Resolving the Exposure Puzzle: The Many Facets of Exchange Rate Exposure

Transcription

1 Resolving the Exposure Puzzle: The Many Facets o Exchange Rate Exposure Söhnke M. Bartram, Gregory W. Brown +, and Bernadette A. Minton # Abstract Empirical research has documented a low stock price reaction to exchange rate movements. We examine a global sample o,6 manuacturing corporations rom 6 countries and show that the weak response is due to a combination o actors at the industry and irm level. Speciically, we extend prior theoretical results, which model a irm s oreign exchange rate exposure as a unction o market share, product substitutability, pass-through, sales and cost in oreign currency, so that they may be applied to global industries. Using this model, we show empirically that corporations are able to manage exchange rate risk via three complementary channels. First, irms pass part o exchange rate changes through to customers. Second, most global manuactures utilize operational hedges (e.g., matching oreign sales with oreign production). Third, corporations employ inancial risk management strategies such as issuing oreign currency denominated debt and entering into oreign exchange rate derivatives transactions. All o these channels are important. We estimate that or a typical irm pass-through and operational hedging each reduce exposure by 0% to 5% and inancial hedging reduces exposure by 45% to 50% (depending on model parameterization). Finally, our evidence suggests that the combination o these actors is suicient or explaining the observed levels o oreign exchange rate exposure. Keywords: Competition, hedging, exposure, derivatives, corporate inance, international inance JEL Classiication: G3, F4, F3 First version: October, 005 This version: June 4, 006 Lancaster University, Management School, Department o Accounting and Finance, Lancaster LA 4YX, United Kingdom, phone: +44 (5 4) , ax: + (45) , <s.m.bartram@lancaster.ac.uk>, Internet: < + Corresponding Author, Associate Proessor o Finance, Kenan-Flagler Business School, The University o North Carolina at Chapel Hill, CB 3490, McColl Building, Chapel Hill, NC USA, phone: (99) , gregwbrown@unc.edu. # Associate Proessor o Finance, Fisher College o Business, The Ohio State University, 834 Fisher Hall, 00 Neil Avenue, Columbus, OH USA, phone: (64) , minton_5@cob.osu.edu. The authors wish to thank Eitan Goldman and Merih Sevilir or their assistance with the enhanced BDM model as well as Keith Brown, Kalok Chan, Joshua Coval, John Griin, Yrjö Koskinen, Stephen Magee, John Hund, Mitchell Petersen, Roberto Wessels and seminar participants at the 006 Financial Intermediation Research Society Conerence, Forum on Corporate Finance, Beijing University, Hong Kong University o Science and Technology, University o North Carolina and University o Texas at Austin or helpul comments and suggestions. They grateully acknowledge research unding by the Center or Financial Research o the FDIC as well as support by Mike Pacey, Global Reports, and Thomson Financial in establishing the dataset. Florian Bardong provided excellent research assistance.

2 Resolving the Exposure Puzzle: The Many Facets o Exchange Rate Exposure Abstract Empirical research has documented a low stock price reaction to exchange rate movements. We examine a global sample o,6 manuacturing corporations rom 6 countries and show that the weak response is due to a combination o actors at the industry and irm level. Speciically, we extend prior theoretical results, which model a irm s oreign exchange rate exposure as a unction o market share, product substitutability, pass-through, sales and cost in oreign currency, so that they may be applied to global industries. Using this model, we show empirically that corporations are able to manage exchange rate risk via three complementary channels. First, irms pass part o exchange rate changes through to customers. Second, most global manuactures utilize operational hedges (e.g., matching oreign sales with oreign production). Third, corporations employ inancial risk management strategies such as issuing oreign currency denominated debt and entering into oreign exchange rate derivatives transactions. All o these channels are important. We estimate that or a typical irm pass-through and operational hedging each reduce exposure by roughly 0% to 5% and inancial hedging reduces exposure by about 45% to 50% (depending on model parameterization). Finally, our evidence suggests that the combination o these actors is suicient or explaining the observed levels o oreign exchange rate exposure.

