The views expressed in the articles are solely those of the author(s) and do not represent the organization to which the author(s) belong or the Research Institute of Economy, Trade and Industry. This study uses micro data on inter-company transactions collected by Tokyo Shoukou Research (TSR), which is provided by RIETI, and the "Basic Survey of Japanese Business Structure and Activities" which is conducted by the Ministry of Economy, Trade and Industry. (METI). To guide the empirical analyses, we use a simpler version of the theoretical framework in Ahn et al.

In the model, two export methods are available to producers (direct and indirect export through intermediaries). Due to fixed costs, the net export profit can be negative for both export modes. In this case, the company does not export and sells only on the domestic market.

The difference in the fixed costs fD fI can be recovered very quickly as it increases. For more detailed explanations and analyzes of the TSR network data, see Bernard et al. We define indirect exporters as manufacturing companies that supply their products to exporting wholesalers to track indirect shipments of the products.

## Descriptive Statistics

Exporting (or international) wholesalers are identified by the 2-digit JSIC code between 50 and 55, and their active export status. We count them as direct exporters as they could overcome the fixed cost of direct exporting, which is the interest of study. Roughly speaking, the median indirect exporter is three times larger than the median domestic firm, and the median direct exporter is four times larger than the median indirect exporter.

The differences in magnitude can also be seen in the last two columns of Table 1, which show the share of total sales and employment of each export type. Direct exporters represent only 6% of the number of companies, but account for 56% and 40% of total turnover and employment respectively. The sorting by company size is not generated by a small portion of companies at the bottom or top.

Direct exporters are larger than indirect exporters, and indirect exporters are larger than domestic firms in the sense of first-order stochastic dominance (FOSD). At any sales percentile, a direct exporter is larger than an indirect exporter, which is larger than a domestic firm. In the empirical analysis, employment is used to calculate labor productivity, which is defined as sales per employee.

In-degree and out-degree are used as covariates in the multinomial logit model in Chapter 4. In-degree and out-degree are also positively correlated with sales, but the correlation is stronger for out-degree. Another way to see the relationship between size and export status is to calculate the share of each type of exporter by revenue percentile group.

In the bottom decile, more than 90% of firms are domestic and there are few direct exporters. The share of exporters in the sales deciles per industry can be found in Figure 10 in the appendix.

## Multinomial Logit Model for the Choice of Export Status

Based on a multinomial logit model, the probabilities of indirect and direct export are written as follows. The results are consistent with the model predicting a lower intercept but a steeper slope of direct export sales. The score of indirect export is higher than that of direct export as the indirect curve starts from a higher intercept.

As sales increase, the gap between the lines of indirect and direct exports decreases due to the higher slope of direct exports. We can see that the probability of not exporting (domestically) decreases monotonically and the probability of exporting directly increases monotonically with sales. When sales become very large, the probability of direct exports begins to dominate the probability of indirect exports.

Like specification (2), the coefficient of out-grade is positive for both indirect and direct exports, and the slopes of timber sales become smaller for both comparisons. The slopes of sales and in-grade are higher for direct export, while the slope of out-grade is smaller. There may be industry heterogeneity in the propensity of indirect and direct exporting due to differences in distribution networks, tariffs, degree of product differentiation, and so on.

With all specifications, the cut is lower and the slope of timber sales is higher for direct export. If a manufacturer has many suppliers, it may be able to split the fixed costs of direct exports and charge an additional margin to each supplier in the form of a smaller payment. More suppliers may also be associated with the producer's greater bargaining power, which lowers marginal costs and increases the incentive to export directly.

In either specification, the cross section is lower and the labor productivity slope is steeper for direct exports. In the last regression of the bottom panel, we can see the same implications, such as higher labor productivity and in-rate slopes, but a lower output-rate slope for direct exports.

