Does Trade Stimulate Innovation? Evidence from Firm-Product Data. November 23, 2009

Transcription

1 Does Trade Stimulate Innovation? Evidence from Firm-Product Data Ana M. Fernandes a The World Bank Caroline Paunov b OECD Development Centre November 23, 2009 Abstract Innovation plays a crucial role in growth and welfare. Exposure to trade may have a significant impact on innovation. This paper investigates whether increased import competion leads firms to improve the qualy of their products. The econometric analysis relies on a rich dataset of Chilean manufacturing firms and their products. Product qualy is measured wh un values (prices) and industry-level transport costs are used as an exogenous proxy for import competion. The estimates show a posive and robust effect of import competion on product qualy. As explic evidence that the estimated increases in un values capture product qualy upgrading, we show that those increases are higher in cases where there is greater scope for qualy differentiation. We find that reduced transport costs on intermediate inputs also lead to qualy upgrading, whereas neher the effects from improved access to export markets nor strategic product market posioning explain our results. Interestingly, we find that import competion provides catch-up opportunies for firms wh relatively lower qualy products, lower productivy levels, and smaller size so as to reduce the gap to frontier producers. Keywords: import competion, transport costs, incremental innovation, product qualy, output un values, firm-level data, multi-product firms, Chile. JEL Classification codes: O31, F14, L6. a Ana Margarida Fernandes. The World Bank. Development Research Group H Street NW, Washington DC, afernandes@worldbank.org. b Caroline Paunov, OECD Development Centre, 2, rue André Pascal, Paris Cedex 16, France. caroline.paunov@oecd.org and caroline.paunov@gmail.com. This paper is a substantially modified version of the World Bank Policy Research Working Paper 4894 entled Does Tougher Import Competion Foster Product Qualy Upgrading?. The authors would like to thank Beata Javorcik, Jonathan Haskel, Richard Disney, and seminar participants at the Edinburgh Workshop on Firm-Level Data Analysis in Transion and Developing Economies, the Third Conference on Micro Evidence on Innovation in Developing Economies, the Chicago Fed, and the 4 th ISGEP Conference at the Universy of Valencia for valuable comments. The authors thank Eric Verhoogen for sharing his data on the advertizing intensy of U.S. industries and Francis Ng for help wh COMTRADE data. The findings expressed in this paper are those of the authors and do not necessarily represent the views of the World Bank or the OECD.

2 1. Introduction What are the dynamic effects of globalization on manufacturing producers in emerging economies? The evidence on trade and growth at the micro level has focused generally on the effects of trade liberalization on firm productivy levels and growth. 1 There are, however, other interesting dimensions along which firms adjust to globalization including innovation which plays a crucial role for growth and welfare (Grossman and Helpman, 1991; Aghion and Howt, 1998). Minor or incremental innovations in particular have been shown to be important drivers of growth in emerging economies (Puga and Trefler, 2009). This paper investigates whether increased exposure to import competion stimulates incremental innovation reflected in product qualy upgrading for firms in an emerging economy. While the idea of linking import competion to product qualy upgrading is appealing, s empirical implementation faces two challenges. The first challenge concerns the measurement of product qualy. To address, we explo a new dataset including rich information from census data on all the products manufactured by all Chilean firms during the period. We follow the empirical trade lerature and use un values (prices) of products to measure their unobserved qualy or sophistication. 2 Increases in un values correspond well to the definion of incremental innovation in the OECD Oslo manual (1997) which covers existing product[s] whose performance has been significantly enhanced or upgraded. 3 The second challenge concerns the difficulty in identifying causal effects of import competion on qualy upgrading as upgrading can self affect whether and how much foreign competors choose to export to the domestic market. This challenge is particularly relevant in the case of our sample period for Chile 1 See Harrison (1994), Pavcnik (2002), and Fernandes (2007) among many others. Tybout (2000) provides a survey of the empirical lerature. 2 Iacovone and Javorcik (2008), Kugler and Verhoogen (2008), and Lelarge and Nefussi (2008) use data on un values of domestic or exported products to proxy for product or export qualy at the plant level, while Kiyota (2008) and Schott (2008) use data on un values of exports to proxy for export qualy at the country level. 3 Pietrobelli and Rabellotti (2006) also equate product qualy upgrading wh innovation to increase value-added incremental innovation. 1

