Liquidity Constraints and Firm s Export Activity

Transcription

1 Liquidity Constraints and Firm s Export Activity Emanuele Forlani Université Catholique de Louvain - CORE Abstract An increasing quota of papers in international trade are dealing with the relation between rm s liquidity constraints and export decision. If there exist xed costs to begin exporting activity then liquidity constraints become relevant for rm choices. To test how much nancial liquidity matters for exporting I use a productivity index (TFP) cleaned out from capital demand heterogeneity. In order to solve the problem of unobserved input (capital) price bias in production function s estimation I introduce credit constraints in the Olley and Pakes method (1996). Then employing liquidity indices used in business economics I verify that not only TFP matters for rm s export decision, but there exists signi cant di erences in terms of liquidity among exporting and domestic rms. Keywords: Productivity, Credit constraints, Heterogenous rms, Trade JEL Classi cations: C14, D24, F10, F12, F13, F19, M40

2 1. INTRODUCTION 2 1 Introduction In this paper I analyze the issue of credit constraints and how them a ect rm s e ciency and exporting decision. In particular I try to de ne if rm s nancial health is linked with the ability to export (once I controlled for e ciency). The paper provides the evidence that rm s liquidity constraints are crucial both for the estimation of production function (and productivity) as well as for rm s exporting choice. In the rst part I will discuss the role of credit constraints for Total Factor Productivity (TFP) and I suggest an estimation method to solve the problem of unobserved price bias for capital input. In the second part, once I controlled for input price bias in TFP, I test empirically if there are any di erences in term of nancial resources for a sample of Italian rms and I verify if " nancial variables" play a relevant role in exporting activity. Since the begin it is important to clarify the terminology of credit constraints: in this paper I de ne credit or liquidity constraints as the absence or scarcity of internal resources which are necessary for rms in order to nance their own investments (machinery, new plants, R&D, exports...). There is a large literature that analyzes the e ects of capital imperfections and rms s activities (Hubbard, 1998; Bond-Van Reenen, 2005). Why credit constraints matter to rm s decisions and empirical trade analysis? Because if a rm is constrained in credit, it incurs in di culties to cover xed costs for investments: scarce internal resources or no access to external nancing means limited options for investments. In addition it is reasonable to assume that credit constrains are a second source of heterogeneity across rms because rms may di er in term of liquidity. If we consider capital s demand it will vary according to sector technology and rm s resources availability; more precisely, the price of capital varies across rms because of internal ( rm nancial health) and external (sector technology) factors. Some sectors are capital intensive hence rms have to invest a huge amount of resources such that they can not be covered with internal resources( steel, machinery, chemicals...); instead other rms do not need credit due to high internal liquidity and low investment requirements. Thus there exists heterogeneity not only in term of productivity but also in terms of nancial needs; rms pay a di erent unit capital price. The heterogeneity in capital price a ects the estimation of productivity as well as any kind of choices of quantity or quality done by entrepreneurs 1. 1 For example a constrained rm can not face a positive demand shock because it is not able to collect 2

3 1. INTRODUCTION 3 Given that nancial constraints narrow investments it is quite straight supposing the existence of a relation with exporting activity due to the existence of sunk costs associated with trade (Melitz, 2003). A plant may be e cient enough to export, but if capital market is not very well developed, it is possible that a rm is not able to exporting because it can not gather enough liquidity to cover the initial xed cost (Greenway et al., 2007). With sunk costs only rms that are highly productive are able to export because they generate su cient cash ows from their sales. Hence the self-selection process is reinforced by liquidity constraints and the development of capital market plays an important role. Credit constraints interact with rms s heterogeneity and reinforce the selection of the most productive rms because these are able to raise higher revenues and consequently they can o er to creditors greater returns or more safe collateral securities. The paper is related to the literature on nancial institutions and this analysis try to contribute in the direction of trade at rm level. There is a small but growing literature on this topic, in particular there are several empirical works that show nancially developed countries 2 export relatively more from nancial vulnerable sectors 3 (Beck, 2002, 2003; Manova, 2005). Di erently from the previous papers which concentrate on product composition of exports at sector level (Manova, 2006), I focus my analysis on micro level data for a survey of Italian rms and I use information about rms s nancial status for measuring rm s liquidity constraints. This allows me to study how credit constraints interact with rm s heterogeneity and how they a ect rm s export status. There three papers that looks at the relationship between international trade and liquidity constraints at rm level, Campa and Shaver (2002), Guariglia and Mateut (2005) and Greenway et al.(2007). Campa and Shaver (2002) use a panel of Spanish manufacturing rms to test if there exists any link between rm s nancial status and its exporting activity. They nd that cash ows 4 are more stable for exporter rather than for non-exporters, probably because exporters earn pro ts in many markets covering themselves from business cycle. Moreover they nd that credit constraints matter with export status but not with rm s intensive margin of trade; it is more likely that the total amount of quantity exported is more related with the productivity enough resources to expand the production. 2 Countries in which is easy the credit access and there are few rigidities in the capital market. 3 Sectors with low internal resources and high demand of capital. 4 Credit constraints are measured as variation of investment in tangible xed assets caused by nancial variables. 3

