III.i. Drivers of Upgrading: Output-Side Drivers

We now turn to our central question: what are the drivers of upgrading? I categorize drivers into three groups, which can be understood with reference to the general framework above: (1) output-side drivers: factors that affect product demand curves (the D(·) functions); (2) input-side drivers: factors that affect input-supply curves (the S(·) functions); and (3) drivers of capabilities: factors that affect the the “know-how” of firms (the Λijkt, Jit, and Kijkt). This categorization is necessarily somewhat loose - some drivers fit in more than one category, and some not quite in any - but the grouping is helpful to organize the review.


3.1 Output-Side Drivers

We begin with the literature on the effects of exporting on upgrading outcomes, because the literature is perhaps the most fully developed, and then turn to other output-side drivers, including the effects of local demand from multinational enterprises, competition in output markets, and other factors.

3.1.1 Exports

Early studies on exporting and productivity - Bernard and Jensen (1995, 1999) using US data and Clerides et al. (1998) using Mexican, Colombian, and Moroccan data - find little evidence that firms increase productivity when they start exporting. Instead, the superior performance of exporters in cross-section is explained by the selection of already-higher-performing firms into exporting. The influential Melitz (2003) model was written with these results in mind and is consistent with them: under monopolistic competition and trade between symmetric countries, firms with a sufficiently high initial productivity draw enter the export market, but increases in exporting have no within-firm effects on productivity, output quality, or wages. More recent evidence, however, has found robust effects of exporting on a number of upgrading outcomes. Exports and Quality

A first-order feature of the world economy, from the perspective of manufacturing firms in developing countries, is that consumers in international markets are on average richer and more willing to pay for product quality than domestic consumers.[1] A natural corollary is that a given firm in a developing country will produce higher-quality goods for export to rich countries than for sale in its own domestic market, to appeal to richer consumers. Verhoogen (2008) develops this idea in a Melitz (2003)-type heterogeneous-firm framework.[2] In addition to non-hometheticity of consumer demand, the key theoretical supposition is that firm capability and input quality are complementary in producing output quality. It follows that more-capable firms use higher-quality inputs to produce higher-quality outputs in equilibrium. As in Melitz (2003), only more-capable firms enter the export market. An exogenous increase in the incentive to export leads plants that are already exporting to shift production toward higher-quality varieties and induces some firms that are not exporting to enter the export market. Average product quality and hence average input quality and average wages increase in more-capable firms relative to less-capable firms.[3] Empirically, the paper tests this prediction at the plant level using initial plant size as a proxy for capability (since more-capable plants grow to be larger) and examining the differential response of Mexican plants to the late-1994 peso devaluation. Initially larger plants increased exports, were more likely to acquire ISO 9000 certification (an international production standard interpreted as a proxy for product quality), and increased wages relative to initially smaller plants within the same industry. The differential response was not evident in periods without devaluations. The differential quality upgrading generates a link between trade and wage inequality, since the initially larger plants already paid higher wages and further increased wages relative to initially smaller plants within industries.[4]

This basic story has held up reasonably well and has been extended by subsequent research. One source of evidence is price correlations in more disaggregated data. Using trade-transactions data from customs agencies, several papers have documented that firms charge higher prices in richer destinations within narrow product categories. Bastos and Silva (2010) first documented this pattern in Portuguese data, and it has been shown to be robust in Chinese (Manova and Zhang, 2012), French (Martin, 2012), and Hungarian (Görg et al., 2017) data. As mentioned above, Kugler and Verhoogen (2012) document positive correlations between output prices, input prices, and plant size that suggest producing highquality outputs requires high-quality inputs, consistent with a general-equilibrium model similar to Verhoogen (2008).[5] Hallak and Sivadasan (2013) document that exporters have higher average output prices and are more likely to have ISO 9000 certification than non-exporters, even conditioning on plant size. These facts are difficult to reconcile with a model where firm heterogeneity is one-dimensional, as in Kugler and Verhoogen (2012), but fit naturally with a model they develop with heterogeneity in two dimensions: in “process productivity,” which reduces variable costs conditional on quality, and in “product productivity,” which reduces the fixed costs of producing quality (i.e. which, in the notation of Section 2.1, reduces the fixed costs, fijt, required to produce high-quality varieties). In Chinese customs data, Manova and Zhang (2012) show that, within industries, firms that export more and charge higher export prices on average also pay higher prices for their imported inputs, and Manova and Yu (2017) show that, across products within firms, export prices are positively correlated with an index of input prices, constructed using a sector-level input-output table.[6] Exploiting barcodelevel scanner data from the US, Faber and Fally (2017) find that richer households purchase products from larger firms than poorer households within detailed product categories, again consistent with the quality story. It would be valuable to investigate whether this pattern holds in barcode-level data in poorer countries as well.

