better measure of technological advance. In addition, considering multiple product or multiple business functionality firms (e.g. firms both import inputs and/or complete products for sales and produce products for local and exporting markets), the effect of import penetration would vary across different types of firms (Bernard, Redding and Schott 2011). Finally, the net benefit of trade liberalization can be observed when we look at the impacts of imports, ex- ports and across-industry effect simultaneously. This is another avenue for future research. Acknowledgements We thank, without implicating, two anonymous reviewers for their helpful comments and suggestions to improve the paper. Any remaining errors in the paper are those of the authors. Disclosure statement No potential conflict of interest was reported by the authors. Notes 1. 2. The concording program is available upon request. 3. In Vietnam, it is not unusual to run a business before having a tax code. 4. But longer time differencing, e.g. three-year or four-year differencing also comes at the cost of losing more observations of short-lived firms;therefore, two-year D ow nl oa de d by [D r T uy en T ran ] a t 0 6:2 9 2 1 M ay 20 15 20 T. Doan et al. differenced specification is the main model specification for our time-differencing approach. 5. The Agreement on Trade in Goods of the China–ASEAN FTA entered into force in July 2005, and the Agreement on Trade in Services came into effect in July 2007. 6. Bloom, Draca, and Van Reenen (2011, p.28) indicate that China is a good experiment of low-wage country trade shock in the recent decades. 7. For this reason, we do not estimate for one-year difference specification. 8. Clustering may be still problematic if the number of clusters (industry-years) is small relative to the number of units per cluster. Cameron, Gelbach, and Miller (2008) suggest cluster bootstrapping techniques for inference. We tried both clustered and clustered bootstrapping for our main estimates and found similar estimated standard errors. We report clustered standard errors in the paper. 9. Firm survival cannot be accurately identified by a firm’s absence from our analysis data alone. A firm is classified as continuing if, in the following year, there is output or intermediate input recorded in the VES data-set, if the firm has positive employment or if any sales or purchases are observed. 10. If including exited firms in the sample, the estimated effect would be more negative, but obviously not positive as exited firms should be the least productive ones. Our estimated effect thus would be underestimated. References Acharya, R.C., and W. Keller. 2009. “Technology Transfer Through Imports.” Canadian Journal of Economics/Revue Canadienne D’e´conomique 42 (4): 1411–1448. Ackerberg, D., K. Caves, and G. Frazer. 2006. “Structural Identification of Production Functions.” Aghion, P., N. Bloom, R. Blundell, R. Griffith, and P. Howitt. 2005. “Competition and Innovation: An Inverted-U Relationship.” Quarterly Journal of Economics 120 (2): 701–728. Aghion, P., S. Redding, R. Burgess, and F. Zilibotti. 2005. “Entry Liberalization and Inequality in Industrial Performance.” Journal of European Economic Association 3 (2–3): 291–302. Altenburg, L., and A. Brenken. 2008. “Effort, Trade, and Unemployment.” Canadian Journal of Economics 41 (3): 864–893. Alvarez, R., and Claro S. 2008. “David Versus Goliath: The Impact of Chinese Competition on Developing Countries.” World Development 37 (3): 560–571. Amiti, M., and J. Konings. 2007. “Trade Liberalization, Intermediate Inputs, and Produc- tivity: Evidence From Indonesia.” American Economic Review 97 (5): 1611—1638. Autor, D., D. Dorn, and G. Hanson. 2013. “The China Syndrome: Local Labour Market Effects of Import Competition in the United States.” American Economic Review 103 (6): 2121–2168. Bernard, A., S. Redding, and P. Schott. 2011. “Multi-Product Firms and Trade Liberaliza- tion.” Quarterly Journal of Economics, 126 (3): 1271–1318. Bloom, N., M. Draca, and J. Van Reenen. 