4 1. INTRODUCTION 4 level rather than credit constraints. Greenway et al.(2007) nds that exporters have a better nancial situation respect to domestic rms; in addition they nd that rms before starting the export s activity have a low liquidity and high leverage. Di erently from Campa and Shaver I will not use cash ows as measure of liquidity constraints, but similarly to Greenway et al.(2007) I use indices from asset and liabilities balance sheet for describing rm s nancial health. The indices (or ratios) are commonly employed in business economics for rm s rating and they are used by banks for evaluating the riskiness of an investment in a speci c economic activity 5. These ratio are widely used since the introduction of "Basel" and "Basel II" agreements. According to the rules, nancial institutions as banks have to follow a standardize procedure in order to manage risk and they consider for part of their risk s analysis indices derived from asset and liabilities value 6. There exists a number of theoretical works in the stream of nancial development which deal with liquidity constraints as a source of comparative advantage (Matsuyama, 2004; Becker Greenberg, 2005). However these models depend on Ricardian comparative advantage framework hence the basic prediction is that either all or no rms export in a given sector. A seminal paper in this eld is by Rajan and Zingales (1998): they show as - nancial development is a relevant determinant for economic growth because it reduces the cost of external nancing for rms. Chaney (2005) introduces liquidity constrains in a heterogenous rm framework where the most productive plants are the less credit constrained. Instead Manova (2006) shows empirically as credit constraints are important determinants both for zeros in bilateral trade ows and variation in the number of products exported and countries reached. Moreover she takes into account also the di erent requirements of outside nancing across sectors due to production technologies. I include heterogeneity in nancing requirements across sectors too, when I estimate production function with credit constraints. I employ for my analysis a dataset that I construct merging two surveys at rm level and a detailed balance sheet dataset provided by Capitalia. The two surveys contain information about a sample of Italian small and medium enterprises (SMEs) while the balance sheet dataset includes detailed information on statement of assets and liabilities. 5 Brealey R., Myers S. (1999): Principles of Corporate Finance, ed. MacGraw-Hill 6 or or ndustria.it 4

5 2. DATA DESCRIPTION: CAPITALIA SURVEYS 5 My strategy can be divided in two parts. First, I estimate production function solving for unobserved capital price; in order to solve it I take into account rm s credit constraints in the estimation and I modify the OP, relaxing some assumptions. The estimated TFP varies more across sectors as a consequence of di erent industry technology and risk level 7. In the second part I link the export status with TFP and several indices that evaluate rm s nancial health. I perform a cross section analysis to verify if exporters are more or less credit constrained and if liquidity matter in the self-selection process. I nd that on the average exporters are not only more productive and e cient but they operate with better nancial situation. Moreover the credit constraints partially explain the self-selection process in the export market; once I controlled for productivity, rms with an higher level of liquid asset available are more likely to export. It is interesting notice as rms which use many banks are more likely to be exporters. Firstly I describe data then I introduce the nancial situation of Italian rms. In the fourth part I provide the estimation techniques while in the last section I illustrate the empirical results; then I conclude. 2 Data description: Capitalia surveys I employ two surveys in the analysis and both of them provide qualitative and quantitative information about a sample of Italian manufacturing SMEs 8. The rst survey (seventh Capitalia survey wave) analyses a period from 1995 to 1997, while the second (eighth wave) covers from 1998 to From the surveys I get information about rm s export status in the two observation periods: more precisely the surveys ask if rm has exported during the period considered ( and ). In the eight survey are provided more information as the number of banks with whom the rm is related. The data from surveys are not time variant but they refer just for a period of three year: for this reason the "selfselection test" is just a cross section because we knows of export status only for three year 7 I proceed in this way in order to disentangle the e ect of credit constraints from the e ciency e ect in rm s self-selection process: with the standard OP, the e ciency index includes unobserved price bias. 8 Private owned rms. 5

6 2. DATA DESCRIPTION: CAPITALIA SURVEYS 6 period 9 ( and ). In addition it is possible to merge the two surveys and the rms, even if some observation are missed. To which concern balance sheet, the dataset contains information about all rms included in the eighth survey ( ) and it collects budget items from 1991 to 2000 in thousands of Euros. It is possible to recover budget information also for a rm in the seventh survey if it is present in the eighth. The balance sheet dataset provides a detailed statement of assets and liabilities as well as data on input values, turnover and average number of employees. The indices mentioned above are constructed using balance sheet data on assets and liabilities; it is possible to obtain information on short and long term debts, credits, assets, equity, amount of leftover stock, pro ts. Firms are classi ed according to four digit ATECO 2002 industrial classi cation. I cluster them along the same industrial classi cation at two digit ATECO 2002; in Appendix A is provided a list with the sectors included and a list of descriptive statistics. We can notice that on the average the rms included in the surveys are of medium or small size in term of employees (less than 250 employees). I merge the three datasets in one, in order to exploit information about export s behavior at two di erent period in time (1997 and 2000) and to relate them with the liquidity constraints ratios. The dataset constructed is an unbalanced panel and it provides qualitative and quantitative information about 7882 Italian rms 10 over ten years ( ). Finally I add in the dataset also de ators data for turnovers, wages, capital and material provided by Eurostat (Ameco). For the estimation of production function the dependent variable is the de ated value of operating revenues while as input the de ated value of material consumption and tangible xed assets (capital) plus the average number of employees. The investment (INV t ) information are recovered with the perpetual inventory method according to a law motion as INV t = K t+1 (1 ) K t 1 using as capital (K t ) the total amount of assets (tangible plus intangible assets) More precisely the survey requested if rm was involved in export activity during the three years of reference. 10 Included in both surveys. 11 = 8% to 11% 6

