The impact of trade costs on firm entry, exporting, and survival in Korea.

Kim, Sooil ; Reimer, Jeffrey J. ; Gopinath, Munisamy 等

I. INTRODUCTION

Declining trade costs is a signature feature of globalization and carries important consequences for the structure and performance of firms in an industry. Firms in most industries vary greatly in size, productivity, and capital intensity (Bernard and Jensen 1999). Falling trade costs therefore have different effects on firms. New models of international trade with heterogeneous firms such as Melitz (2003) and Bernard et al. (2003) formalize these processes and generate a number of testable implications. For example, falling trade costs may drive down goods prices and force firms with low-productivity to exit the domestic market. However, firms with high productivity may increase their domestic market share and be able to start exporting.

The purpose of this study is to evaluate the key predictions of these models in the context of Korean manufacturing firms over the period 1992-2002. Like other nations in this period, Korea made efforts to reduce government intervention in the economy and liberalize policies on imports and foreign investment (Moskovitch and Kim 2008). (1) Our approach lets us examine how the resulting changes in trade costs affected the productivity of industries and individual firms. We examine how trade costs and other factors affect the probability of firm entry and exit, and the probability that firms newly begin to export to foreign markets. In addition, we examine changes in domestic market share and the amount that firms were able to export as trade costs fell.

The data are from Korea's credit-rating agency (the Korea Information Service [KIS]) and describe the activities of an average 5,021 firms over the sample period. Our data series is for 1992-2002 and is somewhat longer than what is typically used in the literature that investigates heterogeneous firms. To the best of our knowledge, these data have not previously been used in published economics research. Our study contributes to the literature that documents the characteristics of individual firms or plants that produce for foreign markets (e.g., Aw, Chung, and Roberts, 2000; Bernard and Jensen, 1999; Bernard, Jensen, and Schott, 2006; Clerides, Lach, and Tybout, 1998; Tybout, 2003). Our study differs from many of these studies in that our data are for the firm level as opposed to the plant level. (2) Korea has been little studied within this literature and is an excellent example of the export-led growth idea. Our study also complements the industrial organization literature, which has long examined firms' entry and exit patterns. This literature suggests that a firm may enter because it has high productivity, is filling a niche market, or is optimal size for the market. Our approach allows for many of these factors but focuses on how changes in trade costs may have affected the competitive environment of an industry, and thus patterns of firm entry and exit. In this study, we verify several key predictions of recent heterogeneous-firm models of international trade, and also find a few results that are new to the literature.

II. TESTABLE HYPOTHESES

The development of trade models with heterogeneous firms has opened up many new research questions. One class of models is based on Bernard et al. (2003), who introduce stochastic firm productivity into the multicountry Ricardian model. Firms use different technologies to produce the same good. Consumers in any given country buy each good from the lowest-cost producer across all countries, based on a stochastic productivity draw. Because of trade costs, several firms producing the same good can survive if they are located in different countries, although each firm is the sole supplier to any given destination.

Another class of models is based on Melitz (2003), who starts with a monopolistic competition framework. Unlike Bernard et al. (2003), firms do not directly compete to be the exclusive supplier of a homogeneous good, as each firm produces its own distinct variety. Firms make an irreversible investment to enter the domestic market while being uncertain about their future productivity. Upon entry, each firm learns about its productivity level, as drawn from a known distribution. The least productive firms face negative profits and have to exit. Among those surviving, only relatively productive firms choose to export because exporting is costly. Trade costs include a fixed cost of entry into the export market, plus a per unit (variable) trade cost. Remaining firms serve the domestic market.

Although the models in this literature have different structures, they make a number of overlapping predictions regarding the role of trade costs. The specific hypotheses that we test concern falls in variable (per unit) trade costs, as opposed to any fixed cost of entry or fixed cost of exporting. Some of the hypotheses concern the cost for Korean firms to export to foreign markets, which we call "export-trade costs." Other hypotheses concern the cost for foreign firms to import into the Korean market, which we call "import-trade costs." Some hypotheses consider both types of costs.

One effect of lower import-trade costs may be to inhibit successful entry of new firms into the domestic market (Hypothesis 1). Entering firms must leap a higher hurdle in terms of productivity due to the presence of foreign competition. The productivity cutoff or breakeven point to enter domestic markets is larger in the case of an open economy relative to a closed one (Melitz, 2003). The effect of a fall in export-trade costs may have a different effect on entry, however. Because their expected profits rise, falling export-trade costs could encourage entry by domestic firms.

Falling import-trade costs may reduce the domestic market share of surviving firms because of the entry of high-productivity foreign firms into the domestic marketplace (Hypothesis 2). In turn, the productivity of individual firms is expected to increase with declining import-trade costs (Hypothesis 3). This may be because a firm changes its product mix, or because increased competition induces plants to improve their productive efficiency, that is, it gives them a "kick in the pants" (Bernard, Redding, and Schott, 2006).

As import- and export-trade costs fall, industry productivity may increase (Hypothesis 4). This could happen if low-productivity, none-xporting firms have to exit due to falling import costs. This could also happen if high-productivity firms expand through exporting due to falling export costs (there are scale economies to be gained as already-efficient firms expand in foreign markets). In short, as trade costs fall, market share goes to more efficient firms (Bernard et al., 2003; Melitz, 2003). Another consequence of falling export-trade costs is a rise in the number of new exporting firms, in particular, highly productive firms that had previously not been able to export (Hypothesis 5). In turn, falling export-trade costs may increase export sales at existing exporters (Hypothesis 6). Existing high-productivity exporters may be able to export more, as falling per unit export-trade costs give them cheaper access to external markets. Falling import-trade costs also raise the probability of firm exit (Hypothesis 7). This is because the productivity threshold for survival in an increasingly competitive marketplace will be higher.

In summary, the hypotheses that we test are:

Hypothesis 1: Falling import-trade costs decrease the probability of entry by new firms, whereas falling export-trade costs increase the probability of entry by new firms.

Hypothesis 2: Falling import-trade costs reduce the domestic market share of surviving firms.

Hypothesis 3: Falling import-trade costs raise the productivity of individual firms.