2011. Trade Induced Technical Change? The Impact of Chinese Imports on Innovation, IT and Productivity. National Bureau of Economic Research. Brambilla, I., A.K. Khandelwal, and P.K. Schott. 2010. “China’s Experience Under the Multi-Fiber Arrangement (MFA) and the Agreement on Textiles and Clothing (ATC).” In China’s Growing Role in World Trade, edited by R. Feenstra and Sh. Wei, 345–387. University of Chicago Press. Bugamelli, M., S. Fabiani, and E. Sette. 2010. “The Pro-Competitive Effect of Imports from China: An Analysis of Firm-Level Price Data.” Working Paper No. 737. Rome: Bank of Italy. D ow nl oa de d by [D r T uy en T ran ] a t 0 6:2 9 2 1 M ay 20 15 The Journal of International Trade & Economic Development 21 Cameron, A., B. Gelbach, and L. Miller. 2008. “Bootstrap-Based Improvements for Infer- ence with Clustered Errors.” Review of Economics and Statistics 90 (3): 414–427. Coe, D.T., and E. Helpman. 1995. “International R&D Spillovers.” European Economic Review 39 (5): 859–887. Doan, T., and P. Stevens. 2012. “Evolution of Competition in Vietnam Industries over the Recent Economic Transition.” Economics – The Open-Access, Open-Assessment E-Journal 6 (19): 1–24. Edwards, L., and R. Jenkins. 2013. The Impact of Chinese Import Penetration on the South African Manufacturing Sector. https://www.uea.ac.uk/documents/439774/ 0/Edwards+ Jenjkins+SALDRU.pdf/82f05bfc-bf8b-4af1-b843-bb522695c6d6 Federico, S. 2014. “Industry Dynamics and Competition from Low-Wage Countries: Evi- dence from Italy.” Oxford Bulletin of Economics and Statistics 76 (3): 389–410. Fernandes, A. 2007. “Trade Policy, Trade Volumes and Plant-Level Productivity in Colom- bian Manufacturing Industries.” Journal of International Economics 71 (1): 52–71. Ferreira, P, and J. Rossi. 2003. “New Evidence from Brazil on Trade Liberalization and Productivity Growth.” International Economic Review 44 (4): 1383–1405. Freeman, R.B., and M.M. Kleiner. 2005. “The Last American Shoe Manufacturers: De- creasing Productivity and Increasing Profits in the Shift from Piece Rates to Continuous Flow Production.” Industrial Relations: A Journal of Economy and Society 44 (2): 307– 330. Griliches, Z., and J. Hausman. 1986. “Errors in Variables in Panel Data.” Journal of Econo- metrics 31 (1): 93–118. Griliches, Z, and J.Mairesse. 1998. “ProductionFunctions: TheSearch for Identification.” In Econometrics andEconomic Theory in the 20thCentury: TheRagnar FrischCentennial Symposium, edited by S. Strøm, 169–203. Cambridge: Cambridge University Press. Grossman, G.M., and E. Helpman. 1995. “Technology and Trade.” Handbook of Interna- tional Economics 3: 1279–1337. Halpern, L., M. Koren, and A. Szeidl. 2005. Imports and Productivity. ssrn.com/sol3/papers.cfm?abstract_id = 785127 Halpern, L., and G. Ko¨ro¨si. 2001. “Efficiency andMarket Share in the Hungarian Corporate Sector.” Economics of Transition 9 (3): 559–592. Holmes, T., and J. Schmitz. 2010. “Competition and Productivity: A Review of Evidence.” Research Department Staff Report no. 439. Minneapolis: Federal Reserve Bank of Minneapolis. Kasahara, H., and J. Rodrigue. 2008. “Does the Use of Imported Intermediates Increase Productivity? Plant-Level Evidence.” Journal of Development Economics 87 (1): 106– 118. Kumbhakar, S. 2003. “Factor Productivity and Technical Change.” Applied Economics Letters 10 (5): 291–297. Levinsohn, J. & A. Petrin. 2003. “Estimating Production Functions using Inputs to Control for Unobservables.” Review of Economic Studies 70 (2): 317–41. Melitz, M. 2003. “The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity.” Econometrica 71 (6): 1695–1725. Mion, G., and L. Zhu. 2013. “Import Competition from and Offshoring to China: A Curse or Blessing for Firms?” Journal of International Economics 89 (2013): 202–215. Moulton, B.R. 1990. “An Illustration of a Pitfall in Estimating the Effects of Aggregate Variables on Micro Units.” Review of Economics and Statistics 72 (2): 334–338. Olley, G.S. and A. Pakes. 