7 3. CREDIT CONSTRAINTS FOR ITALIAN FIRMS 7 3 Credit constraints for Italian rms For technological or strategic reasons rms could need external resources to sustain their projects but not all rms are able to raise nancing in the same way. Fund raising is a relevant problem in particular for small and medium size rms (SMEs) which for several reasons have no access to all potential sources of credit (non big enough for equity market, not enough know-how to use complex nancial instruments). In particular SMEs rely in large part on debts with banks rather than stock markets or internal resources generated by cash ows or shareholders. It is a relevant problem in particular for Italian SMEs which usually do not gather enough liquidity and su er under-capitalization: it means that the internal nancing (equity and cash ows) cover less than one third of total capital stock 12 (Table 1). In addition liquidity constraints may exists regardless rm s characteristics and they can depend on the ability of management to raise founds or the private guarantee o ered by the owner. Therefore the cost and the amount of a loan depends on several factors, some observable as rm e ciency other unobservable as raising found ability. As introduced above, I consider in the analysis a sample of Italian SMEs, which are the bulk of Italian exporting activity (ISTAT, 2005). Usually these rms are "undercapitalized" in the sense that they have not enough internal resources to sustain their economic activities (i.e. investments). The reasons can be several among which the scarce capacity to generate liquidity or the low level of equity, in this case rms have to raise funds somewhere in order to nance themselves. The Basel agreements (I and II), proposed by International Bank of Settlements, introduce for banks standard nancial requirements in relation to the di erent kind of risks assumed by the same bank. The agreement de nes risk management criteria for whom, each bank has to rank its credits (loans) according to risk; in particular "credit risk" principles are source of some concern for Italian SMEs about the accessibility to credit 13 ; the concerns derive from the fact that SMEs are not able so often to generate enough liquidity to satisfy Basel II principles or knowledge to deal with the new conditions in credit market. In addition liquidity constraints of SMEs are an important issue because 12 Brealy-Myers (1999). 13 Since the introduction of Basel II many meetings were organized by Con ndustria (Italian association of rm s owners) or Bank of Italy in order to explain the credit risk and new features of Basel agrrements to SMEs. 7

8 3. CREDIT CONSTRAINTS FOR ITALIAN FIRMS 8 part of Italian SMEs have a closed ownership structure, families own economic activities and they are not in favor to any kind of nancing brought from outside that reduces the ownership control. I present in Table 1 some ratios that are widely used for assessing rm s nancial independency and structural stability as well as for evaluating risk; data comes from Capitalia surveys and balance sheet informations from 1991 to One indicator used 14 is the so called Financial Independency Index (FII) and it is a ratio between internal resources (equity plus cash ows) over the total amount of capital invested (total assets). The optimal ratio should be 33% meaning that at least one third of rm s assets have to be nanced (covered) by internal resources. In the rst line of Table 1 I report the average FII for rms in the sample; we can see as the index is on average below the "suggested" threshold of 33% for rms in dataset. Table 1: Liquidity Constraints Ratios Year FII SR CFI D a t a S o u r c e : C a p it a lia Below FII is presented a second index that measures rm s ability to cover debts and long-time loans; it is a ratio of nancial solvency (SR) and it assesses the rm s ability to cover capital assets and leftover stocks with durable nancing sources as equity and long term debt. The optimal ratio should be one or greater in order to show a stable nancial situation: also in this case the average ratio suggests us a critical nancial condition for Italian rms in the sample. Hence the two indices give the intuition that SMEs in the sample are partially credit constrained. Lower is the value of these indices and lower is the probability for a rm to obtain an external nancing or if it is obtained it will be paid for higher price 15 because it is more risky invest. A third interesting index, mentioned in Table 3.1, is the Cash Flow Index (CFI) (Manova, 2006). It is not commonly employed by nancial institutions to deal with risk, but by economists; it is de ned as a ratio between rm s cash ows (internal nancing) and the amount of new loans negotiated (external nancing). It 14 The list of ratios used in the analysis is provided in Appendix B. 15 As more collaterals required. 8