An important question in this literature has been whether the upgrading response is attributable to the greater willingness of richer consumers to pay for quality or to two other mechanisms: scale effects, if for instance producing high quality requires paying fixed costs; or distance effects, if for instance per-unit shipping costs are higher for more distant destinations. A small literature has used exchange rate movements as a source of exogenous variation in export destinations to separate these channels. Using panel data on Argentinian firms, Brambilla et al. (2012) show that the Brazilian devaluation of 1999 shifted the composition of export destinations of Argentinian firms toward richer destinations, especially for those firms previously exporting to Brazil. They are thus able to separate the effect of exporting to a richer destination from exporting per se, and they find that the former is associated with an increase in skill intensity and wages while the latter is not.[7] In Portuguese data, Bastos et al. (2018) also use the initial composition of destinations together with exchange-rate movements to show that exporting to richer countries leads countries to pay more for their material inputs, again consistent with a quality story. They find no evidence that exogenous changes in exports per se or in average destination distance lead firms to pay more for inputs. Although firms may charge different markups in different markets, and this may in part explain the output-price patterns, the authors argue that differences in markups alone are unlikely to account for the response of input prices to the export shocks.

The above studies have not had access to direct information on quality and have had to draw indirect inferences from prices and other observables. In the absence of direct information on quality, it is difficult to rule out other explanations for the price patterns definitively. A small but promising literature has had access to direct quality measures, and has corroborated several of the above points. Using wine-guide quality ratings of French champagnes, Crozet et al. (2012) show that firms with higher overall quality ratings charge higher prices, are more likely to export, and export higher volumes and export to more countries. Using wine-guide ratings from Chile, research by Ana Cusolito, Álvaro Garcia-Marin, and Luciana Juvenal, summarized in Cusolito and Maloney (2018), shows that higherrated wines carry higher prices and are associated with higher material costs. Among soccer-ball producers in Pakistan, where several quality types are directly reported, Atkin et al. (2015) show that, in the cross-section of firms, larger producers produce a higher share of high-quality balls, at a higher average cost, and charge higher prices and markups. Hansman et al. (forthcoming) show that among fishmeal producers in Peru, where protein content is an observable indicator of quality, processing firms are more likely to vertically integrate by buying fishing boats when demand for quality on the export market is high. This integration arguably solves a quality-assurance problem that arises because of imperfect observability of input quality.[8]

Perhaps the cleanest study of the effect of exporting on quality choices is by Atkin et al. (2017a). The authors convinced a US-based non-governmental organization to randomize initial export contracts to Egyptian rug producers and tracked their responses. They paid a local master artisan to evaluate the quality of rugs on a number of dimensions, including the straightness of corners and how tightly packed the threads were. They find clear increases in product quality and profits among treated firms. They also find effects on productivity, to which we return below.

The idea that demand matters - in particular, that demand from richer end-consumers (at the end of value chains) matters - is reinforced by case studies of Argentinian export industries by Artopoulos et al. (2013), who find that a distinguishing feature of industry pioneers in exporting is that they had direct knowledge of end-consumer tastes in developed-country markets.[9] Relatedly, the Enterprise Maps series by John Sutton and co-authors has found that most large firms in several African countries started out as trading firms, rather than as small producers; these findings are consistent with the idea that knowledge of foreign markets is key to firm growth in developing countries (Sutton and Kellow, 2010; Sutton and Kpentey, 2012; Sutton and Olomi, 2012; Sutton and Langmead, 2013; Sutton, 2014). Exports and Technology Adoption