1996. “The Dynamics of Productivity in the Telecommunications Equipment Industry.” Econometrica 64 (6): 1263–1297. Pavcnik, N. 2002. “Trade Liberalization, Exit, and Productivity Improvements: Evidence from Chilean Plants.” Review of Economic Studies 69 (1): 245–276. D ow nl oa de d by [D r T uy en T ran ] a t 0 6:2 9 2 1 M ay 20 15 22 T. Doan et al. Sivadasan, J. 2009. “Barriers to Competition and Productivity: Evidence from India.” B.E. Journal of Economic, Analysis and Policy 9 (1): 1–66. Van Biesebroeck, J. 2003. Revisiting Some Productivity Debates. National Bureau of Eco- nomic Research. Vogel, A., and J. Wagner. 2010. “Higher Productivity in Importing German Manufacturing Firms: Self-Selection, Learning from Importing, or Both?” Review of World Economics 145 (4): 641–665. Vogel, A., and J.Wagner. 2011. Robust estimates of exporter productivity premia inGerman business services enterprises. Working Paper Series in Economics 207, University of Lu¨neburg, Institute of Economics. Wooldridge, J. 2009. “On Estimating Firm-Level Production Functions Using Proxy Vari- ables to Control for Unobservables.” Economics Letters 104 (2009): 112–114. Yahmed, S., and S. Dougherty. 2012. “Import Competition, Domestic Regulation and Firm- Level Productivity Growth in the OECD.” OECD, Economics Department Working Paper No. 980. Paris: Organisation for Economic Co-operation and Development. Appendix 1: China export over time Total world export CN export Year ($US billion) ($US billion) Growth Note 1998 5158 184 1% 1999 5220 195 6% 2000 6010 249 28% 2001 5830 266 7% (WTO accession) 2002 6190 326 23% 2003 7240 438 34% 2004 8780 593 35% 2005 9940 762 28% 2006 11,600 969 27% 2007 13,200 1220 26% 2008 15,300 1430 17% GFC 2009 11,900 1200 –16% GFC 2010 14,400 1580 32% 2011 15,000 1900 20% Average 10,047 808 19% Source: UN Comtrade database D ow nl oa de d by [D r T uy en T ran ] a t 0 6:2 9 2 1 M ay 20 15 The Journal of International Trade & Economic Development 23 Appendix 2: Vietnam import from the world and China VN import Total VN import %VN from CN VN import VN import from China import in (% in Year ($US billion) growth ($US billion) world trade total import) (1) (2) (3) (4) (5) (6) 1998 11.5 –0.9% 0.51 0.22% 4% 1999 11.7 1.7% 0.67 0.22% 6% 2000 15.6 33.3% 1.40 0.26% 9% 2001 16.2 3.8% 1.61 0.28% 10% 2002 19.7 21.6% 2.16 0.32% 11% 2003 25.3 28.4% 3.14 0.35% 12% 2004 32.0 26.5% 4.60 0.36% 14% 2005 36.8 15.0% 5.90 0.37% 16% 2006 44.9 22.0% 7.39 0.39% 16% 2007 62.8 39.9% 12.71 0.48% 20% 2008 80.7 28.5% 15.97 0.53% 20% 2009 69.9 –13.4% 15.41 0.59% 22% 2010 84.8 21.3% 20.20 0.59% 24% 2011 107.0 26.2% 24.59 0.71% 23% Average 44.2 18% 8.3 0.41% 15% Source: UN Comtrade and GSO website, column 3 is calculated from column 2 capturing nominal growth rates. Appendix 3: Tech level industry groups Group 1: Low technology D15: Food and beverages D16: Cigarettes and tobacco D17: Textile products D18: Wearing apparel, dressing and dying of fur D19: Leather and products of leather; leather substitutes; footwear. D20: Wood and wood products, excluding furniture D21: Paper and paper products D22: Printing, publishing and reproduction of recorded media D23: Coke and refined petroleum products and nuclear fuel D36: Furniture and other products not classified elsewhere D37: Recycles products Group 2: Medium technology D24: Chemicals and chemical products D25: Rubber and plastic products D26: Other non-metallic mineral products D ow nl oa de d by [D r T uy en T ran ] a t 0 6:2 9 2 1 M ay 20 15 24 T. Doan et al. D27: Iron, steel and non-ferrous metal basic industries D28: Fabricated metal products, except machinery and equipment Group 3: High technology D29: Machinery and equipment D30: Computer and office equipment D31: Electrical machinery apparatus, appliances and supplies D32: Radios, television and telecommunication devices D33: Medical equipment, optical instruments D34: Motor vehicles and trailers D35: Other transport equipment D ow nl oa de d by [D r T uy en T ran ] a t 0 6:2 9 2 1 M ay 20 15