Volatility, financial constraints, and trade

Transcription

1 Volatility, financial constraints, an trae by Maria Garcia-Vega Dep. Funamentos el Analisis Economico I, Faculta e CC. Economicas y Empresariales, Campus e Somosaguas, 28223, Mari, Spain an Alessanra Guariglia *+ School of Economics, University of Nottingham, University Park, Nottingham NG7 2RD, Unite Kingom Abstract We construct a ynamic monopolistic moel with heterogeneous firms to stuy the links between firms iiosyncratic income volatility, the egree of financial constraints that they face, an their export market participation ecisions. Our moel preicts that more volatile companies are more likely to fail, nee to be more prouctive to stay in the market, an are more likely to enter export markets. A further implication is that through market iversification, exports ten to stabilize firms total sales. We test these preictions, using a panel of 9292 UK manufacturing firms over the perio he ata provie strong support to our moel. * Corresponing author: BA. + he authors are grateful for etaile comments on an earlier raft of this paper from Parantap Basu, Robert Hine, Maria el Pilar Montero, an participants at presentations at the European rae Stuy Group Financial support from the Leverhulme rust uner Programme Grant F114/BF is also gratefully acknowlege. 1

2 1. Introuction A number of recent stuies have focuse on the causes of the rise in firm level volatility, which characterize US firms in recent years. For instance, Comin an Philipon (2005) argue that this increase is primarily ue to more competition in prouct markets, which in turn might be ue to eregulation, increases in R&D investment, an higher use of ebt an equity. Other papers have focuse on the consequences of the rise in firm level volatility. For instance, Comin et al. (2006) ocument that the rise in firm turbulence explains about sixty percent of the rise in the high frequency volatility of wages. o the best of our knowlege, no stuy has looke at the effects of firm level volatility on firm survival an trae. his paper fills the gap. Specifically, we construct a ynamic monopolistic moel with heterogeneous firms to stuy the links between firms iiosyncratic uncertainty, the egree of financial constraints that they face, an their export market participation ecisions. Our moel can be seen as an extension of Melitz (2003) moel, in which firm level volatility is inclue as an aitional element of firm heterogeneity, in aition to prouctivity. Our moel preicts that more volatile companies are more likely to fail, nee to be more prouctive to stay in the market, an are less likely to enter export markets. A further implication is that through market iversification, exports ten to stabilize firms total sales. We test these preictions, using a panel of 9292 UK manufacturing firms over the perio he ata provie strong support to our moel. he rest of the paper is lai out as follows. In Section 2, we outline our moel, both in the close an in the open economy case. Section 3 escribes our ata. Section 4 tests the main implications of the moel, an Section 5 conclues. 2. he moel 2.1 Deman We assume that the economy has two sectors: one is characterize by a numeraire goo, an the other by ifferentiate proucts. he preferences of a representative consumer are given by the following intertemporal utility function: 2

3 rt ˆ U ( x 0 log Y ) e t 0, (1) where ˆr is the iscount factor, x 0 is the consumption of the numeraire goo, an Y is an inex of consumption of the ifferentiate proucts. As in Dixit an Stiglitz (1977), Y reflects the consumer s taste for varieties, an can be written as: 1 n Y (yz ) z an 0<<1 0 where y z is the quantity of variety z of the ifferentiate prouct emane by the consumer, n is the mass of firms in the stationary competitive equilibrium, an 1 (1 ) is the elasticity of substitution among varieties. As shown by Dixit an Stiglitz (1977), the consumer s behavior can be moele consiering the set of varieties of the aggregate goo Y consume, with aggregate price P Y n 0 p 1 z z. In this set-up, the aggregate eman for any of the varieties of the ifferentiate prouct is given by y p P Y E, where E is aggregate expeniture, an p is the price of the goo (see Grossman an Helpman, 1991, for a similar approach) Prouction We assume that firms are heterogeneous. Firms iffer in their prouctivity an in their income volatility. Firms have ifferent levels of prouctivity, enote by [0, ). As in Melitz (2003), prouctivity affects the firm s variable cost function. More prouctive firms have a lower marginal cost than their less prouctive counterparts. Each goo is prouce only with labour. he technology to prouce each of the ifferentiate goos is given by: l( ) I y, where l is labour, an I is a fixe cost in every perio. his leas to the stanar pricing rule p( ) 1 (. ), where the numerator represents the common wage rate that we normalize to 1. Firms with high levels of efficiency will charge lower prices an obtain higher revenues. 3

4 We introuce a new element of heterogeneity among firms: their iiosyncratic income volatility, enote with [0, ). Volatility is relate to an exogenous eman firm-specific shock that firms face each perio, immeiately after they have chosen their prices an quantities, an prior to their ecision problem of the next perio. Specifically, in each perio, the firm s income is given by z ( ) p( ) y( ), where z t is the eman shock, which is istribute normally with mean one an stanar eviation. Note that if 0 for all firms, the moel specializes to Melitz s (2003) case. We consier that prouctivity an income volatility are both exogenous, inepenent, an constant parameters. his assumption implies that neither prouctivity nor income volatility are ecision variables for the company 1, an that, as in Hopenhayn (1992), in the competitive equilibrium, prices are eterministic. Before entry, firms raw their initial prouctivity g(), an volatility h() from a common istribution with continuous cumulative istribution given respectively by G() an H(). t 2.3. he firm s problem We assume that the firm faces a cash-in avance constraint: in each perio, it requires the bank to finance its fixe cost 2. Companies repay the loan at an interest rate enote with r, that iffers among firms epening on the risk associate with the investment. As in Cooley an Quarini (2001), we assume that there are only one perio ebt contracts signe with the bank. If the firm efaults on the ebt, the bank liquiates the firm, an the company immeiately exits the inustry. he firm s profits can be written as: y( ) t (,, zt ) zt p( ) y( ) 1 r(, ) I (2) 1 In other moels, prouctivity an volatility are consiere as enogenous variables. For instance, in Comin an Mulai (2005), firms choose how much to invest in R&D, an because R&D investments are risky, they increase both the firm s prouctivity an the volatility of its prouctivity. Since our analysis only focuses on income volatility, it is reasonable to assume that both prouctivity an volatility are exogenous. 2 We assume that the fixe cost of prouction is ientical among firms. herefore, in the moel, the firm oes not choose its optimal amount of ebt. In orer to simplify the moel, we consier that ebt is the same for all firms an that it is inepenent of their prouctivity. 4