Hypothesis 4: Falling import-trade and export-trade costs raise overall industry productivity.

Hypothesis 5: Falling export-trade costs increase the number of new exporting firms.

Hypothesis 6: Falling export-trade costs increase export sales at existing exporters.

Hypothesis 7: Falling import-trade costs raises the probability of firm exit.

We also examine the importance of the size and capital intensity of firms, as these are also likely to affect firm entry and exit. For example, larger firms may benefit from scale economies and greater experience. In addition, large firms may have better access to capital than newer, smaller startups, especially if they are part of a conglomerate (Doh and Ryu, 2004; Moskovitch and Kim, 2008).

III. DATA

Firm-level data on Korean manufacturing are obtained from the KIS, the major credit-rating agency in Korea. Our data include all manufacturing sectors for Korea, and account for more than 60% of total and individual industry output. A firm is designated as "in the market" if it reports to KIS in three consecutive years. A firm "enters" in year t if the firm did not report in year t- 1 or any previous year, and if it did report in year t, t + 1, and t + 2. A firm "exits" in year t if it reports to KIS in years t - 2, t - 1, and t, but not in year t + 1 or any following year. (3) Tables 1 and 2 report selected descriptive statistics. For example, in a given year, 23% of firms were newly entering, 10% of firms were exiting, and 3% of firms began exporting (Table 1, bottom row). Entering firms, exiting firms, and new exporters tended to be smaller than other firms (Table 2). Table 2 shows that new exporters had 117.0 employees versus 134.2 employees for all other firms. This reflects the fact that during the 1990s in Korea, many new businesses formed with the goal of exporting beyond the small domestic market (Moskovitch and Kim, 2008). Looking at other data (not reported in the tables), 46% of new exporters in Korea started exporting in the same year that they started their business during 1992-2002; 17% started exporting during the second year of their business. Of those firms that started exporting within 3 yr of starting their business, the average number of employees was 69. These results are largely consistent with evidence for France (Eaton, Kortum, and Kramarz, 2004) and the United States (Bernard and Jensen, 1999), among other countries. Eaton et al. (2007) examine 1996-2005 data for Colombia and find that nearly half of all Colombian exporters were not exporters in the previous year. These new exporters were small in terms of their overall contribution to export revenues, and often failed to continue exporting the following year. These characteristics appear to have been present in Korean manufacturing during 1992-2002.

A. Trade Costs

Asymmetric trade costs are calculated based on the gravity framework of Novy (2007) and Jacks, Meissner, and Novy (2008). (4) Let [t.sub.ij] be the trade cost between country i and j, [x.sub.ii] be i's consumption of domestic goods, [x.sub.ij] be exports from i to j, and [sigma] > 1 be the elasticity of substitution among varieties. (Depending upon whether Korea is an exporter or importer, it may be either i or j.) If we abstract from intranational trade ([t.sub.ii] = [t.sub.jj] = 1) for the moment, we can restate Equation (5) in Novy (2007) to get asymmetric trade costs that distinguish between exporting and importing:

(1) [t.sub.ij] = [[x.sub.ij] / [([x.sub.ii] x [x.sub.jj]).sup.1/2]1/(1-[sigma])

If we assume that bilateral trade costs are symmetric ([t.sub.ij] = [t.sub.ji]), then we can get a tariff-equivalent measure of symmetric "two-way" trade costs ([[tau].sub.ij]), calculated as:

(2) [[tau].sub.ij] = ([x.sub.ii] x [x.sub.jj] / [x.sub.ij] x [x.sub.ji]).sup.1/(2[sigma] -1])) - 1.

This is Equation (8) in Novy (2007). We calculate trade costs using gross domestic product (GDP) data and trade data from the Bank of Korea (2007) for Korea and seven major trading partner countries (we account for countries' respective trade volumes with Korea, and deflate with a GDP deflator). We use [sigma] = 8 as in Novy (2007) and several other studies that are cited in that paper. Results for Equation (1) are illustrated in Figures 1 and 2. Looking at Figure 1, we see a general decline in the costs of exporting over time. Looking at Figure 2, the most striking change is the large increase in the cost of importing in 1998, which is related to the Asian financial crisis and the strong devaluation in Korea's currency during that year. The fall in import costs after 2001 may be partly related to China's gaining Most Favored Nation status upon accession to the World Trade Organization. In calculating Equation (2), we find that the average two-way trade cost declined 10.6% over the sample period. Trade cost measures do not vary by industry.

[FIGURE 1 OMITTED]

[FIGURE 2 OMITTED]

B. Total Factor Productivity

To measure total factor productivity, we use the mathematical programming technique called Data Envelopment Analysis. The approach is described fully in Chambers, Fare, and Grosskopf (1996). It identifies a "best practice" benchmark for efficient use of resources, then evaluates each firm relative to this benchmark. Our measure is actually an "inefficiency" index because as firms have higher total factor productivity, their numerical result is closer to zero. Industry-level inefficiency is taken to be the median of firms' inefficiencies.

IV. RESULTS

A. Entry

Hypothesis 1 is that falling import costs decrease the probability of entry by new firms, whereas falling export costs increase the probability of entry by new firms. The probability that firm f newly enters the domestic market between year t and year t + 2 is denoted Pr([ENT.sub.ft+2] = 1). The change in trade costs between period t and t - 2 between Korea and its major trading partners is denoted [DELTA]C[ost.sub.t-2] (note that when writing out our equations, we will not identify the particular form of trade cost that we are considering because this can be seen in the tables). We use a gap in time between changes in trade costs and changes in entry to help mitigate problems of endogeneity and omitted variables. The most general model is this logistic regression:

(3) Pr([ENT.sub.ft+2] = 1) = [PHI]b([beta][DELTA][Cost.sub.t-2] + [gamma][X.sub.ft] + [[delta].sub.i]).

[X.sub.ft] is a vector of firm characteristics, including the size of the labor force and the capital-labor ratio. [[delta].sub.i] is a coefficient on a dummy variable for industry i, with i = (1, ..., 7) because there are eight industries. (The full representation of industry fixed effects is suppressed for notational simplicity.) The expected sign of the coefficient on import costs is positive because the threshold by which new firms are able to survive in the market is lowered. The expected sign on export costs is negative, however, as firms are more likely to enter if there are profit opportunities abroad. The sign on our two-way trade cost measure will be positive if the import cost story dominates, and negative if the export cost story dominates.

