Credit limits

Размещено на http://www.allbest.ru/

Contents

Motivation

Research Question

Overview of Existing Literature and My Contribution

Data and Summary Statistics

Research Methodology and Results

Conclusion

Reference List

Appendix



Motivation

The main driver of growth of export flows for a country are the new firms entering foreign markets (Melitz and Redding, 2014) and while the new firms contribute up to 80% in developed markets, new Russian exporters only contribute around 50%, which can be partly attributed to the lack of export diversification in the last decade, according to a survey of a sample of Russian firms conducted by CEFIR in 2011. The CEFIR report on the provision of services on collection and analysis of information on the topic "Identifying and assessing barriers to the entry of Russian companies in the manufacturing sectors to foreign markets" for the needs of the federal state educational institution of higher professional education "Russian Presidential Academy of National Economy and Public Administration", Moscow, 2011, supervisor N.A. Volchkova Low export diversification of Russian economy (70-80% are exports of resource-oriented industries, according to Federal State Statistics Service) Federal State Statistics Service: <http://www.gks.ru/free_doc/new_site/vnesh-t/docl/osn_razd/stru-ex.htm> leads to the fact that decreases in oil price act as a huge headwind to the economy.

The survey of potential and existing Russian exporters done by CEFIR shows that the low level of financial markets' development and accessibility of credit resources is one of the most important impediments for Russian exporters. These firms, according to empiric results¹, are facing a number of problems with either entering export markets or sustaining their exports if they have already entered foreign markets, partly due to insufficient availability of credit. Therefore, it is worth to study the credit constraints faced by exporters and compare them to those faced my non-exporters.

The rest of this paper is structured as follows. Section II specifies the research question. Section III provides an overview of existing literature on the topic and my contribution. Section IV describes the data used in the analysis and contains summary statistics. Section V contains the description of the research methods and the results. Section VI concludes, sections VII contains the reference list and section VIII contains the appendix.



Research Question

Studies of export decisions and export costs (such as, for example, Baldwin & Krugman (1989), Dixit (1989)) argue that firms must incur sunk costs in order to enter a foreign market. Literature defines such costs as costs of acquiring skilled labor force (Brambilla, Lederman & Porto (2009), Molina & Muendler (2009)), costs needed to overcome technical barriers, such as updating and enhancing the product standards to make them comply with higher foreign standards (Felbermayr & Jung (2008), Maskus, Otsuki & Wilson (2005)). Other examples include but are not limited to acquiring information about a foreign market, building a foreign distribution chain, marketing expenses, etc.

Firms that start exporting need funding in order to be able to pay for these entry costs. On top of that, in order to get working capital financing, firms usually have to post collateral for such borrowing in the form of their assets or of the product produced. As found by Halldin (2012) on Swedish data, tangible and collateralizable assets are a significant factor of export activity. Muuls (2015) also notes that “firms need to accumulate sufficient collateral before they can borrow enough funds to profitably export”. In export transactions, however, the product can no longer serve as good collateral from the point of view of domestic banks since it leaves domestic borders. This logic is presented in Feenstra, Li & Yu (2011): they argue that credit for export purposes involves more risk since payment enforceability across borders is more difficult. Also, it takes more time to receive payment for an export sale than for a domestic one. Muuls (2015), for example, emphasizes that international shipping increases the time lag between paying trade costs and realization of revenue by 90 days, compared to domestic sales. These factors decrease the opportunity to use a firm's own product as collateral, increase the cost of financing and decrease its availability. These financial constraint considerations shape the firm's decision whether to export or not.

The parameters that most of existing literature uses as proxies for financial constraints are balance sheet metrics such as, for example, ratio of total debt over total assets, cash flow over total assets, tangible assets over total assets (Berman & Hericourt, 2008). financial export status firm

In my research I use metrics that are individual to each loan (interest rate on each firm's most recent loan, maturity of the loan, collateral to loan ratio and the extent to which the granted amount satisfied the amount requested by the firm). I reckon that the use of these specific loan metrics should provide a better estimate of the level of financial constraint than the more general balance sheet metrics pertaining to the whole firm that are used in existing studies. Therefore, I believe that it is of interest to employ these loan metrics that I have to study how a firm's export status affects the external credit constraints that the firm faces. I will also estimate the effect of loan use (domestic or export related) on the parameters of credit constraints faced by exporting firms. On top of the 4 parameters mentioned above, I also have data on each of the firms' own estimates of the level of credit constraints that it faces. This information is of a different nature since it applies to the whole firm and not a particular loan, but is also of interest. So, I will provide results on this metric separately.

Overview of Existing Literature and My Contribution

Traditionally, country-level differences in factor endowments and in productivity have been central to research on international economics. Modern trade theories have introduced firm heterogeneity and analysis of the fixed and variable costs of becoming an exporter. If such costs are very high even some highly productive firms will not be able to enter foreign markets. Many studies have focused on analyzing international trade on a firm level rather than on a country or industry level. The theoretical framework that has shaped a significant part of the studies was developed in Melitz (2003) model in which the heterogeneity of individual firms influences the macro results.