3 Introduction For many corporations oreign exchange rate luctuations represent a major inancial risk. For example, global companies regularly cite exchange rate movements as a cause o earnings surprises. The recent strength o the U.S. Dollar during the 00 recession has even led to the creation o the Coalition or a Sound Dollar, a group o over 00 major trade associations which lobbies the U.S. government or a weaker currency. Theoretical models (such as Bodnar, Dumas, and Marston, 00) also predict that many irms should have signiicant exchange rate exposures. Despite these anecdotal indications and theoretical predictions o signiicant exchange rate exposures, academic studies have documented only weak relationships between exchange rate changes and stock returns (see, or example, Jorion, 990). The goal o this paper is to resolve the discrepancy between theoretical predictions o exchange rate exposure and observed levels o exchange rate exposure in the broad cross-section o global corporations. We undertake this by combining a variety o insights rom the extant literature to orm a more comprehensive analysis o exchange rate exposure. In essence, our analysis examines how irms combine three dierent mechanisms at their disposal or mitigating exchange rate risk. First, irms can (to varying degrees) pass through to customers changes in costs due to exchange rate movements. Second, irms can oten aect their exchange rate exposure by choosing the location and currency o costs (e.g., where actories are located). Third, irms can utilize an array o inancial products, such as oreign currency denominated debt and inancial derivatives, as exchange rate risk management tools. Our results show that each o these actors plays an important role in mitigating observed exchange rate exposure and together they account or the vast majority o the discrepancy between prior theoretical predictions and observed exposures. Our analysis has three parts. First, we examine the global automotive industry as a motivating example. Using a simple model o exchange rate exposure by Bodnar and Marston (00), we show that global automakers have large theoretical exposures to exchange rates based simply on their high levels o oreign sales. Accounting or oreign production reduces the theoretical exposure o automakers by about 50%. Finally, simple corrections or the use o oreign exchange rate derivatives and oreign currency debt reduce the theoretical exposure by an addi- 3

4 tional 40%. Consequently, the inal theoretical exposure o global automakers to exchange rate risk is airly low. Second, we expand the theoretical model o Bodnar, Dumas, and Marston (hereater BDM, 00) to examine the exchange rate exposures o a global irm that can compete and produce in both a oreign and local market. In the BMD model, the exporting irm cannot sell in its own market and the local irm cannot produce abroad. By assuming that global oreign exchange rate exposure is a weighted average o a irm s oreign exchange exposure in the oreign market and the domestic market, we can derive optimal pass-through decisions and the resulting oreign exchange exposures o global irms in competitive industries. Our enhanced BDM model generates exposures as a unction o market share, product substitutability, pass-through, sales and cost in oreign currency that are smaller than the original BDM model under most conditions, and in some cases the enhanced model generates negative oreign exchange rate exposures. Overall, our enhanced model allows or a richer, more realistic set o exposures. Third, we analyze a large sample o global manuacturing irms in 6 countries using the enhanced BDM model. We show that pass-through and operational hedging are important or reducing the level o exchange rate exposure. However, ater accounting or pass-through and operational hedging, theoretical exposures are still larger than observed exposures. We document that irms with high theoretical exposures are both more likely to have oreign currency debt and more likely to use exchange rate derivatives. We estimate the reduction in exposure due to each channel or a typical irm. Depending on the level o product substitutability, passthrough reduces exposure by about 0% to 5%. Operational hedging reduces exposure by similar amounts, while inancial risk management (oreign currency debt and FX derivatives) accounts or a urther 45% to 50% reduction in exposure. Thus, irms reduce their gross exchange rate exposure by about three quarters via the three channels together. Consequently, or reasonable parameter values, it is not possible to reject the enhanced BDM model ater correcting or the estimated eects o inancial risk management. Our study contributes to the literature in several ways. First, we take a more comprehensive approach to studying oreign exchange rate exposure. For example, the model we derive allows or global irms in an imperectly competitive global market. Furthermore, the sample o irms and the number o countries we study is among the largest examined to date and includes 4