There is a small literature on the effect of exports on direct measures of technology and innovation. Bustos (2011) analyzes the behavior of Argentinian firms in response to a regional trade agreement. She first develops a Melitz (2003)-type heterogeneousfirm model in which firms choose between a low-fixed-cost high-variable-cost traditional technology and a high-fixed-cost low-variable-cost modern technology (as previously considered by Yeaple (2005) in a model with perfect competition and ex-ante-homogeneous firms). The theoretical predictions are driven by scale effects: the reduction of tariffs by a trading partner leads exporting firms to expand and to adopt the modern technology. Empirically, Bustos finds that sectors with greater reductions in Brazilian tariffs saw greater increases in exporting, in spending on technology, and in indicators of process and product innovation. Consistent with the theory, these effects are driven primarily by firms in the third quartile of the size distribution (just above the median) in each sector, which in the Argentinian context tend to be the ones that move from non-exporting to exporting. In Canadian data, Lileeva and Trefler (2010) reinforce the basic finding that exports lead to technology adoption. They are able to construct firm-level changes in US tariffs, and find that firms facing greater tariff reductions were more likely to adopt new technologies and to engage in product innovation. They find similar effects on labor productivity, but do not have information on capital stocks with which to estimate TFP. These effects were larger for firms that were initially less productive.[10] Exports and Productivity

In contrast to the literatures on exporting and quality or technology adoption, which consistently find positive effects of exporting, the literature on exports and productivity is mixed, possibly in part because of the measurement issues highlighted in Section 2.2.1.[11] As mentioned above, the early literature found little evidence of within-firm effects on productivity (Bernard and Jensen, 1995, 1999; Clerides et al., 1998). More recently, De Loecker (2007) compares Slovenian firms that start exporting to firms that remain only in the domestic market, matching on the propensity to export and controlling for common trends, and finds that the productivity of new exporters rises significantly, especially for firms that start exporting to richer markets. Notably, the paper modifies the Olley and Pakes (1996) procedure by including export status in the construction of the proxy for unobserved productivity in the first stage. (See also De Loecker (2011).) Other papers that have found positive effects of exporting on productivity among developing-country firms include Bigsten et al. (2004), Van Biesebroeck (2005), Álvarez and López (2005), Blalock and Gertler (2004), and Park et al. (2010). By contrast, Aw et al. (2000) find little evidence for learning-by-exporting in Korea (although they find some evidence in Taiwan), and Luong (2013) implements the De Loecker (2007) approach in China but finds no learning-by-exporting effects. (See also Lopez Cordova (2003) and ISGEP (2008).)

An important caveat about these papers is that standard TFP measures may reflect markups as well as technical efficiency, as discussed in Section 2.2.1. A recent paper by Garcia-Marin and Voigtländer (2019) addresses this issue. Using detailed plant-product data from Chile, the authors implement a variant of methods developed by De Loecker and Warzynski (2012) and De Loecker et al. (2016) (which in turn builds on insights from Hall (1988)) to estimate markups and marginal costs and investigate how they respond to exporting. Under the assumption that a first-order condition holds for at least one flexible input, the product-level markup can be expressed as the output elasticity with respect to the flexible input divided by expenditures on the input as a share of sales of the corresponding product. Assuming that materials are used across products in the same proportion as in total variable costs, the authors are able to calculate input expenditures as a share of revenues at the product level, using materials as the flexible input. After estimating output elasticities using the method of Ackerberg et al. (2015) (using single-product firms with a selection correction, following De Loecker et al. (2016)), they calculate product-level markups and use them to recover product-level measures of marginal costs, which they interpret as a measure of productivity. Using this measure and several different estimators, including a propensity-score matching estimator and an instrumental-variables (IV) estimator using tariff changes in export destinations, they find that marginal costs decline by 15-25% for new exporters. Strikingly, when the authors use a standard TFPR measure, they find no effect of exporting; they argue that because the increases in efficiency are passed on to consumers in the form of lower prices, they do not show up in revenues. This study is a notable step forward for the literature. It is also subject to the concern that it depends heavily on the accuracy of the markup estimates derived from the De Loecker and Warzynski (2012) method, which has recently been criticized by Raval (2019) and Traina (2018). In addition, the criticisms of Gandhi et al. (forthcoming) of the Ackerberg et al. (2015) method of production-function estimation (discussed above) apply here as well (Flynn et al., 2019). However, using the product-level total variable cost and output quantity information, the authors are able to calculate average variable cost at the product level and show that it is highly correlated with the marginal costs they calculate, which provides support for their method.

