WPS6839 Policy Research Working Paper 6839 Why Liquidity Matters to the Export Decision of the Firm Rosanna Chan The World Bank South Asia Region Finance and Private Sector Unit April 2014 Policy Research Working Paper 6839 Abstract Under financial constraints, exporting may have less as it is not under financial constraints. However, under to do with productivity and more to do with financial financial constraints, export status is less dependent on resources. The established relationship between exporting productivity and more dependent on the availability of and productivity would differ when examined through working capital. Empirical results support the model’s the lens of the working capital needs of the firm. The prediction. The relationship between exporting time and hypothesis that working capital matters in the firm’s the need for greater liquidity is also borne out empirically exporting decision is explored in two ways: first, by as shown by a positive and significant correlation between articulating a dynamic working capital model of the firm the amount of working capital and the distance of export that incorporates the firm’s export decision. Secondly, by destination. An important policy implication from the testing the hypothesis empirically using a unique firm analysis is that short term liquidity is critical in allowing level dataset from Bangladesh, where issues of financial productive firms to export and that access to finance constraints are particularly acute. The model shows that may prevent the benefits of trade liberalization within a productivity determines export status of the firm as long country to be fully realized. This paper is a product of the Finance and Private Sector Unit, South Asia Region. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author may be contacted at rchan@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Why Liquidity Matters to the Export Decision of the Firm Rosanna Chan∗ 1 Introduction The growth of exports is a key priority for many developing countries as a means to access foreign earnings and to stimulate economic growth.1 Yet, among developing countries, access to external finance is a major problem. The World Bank Investment Climate Surveys, covering more than 26,000 firms across 53 devel- oping countries, find that the cost and access to finance is considered by firms to be the top five problems they face (?). An important question to ask is how do financial constraints affect the export decision of the firm. This question is particularly relevant in light of the 2008 financial crisis, where the freeze on credit coincided with a drop in world exports of more than 30 percent2 that cannot be entirely explained by the drop in aggregate demand. ∗ Economist, South Asia Finance & Private Sector Development, World Bank. Email rchan@worldbank.org I am grateful to Albert Park, Dan Silverman, Jim Levinsohn, Jan Svejnar, Chris House and Andrei Levchenko for their advice, Mary Hallward-Driemeier at DECRG, World Bank for data and institutional background, Aneeqa Aqeel, Oli Coibon, Jane Dokko, Chris Kurz, Sara Lalumia, Rahul Mukherjee, Joo Youn Park, Bill Powers, Heidi Schierholz, Mine Senses, Lina Walker for their insightful comments and encouragement. I would also like thank seminar participants at the Universities of Michigan, University of Melbourne, Australia National University, University of Wisconsin and the Minnesota International Economic Development Conference for their valuable input. All errors are my own. The views expressed here are those of the author and do not necessarily reflect official views of the World Bank. JEL: D01, D2, D21, D22, D24, D92, F02, F10, F14, F15, F36, O12, G20 Keywords: credit constraints, financial constraints, working capital, firm behavior, international trade, trade costs, transaction costs, entry costs to trade, time, trade barrier. All errors are my own. 1 See ?? 2 Quarterly trade data from WTO, compiled from data sourced from IMF, International Financial Statistics; Eurostat, Comext Database; National statistics; Global Trade Atlas. To understand how financial constraints affect export status, the first step is to ask what are the financing needs of an exporting firm. Take for example, a firm owner who has to decide between selling to domestic or foreign markets. She/he would face export specific costs such as the costs of licensing or getting entry to markets.3 However, beyond these additional costs, the owner also faces a significant difference in timing between domestic sales and foreign sales. In the case of Bangladesh, exporting to Hamburg can take up to 39 days. With domestic sales, goods would be delivered faster and the payment for the goods received within a shorter time. With foreign sales, although the goods will fetch a higher price, the goods will take longer to deliver and the payment will be received with a longer delay. The data show that most firms receive payment only after their goods have been delivered. Thus, the decision to export involves the owner asking themselves whether they would be able to remain liquid during the period before payment is received. Operating liquidity in financial metric terms is working capital. As such, to address how financial constraints affect the export decision of the firm, it is necessary to understand how working capital affects the export decision of the firm. This paper makes two contributions to this question. First, a dynamic working capital model of the firm’s export decision is presented. Second, the empirical predictions of the model are tested using a unique panel data set of Bangladeshi firms and robust evidence is found that financial constraints provide a significant barrier to exports. The theoretical model builds upon the dynamic working capital model of the firm ? and extends it to the export decision of the firm. At the time of writing, there has not been any existing trade models that have formalized the role of working capital in exporting decisions. Working capital underlies two fundamental concepts in trade: costs to entry and the role of distance in determining trade volumes. The extra working capital demands from exporting over domestic sales due to shipping time, formalizes the concept of sunk cost or the cost of entry which many trade models are premised upon (see ?????). Furthermore, as the amount of working capital directly reflects transportation time, the greater the distance, the greater the working capital demands in exporting. Therefore, the model also speaks to the area of trade research on how distance determines trade volumes as well as on how time can act as a trade barrier (see ??? for research in this area. ). The main theoretical result is that the export decision of the firm is determined by both productivity and working capital and their relative importance depends on whether the firm is financially constrained or not. The model shows that exporting depends primarily on productivity if the firm is not financially constrained. When firms are financially constrained, however, a lack of working capital can severely restrict a firm’s willingness and ability to engage in exports. The implication is that among the population of firms, there will exist low productivity firms that never export regardless of their cash positions, and cash-poor firms that do not export regardless of their productivity levels. To assess the importance of this mechanism, I use a unique firm-level data set which allows two key aspects of the model to be examined: a) a direct test of the interaction between financial constraints, work- ing capital and productivity in determining the export status of the firm, and b) the relationship between working capital and the export distance that underlies the rationale behind the theoretical model. The 2002 Bangladesh Investment Climate Survey is one of the few panel data sets available that contain firm level fi- nancial information, their access to credit and detailed information on exports. The empirical results provide support for the model’s predictions that the effects of working capital and productivity will differ between financially constrained and unconstrained firms. Specifically, for financially constrained firms, productivity matters less than for unconstrained firms while working capital matters much more. The correlation between 3 Costs to exporting also include shipping and transportation, custom duties and many more. These are discussed in ?. 2 working capital and export distance is shown empirically to be positive and significant, lending support for the working capital model of the firm’s export decision. The paper is structured as follows: the next section discusses the relevant literature for both theoretical firm models and empirical work in the area of export and finance4 , Section 3 motivates the need for the working capital model of export decision by taking a preliminary look at the data to examine the distribution of productivity between exporting and non-exporting firms and also to look at the additional working capital demands that exporting imposes, Section 4 develops the working capital model of the firm’s export decision, Section 5 examines the empirical results and then the conclusion. 2 Relevant Literature The role of firm productivity as a determining factor for the export status of the firm has been researched extensively both theoretically through the heterogeneous firm models of ?? and empirically in the work of ???. It is, however, ambiguous as to whether the existence of financial constraints changes the relationship between productivity and exporting. The literature on the interaction of trade and financial factors has existed for some time(see ?), but the emphasis has been on incorporating finance into the theory of comparative advantage to explain trade pat- terns across countries. Only more recently has the new trade literature begun to consider the role of finance in heterogeneous firm models (see ????). Existing models of exporting decisions such as ??? assume a set percentage of borrowing and do not account for the accumulation of internal finance that affect financial constraints and reduce the demand for borrowing. Their models assume that all firms are affected by financial constraints, due to the set borrowing requirement, and therefore financial constraints will always influence exporting decisions, regardless of differences in internal financing between firms. This is a restrictive assump- tion as it is important to take into account both good financial health, when firms do not require to borrow, and when firms are financially constrained so as to accurately capture the impact of financial constraints on the distribution of firms in the aggregate. Existing models also assume that allocation of credit is perfectly aligned to productivity. This assumption drives the result of these models: exports with financial constraints simply raises the productivity cutoff across all firms. Existing models assume no time difference between inputs and revenues and as such, the financing of the inputs responds directly to the input choice. This, coupled with the assumption that credit is perfectly aligned with productivity, drives the existing models’ result.5 However, existing literature has shown that the allocation of credit does not correspond perfectly to productivity (see ??). The assumption of direct correspondence is thus too strong as it does not allow for the existence of distortions in the allocation of credit. The assumptions made by existing theoretical models, static borrowing demand and direct correspondence between allocation of credit and productivity, lead to essentially the same result as predicted by ? with the caveat that financial distortion raises the productivity cut off level across all firms. The empirical literature using micro level data on the relationship between exporting and finance is limited due to both the availability of firm level data sets as well as the topic being an emerging area of research. Of note, one of the few studies is ? that uses a large UK panel data set. They show that the financial health of 4 Theliterature on the area of trade credit , such as ????? reviewed in ?. 