5 inancial risk management variables that have not been considered in such a large and diverse sample. Finally, and most importantly, we are able to resolve a major puzzle in inancial economics by careully demonstrating the close relationship between theoretical and observed oreign exchange rate exposures. The paper is organized as ollows. Section provides a review o the extant literature. Section 3 demonstrates in a simple ramework the impact o various actors such as the competitiveness o a irm, operational hedging, and inancial hedging on the exchange rate exposure o irms in the global automotive industry. Subsequently, in Section 4 we develop an extended ramework or assessing oreign exchange rate exposures and pass-through, based on the theoretical model o Bodnar, Dumas and Marston (00). This enhanced BDM model is estimated or a global set o manuacturing irms in Section 5. Finally, Section 6 concludes. Motivation and Related Literature Studies on exchange rate exposures such as Adler and Dumas (984), Hekman (985), Shapiro (975), Flood and Lessard (986), von Ungern-Sternberg and von Weizsäcker (990), and Marston (00) develop theoretical models o irm behavior to examine oreign exchange exposure. These papers do not provide empirical estimates o their models. In a more recent paper, Bodnar, Dumas, and Marston (00) speciy a theoretical model o exchange rate exposure that explicitly incorporates optimal export pricing behavior and provide direct estimates o the structural model. Specially, BDM develop a model o exporting irms under duopoly to study exchange rate exposure and pass-through behavior o irms. An exporting irm competes with a oreign irm in that export market. The costs o the exporting irm are based in the domestic or oreign currency, while the oreign irm only has oreign costs. BDM are able to derive optimal pass-through decisions and the resulting exchange rate exposure. They show that as substitutability between home-produced and oreign-produced goods (holding market shares constant) increases, exposure increases and pass-through declines. Holding product substitutability ixed, increases in market share reduce both exposure and pass-through elasticities. However, their empirical analysis is limited to a ew industries in Japan and the empirical results are mixed. Dekle (005) studies the impact o substitutability between oreign and export products and the type o competition on exchange rate exposure. For 5 Japanese export industries, product substitutability is ound to be high, and ten o these 5 industries are better characterized as 5

6 Cournot competitors in oreign markets than as colluders. While there is heterogeneity in the exposure elasticities, collusive exporters tend to have higher elasticities than competitive exporters. This result is consistent with the theoretical predictions o the paper. In some ways, the analysis o operational hedging by choosing the location o costs and the eects o market competition on pricing policy are simply dierent mechanisms or aligning costs and revenues. For example, locating a production acility in a oreign country aligns costs with revenues, whereas passing-through exchange rate changes to output prices aligns revenues with costs. However, the two methods dier in some respects. The location o production is largely under the control o management (though it may be costly to change) and is a real investment decision or the irm. To the contrary, the ability to pass-through price changes depends on the nature o competition in the product market and is presumably not under the control o management i managers choose a value-maximizing strategy. Beginning with Jorion (990), a number o empirical studies on oreign exchange exposures estimate exposures without a theoretical model and have documented typically small oreign exchange exposures. Exposure studies generally report large dierences in exposures across industry classes and countries (e.g., Campa and Goldberg (999), Bodnar and Gentry (993), Marston (00), and Allayannis and Ihrig (00)). Williamson (00) shows that auto manuacturers have higher oreign exchange rate exposure i they have high levels o oreign sales and ace oreign competition. Bartram and Karolyi (003) ind that the oreign exchange rate exposure o noninancial irms is systematically related to irm characteristics (sales, the percentage o oreign sales in general and in Europe in particular), regional actors (geography, strength o currency) and industry characteristics (competition, traded goods). Other studies also have examined the association between oreign exchange exposures and irm and macroeconomic activities. For example, Parsley and Hopper (00) and Dahlquist and Robertson (00) ind signiicant associations between exchange rate exposures and currency arrangements. Several studies document signiicant negative associations between oreign See also Jorion (99), Bodnar and Gentry (993), Bartov and Bodnar (994), Bartov, Bodnar, and Kaul (996), Choi and Prasad (995), He and Ng (998), Chow, Lee and Solt (997), Griin and Stulz (00), and Bartram and Bodnar (005). 6