Table 3 reports four versions of Equation (3). In this table, as in our tables in general, Variants 1 and 2 distinguish between export- and import-trade costs, whereas Variants 3 and 4 do not. Looking at the results for Variant 1 in Table 3, we get the expected signs on the import cost channel, with statistical significance at the 1% level. This means that as import costs fall, entry by new domestic firms is less likely, because there is more foreign competition. A one standard deviation decrease in import-trade costs decreases the probability of entry by approximately 3.7%. We do not, however, get the expected sign on the export cost channel (it is positive instead of negative).

Variant 2 adds firm characteristics, including the size of the labor force and the K-L (capital-labor) ratio, to Variant 1. Results on trade costs do not change from Variants 1 and 2. The coefficient on the employment variable is negative and statistically nonzero. This means that entering firms are typically smaller than incumbents. This is consistent with our descriptive statistics; Table 2 shows that new firms have 36.6 employees on average, whereas incumbent firms had 161.9 employees on average. The coefficient on the K-L variable is negative and is also statistically significant (Table 3). New firms tend to use relatively labor-intensive techniques.

In Variants 3 and 4, the coefficient on the change on two-way trade costs is positive and statistically significant at the 1% level. This is consistent with the import-trade cost story. (Note that the magnitude of coefficients cannot be directly compared across trade cost measures without calculating marginal effects on probability, which we do not have space to report.) The inclusion of firm characteristics in Variant 4 makes no difference to the results. The signs and significance of the coefficients on firm size and the K-L ratio in Variant 4 are the same as they were in Variant 2.

B. Market Share

Hypothesis 2 states that falling import-trade costs reduce the market share of domestic firms, because high-productivity foreign firms are likely to increase their share. As above, we use a gap in time between changes in trade costs and changes in market shares to help mitigate problems of endogeneity and omitted variables. The change in market share of a surviving firm between year t and year t + 2 is denoted [DELTA][Share.sub.ft+2] = [Share.sub.ft+2] - [Share.sub.ft]. (5) The general form of the regressions is:

(4) [DELTA][Share.sub.ft+2] = [beta][DELTA][Cost.sub.t-2] + [gamma][X.sub.ft] + [[delta].sub.i] + [[epsilon].sub.ft],

where remaining variables and coefficients are defined as for Equation (3). (Again, the full set of industry fixed effects is suppressed for notational convenience.) The expected sign of [beta] is positive because increases in import-trade costs ([DELTA][Cost.sub.t-2] > 0) are associated with increases in market shares by Korean firms. To the extent that the import-trade cost story dominates, we also expect a positive sign on the coefficient of our two-way trade cost measure.

Table 4 reports four variants of Equation (4). In Variant 1, the sign of the coefficient on change in import-trade costs is positive and statistically nonzero. This is consistent with expectations. Variant 2 adds firm characteristics, including the size of the labor force and the K/L ratio to Variant 1. The coefficient on import-trade costs is again positive and statistically nonzero, consistent with expectations. The coefficient on employment is negative but not statistically significant. The coefficient on the K-L ratio is positive but not statistically significant. Variants 3 and 4 differ in that trade costs do not distinguish between importing and exporting. In both versions, the coefficient on the change in two-way trade costs has the expected positive sign and is statistically significant at the 1% level. We thus have strong support for Hypothesis 2. This distinguishes the case of Korea from that of the United States, for which Bernard, Jensen, and Schott (2006) find that changes in industry-level trade costs are not correlated with changes in plant-level domestic market share.

C. Productivity of Individual Firms

Hypothesis 3 is that falling import-trade costs will force existing firms to improve their productivity, perhaps by changing the scale of their product mix. To test this idea, we denote the change in a firm's inefficiency (the inverse total factor productivity) from period t to period t + 2 as [DELTA][INEFF.sub.ft+2]. The most general regression is:

(5) [DELTA][INEFF.sub.ft+2] = [beta][DELTA][Cost.sub.t-2] + [gamma][X.sub.ft] + [[delta].sub.i] + [[epsilon].sub.ft].

As mentioned above, total factor productivity is measured by an inefficiency index (INEFF) in which values closer to zero imply higher productivity. Therefore, [DELTA]INEFF > 0 means productivity has worsened. The expected sign of [beta] is positive--productivity worsens when domestic firms are more sheltered from foreign competition. We expect a positive sign irrespective of whether we use import-specific trade costs or our two-way trade cost measure.

Table 5 reports four variants of Equation (5). The sign on trade costs is positive--as expected--in both Variants 1 and 2, although statistically significant only in the former. Thus, we have some support for the hypothesis. As above, Variant 2 adds firms' size of labor force and K-L ratio as explanatory variables. The coefficient on employment is negative and statistically significant. This implies that large firms are more likely to improve their productive efficiency. The result concerning capital intensity is not statistically significant.

Variants 3 and 4 differ in that the measure of trade costs no longer distinguishes between importing and exporting. In both versions, the coefficient on the change on two-way trade costs is unexpectedly negative and statistically significant. The latter may arise due to the trade cost being dominated by what was happening to export costs, and there is no clear channel between that and our dependent variable (change in firm-level productive inefficiency). We conclude that Variants 1 and 2 provide reasonable support for the hypothesis that falling import-trade costs provide a "kick in the pants" to firms.

D. Productivity of Industries

Hypothesis 4 is that falling export- and import-trade costs lead to higher productivity in manufacturing as a sector. One way to test this is with a regression of the form:

(6) [DELTA][INEFF.sub.it+2] = [c.sub.t] + [beta][DELTA][Cost.sub.it-2] + [[delta].sub.i] + [[epsilon].sub.it],

where [INEFF.sub.it+2] is the median level of inefficiency (i.e., inverse total factor productivity) in industry i. The expected sign of [beta] is positive--inefficiency is higher when domestic firms are more sheltered from foreign competition. We expect this sign on both import- and export-trade costs, and also when we use our two-way trade cost measure.