9 3. CREDIT CONSTRAINTS FOR ITALIAN FIRMS 9 results that rms in the sample rely increasingly on debt rather than on their ability to generate liquidity/resources 16. There is evidence that rms in the sample su er of liquidity constraints and rely on external nancing rather than internal. Moreover it seems reasonable assume that there exists heterogeneity in term of nancial health (Chaney, 2005; Greenway et al., 2007) between rms and across sectors. In Table 2 I cluster rms in two di erent groups (exporters and non exporters) and then I show the same ratios, with the addition of others. In the dataset it is possible to observe the export status for year 1997 and year 2000 in two di erent surveys waves of Capitalia 17 in which is possible to track rms ( rms in survey 1997 are a partially included in survey 2000). Table 2: Indices analysis Expo97 Dom97 Exp00 Dom00 Expo New00 Dom 1 Current Acid Quick FII SR CFI Cash Flows Banks D om :D om estic. E xp o :E xp orters. N ew 00: E xp orter in the eight survey and dom estic in the seventh. D ata source:c apitalia. In each row is mentioned a rating index that assesses rm liquidity situation and nancial independency; every cell reports the average across time for each ratio 18. The rst three indices (current, quick and acid) are three liquidity ratios 19 that evaluate rm s ability to repay debts in the short run with di erent kinds of resources (cash, credits, leftover stock of nal products). We can notice as exporters show higher ratios than non exporters in both period of observation. In addition when I track rms across two surveys the previous 16 In the Appendix C I provide tables with indices in Table 3.1 (FII, SR, CFI) for each sector. 17 More precisely the surveys ask if rm exported during the period considered ( and ). 18 In the rst two columns the averages refer for a period from 1991 to In the third and fourth columns the averages refer for a period from 1998 to In the last three columns the averages refers to all periods available in the dataset. Remember that rms 1997 are a sub-sample of rms in De nition of indeces in Appendix B 9

10 4. CREDIT CONSTRAINTS IN PRODUCTION FUNCTION ESTIMATION 10 intuition is con rmed: on average the exporters (Exp) in both surveys have less problems of liquidity rather than domestic rms (Dom); it is in line with the results obtained by Greenway et al.(2007). Instead the position of new exporters in term of credit availability is in the between; even if this is a simple descriptive statistics, di erently from Greenway (2007) I do not nd that new exporters show high leverage and low liquidity 20. It is noteworthy to observe that also exporter from eight survey (Exp00) are supported by more banks rather domestic ones (row 8). The other indices (FII, SR, CFI) in Table 2 suggest us that there are no substantial di erences in term of nancial independency amongst exporters and domestic rms (probably exporters show higher leverage, row 4 and 5). Finally exporters generate more cash ows than domestic rms (row 7) as in Campa-Shaver (2001). I will use in the next sections some of these indicators to evaluate the unobserved capital constraints in production function and to test the relevance of liquidity for the exporting activity. 4 Credit constraints in production function estimation Now I turn to production function estimation and I deal with bias generated by unobserved input price. Firm has to face many choices in order to produce, as output quality, quantity, prices, level of investment, or bargain loan contracts and so on. In each of these tasks rm could nd di erent situations that a ect production and consequently also productivity. One fundamental variable is rm s investment demand and it depends on plant s e - ciency, capital stock and resources of whom rms are endowed. The availability of internal resources is fundamental to self nancing investments for new projects: but not all rms are able to cover expenses. The data showed in Table 1 depict a situation for which rms may need external nancing. Hence the price paid by rms for capital in credit market is a relevant variable, as well as the amount of resources borrowed. Usually with the estimation of production function the price of capital is assumed constant across rms but it generates biased coe cients and TFP values (De Loecker, 2007b). It seems more reasonable assumes that each rm pays a di erent price for capital depending on rm s nancial health. Why should rms pay di erent price for the same amount of capital? Why some rms 20 I am able to track just 70 rms which pass from domestic to exporter using both surveys. 10

11 4. CREDIT CONSTRAINTS IN PRODUCTION FUNCTION ESTIMATION 11 can not obtain all liquidity that they need once provided a certain e ciency level? In a real world nancial institutions as banks can be reluctant to lend money in some cases, or lend less than the requested amount. A bank could demand di erent conditions for the same loan between rms which are di erent in term of risk and collateral securities: capital price may depend not only on capital demand but on other factors as moral hazard, bargaining ability or unobservable personal guarantee 21 (especially for private owned SMEs). Besides Wasmer and Weil (2004) show as frictions in capital market a ect rm s hiring and production decision (as well as level of pro ts): the rigidities in capital market could generate as outcome di erent capital price for two similar rms. 4.1 Identi cation of production function parameters In this section I propose a solution to solve the bias due to unosbervable capital price; I modify the model of Olley and Pakes (1996) following the approach suggested by Ackerberg et al.(2005). The solution for input price bias entails the modi cation of one fundamental OP assumption, namely the assumption on the scalar nature of unobservable term; I introduce a second unobservable term that outlines the heterogeneity of liquidity constraints among rms. Let s start with a classic production function, where rm i at time t produces according to a Cobb-Douglas function where A = exp( 0 +! it + u it ) is the e ciency term. Q it = L l it K k it M m it A it ; (1) The term Q it is the quantity produced, L l it, K k it, M m it, are the three inputs, labor, capital and material with respective coe cients. The constant term 0 productivity and u it is the i:i:d: term. is the mean productivity,! it is rm unobserved The large part of datasets do not report information on input and output quantity but just values. For this reason, the standard procedure is to use data on values and divide them for a speci c price index in order to obtain quantities. For example the output quantity Q it is substituted with the de ated values of revenues( ~ R it ) 22 : Q it = ~ R it = R it =P it given that 21 Unobservable for econometricians. 22 We could also raise the issue of multiproduct rms relation of product market competition (De Loecker, 2007b). However this is paper s purpose. 11