Bernard, Jensen, Redding and Schott (2011) summarize the empirical papers that use micro-level data on firms and one of their conclusions is that “firm participation in international trade is exceedingly rare”. They quote a number of previous studies that have found that “exporters and importers are larger, more productive, more skill- and capital-intensive, and pay higher wages prior to their entry into international markets than non-trading firms”. So, according to them and the Melitz model, the low participation of firms in international trade might be due to the low levels of productivity that are insufficient to combat the high entrance costs.

Roberts and Tybout (1997) study data on Colombian manufacturers gathered over the 1980s, a period when the country experienced significant shifts in exchange rate and in demand. One of the noteworthy results they get is that prior export status is an important determinant of decision to export but that it wanes quickly over time, i.e. a firm whose last export was one year before is much more likely to export in the current year than a firm whose last export was two years ago. More importantly, they find that sunk costs of exporting are statistically significant and that “only the most productive firms find it profitable to incur” sunk costs of entry into export markets. They state that the causality effect runs in this particular direction and effects of learning by exporting are less clear.

Historically, however, it has been difficult to capture the internal problems faced by firms precisely. It also was not easy to exactly measure differences in funding availability among different countries, so there are not too many works in international economics that offer significant insight into these more corporate finance related factors. Most of them looked only at physical, as opposed to financial, capital and considered it immobile across countries. Most corporate finance literature, on the other hand, has been focused on firms that operate in one country. However, some deviations from the traditional frameworks started to appear that took two important factors into consideration: physical capital that is mobile across borders and financial capital which availability differs among countries. For example, La Porta et al. (1998) studied the differences in financial capital accessibility across countries and found that they could be explained by different levels of legal protection. Rajan and Zingales (2003) added political factors to this explanation. Further research showed that, despite the recent surge in international debt issuance, firms that operate only in one country often depend on raising funding locally. So, the level of financial development of a country can be crucial for local firms considering engaging in some international activity. Manova and Foley (2014) provide a comprehensive summary of the work done in this field and conclude that “the cumulative evidence convincingly establishes the significance of well-functioning financial markets for global trade”.

Manova (2013) finds that the impact of improvement in financial development, and thus in availability of credit, is of the same order of magnitude as that of a similar rise in human capital endowment and significantly more than that of an increase in stock of physical capital. Manova studies cross-sectoral data on 161 countries and 27 sectors from 1995. She builds a multi-industry, multi-country model that incorporates heterogeneous firms that incur fixed and variable production and trade costs, as well as different levels of financial development among countries and different levels of financial vulnerability among sectors. Her results show that if enterprises need external funding to cover fixed costs of trade only, then the presence of credit constraints increases the minimal level of productivity required to become an exporter; and if external funding is also needed to cover variable costs, companies will reduce their export sales below their first best option in the presence of credit restrictions. These effects are exacerbated in sectors highly dependent on external finance and in countries with poor financial institutions. Besides confirming the importance of a country's financial development for export decisions, she concludes that the impact of development on exports in particular is greater than that on overall production, with a firm's decision to start exporting accounting for a third of the effect on exports and decrease in foreign sales of existing exporters accounting for two thirds. Higher level of financial development of a country also leads to that country exporting a richer variety of products, exporting to more destinations and to smaller counterparties, as well as having more exporters from sectors that are deemed financially vulnerable. The cross-industry comparison that she also conducts shows that these effects are more pronounced in sectors that heavily depend on external finance versus those that do not. This result highlights the importance of studying credit constraints that firms in developing countries with rather poor level of financial institutions face and, in particular, it reinforces my motivation to study results gathered on Russian firms.

Muuls (2008) analyses firm-level balance sheet data on Belgian companies on top of country-level trade data for Belgium from 1999-2005. One metric of credit constraints used by Muuls that is worth noting is the Coface score that is used as a measure of credit constraints faced by the firm. This metric is computed by a special agency Coface International and is calculated based on measures similar to those used by banks in their lending decisions, so is a relevant measure for testing how credit constraints affect export behaviour. Muuls finds that firms have a higher probability to start exporting if they are more productive and less constrained in access to funding. Another finding is that credit limitations influence only extensive margin of trade in regards to destination market. Muuls also finds that “pecking order of trade holds” in his data, i.e. “firms exporting to the smallest and furthest away economies are more productive and less credit-constrained”.

Minetti and Chun Zhu (2011) also focus on analysing the effect of credit constraints on exporting behaviour of companies. They use data based on answers to a survey conducted on more than 4,500 Italian firms rather than any balance-sheet metrics or any credit agencies scores. Using necessary controls for endogeneity of credit constraints they conclude that credit rationing significantly reduces the probability of being an exporter and, for firms that already export, credit rationing substantially diminishes their sales abroad. They note that such rationing has a much higher impact on foreign sales than on domestic sales. Moreover, they also analyse different industries' sensitivities to credit restrictions and find that high-tech firms and firms in industries that have a high reliance on external funding are those that are most heavily influenced by credit rationing. The main finding of their work is that “limited access to liquidity appears to impact both the probability that a firm exports (the “extensive margin”) and firm-level exports, conditional on exporting (the “intensive margin”).