The top half of Table 6 reports four variants of Equation (6), differing according to how trade costs are calculated, and (in contrast to prior tables) whether industry fixed effects are included. Regardless of the specification, the coefficient on trade costs is positive. The only coefficients to be statistically different than zero, however, are those on import-trade costs. Note that Variants 2 and 4 differ from the others in that they include industry-specific fixed effects. This improves the [R.sup.2]-values considerably.

Because Hypothesis 4 is based on cross-firm reallocations of product-market shares (Bernard et al., 2003; Melitz, 2003), a more direct test may be to regress firm-level sales on trade costs interacted with lagged firm-level productivity. (6) Let the log difference in sales for firm f from t and t + 2 be denoted [DELTA] ln [Sales.sub.ft+2]. We regress this on two-way trade costs ([DELTA][Cost.sub.ft-2]) interacted with our measure of lagged firm-level inefficiency ([INEFF.sup.ft-1]):

(7) [DELTA] ln [Sales.sub.ft+2] = [c.sub.t] + [beta][DELTA][Cost.sub.ft-2] x [INEFF.sub.ft-1] + [[delta].sub.i] + [[epsilon].sub.ft],

Holding INEFF constant, the derivative of [DELTA] ln Sales with respect to [DELTA]Cost is negative, meaning that as the increase in trade costs gets larger, the change in sales is more strongly negative. Holding [DELTA]Cost constant, the derivative of [DELTA] ln Sales with respect to INEFF is negative, meaning that as inefficiency increases, the change in sales is more strongly negative. For these reasons we expect [beta] to be negative.

The bottom half of Table 6 reports four variants of Equation (7), differing according to whether industry fixed effects are used and the measure of trade cost. We get the expected sign, as well as statistical significance at the 1% level, when export-trade costs are used. However, we get an unexpected sign when two-way trade costs are used. Given that the latter measure encompasses import-trade costs, we feel most confident in Variants 1 and 2. Altogether, we have reasonably strong evidence that falling trade costs are associated with increased industry productivity. This is consistent with what Bernard, Jensen, and Schott (2006) find for the United States.

E. New Exporters

Hypothesis 5 is that falling export-trade costs increase the number of firms who are newly able to export. Let Pr([EXP.sub.ft+2] = 1) denote the probability that nonexporting firms become exporters. The most general logit specification is:

(8) Pr([EXP.sub.ft+2] = 1) = [PHI]([beta][DELTA][Cost.sub.t-2] + [gamma][X.sub.ft] + [[delta].sub.i]).

The expected sign of [beta] is negative because increases in export-trade costs are associated with fewer exporting firms. We also expect a negative sign if we use our two-way trade cost measure.

The top half of Table 7 reports four variants of Equation (8). The coefficient on export-trade costs is negative in both Variants 1 and 2. This is consistent with heterogeneous-firm models of international trade. However, this coefficient is statistically insignificant--as is the coefficient on two-way trade costs in Variants 3 and 4. The coefficients on the employment variable and the K-L ratio in Variants 2 and 4 are also statistically zero. Thus, we do not have statistical evidence that smaller firms are most likely to enter the export market, which we might have expected given the relatively small size of new exporters (Table 2). By way of comparison, Bernard, Jensen, and Schott (2006) find that larger and more capital-intensive U.S. manufacturers are more likely to become exporters.

We also investigate Hypothesis 5 by regressing export status on the absolute level of trade costs instead of the change. This regression would be closer in spirit to the descriptive evidence in Bernard et al. (2007) for the United States and related evidence for other countries, where export status is associated with size and capital intensity. (7) The regression is:

(9) Pr([EXP.sub.ft+2] = 1) = [PHI]([beta][Cost.sub.t] + [[delta].sub.i]).

We expect [beta] to be negative, meaning that export status is associated with low costs. The bottom half of Table 7 reports four variants of Equation (9). In all four specifications, we get an unexpected sign on trade costs and a lack of statistical significance. Similar to our other test of Hypothesis 5, lower export costs are not associated with export status. As with Hypothesis 4, the fact that we find this with two distinct approaches adds some robustness to our results.

F. Export Growth

Hypothesis 6 states that a decrease in export-trade costs raises export sales at existing, high-productivity exporters--the so-called intensive margin. The log difference in exports from t and t + 2 is denoted [DELTA] ln [Exports.sub.ft+2]. The most general regression is:

(10) [DELTA] ln [Exports.sub.ft+2] = [beta][DELTA][Cost.sub.t-2] + [gamma][X.sub.ft] + [[delta].sub.i] + [[epsilon].sub.ft].

The expected sign of [beta] is negative because increases in export-trade costs are associated with lower export sales by existing exporters. We expect a negative sign irrespective of whether we use export-trade costs or our two-way trade cost measure.

Table 8 reports four variants of Equation (10). The coefficient on export-trade costs is positive in all four variants, which is not consistent with heterogeneous-firm models of international trade. One possible explanation is that the financial crisis of the late 1990s prevented firms from expanding export operations, even with decreases in trade costs and devaluation of the Won. Interestingly, in Variants 2 and 4 we get a negative and statistically significant coefficient on firm size. This implies that increases in exports tended to occur for smaller firms. This is consistent with some of the descriptive statistics in Table 2, and also with what Bernard, Jensen, and Schott (2006) find for the United States; exporter size is negatively and significantly associated with export growth.

G. Firm Exit

Hypothesis 7 is that declining import-trade costs increase the probability of a firm exiting the domestic market. Let the probability of firm f's exit between year t and year t + 2 be denoted Pr([D.sub.ft+2] = 1). The most general logit specification is:

(11) Pr([D.sub.ft+2] = 1) = [PHI]([beta][DELTA][Cost.sub.t-2] + [gamma][X.sub.ft] + [[delta].sub.i]).

The expected sign of [beta] is negative because increases in import-trade costs are associated with lower rates of firm exit. We also expect a negative sign if we use our two-way trade cost measure.