3 during which a gradual and continuing process of trade integration was under way. To address, we rely on an effective trade barrier measure - transport costs - which proxies for differences in import competion across industries that are exogenous to qualy upgrading. While we are not the first to use transport costs, we demonstrate why that measure is indeed a useful proxy for import competion both in general as well as in the specific case of Chile. Our exercise contributes to the lerature by providing evidence on the impacts of trade openness on innovation beyond the inial effects of radical liberalization. The econometric approach explos variation in transport costs across 4-dig industries and over time and consists of regressions of product un values on a lagged measure of transport costs, firm-product and industry-year fixed effects, as well as a set of firm and industry control variables. Importantly, our specifications identify the impact of transport costs by comparing un values of a given product whin firms as transport costs change. No attempt is made to distinguish higher-qualy from lower-qualy products since differences in uns of measurement and other characteristics across products preclude a direct comparison of their un values. Our main finding is that stronger import competion leads to significant qualy upgrading by Chilean firms. This result is obtained in regressions that use transport costs measures directly and is also confirmed in regressions that use import penetration ratios instrumented by transport costs measures. As explic evidence that the estimated increases in un values due to tougher import competion indeed pick up improvements in product qualy, we show that those increases are higher in industries wh greater scope for qualy differentiation and for firms wh greater absorptive capacy. We show that our main finding is robust to a variety of tests such as the use of alternative outlier creria or alternative transport costs measures, and the inclusion of addional or alternative control variables. Interestingly, we find that import competion provides 2

4 catch-up opportunies for firms wh relatively lower qualy products, lower productivy levels, and smaller size so as to reduce the gap to frontier producers. Our findings can have possible alternative interpretations. We consider these in turn and evaluate their validy. First, our results could be picking up the impact of the access to better production inputs on product qualy. Accounting explicly for the effects of intermediate inputs transport costs, we show that the importance of stronger import competion for qualy upgrading is maintained, even if the access to better inputs also has a significant posive effect. Second, our main findings could be explained by a strategic product market posioning story consisting of a decline in the qualy of Chilean imports due to reductions in transport costs (as predicted by Hummels and Skiba (2004)) and Chilean firms improving the qualy of their products to fill out the product space (as predicted by Vogel (2008)). However, the story is not supported by our data as the qualy of Chilean imports (proxied by the un values) did not change wh transport costs in a manner consistent wh the Hummels and Skiba (2004) prediction. Third, our results could reflect an increase in access to export markets if reductions in transport costs for Chilean imports are correlated wh reductions in transport costs for Chilean exports. We do not find supportive evidence for this theoretically plausible explanation as the effects of import competion on product qualy are actually stronger for firms that do not export and for products that are not exported. Fourth, a further concern about our results is that they seem at odds wh the importsas-market-discipline hypothesis which predicts a negative effect of import competion on pricecost margins (Levinsohn, 1993; Melz and Ottaviano, 2008). Since radical trade liberalization in Chile occurred in the early 1980s, we would not expect the pro-competive price-lowering effects as a reaction to imports to still play a major role during our sample period. Indeed, we are able to dismiss those concerns based on our estimation of the link between transport costs and 3

5 price-cost margins of Chilean firms following the widely used methodology proposed by Roeger (1995). Our paper relates to the debates in two strands of the lerature. First, theoretical and empirical studies on product market competion and innovation are unclear about the sign of that relationship (Ahn, 2002). In a seminal contribution, Schumpeter (1942) argues that producers facing less competion are best placed to innovate since getting adequate returns for one s innovation requires some form of temporary monopoly power. In contrast, strong competion may foster innovation as producers need to escape their innovating peers to stay in business (Nickell, 1996). Aghion et al. (2005, 2006) predict and show evidence of an inverted U-shaped relationship between competion and innovation based on a model which allows for counteracting escape competion effects as well as Schumpeterian effects of competion on innovation depending on firm or industry distance to the technological frontier. 4 Gorodnichenko et al. (2009), however, find no support for the inverted U-shaped relationship. Second, the theoretical lerature on whin-firm adjustment to increased import competion is ambiguous about the incentives for firms to invest in productivy-enhancing technology and to innovate. In Goh (2000) import competion increases these incentives by reducing the opportuny cost of technological effort and in Thoenig and Verdier (2003) results in defensive skill-intensive innovations by firms desiring to reduce future threats of imation or leapfrogging by competors. 5 Considering dynamic heterogeneous firm models, Ederington and McCalman (2008) show that trade openness increases the rate of technology adoption by firms while Atkeson and Burstein (2007) show that trade openness can have a substantial posive impact on process innovation decisions. In contrast, Rodrik (1992) argues that by reducing the firm s 4 Focusing on managerial effort (which may influence innovation outcomes), Schmidt (1997) equally shows that the relationship wh product market competion is ambiguous: a posive incentive to exert more effort is provided by the increased probabily of ex but that is counteracted by a negative incentive from the reduction in profs. 5 Goh (2000), Thoenig and Verdier (2003), and Rodrik (1992) consider representative firm models. 4

