Market Access and the Evolution of Within Plant Productivity in Chile

This paper studies the impact of trade reforms on the evolution of plant's productivity in Chile (1979-2000). The main contribution of the paper is to construct detailed measures of trade liberalization disentangling the impact of export and import oriented policies. We find evidence of a positive impact of export oriented policies on productivity of traded sectors relative to non traded. On the other hand, the reduction of import barriers might have a positive impact on productivity in export oriented sectors, but it hurts local firms in import competing ones probably due to the existence of increasing returns.


Introduction
Trade liberalization was at the core of reform packages implemented in many developing economies during 1980's. Several empirical works have found evidence of a positive correlation between trade liberalization and productivity both at aggregated and plant's level (see Pavcnik (2002), Bernard and Jensen (2001), Clerides et al. (1998), Bernard, Jensen and Schott (2003)). In this paper we revisit the case of Chile, one of the earliest and most radical examples of trade liberalization in developing countries. We aim at testing the link between this reform and productivity in Chilean manufacturing plants. To test this relationship we rst estimate plant's TFP taking into account rm heterogeneity in terms of productivity levels and then estimate the impact of market access on plant's productivity. To do so we use the anual industry ENIA survey (encuesta anual industrial) of manufacturing Chilean's plants and the Trade and Production database provided by CEPII (centre d'etudes prospecives et d'information internationalles).
Our contribution to previous works is to disentangle policy implications on both export and import sides. We construct a measure of trade liberalization which takes into account the evolution of market access over time and across industries between Chile and its main trading partners. These measures, usually referred as "border eects", essentially capture trade diculties from the fact of crossing national border and selling or buying abroad. Considering only taris or year dummies may neglect two important features of trade policy. First, unilateral import tari reduction may not be symmetric among trade partners. Second, the role of non taris barriers, xed export costs and bilateral agreements may be an important determinant of trade ows. Taking into account the evolution of market access allow us identifying which industries benet the most, in terms of productivity gains, from import or export oriented policies.
Arguments concerning the consequences of trade liberalization do not always go in the same direction. Foreign competition is usually highlighted as a positive engine of productivity. It would press less productive rms to exit the market and surviving rms to trim their ineciencies. However, the presence of increasing returns to scale (IRS) and imperfect competition may introduce new ingredients to the model (Devarajan and Rodrik (1999), Rodrik (1992)). In particular one of the most important features of the Krugman (1980) model of economies of scales is precisely that average cost falls as output increases. In that sense, the size of local market plays an important role mapping cost structures. The monopoly power means that rms integrate their demand in their decisions and so the advantage of price setting 1 . In a country like Chile, with a population nowadays of 16 million (11 million in the 1982 census), the opportunities for scales economies in import competing sectors after a radical change in foreign market competition are likely constrained. On the empirical grounds, Antweler and Treer (2002) show that indeed scales economies do matter to explain trade patterns.
The literature also suggests learning by exporting as a plausible mechanism to explain the eects of trade liberalization on plant's productivity. Several empirical works nd evidence of ex post increasing productivity gains arising from selling in foreign markets (See Aw, Chung, Roberts (1999) for Korea, Kray (1999) for Chinese rms, and Alvarez and Lopez (2005) for Chile). The undermining theoretical channel focuses on productivity improvements resulting from knowledge and expertise gained in the export process. One possible explanation is that exporters learn from their contacts in the export market, and as a result they adopt better production methods and achieve higher productivity. Clerides, Lach and Tybout (1998) construct a dynamic model based on Baldwin (1989), Dixit (1989) and Krugman (1989). In their model rm's current productivity depends on prior export experience. As a result, learning by exporting widens the gap between the productivity of rms that enter the export market and those that sell only to the domestic market. Related to this view Grossman and Helpman (1991) develop a trade and innovation model emphasizing the role of international spillovers in the growth process. By disentangling the nature of trade policies, using specic export and import oriented policies, We might capture benets stemming from this mechanism.