7 exchange exposures and the use o inancial derivatives and oreign debt (e.g. Allayannis and Oek (00), Wong (000), Simkins and Laux (996) Hagelin and Prambourg (00)). Recent studies also examine the associations between exposure and proxies or operational hedging (Pantzalis, Simkins and Laux (00) and Carter, Pantzalis and Simkins (003)). In a recent paper, Carter, Pantzalis and Simkins (003) examine the use o derivatives and operational hedges on oreign exchange exposures and examine whether operational hedges act as real option strategies. A similar analysis o two gold mining irms by Petersen and Thiagarajan (000) reveals that operating lexibility can substitute or inancial risk management (i.e., hedging gold prices). As Triantis (000), among others, points out, the real option strategy, such as locating a plant in a oreign country, is more valuable when exchange rates are more volatile (i.e., uncertainty increases). Carter et al. (003) document that exposures vary not only as to whether a irm is a net exporter or net importer but also across weak and strong dollar states. They interpret these results as evidence that operational hedges serve as real options. The use o inancial risk management tools has grown substantially over the last 30 years to the point where the majority o large corporations with oreign operations use some type o inancial instrument that can mitigate FX risk. Two o the most common methods are issuance o oreign currency (FC) denominated debt and the use o FX derivatives. FC Debt is widely regarded by inancial executives as an eective method or mitigating FX risk. For example, Graham and Harvey (00) report that 85.8% o surveyed Chie Financial Oicers respond that the ability to provide a natural hedge is important or very important in the decision to use FC Debt. Additional evidence that FC debt is used as a inancial risk management tool is provided by Allayannis and Oek (00). Similarly, Bartram, Brown, and Fehle (003), among other studies, show that FX derivative use is widespread among global irms with oreign sales. Financial risk management tools have some distinct advantages over the other types o FX risk mitigations techniques. First, inancial decisions are very much under the control o inancial managers and less subject to constraints in the product market (e.g., geographic availabil- Because local currency debt can be eectively converted to oreign currency debt with FX derivatives, and viceversa, there is not necessarily a clear distinction between these two strategies (see Allayannis, Brown, and Klapper, 003). For example, synthetic FC debt can be created using a currency swap. Synthetic FC debt might be preerable to natural FC Debt or tax, accounting, and regulatory reasons. 7