6 market share, import competion actually decreases s incentives to innovate, reviving the arguments of Schumpeter (1942). Empirical studies examining the effects of import competion on innovation outcomes at the micro level are rare and those available differ in important aspects from ours. Bertschek (1995) and Baldwin and Gu (2004) examine the effect of import competion on German and Canadian firms involvement in product upgrading or innovation measured by an affirmative answer to the question: Did you introduce new or significantly improved goods. 6 Lelarge and Nefussi (2008) study the link between import competion from low-wage countries and French firms research and development (R&D) spending. Bloom et al. (2009) examine the effects of competion from Chinese imports on patents for large European firms. To the best of our knowledge only two studies - Bustos (2009) and Teshima (2009) examine the effects of import competion on the innovation behavior of firms in developing countries relying on survey data and focusing mainly on firm-level R&D spending. 7 Our study s contributions to the lerature are three-fold. First, ours is the first paper to examine the impact of import competion on a measure of incremental innovation at the firm level for an emerging economy. In emerging economies where most producers lag behind the world s technology frontier and often improve upon products imported from developed countries, incremental innovation is a much more prevalent type of innovation than radical R&D-intensive innovation as considered in previous studies. 8 We measure incremental innovation using direct quantative information on product prices rather than subjective perception-based measures of 6 Alvarez and Robertson (2004) use a similar question when relating innovation outcomes to alternative dimensions of openness (foreign direct investment and exports) for a cross-section of Chilean and Mexican plants. Gorodnichenko et al. (2009) link these alternative dimensions of openness to a broader but also perception-based definion of innovation outcomes that includes also the adoption of new technologies for a cross-section of firms across several Eastern European countries. 7 Bustos (2009) uses a measure of expendures on R&D, computers, software, technology transfers, and patents. 8 Moreover, changes in un values capture an output from incremental innovation in contrast to R&D spending which is an innovation input which may or may not generate an innovation output. 5

7 product upgrading as in previous studies. Second, we analyze the effects of import competion on qualy upgrading for the universe of Chilean manufacturing products whereas most previous studies focus on exported products. Since 86 percent of the products manufactured by Chilean firms are sold only in the domestic market, this feature of the analysis is important. Furthermore, exported products may differ in many respects from domestically sold products, thus estimates obtained focusing exclusively on the former may be biased. Third, our identification of the effects of import competion on product qualy relies on the use of a measure of transport costs that can be considered to be exogenous to qualy upgrading. Our findings indicate that increased exposure to import competion may be beneficial by encouraging producers to follow the high road to competiveness (Pietrobelli and Rabellotti, 2006). Considering potential labor market outcomes, the evidence suggests, along the lines of Verhoogen (2008), that globalization may result in an increase in whin-firm wage inequaly if qualy upgrading self requires an increase in the demand for skilled labor. Taking into account the evidence provided by Iacovone and Javorcik (2008) that Mexican firms invest in product qualy upgrading before they export, our findings also suggest that over time firms - including those wh no export experience - may be able to progressively target more sophisticated export markets. Moreover, our results indicate that competive pressure from imports may strengthen catching-up opportunies and does not have to be a force leading to divergence in firm innovation outcomes. The remainder of the paper proceeds as follows. Section 2 describes the data. Section 3 presents the empirical specification and discusses the use of transport costs as a proxy for import competion and the exogeney of the measure. Section 4 discusses our main results, robustness tests, and evidence of qualy upgrading. Section 5 presents alternative explanations for our results. Section 6 examines heterogeney in the effects of transport costs. Section 7 concludes. 6

8 2. Data 2.1. Firm Un Values and Other Information In our analysis, we use a dataset wh information on products at the firm level from 1997 to 2003 that is merged wh the annual manufacturing census of Chilean firms wh more than 10 employees (ENIA). 9 Both datasets are provided and collected by the Chilean National Statistical Office. The products dataset includes information for each firm and year on the physical quanty sold and the sales value of each of 1,817 products at the 7-dig ISIC level (revision 2). Appendix Table 1 provides some examples of 7-dig ISIC level categories to illustrate the level of detail of the products. The ENIA census described in detail in Fernandes and Paunov (2008) is an unbalanced panel of firms capturing entry and ex that includes information on basic firm characteristics such as employment or ownership, and on accounting variables such as sales. For each product p7 of firm i in year t we construct a un value as UV S Q, p7 p7 p7 where S is the value of sales and Q is the physical quanty sold. A un value measures the average price charged by a firm for each product in a year. We assume that an increase in un values proxies for firm product qualy upgrading. Our dataset reports the physical quanties of the 1,817 products in 20 different measurement uns, some of which are shown in Appendix Table 1. The un values of products measured in different uns (e.g., kilogram, ler) are not comparable. To obtain our final estimating sample, we address two issues on the measurement uns of the products physical quanties: (i) some firms do not report the measurement un of their products quanty, and (ii) some firms report their products quanty in a different un than the un in which the majory of firms report product quanties. The un values of both types of firms cannot be compared to those of other firms producing the same 7-dig product and are thus 9 Note that the ENIA dataset provides information by plant and not by firm. However, according to Pavcnik (2002) more than 90% percent of firms during the period are single-plant firms. Hence plant information corresponds to a large extent to firm information and we use the term firm throughout the paper. 7