5 The working capital model differs from the existing models in that the firm’s maximization problem at time t is bounded by predetermined internal revenues- that is, revenue is a state variable. Even if credit is perfectly aligned to productivity in the working capital model, this would not detract from the predictions that are driven uniquely by the timing difference between inputs and receipt of revenues. 3 firms positively affects export decisions. Another more recent study, ? uses cross section of firm level data across 9 developing countries and they find that access to finance is important to the firm’s decision to enter exports. Furthermore, they find that financial constraints create a disconnection between firms’ productivity and their export status: productivity is only a significant determinant of the export decision if the firm has a sufficient access to external finance. While the existing empirical work show that there are significant relationships between exporting and financial factors, neither studies relate these empirical findings to a model that explains the underlying economic relationships that drive these results. 3 Why might working capital restrictions affect exporting decisions? The Bangladesh Investment Climate Assessment 2002 surveyed 990 firms and collected annual recall data for 2000 and 2001 for key variables, except for access to finance questions that are only reported for 2002. A total of six manufacturing industries are represented: Garments, Textiles, Food, Leather, Electronics and Chemicals. The distribution of firms within each of these industries is representative of the composition of the Bangladeshi economy, with 57 percent of the observations in the garments and textiles industry. The data set contains financial information that allows working capital to be measured at the firm level. Working capital is defined in the data as the sum of inventories, accounts receivable and cash on hand.6 Working capital can be thought of as current assets or the liquidity of the firm. A unique characteristic of the data set is that it contains country destinations of exports. This allows us to look at whether the distance to country destinations is correlated with working capital at the firm level. There are three major observations from the data that indicate working capital is particularly important for exporting. First, there are substantial additional costs to exporting in the form of custom duties and procedures. Table 1 shows that average official costs are 84,935 takas (approximately $1,456 USD) and average unofficial costs total 32,895 takas (approximately $564 USD). In light of the average value of exports is only 139,981 takas (approximately $2,399 USD), total custom costs are a substantial cost to firms in Bangladesh. In addition to these costs, exporting firms also hire clearance agents to help get them through customs -around 90% of exporting firms use a clearance agent at an average cost of 0.82% of the value of the freight. These costs plus the time delays in customs would require the firm to have adequate working capital in order to just get their goods through customs. Second, exporters do not get paid until their goods are delivered to the export destination for a much larger proportion of their sales than non-exporters. Table 2 shows that on average, 89 percent of the sales for exports is paid upon delivery (median is 100 percent) compared to 65.74 percent for non-exporting firms (median 75 percent).7 Therefore, exporters would require to cover a greater percentage of their costs with their own working capital than non-exporters. Furthermore, the longer the time period is between production and delivery, the greater the amount of working capital is required. Last, delivery times are longer for exporters than non- exporters. Turnover time differs from industry 6 Working capital is measured according to the accounting definition: current assets (inventories, accounts receivable, cash and short- term credit) minus current liabilities (accounts payable and any short-term debt). Note that working capital data is only reported for 2002 7 A proportion of sales is also bought on credit, and here, non-exporting firms appear to extend more credit than exporting firms. Extending credit to customers would further delay the receipt of revenues from when the cost of production was incurred. On average, amongst those that do extend credit to customers, 43.77 days of credit is extended with a median of 30 days. This is funded mainly from retained earnings ( median of 60%, with the rest from micro lenders or informal lenders). This implies that those firms that extend credit to their customers are doing so because they have enough cash or access to finance. They charge their customers extra for the credit for the delay in the receipt of payment and the delay is not crucial for completing the exchange. Unlike the delays caused by shipping across large geographical distances which are unavoidable and crucial to completing the transaction between the exporter and the customer. 4 Table 1: Costs to Customs 2002 Bangladesh ICA Mean Standard Deviation Average days to clear customs 8.87 10.69 Maximum number of days to clear customs 14.10 13.76 Official costs (takas) 84935 440869 (USD) 1456 7556 Unofficial costs (takas) 32895 221242 (USD) 564 3792 Value of exports (takas) 139981 321491 (USD) 2399 5510 Percent that use clearing agent 89 Cost of agent as % of the value of freight 0.8132 3.109 Source: Bangladesh 2002 Investment Climate Assessment Survey, World Bank Note: The exchange rate in 2002 average is 0.017137961 takas=1 USD Table 2: Timing of payment for sales % of Sales paid % Sales paid at % Sales bought Number Mean before delivery delivery on credit of Obs Non-Exporting 5.57 65.74 28.43 561 Firms Exporting 2.87 89.42 7.66 421 Firms Total 4.41 75.89 19.53 982 Source: Bangladesh 2002 Investment Climate Assessment Survey, World Bank to industry. It ranges from a month in garments to 4 months for shrimp farming (?). For firms that export, clearing customs, transiting to overseas destination and clearing foreign customs all add to the amount of time firms have to wait before receiving payment for their goods. The time to clear domestic customs is on average around 9 days, with a maximum of 14 according to the 2002 Bangladesh ICA (see Table 1). For transiting time, Bangladesh exports are shipped using international shipping lines that run on regular schedules. The time in transit consists of: overland to port, a feeder journey from Chittagong or Dhaka to Singapore to meet up with the international shipping line, and then the destination port to customer location. According to ? , the greatest delays are caused in the exchange from feeder to mainline vessel as schedules may not be synchronized and containers may wait in the transshipment port for several days. His estimate of ocean transit times from Bangladesh, using a variety of shipping companies, is between 25-35 days. Figure 1 shows the transit times for APL shipping (the world’s sixth largest container transportation and shipping company) which corroborates the ? estimate. Without even counting for time needed to clear foreign customs and inland delivery, selling to European markets from Bangladesh can add an extra month to the product cycle. Considering that production of garments only requires about a month to process, exporting essentially would double the amount of working capital required. Among exporting firms, Figure 2 shows that the amount of working capital is increasing with the distance of export destination. This relationship is particularly stark with Europe and the EU - destinations that have the longest shipping transit times from Bangladesh. This suggests that if firms require additional working capital the longer the transit time or the greater the distance of the export destination. 5 Figure 1: Ocean Transit Times (Days) from Chittagong Source: APL Limited (American President Lines Ltd.) at http://www.apl.com/services/documents/sells_mkt_ia_chtct2.pdf Figure 2: Export Destination and Working Capital Source: Bangladesh 2002 Investment Climate Assessment Survey, World Bank Notes: Average of the percentage of working capital to total costs (sum of total wage cost, raw material cost, energy costs and other costs). Other sectors and regions drop due to lack of observations. 4 A Model of Export Decision with Working Capital The previous section motivates the importance of working capital in the firm’s export decision. This section presents a dynamic model of the firm’s export decision that allows for differences in liquidity to affect export decisions. The purpose of articulating an export model where working capital is required for production is to show how productivity and cash interact to jointly determine export status of the firm, and as a result, the presence of financial constraints distorts the selection of the most productive firms into exports through the time demand for liquidity required for exporting. The interaction is highly intuitive: the firm exports if it has met both productivity and cash requirements for exporting. This implies that among the population of firms, there will exist low productivity firms that never export regardless of their cash positions, and likewise, cash poor firms that do not export regardless of their productivity levels. The interaction of productivity and cash as necessary conditions to enter exports would give rise to lower correspondence between productivity and export status of the firm than the correspondence predicted by earlier models. Take a representative firm within an industry with the production function F (Kt , Lt ). At each period, the firm chooses inputs in labor and capital as well as the percentage of output, ψ , to sell to international markets8 to maximize all future stream of profits. Capital takes one period to install. Labor is inelastically supplied. Productivity, Ai , is exogenously given and differs across firms. Firms know their productivity level and makes 8 Likewise, (1 − ψ ) percent is sold to domestic markets 6 Figure 3: Time line for Production and Receipt of Revenue t-(1+S) … t-1 t t+1 … t+(1+S) Receive domestic sales(t- Receive domestic sales Receive domestic sales 1) shipped at t (t+S) Receive export sales(t- Receive export sales Receive export sales(t-S) (1+S)) shipped