The most direct evidence of an effect of exporting on productivity is provided by the study by Atkin et al. (2017a) on Egyptian rugmakers, mentioned above. In part for analytical convenience, Verhoogen (2008) models quality upgrading as a shift between lower- and higher-quality goods that a firm already knows how to produce. But Atkin et al. (2017a) argue, convincingly, that the rugmakers learned something in the process of exporting, using two main approaches. In the first, they estimate the effect of treatment on productivity controlling for detailed product attributes and find that it raises TFP. A possible concern, acknowledged by the authors, is that producers choose the product attributes in response to treatment.[12] This concern does not apply to their second approach, in which they had rugmakers produce identical rugs using the same looms in a laboratory. They find that treated producers make rugs that score more highly on observable quality dimensions but take no less time to produce them. This is already strong evidence for learning. The authors also document an association between messages between the intermediary and producers about quality issues and improvement on those dimensions. One could raise the question of whether producers gained a pure increase in capability applicable to all types of rugs or learned something specifically about the tastes of foreign buyers. But the constellation of evidence strongly supports the idea that the producers have learned by exporting. This study is a nice example of the advantages of collecting direct information on quality and productivity in a controlled setting (as well as on communications between buyers and producers).

3.1.2 Demand from Local Buyers, Foreign and Domestic

The literature on domestic demand conditions and upgrading has tended to focus on the effects of the presence of multinational corporations (MNCs) in local markets. The entry of foreign firms through foreign direct investment (FDI) is considered by many to be one of the primary drivers of upgrading. But foreign entry may have several effects on local firms. On one hand, foreign entry may generate technological learning spillovers or increased demand (especially for high-quality products) from local firms. On the other hand, foreign firms may have a “business-stealing” effect, gaining market share at the expense of local firms and making it harder for them to reap scale economies.

Early papers using firm-level data found mixed results. In Venezuelan data, Aitken and Harrison (1999) find a negative effect of FDI on the TFP of domestic firms in the same sector, consistent with a business-stealing effect. In Lithuanian data, Javorcik (2004) uses a sector-level input-output matrix to construct measures of exposure to FDI in a firm’s own sector, downstream sectors, and upstream sectors. She finds that firms in sectors that supply the FDI sector experience productivity gains (“backward” spillovers),[13] but that there is little evidence of a productivity effect in the same sector (“horizontal” spillovers) or in sectors that buy from the FDI sector (“forward” spillovers). In a related study in the US, Greenstone et al. (2010) compare counties that win competitions to host large plants, many of them foreign, to counties on the shortlists of candidate locations that lose the competitions. They find that incumbent plants in winning counties see significant TFP increases, and that the spillovers appear to pass through worker-flow and technological links, rather than supplier links. Using the same strategy, Bloom et al. (2019) find spillovers in management practices, but only for firms in sectors with high rates of cross-migration for managers in household data. Abebe et al. (2019) pursue a similar strategy in Ethiopia, comparing TFP outcomes in regions that received foreign investment to regions where firms planned to invest but for bureaucratic reasons were delayed; they find positive effects of nearby FDI on the level of TFP in local firms.

Several papers have examined the effects of the entry of big-box retailers on local suppliers. In a detailed case study of Wal-Mart’s entry into Mexico, Javorcik et al. (2008) argue that there was a heterogeneous effect on local suppliers in the soap and detergent industry: the best suppliers began selling to Wal-Mart and faced pressure to reduce prices but also received input on how to upgrade; weaker suppliers continued to sell through traditional retail channels and just faced increased price competition.[14] Iacovone et al. (2015) develop a dynamic industry-evolution model that captures this effect and find reduced-form evidence consistent with it: in regions with more Wal-Mart stores, and in sectors more likely to be selling to Wal-Mart (e.g. frozen foods), larger plants (presumed to produce products of greater “appeal”) increased sales, R&D spending, wages, and imported input shares (presumed to be correlated with product quality) relative to smaller plants. In Romania, Javorcik and Li (2013) estimate the effect of the entry of global retail chains on local suppliers, using a summary measure of distance from foreign retailers as a driving variable, and find positive effects on the estimated TFP of affected upstream firms.