5 where (,, z ) represents the profit of a firm of type (, ), at perio t after the t t realization of its shock z t. We assume that risk-neutral banks perfectly observe the firm s characteristics. As in Cooley an Quarini (2001), we consier that banks issue funs at an interest rate r, such that the expecte repayment from the loan is equal to the repayment of a riskless loan. his is summarize in the following Equation: (1 r ) I (1 r(, )) I (1 (,, r)) C( ) (,, r) where (1 r) I C. (3) 0 he left-han sie of Equation (3) gives the return of the loan at the riskless interest rate r 0. he right han sie of the equation says that with probability 1 the company can repay its ebts, an with probability, it goes bankrupt. In the case of bankruptcy, the bank gets the firm s collateral (C), which epens negatively on the firm s income volatility 3. istribution: he firm s probability of bankruptcy, is given by the following cumulative 2 z(, r) z(, r) ( z1) (4) (,, r) f ( z, ) z e z 2 he probability of bankruptcy given by Equation (4) involves a threshol shock, enote by z(, r) 1. We assume that a firm will exit the market as soon as it experiences an income shock below the threshol. We assume that the threshol shock solves the y( ) following equation: zp( ) y( ) (1 r) I 0. 3 It is assume that the company efaults an goes bankrupt if it is unable to repay its loan. his hols inepenently on the amount of the loan. here is therefore no possibility to renegociate the loan. However, the bank is more likely to recover a higher part of the loan when the firm efaults, if the loan is relatively small. Since more volatile firms are more likely to obtain more extreme returns, when they go bankrupt, it is more likely that they ask for larger loans. For this reason we assume that from the point of view of the bank, the collateral is a negative function of the company volatility. 5

6 he next proposition, prove in Appenix 1, gives the relationship between the firm s volatility an its probability of bankruptcy. Proposition 1: Controlling for prouctivity, the collateral to ebt ratio, an the interest rate, firms with high iiosyncratic income volatility have a higher probability of bankruptcy than firms with low volatility. Consequently, it is more costly for these firms to obtain external finance. he moel suggests that firms characterize by higher income volatility will have a higher probability of bankruptcy. his channel rives to an increase in the costs of accessing external funing. his implication is consistent with the empirical result of Minton an Schran (1999), who fin that cash flow volatility is positively relate with the costs of accessing external capital. As in Hopenhayn (1992) an in Melitz (2003), in every perio an incumbent firm makes two ecisions: whether to leave the inustry or to stay, an in the latter case, how much to prouce. he value of a firm of type (,) at perio t after the realization of its shock in the stationary competitive equilibrium is enote with v (,, z ), an is given by: t t vt (,, zt ) t (,, zt ) (1 (, )) max 0, Et[ vt 1] 1 (,, ) (1 ((, )) (,,1) (,, ) (,,1). (5) (, ) st t zt s t zt st1 he firm will stay in the inustry if its expecte future value is positive, in the stationary competitive equilibrium, i.e. if Et[ vt 1] 0. his is equivalent to saying that its expecte future profits in each perio are positive, i.e. that (,,1) 0. 4 Let min be the level of prouctivity for which the firm obtains zero profit when its volatility is zero 5, an for which the income shock is equal to one. he following conition, prove in Appenix 1, 4 In orer to simplify the notation, we write (, ) instea of (,,1). 6

7 shows the characteristics in terms of prouctivity an volatility of the firms that can obtain a loan an therefore stay in the market. Zero-profit conition: For any level of prouctivity [, ), there exits a function s satisfying s( ), such that (, s( )) 0. hus, if s( ), the firm cannot obtain a loan an exits the market; while if s( ), the firm remains in the market. min Graphically the zero-profit conition can be represente as in Figure 3. his leas to the following Proposition: INSER FIGURE 3 HERE Proposition 2: Firms characterize by high income volatility nee to have a high level of prouctivity; otherwise they are force to exit the market. However, there is no correlation between volatility an prouctivity for low levels of volatility. Firms with low volatility can have either a high or a low prouctivity level an stay in the market. he intuition behin Proposition 2 is that volatility is costly. herefore, only firms with high prouctivity levels can counter the high cost associate with volatility, an obtain on average positive profits. a Let us enote with f e the entry cost, an with E t [ vt 1] the expecte ex-ante firm value which is given by E [ v a t s t Pin 1] (1 ) s, where in st1 P is the ex-ante probability of successful entry, is the average probability of bankruptcy, an is the average exante profit 6. New firms will enter in the inustry until the expecte ex-ante firm s value 5 In this case, the interest rate that the company has to pay is the riskless interest rate an the moel specializes to Melitz s moel. 6 he ex-ante probability of successful entry is equal to min min min ) P Pr ob( ) Pr ob( f ( ) / ) 1 G( 1 H ( s( )) in. 7