8 ity o skilled labor and raw materials or the competitive landscape o the local market). Second, inancial hedges are likely to have low implementation costs and positions are usually reversible or easily adjusted. Third, inancial hedges are unlikely to introduce additional new risks that might be associated with some operational hedges (e.g., the risk o oreign assets be expropriated or nationalized). On the other hand, inancial risk management tools, especially FX derivatives, are likely to have relatively short horizons. 3 Exchange Rate Exposure in the Automotive Industry In this section, we present a brie industry case study o global automotive production to motivate our subsequent, broader analysis. The goal o this section is to illustrate in a simple, yet powerul, way the potential eects o competition, operational hedging, and inancial hedging on the FX exposure o irms. First, we show that Gross Exposures (estimated rom oreign sales and proit margins) are oten quite large or global automakers. Next, we show that Model Exposures, which account or oreign costs (i.e., operational hedges), are signiicantly lower. Finally, we estimate the eect o inancial hedging with oreign currency debt and oreign exchange rate derivatives and show that these Residual Exposures are much closer to the actual oreign exchange rate exposures estimated rom stock price data. We choose the auto industry or several reasons. First, it is a well-known, mature, and competitive industry. In addition, the industry is truly global with major companies headquartered (as well as manuacturing) in Asia, Europe, and North America. The auto industry has a strong anecdotal history o being aected by exchange rates and consequently taking exchange rate risk seriously. 3 Finally, we are by no means the irst to study the oreign exchange exposure o the auto industry, so we can rely on existing results or some o our discussion. In particular, Williamson (00) studies the auto industry rom 973 to 995 to examine the eect o real exchange rate changes on the value o irms. In his analysis, Williamson takes into account the eect o industry structure and competition among U.S., German, and Japanese irms. Among other indings, Williamson documents statistically signiicant but economically small exchange rate exposures that change over time. 3 See, or example, Foreign Exchange Hedging Strategies at General Motors (004 Harvard Business School Case) by Mihir Dasai and Mark Veblen. 8

9 Consistent with theoretical predictions, changes in exposure are related to changing industry structure (e.g., competition) as well as large and extended movements in real exchange rates. Our subsequent analysis diers rom Williamson on various dimensions. First, we examine a somewhat more global auto industry by expanding the sample o irms in Williamson s analysis rom 0 to 6 irms. Second, in addition to industry structure and competition, we explicitly examine the impact o inancial hedging in the orm o oreign exchange rate derivatives and oreign currency debt on irms oreign exchange exposures. Third, we extend the measures o industry structure, competition, and export sales to examine the dierences in global and regional competition measures. Table reports selected descriptive statistics on the global auto industry. The sample includes 6 irms, representing six countries (France, Germany, Italy, Japan, South Korea, and the United States). Panel A reports irms sales or North America, Europe, Japan, and all other regions o the world. The data indicate that or most auto manuacturers the largest ractions o sales occur in their own region. For example, except or DaimlerChrysler, European irms sales are predominantly in Europe with negligible sales volume in Japan. Panel B o Table reports irms production volume around the world. The data show how irms have geographically diversiied production with roughly one third o output (on average) produced outside their home country. Panel C o Table reports the dierence between sales and production percentages or each irm and each region o the world. A positive (negative) number indicates that in the region o interest, a irm has more (less) sales volume than production volume. In some cases irms have larger ractions o oreign production than ractions o oreign sales. However, most irms produce more than they sell locally (e.g., all Japanese irms). These values are o interest because they are closely related to net operating exposures. As discussed previously, irms can operationally hedge to manage exchange rate exposure and as expected, the values show that irms will to varying degrees locate production in regions o the world in which they have sales. For example, the North American production, in part, acts as an exchange rate hedge to the European irms North American sales. 9

10 More interesting is the possible eect o global competition on the regional markets. Panel A shows that North American irms sell the second largest volume o autos in Europe. 4 Thus, European irms can have signiicant oreign exchange exposure rom their import competition. However, the nearly balanced production and sales or both U.S. and European manuacturers in Europe will tend to limit large competitive exposures. In contrast, other irms choose to have relatively low production relative to sales in certain regions. The most extreme examples are Mazda and Hyundai, which each have sales in North America and Europe exceeding production in those regions by 6.% to 3.%. As noted already, we seek to understand how these operating hedges (or lack thereo) combine with competitive orces and inancial risk management decisions to determine an overall oreign exchange exposure. Similar to Williamson (00) and others, we base our analysis on theoretical measures o exposure. For this part o the analysis, we estimate the simple model o exchange rate exposure derived in Bodnar and Marston (00), which show that the exchange rate exposure elasticity (δ) can be expressed as δ = h + ( h h ) () r where h is the oreign currency denominated revenue as a percent o total revenue, h is oreign currency denominated costs as a percent o total costs, and r is the irm s proit margin. In Table, we examine the oreign exchange exposure o global automakers based on this model. The irst two columns o Table report the total percentage o oreign sales and the gross proit margin. On average, more than hal o automakers sales are oreign. Gross proit margins are airly similar across irms with the spread between the lowest (Isuzu) and the highest (Honda) equal to 0.6. The next column reports a value or Gross Exposure, which is an estimate o the oreign exchange rate exposure beore accounting or oreign production (i.e., operational hedging) and is obtained by evaluating equation () or each irm with h set equal to zero. These Gross Exposure values tend to be large. The average gross exposure o.70 indicates that i automakers produced only in their home country, their stock prices should decline on average 4 Ford and General Motors both have acquired European automakers. For example, Ford acquired Jaguar in 989 and Volvo in 999. General Motors acquired Adam Opel in 99 and 50% o SAAB in