Chaney (2013) studies the relation between funding availability and exporting decisions by building a theoretical model where he adds liquidity constraints to the Melitz model from 2003. His results support the prevailing hypothesis that total exports will be higher the better access to liquidity there is and the more equally this liquidity is distributed. Chaney's work shows that the positive correlation between productivity, availability of funding and export activity could sometimes be less straightforward than in most of the models. If, in a Melitz framework, firms draw not only their productivity levels but also “liquidity endowments”, then some firms with low productivity might end up having a high liquidity endowment which will allow them to engage in export activity.

Manova, Wei and Zhang (2014) use official Chinese trade data from 2005 (value of firm exports by product and destination country) to estimate the effect of credit restrictions on export and FDI decisions. They find that joint ventures and foreign affiliates export more than pure domestic firms in sectors dependent on external finance, the reason for that being better access to finance of the former. This effect is very substantial and far larger than effects of any other factors such as contractual imperfections, input costs and property rights.

Berman and Hericourt (2008) study data on 5,000 firms from 9 countries and find that both availability of external finance and productivity increase the probability of exporting, noting that credit constraints (proxied by three balance sheet metrics) is only a significant determinant of entry probability and not of amount of exports or probability of continuing to export. They also find that higher productivity increases exporting probability but only as long as there is sufficient credit availability.

Manova and Chor (2012) study data on US imports during 2006-2009 and find that during the crisis the countries that had more credit constraints (as measured by higher interest rates) exported less to the US than those with less credit constraints (lower rates) and that this effect is even more significant for financially vulnerable industries.

The studies summarized above all indicate that credit constraints influence export decisions of firms (in some papers this is called the self-selection hypothesis). However, there are a number of studies that find that the link in another direction also exists, i.e. that export status influences financial and credit constraint parameters of firms. Below is a summary of some of the studies that support this ex-post effects hypothesis.

Campa and Shaver (2002) study data on Spanish manufacturers from the 1990s and find that exporters face less liquidity constraints (as measured by the sensitivity of their investments to cash flows) than non-exporters thanks to the fact that their diversified exposure to business cycles helps them receive more stable cash flows. They argue that the positive signal about cash flow stability that export status conveys also decreases liquidity constraints faced by exporters.

Greenaway, Guariglia and Kneller (2007) show that continuous exporters (though not those who recently started) are more financially sound, i.e. have lower leverage and better liquidity, than non-exporters based on panel data of more than 9,000 UK manufacturing firms over the 90s. They find that that those firms that have better financial metrics are not more likely to become exporters. They argue that the linkage goes another way round: exporting activity benefits a firm's financial characteristics. The authors note three channels that could possibly be at work here. First explanation is that exporters can obtain loans from both domestic and foreign sources and this kind of diversification is beneficial for financial health. Tornell and Westermann (2003) provide a similar explanation arguing that foreign currency denominated revenues can act as collateral needed to tap international funding markets. Second, export status can act as a signal that the firm is productive enough to cover the sunk costs of exporting. This signal lowers the liquidity constraints faced by the firm. Ganesh-Kumar, Sen and Vaidya (2001) obtain a similar result on Indian data and note that this channel is particularly important for emerging markets with poor quality institutions. Third, exporters are less dependent on the macroeconomic situation in the home market and thus their risks are more diversified, an argument in line with that used by Campa and Shaver (2002), mentioned above.

Feenstra, Li and Yu (2015) study the link between export status and parameters of credit constraints by focusing on the problem of longer time that is needed to receive revenue for export sales versus domestic sales, of greater risks associated with exports and of extra costs needed to participate in international trade. They assume that a firm's productivity level is unknown to lending institutions and that they will set such terms of credit that make the firm reveal this information correctly. Such a setting leads to the presence of credit constraints. Credit constraints are defined in the empirical part of the paper as the amount by which the actual sales revenue of a firm is less than the revenue it would have, had it not had working capital financing needs. An important result is that the longer time lag needed for export sales is the most important determinant of credit constraints and that it negatively affects both extensive and intensive margins of trade. Also, since in their model banks can't discern whether a loan by an exporting firm is taken out for export or domestic activity, exporters are more credit constrained than non-exporters in both their export and domestic sales. This model also predicts that an exporter should face the same level of credit limitations for its domestic and international activity loans. The authors show that these results hold by analyzing a large panel data set on Chinese manufacturing companies from 2000-2008. This is consistent with some earlier results, for example, of Behrens, Corcos and Mion (2010) for Belgian firms but contradicts other researchers' results, like those of Amiti and Weinstein (2009) for Japanese firms. The latter show that financial constraints, which they measured as the health of the main bank, impacts export activity 5 times as much as domestic activity. One explanation proposed by Feenstra is that these results are focused on loans extended specifically for export purposes while in his work he studies all working capital loans. The different views on the issue of the purpose of the loan make it worth to study this further and, as I discuss in more detail later, my research shows that Russian exporters have similar interest rates (i.e. similar credit constraints) on their export-related loans and loans for purely domestic purposes.