8 net of the entry cost is zero (Free entry conition). he following Equation will therefore hol: E [ v ] f 0. a t t 1 e (6) As in Melitz (2003), a stationary equilibrium is efine by constant aggregate variables over time an free entry for firms into the inustry. We show in Appenix 1 that such an equilibrium satisfies the zero cut-off profit conition, the free entry conition, an an aggregate stability conition that requires that the mass of successful entrants in each perio exactly replaces the mass of incumbents who are hit by a ba shock an exit. In the equilibrium the istribution of firms can therefore be represente as follows: g( ). h( ) if min, an s( ) (, ) Pin 0 otherwise Open economy moel Effects of trae on the probability of bankruptcy We now assume that there are two ientical countries that trae the varieties of Y. rae involves two types of costs. Firstly, there is a sunk entry cost into the foreign market, As in Melitz (2003), we assume that every perio, the firm pays the amortize per-perio portion of this cost, enote by x f e. x I. As in the close economy framework, every perio, exporters borrow the fixe cost from the bank. Seconly, there is a variable cost per unit of prouct that is transporte. As typically assume in the literature, this variable cost takes the form of an iceberg cost, so that for one unit of a goo to arrive to the final he average probability of bankruptcy is given by z ( ) 1. e 2 volatility, an. g( ) 2 ( z 1) 2 2 z, where is the average 1 G( ) 1 H ( s( )) is the average prouctivity. min s( ) g( ) h( ) 0 8

9 estination, 1 units of the goo nee to be shippe. While prices in the omestic market (enote with p ( ) 1 ) are the same as before, exporters set higher prices in. the foreign market, ue to the increase in the marginal cost. hese are given by: x p ( ). he profits of an exporter in the omestic an the foreign market are. respectively given by: x y x x x x y x t zt p y (1 r ) I, an t zt p y (1 r ) I, where r is the interest rate pai by an exporter on its total loan, I I x, an x z t is the income shock in the foreign market. We assume that the shock in the foreign market is normally istribute with mean one an variance the omestic shock, has mean one an variance x. As in the close economy moel,. When firms export, their probability of bankruptcy an their access to external funs change. If national an international shocks are correlate, we can aggregate both shocks an moel them as a total income shock that is normally istribute with variance (given by the sum of the national an international income variances plus two times the covariance between the two shocks). he probability of bankruptcy of an exporter is therefore given by the cumulative normal istribution shown in the following expression (which has mean two an total variance ): z (, r ) z (, r ) 2 ( z2) 2 2( ) 1 (,, r ) f z e z 2. (7) otal volatility ecreases with trae if the covariance between shocks is negative an larger in absolute value than x 2 ( ) / 2. In such case, accoring to Equation (7), the probability of bankruptcy involves a threshol shock z (, r ) that solves the following 9

10 x x x x equation: z ( p y p y ) ( y y ) / ( I I )(1 r ) 0. he threshol shock of an exporter is lower than that of a non-exporter if x I (1 r ) I 1 (r r ) (1 r) ( 1)y, i.e. if the fixe cost involve in exporting is not too high compare with the ecrease in the fixe cost in the national prouction, an the increase in sales. If the above conition hols, an the total income variance of a company when it exports is lower than its omestic income variance, then the probability of bankruptcy of exporters will be lower than that of non-exporters 7. he following Proposition summarizes the impact of volatility on the cost of trae: Proposition 3: If the volatility shocks in the national an international market are negatively correlate 8, trae can lea to a reuction of total income volatility, through market iversification. rae can therefore reuce the firm s probability of bankruptcy if the firm iversifies its total income volatility, an if the fixe cost to export is not too high. he intuition behin Proposition 3 is that trae can reuce financial risks but only if the income shocks are negatively correlate, an if it is not too costly to export compare to proucing only omestically. Uner these circumstances, by exporting the firm can reuce its probability of bankruptcy. Even if its probability of bankruptcy is lower, this oes not mean that exporters obtain a lower interest rate than non-exporters. he bank charges in fact an interest rate r to exporters, such that the expecte repayment of the loan equals the repayment of a riskless loan, solving the following Equation: x x (1 r )(I I ) (1 r )(I I ) (,,r ) C( ).(1 (,,r )). 0 7 he probability of bankruptcy is lower because both the threshol shock an the total income volatility ecrease uner these two assumptions. hese two conitions are sufficient conitions but not necessary conitions for trae to reuce the firm s probability of bankruptcy. A less restrictive necessary conition is: z z f ( z) f ( z) f ( z) i i x z z i 8 Hirsch an Lev (1971) also foun that exports ten to stabilize firms sales through market iversification. 10

11 Having to borrow more, exporters might have a lower ratio of collateral over ebt than non-exporters. rae has therefore an ambiguous effect on the exporters cost of external financing as it both ecreases their probability of bankruptcy (uner the assumptions of Proposition 3), which lowers the interest rate; while simultaneously increasing the ebt, which raises the interest rate. Greenaway et al. (2006) have shown empirically that trae improves the financial position of firms. Since collateral is a ecreasing function of volatility, it is plausible that exporters systematically have higher collateral than non-exporters. hen by Proposition 3, income iversification woul ecrease the exporters interest rate, an trae woul unambiguously reuce their costs of accessing external finance, improving their financial situation. his argument coul provie a theoretical founation for the outcome escribe in Greenaway et al. (2006). Volatility an the incentive to trae We now turn to a relate question: o firms with high national income volatility have more incentives to trae than firms with low national income volatility? Denote with (, ) the profits of exporters, with (, ) the profits of non-exporters, with the interest rate of exporters, an with profit ifferential is given by the following Equation: r the interest rate of non-exporters. hen the r (, ) (, ) (1 r ).I (1 r ).I p y x x x y (1 r ).I x. (8) Firms with high national income volatility have more incentives to trae than firms with low volatility if the profit ifferential is increasing with omestic income volatility. Differentiating Equation (8) with respect to omestic volatility leas to: ( ) r I. x r (I I ). his implies that firms with high national income volatility have more incentives to trae than firms with low national income volatility if the following inequality hols 11