at t Discover price Pt Discover price Pt+1 Determined domestic Determine domestic Determine domestic sales(t-1) salest salest+1 Determined export Determine export Determine export salest sales(t-(1+S)) salest+1 Note: Price Pt represents both domestic price PD and foreign price PF . production decisions based on the uncertainty of output and prices in the domestic and foreign markets.9 The firm also has access to financial services where it can borrow or save, b, at interest rate r.10 The standard assumptions apply where wages, w, the price of capital, pk , and the interest rate, r are assumed to be 1 exogenous and non time varying. The discount factor given by β = (1+r) and the depreciation rate of capital given by δ . Entry to export markets requires a fixed cost, f , each period, which can be viewed as a license, permit or agent’s fee to export. To incorporate the need for working capital, time delays between production and receipt of revenue are introduced. The time delays are as follows: selling to domestic markets requires one period to receive the revenue and selling to international markets requires S periods in addition to the one period delay that is normally incurred through domestic sales. That is, exporting requires additional time for products to be shipped and received overseas over that of domestic sales. Figure 3 shows the timing of production and receipt of revenue. At each period, the firm receives the revenue from past sales: past sales includes any domestic sales from period (t − 1) and any export sales from period (t − (1 + S)). At each period, the firm also decides on production to be sold domestically and/or exported - the revenue from which will be received either at the next period (t + 1) or at t +(1 + S) periods later. All costs of production are incurred at the time of production, regardless of whether products are going to export or domestic markets, but the production choice affects the transitional equation at the present period and at (1 + S) period when export revenues arrive. At each period, the firm discovers the prices for output; the domestic price, PtD and the foreign price, PtF . Both prices are stochastic over time. For simplicity, the present value of the foreign price is assumed to be always greater than domestic price so that there is an incentive for firms to enter exports, β S PF − PD > 0.11 The availability of export markets can be thought of as an additional mechanism to transfer resources over time: domestic sales get PD and returns are delivered at time t + 1 while international sales get PtF but incurs a fixed cost and returns are delivered at time t + 1 + S.12 Intuitively, the extra time required for exports costs 9 This departs from current literature where firms only realize their productivity after paying a fixed cost ???. The rationale is that firms are privy to more information about their own operations and how productive they are more so than knowing the price conditions on foreign markets. 10 Note, this follows the international finance literature in that b represent bond holdings and positive implies savings and negative implies borrowing. 11 Relaxing this assumption will simply allow more instances where firms will choose not to export because it isn’t as profitable. The objective of this paper is to highlight how financial constraints affect export behavior, and thus it is important for firms to have an incentive to export in the first place. 12 The firm also has available bonds, that earns interest, r , that is received also at t + 1. It is implicitly assumed that the rate of return from domestic production is greater than the interest rate - or else there is no incentive to produce. In other words, PD is bounded from below such that there is always an incentive to produce. 7 the firm the discount rate β S on revenue which is increasing with time S. The delays imply that at each period, the firm has a stock of wealth made up of revenues that are due from past production and any bond holdings from last period. This stock of wealth is the firm’s working capital, Xt : Xt = (1 − ψt −1 )PtD F −1 AF (Kt −1 , Lt −1 ) + ψt −(1+S) Pt −(1+S) AF (Kt −(1+S) , Lt −(1+S) ) + (1 + r )bt −1 The cash the firm has at any particular point will not include all export sales pending from all previous periods but only the export sales due at that period. The revenue for export sales is only realized as cash at the period the firm receives it. Working capital is liquidity, and pending payments that are not received cannot be used to pay for inputs.13 The firm chooses labor, investment and percentage of production to export according to how much work- ing capital it has. Therefore, the firm’s budget constraint is given by: wLt + pk (Kt − (1 − δ ) Kt −1 ) + { ft |ψt > 0} + bt = Xt Financial constraints are introduced as a limit on how much the firm can borrow at any particular time. The firm is restricted to borrow only up to a percentage, ϕ , of it’s total wealth: bt ≥ −ϕ Xt .14 Define π D as the current value profit from only domestic sales; π D = β PD Ai F (K , L) − C(K , L, K−1 ). Likewise, define π E as the current value profit from sales where exports are non-zero; π E = (1 − ψ )β PD + ψβ 1+S PF AF (K , L) − C(K , L, K−1 ) − f . Variable costs are denoted by C (K , L, K−1 ) = wL + pk (K − (1 − δ )K−1 ). The dynamic maximization problem of the firm can be described by the value function defined below: V (X , K−1 ) = max max π D + β E V X , K + ν [(1 + ϕ )X − C(K , L, K−1 )] ψ =0,ψ >0 L,K F s.t . X = PD Ai F (K , L) + ψ−s P−s Ai F (K−s , L−s ) + (1 + r ) [X − C (K , L, K−1 )] , max π E + β E V X , K + ν [(1 + ϕ )X − C(K , L, K−1 ) − f ] L,K ,ψ F s.t . X = (1 − ψ )PD Ai F (K , L) + ψ−s P−s Ai F (K−s , L−s ) + (1 + r ) [X − C (K , L, K−1 ) − f ] Following ?, the borrowing constraint for domestic,(1 + ϕ ) X − C (K , L, K−1 ) ≥ 0 and for exports, (1 + ϕ ) X − C (K , L, K−1 ) − f ≥ 0, is embedded in the value function where ν denotes the shadow value of loosening bor- rowing constraint. The firm’s maximization problem involves 2 steps due to discontinuity introduced by the fixed cost of exporting. First, given the cash available and output prices, the firm determines the output levels that will maximize profits under domestic sales as well as for export sales. Under export sales, the firm also determines 13 It is possible that pending payments could be used as collateral to borrow more, but this scenario will be left for further research. The focus of this paper is to look at cash strap firms in developing countries where any kind of credit is difficult to obtain even on cash collateral. 14 This setup differs from the existing literature in three ways: 1) borrowing is endogenously chosen, 1) the probability of exiting export markets is not assumed a priori, 2) the amount the firm produces for domestic and international markets is chosen endogenously by the firm in response to relative prices and 3)the time required to trade and the resulting liquidity requirements are modeled explicitly. I take out the assumption of monopolistic competition, where exporting firms can influence price levels, and substitute this with financial constraints as this may be a more realistic assumption for firms in developing countries. 8 the optimal amount to export if exporting. Second, the firm then chooses between the two profits to determine whether to export. 4.1 Domestic Sales The firm’s dynamic problem in the domestic sales case reduces to only two possible solutions, as established in ?; the firm is either not financially constrained, in which case it produces at the optimal, or it is constrained, in which case, it produces as much as it can given cash on hand. The first order conditions are repeated below: Unconstrained: (1) β PD FL (K , L) = w (2) β PD FK (K , L) = pk . (r+δ ) (1+r) LD∗ and K D∗ is the solution to the firm’s maximization problem if and only if: Equation (1) and (2) hold and (1 + ϕ )X − wLD∗ − pk (K D∗ − (1 − δ )K−1 ) > 0. Otherwise, the solution is given by Equations (3) and (4) below: Constrained: β PD FL (K ,L) D K (K ,L) 1−δ ) (3) w = β P Fp k +( (1+r) 1 k (4) X = (1+ ϕ ) wL + p (K − (1 − δ )K−1 ) Let V ∗D (Ai ) denote the value of the value function of optimal domestic profits as a function of A and let X D∗ (Ai ) denote the minimum cash requirement for unconstrained profits to occur. Note that both profits and cash requirements are increasing with technology, A. Let V D (Ai , Xi ) denote the value of the value function where profits are at constrained optimal (where Xi < X D∗ (Ai )). It is necessarily so that constrained optimal profits is lower than unconstrained profits: V ∗D (Ai ) > V D (Ai , Xi ) . 4.2 Export Sales The firm’s dynamic problem in the export sales case can also be similarly reduced to two scenarios. Let Σ denote the expected sum of all future shadow values up till the period export revenue is received; Σ = E [νS−1 (1 + ϕ )] + .....E [ν (1 + ϕ )]. This can be interpreted as the opportunity cost of liquidity due to export- ing. The first order conditions under export sales are given by: 9 ∂V ∂V (5) : (1 − ψ )PD + ψβ S PF β AFL − w 1+E = ν w + ψβ 1+S PF AFL .Σ ∂L ∂X ∂V ∂V : (1 − ψ )PD + ψβ S PF β AFK − β (r + δ ) pk 1+E ∂K ∂X (6) = ν pk + ϕβ ν (1 − δ ) pk + ψβ 1+S PF AFK Σ ∂V ∂V ∂ X ∂ V (1+S) ∂ X (1+S) ∂ RF (7) : (−β PD + β 1+S PF AF (K , L) + β E + β 1+S E ∂ψ ∂X ∂ψ ∂ X (1+S) ∂ RF ∂ ψ either ν = 0 ⇒ (1 + ϕ )X − C(K , L, K−1 ) − f = 0, or ν = 0 ⇒ (1 + ϕ )X − C(K , L, K−1 ) − f > 0 Identical to the case of optimization of domestic sales, the choice of labor and capital is governed by their respective marginal cost when unconstrained and governed by the binding budget constraint when con- strained. Unlike the case of domestic optimization, optimizing export sales involves taking into consideration of the additional future periods until export sales revenue are received. The first order condition for exports dynamically captures the timing differences between domestic sales and export sales and also show why this matters under financial constraints. One way to see this, is by ∂X ∂ X (1+S) ∂ RF simplifying the first order condition by substituting in ∂ψ = −PD AF (K , L) and ∂ RF ∂ψ = PF AF (K , L) into the first order condition for ψ , Equation (7). This yields: ∂V ∂V ∂ V (1+S) F (8) = Aβ F (K , L) β S PF − PD − E PD + β S E P ∂ψ ∂X ∂ X (1+S) The entire term in the square brackets can be interpreted as the net price of exporting which has two com- ponents: i) the difference in price between exporting and domestic, given by β S PF − PD , and ii) the value ∂V ∂ V (1+S) of the timing of the different revenue streams, given by −E ∂X PD + β S E ∂ X (1+S) PF . An increase in the ∂ V (1+S) percentage of exports will increase export revenue via the +β S E ∂ X (1+S) PF term at time (1 + S) while a de- ∂V crease in the marginal domestic revenue, via the −E ∂X PD term at the next period. Each of these revenue ∂V streams are weighted by the corresponding value of cash at that point: ∂X , the expected change in the value ∂ V (1+S) function due to additional cash at the next period and ∂ X (1+S) , the expected change in the value function due ∂V ∂ V (1+S) to additional cash at time exports arrive, (1 + S). The second component, −E ∂X PD + β S E ∂ X (1+S) PF , is only relevant when financial constraints are binding as that is the only time when additional liquidity brings value to the value function.15 The additional time dimension of the export sales maximization problem makes the solution less tractable than the domestic case. However, the export maximizing solution can be characterized similarly as domestic sales into unconstrained and constrained scenarios: Unconstrained If constraints are never binding during the S periods it takes for international sales to be received, i.e. ν = ν = .....