Joachim Wagner (2013) summarizes the empirical studies that have been conducted using firm level data and finds that most of the results more or less confirm that exporters face less financial limitations than non-exporters. Most studies find that the firms that face less constraints self-select into exporting and that exporting does not exert a positive influence on a firm's financial metrics. However, Wagner warns that these results “should not be considered as stylized facts” because of the different metrics used as proxies for financial constraints, different empirical models and the insufficient of “export status switchers”.

My Contribution

First, as mentioned above, there is no consensus in existing papers regarding the direction of the link between export status and credit constraints, so more research is necessary to shed the light on the issue. My paper studies the effect of export status on credit constraints (i.e. tests the ex-post effects hypothesis) and finds empirical evidence in favour of the effect. Moreover, there is also ambiguity in the literature regarding whether exporters face different levels of credit constraints for their domestic and export purpose loans. I find evidence that Russian exporters face a similar level of credit constraints on their export-related loans as on domestic ones.

Second, existing literature does not discuss individual loan metrics and, as I explain in the literature overview section, focuses instead on some other parameters (such as balance sheet data or scores calculated by some special agencies on the firm level). In my paper, I will focus on the metrics of credit rationing that have not been studied in the existing literature: the extent to which the loan was granted, the effective interest rate, percent of collateralization and maturity of the loan. I will estimate which ones differ the most between exporters and non-exporters.

Finally, I will analyse firm-level data on Russian companies (previous studies have mainly been based on data from other countries).



Data and Summary Statistics

My analysis is conducted using firm-level data collected in 2011 via a survey of two groups of Russian companies: existing exporters (1108 companies) and firms that have the potential to start exporting but currently do not (406 companies) The CEFIR report on the provision of services on collection and analysis of information on the topic "Identifying and assessing barriers to the entry of Russian companies in the manufacturing sectors to foreign markets" for the needs of the federal state educational institution of higher professional education "Russian Presidential Academy of National Economy and Public Administration", Moscow, 2011, supervisor N.A. Volchkova .

The survey's aim was to determine what are the main obstacles that undermine export practices of the companies, to ascertain their scale and to establish suggestions for economic policy. The method of collecting the information was a formal interview with the company representatives that was approximately 60-90 minutes long and covered not less than 100 questions. Data was collected according to strict guidelines and was tested for potential mistakes.

The survey has a number of blocks on different types of issues and associated costs: personnel policies, economic and financial barriers, currency risks, administrative barriers, informational barriers, support of export activities from government-related entities, logistics issues, technical problems, legal costs and others. I will focus on the block containing questions and answers regarding financing issues.

In this block I have data on: employment, turnover, age, industry, region where the firm is based, the ownership structure (percent of the firms that belongs to foreigners and percent belonging to state), breakdown of costs by currency etc. Also, I have data on the parameters of the latest loan that was granted to the firm: extent loan granted, interest rate, collateral to loan ratio, number of months the loan was granted for, severity of financial constraints, year when the last loan was granted, type of bank that gave the last loan (private commercial bank, private commercial bank with foreign participation or government bank/bank with government participation), currency of the loan, business dimension for which the last loan was taken (developing business in home market, developing business in existing export markets, expanding the range of export products in existing export markets or expanding the number of export destinations), type of activity for which the last loan was used (working capital, capex, acquisition of a stake in another firm, repayment of other debts), etc.

Summary Statistics

The main variables I am going to use are defined as:

· Exporter = 1 if a firm is an exporter and 0 otherwise.

· Employment is the number of employees the firm had in 2009.

· Turnover is the total sales of the firm in 2009.

· Age is the total number of years the firm is in existence.

· Industry: I used dummy variables (din1 din2 din3 din4 din5) to form the 5 broad industry categories out of the 21 that were available directly from the data. I did this redefining in order to get rid of collienearity issues that arise when using the initial industry specification. I also drop industry group 3 (din3) to omit collinearity since it is the most frequent type.

The industry groups are: 1) food and textiles; 2) wood, chemicals and minerals; 3) metals and machinery; 4) electronics and medical equipment; 5) production of vehicles and other categories not included elsewhere.

· Region: dummy dreg = 0 if a firm is based in Moscow or Moscow region and 1 otherwise.

· Year when the last loan was granted.

· Ruble costs is the share of Ruble-denominated costs in total costs.

· Foreign ownership = 1 if percent of foreign ownership is higher than zero and 0 otherwise.

· Use = 1 for exporters: if the loan's purpose is developing business in existing export markets, expanding the range of export products in existing export markets or expanding the number of export destinations and 0 - if developing business in home market.

Use = 1 for non-exporters: if the loan's purpose is trying to enter a foreign market and 0 - if developing business in home market.

· Currency is the currency in which the loan was given. Dummy cy = 1 if currency of the loan is foreign and 0 otherwise.

· Bank: bank = 1 if the bank that gave the loan has government participation and 0 otherwise.

· Export sales is the ratio of export sales to total sales (defined only for exporters).