11 .7% in response to % appreciation o the local currency. The simple model by Bodnar and Marston (00) illustrates how important industry competition and proit margin can be to overall exchange rate exposure. For example, i we assume that a less competitive irm has a lower proit margin, it will also have a higher exposure, ceteris paribus. 5 Consequently, the automakers with the lowest gross proit margins (Isuzu, DaimlerChrysler, and Mitsubishi) are the ones with the highest Gross Exposures. The next column o Table labeled Model Exposure reports results obtained rom evaluating equation () using actual values o oreign production (h ). The important role o oreign production as an operational hedge is evident rom the substantial drop in estimated exposure or each irm. Several companies have estimated exposures close to zero, and the average o all irms drops by more than hal to.6. O course, automakers also use inancial tools to manage oreign exchange rate risk. Two common techniques involve issuing oreign currency (FC) denominated debt and the use o oreign exchange rate (FX) derivatives. The next columns report the outstanding value o oreign currency debt and oreign exchange rate derivatives as a percent o irm value. All o the automakers use oreign exchange rate derivatives to some extent, though some (like Mitsubishi and Renault) are clearly much larger users than others (like Ford and Toyota). On average, automakers hold derivatives with notional values equal to 0.4% o irm value. Automakers use o oreign currency debt, while relatively low, also reduces their exchange rate exposure. 6 The next column o Table adjusts the Model Exposure numbers by subtracting the level o oreign exchange rate derivatives and oreign currency debt to obtain an estimate o Residual Exposure. These values or Residual Exposure are our estimates o theoretical oreign exchange rate exposure taking into account the level o oreign sales, industry competition (via gross proit margin), operational hedging (via oreign costs), and inancial hedging (via oreign exchange rate derivatives and oreign currency debt). On average, these values are considerably lower than the estimates or the Model Exposure. The Residual Exposures range rom negative values 5 This is because the less proitable irm will have the same nominal exchange rate exposure, but a lower irm value. We also note that this analysis stretches the application o the Bodnar and Marston (00) model, which is or a monopolist. We explicitly consider exposure in a model with a competitive product market in the next section. 6 The use o FC debt is sometimes diicult to determine or automakers because o poor disclosure by some irms, so we make conservative assessments in obtaining these values.