12 r I x r (I I ). (9) In the absence of trae, an increase in volatility leas to an increase in the interest rate, r 0, as shown in Proposition 1. However, this is not necessarily the case with trae if there is iversification in income shocks. We can observe two cases when firms export. First, if the total interest rate oes not change for ifferent egrees of national income r volatility (i.e. if 0 ), then inequality (9) hols an firms with a high omestic income volatility have more incentives to trae than firms with low omestic income volatility. In this case, firms with high national income volatility are able to iversify their income shocks relatively more than low volatility firms. Exporters an nonexporters en up therefore having a similar total volatility, inepenently of the volatility of their omestic income volatility. For this reason more volatile firms have more incentives to trae. Secon, even if the interest rate rises with omestic volatility for r exporters (i.e. if 0 ), when there is iversification, the increase in the interest rate associate with increases in volatility is lower for exporters than for non-exporters (i.e. r r ). If this effect is sufficiently large as to counter balance the rise in the fixe cost necessary to export (i.e. if r r I I I x ), then firms with high omestic income volatility have more incentives to export than firms with low omestic income volatility. Proposition 4 summarizes the relationship between national income volatility an incentives to trae: Proposition 4: Firms with high national income volatility have more incentives to trae than firms with low national income volatility provie that when exporting, their interest rate ecreases more than proportionally to the increase in the fixe costs of prouction. 12

13 Intuitively, before exporting, firms with low omestic income volatility can obtain a loan at a lower cost than firms with high omestic income volatility. his generates incentives for firms with low volatility to start exporting. However, for highly volatile firms, the prospective reuction in the total volatility is larger than for firms with low volatility, which in turn reuces their interest rates an encourages them to trae. Effects of a trae agreement lowering trae barriers between countries A trae agreement that reciprocally lowers trae barriers between countries can, on the one han, reuce the variable costs of trae, an on the other, ecrease the sunk entry cost. A ecrease in the variable trae cost reuces the probability of bankruptcy of the exporters by ecreasing the threshol shock. his leas to a ecrease in the interest rate. he moel suggests that the reuction in the cost of accessing external funing can inuce a reuction in the firm s financial constraints. he ecrease in the sunk costs has two effects. First, as in the case of a reuction in the variable cost to trae, it reuces the firm s probability of bankruptcy, leaing to a ecrease of the firm s interest rate. Secon, it increases the incentives of high volatile firms to trae by reucing the left han sie of inequality (9). Uner the assumptions of Proposition 4, more volatile firms ten to trae more than firms with low volatility, when there is iversification of income shocks: trae reuces in fact the total volatility of these firms. he ecrease in the total volatility leas to an increase in the expecte profits for the potential entrants, since both the average inustry volatility an the average probability of bankruptcy ecrease. As in Melitz s (2003) moel, the change in the expecte profits affects the free entry conition an leas to a stationary equilibrium with less firms near the exit function s (). his mechanism, that is similar to a ecrease in the fixe entry cost in Hopenhayn s (1992) moel, prouces a subsequent ecrease in the exit function s (), which, as shown in Figure 5, implies that in the new equilibrium, the inustry has on average a higher level of prouctivity an a lower level of average volatility. INSER FIGURE 5 HERE 13

14 2.5 estable implications From the analysis of our theoretical moel, several testable implications emerge. First, the moel suggests that controlling for prouctivity an collateral over ebt, firms with high volatility have a higher probability of bankruptcy. It is consequently more costly for them to obtain external finance. Secon, the moel implies that there is a positive correlation between volatility an prouctivity for high levels of volatility, while there is no correlation for low levels. Highly volatile firms nee in fact to be more prouctive to stay in the market. hir, the open economy moel suggests that, through market iversification, exports ten to stabilize firms total sales. Exporters total income volatility will therefore be lower than their national income volatility. Finally, firms characterize by high national income volatility will have more incentives to start exporting than firms with low volatility. In the Sections that follow, we will test these implications using a panel of 9292 UK firms over the perio Data an summary statistics 3.1 he ataset We construct our ataset from profit an loss an balance sheet ata gathere by Bureau Van Dijk in the Financial Analysis Mae Easy (FAME) atabase. his provies information on companies for the perio It inclues a majority of firms which are not trae on the stock market, or are quote on other exchanges such as the Alternative Investment Market (AIM) an the Off-Exchange (OFEX) market 10. Unquote firms are more likely to be characterize by averse financial attributes such as a short track recor, poor solvency, an low real assets compare to quote firms, which are typically large, financially healthy, long-establishe companies with goo creit ratings. he firms in our ataset operate in the manufacturing sector. We exclue companies that change the ate of their accounting year-en by more than a few weeks, so that ata refer to 12 month accounting perios. Firms that i not have complete 9 A maximum of ten years of complete ata history can be ownloae at once for each firm. 10 We only selecte firms that have unconsoliate accounts: this ensures the majority of firms in our ataset are relatively small. Moreover, it avois the ouble counting of firms belonging to groups, which woul be inclue in the ataset if firms with consoliate accounts were also part of it. 14