· Extent (of loan granted) is the ratio of the notional of the last loan given to the entity to the value that the entity had initially requested.

· Interest rate is the interest rate the firm reported it pays on the loan.

· Collateral to loan is the ratio of collateral pledged for the loan to the loan's notional.

· Maturity is the number of months the loan was granted for.

· Financial constraints estimate is the extent to which financial constraints are a hindrance to the overall business activity of the firm as estimated by the firms themselves, ranging from 1 to 4. 1 means that financial constraints are an insignificant hindrance and 4 means they are a very severe one.

In most of the models I use log-transformed versions of the variables above.

Table 1: Summary statistics

Variable	Mean	St.D.	Min	Max	# of obs.
Exporter	0,7	0,4	0	1	1536
Employment	340,6	944	1	16539	1493
Turnover	2642	20649,8	0	531644,1	1493
Age	27,7	27,8	2	279	1525
Ruble costs	87,8	21,4	0	100	1473
Foreign ownership	0,1	0,3	0	1	1519
Use	0,2	0,4	0,0	1,0	564
Currency	1,1	0,4	1,0	4,0	580
Bank	0,4	0,5	0,0	1,0	571
Extent	96	12,7	5	100	511
Interest rate	14,3	4,1	1	36	366
Collateral to loan	101,4	69,9	0,5	550	291
Maturity	27,8	18,1	1	120	488
Financial constraints estimate	1,4	1,2	0	4	1488

Below is the table containing simple means of the parameters of interest: for exporters average extent of loan granted is higher, average interest rate is lower, collateral required higher, maturity is shorter and severity is lower than that of non-exporters.

Table 2: Summary statistics of parameters of interest, split by export status

	Exporters	Non-exporters
Avg Extent	96,4	95,0
Avg Interest rate	13,9	15,4
Avg Collateral	101,5	101,2
Avg Maturity	27,0	29,8
Avg Severity	1,3	1,5

Finally, it is also worth looking at the distribution of the business dimension for which the last loan was taken in a bit more detail. Both exporters and non-exporters tend to take loans mostly for developing domestic presence (75% of exporters and 92% of non-exporters). Among the exporters who stated developing foreign activity as their goal, 60% focus on developing business in existing export markets, 31% - on expanding the range of export products in existing export markets and only 9% said they needed the loan to expand the number of export destinations. In terms of the type of activity the loan was used for, the most popular was working capital (67% of exporters and 76% of non-exporters said they used the part of the borrowed funds for this activity). Capex is slightly less popular scoring 56% and 50% with exporters and non-exporters respectively. A fifth of exporters and a quarter of Non-exporters used the funds to repay existing debts, while only 5.2% and 2.7% directed the money to M&A activity.

Research Methodology and Results

I split the rest of my paper into two parts: first, modelling the average treatment effect (ATE) of export status on credit constraint parameters and, second, analysing data only on exporters to ascertain what are the determinants of their credit constraints and whether loan use is one of them. The estimation strategy for the first part is propensity-score matching. The estimation strategy for the second part is propensity-score matching and ordinary least squares (OLS).

Non-exporters vs. Exporters:

In order to estimate the average treatment effect of export status on the parameters of interest, I use treatment effects with propensity-score matching method (teffects psmatch command in Stata). A logistic model is used to calculate each observation's probability of being an exporter, i.e. the propensity score, using a set of covariates. Then the method estimates the level of similarity between observations for exporters and non-exporters based on these estimated treatment probabilities. For each non-exporter the method finds some similar exporters. So, we have a list of non-exporters and to each one of them we have one or several matched exporters. Then for each of the non-exporters the method calculates the average of the parameter of interest for the matched exporters. ATE within a pair is the difference between the average value of the parameter for the matched exporters and the value of the parameter for the non-exporter. And overall ATE is calculated as the average of the differences for each pair. The motivation for using propensity score weighting is that it gives a robustness advantage for estimating ATE over other methods that don't use weighting (Imbens & Wooldridge, 2009).

There are a couple of important assumptions that we are making here to identify treatment effects. First is the unconfoundedness assumption which requires that “conditional on observed covariates there are no unobserved factors that are associated both with the assignment and with the potential outcomes”. It implies that we have a large enough number of predictors for the treatment indicator, such that “adjusting for differences in the covariates leads to valid estimates of causal effects” (Imbens & Wooldridge, 2009).

The second is the overlap assumption which requires that both control and treated units exist for all plausible covariates' values. In some specifications of my models this assumption fails, so results can't be computed. But in most specifications it holds, which can be illustrated by the overlap plots The graph displays the estimated density of the predicted probabilities that a non-exporter is a non-exporter and the estimated density of the predicted probabilities that an exporter is a non-exporter.. Neither has too much probability mass near 0 or 1 and the densities have most of their masses in the area where they overlap each other. Therefore, there is no evidence of violation of the overlap assumption http://www.stata.com/manuals13/teteffectsoverlap.pdf.