12 4. CREDIT CONSTRAINTS IN PRODUCTION FUNCTION ESTIMATION 12 Q it P it = V (Q it ) where V (Q it ) is the output value (revenues) and P it is the output price for rm i at time t. Input quantities should be substituted with de ated values, using rm speci c de ators l k m V (Q it ) V (Lit ) V (Kit ) V (Mit ) = A it; (2) P it w it r it where w it, r it and c it are respectively the price for labor capital and material. I take the logs of this equation (2) to implement the estimation and recover production function coe cients. Let s assume I observe all quantities, excluding capital, for which we observe only value. Then (2) in logs is q it = 0 + l l it + m m it + k (vk it r it ) +! it + u it ; with vk it is the log of capital value (not de ated) and r it is rm capital price. As I mentioned above there are no information about capital s price of paid by rm, hence I am obliged to a use capital de ator (r It ) common across rms. If we add and subtract in the right hand side k r It we get 23 (vk it q it = 0 + l l it + m m it + k (vk it r It ) k (r it r It ) +! it + u it : (3) The equation (3) is similar to OP starting equation with an additional factor. The term r It ) is the standard proxy for capital (the de ated value) that is used to identify capital coe cient k. However there is a second term k (r it r It ), that is unobserved due to unobservability of rm s capital price: it is cause of bias in production function estimation (Hausman-Griliches, 1986; Klette-Griliches, 1996). It determines the di erence between capital price paid by rm i and the average capital price de ator r It ; if r it > r It it means that rm pays an higher cost for capital than the average 24 and it has a negative impact on the total output produced (negative sign in front). c it Given that do not observe k (r it r It ), it enters in the error term making our estimation biased for several reasons. Firstly any variations in capital price is correlated with input value namely [E ((vk it r It ) (r it r It )) 6= 0]. The correlation can generate an additional bias in the 23 k It is equivalent to multiply the left hand side of (2) for rit r : It 24 It looks like the beta of Capital Asset Price Model (CAPM). The beta measures the rm s extra return respect to stock market returns. 12

13 4. CREDIT CONSTRAINTS IN PRODUCTION FUNCTION ESTIMATION 13 estimation of capital coe cient (Olley Pakes, 1996); higher is the price for capital and less capital will be used in the production process for a given level of! i : In addition (r it r It ) term is positively correlated with the other two inputs because an increase in capital price will correspond to an increase in the use of other inputs and it will generate biased coe cients also for free inputs (labor and material). Thus even if we take into account unobserved productivity shock, the capital price r it generates biased estimations. Moreover if the estimated coe cients are e cient the calculated productivity (TFP) is not reliable. We can notice that tfp it =! it + u it k (r it r It ) (4) the TFP for rm i includes also price gap for capital. It generates two type of problems, one related to the evaluation of economic policies on rm s productivity, another concerning the role of TFP in the self-selection process. The former problem can be easily illustrated by tfp m it = x t + it ; with which it is possible to evaluate the e ects of a policy x (i.e. trade liberalization) on the e ciency of rm i. If we do not observe the capital price, the empirical analysis can give incorrect results because it is not possible to disentangle the e ect on the e ciency from the e ect on capital price, namely k (r it r It ) k (r it r It ) =@x is the bias due to unobserved price. The second concern is related to self-selection and the role of liquidity constraints for export choice; the TFP as de ned in (4) includes in itself with r it term information about rm s nancial situation. Therefore when I introduce TFP in the right hand side of a probit model (10) the estimated coe cient for e ciency is misleading because it includes capital price: it is not possible to disentangle the e ect of e ciency from an e ect due to the variation of capital price. For these reasons the unobserved capital price is a relevant issue in production function s estimation. However as I mentioned above there exists a substantial heterogeneity in term 13