9 excluded from the final sample. Further, to eliminate potential outliers we exclude the top and bottom 5% of the distribution of un values for any 7-dig product. 10 Appendix 2 describes further the cleaning procedures used for the products dataset and some tests performed to assess the goodness of the data. Our final sample combining the products dataset wh the ENIA census includes 51,349 firm-year-product observations corresponding to 5705 firms wh the average number of products manufactured per firm being 2.1. Navarro (2008) shows that many stylized facts based on the Chilean products dataset are similar to those obtained for a U.S. products dataset by Bernard et al. (2009) and an Indian products dataset by Goldberg et al. (2009). 11 Table 1 shows average coefficients of variation in un values for selected 4-dig industries. The statistics show a substantial degree of heterogeney in un values across firms and point to some interesting differences across industries. Industries wh homogeneous products and thus less scope for qualy differences such as cement or petroleum refineries are characterized by low average coefficients of variation. In contrast, industries where qualy is expected to play a more important role such as electrical machinery, motorcycles, and professional equipment are characterized by higher coefficients of variation Transport Costs Our measure of transport costs is based on detailed information provided by the Latin American Integration Association (ALADI) on freight costs excluding insurance costs and the free on board customs value (fob) of Chilean imports for each 8-dig Harmonized System (HS) 1996 nomenclature code, exporting country, and year from 1997 to First, we compute for 8-dig HS code i from exporting country c in year t ad-valorem freight rates as the ratio of freight 10 While we base our main results on the exclusion of outliers for product categories, our main findings are maintained when the exclusion of outliers is done for product-year categories. 11 For example, the average shares of the most important product, the second most important product, and so on, in total sales of Chilean multi-product plants are strikingly similar to those of U.S. and Indian multi-product plants. 8

10 costs ( freight ict ) to the fob value of Chilean imports ( fob ict ): TC freight / ict ict fob ict. Second, we aggregate these freight rates from the 8-dig HS code, exporting country, and year level to the 4-dig ISIC (revision 2) and year level using (i) a concordance between 8-dig HS and 4-dig ISIC codes and (ii) weights given by Chile s 8-dig HS fob imports from each exporting country and year as a ratio to Chile s total imports in the corresponding 4-dig ISIC code in that year. Appendix 2 provides more details on the construction of the freight costs measure hereafter referred to as transport costs measure Empirical Framework 3.1 Baseline Specification Using the un values and transport costs measures as defined above we propose the following specification to examine the impact of import competion on product qualy: loguv p7 k 4 p7 m3 p7 * TCt 1 * X fi * I I * It (1) where loguv is the log of the un value for 7-dig product p7 manufactured by firm i in year p7 t, TC are transport costs for 4-dig industry k4 to which the firm s product p7 belongs in year k 4 t 1 t-1, X is a vector of controls to be specified below, p7 f * I are firm-7-dig product fixed i effects, m3 p7 I * I t are 3-dig industry m3-year fixed effects, and is an independent and identically distributed (i.i.d.) residual. A negative indicates a posive impact of import competion on product qualy. 3.2 Transport Costs as a Measure of Import Competion 12 Since some countries may not export a product to Chile due to prohibive transport costs, our measure is a lower bound for transport costs accounting only for those of exports that actually occur (Hummels, 2001). However, as this feature of our measure is common to products in all industries, does not impair our analysis which focuses on differences in the relative, rather than the absolute, magnude of transport costs across industries and time. 9

11 Our paper is not the first to use transport costs as a proxy for import competion: Bernard et al. (2006a) use transportation rates (combined wh tariffs) in a study of the responses of U.S. manufacturing industries (in terms of export growth, ex, and productivy growth) to stronger import competion. Notwhstanding, is important to discuss the choice of transport costs as a proxy for import competion. We present multiple reasons that support the view that transport costs are an adequate measure of import competion in general as well as in our specific case for Chile during our sample period. First, in general terms, transportation costs are an important friction to international trade. Given their size and variabily across trading partners, they play an important role in shaping patterns of trade across goods and partners. For example whin disaggregate product categories, exporters wh the lowest freight rates are shown to have the largest import shares based on data for the U.S., New Zealand, Argentina, Brazil, Chile, Paraguay, and Uruguay (Hummels, 2001). Transport costs represent currently a greater share of total trade costs than the tariffs for most countries given the trade liberalization efforts of the last decades (Anderson and Wincoop, 2004). For Latin American importers transport costs represent between 31 and 63 percent of total trade costs for the median product (Hummels et al., 2009). Second, Chile reduced tariffs significantly since s trade liberalization in the 1980s and imposed a uniform tariff structure across industries. 13 As such, tariffs do not capture differences in import competion across industries. Hence, transport costs capture more appropriately the differential obstacles to trade faced by Chilean industries (Moreira and Blyde, 2006). More importantly, focusing on Chilean industries during our sample period, we show that reductions in transport costs led indeed to increases in import competion. Table 2 presents the results from 13 During our sample period Chile s uniform tariff continues to decline from 11% in 1996 to 7% in 2003 but wh no variation across industries. The multiple preferential trade agreements that Chile entered into since the 1990s discussed in Chumacero et al. (2004) introduced a complex set of product- and country-specific exceptions to that uniform tariff structure that could provide useful variation for our analysis. However, such exceptions may have been subject to polical economy pressures and thus might be related to changes in product qualy in an industry. To avoid endogeney problems, we prefer not to consider those in our analysis. 10