ν S = 0, then: 15 Cash, X , in of itself does not contribute to the value function but only when financial constraints bind. 10 ∂V = Aβ F (K , L) β S PF − PD ψ =1 ∂ψ ψβ 1+S PF AFL = w (r + δ ) ψβ 1+S PF AFK = pk . (1 + r) Constrained ∂V (9) β S PF − PD 1+E = β S PF Σ ∂X (10) wL + pk (K − (1 − δ )K−1 ) + f = (1 + ϕ )X (1 − ψ )PD + ψβ S PF β AFL − w ν w + ψβ 1+S PF AFL Σ (11) = (((1 − ψ )P + ψβ P ) β AFK − β (r + δ ) pk ) D S F ν pk + ϕβ ν (1 − δ ) pk + ψβ 1+S PF AFK Σ Equation 9 comes from a derivation of the first order condition from exports, given by: ∂V ∂V (12) = Aβ F (K , L) β S PF − PD 1+E − β S PF Σ ∂ψ ∂X 16 Thus the solution to the export maximization problem is as such: • If there are no binding future constraints for all future periods up to (1 + S), i.e.Σ = 0, then the marginal value of increasing export sales is strictly positive. Exports in this case will be the maximum amount possible which is 100 percent; ψ = 1. • If there are binding future constraints in the period up to (1 + S), the extent that financial constraints are binding during that time will determine the amount of exports. The choice of ψ affects the size of Σ, ∂V the shadow values ν and ν as well as E ∂X . The choice of labor, capital and export percentage will be determined where all three equations are satisfied. Exports in this case can take on values between 0 and 100 percent due liquidity constraints. Let V ∗E (Ai ) denote the value of the value function of optimal domestic profits as a function of firm specific technology Ai and let X E ∗ (Ai ) denote the minimum cash requirement for unconstrained profits to occur. Additionally, let V E (Ai , Xi ) denote the value of the value function where profits are at constrained optimal (where Xi < X E ∗ (Ai )). It is necessarily so that constrained optimal profits are lower than unconstrained profits: V ∗E (Ai ) > V E (Ai , Xi ) The cash requirement, X E ∗ (Ai ), to satisfy Σ = 0 can be calculated by iterating backwards the budget constraint from the period when export revenue is received. At that period, cash on hand must be greater than 1 expected cost. ie Xt +s > (1+ϕ ) Et [Costt +s ] for constraints to not bind. Likewise for the period before that, and so on and so forth. The minimum cash required for constraints to not bind for each period thus can be given as: 17 16 This ∂V ∂V utilizes a backward iteration of ∂X =E ∂X + ν (1 + ϕ ) that allows us to summarize the change in the value function at time ∂ V 1+S ∂V (1 + S) in terms of the history of shadow values leading up to it: E ∂ X 1+S =E ∂X − Σ. 17 The cash threshold is defined for where the firm also expects to export in the future. An alternative scenario is that, due to 11 Let CF denotes the cost to produce optimally at foreign prices. X E ∗ (Ai ) = CF + f − β 1+S PtF Ai F (K , L) S +Et CF + f ∑ βs s=1 1 (13) +Et CF + f β 1+S − ω (1 + ϕ ) S−1 F Given : F ω = β S P−1 Ai F−1 + β F P−2 Ai F−2 + . . . + P−(1+S) Ai F−(1+S) The characteristics of the cash threshold as defined by Equation 13 is intuitive as all costs enter positively; an increase in production costs and fixed costs will raise the cash threshold. Revenues enter negatively; if foreign price, PF , increases, the amount of cash required to export will decline. ω denotes the current value of foreign sales revenue that is going to be delivered in the future, between t = 0 and t = (1 + S), from past production. If the firm has never exported before then ω = 0. If the firm has exported before (and the greater the value of ω is), the lower the initial amount of cash required to export. This captures how firms are more likely to export, if they have exported before. The motivation here, however is not a learning story as is often forwarded by existing trade story (ie ?), but one of liquidity and the timing of when cash arrives - firms that have exported before will have more liquidity to export in the present period. The cash requirements also show that in addition to the fixed cost f required to export, the firm requires additional liquidity to cover the longer delays in exporting. The term Et CF + f ∑S s=1 β s is the cost of production for exporting sales during the interim. The greater the distance, S, the greater the cash required for the firm to not run into binding liquidity constraints. Thus it captures the role of distance in exporting and relates this to the liquidity demands. The borrowing allowance ϕ also affects the cash threshold: if it decreases, the threshold also increases as firms cannot rely on external borrowing to finance production but need to finance internally. The cash threshold is an increasing function of firm productivity Ai as cost of production, C, increases with inputs L and K which themselves are increasing functions of productivity. Intuitively, greater productivity increases the level that optimal scale of production is reached. 4.3 Exporting Decision The exporting decision rests on choosing between domestic sales or export sales depending on which value stream is greater. The value streams depend on the state variables Ai and Xi .18 First, looking at technology Ai , let’s assume that the firm’s cash level Xi is greater than the minimum cash requirement to export, X E ∗ (Ai ). The firm picks the maximum of the two unconstrained optimal value streams: V ∗D (Ai ) or V ∗E (Ai ) . As cash is not a binding constraint both at the present time period and in the future, the firm need only compare current profits: π D∗ = β PD Ai QD∗ − CD∗ to π E ∗ = β 1+S PF Ai QE ∗ − CE ∗ − f . The higher than mean expected price realizations, the firm is able to export once but expects to return to domestic sales only as prices E ∗ (A ), will be lower. X E ∗ (A ) = CF + f − β 1+S PF A F (K , L) − return to expected mean. In this scenario, the cash threshold , XFD i FD i t i S 1 1+S − ω Et β PtD +1 A i F (K , L ) − C D ∑s=1 β s + E CD t (1+ϕ ) β Price expectations are iid such that: Et [Pt +1 ] = Et [Pt +2 ] = . . . = Et [Pt +T ]. Expected costs at time t are the same: Et CtF +1 = Et CtF F F D +2 = . . . = Et Ct +T where C denotes the cost to produce optimally at foreign prices and C the cost at domestic prices. 18 Recall that all firms are assumed to face the same prices, wages and interest rates. 12 ¯ , is defined where π D∗ = π E ∗ : productivity threshold, A ¯ C E ∗ − C D∗ + f A = β 1+S PF QE ∗ − β PD QD∗ The productivity threshold exhibits some intuitive characteristics in terms of it’s parameters. An increase in the foreign price PF or a decrease in the domestic price, PD , lowers the threshold as this causes exporting to be more profitable relative to domestic sales. Likewise, an increase in the fixed cost, f , raises the productivity bar for firms to be profitable exporting over domestic sales. An increase in the delay, S, such as through shipping distance, increases the threshold through the discount rate. The productivity threshold is important in that it defines the minimum technology level required for the firm to be profitable in exporting. It would not be profitable for a firm with technology below the threshold to export, regardless of it’s cash position. This threshold holds even under financial constraints as a firm that is not profitable exporting without constraints, will not be profitable exporting with constraints. However, productivity alone does not determine whether the firm exports. While the productivity thresh- old separates firms with the ability to export from those that cannot, whether the firm has the means to do so will depend on the level of cash, Xi . Clearly, if the firm meets both productivity and cash requirements, Ai > A¯ and Xi > X E ∗ (Ai ), the firm will export. Also, if the firm does not meet productivity requirements, Ai < A¯ , then the firm will not export regardless of the cash position, Xi . Therefore, out of the set of solu- tions dependent on the state variables Ai and Xi , the situation where the firm meets productivity requirements, Ai > A¯ , but does not have the cash, Xi < X E ∗ (Ai ) , is the only combination left to fully map out the solution. Previously, in the sections on domestic sales and export sales, the minimum cash levels required for unconstrained optimal were defined as X D∗ (Ai ) and X E ∗ (Ai ) respectively. If the firm has cash levels below the minimum domestic “unconstrained optimal”, that is Xi < X D∗ (Ai ), then the firm would necessarily not choose to export as the value of exporting is lower than domestic sales, V D (Ai , Xi ) > V E (Ai , Xi ) . This will hold regardless of productivity levels, Ai . Intuitively, if the firm cannot afford to produce at optimal for domestic markets, it would not produce for export markets where there are even greater demands on liquidity ¯. due to fixed cost of exporting and the longer delays in the receipt of revenues, even if Ai > A Take the situation when the firm’s cash level is above the minimum domestic unconstrained optimal but below the minimum export unconstrained optimal, that is X D∗ (Ai ) < Xi < X E ∗ (Ai ) and Ai > A ¯ . In this case, the firm has more than enough liquidity to sell at domestic optimal, but not enough to sell at the export unconstrained optimal. The firm needs to compare between unconstrained optimal domestic sales, V D∗ (Ai , Xi ), with constrained export sales, V E (Ai , Xi ), where export percent is less than 100 percent. This situation gives rise to two interesting outcomes: either a) the firm chooses to export, but it will export at less than 100 percent which is unusual to observe when there is a fixed cost to export, or b) the firm chooses to sell to domestic markets even though it meets productivity requirements. The export decision mapped accordingly to the interaction between the two state variables Ai and Xi can be summarized by Figure 4. All firms below the productivity threshold needed to be profitable in export sales do not export regardless of cash levels; that is, if Ai < A , then exports will equal zero, ψ = 0, for all values of Xi (Areas IV, V and VI in Figure 4). All firms that above the productivity threshold, and are not financially ¯ and Xi > X E ∗ (Ai ), then exports will equal one, constrained to export, will export 100 percent; that is, if Ai > A ψ = 1 (Area III in Figure 4). Area I in Figure 4 demarcate values of cash that constrain firms from producing at domestic unconstrained optimal levels and thus will not export despite productivity levels that are above ¯ and Xi < X D∗ (Ai ), then exports will equal zero,ψ = 0. Area II demarcate values of the threshold; if Ai > A 13 Figure 4: Summary of Model Results Note: FC -Financially constrained, NFC - Not financially constrained. cash that constrain firms from producing at unconstrained export optimum, but allow the firm to produce at unconstrained domestic optimum. Here, export able firms will choose to mix