14 4. CREDIT CONSTRAINTS IN PRODUCTION FUNCTION ESTIMATION 14 of capital demand and prices paid by rms. Investment needs in physical capital, as machinery, change across sectors and rms due to technological reasons. Firm can nd these resources for its projects inside, as cash ows, or outside, borrowing money. Firm which is credit constrained, is not able to raise enough resources (internal and external) to nance investments or it could pay an higher price for capital, forcing to invest less than original project. In order to solve the problem of unobserved capital price I assume that r it (price paid by rm i) is a function of two components, an average capital price r It (common across industries) and a second term that captures rm s ability to raise funds, i.e. r it = f (r It ; it ). The term r It is the capital de ator and it assesses the general condition of capital market in which rms nd resources for investments; it may include information about capital market rigidities (which remain common across rms). The second term i is rm speci c and it represents the heterogeneity of nancial health across rms; it is an unobserved by econometricians and it a ects the investment s choice. Formally I assume that the investment s decision depends also on capital price, namely i = h t (k it ;! it ; r it ) = h t (k it ;! it ; it ): (5) where the last equality derives from an underlying assumption of OP for which the investment price does not vary across rms 25 (in this case the term r It ) A positive shock of i means that for rm is easier to raise funds (i.e. capital s price reduces). The easiness of credit access may depend on external factors independently from rm e ciency: more bank s windows next to the rm, new personal collateral security, better nancial situation, cash ows generated by the sell of rm assets 26. The assumption is realistic especially for Italian SMEs that do not enter in capital markets and have business relation with banks based on strong personal contacts. hence To be more precise I assume that the unobserved capital price is linear in the r I and i, r it = f (r It ; it ) = r It it ; then if I substitute the log capital price in (2) I eliminate rm speci c capital price and 25 Ackerberg et al.(2006) 26 Moreover the business cycle can a ect di erently each rm in term of liquidity. 14

15 4. CREDIT CONSTRAINTS IN PRODUCTION FUNCTION ESTIMATION 15 the production function that has to be estimated is q it = 0 + l l it + m m it + k (vk it r It ) +! it + it + u it : (6) With the equation (6) I recover production function coe cients ( i with i = l; k; m), controlling for unobserved productivity (simultaneity bias), and credit constraints shock ( it ) in order to deal with unobserved capital price. It is important to underline that the de ated value of capital k (vk it r It ) is used now to estimate production function but it is no more a source of unobserved price bias. I will follow Ackerberg et. al (2005) to estimate production function with two unobservable components (! it ; it ). 4.2 Estimation strategy The main problem is to estimate production function taking into account unobserved productivity and credit constraints shocks. Formally I relax the assumption about the scalar nature of unobserved term done by OP; I allow that investment ( it ) depend not only on unobserved productivity (! it ) but also on unobserved credit constraints ( it ) as in (5). Poorly speaking I estimate(6) similarly to OP with the addition of a second unobservable term. Introducing it I capture in the estimation, rm speci c characteristics, as liquidity constraints that could a ect production decisions. It is necessary to assume that both components follow a rst order Markov process! it = E [! it j! it 1 ] + " it it = E it j it 1 + it : Another key assumption is the independency between the two unobserved terms (! i and i ) in order to obtain the inversion of investment function, E [ it! it ] = 0 8 i and t: I group both unobservables in one! it =! it + it as new unobserved state variable (De- Loecker, 2007b) and the investment has to be a monotonic function with respect of! it in order to be invertible: However in order to invert the investment function I need a second control variable to proxy i. For my purpose I use the Solvency Ratio ( i ) among all the other indices as control 15

16 4. CREDIT CONSTRAINTS IN PRODUCTION FUNCTION ESTIMATION 16 variable for unobserved credit constraints shock. The SR is an index which includes stocks, debt and equity: these are variables directly controlled by rm. It is the best candidate to be a control variable among all the other indices 27 because it is better "controlled": rms can modify SR reducing or increasing the value of goods in stock, or changing debt and credit composition. Then the bivariate policy function is it = t (k it ;! it ; it ); (7) it and I assume that it is a bijection 28 (one-to-one mapping) in (! it ; it ) conditional on (k it ). Now I can invert investment demand as! it = 1 t (k it ; it ; it ): (8) I proceed as OP in the rst stage as usual I obtain consistent coe cients for labor and material ( l ; m ). The rst stage changes to include in the polynomial function it also the second control variable it q it = l l it + m m it + t (k it ; it ; it ) + u it ; where t () = 0 + k k it + t 1 (k it ; it ; it ) is the non-linear term approximated with a polynomial expression of third order 29. In the second stage does not change respect to OP since the process of credit constraints shock is assumed to be independent of the productivity shock. Due to the fact that both unobservables follow a rst order Markov process, i.e.! it = E! it j! it 1 + it I obtain a more precise estimates for capital stock ( k ) because ^ includes also credit constraints Formally the second stage is y it ^ l l it ^ m m it (9) = 0 + k k it + g(^ it 1 k k it 1 ) + it + u it ; 27 Or all other indicator used by banks to evaluate rm risk. 28 I continue to assume strict monotonicity. A bijection is a function surjective and injective at the same time. 29 As Ackerberg et al.(2005) make notice, the it () is a complicated dynamic programming problem that depend on all primitives of the model: my methodology consists to consider among those primitives also credit market rigidities 16