Chilean dictatorship in power from 1973 to 1990, implemented a deep package of market reforms concerning every economic eld. Among them trade liberalization took place in the second half of seventies. Since the beginning of the period, all trade barriers and restrictions to trade were removed. Average nominal tari rates decreased gradually from 98% in 1973 to 10% in 1979. Specially, during nineties one of the main trade strategies of Chile was to pursuing several trade agreements with dierent countries and regions, without being tied to only one regional customs unions. Chile has signed trade agreements not only with almost all Latin American countries, but also with United States, European Union and Asia in recent years. Decomposition of labor productivity evolution; 1979=100 Graph 5.2 plots the average evolution of the weighted (by market share) and unweighted average labor productivity between 1979 and 2000. While the unweighted average labor productivity is directly related to within plant productivity, the weighted measure takes into account the gains due to the reallocation of market shares towards the most productive rms. As the graph shows, this evidence indicates that after 1987 within plant productivity gains 5 become a key mechanism rather than the reallocation process. Consequently we focus on mechanisms that modify individual productivity after a change in the exposure to trade rather than to ex-ante self-selection mechanisms that, holding individual productivity constant, alter the composition of the average weighted productivity.
Several works have estimated the eects of trade reforms on rm's productivity using specic methodologies to estimate production functions, such as Olley and Pakes (1996) (OP) or Levinsohn and Petrin (2003) (LP) (See section 4.1.1). These methodologies allow estimating production functions in a framework of rm heterogeneity. We are particularly interested in two works that have also studied the Chilean experience. Based on the ENIA Survey, Pavcnik (2002) estimates the impact of trade on plant's productivity in Chile during the period 1979-1986. She applies OP strategy and controls explicitly for simultaneity and selection issues. Using year dummy variables as proxies of trade reform (treatement eect in a dierence in dierence framework), the conclusion of Pavcnik (2002) is that aggregate productivity improvements are explained by two factors induced by trade liberalization: (a) the growth of within plant productivity in importing-competing industries and (b) the exit of less productive rms and the corresponding reallocation of market share towards most productive ones. However, as Bergoeing, Hernando and Repetto (2006) note taris rose between 1983 and 1985 (see graph 5.1 in Appendix 1). Due to the debt and recession crisis in 1982, the government increased taris and nominal averages to 26% between 1983 and 1985. Moreover these year dummies are also supposed to be a control of other macro economic shocks, namely the debt crises and other market reforms. Using year indicators in interactions with trade orientation sectors implies the implicit assumption that these macro economic shocks aect all sectors in a uniform way.
Chilean market reforms were recently revisited by Bergoeing, Hernando and Repetto (2006). They study the impact of trade and nancial liberalization on productivity gains in Chile in a longer period of time  using the LP strategy. Their results show that aggregate productivity gains come from within plant improvements over time in traded industries relative to non traded ones (during the nineties) and from the entry of more productive rms than the exiting ones. They also nd that the process of resources reallocation among incumbents play a minor role enhancing aggregate productivity. When explicitly regressing by eective taris productivity advantages of traded industries are not signicant and import-competing sectors get (signicantly) productivity gains from protection.
Unfortunately for identication issues the drop of Chilean taris was quite radical but homogeneous across industries. This is probably the reason why Pavcnik (2002) is constrained to use year dummies and the reason why Bergoeing, Hernando and Repetto (2006) can not get enough variance for their estimates. Estimating the evolution of market access (border eects) between Chile and its trading partners also allows us to identify heterogeneity across industries and time. In that sens this paper yields new ndings of trade policy implications. In order to facilitate the comparison of the results with previous works we also distinguish between export oriented, import competing and non traded sectors. We start reproducing Pavcnik's (2002) results for the full sample period. Then we run the regressions of productivity using the measures of border eects in interaction with traded sectors relative to non traded ones. First, we nd a positive signicant eect of a reduction of export barriers on productivity in both export oriented and import competing sectors. Second, we also nd evidence of a positive impact of the reduction of import diculties on productivity in export oriented industries. Finally, the regressions show that the decrease in import barriers might have a negative impact on productivity in import competing industries. This latter result implies that industries in import competing sectors may actually suer from foreign competition probably by reducing their domestic market size and the possibility to exploit increasing returns.
The rest of the paper is organized as follows. In Section 2 the estimation strategy in three steps is presented. Section 3 discusses the main estimation results. Section 4 concludes.