An important limitation of the above studies is that until recently it has not been possible to see input-output links at the firm level, and the measures of linkages have had to be constructed using sector-level and/or region-level information. A recent paper by Alfaro-Urena et al. (2019) takes advantage of administrative tax data from Costa Rica, which contains firm-level input-output links. The authors compare firms that start supplying to a multinational corporation (MNC) in Costa Rica to firms that never supply to a MNC and find positive effects on sales to other firms, employment, and standard TFP measures. In a supplemental survey of new MNC suppliers, firms report that the MNCs demand high product quality, which in turn requires using high-quality inputs and changes in hiring, sourcing, and organizational practices.[15]

A persistent challenge in this literature has been to estimate effects on local firm performance that are not confounded by the effects of demand shocks on markups. A new MNC coming to town can be expected to increase demand for local firms, which may in turn induce local firms to increase markups, which are captured by standard TFP measures. Since the process also often involves quality upgrading, simply estimating TFPQ, if quantity information were available, would not solve the problem. One potential way forward is to use natural experiments to analyze the effect of shocks to domestic demand per se, as opposed to shocks to demand from MNCs. The former are typically not expected to raise product quality, and therefore a comparison between TFPR and TFPQ might be more informative about the role of markups than in settings with larger shocks to the demand for quality. Although not focused on upgrading outcomes, several recent studies examine the effects of arguably as-good-asrandom or literally random allocation of government procurement contracts to local firms, for instance in Brazil (Ferraz et al., 2015) and Ecuador (Carrillo et al., 2019). This line of research seems promising.

Another sort of buyer-driven effect arises when customers have preferences directly over the technologies used by firms. One example is provided by Higgins (2019), who shows that when a large Mexican social program (Progresa/Prospera) began disbursing funds on debit cards, corner stores responded by adopting electronic payment technologies, to make payment more convenient for the beneficiaries. (Supermarkets were already largely saturated with the technologies.) Interestingly, the greater use of electronic-payment technologies by corner stores increased demand by other (non-beneficiary) consumers for debit cards, creating a two-sided feedback loop. Another example is provided by the preferences of multinational buyers of consumer goods over working conditions: several studies have found evidence that anti-sweatshop pressure has increased wages and improved working conditions (Harrison and Scorse, 2010; Tanaka, forthcoming).[16]

3.1.3 Competition in output markets

The degree of competition in output markets is another potential driver of upgrading. The key question in this literature, as memorably phrased in the title of Lawrence (2000), is “Does a kick in the pants get you going or does it just hurt?” The conceptual link between increased competition and upgrading is not obvious. One common argument is that firms do not maximize profits prior to the increase in competition[17] and are spurred to do so (to increase “X-efficiency” in the terminology of Leibenstein (1966)) by the competitive threat. But this argument also needs to explain why firms were not maximizing profits in the first place. One also needs a mechanism strong enough to overcome the possible reduction in scale - and hence in scale economies - by firms facing stronger competition. Empirically, the challenge is to separate the effect of competitive pressure to upgrade from the effect of killing off firms that fail to upgrade. Holmes and Schmitz (2010) review the theoretical and empirical research on these issues, focused mainly on developed countries. Although they discuss a number of ideas, they acknowledge that there is little consensus in the literature about theoretical mechanisms.

Empirically, there is reasonably convincing evidence of a positive effect of competition on firm performance in particular cases. One leading study is Schmitz (2005), which tracks the response of US iron ore firms to the lower prices of Brazilian ore in the 1980s. Schmitz finds significant increases in productivity and argues that they were mainly due to changes in work practices, made possible in part because the competitive threat led unions to be more flexible about work rules. He marshals direct evidence from collective bargaining contracts and staffing levels, which reinforces the findings from more conventional productivity estimation. In a developing-country context, Das et al. (2013) focus on a public-sector rail mill in India which was for many years the exclusive producer of long rails for Indian railroads. In the late 1990s, the Indian government invited private companies to begin production and a large private conglomerate announced its intention to enter. Output per shift in the rail plant, measured in physical units, rose by 30% in a matter of months. Another example is provided by Jensen and Miller (2018), who study boat-builders in Kerala, India. The expansion of cellphone coverage led fishermen to travel further so they could sell their fish at the best prices. This increased their knowledge of boat-builders in other villages and arguably increased competition in the boat-building market. In turn, increased competition led to an expansion of the businesses of higher-skilled (and higher-quality) boat-builders and a contraction of those of lower skill, raising average quality. The greater scale for higher-skill builders also arguably enabled greater capacity utilization and greater labor specialization within firms, reducing costs. Another interesting example, from the Chinese footwear industry, is offered by Qian (2008). Following a shift of intellectual property rights enforcement resources away from counterfeiting in 1995, the industry saw a sharp increase in the entry of low-quality producers selling counterfeit brands. To differentiate themselves, more-productive, higher-quality producers upgraded quality and vertically integrated downstream by opening company stores.[18]