17 4. CREDIT CONSTRAINTS IN PRODUCTION FUNCTION ESTIMATION 17 The key identifying assumptions are two, rst the independency between the two unobservable terms, second the uncorrelation between k i and the error term it (standard assumption in Olley and Pakes). The equation (9)can be estimated with Non linear least square (NLLS), with a polynomial approximation of g (treating it not parametrically). Finally with the estimated coe cient I can calculate productivity as tfp it = q it ^ l l it ^ m m it ^ k k it : In the Appendix D I provide production function s coe cients for traditional OP method (Table D.1) and for the extension with the inclusion of rm unobserved credit constraints (OP-CC)(Table D.1). In Appendix D we can notice from Figure D.1 and from Tables D.1 and D.2 that on the average the inclusion of credit constraint in production function generates higher capital coe cients especially in some sectors as metallic production (DM), and motor vehicles (DJ), for whom it is probably required an higher level of investments to be active in the market. As mentioned above the production technology generates di erent demand of liquidity depending on the industry type, so rms in sectors with higher capital requirements may rely more on external nancing. In Figure 1. are represented the distribution of productivity obtained with the two methods 30. Figure 1: Productivity distribution 30 I use only observation from the 2nd percentile to the 98th. 17

18 5. EXPORT ACTIVITY AND CREDIT CONSTRAINTS 18 It is interesting to notice as plant s productivity distribution calculated with credit constraints (OP-CC) has a more dispersed distribution respect to the standard OP productivity (dashed line); it could depend on the fact that a second source of rm s heterogeneity has been included in the estimation of production function and the result brought is an higher variability in the estimated coe cients across sectors. As yet mentioned it is caused by di erences in sector technologies that imply a di erent demand of capital investment. Finally it is possible to notice in Table 3 as the estimated productivity shows the same features of TFP in empirical trade models, namely on average the exporters are more productive than domestic rms. Table 3: Productivity Averages Average LogTFP(Credit) LogTFP Mean s.d. Obs Mean s.d. Obs All Firms Exporter Domestic Exporter Domestic Exporter Domestic : M ean pro ductivity from 1991 to : M ean pro ductivity from 1998 to : M ean pro ductivity from 1991 to 2000 Both for rms in the survey of 1997 and in survey of 2000 the exporter are more productive than non exporter; the grater e ciency of exporting rms respect to domestic is maintained with OP-CC 31. Now the purpose of research is to verify if there are any di erences in term of nancial situation amongst domestic and exporting rms. 5 Export activity and credit constraints In this section I provide the evidence that credit constraints are linked with the export status namely it is more likely that rms involved in exporting market show higher level of 31 You can also notice that the average productivity decrease over time. The rms in survey 1997 which are also in survey 2000 have reported balance sheet data (hence productivity). 18

19 5. EXPORT ACTIVITY AND CREDIT CONSTRAINTS 19 liquidity. The rst empirical test which I perform is not a properly test of self-selection given that I have only information about export status in 1997 and 2000: I need to bring to the attention that surveys consider a period of three years. Unfortunately just 4638 rms report export status and only 129 change between the two surveys (just the 2.8%, 70 entrants and 59 quitters). For this reason, rstly I consider the export status in year 2000 without using any information about the previous export status. Proceeding in this way I estimate if on average exporters are more or less credit bounded than domestic rms. In order to evaluate these di erences I implement a standard maximum likelihood estimation (probit) using as dependent variable the export status in period ; the dataset is a cross section because I do not exploit any type of xed e ect information nor I test hysteresis of exporting activity (Irarrazabal Opromolla, 2005). Formally the estimated equation (10) follows the non structural approach of Bernard and Jensen (1999) and the model can be written as P 1 if >< t X it P t C it + B i + i > 0 t=1998 t=1998 Y i00 = >: 0 otherwise with Y i00 being the export status of rm i in (1= exporter; 0=domestic) and X i a set of rm s characteristics as productivity or capital intensity referred to year (from 1998 to 1999). The third term C i captures rm s nancial health and as for rm s characteristics I consider values in 1998 and Both X i and C i are time variant (panel dimension) because they are provided by balance sheet dataset. Finally B i is the number of banks with whom rm i has a relation in survey period (in log term). I include in the analysis also a "R&D" variable: it is a dummy equal to one if rm faces research and development investments in period otherwise zero. The term i is the i.i.d. error term. The coe cients of interest are t, which measures the e ect of liquidity constraints on exporting probability. In order to evaluate nancial conditions I use three ratio about liquidity, Quick Ratio, Acid Ratio and Current Ratio (Appendix B); a positive coe cients means that on average the exporters are less credit constrained (better nancial status) than domestic rms because higher value for the ratio suggests more liquidity available for rm i. In addition I interpret a positive coe cient for B i as clue in favor of greater availability 19 (10)

20 5. EXPORT ACTIVITY AND CREDIT CONSTRAINTS 20 of credit for exporters. In Table 4 I report the results of probit estimation (10); I consider in the regressions the observations from the 2nd to the 98th percentile in order to eliminate outliers. I include also in the analysis sector dummies to control for industry technology components and business cycles. Each column represents a speci cation while in parenthesis are reported the robust standard errors. Using the productivity estimated as suggested in the previous section, we can notice that exporters are more e cient than domestic rms (as expected) and show higher capital intensity. It is noteworthy observe that all liquidity indices have a positive impact on export status; exporters enjoy a better nancial situation, in the sense that they own higher internal resources. The importance of credit constraints is highlighted also by Bank variable: when rm works with many banks it is probably less tied in credit because it may diversify the nancing s demand 32. The bank coe cient is positive and signi cant even if we consider it with R&D dummy (which is positive and signi cant too). 32 The statistical relation is maintained also considering all variables together. Look Appendix E. 20