export sales with some domestic ¯ and X D∗ < Xi < X E ∗ (Ai ), then export values sales depending on the level of cash on hand; that is, if Ai > A will take on values between zero and one, 0 ≤ ψ ≤ 1. Within Area II, cash is the greater driver of the export status of the firm. For example, take two firms with the same cash level Xi where X D∗ < Xi < X E ∗ (Ai ) and both have productivity levels higher than the threshold. The firm that has lower Ai would have a higher likelihood of exporting as it has lower liquidity requirements to export. (See cash threshold requirements Equation 13 that the higher the Ai the greater the cash is required to produced at unconstrained optimal.) The results of the working capital model with exports is unique in that it captures the tension between time and payoff that arises under financial constraints. The firm faces a tradeoff between getting a higher price but waiting longer, or getting a lower price but getting it sooner in its decision between exporting and domestic sales when financial constraints are present. Under severe financial constraints, such as in Area I in Figure 4, firms choose to sell to domestic sales as liquidity constraints demands that payoffs be received sooner - even though the option of selling at a higher price is available. Under severe financial constraints, firms cannot afford to wait the longer time it takes to export. As financial constraints become less restrictive, such as in Area II in Figure 4, the firm will mix both domestic and export sales to maximize payoff and the timing of when revenue is received to ensure adequate working capital for future production. The decision to export under financial constraint is analogous to a decision in investing in an illiquid in- vestment. This occurs as liquidity becomes important when firms are financially constrained. When liquidity is important, firms would not strictly specialize in domestic or exports so as to not put all their liquidity in one longer term illiquid investment. Instead, they would choose to spread the ‘liquidity investment’ over both domestic and export sales. The working capital model with exports is essentially a model of liquidity for the firm, that draws parallels with models of liquidity in the household savings literature (ie. ?). Additionally, the working capital model provides an alternative explanation to recent trade research that attributes the negative correlation of domestic sales with exports to decreasing returns to sales; see ? and ? . 14 The results of the export working capital model of the firm show that the export decision of the firm is determined by both productivity and working capital. The model exhibits the characteristic results of ? in heterogeneous firm models of export, where exporting depends on productivity but also highlights the fact that this only occurs if the firm is not financially constrained. The working capital model shows that under financial constraints, the lack of working capital can yield different export outcomes even if the firm is productive enough to export. Liquidity constraints can constrain export potential firms to not export and/or export less than 100 percent. Therefore, the correspondence between productivity and exporting is much weaker. 5 Empirics This section examines empirical evidence in support of the model. First some stylized facts are presented: 1) the most productive firms do not necessarily self select into exporting and there does not appear to be a productivity cut-off above which all firms export, 2) significant differences exist in physical and working capital between exporters and non- exporters even within the same industry and 3) access to finance differs significantly between exporters and non-exporters and credit does not appear to be allocated according to productivity. The Bangladeshi data set allows two key aspects of the model to be empirically tested: a) whether the export status of the firm is determined by both productivity and working capital and that their effects depends on whether the financial constraints are present, and b) whether there is a relationship between working capital and the export distance. 5.1 Stylized Facts 5.1.1 Productivity of exporters and non-exporters in the data: Self-selection not evident Productivity is measured as total factor productivity estimated according to ? as the method allows for more generally plausible assumptions as to the dynamic data generating process of the inputs used to estimate the production function. 19 A possible caveat to estimated productivity is that estimates of the production function with a large proportion of financially constrained firms could potentially underestimate factor coefficients as firms are not operating at optimal scale. This potentially could bias the size of the TFP estimate but not the distribution of tfp overall. However, the consequences of estimating production functions with firms operating sub-optimally is not clear and is left as possible future research. The dominant trade theory at the firm level (??) predicts that, within each industry, the firms at the upper distribution of productivity above a certain cut off will export while those that are below will produce only for the domestic market. Trade liberalization induces a self-selection of the most productive firms into exports. However, this self selection is not evident in the data and it does not appear that export firms are necessarily the most productive. Figure 5 on the next page shows the distribution of productivity for exporters and non exporters in each industry.20 If there is evidence of self selection, the productivity levels of exporting firms 19 TFP was also calculated using OLS, RE, FE and using the ? method. The OLS estimate suffer from endogeneity issues and both fixed effects and random effects estimates relies on strong assumptions on ωit and has not worked well in practice. The LP estimates relies strictly on specific assumptions on the data generating process of the inputs to production that may not hold generally (see ?) and is prone to suffer collinearity. See Appendix A.1 on page 25 20 Furthermore, Table 12 in Appendix A.2 shows that the median productivity level do not substantially differ between exporters and non-exporters and, in fact, shows that mean productivity is higher for domestic firms than for exporting firms in all sectors except for Leather and Chemical. 15 Figure 5: Distribution of Productivity and Export Status Source: Bangladesh 2002 ICA, World Bank. Productivity as measured by estimated total factor productivity using ?. See A.1 on page 25 For table of inter quartile range would be within a range that is distinctively higher than the range of non exporting firms. However, the productivity range of exporters overlap non-exporters in each industry. The distortion to self selection due to financial constraints as articulated by the working capital model (see Figure 4) in fact provides an explanation to the non correspondence seen in in the date (Figure 5). 5.1.2 Differences between Exporting and Non-Exporting Firms: Working Capital, Physical Capital and Size Table 3 examines firm characteristics that differentiate exporters and non-exporters. Along with the estab- lished differences in size and age21 , the difference in means between exporting and non-exporting firms in output, physical capital and working capital are also examined. Two main patterns emerge: First, the direction of the differences between exporting and non-exporting firms are not the same across all sectors. Exporting firms are on average older than non exporting firms except for Electronics and Chemicals sector. The same reversal of the difference is seen in output, physical capital, physical capital per worker and working capital per worker.22 This suggests that between industries, age and physical capital may have different effects in determining whether a firm exports. Secondly, working capital and labor are the only two variables where exporters consistently have a higher mean than non-exporters across all sectors. The significance of the dif- ferences are even sharper when samples are taken according to productivity cutoffs and the direction of the differences remain unchanged. 5.1.3 Access to Finance: Non-exporters are more financially constrained than exporting firms and credit is not necessarily allocated according to productivity. Under perfect financial markets, firms are able to borrow to facilitate production and to overcome the time delays in the receipt of revenue. However, when access to credit is limited, the amount of cash on hand will 21 Differences in firm size and age has been shown to be persistent in different empirical studies both of the US and in developing countries (??). 22 While this may not be the case for physical capital, physical capital per worker and working capital per worker if we considered Textiles as an abnormality, for age and output this remains true. 16 Table 3: Differences between non-exporters and exporters by sector Garments Textiles Non-Exporters Exporters Total diff t-stat Non-Exporters Exporters Total diff t-stat Age 7.311 7.465 7.442 -0.154 -0.302 9.516 11.53 9.979 -2.014 -2.470 Log Output 11.41 11.26 11.28 0.15 1.830 11.14 11.05 11.12 0.09 0.633 Number of Workers 418.5 471.1 463.1 -52.6 -1.099 340.5 405.7 355.9 -65.2 -1.329 Log Physical Capital 8.771 9.19 9.126 -0.419 -3.759 10.46 9.894 10.33 0.566 3.531 Log Working Capital 8.974 9.107 9.087 -0.133 -1.163 9.458 9.532 9.475 -0.074 -0.469 Food Leather Non-Exporters Exporters Total diff t-stat Non-Exporters Exporters Total diff t-stat Age 15.31 33.37 17.74 -18.06 -4.071 15.36 18.3 17.38 -2.94 -2.173 17 Log Output 9.884 11.07 10.04 -1.186 -4.223 10.01 11.33 10.92 -1.32 -6.630 Number of Workers 126.4 426.9 167.3 -300.5 -3.839 107.8 189.7 164.8 -81.9 -3.013 Log Physical Capital 8.439 10.49 8.715 -2.051 -7.953 9.027 10.42 9.985 -1.393 -5.156 Log Working Capital 7.655 10.08 7.983 -2.425 -8.876 8.605 9.786 9.413 -1.181 -4.810 Electronics Chemicals Non-Exporters Exporters Total diff t-stat Non-Exporters Exporters Total diff t-stat Age 13.18 12 13.15 1.18 0.402 20.38 18.97 20.17 1.41 0.520 Log Output 9.678 11.21 9.719 -1.532 -4.311 10.41 12.58 10.78 -2.17 -7.718 Number of Workers 100.8 198.8 103.5 -98 -0.765 159.5 636 241.3 -476.5 -5.319 Log Physical Capital 8.137 9.371 8.17 -1.234 -6.922 9.505 11.82 9.911 -2.315 -8.790 Log Working Capital 7.886 10.06 7.943 -2.174 -3.850 9.046 11.64 9.501 -2.594 -8.841 Source: Bangladesh 2002 Investment Climate Assessment Survey, World Bank Table 4: Indicators of Financial Constraints Non-exporting Exporting Total Access to Finance: Moderate to Severe 0.6643 0.6095 0.6407 (2-4) Problem Access to Finance: Major to Severe 0.4506 0.3619 0.4125 (3-4) Problem Finances for Investment 100% 0.262 0.1887 0.2305 internally financed∗ Finances for Working Capital 100% 0.1907 0.1321 0.1655 internally financed∗ Overdraft facility or line of credit 0.6715 0.6506 0.6626 If yes, % currently unused 22.31 30.36 25.63 Source: Bangladesh Investment Climate Assessment survey: 2002, World Bank. Note: All variables are indicator variables except for percent of credit line unused. ∗ 100% internally financed and also reported access to finance as a problem. affect the firm’s decision to enter international