12 to positive values, much like the values or Actual Exposure derived rom a regression model (and tabled in the ollowing column). 7 Figure shows that there is a clear positive relationship between these estimates even though the theoretical estimates o Residual Exposure are signiicantly more volatile than the statistical estimates o Actual Exposure. 8 These results are consistent with the inding in Bartram (005) that the insigniicance o oreign exchange rate exposures o aggregate perormance measures, such as total cash low or stock price, can be explained by hedging at the irm level. In summary, this section has used global automakers to illustrate how various actors such as the competitiveness o a irm, operational hedging, and inancial hedging can have dramatic eects on oreign exchange rate exposure. Speciically, our quick estimates o theoretical oreign exchange rate exposure that account or risk management are quite similar to estimates o actual oreign exchange rate exposure obtained rom stock returns. 4 An Expanded Empirical Framework Despite the encouraging results presented in the previous section, the calculated Residual Exposure still (i) overestimates exposures on average and (ii) predicts a wider range o exposures than actually observed among automakers. One potential explanation or these results is our use o a relatively simple model as an illustrative example. Speciically, the Bodnar and Marston (00) model is or a monopolist with exogenously speciied cash lows. In our analysis o automakers, we analyze the eect o product market competition on oreign exchange rate exposure by assuming that the gross proit margin is a suicient statistic or describing a irm s level o overall competitiveness. In this section, we present a more realistic model o the oreign exchange rate 7 Following the existing literature on exchange rate exposure (see Bartram and Bodnar (004) or a summary o this research), we regress separately or each irm its excess stock return on the excess return o the local market index (Market) and the percentage change in a trade-weighted exchange-rate index (FX). The regressions are estimated using weekly data or the period rom 000 to 004. Overall, the results in Table show that irms have either no statistically signiicant oreign exchange exposure or positive exchange rate exposure. The exchange rates are deined in terms o local currency per unit o the oreign currency. Thus, a positive coeicient estimate is consistent with extant research results on the exposure o exporting or import-competing irms (See Bartram and Bodnar, 004; Bartram, 004). The coeicient on the trade-weighted exchange-rate index is positive and signiicant or only ive o the 6 irms. 8 The estimated slope obtained rom a regression o Residual Exposure on Actual Exposure is 0.66 with a p-value o 0.09.

13 exposure o a global irm in a globally competitive market based on the results o Bodnar, Dumas, and Marston (00). BDM examine a irm selling in a oreign market. The irm in their model can produce (i.e., have costs) in both the local and the oreign country, and it competes with another irm that produces (i.e., has costs) only in the oreign market. The authors examine pass-through o exchange rate changes or irms based on the type o competition (price or quantity), the relative cost structures, and product substitutability. While the model very precisely examines the eect o competition and the currency denomination o costs on exchange rate exposure, its empirical relevance is limited by simpliying assumptions. To acilitate our empirical analysis, we extend the BDM model in two straightorward ways. First, we consider a market where both irms have costs in local and oreign currency. Second, we consider irms that sell both locally and in oreign markets. These are relatively simple extensions o the BDM model that make it applicable to a much broader sample o global irms. 9 In addition, these extensions produce a substantially broader set o predicted exchange rate exposures. For example, our model considers the case o a irm that produces in a oreign market, sells in the local market, and aces import competition. This irm can have a negative exposure to exchange rate changes something not allowed or in the original BDM model. We reer the reader to BDM or a ull discussion o the base model and begin our analysis with a derivation o a global irm s exchange rate exposure Exchange Rate Exposure o a Global Firm that Competes Globally Total exchange rate exposure or a global irm (δ) is deined as the sales-weighted average o exchange rate exposures rom oreign operations (δ ) and domestic operations (δ d ) so that ( ) δ = φδ + φ δ () where φ is the percentage o oreign sales. To keep the analysis tractable, we proceed by assuming that two symmetric irms compete in a single oreign market and derive the exposure o each irm. We then treat the exposure o the oreign irm as equivalent to δ and the exposure d 9 Because our goals are ultimately empirical (and to keep the exposition parsimonious), we examine only quantity competition. However, expanding the analysis so as to also examine price competition would be straightorward. 0 The ull model corresponding to BDM (00) is available rom the authors upon request. This is equivalent to assuming that a global irm is made up o completely separate domestic and oreign divisions that can be examined independently. 3