A large number of papers have explored the consequences of reductions of import tariffs on within firm productivity changes. These studies have typically considered many sectors together, and do not have the sort of detailed information on business practices or physical output that the papers discussed above have. An early paper by Pavcnik (2002) used the Olley and Pakes (1996) methodology to estimate TFP in Chilean data and found that productivity increased in import-competing industries relative to non-traded industries following Chile’s unilateral liberalization in the late 1970s. (See also Tybout et al. (1991).) Amiti and Konings (2007), in one of the first papers to separate the effects of tariffs on a firm’s outputs and inputs, apply the Olley and Pakes (1996) methodology to estimate TFP in Indonesian data and estimate separately the effects of tariffs on outputs and inputs. The effects of output-tariff reductions on productivity are positive but modest, especially relative to the input-tariff effects (mentioned in Section 3.2.1 below). Papers that have found a positive effect of output-tariff reductions on productivity include Schor (2004) and Muendler (2004) in Brazil, Fernandes (2007) in Chile, Lopez Cordova (2003) and Iacovone (2012) in Mexico, Yu (2015) in China, and Topalova and Khandelwal (2011), Nataraj (2011), and De Loecker et al. (2016) in India.[19] A small literature has also found effects of output tariff reductions on R&D expenditures and/or other innovation outcomes in developing and emerging countries (Teshima, 2010; Gorodnichenko et al., 2010).

But there is reason for caution in concluding that trade competition has an unambiguously positive effect on productivity. In the corrected version of the study of WTO accession on Chinese firms by Brandt et al. (2017, 2019), the effect of output-tariff reductions on the productivity of incumbent firms is not statistically distinguishable from zero. In detailed Ecuadorean data, Bas and Paunov (2019) find mostly statistically insignificant results of output tariffs on TFP measures. Holmes and Schmitz (2010) note that studies often focus on tariff effects on productivity changes in surviving firms, which may be a selected sample.[20] The extent to which import competition raises productivity by killing off lesscapable firms versus stimulating firms to improve their performance remains a persistent question. The issues with standard TFP measures discussed in Section 2.2.1 continue to be concerns in many studies. There is also well-identified historical evidence that temporary protection from British imports during the Napoleonic wars promoted adoption of mechanized cotton spinning in Northern France (Juhász, 2018), suggesting that reduction of competition can also increase productivity.

Overall, the evidence on the effects of competition on upgrading seems somewhat inconclusive. It is clear that increased competition can have positive effects on firm performance in some cases, but the effects vary significantly across settings. More research is needed to better understand the conditions under which competition stimulates upgrading. One interesting idea, which has not been well explored empirically at the firm level, is that competition plays more of a stimulating role for firms closer to the world technological frontier than for those further away (Aghion et al., 2005a,b; Amiti and Khandelwal, 2013).

3.1.4 Reputation in Output Markets

The quality models discussed above treat quality as observable and enforceable in contracts. But in the real world, information is often asymmetric. Buyers may only learn about the quality of a good after a transaction has taken place, and, if the quality is lower than contracted, may have difficulties getting a court to enforce the contract. The same goes for other product characteristics (broadly construed) such as the timeliness of delivery. These issues are especially severe in developing countries, where quality and reliability vary greatly across firms and legal institutions are weak.[21]

In such settings, firms typically rely on repeated interactions and the threat of discontinuing a relationship to enforce agreements; in other words, they enter into relational contracts (MacLeod and Malcolmson, 1989; Baker et al., 2002). But establishing a relational contract, and developing a reputation for quality and reliability, can take time and require up-front investments. This can be especially challenging in developing countries, because buyers often use average quality in a country or country-sector to form expectations about the quality of a particular firm. Given this collectivereputation issue, it may not optimal for individual firms to upgrade: there may be a low-quality equilibrium trap (Tirole, 1996). In such situations, mechanisms that allow firms to build individual reputations may stimulate upgrading. In addition, networks of firms may facilitate contracting, by providing information about potential trading partners, enhancing a firm’s ability to sanction partners who renege, and giving the group an incentive to sanction its own members in order to maintain a group reputation.