markets where the time delays are much longer than domestic sales. Table 4 compares financial indicators between exporting and non-exporting firms. Access to finance is a subjective variable where respondents are asked to rank the problem from 0, being no problem to 4 as a severe problem. There is a smaller percentage of exporters who reported access to finance as a problem than the percentage of non-exporters. Furthermore, non-exporters report greater severity of the problem. Respondents are also asked where they sourced their financing for investment and for working capital and report the percentage from each category.23 Exporters source a smaller percent of their finances internally for both investment and working capital; indicating that exporters do indeed have better access to credit. The last financial indicator is whether the firm has an overdraft facility or a line of credit available to them, and if yes, to report the percent of the line unused. Here, a larger proportion of non-exporters than exporters have overdraft facilities or credit lines however, out of those that do have credit lines, the average percent of credit lines used by non-exporters is higher. This suggests that non-exporters are closer to their borrowing limit than exporters or that exporters generally do not use this type of credit as much. Overall, these statistics suggest that non-exporters have less access to finance than exporters. It is not only important to verify whether exporters have better access to finance than non-exporting firms but also to see whether access is allocated according to firm productivity. As discussed in the introduction, the results of existing theoretical models is driven by a direct correspondence between productivity and the amount of credit the firm has access to. It is this correspondence that supports the self selection of the most productive firms into exports even under financial distortion. Table 5 shows the average productivity ac- cording to each financial variable. There is no significant difference in average productivity between firms that report access to finance as a problem and between firms that fund investment wholly from internal fi- nance. Productivity is slightly higher for those who can borrow to fund working capital24 and those who have overdraft facilities. The difference in productivity however, does not appear large and the size of the stan- dard deviations imply considerable overlap in the TFP distribution between constrained and not constrained categories. Therefore there is no strong evidence to suggest that credit is allocated along productivity lines. 23 There are 14 categories the respondents are asked to choose to fill in: a. Internal funds or retained earnings b. Local commercial banks (loan, overdraft) c. Foreign-owned commercial banks d. Leasing or hire purchase arrangement e. Government subsidies f. Investment Funds/Special Development Financing/ Or Other State Services g. Trade credit (Supplier or customer credit) h. Credit cards i. Equity (Capital, sales of stock) j. Family, friends k. Informal sources (e.g. money lender) l By selling other assets m. Letters of credit n. Other (specify source): 24 That is, firms that do not fund working capital wholly internally. 18 Table 5: Productivity by Financial Constrained Indicator Variables Average Productivity Indicator variable= No Yes t-stat Access Mod-Severe 4.212 4.127 1.123 (1.15) (1.068) Access Major-Severe 4.18 4.126 0.7616 (1.131) (1.05) 100% Internally financed Investment 4.177 4.115 0.7433 (1.107) (1.105) 100% Internally financed Working Capital 4.193 4.012 1.906 (1.108) (1.09) Overdraft Facility/Credit line available 3.978 4.256 -3.8719 (0.9985) (1.151) Source: Bangladesh 2002 Investment Climate Assessment Survey, World Bank 5.2 Working Capital and Distance to Export Destination The 2002 Bangladesh data set is unique in that the survey asked firms to list their main export destinations and amount exported to each of these destinations for years 2000, 2001 and 2002. From this, a weighted export distance can be calculated for each firm. The availability of data on the distance of the export destination allows a simple test of the relationship between working capital and the export distance by exporting firms. This relationship is a key component of the model: the greater the export distance, the longer the delay in the receipt of payments and the greater the demand for working capital required.25 The empirical question is whether the amount of working capital, X , is significantly and positively cor- related with export distance among exporting firms controlling for factors that also determine the amount of working capital. If working capital does not vary with distance, than this puts into doubt whether the working capital is driven by delays in the receipt of revenues, or is simply a function of costs. The empirical specification is motivated by the determinants for the amount of cash required to export from Equation 13. Working capital demand increases with time delay S , cost of production Ct , expected cost of production Ct +1 (to ensure that costs are covered in the interim between production and when revenue is received upon deliv- ery) and with productivity A. The time delay, S , is proxied by the distance to export destination, distance. Whether the firm will export in the interim periods during the delays also affect how much working capital is needed. Therefore, working capital demands would also increase with expected relative foreign to domestic price, p. The stock of working capital kept on hand would decrease with supply of liquidity: the amount the firm can borrow which is measured as short-term liabilities, and past export production, proxied by years of exporting, years exporting.26 The empirical specification is as follows: Xt = α0 distancet +1 + α1Ct + α2Ct +1 + α3 At + α4 years exporting + α5 creditt + α6 pt +1 + εt + µi Distance enters the specification as one period ahead as the export distance is realized after the firm 25 See previous sections as well as refer to Table 2 on page 5 that shows when payments are made and Figure 1 on page 6 for transit times. 26 Financial constraint variable was not included as this would restrict the sample to one year, and cannot estimate the relationships. Furthermore, the discount rate β could be proxied by reported interest rates but interest rates were also only available for 2002. 19 Table 6: Working Capital and Distance to Export Destination Lagged Log Working Capital (1) (2) (3) (4) Log weighed distance to export 0.382 0.464* 0.668** 0.739** destination (0.237) (0.241) (0.338) (0.349) Controls for Cost, Productivity, Yes Yes short-term Liabilities, Years Exporting, Export Price to Domestic Price, Interact sectors*Log distance No No Yes Yes Observations 848 780 848 780 R-squared 0.007 0.091 0.017 0.099 Number of firms 459 438 459 438 Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1 Notes: 2002 ICA World Bank (with recall data from 2000, 2001.). Regressed using fixed effects for the sample exporting firms only. Distance data from Centre d’Etudes Prospectives et d’Informations Internationales (CEPII), at http://www.cepii.fr/anglaisgraph/bdd/distances.htm using distance measured from the capital city. Up to five countries were reported in the survey (only for 2001 and 2002) and the weighted average distance was calculated by the proportion of export revenue of each country. Productivity calculated from structural estimation of the production function (see ? ). None of the coefficients for the interactive terms for sector and log distance were significant and are not reported. Log Yrs Exporting is averaged across products. Full regression in Appendix 14 in Table 14 makes it’s export decision based on current working capital levels.27 The destination for exports would systematically differ according to industry, and the interaction between sectors and export distance were also included to improve fit. The specification was estimated using fixed effects and the results are shown in (the full set of estimates are shown in Appendix 14 in Table 14). Column (1) shows the estimation with no other control variables and there is no significant correlation between working capital and the distance to export destination. Column (2) includes controls for cost, productivity and other variables, and columns (3) and (4) include interaction between sectors and export distance. The estimated results shows that the correlation between working capital and export distance is positive and significant when relevant controls are included. Furthermore, the coefficients for the control variables: Lagged cost, current cost and productivity are correctly signed and significant (see in full estimate results in Appendix 14 in Table 14). The results show that the size of the estimated coefficient for distance is much larger than the estimated coefficients for cost and productivity. The distance variable may also be proxying for additional transport costs associated with shipping longer distances. However, the significant and positive coefficient suggests that the distance to the export destination could be the largest factor in determining the liquidity needs of the firm, and this could be due to the delays in the receipt of revenues associated with longer shipping times. 5.3 Export Status of the Firm The model in the Section 4 provides a set of theoretical predictions that could be put to the data: i) when firms are financially constrained, export status is driven by the level of working capital and less by productivity 27 This specification is an approximation given different industries will have different turnover time. The Bangladesh data has a majority of garment and textile firms reflecting the Bangladesh economy and the specification on the timing of working capital and export distance one year ahead fits this particular data set. Specification using the same time period, i.e. Xt and distancet yields a negative and insignificant coefficient - that supports the specification for distance as a forward variable. 