15 recors on variables use in our regressions were also roppe. Finally, to control for outliers, we exclue observations in the 1% tails for each variable 11. Our panel therefore comprises a total of annual observations on 9292 companies, covering the years It has an unbalance structure, with an average of 7 observations per firm. By allowing for both entry an exit, the use of an unbalance panel partially mitigates potential selection an survivor bias. 3.2 Summary statistics Summary statistics of the main variables use in our empirical analysis are presente in able 1. Column 1 refers to the entire sample; column 2 an 3, to surviving an faile firms, respectively. As in Bunn an Rewoo (2003), we efine a firm as faile (bankrupt) in a given year if its company status is in receivership, liquiation, or issolve 12. Columns 4 an 5 of able 1 refer respectively to low an high volatility firms; an columns 6 an 7, to non-exporters at time t-1 that entere an i not enter export markets at t. INSER ABLE 1 HERE As in Comin et al. (2006) an Comin an Philippon (2005), our main volatility measure is calculate as the stanar eviation of the firm s total real sales growth, measure over a rolling winow of 5 years. Specifically, enoting with otalvol it this stanar eviation for firm i at time t; with srgr it, the growth rate of the real sales of firm i at time t, an with it, the average growth rate between t-2 an t+2, we have: 1 otalvol it = srgr i ( t ) it 1/ 2 11 hese cut-offs are aime at eliminating observations reflecting particularly large mergers, extraorinary firm shocks, or coing errors. See Appenix 2 for more information on the structure of our panel an complete efinitions of all variables use. 12 Liquiation an receivership are two types of reorganization proceures, which can take place when a company becomes insolvent. In liquiation, the assets of the company are sol so as to meet the claims of creitors. In receivership, the receiver can ecie whether it is in the creitors interests to sell the company s assets. Generally, it is in the creitors interests to liquiate if the liquiation value of the company excees its going concern value (Lennox, 1999b). 15

16 We also provie measures of volatility only base on national sales growth an overseas sales growth, which we enote respectively with Nationalvol it an with Overseasvol 13 it. Comparing total sales growth volatility at faile an surviving firms (columns 2 an 3), we can see that the former isplay a higher volatility (0.207) than the latter (0.199). he ifference between the two figures is marginally statistically significant (tstatistic: 1.69). In accorance with the first testable implication of our moel, there is some evience that faile firms are more volatile than their surviving counterparts. More formal tests of this hypothesis will be provie in the section that follows. Focusing now on columns 2 an 3, with emphasis on prouctivity (FP), which is calculate using the Levinsohn an Petrin (2003) metho 14, it appears that both high an low volatility firms isplay very similar levels of prouctivity (5.820 an 5.826, respectively). Yet the correlation between FP an volatility is positive for highvolatility firms (0.0650) an negative for low-volatility firms ( ). his seems to support our moel s secon preiction, accoring to which one shoul observe a positive correlation between iiosyncratic volatility an prouctivity for high levels of volatility only. Next, in relation to our open economy moel s preictions, we compare the firm s total, national, an overseas sales growth volatility, base on the entire sample (column 1). We can see that, as suggeste by our moel, the volatility of total sales growth (0.1999) is lower than that of national sales growth (0.238). he ifference between the two means is strongly significant (t-statistic = 29.86). Also consiering that overseas sales isplay the highest volatility (0.482), this provies some preliminary support for the hypothesis that through market iversification, exports ten to stabilize total sales It shoul be note that given the way in which we calculate volatility, this variable is not available for the years 1993, 1994, 2002, an For this reason, all regressions which contain our main measure of volatility are base on the sample A key issue in the estimation of prouction functions is the correlation between unobservable prouctivity shocks an input levels. Profit-maximizing firms respon to positive prouctivity shocks by expaning output, which requires aitional inputs; an to negative shocks, by ecreasing output an input usage. Olley an Pakes (1996) estimator uses investment as a proxy for these unobservable shocks. his coul cause problems as any observation with zero investment woul have to be roppe from the ata. Levinsohn an Petrin (2003), by contrast, introuce an estimator which uses intermeiate inputs as proxies, arguing that these (which are generally non-zero) are likely to respon more smoothly to prouctivity shocks. 15 If we limit the sample to exporters, the volatility of total sales growth is given by 0.198; that of national sales growth, by 0.248; an that of overseas sales growth, by

17 Finally, focusing on columns 6 an 7 of able 1, we can see that starters isplay a much higher national sales volatility compare to non-starters (0.309 versus 0.193). he ifference between the two means is statistically significant (t-statistic = 7.63). Although this comparison is simply base on unconitional means, it provies some strong support for our moel s last testable implication. More formal tests of all the implications of our moel are provie in the next Section. 4. Specifications an results 4.1 Are more volatile firms more likely to go bankrupt? In orer to test the first implication of our moel, namely that more volatile firms are more likely to fail, we will estimate a ranom-effects Probit specification of the following type: Pr(FAIL it =1) =( a 0 + a 1 size it + a 2 age it + a 3 group i + a 4 tfp it + + a 5 Collateral it /Debt it *+ a 6 otalvol+ u i + u j + u t ) (10) FAIL it is a ummy variable equal to 1 if firm i faile in year t, an 0 otherwise. (.) enotes the stanar normal istribution function. As typically one in the literature (see for instance Bunn an Rewoo, 2003, an Disney et al, 2003), our Equation controls for firm s size, age, prouctivity, an for whether the firm is part of a group. In accorance with our moel we also inclue the firm s collateral to ebt ratio an the volatility of its total sales growth among the regressors. Since the average length of time between the final annual report of a failing company an its entry into bankruptcy is usually 14 months (Lennox, 1999a), our specification inclues regressors evaluate at time t. Yet, all our results were robust to using lagge regressors. u t, represents a full set of time ummies aime at accounting for business cycle effects; an u j, a full set of inustry ummies Firms are allocate to the following inustrial groups: metals an metal goos; other minerals, an mineral proucts; chemicals an man mae fibres; mechanical engineering; electrical an instrument engineering; motor vehicles an parts, other transport equipment; foo, rink, an tobacco; textiles, clothing, leather, an footwear; an others (Blunell et al., 1992). 17