Estimation Strategy
The estimation strategy consists of three steps. In the rst step we estimate the production function using OLS, Fixed Eects and LP methodology to obtain the factor elasticity coecients and to calculate total factor productivity (TFP) of Chilean manufacturing plants. In the second step, we construct the measure of trade liberalization estimating the border eects coecients using a gravity model developed by Fontagne, Mayer and Zignago (2006). Finally, in the third step we estimate the impact of trade diculties regressing productivity on border eects coecients in interaction with sectors dened by trade orientation (export, import competing and non traded industries).

Step 1: Specication of production function estimations
As usually plant's TFP is calculated as the residual between the observed value added and the estimates of factors contribution. In order to do so we must estimate the production function at two digit industry level. When estimating production functions using rm panel data eventual problems concerning simultaneity and selection should be considered. Simultaneity arises because inputs demand and productivity are positively correlated. Firm specic productivity is known by the rm but not by the econometrician and panel data information usually shows that productivity is heterogeneous among rms and it evolves over time. A high productivity shock implies greater demand and consequently rms must purchase more inputs. OLS will tend to provide upwardly biased estimates of labor coecients. If capital is positively correlated with labor and labor's correlation with the productivity shock is higher than capital one, which is the realistic case, then the coecient of capital may be underestimated.
Selection problems are likely to be present because unobserved productivity inuences the exit decision and we only observe those rms that decide to stay. On the other hand, if capital is positively correlated with prots, rms with larger capital stock will anticipate higher prots and decide to stay in the market even for low realizations of productivity shocks. So at the end, there is a potential source of negative correlation in the sample between productivity shocks and capital stock. This negative correlation means a downward bias in capital elasticity estimates.
Olley and Pakes (1996) (OP) propose a methodology of three stages to control for unobserved productivity incorporating exit and investment rules derived from rm optimal behaviour. In the rst stage they use the investment rule, a function of capital stock and unobserved productivity, to address simultaneity. OP invert this investment rule to express unobserved productivity as a function of investment and capital. This inverted function is used as a proxy for productivity in the estimation. In the second stage, based on the exit rule they estimate the probability of survival conditional on available information to the rm. Following an optimal behaviour their exit rule states that rms decide to exit the market if productivity realization shocks are lower than a specic productivity cut-o which in turn is determined by capital stock and productivity. The estimates of this survival probability are used to control for selection bias. To obtain the capital coecient, we substitue the estimates of labor coecient (stage 1), the productivity function (inverted investment decision) and the survival probability (stage 2) into the production function equation.
Besides some technical dierences (such us the use of GMM criterion and bootstraps), Levinsohn and Petrin (2003) (LP) make use of this strategy and extend it showing that inputs (like electricity or materials) can be better proxies to control for unobserved productivity when one deals with simultaneity. Inputs adjust in a more exible way, so they are more likely to have better responsive to productivity shocks. Moreover, inputs usually have more non-zero observations than investment, a property that has consequences on estimation eciency. In the case of the ENIA this property is important. As LP show the risk of selection biases are signicantly reduced by considering a non balanced panel.
There are some advantages of OP-LP strategy over alternative methods. Firstly, it performs better than xed eects because it does not assume that the unobserved individual eect (productivity) is constant over time when controlling for simultaneity. Secondly, approaches based on instrumental variables can be limited by the instruments availability. Finally, it does not assume restrictions on the parameters. For instance, an alternative approach is the one developed by Katayama, Lu and Tybout (2005). They show how misleading can be the use of sale revenues to measure output in productivity accounting. Factor prices and mark-ups can produce important distortions if they are not homogeneous. However, the methodology proposed in their paper assumes constant returns to scales and neglect entry-exit process to facilitate likelihood estimates. Again both assumptions are not neutral in the case of the ENIA.
In order to maximize sample size with a reduced risk of selection we keep the LP strategy and use electricity as a proxy for unobserved productivity. In the rst step, we will estimate the following specication of a Cobb Douglas production function: (1) y it = 0 + 1 x it + 2 k it + " it Where all variables are expressed in natural logarithmics, "y it " is the value added of plant "i", "x" are variable inputs (skilled and unskilled labor) and "k" is the stock of capital. Consequently, TFP in log (a it ) is computed as the residual of this function, given by:

Step 2: Specication of Border Eects estimation
It is well known that the reduction of taris in Chile was homogeneous across industries. As a consequence we do not have variance in taris measures among industries. Even their rise in early eighties, during the deep Chilean debt crisis, was homogeneous. On the other hand, taris are not the only measure that matter in trade. One should consider bilateral agreements, asymmetries between export and import costs and industrial specic diculties to trade, not only concerning direclty policies but also home biases, tastes and the like. Actually, by considering all these issues we obtain heterogeneity in both industrial and time dimensions.