21 5. EXPORT ACTIVITY AND CREDIT CONSTRAINTS 21 Table 4: Probit Estimation in cross section. Dependent variable export status year 2000 I II III IV V VI OP-CC (0.255) (0.253) (0.249) (0.266) (0.254) (0.157) OP-CC (0.316) (0.313) (0.301) (0.325) (0.329) (0.318) KL (0.132) (0.133) (0.132) (0.135) (0.134) (0.133) KL (0.133) (0.133) (0.132) (0.136) (0.134) (0.134) ACID (0.169) CURRENT (0.157) QUICK (0.162) BANK (0.072) (0.07) R&D (0.069) Obs Pseudo R R obust standard errors in parenthesis. *,* *, * * * indicate statistical signi cance a t 1 0,5 a n d 1 % le v e l. S e c t o r d u m m ie s in c lu d e d. With the results reported in Table 4 we can infer that on the average exporters own more internal nancial resources, once we controlled for productivity and capital intensity. The importance of credit constraints for exporting is con rmed also by Bank variable. Now I verify with a rough method if e ciency and credit constraints interact among them reinforcing the process of selection. In the second empirical test I introduce as explanatory variable the past export status in cross section estimation (10): the estimated relation is more similar to Bernard and Jensen 21

22 5. EXPORT ACTIVITY AND CREDIT CONSTRAINTS 22 (1999) approach than the previous. I consider rms which appear in both surveys and I slightly modify (10) including productivity (OP-CC), capital intensity and ratios for year 1999 (as Greenway et al., 2007): namely I regress the export status for over rm characteristics X i and credit constraints C i plus export status for (Y i97 ). estimated equation is 8 >< 1 if Y i97 + X i99 + C i99 + B i + i > 0 Y i00 = >: 0 otherwise I continue to use also information about banks and R&D dummy for period In Table 5 I provide the results for this second speci cation and for each column is de ned a di erent speci cation. The 22

23 5. EXPORT ACTIVITY AND CREDIT CONSTRAINTS 23 Table 5: Probit Estimation in cross section. Dependent variable export status year 2000 with past export status I II III IV V VI VII Export (0.114) (0.115) (0.114) (0.115) (0.115) (0.115) (0.115) OP-CC (0.144) (0.147) (0.147) (0.149) (0.146) (0.147) (0.148) KL (0.100) (0.102) (0.101) (0.101) (0.102) (0.100) (0.100) Current (0.179) (0.180) Acid (0.202) (0.204) Quick (0.196) (0.197) Banks (0.122) (0.122) (0.123) (0.123) (0.122) (0.115) (0.123) R&D (0.123) (0.123) (0.123) Obs Pseudo R R obust standard errors in parenthesis. *, * *, * * * indicate statistical signi cance a t 1 0, 5 a n d 1 % le v e l. S e c t o r d u m m ie s in c lu d e d Credit constraints are relevant for the self selection process in trade; the number of banks is a determinant variable to explain the exporting behavior of Italian rms but the signi cance is less strong. The current ratio remains positive and signi cant instead acid and quick ratio no. It could depend on two factors; rstly the export status for predicts in large part the dependent variable and it captures all the variation (just the 2.8% of rms in the sample change status from on survey to the other). Secondly, current ratio is a broader measure of liquidity because it includes (in the numerator) the leftover stock of intermediate input and output 33 ; it could suggests that all kind of potential resources 33 Look Appendix C. 23

24 6. CONCLUSIONS 24 available to generate liquidity are used by rms to expand their activities. The results are in line both with Campa and Shaver (2001) and Manova (2006), as them I nd that credit constraints limit the exporting activity: di erently from Greenway et al.(2007) I nd that e ciency still matters in the self-selection process. It is probably due the estimation method used for calculating TFP instead of Levinsohn-Petrin (2003) procedure used by them. 6 Conclusions In this paper I highlighted the relevance of credit constrains as key variable, both for production function s estimation and for rm s export choice. The availability of nancial resources is a second source of rm heterogeneity and it can be a key variable in the rm s decision process. If resources are available rms are able to invest and expand their production, or are able to overcome sunk cost of exporting. The introduction of credit constraints in production function estimation allows us to solve the problem of unobservable input price bias for capital. The estimated productivity is more disperse respect to OP because now it is possible to take into account roughly the capital demand for each industry in the estimation; capital intensive sector as production of motor-veicheles, can rely more on external nancing rather than other industry and it can generate heterogeneity in price for capital input. With the evaluation credit constraints through liquidity indices and qualitative variables (number of banks per rm) I nd that credit constraints are important for rm exporting activity. Exporters are not only more e cient than domestic rms but they own also more internal resources. Finally I nd that self-section process in export market is a ected by rm credit constraints even if I consider hysteresis in a naive way. References [1] Ackerberg D., Benkard C.L., Berry S., Pakes A. (2005): Econometrics tools for analyzing market outcomes, Handbook of econometrics vol.6 ed. Heckman J.J., Leamer E.E. 24