A small but growing literature has explored these issues empirically in developing countries. Using a tailored survey of Vietnamese firms, McMillan and Woodruff (1999) document that, consistent with models of relational contracts, firms’ willingness to supply trade credit (an indicator of how much the firm trusts a trading partner) depends on a number of features of the relationship: how easy it is for the partner to find another supplier, how long the two parties have been transacting, and the density of network links. In data on contracts of Indian software firms, Banerjee and Duflo (2000) show that older firms and firms with a very long-term, open-ended relationship with the buyer - characteristics plausibly associated with the reputation of the Indian firm - are offered more attractive contracts, in the sense that the buyer is more willing to accept responsibility for cost overruns. Macchiavello (2010) shows that Chilean wineries receive more attractive terms from UK wine distributors over time, controlling for such factors as quality and winery-distributor match effects, suggesting that the wineries acquire improved reputations over time.

Macchiavello and Morjaria (2015) examine the response of Kenyan rose exporters to a major supply disruption brought about by ethnic violence in 2008 and find patterns consistent with a reputation model. In particular, they find an inverted-U relationship between relationship age and the exporters’ compliance with agreements to provide flowers during the violence (which raised the cost of supplying flowers). Compliance initially increases with age because the value of the relationship increases with age. But at a certain point, sellers have established their reputations with the buyers, and do not have to worry as much about damaging their reputation by not complying.[22]

A recent experiment by Bai (2018) with watermelon sellers in China highlights the importance of branding for the development of reputations: simply giving sellers a hard-to-counterfeit way of marking their watermelons was sufficient to induce them to upgrade the quality of goods sold with that mark. A somewhat contrasting case is offered by Bold et al. (2017), who calibrate a learning model using data from agricultural trials in Uganda and argue that, given the noise in the environment and the difficulties that consumers have in inferring fertilizer quality, it would be very costly for a seller of fertilizer to develop a reputation for supplying high quality. This may explain the fact that the fertilizer market appears to be stuck in a low-quality equilibrium. Bai et al. (2017) provide evidence for the role of group reputation in the Chinese dairy industry. In 2008, a subset of producers were found to have sold adulterated baby formula by adding the industrial chemical melamine. Exports dropped by 68% following the scandal, and, perhaps surprisingly, firms that were inspected by the Chinese authorities and found to be innocent saw similar declines as those found to be guilty. The group reputation effects appear to have been particularly strong in this case.

Overall, despite these notable contributions, we are still at an early stage of learning about the causal mechanisms linking the costs of acquiring a reputation in output markets and upgrading by industrial firms. Newly available data from online platforms are making it possible to investigate reputation mechanisms at a level of detail not previously possible; see Tadelis (2016) for a review. This area seems to be very fertile ground for research.


[1] In trade, the idea that consumers in richer countries are more willing to pay for quality is commonly attributed to Linder (1961). In the consumption literature, the idea is regarded as so well established as to be unremarkable; see e.g. Deaton and Muellbauer (1980).

[2] Several earlier empirical papers explore the role of quality in trade at a more aggregate level. In addition to Hummels and Klenow (2005), cited above, Schott (2004) shows that the US imports higher-priced products within narrow trade categories from richer countries, suggesting quality differences. In a cross-country setting, Hallak (2006) shows that richer countries tend to demand relatively more from exporters with higher prices (and presumably higher quality). Notable early theoretical papers on quality in trade include Gabszewicz et al. (1982) and Flam and Helpman (1987). It appears that Verhoogen (2008) was the first to use a heterogeneous-firms model to formalize the idea that a given firm will sell a higher-quality variety in a richer market and to explore its implications in firm- (or plant-) level data. The related but distinct idea that firms’ quality choices respond to per-unit trade costs (as in the famous example of Washington apples from Alchian and Allen (1964)) has been developed by Rodriguez (1979), Feenstra (1988), Hummels and Skiba (2004), Feenstra and Romalis (2014) and others.