18 INSER ABLE 2 HERE he estimates of Equation (10) are presente in column 1 of able 2. As typically foun in the literature, size an FP have a negative effect on the firm s probability of failure. he coefficient associate with the ratio of the firm s collateral to total ebt is poorly etermine suggesting that this variable oes not play a statistically significant effect on firm survival. In accorance with Proposition 1 in our moel, the volatility of the firm s total sales growth is positively associate with the chances that the firm will go bankrupt: more volatile firms are therefore more likely to ie. As an alternative test of this first implication of the moel, we make use of the Quiscore measure prouce by Qui Creit Assessment Lt., which assesses the likelihoo of company failure in the 12 months following the ate of calculation. he lower its Quiscore, the more risky the firm, an the higher its chances of failure. We estimate the following fixe-effects regression: QUISCORE it = a 0 + a 1 size it + a 2 age it + a 3 tfp it + + a 4 Collateral it /Debt it + a 5 otalvol+ u i + u j + u t (11) he results are presente in column 2 of able 2. FP is positively associate with Quiscore, suggesting that more prouctive firms are less risky, an less likely to fail. Our collateral to ebt ratio is also positively associate with Quiscore, inicating that the more collateral a firm has relative to its total ebt, the less risky it is. Surprisingly, our size variable is negatively associate with Quiscore, suggesting that larger firms are more risky. Finally, our volatility measure isplays a negative coefficient: more volatile firms are more risky, an therefore more likely to fail, which is in accorance with our Proposition 1. As banks generally look at firms creit ratings such as Quiscore when eciing the terms of the loans they make to firms, it is likely that they will charge higher interest rates to the riskiest firms. hus, as preicte by our moel, more volatile firms are more likely to fail, an to be charge higher interest rates by their leners. 18

19 4.2 Is there a positive correlation between prouctivity an volatility for highly volatile firms only? he secon main implication of our moel is that there shoul be a positive correlation between prouctivity an volatility for highly volatile firms only. In orer to test this implication, we construct the following two ummies: LOWVOL it, which is equal to one if firm i s total real sales growth volatility in year t is in the lowest half of the istribution of the volatilities of all firms operating in the same inustry as firm i s in year t, an 0 otherwise; an HIGHVOL it, which is equal to one if firm i s volatility in year t is in the highest half of the istribution, an 0 otherwise. We then interact our volatility measure with the two ummies an estimate the following Equation, using a fixe-effects specification: FP it = a 0 + a 1 totalvol it *LOWVOl it + a 2 totalvol it *HIGHVOl it + u j + u t + e it (12) he coefficient a 1 can be interprete as the effect of volatility on FP for firms with low volatility; an a 2, as the effect for firms with high volatility. he estimates are reporte in column 1 of able 3. We can see that only a 2 is statistically significant. his suggests that volatility only affects the prouctivity of those firms characterize by a high volatility. his is consistent with our moel s secon preiction. Column 2 reports the results when firms are ivie in three categories base on their volatility: there are three interaction terms: one for low-volatility firms, one for meium-volatility firms, an one for highvolatility firms 17. Once again, there is a positive association between volatility an FP only for high-volatility firms, which supports our moel s secon preiction. INSER ABLE 3 HERE 4.3 Are more volatile firms more likely to start exporting? 17 In this case the interaction ummies are efine as follows: LOWVOL it is equal to one if firm i s total real sales growth volatility in year t is in the lowest 33% of the istribution of the volatilities of all firms operating in the same inustry as firm i s in year t, an 0 otherwise; MIDDLEVOL it is equal to one if firm 19

20 Lastly, our moel preicts that it is those firms that isplay highest volatility of national sales growth that shoul have more incentives to start exporting. In orer to test this preiction, we estimate the following ranom-effects Probit equation for the probability that a non-exporter at time t-1 becomes an exporter at t: SAR it = a 0 + a 1 size it + a 2 age it + a 3 tfp it + a 4 group i + +a 5 nationalvol it + u j + u t + e it (13) he epenent variable, SAR it, is equal to one for those firms that exporte at t, but not at t-1, an 0 otherwise. As in the regression for firm failure, our right-han sie variables inclue the firm s size, its age, an its FP. o test our moel s preiction, we have also ae the volatility of the firm s national sales growth among our regressors. he results are presente in column 1 of able 4. We can see that larger firms are more likely to enter export markets, an that the volatility of national sales growth is also positively associate with the probability that the firm starts exporting. Yet, it shoul be note that our previous measure of volatility, calculate over a rolling winow of five years is base on the firm s national sales before an after entry in the foreign market. his coul introuce bias in the regression. We therefore verify whether our results are robust to using two ifferent measures of national sales volatility: the first is the stanar eviation of the firm s real sales calculate over the five years preceing an incluing year t. he secon one is calculate in a similar way but using all years preceing an incluing year t. he results base on these two alternative measures of volatility are presente in columns 2 an 3 of able 4. We can see that in both cases, a higher volatility is still positively associate with a higher probability to start exporting. hus, as preicte by the moel, those firms isplaying high volatility in their national real sales are also more likely to start exporting. In summary, the ata seem to len strong support tour moel. INSER ABLE 4 HERE i s volatility in year t is in the mile 33% of the istribution, an 0 otherwise; an HIGHVOL it is equal to one if firm i s volatility in year t is in the highest 33% of the istribution, an 0 otherwise. 20