Graph1:Typical overlap plot for most of my model specifications

I have two major set-ups for my models: first is where I only look at the loans denominated in rubles and taken out for expanding domestic activity (as opposed to export activity); second is where currency and use variables are used as dummy controls. And for each of these two set-ups I have 3 sets of model specifications: the first set is the most basic and then the other two add some specific details. The first set estimates the average treatment effect of exporting status on each of the 4 parameters of interest: extent, interest rate, collateral and maturity. I also provide results for a model where the independent variable is severity - I do this separately because severity is not a parameter of credit constraints in a strict sense since it is not an observable loan metric. Nevertheless, I believe it is worth estimating the ATE of export status on this parameter using propensity score matching method as well, because it provides an insight into how differently exporters and non-exporters perceive their own degree of financial constraint. For controls I take: industry and region dummies, employment and turnover to control for size, foreign ownership, total experience and ruble costs.

The second set of models adds a control for bank type to the list of controls. For this, I use the bank type dummy. The third set is models that are estimated for each of the two types of the banks separately.

Let's look at the results of the three sets of models from the first set-up. Below are results from my first set of models (model I) that does not include controls for the type of bank.

Table 3: Model I

	(1)	(2)	(3)	(4)
Variables	Extent	Interest rate	Collateral	Maturity
r1vs0.Exporter	0.0580	-0.100***	0.0417	-0.364**
	(0.0616)	(0.0362)	(0.0957)	(0.156)
Observations	355	261	202	333

The model is estimated using propensity score matching method

This table shows that coefficients on interest rate and maturity are significant. The coefficient on maturity shows that exporters receive loans with maturities shorter than those of non-exporters by on average 31% The 31% is calculated as: . Similar calculations apply for the rest of the results as well.. Since the average loan maturity for non-exporters in the subsample used for this model is 23.2, it means that the 31% difference translates into 7.2 months difference. The coefficient on interest rate shows that exporters on average have an interest rate lower than non-exporters by 10%, which translates into 1.5 percentage points difference. It could, of course, be argued that it is only natural that shorter maturities go hand in hand with lower interest rates. While this is true, my result holds even after controlling for maturity, as I discuss in a bit more detail a few paragraphs later. The other coefficients in this specification are insignificant.

In the second set of models (model II), where we start to have a control for bank type, the only significant coefficient is the one on interest rate and it shows a result similar to what we received above in the first set (here the interest rate of exporters is on average lower than the interest rate of non-exporters by 14%, which is 2.1 percentage points). The coefficient on maturity is also negative as in the first set, but here it is not significant.

Table 4: Model II

	(5)	(6)	(7)	(8)
Variables	Extent	Interest rate	Collateral	Maturity
r1vs0.Exporter	0.0604	-0.147***	0.128	-0.227
	(0.0420)	(0.0279)	(0.138)	(0.180)
Observations	353	260	202	330

The model is estimated using propensity score matching method

Finally, the last set of models (model III) produces the following results:

Table 5: Model III

	(9)	(10)	(11)	(12)
Variables	Extent_b1	Extent_b0	Interest rate_b0	Maturity_b0
r1vs0.Exporter	0.0675	0.0233	-0.187***	-0.457***
	(0.0783)	(0.0221)	(0.0297)	(0.139)
Observations	146	207	168	200



The model is estimated using propensity score matching method

Subscript _b1 here means the model was run for banks with government participation, _b0 - for banks without government participation. Some model specifications are omitted here (for example for collateral for both banks types and for interest rate and maturity for government banks). It means that the results could not be estimated because the distributions of controls variables are not suitable for the estimation methodology used. More precisely, there is a violation of the treatment overlap assumption when running the treatment effects propensity score-matching estimation. The same holds for some of the tables further on in this paper.

For model III here only results for private bank group are significant. Interest rate results are in line with the results from the first two sets of models: i.e. interest rates for exporters taking loans from private banks are on average less than for non-exporters by 17%, i.e. by 2.6 percentage points. Results for months are consistent with our earlier findings as well: loan maturities are shorter for exporters by 37%, i.e. by c. 8.7 months.

In order to check that these results do not simply stem from the fact that shorter loans usually carry a lower interest rate, I run an additional set of models - same as above but now including maturity as a control. This method has a disadvantage because banks that give loans decide on the maturity of the loan at the same time as they decide on the interest rate, collateral and extent. But the fact that the results are virtually the same as without controlling for maturity (with coefficients on interest rate only marginally lower) is useful since it proves that the above result on lower interest rates for exporters is not only due to the fact that these loans are for shorter maturities (see Tables A8-A10).

The fact that these results exhibit significant differences in some parameters of credit constraints between exporters and pure domestic firms is in line with the results of most of the existing papers that study the effect of export status on credit constraints. In particular, the result that exporters have lower interest rates than non-exporters is consistent with the results provided in Greenaway, Guariglia and Kneller (2007), Campa and Shaver (2002), Tornell and Westermann (2003), Ganesh-Kumar, Sen and Vaidya (2001). As mentioned above, one of the explanations of lower credit constraints for exporters provided in these works is that export status can act as a signal that the firm is productive enough to cover the sunk costs of exporting and that exporters are also less affected by the idiosyncratic risks of the domestic market. At the same time, my result is not consistent with the result of Feenstra, Li & Yu (2011) who predict higher credit constraints for exporters. It may not be entirely applicable though to compare the results obtained in my work with those obtained in the other papers. The reason is that the four measures that I use as proxies for credit constraints are different from the measures used in the other papers, and so is the estimation methodology.