Using a gravity equation framework, we measure the diculties of bilateral trade explained by the fact of crossing the border between two countries. We apply to Chile and Chilean's trade partners the methodology developed by Fontagne, Mayer and Zignago (2005) 2 . The gravity bilateral trade equation that we estimate is based on a comparison between international (m ijs ) and intra-national trade ows (m iis ), the latter being a kind of ideal type of free trade.
m ijs : imports of industry "s" in country "i" from country "j" m iis : volume of trade within a country measured as the overall production minus total exports in industry "s" in country "i". a ij : consumer preferences of country "i" with respect to varieties produced in j. v js : the value of production in industry "s" in country "j" p j : Index price in country "j" Trade costs ( ij ) are composed by distance (d ij ) (related to transport costs), advalorem taris (t ij ) and "tari equivalent" of non tari barriers The structure of protection varies across all partner pairs and depends on the direction of the ow for a given pair. To capture this protection framework, taking the example of the US as trade partner, the following dummy structure is dened: USA-CHL ij : is a dummy variable set equal to 1 when j (T = i) is Chile and i the USA (related to imports of USA from Chile).
CHL-USA ij : is a dummy variable set equal to 1 when j (T =i) is the US and i Chile (imports of Chile from USA).
Preferences a ij are composed by a random component e ij and the coecient which represents a systematic preference for goods produced in the home country. This "home market bias" is reduced to ( i ) when the countries share the same language (L ij = 1): Combining the previous equations, we obtain the estimable equation: This equation can be estimated at country level (considering all industries) and also at industry level. From the latter estimation we obtain the global border eect measure for each industry as a weighted average of all trading partners. The border eect coecient of each import (export) ow will be weighted by the share of the ow on total imports (exports). Since we drop the constant and incorporate dummy variables for each combination, the coecients of the dummy variables can be interpreted as the border effect of each combination. For example, the exponential of the coecient of USA CHL ij multiplied by -1, exp (( 1) [ i + ]) indicates the diculty for Chilean's exporters in accessing to the US markets. The part of missing trade mainly caused by trade policy is captured in these coecients.
In the estimation we consider not only the US but also other countries trading with Chile. To determine the main trading partners of Chile we use the aggregated trade ows data of ECLAC. Between 1990 and 1999 the main destination countries of Chile manufacturing exports are Latin America (AL), the United States (USA) and the European Union (UE). At the same time, most manufacturing imports from Chile come from these countries. In the Border Eects regressions we consider nine countries of European Union , which were members through out the whole period . 3 Finally, there is a potential source of endogeneity since in step three we will use these border eect coecients to estimate the impact of trade liberalization on plant's TFP in dierent sectors. Most productive sectors or those producing high quality goods will tend to increase their trade ows and to have a smaller border eect. To address this issue we use relative wages and productivity as control variables (Zij) in the estimation of border eects. In that sense the residual measure of missing trade that is captured by the border eect will be free of productivity concerns.