21 5. Conclusion We have constructe a ynamic monopolistic moel with heterogeneous firms to stuy the links between firms iiosyncratic uncertainty, the egree of financial constraints that they face, an their export market participation ecisions. Our moel, which can be seen as an extension of Melitz (2003) moel, where firm level volatility is inclue as an aitional imension of firm heterogeneity, preicts that more volatile companies are more likely to fail, nee to be more prouctive to stay in the market, an are less likely to enter export markets. A further implication is that through market iversification, exports ten to stabilize firms total sales. We have teste these preictions, using a panel of 9292 UK manufacturing firms over the perio , an foun strong support for our moel. 21

22 APPENDIX 1: Proofs of Proposition 1 an of the Zero-Profit Conition Proof of Proposition 1 We want to show that given two firms (i an j) which only iffer in their volatility (for example ), then the more volatile firm has a higher probability of bankruptcy i j than the less volatile company, i.e.. i j Let us enote with f(z) the istribution of shocks of firm i, an with g(z) the istribution of shocks of firm j. Let ẑ be the point where both istributions intersect. Since the firms have the same prouctivity level an interest rate, from the point of view of the bank, the threshol shock is the same for both firms, i.e. z(, r) z(, r) z. As i j shown in Figure 1, if ẑ z, then A>B meaning that. i j z z f z gz, an therefore INSER FIGURE 1 HERE Focusing on Figure 2, if ẑ z, then the probability of bankruptcy can be expresse as A B C for company i, an as B C D for company j. If A>D, then i i j j. he properties of the normal istribution imply that A+B+C+F=B+D+C+E+F=1/2, therefore A=D+E. Since we have assume that z 1, this leas to E>0, therefore A>D. Q.E.D. INSER FIGURE 2 HERE Proof of the Zero-Profit Conition he firm s expecte profits are given by the following expression: y( ) Et ( t 1) p( ) y( ) (1 r(, )) I ( ) (1 r(, )) I, where ( ) is the firm s revenue. he maximum interest rate the company is willing to pay is the interest 22

23 rate, which make expecte profits equal to 0. We enote this interest rate with r, which satisfies: 1 r ( ) I. Note that if the company is charge the interest rate r, the threshol shock is z( r ) 1, an the probability of bankruptcy is ( r ) 1/ 2. he minimum interest rate that the bank will charge is the riskless interest rate r 0. herefore, r [ r0, r ]. We want to show that there exists an unique r 0 * [ r, r ] such that Equations (3) an (4) in the text. Accoring to these Equations, the bank will issue funs at an interest rate such that the expecte repayment from the loan is equal to the repayment of a riskless loan for firms with probability of bankruptcy (,, r) ; an collateral is a negative function of the firm s volatility. hus, the following Equation hols: I ( r * r ) ( r*)[(1 r*) I C] 0 (A1) 0 Let us efine j1 ( r) I( r r0 ), an j ( r) ( r)[(1 r) I C] 2. We want to show that these functions intersect at a unique point. Both j ( r ) an j ( ) 1 2 r are increasing functions of r in the interval [ r0, r ]. We therefore have: j ( ) 1 r j2 I 0 an ( r ) ( r ) [(1 r) I C] I ( r) 0 r r r ( r) I 1 e r p. y ( z1) 2 0 since Moreover, the secon erivate of j ( ) 2 r with respect to r is positive for z 1 or its equivalent if r r. his can be represente as: 2 2 j2( r) ( r) ( r) [(1 r) I C] 2 I ( r) 2 2 r r r with 2 r I 2 2 ( z1) 2 z e ( ) 1 1 r p. y 2 23

24 We know that at r 0, j1 ( r0 ) 0 j2( r0 ) ( r0 )[(1 r0 ) I C] when (1 r0 ) I C. On the 1 other han, j1 ( r ) I( r r0 ) an j2( r ) ( r )[(1 r ) I C( )] [ ( ) C( )]. It 2 follows that at r, j1 ( r ) j2( r ) if ( ) 2 I (1 r0 ) C( ). herefore by the intermeiate value function, there exists an unique r* [ r0, r ] in which both functions intersect when ( ) 2 I (1 r ) C( ), an Equation (14) hols. Now we have to show that the curve 0 j ( r ) always intersects with the line 2 j ( ) 1 r from bellow, so that they never intersect at two points in the interval [ r0, r ] an there cannot be two possible interest rates for a given firm. his is equivalent to showing that: j2( r*) j1 ( r*) r r (A2) or equivalently that I ( z( r*) 1) 2 e.[(1 r*) I C] I ( r*) I. Since p. y 2 z( r*) 1 (0,1) then e 2 2 ( z( r*) 1) 2 I 1. We also know that 1 p. y an that the maximum value of ( r*) is 1/2. herefore our conition (A2) can be expresse as follows: I [(1 *) ].[(1 *) ] 2 C 2 2 C. his inequality hols if 2 (1 r*) I C. I here are two possible cases. First, if the collateral is such that C (1 r0 ) I, then there is no risk an the bank will set r* r0. Secon, if C (1 r0 ) I, the bank will set an interest rate r* ( r0, r ) for those firms with ( ) 2 I (1 r0 ) C( ). For firms with ( ) 2 I(1 r ) C( ), the bank will have incentives to charge an interest rate that the 0 24