12 estimating the specification IMP k 4 t TC * TC k 4 t 1 I I t m3 u k 4 t, where k 4 IMP t is the import penetration ratio of 4-dig industry k4 in year t, TC are transport costs as above, I t are year k 4 t 1 fixed effects, m3 I are 3-dig industry fixed effects, and k 4 u is an i.i.d. residual. 14 The results from OLS estimation in column (1) shows that as transport costs decline, import penetration ratios increase significantly. An expansion in an industry s volume of imports may lead to a reduction in s transport costs if the transportation of larger shipments benefs from economies of scale. 15 To migate this potential reverse causaly problem, we present in columns (2) and (3) the results from instrumental variables (IV) estimation where transport costs are instrumented by a variable that captures important time-varying shocks to the cost of transportation (fuel prices) interacted wh variables that allow those shocks to affect products differently depending on their transportation intensy (transport costs in the first sample year or distance weighted by the import share in the first sample year). Appendix 2 describes the instruments in detail. The instrumental variables results show a negative and significant effect of transport costs on import penetration of Chilean industries. We can therefore argue wh confidence that transport costs are an adequate proxy for import penetration across Chilean industries. 3.3 Endogeney Concerns An important observation about our transport costs measure is that is computed based on freight costs that exclude insurance costs. This is an advantage relative to the measure used by Bernard et al. (2006a) and avoids one source of endogeney as insurance costs increase wh the value - and likely wh the qualy - of exported products. We should also note that our transport costs measures capture the costs incurred by imports from the exporting country until 14 The import penetration ratio is given by the ratio between imports and the sum of total imports and total domestic sales in the industry as detailed in Appendix Hummels (2007) argues that scale economies led to important reductions in shipping prices over the last decades. 11

13 arrival to the point of entry into Chile. They are, therefore, immune to polical economy forces in Chile that could pressure for increased trade openness and/or investments in domestic infrastructure as a result of domestic improvements in product qualy. This is an advantage of our measure over tariff-based analyses of trade liberalization. Our focus in this paper is on a gradual and continuing process of trade integration captured by the variabily of transport costs across industries and over time. Table 3 illustrates the substantial variation in our transport costs measure over time and across 4-dig industries. 16 It is important to understand where this variabily originates and whether is exogenous to product qualy upgrading in Chile. One explanation is that originates in the variabily in the underlying freight rates TCict across exporting countries and 8-dig HS products. 17 If technological progress (e.g., containerization in marime shipping and cost reductions in air shipping) was the only explanation for differences in freight rates, then one could be concerned that the large variabily observed across exporting countries and products could be due to measurement error in the data on Chilean freight rates and more importantly that could be spuriously correlated wh our measures of product qualy through a general effect of technological progress. However, that concern is not valid for two reasons. First, the variabily in freight rates across products and partner countries is not a Chilean but rather a universal feature (Hummels et al., 2009). Second, many factors exogenous to Chilean product qualy help explain that variabily in those freight rates: infrastructure qualy in the exporting country, the distance shipped, the mode of transportation used and s qualy, market power in the shipping industry, 16 For certain industries transport costs increase in parts of the sample period which could be due to increases in fuel costs, port congestion, and their differential effect across countries and products despe technological progress. 17 The coefficient of variation of the ratio of TCict to the average of TCict by 8-dig HS product and year is 0.99 and that of the ratio of TCict to the average of TC ict by exporting country and year is

14 and the type of product being shipped (e.g., heavy versus light) (Limao and Venables, 2001; Hummels, 2007; Hummels et al., 2009). 18 Another explanation is that this variabily partly originates in composional effects given the way in which the 4-dig transport costs measure is constructed (as described in Section 2.2). A change in the composion of the exporters or of the products exported to Chile can change the transport costs measure even if all disaggregate freight rates are unchanged. This would be problematic if the change in exporters or in products responded directly to qualy upgrading in Chile. We examine two resulting reverse causaly arguments below. A first reverse causaly argument would be that domestic improvements in product qualy motivate certain countries to stop exporting or to export smaller volumes to Chile. This would bias downward our estimated effect of transport costs on qualy - making appear smaller than what might be in realy - only if these changes led to a reduction in our transport costs measure. By construction our transport costs measure would decline if (i) the countries that stop exporting had relatively high freight rates and represented a large share of imports in the industry, and/or (ii) the countries sending smaller volumes faced as a result lower freight rates and represented a substantial share of imports in the industry (that would remain large even as exports were reduced). Regarding (i), our data shows that few of Chile s important import relationships cease during the sample period. 19 Regarding (ii), since we would expect economies of scale in transportation to be important, a reduction in shipment size would very likely be associated wh higher rather than lower freight rates. Hence this reverse causaly argument would if anything result in an increase in our transport measure and thus would bias our estimates upward making more difficult to obtain a negative effect of transport costs on qualy. 18 The mode of transportation self depends on the product exported. Moreover, different countries adopt different modes of transportation at different times due to scale and factor price reasons (Hummels, 2007). 19 Considering all of Chile s import relationships at the country-4-dig industry level 4,960 pairs, only 49% last the entire sample period. Focusing on the top 10 exporting countries to Chile for each 4-dig industry, 4,400 out of 4,764 observations (94%) correspond to relationships that last the entire sample period. 13