## Industry Heterogeneity

We also perform the same multinomial logit analyzes with labor productivity (sales per employee) instead of sales. Although there is a one-to-one relationship between productivity and sales in the model due to CES demand and monopolistic competition, it is important to confirm the productivity predictions in the empirical analysis as well. As before, the top panel shows the baseline results without fixed effects and the bottom panel shows the results with industry and prefecture fixed effects.

JSIC FE 2 digits No No No No. Prefecture FE No No No No JSIC FE 2 digits Yes Yes Yes Yes Yes. Prefecture FE Yes Yes Yes Yes Yes. b) industry and prefecture fixed effects. From Table 3, we know that the industry fixed effects do not absorb the variations in log sales, at scale and out of scale. To examine the differential effects of size on the choice of export mode by industry, we estimate the multinomial logit model with only log sales (this corresponds to specification (1)) for each industry.

The upper graph indicates that there is a strong positive correlation between the intercepts of indirect and direct exports with a correlation coefficient of 0.666, but the intercept of direct exports is lower than that of indirect exports in all industries (they all lie below the 45 -degree line). 6 Investigating the source of industry heterogeneity in the shares of indirect and direct exporters is an important agenda for future research. We can also calculate the upstreamness of industries using the method described in Antras et al. 2012) and see its correlation with the shares of indirect and direct exporters.

There is also a positive correlation between the slope coefficients, as shown in the bottom graph. As predicted, all points are above the 45 degree line, indicating that the estimated slope of direct exports is higher across all industries. It is interesting to see that the electronic machinery and chemical products sectors are smaller for both indirect and direct exports, yet have a higher share of exporters.

From this analysis we can see that they have greater export intensity because of lower fixed export costs, and not because their size distribution is larger. The opposite is true for the iron and steel industry, which is characterized by higher fixed costs but steeper sales slopes for both indirect and direct exports.

## Logit Model for Wholesalers’ Export Decision

Nevertheless, we can see from Figure 7 that the iron and steel industry has a higher share of exporters. Wholesalers who have more suppliers tend to export after checking size, sector and prefectural averages. Because exporting wholesalers generally have more suppliers, the influence of indirect exports through international wholesalers is greater.

In contrast to the results for the export status of producers, the estimated coefficient of the deregistration rate is negative and significant in all specifications. This means that wholesalers who have more customers in the domestic market are less inclined to export. This also refers to the role of exporting producers as wholesalers in the so-called carry trade.

This paper reveals the characteristics of direct and indirect exporters relative to domestic firms and examines which factors are related to firms' choice of export mode. The distributions of sales, employment, labor productivity, on-scale and off-scale by export status are listed in terms of FOSD. Multinomial logit analysis for firms' choice of export mode provides strong evidence of the higher fixed cost but lower variable cost of direct exporting compared to the alternative of exporting indirectly through intermediaries.

We also find that in-degree increases the probability of direct export, which implies a cost-sharing mechanism for companies with several suppliers. This means greater product appeal and a broader demand base for companies that have more customers in the domestic market. Product industries with a high degree of processing or a high degree of product differentiation, such as chemical products and electronic machinery, have lower fixed costs but higher variable costs of exporting.

Compared to industries upstream in supply chains, these industries have a higher share of exporters and a disproportionately higher share of direct exporters. Khandelwal and Shang-Jin Wei (2011) “The Role of Intermediaries in Facilitating Trade”, Journal of International Economics, Vol. Antras, Pol and Arnaud Costinot (2010) “Intermediation and economic integration”, American Economic Review, Vol. 2011) “Intermediated Trade”, Quarterly Journal of Economics, Vol.

Crozet, Matthieu, Guy Lalanne and Sandra Poncet (2013) "Wholesalers in International Trade", European Economic Review, Vol. 2003) “The Impact of Trade on Intra-Industry Reallocations and Total Industry Productivity”, Econometrica, Vol. 1999) "Networks versus Markets in International Trade", Journal of International Economics, Vol.