[3] Subsequent papers that have developed heterogeneous-firm models with endogenous output and input quality choice include Kugler and Verhoogen (2012), Hallak and Sivadasan (2013) (discussed below), Johnson (2012), Antoniades (2015), Fan et al. (2015), Bastos et al. (2018), and Blaum et al. (2019).

[4] The within-plant wage change was stronger for white-collar workers than blue-collar workers, hence wage inequality also increased within plants, a finding further explored in employer-employee data in Frías et al. (2012).

[5] In value-added-tax data from Turkey, Demir et al. (2019) find assortative matching between high-wage buyers and high-wage suppliers, again consistent with the idea that producers of high-quality outputs buy high-quality inputs.

[6] In Chinese and US data, Bloom et al. (forthcoming) show that many of the relationships previously documented between exports, inputs, and plant size also hold between exports, inputs and measures of management practices, consistent with the idea that larger plants tend to have higher capability than smaller plants, and that higher-capability plants tend to select higher-scoring management practices. Eckel et al. (2015) show that the correlation between sales and output prices documented across firms by Kugler and Verhoogen (2012) also holds across products within firms in Mexican data, consistent with a model in which firms invest more in the quality of their core products.

[7] See also Rankin and Schöer (2013).

[8] This argument echoes earlier research by Woodruff (2002), who found in cross-sectional data among Mexican footwear producers that vertical integration is more likely in firms producing higher-quality shoes.

[9] See also Sabel et al., eds (2012).

[10] There is also a small structural literature on exporting and investments in innovation by firms, which is beyond the scope of this review. See e.g. Aw et al. (2011).

[11] Readers interested in greater detail are referred to the reviews by De Loecker and Goldberg (2014) and Shu and Steinwender (2019).

[12] Conditioning on a set of covariates that respond to treatment breaks the balance on unobservables between treatment and control groups; see e.g. Angrist and Pischke (2009, Section 3.2.3).

[13] Javorcik suggested that pressure on local suppliers to raise the quality of goods sold to foreign-owned firms may have been part of the reason for this effect.

[14] Atkin et al. (2018) document that foreign retailers in Mexico charge prices that are on average 12% lower than modern domestic retailers, for the same barcode-level product in the same location.

[15] In related work in the coffee sector of Colombia, Macchiavello and Miquel-Florensa (2019) show that a qualityupgrading program of a large multinational buyer, which both provided training to farmers and guaranteed a price premium for coffee fulfilling quality (and traceability) requirements, was successfully in increasing the supply of highquality coffee.

[16] Relatedly, Boudreau (2019) randomized enforcement of local labor laws by multinational companies in Bangladesh, and found positive effects on compliance with a local requirement to maintain worker-manager safety committees.

[17] This could be either because they fail to optimize altogether or that they optimize an objective other than profits. This issue is discussed further in Section 3.3.1 below.

[18] Using case studies of the construction equipment, automotive, and machine tools industries in China, Brandt and Thun (2010) develop the related and interesting idea that competition at the low-quality end of industries induced domestic firms to upgrade to the middle-quality segment to escape competition. The fact that China has a large domestic market meant that firms were shielded somewhat from foreign competition even in the middle-quality segment, because foreign firms had higher costs and less knowledge of domestic consumers. See also Medina (2018).

[19] In a similar spirit, Bloom et al. (2016a) find positive effects of competition from China on patenting, information technology use, and TFP in twelve European countries. In Spanish data, Chen and Steinwender (2019) find positive effects of import competition on productivity for initially less-productive, family-managed firms. By contrast, Autor et al. (forthcoming) find negative effects of Chinese competition on patenting in the US.

[20] See also Yang et al. (2019).

[21] For a useful overview of the international dimensions of these contracting issues, see Antràs (2015).

[22] In related work, Ghani and Reed (2019) examine how relational contracts between ice sellers and fishermen in Sierra Leone evolve in response to an increase in upstream supply of ice.

Previous Chapter II.ii. What Do We Mean by Upgrading? Measurement Issues
Back to top
Next Chapter III.ii. Drivers of Upgrading: Input-Side Drivers