15 A second reverse causaly argument would be that domestic improvements in product qualy lead countries to increase the qualy of their exports to Chile. But equally likely would be the possibily that countries start exporting lower qualy products to Chile by considering that the market there had become too tough on the higher-qualy end. As in the previous argument, changes in Chilean import qualy would bias downward our estimated effect of transport costs on qualy - making appear more negative than what might be in realy - only if they led to a reduction our transport costs measure. An increase (decline) in the qualy of Chilean imports proxied by an increase (decline) in the imports un values - the ratio of imports fob to the quanty imported - would occur if imports fob increased (declined) more than the quanty imported. Since imports fob are also the denominator of our ad-valorem freight rates, our transport costs measure would therefore decline (increase) in the case of an increase (decline) in the qualy of imports. As mentioned earlier a change in response to domestic qualy upgrading that results in an increase in our transport measure - which is the case if Chilean import qualy declined - would simply make more difficult to estimate a negative effect of transport costs on qualy and is thus not a source of concern. To address the potential reverse causaly concern from an increase in Chilean import qualy, we examine the actual changes in Chilean imports qualy during our sample period using un values at the 4-dig ISIC and exporting country level based on data from COMTRADE. 20 The median and average annual changes in import un values are 0% and 0.9%, respectively. Hence, does not appear as if the qualy of imports to Chile changed substantially over the sample period as a reaction to domestic qualy upgrading. 20 Un values are computed as the ratio of Chilean imports fob to the quanty imported from COMTRADE for each 6-dig HS product and each exporting country between 1997 and The COMTRADE data on imports fob is similar to the ALADI data used to construct our transport costs measure summed up from the 8-dig to the 6-dig HS level. Note that quanties are not available for all 6-dig HS products and exporting countries. Over the sample period quanties are missing for an average of 51% of the product-exporting country pairs for which imports fob are available. Hence while the results based on un values from COMTRADE are the only evidence we can provide on the qualy of imports, they need to be viewed wh caution. 14

16 This suggests that concerns about reverse causaly based on the changing qualy of imports are not too serious in the case of Chile. Our conclusion is that the two reverse causaly arguments linked to the construction of the transport costs measure are eher unlikely or work in our favor i.e., making more difficult to estimate a negative effect of transport costs on qualy. Notwhstanding, in order to migate any remaining potential endogeney concerns and to allow the effect of import competion to occur wh a lag, we follow Bernard et al. (2006a) and include a one-year lag of the variable TC as shown in Equation (1). 3.4 The Level of Disaggregation of Transport Costs Transport costs in Equation (1) are measured at the 4-dig ISIC level. A measure at a more aggregate level may not adequately capture the degree of import competion faced by firms. For example, 3-dig ISIC industry 311, food manufacturing, includes 4-dig industries ranging from fru and vegetable canning to bakery. If we considered a transport costs measure at the 3-dig ISIC level, a reduction in the transport costs of imported bakery products would erroneously suggest that fru and vegetable canning products also faced stronger import competion, when such products are not exactly substutes. Certainly, one could argue that measuring transport costs at the 4-dig ISIC level for bakery products (ISIC 3117) is still too aggregate. At that level, the measure implies that a reduction in the transport costs of cookie products strengthen the competion faced by cake products too. Cake products may indeed be challenged by imports of cookie products because consumers may decide to substute cake for cookie products. If import competion was measured at a more disaggregate level - i.e., distinguishing cake from cookie products - then one might wrongly ignore that cross-effect. 15