25 firms will not have incentives to take, preferring to stay out of the market. Finally for firms with: ( ) 2 I (1 r ) C( ) the bank will charge r* 0 (A3) r an these firms profit will be zero. Furthermore, we assume that the function C( ) is invertible. here exits therefore a function s( ) that solves equation (A3), such that the firm is charge an interest rate leaing to zero profits. INSER FIGURE 4 Proof of the Stationary Equilibrium in autarky he Zero profit conition (,s( )) 0 can be expresse as ( ) (1 r(,s( ))) I. he free entry conition is as: (, ).P (,s( ))/ (, ) f. he average profits can be written in e (, ) ( ) (1 r(, )I ( ) 1 (1 r(, )I I 1 (1 r(,s( ))) (1 r(, ). Substituting this expression into the free entry conition, we obtain the following Equation which implies a cutoff level: ( ) I 1 (1 r(,s( ))) (1 r( (,s( )), (,s( ))).Pin (,s( )) / ( (,s( )), (,s( ))) fe he left han of the equation is a ecreasing function that goes to infinity when goes to zero an goes to zero when goes to infinity. 25

26 Appenix 2: Data Structure of the unbalance panel: Number of observations per firm Number of firms Percent Cumulative otal Definitions of the variables use: FAIL it : ummy variable equal to 1 if firm i faile in year t, an 0 otherwise. We efine a firm as faile (ea) in a given year if its company status is in receivership, liquiation, or issolve. Size it : logarithm of the firm s total real assets. otal assets are given by the sum of fixe (tangible an intangible) assets an current assets, where current assets are efine as the sum of stocks, work-in-progress inventories, trae an other ebtors, cash an equivalents, an other current assets. Sales: inclues both UK an overseas turnover. Collateral it /Debt it : the ratio between the firm s tangible assets an its total (long- an short-term) ebt. FP it : total factor prouctivity calculate using the Levinsohn an Petrin (2003) metho. 26

27 Group i ummy variable equal to 1 if the firm is part of a group, an 0 otherwise. A company is sai to be part of a group if it is a subsiiary of one or more holing companies (UK or foreign) 18. Quiscore it is given as a number in the range from 0 to 100. he lower its Quiscore, the more risky a firm is likely to be. he inicator is constructe taking into account a number of factors, incluing the presence of any averse ocuments appearing against the company on the public file, an the timeliness of getting the accounts file. However, the most important factors relate to the financial performance of the company as evience by its balance sheet an profit an loss accounts. he key financial items use inclue turnover, pre-tax profits, working capital, intangibles, cash an bank eposits, creitors, bank loans an overrafts, current assets, current liabilities, net assets, fixe assets, share capital, reserves an shareholers funs. he unerlying economic conitions are also taken into account. SAR it : ummy variable equal to 1 if firm i exporte a positive amount in year t, but not in year t-1, an 0 otherwise. otalvol it : stanar eviation of the firm s total real sales growth. he stanar eviation is measure over a rolling winow of 5 years. Nationalvol it : stanar eviation of the firm s real national sales growth. he stanar eviation is measure over a rolling winow of 5 years. Overseasvol it : stanar eviation of the firm s real overseas sales growth. he stanar eviation is measure over a rolling winow of 5 years. Prenationalvol it : stanar eviation of the firm s real sales calculate over the five years preceing an incluing year t. Prenationalvol1 it : stanar eviation of the firm s real sales calculate over all years preceing an incluing year t. LOWVOL it : ummy variable equal to one if firm i s otalvol in year t is in the lowest 50% of the istribution of the otalvols of all firms operating in the same inustry as firm i s in year t, an 0 otherwise. 18 Information about whether a firm is part of a group is only provie in the last year of observations available for each firm. We therefore assume that a firm which was part of a group or foreign owne in its last available year was part of a group or foreign owne throughout the perio in which it was observe. Given the short sample that we analyze, this is a reasonable assumption. 27

28 HIGHVOL it : ummy variable equal to one if firm i s otalvol in year t is in the highest 50% of the istribution of the otalvols of all firms operating in the same inustry as firm i s in year t, an 0 otherwise. Deflators: all variables are eflate using the aggregate GDP eflator. REFERENCES Blunell, R., Bon, S., Devereux, M., an F. Schiantarelli (1992). Investment an obin's Q: evience from company panel ata. Journal of Econometrics, 51, pp Bunn, P. an V. Rewoo (2003). Company accounts base moelling of business failures an the implications for financial stability. Bank of Englan Discussion Paper No Comin, D., Groshen, E., an B. Rabin (2006). urbulent firms, turbulent wages? National Bureau of Economic Research Working Paper No Comin, D. an Mulai, S. (2005). A theory of growth an volatility at the aggregate an firm level. NBER Working Paper no Comin, D. an Philippon,. (2005). he rise in firm-level volatility: causes an consequences. NBER Working Paper no Cooley,. an Quarini, V. (2001). Financial markets an firm ynamics. American Economic Review, 91(5) pp Disney, R., Haskel, J., an Y. Heen (2003). Entry, exit an establishment survival in UK manufacturing. Journal of Inustrial Economics, 51, Dixit, A. an Stiglitz, J. (1977). Monopolistic competition an optimum prouct iversity. American Economic Review, 67(3), pp Greenaway, D., Guariglia, A. an Kneller, R. (2006). Financial factors an exporting ecisions. Mimeograph, University of Nottingham. Grossman, G. an Helpman, E. (1991). Innovation an growth in the global economy. Cambrige Mass: MI Press. Hirsch, S. an B. Lev (1971). Sales stabilization through export iversification. Review of Economics an Statistics, 53(3), pp