Table 9 reports four variants of Equation (11). The coefficient on import-trade costs in Variants l and 2 is positive and therefore not consistent with expectations. However, the coefficient on two-way trade costs is negative and statistically significant in Variants 3 and 4, as expected. We therefore find moderate support for this aspect of heterogeneous-firm models of international trade. This is also what Bernard, Jensen, and Schott (2006) find for United States manufacturing; as trade costs fall, exit is statistically more likely.

The K/L ratio has no statistically significant impact on probability of exit (Table 9). This contrasts with the descriptive statistics that we saw in Table 2, and may be explained by a large outlier in the exiting firms' K/L ratio. By contrast, firm size is negatively and significantly correlated to firm exit. This is consistent with the results of Table 2, which shows that exiting firms have an average of 100.4 employees, whereas continuing firms have an average of 158.9 employees. One potential reason for this is that larger firms may have an established record and more knowledgeable management. In addition, large firms may have more access to credit, especially if they are part of a Chaebol conglomerate (Doh and Ryu, 2004). Unfortunately, we have no way of knowing whether a particular firm in our sample is part of a Chaebol.

V. SENSITIVITY CHECKS AND LIMITATIONS

We considered several additional variants of the above hypotheses. First, in some cases we allowed for a right-hand-side interaction term between trade costs and export status, or one between trade costs and productivity. This allows us to consider additional aspects of the theory, such as that as trade costs fall, firms with higher relative productivity are more likely to enter the export market (this example regards Hypothesis 5). The coefficients on the interaction terms are typically not statistically different than zero.

The second general sensitivity check concerns the Asian financial crisis. In the fall of 1997, the Korean Won had a large fall in value. To the extent that firms had foreign-denominated debt they may have been hurt, yet, however, they may have been able to export more. The crisis also affected credit markets. To account for these effects, we included trade-weighted exchange rates and interest rates as additional explanatory variables. Allowance for these factors did not generally improve goodness of fit or change the qualitative nature of the results. Analysis of descriptive statistics suggests that the financial crisis had only limited effect on the variables we study. One might think that the crisis would have caused many firms to exit; however, in the years after the crisis (1998-2000), the total number of entering firms was 50% higher than the number of exiting firms. In particular, 6,688 firms entered and only 4,461 firms exited (Table 1). This is consistent with findings by Moskovitch and Kim (2008), who suggest that the Asian financial crisis may have contributed to the upsurge of startups and related changes.

These specifications (and all others in the paper) are not immune to a number of potential econometric limitations. We exercise caution in claiming causal relationships due to issues of circularity related to our use of gravity-based trade cost measures, and due to other problems related to endogeneity. For example, productivity and export status may be correlated with the error terms due to factors such as learning-by-exporting. There may be hysteresis in market presence because of sunk entry and exit costs, and self-selection into export status based on unobserved variables. (8) In addition, there may be omitted variables related to changes in Korean industrial policy and certain aspects of the Asian financial crisis that are not adequately proxied for by average interest rates or exchange rates. Such problems might have been partly alleviated through use of time fixed effects. However, this was not possible because it would lead to perfect correlation with our trade cost measures. In this sense, the analysis would have benefited from an industry-specific trade cost measure such as tariffs, as done in Bernard, Jensen, and Schott (2006), for example. However, we emphasize that our ability to control for export-trade costs distinguishes our study from those that rely solely on import-trade cost measures.

VI. CONCLUSIONS

This study shows that falling trade costs have had important consequences for the structure of manufacturing activity in Korea. Some of these effects were "negative" in that many firms went out of business. However, Korea's manufacturing sector appears to have experienced a great deal of entrepreneurial activity and to have made gains in productivity during 1992-2002.

We find that firms differ substantially in export participation, with by far the most selling only at home. Only about 3% of firms newly enter export markets in a given year. In this respect, our results are consistent with those for countries such as France, Colombia, and the United States.

Our results support several key predictions from the evolving literature on heterogeneous-firm models of international trade. Falling import-trade costs are associated with less entry by new domestic firms, and with lower market shares among existing domestic firms. Falling import-trade costs are associated with higher total factor productivity for individual Korean firms. In turn, falling import- and export-trade costs are associated with higher productivity for Korean manufacturing as a whole. These results are largely consistent with predictions made in the emerging theoretical and empirical literature on heterogeneous firms.

Other results are less known in the heterogeneous-firm literature, although not necessarily inconsistent with it. Large firms are statistically less likely to exit, and are statistically more likely to have an increase in total factor productivity. New domestic firms are statistically smaller than incumbent firms. Some of our results differ from those reported for the United States and other countries, as well as the predictions of some theoretical models. For example, falling export-trade costs appear to increase the probability of entering the export market, but this relationship is not statistically significant. Moreover, export status appears to not have a statistical association with firm size or capital intensity. Likewise, falling export-trade costs are not statistically associated with higher export sales by existing exporters. Surprisingly, increases in exports are statistically more likely to occur for smaller firms, whereas falling import-trade costs are not statistically associated with a higher probability of exit.

Some of these results may reflect particular trends in Korean manufacturing in the 1990s. During our sample period, there was a surge of entrepreneurial activity in Korea. This may have been facilitated by reduced government intervention in the economy, or because large firms and alliances were weakened by the 1997 Asian financial crisis. Whatever the reasons, the total number of manufacturing firms increased steadily from 1992 to 2002, with firm entry rates remaining high even after the 1997 financial crisis. During 1992-2002, 46% of new exporters in Korea started exporting in the same year that they started their business, and 17% started exporting during the second year of their business. Many of these were very small firms and did not continue to export (or exist) for long. In this sense, our results bear semblance to those reported for some other developing countries such as Colombia, where nearly half of all exporters were not exporters in the previous year.

In general, we conclude that trade costs have had important consequences for the structure and performance of firms in Korean manufacturing. Continued liberalization is likely to strengthen some of the effects that we observe in this study. There is much scope for future research to examine aspects of the adjustment process associated with trade liberalization beyond those we have considered.

ABBREVIATIONS

GDP: Gross Domestic Product

KIS: Korea Information Service

doi: 10.1111/j.1465-7295.2009.00286.x

REFERENCES

Aw, B. Y., S. Chung, and M. J. Roberts. "Productivity and Turnover Patterns in the Export Market: Firm-Level Evidence from Taiwan and South Korea." World Bank Economic Review, 14(1), 2000, 65-90.