17 Hence, we consider 4-dig to be an adequate level at which to measure the degree of import competion as allows for a reasonable degree of substutabily across products. Importantly, using a transport costs measure at the 4-dig ISIC level also reduces potential concerns of a spurious correlation between reductions in transport costs and qualy upgrading in Chile due to demand shocks. Suppose that pencils are a 7-dig ISIC product manufactured by Chilean firms and that a posive demand shock for pencils increases their prices (un values) domestically. 21 Suppose also that transport costs were measured at the 7-dig ISIC level. As response to the demand shock foreign producers could react by exporting more pencils to Chile, this would likely reduce the transport costs on pencils (eher because freight costs themselves decline due to scale economies or because imports fob increase). This would result in transport costs and qualy (both at the 7-dig ISIC level) being spuriously correlated due to the demand shock. The consideration of a transport cost measure at the 4-dig ISIC level rather than at the more disaggregated 7-dig ISIC level helps to avoid this type of spurious correlation problem. Given our inabily to account for demand shocks in any other systematic way, this constutes a central reason for our choice of a transport costs measure at the 4-dig ISIC level. 3.5 Addional Remarks on the Empirical Specification We now discuss other issues associated wh the estimation of Equation (1). First, we need to account for the fact that 49 percent of Chilean firms manufacture multiple 7-dig products. Among these multi-product firms in any given year, 55 percent manufacture products whin a single 4-dig industry whereas the rest manufacture products across at least two different 4-dig industries. 22 Since transport costs are measured at the 4-dig level, our specification allows firms 21 A similar argument can be made for a worldwide posive demand shock for pencils. 22 Thus, in any given year about 78% of Chilean plants manufacture products whin a single 4-dig industry. 16

18 manufacturing 7-dig products in various 4-dig industries to face a different degree of import competion in each of the 4-dig industries to which their products belong. Second, Equation (1) controls for crucial firm-product fixed effects. Firms differ in the diversy of products they manufacture, in the ways in which they manufacture them, and in the type and qualy of management which can affect their incentives and possibilies for qualy upgrading, possibly differently across products. Firm-product fixed effects help to account for this unobservable firm-product heterogeney. Note that the consideration of this type of fixed effects is motivated by the presence of multi-product firms in the sample. For these firms, the fixed effects allow us to account for unobserved firm-product specific expertise. The use of firmproduct fixed effects implies that our main coefficient of interest is identified on the basis of whin-firm changes in the un value of a given product as transport costs change. Third, un values reflect a combination of qualy and cost attributes such as input prices. Specifically, higher costs of production at the firm level may, depending on the level of competion in the market, lead to increases in un values unrelated to qualy improvements. Controlling for costs is therefore important for our main specification and we include the following proxies for production costs: average wages paid by the firm, the share of skilled labor in the firm s total workforce, un prices paid for electricy by the firm, and the share of imported material inputs in total firm material inputs. 23 However, the downside to including these controls is that doing so raises some endogeney concerns. For this reason we will examine whether our results hold when these cost controls are excluded and we will show that this is indeed the case. Fourth, since un values are prices, their increase may reflect to some extent an increase in a firm s market power. Moreover, firm size may play a role for qualy upgrading by allowing the corresponding fixed costs to be spread over a larger scale and granting easier access to the 23 Appendix 2 provides details on these four variables. 17

19 financing necessary for upgrading, mimicking the role that size plays for radical innovation (Cohen, 1995; Cohen and Klepper, 1996). To address these possibilies, the vector of controls includes a measure of the firm s market share in each of the 4-dig industries to which s products belong and three size dummies based on the firm s total employment. 24 Fifth, omted variables at the industry level correlated wh transport costs but also wh product qualy could bias the estimate of. The knowledge spillovers generated by FDI in an industry could drive firms, particularly those domestic-owned, to upgrade product qualy. In this case omting FDI from our specification could bias downward the estimate of. However, higher FDI in an industry could also have a negative effect on qualy upgrading by domesticowned firms through market-stealing effects. In this case omting FDI from our specification could bias upward the estimate of. Import competion may also be correlated wh domestic competion in the industry. If stronger domestic competion in an industry has escape effects as in Aghion et al. (2005), then is likely associated wh qualy upgrading in that industry. Foreign exporters have an incentive to send to Chile products for which local substutes have lower qualy since is easier to compete wh those. Thus, omting domestic competion from our specification could also result in a downward bias in the estimate of. To control for these possibilies, we include measures of FDI and domestic competion in the vector of controls: the share of total employment in the firm s main 4-dig industry accounted for by foreign-owned firms and the Herfindahl index for each of the 4-dig industries to which the firm s products belong. 25 Similarly, omted variables at the firm level correlated wh transport costs but also wh product qualy could bias the estimate of. The recent trade models of Bernard et al. 24 The size dummies are defined in Appendix Since total employment of a plant is not allocated across the production of each of s products, the share of total employment accounted for by foreign-owned plants is computed for the plant s main 4-dig industry, which is for multi-product plants the industry to which the major product belongs. The major product accounts for the largest share (which could be less than 50%) of the plant s total sales. 18