Finally, I look at the results for severity for the three model specifications (models I,II and III). Only in model specification III does the coefficient on severity become significant. It shows that severity of financial constraints is higher for exporters. This may be due to the fact that exporters compare themselves to their peer exporters from Europe who have lower rates, and, thus, feel that they are more constrained, while non-exporters do not have such a benchmark. These results hold for different specifications of severity: when severity is a discrete variable from 0 to 4, as well as when severity is transformed to a dummy, with 0 corresponding to 0, 1 and 2 (meaning no, insignificant or average reported level of severity) and 1 corresponding to 3 and 4 (meaning high or very high reported level of severity).

Table 6: Results on severity for models I, II and III

(13)	(14)	(15)	(16)
MI: Severity	MII: Severity	MIII: Severity_b1	MIII: Severity_b0
0.163	0.149	-0.349	0.502***
(0.203)	(0.168)	(0.253)	(0.182)
380	376	149	227



The second set-up expands the sample to include all firms (i.e. not only those firms that took ruble-denominated loans for domestic purposes) and, as explained above, uses loan currency and loan use as dummy controls. It turns out that for estimating the ATE of export status on credit constraints it produces virtually the same results, except for marginally changing significance of some of the coefficients, so I will not describe it at length but just provide the result in Appendix in the Tables A1-A3. The main conclusion from this is that export status leads to lower interest rates and shorter maturities not only for ruble-denominated domestic loans but also those denominated in foreign currencies and taken for export purposes.

Exporters:

In the second part of my paper I focus only on exporters and study how different firm metrics influence the four parameters of credit constraints and the severity parameter.

First, I use OLS method. I have three model specifications as above: a model without control for bank type, a model with control for bank type and a model looking at the two bank groups separately. I find that age, turnover, foreign ownership, industry and region are significant determinants of some of the credit constraint parameters.

For model specification I (where only loans denominated in rubles and taken out for domestic purposes are taken into account) I arrived at significant results for the following:

· Age: older companies have significantly lower interest rates, probably because banks consider older companies to be safer borrowers. The coefficient on collateral is also negative but not significant.

· Size: the coefficient on turnover is negative in the interest rate regression. This means that exporters with higher turnover receive loans with lower interest rates.

These results on age and size are in line with existing corporate finance literature. For instance, older firms have been found to have lower borrowing costs in such papers as Berger & Udell (1995) and Petersen & Rajan (1993). Also, Sharpe (1990) finds that the longer a relation of the borrower with a creditor, the lower will the collateral requirements be. Regarding the size, Beck, Demirguc-Kunt & Maksimovic (2005) find that small firms experience greater financial obstacles than large firms.

· Foreign ownership: companies with foreign ownership have significantly lower rates and longer maturities.

· Industry: use of industry fixed effects shows that companies from wood, chemicals and minerals, as well as electronics and medical equipment, industry groups have significantly lower interest rates as compared to those from metals and machinery industry (the omitted group)

· Region: companies from regions other than Moscow and Moscow region have significantly lower interest rates but at the same time have to post significantly larger collateral as percent of loan and are given funds for significantly shorter periods of time.

· Ruble costs' prevalence in the total cost structure is not a significant determinant for any of the credit parameters we look at.

Table 7: Model I

	(1)	(2)	(3)	(4)
Variables	Extent	Interest rate	Collateral	Maturity
Age	-0.0216	-0.0968**	-0.0142	0.0520
	(0.0210)	(0.0462)	(0.122)	(0.0916)
Export Sales	0.00105	0.00996	-0.112	0.00285
	(0.0152)	(0.0303)	(0.0785)	(0.0641)
Employment	0.0106	0.0228	-0.0113	-0.0895
	(0.0133)	(0.0294)	(0.0890)	(0.0604)
Turnover	-0.000614	-0.0360*	-0.0736	0.00885
	(0.00798)	(0.0183)	(0.0483)	(0.0350)
Ruble costs	0.0642	-0.0770	0.461	0.0351
	(0.0717)	(0.162)	(0.440)	(0.305)
Foreign ownership	0.0214	-0.228*	0.455	0.439*
	(0.0544)	(0.129)	(0.343)	(0.247)
bank
Constant	4.283***	3.674***	2.034	3.448**
	(0.343)	(0.769)	(2.064)	(1.455)
Industry f.e.	yes	yes	yes	yes
Region f.e.	yes	yes	yes	yes
Obs	199	148	116	191
Adj R^2	-0.026	0.133	0.184	0.032

The model is estimated using OLS method

In model II where we add a dummy for bank type to the controls all of the above results stay the same and coefficient on the bank dummy is not significant.