Step 3: Specication of the estimation of the impact of trade liberalization on TFP
In this nal step we use the (weighted average) border eect estimated for each industry to measure the impact of trade liberalization on productivity across export and import competing sectors relative to non traded industries. We estimate the following reduced equation similar to the dierence in dierence framework implemented by Pavcnik (2002) The excluded categories are non traded sector, the year 1979 and the sector 385 5 . We are principally interested in the estimates of the vector coecient . It is usually expected a negative and signicant coecient meaning that a reduction of trade barriers has a higher positive eect on productivity in traded industries (export oriented and import competing) than in non traded ones. The vector coecient informs about the relative productivity advantage of traded industries in Chilean manufacture. 4 We classied sectors by trade orientation using 4 digit industry classication. Plants belonging to 4 digit industry which have more than 15% of exports over total production are classied as exported oriented plants; while plants belonging to 4 digit industry which have more than 15% of import penetration indicator are classied as import competing plants. The rest are considered as non traded plants. See Pavcnik (2002) for further details concerning this classication. 5 Manufacture of professional and scientic, measuring and controlling equipment not elsewhere classied, and of photographic and optical goods 13

Data
In the rst step we use manufacturing plant level data from the ENIA Survey provided by the Chilean institute of statistics INE (Instituto Nacional de Estadisticas). This survey is a manufacturing census of Chilean plants with more than 10 employees. Our data covers the period 1979-2000 and contains information concerning mainly added value, materials, labor, investment and exports (only available from 1990). The ENIA survey has been used in previous studies such as Pavcnik (2002), Liu and Tybout (1996), Levinsohn and Petrin (2003) and Bergoeing, Hernando and Repetto (2006) for dierent sample periods. We used several specic deators (at three digit Isic-Rev2 and year base 1992) for added value, exports, materials and investment. Capital series were constructed using the methodology developed by Bergoeing, Hernando and Repetto (2006). Table 5.2 (Appendix 1) shows general descriptive statistics of the sample.
In the second step we use data from "Trade and Production Database" constructed by CEPII using mainly data of the World Bank. In this compilation, production variables are completed with the UNIDO and the OECD STAN databases, and trade variables with the international trade database (BACI) available from CEPII. This database is provided at the ISIC rev2 3-digit industry level over the period 1976-1999 for 67 developing and developed countries. For distance variables, contiguity and common language, we also used the CEPII database of internal and external distances. Distances variables between two countries are measured based on bilateral distance between cities weighted by the share of the city in the overall country's population. Finally, price indexes steam from Penn World Table as price level of GDP expressed relative to United States. Following Fontagne, Mayer and Zignago (2005), we use aggregate price indexes instead of industrial wages or (the unavailable) prices at industry level to reduce potential endogeneity problems. 6 3 Results

Results step 1: T.F.P. estimations
In this step we estimate the equation (1) a Cobb Douglas production function at 2 ISIC industry level using OLS, Fixed Eects and LP strategy. Table 3.1. shows the results. As expected, LP estimates of unskilled labor elasticities are generally the lowest and those of capital elasticities the highest, meaning that the biases induced by the higher responsiveness of the labor input respect to capital are addressed. Considering LP estimates, in ve industries 7 , among them the main exporters, we can not reject at 5% the null hypothesis of constant returns to scale in the Wald test. On the other hand, industries presenting increasing returns to scale are mainly importers. For these industries the size of the market may aect positively their productivity.
After estimating production function coecients we calculate TFP as a residual measure. In Graph 3.1 the evolution of dierent measures of plant's productivity is presented: xed eects (TFP fe), LP (TFP lp), OLS (TFP ols) and the sample mean of valued added over labor (lnproductivity) . 7 Food (31) 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 tfp_lp tfp_fe tfp_ols lnproductivity As a rst robustness check of our performance measures, the graph shows that labor productivity and all TFP measures depict similar evolutions. Although the elasticities estimated by xed eects and LP show some dierences, the TFP path illustrated by both measures is very similar. Even if xed eects TFP may overestimate capital coecient and underestimate labor coecient, after computing all factors contribution the evolution of the residual is not drastically aected. Graph 3.1.2. shows the evolution of TFP (LP) by sector classied by trade orientation. Plants in export oriented sectors are in average more productive than those in import competing sectors. The productivity of non traded plants slows down during the eighties and it slightly recovers during the nineties but it is always behind the TFP of traded sectors.  1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 Export oriented Import competing Non traded

Results step 2: Estimations of Border Eects
In the second step we construct our measure of market access by estimating equation 2 to obtain border eects for ve periods at two digit industry level. 1 shows the evolution of export border eects measuring the difculties for Chilean's exporters to access foreign market. Export barriers were almost constant during the rst half of eighties and even in some industries like woods they increased during this period. Reecting asymmetries between import and export policies, export diculties have considerably diminished in all industries during nineties (except for food industry) even if taris were already very low.