20 (2009) and Nocke and Yeaple (2009) predict differences in productivy across multi-product and single-product firms. Goldberg et al. (2009) show important differences across the two types of firms in India in terms of export orientation, productivy, and sales. Moreover, Bernard et al. (2006b) show that multi-product firms can adjust their product range and thus respond differently to increased trade openness relative to single-product firms. We include in the vector of controls a dummy identifying multi-product firms to account for potential differences in their un values relative to those of single-product firms. Finally, technological progress, inflation, or other shocks experienced by Chilean industries are accounted for by including 3-dig industry-year fixed effects. These may in particular account for different trends in the prices of materials or capal goods faced by firms in different 3-dig industries which could affect the prices at which they sell their final products. 26 We believe that Equation (1) allows us to identify an unbiased effect of import competion on product qualy upgrading at the firm level due to the exogenous nature of transport costs and the set of control variables and fixed effects included Results 4.1. Main Results Table 4 presents the results from estimating Equation (1) considering all firms and products in Panel A. We consider two types of Huber-Whe standard errors robust to 26 IMF (2008) shows that the recent commody price boom (wh the exception of copper and oil) began only after the end of our sample period. Our year and industry-year fixed effects account for possible increases in the prices of copper and oil in the last two sample years which could have affected final products un values. Regarding oil, we also estimate Equation (1) for a sample excluding industries 353 (petroleum refineries) and 354 (manufacture of miscellaneous products of petroleum and coal) and find similar results relative to those discussed in Section Active innovation promotion programs may affect firms incentives and possibilies to engage in qualy upgrading. However, our specification would need to account for such programs only if they targeted specific industries and could therefore be systematically correlated wh transport costs. The Chilean National Fund for Technological and Productive Development (FONTEC) - a public program in place since helped finance innovation projects for manufacturing firms (Benavente et al., 2007). However, the program did not target specific industries whin the manufacturing sector. 19

21 heteroskedasticy. Since our regressions explain firm-product un values wh more 4-aggregate transport costs measures, we follow Moulton (1990) and use standard errors clustered at the 4- dig industry and year level. But we also allow inference to be done based on standard errors clustered by 4-dig industry that allow for serial correlation whin 4-dig industries. Both types of standard errors are shown in Table All specifications include firm-product fixed effects and 3-dig industry-year fixed effects. The estimates in column (1) show that import competion has a posive effect on product qualy when firm cost controls, other firm characteristics, and industry characteristics are ignored. In column (2), the specification includes only firm characteristics in addion to transport costs. The estimate of is negative and significant and s magnude is unchanged. Columns (3) and (4) show the results from specifications where in addion to transport costs eher only industry characteristics or only firm cost controls are included, respectively. The estimates of are negative, significant, and very close to that in column (1) suggesting that these factors do not substantially affect the results. Column (5) shows our preferred specification which includes the three types of controls. 29 The estimate of implies that a one percentage point reduction (slightly less than a one standard deviation) in transport costs would lead to an average increase in un values of almost 5.2% whin firms and products. 30 Such reduction in transport costs would be associated wh the following increases in actual un values: e.g., (i) from an average of USD 86 to USD 108 for bicycles, (ii) from an average of USD 227 to USD 299 for domestic ovens and (iii) from an average of 16,454 USD to 28 Our results are also robust to the consideration of standard errors clustered by product-year and by firm-year. 29 Including firm controls and specifically cost controls raises potential endogeney concerns. However, our preferred results in column (5) are maintained excluding these controls, as shown in columns (1)-(3). Note also that the control variables are contemporaneous relative to firm un values but qualatively similar results are obtained when one-year lagged control variables are considered. 30 Un values are in logarhms and transport costs are measured in fractional terms, thus 5.2% is obtained by taking the exponential of the product of 3% by

22 USD 26,996 for fabricated motor vehicles. 31 By focusing on incremental product innovation outcomes, our main findings complement nicely those obtained for radical innovation outcomes as a response to increased import competion from China in Bloom et al. (2009), from Brazil in Bustos (2009), and from the rest of the world in Teshima (2009). While for brevy Table 4 and the tables thereafter do not report the estimated coefficients on the control variables included in our regressions, two findings are noteworthy. Firms wh larger market shares exhib significantly higher un values. However, this market power effect does not eclipse the importance of increased import competion in generating qualy improvements. Also, firms in industries wh stronger foreign presence exhib on average higher un values. These potential knowledge spillovers from FDI seem to complement - not eliminate - the effects of import competion on qualy upgrading. Panels B and C of Table 4 show the results from estimating Equation (1) for two different sub-samples. In Panel B, we use a sub-sample of all firms but only the products that firms neher start producing nor discontinue during their years in the sample (continued products). The effect of transport costs on product qualy is significant and more negative than in Panel A. The difference in magnudes suggests that products wh less upgrading potential are likely to be discontinued by firms and new products are also less subject to upgrading as a result of import competion after their inial introduction. In Panel C, we use a sub-sample including only firms that are in the sample during the entire sample period (continuing firms) including for each of those firms only their continued products. Transport costs have again a significant effect on product qualy that is more negative than in Panel B. This difference in magnudes suggests that the well-established products of continuing firms are more prone to qualy upgrading as a 31 These averages are for year 2000 and the un values are expressed in USD using the corresponding average peso- USD exchange rate obtained from the Central Bank of Chile. Providing an economic magnude for the average product is not possible due to the lack of comparabily of un values across products measured in different uns. 21