Bank of Korea. Economic Statistics System on the World Wide Web. 2007. http://www.ecos.bok.or.kr

Bernard, A. B., and J. B. Jensen. "Exceptional Exporter Performance: Cause, Effect, or Both?" Journal of International Economics, 47(1), 1999, 1-25.

Bernard, A.B., J. Eaton, J.B. Jensen, and S.S. Kortum. "Plants and Productivity in International Trade." American Economic Review, 93(4), 2003, 1268-90.

Bernard, A. B., J. B. Jensen, and P. K. Schott. "Trade Costs, Firms and Productivity." Journal of Monetary Economics, 53(5), 2006, 917-37.

Bernard, A. B., J.B. Jensen, S.J. Redding, and P.K. Schott. "Firms in International Trade." Journal of Economic Perspectives, 3, 2007, 105-30.

Bernard, A. B., S.J. Redding, and P. K. Schott. "MultiProduct Firms and Trade Liberalization." NBER Working Paper 12782, 2006.

Chambers. R. G., R. Fare, and S. Grosskopf. "Productivity Growth in APEC Countries." Pacific Economic Review, 1(3), 1996, 181-90.

Clerides, S., S. Lach, and J. Tybout. "Is 'Learning-by-Exporting' Important?" Quarterly Journal of Economics, 113(3), 1998, 903-48.

Doh, T., and K. Ryu. "Analysis of Loan Guarantees among the Korean Chaebol Affiliates." International Economic Journal, 18(2), 2004, 161-78.

Eaton, J., S. Kortum, and F. Kramarz. "Dissecting Trade: Firms, Industries, and Export Destinations.'" American Economic Review Papers and Proceedings, 94, 2004, 150-54.

Eaton, J., M. Eslava, M. Kugler, and J. Tybout. "Export Dynamics in Colombia: Firm-Level Evidence." NBER Working Paper No. 13531, 2007.

Jacks, D., C. Meissner. and D. Novy. "Trade Costs, 1870-2000." American Economic Review Papers & Proceedings, 98(2), 2008, 529-34.

Melitz, M. "The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity." Econometrica, 71(6), 2003, 1695-725.

Moskovitch, I., and D. J. Kim. "Building an Entrepreneurial Economy: The Case of Korea 1998-2005." Global Economic Review, 37(1), 2008, 63-73.

Novy, D. "Gravity Redux: Measuring International Trade Costs with Panel Data." November 2009. http://www2.warwick.ac.uk/fac/soc/economics/staff/academic/novy/redux.pdf.

Tybout, J. "Plant- and Firm-Level Evidence on the 'New' Trade Theories," in Handbook of International Trade. edited by E. K. Choi and J. Harrigan. Oxford: Basil Blackwell, 2003, Chapter 13.

U.S. Department of State. Korea Economic Policy and Trade Practices. Washington, DC: Bureau of Economic and Business Affairs, U.S. Department of State, 1994.

World Trade Organization. Korea Trade Policy Review 1996. http://www.wto.org/english/tratop_e/tpr_e/tp40_e.htm.

(1.) During the Uruguay Round Korea reduced tariffs by an average of 54% on a trade-weighted basis, and expanded bindings to 91% of tariff lines (WTO 1996). Import prohibitions on a number of sensitive items were abolished. As part of its bid to join the Organization for Economic Cooperation and Development in 1996, Korea carried out a 5-year program during 1991-1996 to reduce state control of the economy, abolish an array of regulations and restrictions, increase transparency of trade-related policies, and align domestic laws with international regulations (U.S. Department of State 1994; WTO 1996).

(2.) Although both types of data have been used to explore hypotheses in this literature, plant-level studies are somewhat more common due in part to data availability. The basic difference between the two types of data arises from the fact that a firm can have more than one plant. As a result, the exit of a plant does not necessarily imply the exit of a firm. A plant may also produce a narrower range of products than a firm, which may be advantageous for certain questions in the emerging product-focused heterogeneousfirm literature.

(3.) We do not consider firms that reported for 1 yr only because we need two-period lagged measures to minimize endogeneity problems. Less than 2% of firms reported for l yr only.

(4.) The hypotheses that we test concern variable trade costs. Although Novy's (2007) method measures of total trade frictions, it has a time dimension. We use that measure to proxy variable trade costs.

(5.) Market share is calculated using total sales plus imports in the denominator (industry-specific).

(6.) We owe this idea to an anonymous reviewer.

(7.) We owe this idea to an anonymous reviewer.

(8.) We thank anonymous reviewers for some of these observations.

SOOIL KIM, JEFFREY J. REIMER, and MUNISAMY GOPINATH *

* We acknowledge financial support from the Oregon Agricultural Experiment Station and the USDA Cooperative State Research, Education and Extension Service. We thank Hanho Kim of Seoul National University for providing access to Korean firm level data. We appreciate helpful feedback from anonymous reviewers and from participants at several conferences and seminars

Kim: Ministry for Food, Agriculture, Forestry and Fisheries, Government Complex Gwacheon, Jungang-dong, Gwacheon, Gyeonggi-do, South Korea. Phone 82-02503-7200, Fax 82-02-503-7249, E-mail [email protected]

Reimer." Department of Agricultural and Resource Economics, Oregon State University, 213 Ballard Extension Hall, Corvallis, OR 97331. Phone 541-737-1415, Fax 541-737-2563, E-mail [email protected]

Gopinath: Department of Agricultural and Resource Economics, Oregon State University, 213 Ballard Extension Hall, Corvallis, OR 97331. Phone 541-737-1402, Fax 541-737-2563, E-mail [email protected]

TABLE 1
Number of Firms to Enter, Exit, and Export

 No. of Entering Firms No. of Exiting Firms
Year (Share of Total) (Share of Total)

1992 226 (0.127) 121 (0.068)
1993 137 (0.081) 123 (0.073)
1994 264 (0.148) 135 (0.075)
1995 463 (0.225) 132 (0.064)
1996 1,712 (0.474) 271 (0.075)
1997 1,844 (0.367) 266 (0.052)
1998 1,611 (0.253) 623 (0.098)
1999 1,986 (0.265) 910 (0.121)
2000 1,655 (0.205) 1,376 (0.170)
2001 1,436 (0.174) 1,552 (0.189)
2002 2,049 (0.222) N/A
Average 1,217 (0.23) 551 (0.10)