Graph 3.2.2 shows the evolution of the weighted average of import border eect measuring barriers faced by UE, LA and USA to access Chilean markets. In this case, in many industries the market access diculties increased during the rst half of eighties, which seems very consistent with the raise in import taris during this period. From 1987 to the end of nineties import border eects have been drastically reduced in almost all industries with the exception of basic metals, a traditional exporter industry.  1979-1982 1983-1986 1987-1990 1991-1994 1995-1999  In order to give baseline estimation, we rst run the regressions in xed eects using year dummy indicators as a measure of trade liberalization and we obtain similar results to Pavcnik (2002). Once controlling for exit and plant's specic characteristics, trade liberalization (if captured by year dummies) has a positive impact on productivity of traded sectors (export oriented and import competing) relative to non traded ones (see Appendix 2). Now we use the weighted measures of border eects for each industry estimated in step 2 to analyse the impact of the outcome of export and import trade policies on plant's TFP. We check robustness of our results using as dependant variable TFP measured by xed eects (Table 2) and LP  (table 3) estimations. Once we use year and industry indicators to control for industry specic eects and macro economic shocks, the coecients of the rest of variables will only capture the eects of within industry productivity variation. We also use Huber-White estimator of variance to correct standard errors.
In the rst column (table 2 and table 3) we run the regression without controling for exit, entry indicators and domestic competition. Giving our framework we interpret the coecients of interactions relative to non traded sector, the omitted category. The export border eect in the two interactions terms (with the export oriented and the import competing dummies) presents a negative and signicant coecient. This suggests a positive impact of a reduction of export barriers on plant's productivity in both export oriented and import competing sectors. The regression aims at capturing within plant productivity improvements as a consequence of trade policies rather than aggregate productivity improvements coming from reallocation and rm renewal. In that sense what we observe may be related to externalities captured after exporting such as learning by exporting and knowledge spillovers coming from international markets (Aw, Chung, Roberts, 1999).
Regarding the called "import competing sectors", this positive eect of the reduction of export barriers can be driven by exporters inside these sectors. It is well documented in rm heterogeneity literature that even in narrow dened industries exporters and importers compete with some degree of dierentiation. The reduction of export costs will allow new rms start exporting. Bergoeing, Micco and Repetto (2005) show that there were many plants that started exporting during the nineties in Chilean industries having a small aggregate export share.
Concerning the impact of import barriers, the results depend on the sector. We nd evidence of a negative eect of a reduction of import barriers on productivity of plants belonging to import competing sectors (the interaction between import border eect and import competing sector). Therefore, increasing foreign competition will dampen the productivity of plants in these sectors. As suggested by Bergoeing, Hernando and Repetto (2006) this fact may be related to increasing returns. We observed in the estimates of production functions that import competing industries present in general increasing returns to scale. Foreign competition reduces the market shares of domestic rms shrinking the opportunities to exploit economies of scales.
On the other hand, the reduction of the diculties of foreign exporters to access the Chilean market (the reduction of import barriers) has a positive impact on productivity in export oriented sectors (the interaction between import border eect and export oriented sector). A better access to new technologies and to high quality inputs may explain this within plant productivity improvement. In the case of import competing sectors, the negative eect of market size may be negative enough to counteract this positive outcome.
Once we control for exit (column 2), entry (column 3) and domestic competition using a Herndahl indicator of market concentration (column 4) the results remain almost unchanged. As expected, the exit indicator has a negative coecient meaning that exiting rms are less productive than those that decide to stay in the market. Exiting plants are on average 17% less productive than surviving plants. The coecient of the entry indicator is also negative indicating that new rms are roughly 5% less productive than incumbents. The coecient of domestic competition is negative (though less signicant), implying that a reduction of market concentration will enhance productivity. Finally note that the coecients of both border eects (without interaction) are positive and signicant, meaning that the improvement of market access in both sides may have negative eects on productivity in non traded sectors (put zeros in import competing and export oriented dummies). This may be explained by general equilibrium eects that should be studied in more detail 8 . By the moment we will concentrate on relative eects.