20 (Areas I, II and IV in Figure 4 ) and ii) when firms are not financially constrained (Areas III, V and VI in Figure 4 ), export status of the firm is driven by productivity and changes in working capital does not change the export status of the firm. Thus the main test of the theoretical model is to see whether cash only affects export decisions when firms are financially constrained and at the same time, whether productivity affect export decisions when firms are not financially constrained. Let f c be a dummy that indicates whether the ¯ .28 29 The empirical specification is as follows: firm is financially constrained, that is whether Xi < X Ei = β0 f ci + β1 f ci ∗ Xi + β2 Xi + β3 f ci ∗ Ai + β4 Ai + controlsi The test involves four predictions on the estimated coefficients: a) the interaction between financial con- straints and working capital, β1 , is positive and significant, b) the coefficient for working capital only, β2 , should not be significantly different from zero, c) the coefficient for productivity when not constrained, β4 , should be positive and significant and d) the coefficient for productivity under constraints β3 is the negative of β4 and significant in order for productivity to not affect exports under financial constraints. That is: H0 : β1 > 0 β2 = 0 β3 = −β4 β4 > 0 The 2002 Bangladesh data set contains information on the percentage of sales that are exported. Out of the 974 firm observations, only 76 firms reported exporting less than 100 percent. The majority of firms report zero exports with about a third of the firms reporting 100 percent exports. The model does not offer any qualifications on how many firms in the population would export less than 100 percent. However, the fact that the model does predict that there will exist firms who will choose to export less than 100 percent, even in the presence of exporting fixed cost, is a unique and important characteristic of the model. The small number of observations of export percentage makes identification difficult and as a result, the dependent variable used is a export status dummy, E , that is equal to one if the firm engaged in any exports. The financial constraint variable, f c, is a dummy that is equal to one if the firm reports access to finance as a problem (from minor to major severity) and finances working capital entirely from internal finance (that is, does not use any borrowing from any other sources). The rationale behind interacting these two conditions is that firms may report access to finance as a problem when they actually do have access to borrowing. The measure does however, leave out firms that are able to borrow but remain financially constrained due to reaching borrowing limits or quotas. In this sense, it may not capture all possible financially constrained firms. Working capital, X , is measured according to the accounting definition: current assets (inventories, accounts receivable, cash and short-term credit) minus current liabilities (accounts payable and any short-term debt). Productivity, A, is measured as total factor productivity estimated according to ? and the production function is estimated for each sector.30 Controls for other variables in the model that were available from 28 Note that financial access questions are only available for 2002, and thus the number of observations available for estimation is 990. 29 See ? for further examination of the financial access variable including sources of variation that determines access to finance. 30 Estimates using Levinsohn- Petrin (LP2) does not change sign nor significance, but do to some extent change the size of the 21 Table 7: Export Status Estimation Results Probit Export Status (1) (2) (3) Financially Constrained -0.4867*** -0.420*** -0.314 (0.0917) (0.137) (0.202) FC*Log Working Capital 0.1070*** 0.101*** 0.0682** (0.0317) (0.0281) (0.0298) FC*Productivity -0.1284 -0.169* -0.126 (0.0937) (0.0966) (0.107) Log Working Capital 0.0162 0.0288** 0.0298** (0.0112) (0.0123) (0.0124) Productivity 0.3353*** 0.355*** 0.0621 (0.0357) (0.0477) (0.0419) Controls No Yes Yes Sector dummies No No Yes Observations 959 936 936 Pseudo Rsqr 0.149 0.261 0.391 Log pseudolikelihood -557.6 -472.7 -389.6 Source: Bangladesh 2002 Investment Climate Assessment Survey, World Bank Notes: Only 2002 data used for estimation, Bootstrap standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1 Reported marginal effects, Base line for Equation (3) is Garments. the data set are: export price to domestic price ratio, log labor cost, log weighted input price, age and sector dummies are also included (base sector as Garments).31 The results of export status is shown in Table 7 (full results in Appendix A.4 in Table 13). Results show that under financial constraints, working capital is significant and positive in determining export status of the firm, and this result remains robust even with additional controls across all three specifications. The effects of working capital, without the interaction with financial constraints, is small or insignificant in determining export status. This suggests that working capital increases the likelihood of a firm exporting mainly when the firm is financially constrained. A 10 percent increase in working capital raises the probability of exporting by over 10 percent for a financially constrained firm but by only about 3 percent for a firm that is not constrained. All estimates pass joint significance tests of β1 and β3 . A possible explanation as to why working capital for non financially constrained firms remains significant under additional controls, even though the size is small (in regressions (2) and (3)), could be due to the financial constraint measure not capturing all possible firms that are constrained. The estimates for productivity for all regressions are signed according to the model’s prediction, that is, β2 is negative and β4 is positive. However the size and significance of the coefficients vary across the 3 regressions. Joint significance test of β2 andβ4 can reject the null for specifications (1) and (2). In the specification (3), both coefficients for productivity are jointly insignificant when sector specific dummies are included. This could be due to the small sample size within each sector in the data set. However, in other empirical studies, such as ? where empirical specifications always include controls for sectors, they too find that the estimated coefficient for productivity is not significant and their sample contains 23,641 observations. This could be due to the fact that there is more variation in productivity between sectors than between firms within a sector, and this may be in part due to how productivity is estimated. Specification (3) show that coefficients. See Appendix for productivity measures. 31 See Appendix A.3 for table of variables used in estimation. 22 within sectors, productivity does not affect export status at all -regardless of whether firms are constrained or not. In the first and second regression, without sector dummies, the productivity coefficient is large and highly significant, in accordance to theory. The coefficients for the interacted productivity and financial constraint is substantially sized, and only significant in the regression with additional controls. The estimates with additional controls resonates more due to specification and better fit. The results from regression (2) suggest that productivity matter less to the export status of financially constrained firms than unconstrained firms, but the effect does not completely negate the effects of productivity. Looking at the estimates overall, results provide support for the model’s predictions that the effects of working capital and productivity will differ between financially constrained and unconstrained firms. In- tuitively, working capital matter more for financially constrained firms, and productivity matters less when firms are unconstrained. 6 Conclusion Empirical evidence suggests that, particularly in developing countries, financial factors matter for exporting. Extending firm trade models to include dynamic borrowing constraints points to the importance of working capital, as well as productivity for determining the decision to export, the percentage of output that will be exported and its destination. As such, the working capital model adds an important caveat to the established literature on the relationship between productivity and the export decision of the firm. Testing this model empirically with a unique data set from Bangladesh supports the proposition that working capital and pro- ductivity affect export status of the firm and their effects are differentiated by financial constraints. Empirical results also confirm the relationship between working capital and the distance to export destination. The em- pirical results provide support for the model’s predictions that the effects of working capital and productivity will differ between financially constrained and unconstrained firms. Specifically, for financially constrained firms, productivity matters less than for unconstrained firms while working capital matters much more. The correlation between working capital and export distance is shown empirically to be positive and significant, lending support for the working capital model of the firm’s export decision. There are many exciting directions future research in this area will need to explore, both at the micro and macro level. On a macro level, the partial equilibrium model of working capital of the firm can be incorporated into general equilibrium models to look at the propagation of the effects of financial constraints across the economy and across countries. Further research and exploration into the dynamic interaction between the financing needs due to time delays and the financial structure of countries has the potential to add insights to international trade patterns. The distortion that financial constraints introduce to the self-selection of the most productive firms into exporting warrants further investigation as to how this may affect overall industry productivity and thereby the aggregate growth of the economy. In terms of theories of development, if the results of the model are drawn analogously for an economy represented by a single firm, this could potentially prescribe a development path where trade liberalization may play a very minor role at low levels of wealth. This may provide further impetus for developing countries to move quickly in the development of functioning financial systems that allocate financial resources closely in line with productivity and the availability of short-term credit and trade credit so that the most productive firms enter exports, and aggregate industry productivity improvements can be reaped from trade liberalization. On the micro level, the working capital model exhibits distortions to factor ratios (also see ?) and this 23 could potentially lead to different sectors that export within an economy than those predicted by comparative advantage trade theories. Of particular significance is the need to develop theoretical models and empirical tools that will allow the analysis of the firm at sub optimal production levels, and this is crucial in the study of developing economies where financial constraints are the norm. The estimation of productivity under these conditions warrants urgent attention. References 24 A Appendix A.1 Production Function Estimation The production function is estimated using the ACF method32 , ?33 (LP) as well as OLS, fixed effects, random effects by industry. The OLS estimate suffer from collinearity and is shown as a comparison. Fixed effects and random effects estimates relies on strong assumptions on ωit and has not worked well in practice. The LP estimates relies strictly on specific assumptions on the data generating process of the inputs to production that may not hold generally (see ?) whereas the ACF estimates allows for more generally plausible assumptions as to when inputs are chosen, and that certain inputs, such as labor, is ’less variable’ than other inputs, such as materials. The ACF method is derived from ?. The production function is assumed to take the form: yit = βk kit + βl lit + ωit + εit Capital is decided at time t − 1. Assuming that labor is “less variable” than materials, suppose that lit is chosen at time t − b after kit was chosen at t − 1 but before mit was chosen at t . Assume that ωit evolved ac- cording to a first order Markov process between sub periods t − 1,t − b and t . Given these timing assumptions, the firm’s material input demand at t will depend on lit and kit chosen prior to period t : mit = ft (ωit , kit , lit ). Inverting this function will yield: yit = βk kit + βl lit + ft−1 (mit , kit , lit ) + εit ˆ it of the composite term Φt (mit , kit , lit ) = βk kit + βl lit + ft−1 (mit , kit , lit ) can be obtained at An estimate, Φ this first stage. This represents output net of εit . Given the first-order Markov assumption on ωit , ωit = E [ωit |Iit −1 ] + ξit = E [ωit |ωit −1 ] + ξit ξit is mean independent of all information known at t − 1. The first moment condition is derived on the timing assumption that kit was decided at t − 1 and thus uncorrelated with information set Iit −1 . E [ξit |kit ] = 0 The second moment conditions is derived on the timing assumption that lagged labor lit −1 was chosen at time t − b − 1 and is thus also in the information set Iit −1 . Thus we get two moment conditions to identify βk and βl : kit E [ξit | ]=0 lit −1 kit (14) E [ξit · ]=0 lit −1 ˆ it − βk kit − βl lit . Secondly, First, use OLS to estimate initial values of (βk , βl ) and compute: ωit (βk , βl ) = Φ non-parametrically regress ωit (βk , βl ) on lag ωit −1 (βk , βl ) and constant term to obtain ξit (βk , βl ). Finally, we 32 Stata code developed by ? 