 No. of New Exporters
Year (Share of Total) Total No. of Firms

1992 96 (0.053) 1,778
1993 53 (0.031) 1,679
1994 63 (0.035) 1,782
1995 83 (0.040) 2,055
1996 140 (0.038) 3,611
1997 181 (0.036) 5,021
1998 210 (0.033) 6,351
1999 268 (0.035) 7,489
2000 223 (0.027) 8,062
2001 182 (0.022) 8,210
2002 215 (0.023) 9,195
Average 156 (0.03) 5,021

Notes: Our data include all manufacturing sectors for Korea, and
account for more than 60% of total and individual industry
output. We are unable to observe exit data for 2002 because we
would need information for subsequent years.

TABLE 2
Firm Characteristics (Averages)

 Capital/labor
 (K/L) Ratio
 Employees (Thousand Won per
 (Number) Worker)

Entering firms 36.6 40,341
All firms except entering firms 161.9 62,202
Exiting firms 100.4 65,544
All firms except exiting firms 158.9 56,023
New exporters 117.0 62,252
All firms except new exporters 134.2 57,425
All firms together 133.7 57,564

 Gross Sales Ratio of Capital/
 (Million Won) Gross Sales

Entering firms 3,735 3.13
All firms except entering firms 27,071 0.49
Exiting firms 18,789 0.48
All firms except exiting firms 41,760 0.45
New exporters 28,742 0.36
All firms except new exporters 35,884 0.45
All firms together 21,477 1.12

TABLE 3
Hvnothesis 1

 Expected
Variable Sign Variant 1 Variant 2

Dependent variable: Probability of
entering the domestic market

Change in import-trade costs + 0.62 *** 0.44 ***
 (<0.01) (<0.01)
Change in export-trade costs - 7.91 *** 6.56 ***
 (<0.01) (<0.01)
Change in two-way trade costs [+ or -]
 ([dagger])
Log(employment) -1.31 ***
 (<0.01)
Log(K/L) -0.30 ***
 (<0.01)
Constant -1.10 *** 2.08 ***
 (<0.01) (<0.01)
Industry fixed effects Yes Yes
Observations
 55,942 55,942
Log likelihood -30,636 -28,407

Variable Variant 3 Variant 4

Dependent variable: Probability of
entering the domestic market

Change in import-trade costs

Change in export-trade costs

Change in two-way trade costs 19.58 *** 27.17 ***
 ([dagger]) (<0.01) (<0.01)
Log(employment) 1.40 ***
 (<0.01)
Log(K/L) -0.27 ***
 (<0.01)
Constant -0.93 *** 2.31 ***
 (<0.01) (<0.01)
Industry fixed effects Yes Yes
Observations
 55,942 55,942
Log likelihood -30,523 -28,123

Notes: p value is in parenthesis. *, **, *** denote significance
at the 10%, 5%, and 1% level, respectively.

([dagger]) This is a tariff-equivalent measure of trade costs
that does not distinguish between importing and exporting. The
expected sign may be positive or negative depending on the extent
to which the measure reflects export-trade costs versus
import-trade costs.

TABLE 4
Hypothesis 2

Variable Expected Sign Variant 1 Variant 2

Dependent variable: Change in domestic market share

Change in import-trade costs + 0.0004 ** 0.0004 *
 (0.03) (0.08)
Change in two-way trade +
costs ([dagger])

Log(employment) -0.0001
 (0.28)
Log(K/L) 0.0001
 (0.26)
Constant 0.0004 ** 0.0002
 (0.01) (0.57)
Industry fixed effects Yes Yes
Observations 13.576 13,576
Log likelihood 57,133 57,134

Variable Variant 3 Variant 4

Dependent variable: Change in domestic market share

Change in import-trade costs

Change in two-way trade 0.01 *** 0.01 ***
costs ([dagger]) (<0.01) (<0.01)

Log(employment) -0.0001
 (0.06)
Log(K/L) 0.0001
 (0.21)
Constant 0.0005 *** 0.0003
 (<0.01) (0.42)
Industry fixed effects Yes Yes
Observations 13,576 13,576
Log likelihood 57,142 57,144

Notes: p value is in parenthesis.

*, **, *** denote significance at the 10%, 5%, and 1% level,
respectively.

([dagger]) The expected sign on this variable holds to the extent
that it corresponds to import-trade costs.

TABLE 5
Hypothesis 3

Variable Expected Sign Variant 1 Variant 2

Dependent variable: Change in firm-level productive inefficiency

Change in import-trade costs + 0.42 ** 0.03
 (0.01) (0.11)

Change in two-way trade +
costs ([dagger])

Log(employment) -0.04 ***
 (<0.01)

Log(K/L) -0.01
 (0.26)

Constant 0.004 0.12 ***
 (0.77) (<0.01)

Industry fixed effects Yes Yes
Observations 3,209 3,209
[R.sup.2] 0.05 0.07

Variable Variant 3 Variant 4

Dependent variable: Change in firm-level productive inefficiency

Change in import-trade costs

Change in two-way trade -0.71 *** -0.58 ***
costs ([dagger]) (<0.01) (<0.01)

Log(employment) -0.04 ***
 (<0.01)

Log(K/L) -0.01
 (0.15)

Constant 0.005 0.13 ***
 (0.72) (<0.01)

Industry fixed effects Yes Yes
Observations 3,209 3,209
[R.sup.2] 0.06 0.07

Notes: p value is in parenthesis. *, ** *** denote significance
at the 10%, 5%, and 1% level, respectively.

([dagger]) The expected sign on this variable holds to the extent
that it corresponds to import-trade costs.