Table 8: Model II

	(5)	(6)	(7)	(8)
Variables	Extent	Interest rate	Collateral	Maturity
Age	-0.0229	-0.0797*	-0.0232	0.0600
	(0.0218)	(0.0472)	(0.123)	(0.0939)
Export Sales	0.00138	0.00732	-0.109	-0.000948
	(0.0154)	(0.0301)	(0.0786)	(0.0648)
Employment	0.0111	0.0151	-0.00866	-0.0919
	(0.0135)	(0.0296)	(0.0891)	(0.0610)
Turnover	-0.000954	-0.0334*	-0.0761	0.00881
	(0.00825)	(0.0182)	(0.0484)	(0.0361)
Ruble costs	0.0645	-0.0667	0.457	0.00263
	(0.0723)	(0.161)	(0.440)	(0.307)
Foreign ownership	0.0193	-0.208	0.468	0.449*
	(0.0552)	(0.129)	(0.343)	(0.249)
bank	0.0132	-0.0900	0.147	-0.102
	(0.0289)	(0.0606)	(0.162)	(0.125)
Constant	4.284***	3.603***	2.101	3.582**
	(0.346)	(0.764)	(2.067)	(1.462)
Industry f.e.	yes	yes	yes	yes
Region f.e.	yes	yes	yes	yes
Obs	197	147	116	188
Adj R^2	-0.031	0.142	0.182	0.026

The model is estimated using OLS method

Specification III of the regression (with groupings by bank type) doesn't add any significant insights but some observations worth noting are:

· In this specification, direct exports as percent of total sales become important for the private banks groups: the higher this percentage, the higher the interest rates that exporters are charged. This may be due to the fact that more export activity is associated with more risks and these export-related risks affect loans for domestic purposes.

· Industry, region effects and turnover are also significant for interest rate (in line with the first two model specifications).

· Region effect remains to be an important factor for collateral to loan and months outcomes but only for private banks.

Table 9: Model III: For private banks type

	(9)	(10)	(11)	(12)
Variables	Extent	Interest rate	Collateral	Maturity
Age	-0.0159	-0.0576	-0.107	0.0679
	(0.0229)	(0.0442)	(0.160)	(0.123)
Export Sales	0.00566	0.0847***	-0.0279	-0.000329
	(0.0180)	(0.0319)	(0.129)	(0.0943)
Employment	0.00568	0.0268	0.0641	-0.131
	(0.0146)	(0.0288)	(0.125)	(0.0853)
Turnover	0.0102	-0.0316*	-0.113*	0.00846
	(0.00845)	(0.0168)	(0.0623)	(0.0466)
Ruble costs	0.0983	0.153	-0.111	-0.107
	(0.0904)	(0.166)	(0.697)	(0.479)
Foreign ownership	-0.00241	-0.225	0.483	0.230
	(0.0645)	(0.138)	(0.459)	(0.363)
Constant	4.112***	2.289***	4.404	4.268*
	(0.432)	(0.787)	(3.282)	(2.259)
Industry f.e.	yes	yes	yes	yes
Region f.e.	yes	yes	yes	yes
Obs	119	96	76	119
Adj R^2	0.004	0.259	0.216	0.018

The model is estimated using OLS method

Table 10: Model III: For government banks type

	(13)	(14)	(15)	(16)
Variables	Extent	Interest rate	Collateral	Maturity
Age	-0.0152	-0.111	-0.0109	-0.0179
	(0.0474)	(0.115)	(0.205)	(0.165)
Export Sales	-0.00712	-0.110	-0.102	-0.0324
	(0.0289)	(0.0687)	(0.122)	(0.100)
Employment	0.00508	-0.0224	-0.0974	-0.000748
	(0.0276)	(0.0655)	(0.132)	(0.0979)
Turnover	-0.0177	-0.0242	0.0148	0.0472
	(0.0184)	(0.0446)	(0.0837)	(0.0668)
Ruble costs	0.0908	-0.333	1.090*	0.0510
	(0.125)	(0.320)	(0.577)	(0.435)
Foreign ownership	0.0600	-0.0874	0.108	0.648*
	(0.100)	(0.245)	(0.553)	(0.365)
Constant	4.127***	4.885***	0.248	2.940
	(0.597)	(1.541)	(2.731)	(2.067)
Industry f.e.	yes	yes	yes	yes
Region f.e.	yes	yes	yes	yes
Obs	78	51	40	69
Adj R^2	-0.082	0.042	-0.070	-0.031

The model is estimated using OLS method

As in models in the first part (non-exporters vs. exporters), I also provide results for severity separately. I have received significant result on industry and region. For industry: use of industry fixed effects shows that companies from wood, chemicals and minerals, as well as electronics and medical equipment, industry groups report having significantly lower severity of financial constraints. For region: companies from regions other than Moscow and Moscow region consider themselves under significantly more financial constraints versus the companies from Moscow and Moscow region. For model specification III, coefficients on age and foreign ownership are also significant but have different signs for private and government banks.

Table 11: Regressions MI, MII, MIII with Severity as dependent variable