33 Utilizing the Stata program levpet developed by authors. 25 can obtain estimates (βk , βl ) from minimizing the sample analogue to the moment conditions in 14 using the implied ξit (βk , βl ). 1 1 kit T N ∑t ∑i ξit (βk , βl ) · lit Implementation with Bangladesh Data: In ?, the ACF method is implemented using capital, skilled labor and unskilled labor with electricity as the intermediate input/proxy. The Bangladesh data has labor, capital, raw materials and well as energy costs. The variability over time of each of these variables within firms in increasing order are: labor, capital, materials and energy. I estimate the production function using two specifications: 1) using materials as an intermediate input and 2) including materials as an input and using energy as the proxy: 1. yit = βk kit + βl lit + ωit + εit mit = ft (ωit , kit , lit ) 2. yit = βk kit + βl lit + βm mit + ωit + εit eit = ft (ωit , kit , lit , mit ) The estimates are given in Table 8 and Table 9 using the full panel Bangladesh data from 2003-2006 (semi-annual). The estimated coefficients for OLS and FE do not differ greatly in their magnitude. The estimates produced ACF1 using materials as an intermediate input yield unusually large coefficients and ACF2 using electricity as proxy appear to produce more realistic estimates. Average productivity calculated by each estimation method is reported in Table 10 and the correlation matrix is given in Table 11. The productivity estimate from RE is almost perfectly correlated with the OLS estimate; and the OLS estimate is the naive estimate that suffers from collinearity. Thus, ACF2 estimate and possibly LP2 may provide good proxy for productivity. 26 Table 8: Production Function Estimates by Sector Garments ACF1 ACF2 LP1 LP2 OLS FE RE lnl 0.925 0.311 0.227*** 0.0718 0.258*** 0.202*** 0.249*** . . (0.0532) (0.131) (0.0568) (0.0396) (0.0243) lnk 0.149 0.085 0.181*** 0.0413*** 0.0684*** 0.0167 0.0642*** . (0.0603) (0.0141) (0.0163) (0.0276) (0.0136) lnm 0.603 0.466*** 0.596*** 0.620*** 0.552*** 0.598*** . (0.109) (0.0269) (0.0275) (0.0193) (0.0135) Observations 861 861 861 861 861 861 861 # of firms 303 303 Textiles ACF1 ACF2 LP1 LP2 OLS FE RE lnl 0.413 0.245 0.327*** 0.399** 0.326*** 0.126*** 0.295*** . . (0.0438) (0.167) (0.0390) (0.0442) (0.0267) lnk 0.753 0.114 0.0536 0.0506* 0.0534** -0.0233 0.0868*** (0.0897) (0.0284) (0.0250) (0.0393) (0.0187) lnm 0.616 0.462** 0.653*** 0.652*** 0.290*** 0.620*** . (0.209) (0.0298) (0.0271) (0.0415) (0.0180) Observations 730 730 730 730 730 730 730 # of firms 250 250 Food ACF1 ACF2 LP1 LP2 OLS FE RE lnl 1.176 0.315 0.300*** 0.698** 0.345*** 0.421*** 0.363*** . . (0.0314) (0.335) (0.0311) (0.101) (0.0326) lnk 0.294 0.132 0.0236 0.0683*** 0.0708*** -0.0277 0.0725*** . (0.157) (0.0231) (0.0263) (0.0593) (0.0236) lnm 0.665 0.565*** 0.374 0.669*** 0.511*** 0.643*** . (0.207) (0.272) (0.0316) (0.0451) (0.0212) Observations 427 427 427 427 427 427 427 # of firms 147 147 Notes: *** p<0.01, ** p<0.05, * p<0.1 Standard errors in parentheses 27 Table 9: Production Function Estimates by Sector Leather ACF1 ACF2 LP1 LP2 OLS FE RE lnl 0.927 0.192 0.166*** 0 0.184*** 0.261 0.189*** . . (0.0556) (0.367) (0.0540) (0.186) (0.0617) lnk 0.306 -0.005 0 0.0247 0.0394 0.106 0.0467 . (0.222) (0.0352) (0.0314) (0.115) (0.0356) lnm 0.705 0.671*** 0.646*** 0.680*** 0.674*** 0.670*** . (0.0480) (0.0586) (0.0405) (0.0832) (0.0352) Observations 275 275 274 274 275 275 275 # of firms 99 99 Electronics ACF1 ACF2 LP1 LP2 OLS FE RE lnl 0.722 0.422 0.275*** 0.285* 0.279*** 0.0576 0.169*** . . (0.0383) (0.168) (0.0328) (0.0362) (0.0277) lnk 0.348 0.001 0.0496 0.00273 0.0168 -0.0454 0.0183 . (0.135) (0.0239) (0.0209) (0.0343) (0.0193) lnm 0.853 0.704*** 0.691*** 0.733*** 2.568*** 1.943*** . (0.212) (0.0308) (0.0253) (0.506) (0.182) Observations 268 268 268 268 268 268 268 # of firms 91 91 Chemicals ACF1 ACF2 LP1 LP2 OLS FE RE lnl 0.775 0.143 0.225*** 0 0.228*** 0.237*** 0.346*** . (0.0731) (0.260) (0.0652) (0.0845) (0.0579) lnk 0.528 0.082 5.84e-08 0.168*** 0.140** 0.321*** 0.278*** . (0.245) (0.0649) (0.0643) (0.0864) (0.0386) lnm 0.759 0.284 0.660*** 0.660*** 0.334*** 0.438*** . (0.193) (0.0744) (0.0564) (0.0277) (0.0268) Observations 222 222 222 222 222 222 222 # of firms 80 80 Notes: *** p<0.01, ** p<0.05, * p<0.1 Standard errors in parentheses Table 10: Productivity Measures: Summary Statistics Calculated TFP ACF1 ACF2 OLS FE RE LP1 LP2 Garment 4.546 2.300 2.567 4.090 2.894 3.359 4.155 (0.668) (0.316) (0.314) (0.344) (0.316) (0.363) (0.352) Textile 2.896 2.221 2.042 7.664 2.201 4.020 1.678 (0.816) (0.401) (0.391) (1.030) (0.395) (0.523) (0.399) Food 2.782 1.479 1.860 3.873 2.010 3.405 3.198 (1.488) (0.499) (0.486) (0.663) (0.489) (0.589) (0.780) Leather 3.684 2.825 2.680 1.729 2.688 3.247 4.001 (0.998) (0.506) (0.503) (0.532) (0.503) (0.512) (0.562) Electronic 4.140 0.589 2.054 2.568 1.946 2.060 2.517 (1.010) (0.492) (0.352) (0.463) (0.369) (0.356) (0.364) Chemical 1.833 1.781 1.781 3.156 2.037 6.882 2.585 (0.908) (0.486) (0.472) (0.682) (0.542) (1.129) (0.534) Total 3.502 1.999 2.220 4.540 2.397 3.684 3.060 (1.300) (0.724) (0.515) (2.114) (0.563) (1.212) (1.118) Observations 2786 2783 2783 2783 2783 2783 2783 Notes: Reported means with standard deviation in parentheses 28 Table 11: Correlation Matrix of TFP Measures ACF1 ACF2 OLS FE RE LP1 LP2 ACF1 1 ACF2 0.2486* 1 OLS 0.5947* 0.7498* 1 FE -0.1046* 0.2700* -0.0402* 1 RE 0.6124* 0.7743* 0.9582* 0.0344* 1 LP1 -0.2637* 0.3435* 0.0323* 0.2835* 0.1374* 1 LP2 0.6113* 0.3989* 0.7011* -0.5176* 0.7352* -0.0918* 1 Table 12: Interquartile Range and Median of Productivity by Industry Productivity:TFP 25th percentile Median 75th percentile Garments Non-Exporters 2.14 2.24 2.47 Exporters 2.09 2.24 2.46 Textiles Non-Exporters 2.06 2.2 2.34 Exporters 2.04 2.22 2.42 Food Non-Exporters 1.19 1.42 1.71 Exporters 1.02 1.35 1.62 Leather Non-Exporters 2.55 2.77 2.88 Exporters 2.68 2.83 3.03 Electronics Non-Exporters 0.31 0.517 0.892 Exporters 0.293 0.363 1.06 Chemical Non-Exporters 1.48 1.71 1.95 Exporters 1.48 1.75 2.18 29 A.2 Productivity Distribution Between Exporters and Non Exporters by Sector A.3 Variable Description Variable Description FC Indicator variable: Finances for Working Capital 100% internally financed and the firm access to finance as a problem (of any levels) lnwc Log Working Capital. Working capital as defined by accounting definition: current assets (inventories, accounts receivable, cash and short-term credit) minus current liabilities (accounts payable and any short-term debt) FC*Log Working Capital Interactive term: financial constraint and log working capital. Productivity: ACF Total factor productivity estimated according to ? FC*Productivity Interactive term: financial constraint and productivity total factor productivity estimated according to ? Export Price to Domestic Price ratio Ratio of export price to domestic price. For firms where ratio is missing, filled in with average ratio taken across year and sector. Log Labor Cost Total wage bill divide by the number of employees. Log Input Price Weighted average price of top three most important inputs. Age Years since establishment. A.4 Export Status A.5 Working Capital and Distance to Export Destination 30 Table 13: Export Status Estimation Results Probit Export Status (1) (2) (3) Financially Constrained -0.4867*** -0.420*** -0.314 (0.0917) (0.137) (0.202) FC*Log Working Capital 0.1070*** 0.101*** 0.0682** (0.0317) (0.0281) (0.0298) FC*Productivity -0.1284 -0.169* -0.126 (0.0937) (0.0966) (0.107) Log Working Capital 0.0162 0.0288** 0.0298** (0.0112) (0.0123) (0.0124) Productivity 0.3353*** 0.355*** 0.0621 (0.0357) (0.0477) (0.0419) Export Price to Domestic Price ratio 0.185*** -0.0506 (0.0207) (0.0984) Log Labor Cost -0.0479 0.00266 (0.0430) (0.0349) Log Input Price: Weighted 0.0163* 0.0320*** (0.00990) (0.0100) Age -0.000668 0.00246 (0.00142) (0.00160) Textile Sector -0.574*** (0.0881) Food Sector -0.519*** (0.0284) Leather Sector -0.333** (0.132) Electronics Sector -0.512*** (0.0375) Chemical Sector -0.461*** (0.0265) Observations 959 936 936 Pseudo Rsqr 0.149 0.261 0.391 Log pseudolikelihood -557.6 -472.7 -389.6 Source: Bangladesh 2002 Investment Climate Assessment Survey, World Bank Notes: Only 2002 data used for estimation, Bootstrap standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1 Reported marginal effects, Base line for Equation (3) is Garments. 31 Table 14: Working Capital and Distance to Export Destination Lagged Log Working Capital (1) (2) (3) (4) Log weighed distance to 0.382 0.464* 0.668** 0.739** export destination (0.237) (0.241) (0.338) (0.349) Lagged Log Total Cost 0.339*** 0.322*** (0.107) (0.108) Log Total Cost 0.233** 0.222** (0.0917) (0.0929) Lagged Total Factor 0.212* 0.205* Productivity (0.111) (0.112) Log short-term Liabilities -0.00735 0.00681 (0.116) (0.117) Log Yrs Export 0.160 0.174 (0.117) (0.120) Export Price to Domestic 0.0230 0.0229 Price ratio (0.0220) (0.0222) Interact sectors*Log distance No No Yes Yes Constant 5.971*** -1.877 4.267* -3.054 (2.112) (3.031) (2.479) (3.664) Observations 848 780 848 780 R-squared 0.007 0.091 0.017 0.099 Number of firms 459 438 459 438 Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1 Notes: 2002 ICA World Bank (with recall data from 2000, 2001.). Regressed using fixed effects for the sample exporting firms only. Distance data from Centre d’Etudes Prospectives et d’Informations Internationales (CEPII), at http://www.cepii.fr/anglaisgraph/bdd/distances.htm using distance measured from the capital city. Up to five countries were reported in the survey (only for 2001 and 2002) and the weighted average distance was calculated by the proportion of export revenue of each country. Productivity calculated from structural estimation of the production function (see ? ). None of the coefficients for the interactive terms for sector and log distance were significant and are not reported. Log Yrs Exporting is averaged across products. 32