TABLE 6
Hypothesis 4

 Expected
Variable Sign Variant 1 Variant 2

Dependent variable: Change in industry-level productive inefficiency

Change in import-trade costs + 0.081 *** 0.081 ***
 (<0.01) (<0.01)

Change in export-trade costs + 0.069 0.069
 (0.85) (0.85)

Change in two-way trade costs +

Constant 0.019 *** 0.023 *
 (<0.01) (0.09)

Industry fixed effects No Yes
Observations 80 80
[R.sup.2] 0.09 0.20
Dependent variable: Firm sales
Change in export-trade costs - -5.52 *** -6.01 ***
 x inefficiency (<0.01) (<0.01)
Change in two-way trade costs -
 x inefficiency
Constant 0.09 *** 0.11 ***
 (<0.01) (<0.01)
Industry fixed effects No Yes
Observations 8,284 8,284
[R.sup.2] 0.03 0.05

Variable Variant 3 Variant 4

Dependent variable: Change in industry-level productive inefficiency

Change in import-trade costs

Change in export-trade costs

Change in two-way trade costs 0.447 0.447
 (0.17) (0.17)

Constant 0.022 *** 0.026 **
 (<0.01) (0.048)

Industry fixed effects No Yes
Observations 80 80
[R.sup.2] 0.02 0.14
Dependent variable: Firm sales
Change in export-trade costs
 x inefficiency
Change in two-way trade costs 2.39 *** 2.63 ***
 x inefficiency (<0.01) (<0.01)
Constant 0.12 *** 0.13 ***
 (<0.01) (<0.01)
Industry fixed effects No Yes
Observations 8,284 8,284
[R.sup.2] 0.01 0.02

Notes: p value is in parenthesis. *, **, *** denote significance
at the 10%, 5%, and 1% level, respectively.

TABLE 7
Hypothesis 5

 Expected
Variable Sign Variant 1 Variant 2

Dependent variable: Probability that nonexporting firms become
exporters

Change in export-trade costs - -3.42 -3.30
 (0.40) (0.43)
Change in two-way trade -
 costs ([dagger])
Log(employment) 0.007
 (0.95)
Log(K/L) -0.104
 (0.34)
Constant -3.41 *** -2.94 ***
 (<0.01) (<0.01)
Industry fixed effects Yes Yes
Observations 20,610 20,610
Log likelihood -1,928 -1,927
Dependent variable: Indicator for being an exporter
Export-trade costs - 0.94 (0.50) 1.21 (0.42)
Two-way trade costs ([dagger]) -

Log(employment) -0.04
 (0.69)
Log(K/L) -0.10
 (0.37)
Constant -5.11 ** -5.10 *
 (0.04) (0.07)
Industry fixed effects Yes Yes
Observations 20,610 20,610
Log likelihood -1,928 -1,928

Variable Variant 3 Variant 4

Dependent variable: Probability that nonexporting firms become exporters

Change in export-trade costs

Change in two-way trade 1.23 1.26
 costs ([dagger]) (0.70) (0.70)
Log(employment) -0.016
 (0.87)
Log(K/L) -0.10
 (0.35)
Constant -3.35 *** -2.86 ***
 (<0.01) (<0.01)
Industry fixed effects Yes Yes
Observations 20,610 20,610
Log likelihood -1,928 -1,928
Dependent variable: Indicator for being an exporter
Export-trade costs
Two-way trade costs ([dagger]) 1.40 1.51
 (0.52) (0.50)
Log(employment) -0.03
 (0.80)
Log(K/L) -0.10
 (0.37)
Constant -4.55 ** -4.16 **
 (0.02) (0.04)
Industry fixed effects Yes Yes
Observations 20,610 20,610
Log likelihood -1,928 -1,928

Notes: p value is in parenthesis. *, **, *** denote significance
at the 10%, 5%, and 1% level, respectively.

([dagger]) The expected sign on this variable holds to the extent
that it corresponds to export-trade costs.

TABLE 8
Hypothesis 6

 Expected
Variable Sign Variant 1 Variant 2

Dependent variable: Change in log exports

Change in export-trade costs - 1.38 3.00 *
 (0.35) (0.05)
Change in two-way trade -
 costs ([dagger])
Log(employment) -0. 16 ***
 (<0.01)
Log(K/L) -0.04
 (0.33)
Constant 0.10 0.58 ***
 (0.34) (<0.01)
Industry fixed effects Yes Yes
Observations 20,610 20,610
[R.sup.2] 0.001 0.002

Variable Variant 3 Variant 4

Dependent variable: Change in log exports

Change in export-trade costs

Change in two-way trade 0.61 1.25
 costs ([dagger]) (0.61) (0.29)
Log(employment) -0.15 ***
 (<0.01)
Log(K/L) -0.04
 (0.34)
Constant 0.09 0.52 **
 (0.42) (0.01)
Industry fixed effects Yes Yes
Observations 20,610 20,610
[R.sup.2] 0.001 0.002

Notes: p value is in parenthesis.

*, **, *** denote significance at the 10%, 5%, and 1% level,
respectively.

([dagger]) The expected sign on this variable holds to the extent
that it corresponds to export-trade costs.

TABLE 9
Hypothesis 7

 Expected
Variable Sign Variant 1 Variant 2

Dependent variable: Probability of exit

Change in import-trade costs - 0.98 *** 0.06
 (<0.01) (0.73)
Change in two-way trade -
 costs ([dagger])
Log(employment) -0.81 ***
 (<0.01)
Log(K/L) -0.06
 (0.23)
Constant -2.39 *** -0.60 **
 (<0.01) (0.04)
Industry fixed effects Yes Yes
Observations 13,576 13,576
Log likelihood -5,120 -4,986

Variable Variant 3 Variant 4

Dependent variable: Probability of exit

Change in import-trade costs

Change in two-way trade -6.62 *** -7.19 ***
 costs ([dagger]) (<0.01) (<0.01)
Log(employment) -0.82 ***
 (<0.01)
Log(K/L) -0.07
 (0.19)
Constant -2.20 *** -0.55 *
 (<0.01) (0.06)
Industry fixed effects Yes Yes
Observations 13,576 13,576
Log likelihood -5,136 -4,981

Notes: p value is in parenthesis.

*, **, *** denote significance at the 10%, 5%, and 1% level,
respectively.

([dagger]) The expected sign on this variable holds to the extent
that it corresponds to import-trade costs.