We also check robustness of these results using a moving average of border eects to take into account the possible "`lagged"' impact of trade reform on plant's productivity. It might be the case that plants do not react instantaneously to changes in trade policies. To control for this issue, we construct a moving average of four years of each border eect at industry level. For example, the border eect of the year 1982 is an average of border eects from 1979 to 1982. Therefore, in the regression of TFP on border eects we lose the three rst years (1979,1980 and 1981). The last column reports the results of this estimation. In the case of the TFP measured by xed eects (table 2) the coecients of all interactions between border eects and sector trade orientation remain signicant and with the same sign but they have a lower value. Nevertheless, when we used the TFP estimated by LP strategy (table3), the coecients of the interaction between import barriers and import competing sector is non signicant. In this last specication, all other coecients remain signicant and with the same sign.
To sump up, we nd robust evidence that traded sectors increase their productivity, relative to non traded sectors, as a consequence of export oriented policies. In the case of import oriented policies the eects on productivity depends on sectors. While export oriented sectors improve their productivity, probably thanks to the increase in the foreign demand and the easier access to imported inputs and technology, domestic plants competing with imports may suer from this foreign competition.  As we mentioned, there is a potential endogeneity problem if we want to measure the impact of border eects on plant's productivity because border eects may depend on productivity. We already addressed this issue in the previous section when we estimated the border eects controlling by relative wages and productivity measures. Therefore, we expect that our estimates of border eects are free of the impact of industry productivity. Furthermore, we use border eects estimates at 2 digit industry level, while the dummy of industry trade orientation is dened at 3 digits. As an additional check we run quantile regressions (Koenker and Hallock, 2001). The idea is to estimate models for conditional quantile functions, that is, quantiles of the conditional distribution of TFP expressed as functions of the observed covariates. This allows asking whether the conditional t of the mean is also representative for the median or other conditional quantiles. In that sense, if the weight of the border eect in the regression is driven by the most productive rms, this disparity should be reected at dierent quantiles. Table 4 compares the coecients in the estimation at the 25, 50 and 75 quantiles with the xed eect estimation (around the mean) and shows that the magnitude, the signicance and the sign of coecients among the dierent ts do not change our conclusions. Export*BE X -0.027*** -0.031*** -0.025*** -0.028*** (0.008) (0.006) (0.007) (0.007) Import*BE X -0.089*** -0.095*** -0.097*** -0.103*** (0.008) (0.006) (0.008) (0.007) Export*BE M -0.043*** -0.041*** -0.030*** -0.033*** (0.011) (0.009) (0.010) (0.009) Import*BE M 0.046*** 0.056*** 0.059*** 0.063*** (0 Another strategy would be to instrument the border eect with policy variables such as taris in order to keep only the (exogenous) information of trade policies. However, at the industry level we do not have too much variance for imports and for exports the data is only available from 1996 in the CEPII compilation. Moreover, as exposed above, the fact that trade policies have gone far beyond taris reduction arise the question about the potential missing information in this instrument.

Conclusions
This study addresses the eect of import and export oriented policies on the evolution of plant's productivity using Chilean data of manufacturing plants. To measure plant's TFP we estimate the production function of each two digit industry using a semiparametric methodology that takes into account the heterogeneity of productivity among rms. The main contribution of the paper is to construct an accurate measure of the outcome of trade liberalization at the industry level to disentangle the impact of the reduction of export and import barriers on plant productivity.
The incorporation of a more detailed measure of trade liberalization introduces two new results. First, the reduction of export barriers improves productivity of plants belonging to both export oriented and import competing sectors, relative to non traded sectors. As the export costs fall, more rms are able to export increasing their size and probably beneting from knowledge spillovers arising from the foreign market.
Second, the reduction of import barriers shows a positive impact on the evolution of plant's productivity of export oriented sectors relative to non traded. However, this is not the case for plants competing with foreign exporters. It seems that the reduction of import barriers might hurt the benets from increasing returns since it reduces the domestic share of local rms. This has consequence on policy implications. Trade policy should by rather focused on export oriented measures such as reinforcement of bilateral or multilateral trade agreements and reducing the xed export costs of adapting the product to the new market.
To our point of view, further research should be oriented in two directions. Firstly, the possible endogeneity issue mentioned in the previous paragraphs should be completely tackled by means of good instruments for border eects such as policy indicators at the industry level for exporter and importers over the full sample period. Secondly, theoretical and empirical eorts should be focused on general equilibrium eects by which the consequences of specic oriented policies are transmitted to the rest of the economy.