14 o the local irm as δ d. O course, the essential dierence in exposure between the local irm and the oreign irm is the reported currency denomination o proits, since each is potentially global with regards to its production and sales. To acilitate the exposition, we conorm to the notation o BDM and deine the proits ( π ) in home currency as i i i i ( C SC ) π = SP X + X (3) where S is the exchange rate (in domestic currency relative to oreign currency), P i is the equilibrium price o irm i s product, X i is the equilibrium quantity o product sold by irm i, and C i (C i ) are the marginal costs in home (oreign) currency. However, we note that unlike BDM, we allow both irms to have costs in oreign currency so that the relative cost ratio in home currency is i i i C + SC R =. (4) C + SC to In Appendix A, we show that the equilibrium exposure measures δ and δ d are equivalent ( λ ) ρ ( λ ) dlnπ λρμ δ = = + ( λ ) ρ μ + dln S (5) ( λ ) ρ ( λ ) dlnπ d λρμ d d d δd = = ( λd) ρdμd + dln S d d (6) ( ) where or j d,, ( ) ( μ = γ γ γ γ, j j j j j ) λ j is the equilibrium market share in market j, γ ji is the raction o marginal costs in oreign currency in market j or irm i, and ρ j ( 0, ) measures substitutability between the irms products in market j. Thus, the overall exchange rate exposure or a global irm can be expressed as 4

15 ( ) ρ ( λ ) ( d ) ρ ( λd ) λ λρμ λ λdρdμ d δ = φ + ( λ ) ρ μ + + ( φ) ( λd ) ρdμd +. d (7) While our exposure expression is signiicantly more complicated than the original BDM model, the underlying actors determining exposure are the same. In act, the exposure equation rom BDM is a special case o the above equation where φ= and γ =. The terms μ and μ d capture the combined relative eects o marginal costs in oreign currency or irm (γ ) and irm (-γ ). Since 0 γ j and γ j, it ollows that - μ j. Thereore, in our expanded model the lowest obtainable exposure is -.0 as opposed to +.0 in the BDM model. This quite negative exposure would be predicted or a irm that produces only abroad, sells only locally, has competitors that do the same, has a high degree o product substitutability, and low market share. The negative exposure in this case is intuitive since this type o irm would beneit rom an appreciating local currency. Overall, our model allows or a wide range o negative exposures and positive exposures (up to +3.0), which are to be expected or most irms that have exporting activities and/or ace import competition. 4. Exchange Rate Pass-Through o a Global Firm that Competes Globally In a similar way, we derive the pass-through o exchange rate risk or a global irm as the oreign sales-weighted average o pass-through rom oreign operations (η ) and domestic operations (η d ) ( ) η = φη + φ η. (8) d Pass-through is deined as the partial derivative o the equilibrium price o irm i s product, P i, to the exchange rate. Appendix B shows that the equilibrium pass-through measures η and η d are obtained as dln P dln S η = = ( γ ) + λ ρ ( γ ) γ γ ( γ ) dln P dln S η = = ( γ ) + λ d ρ ( γ ) γ γ ( ) d d d d d d γ d (9) (0) as Thus, the overall pass-through o exchange rate risk by the global irm can be expressed 5

16 ( ) ( ) ( ) η = φ γ λρμ φ γd λd ρdμd, () where all parameters are deined as above or the oreign exchange rate exposure o the irm (δ). This pass-through expression is also more involved compared to the BDM model, but again the original model is subsumed and a special case o parameter values φ= and γ =. (BDM multiply their pass-through estimates by minus one so that the elasticity is positive.) In the BDM model, pass-through is between zero and one. In contrast, in the expanded BDM model, passthrough, η, ranges rom negative one to positive one (- < η < +). Overall, our expanded model allows or a wider range o pass-through values. In the extreme case o η nearly equal to minus one, an appreciation o the exporter s currency (dlns < 0) leads to a nearly 00 percent passthrough to the price o the good, i.e., the irm increases its output price by the extent o the exchange rate move in order to o-set the currency conversion loss in local currency proits. Since exchange rate changes may change marginal costs in a irm s home currency in case it has oreign-currency based costs and may cause irms to change their markup, pass-through is likely to be less than proportionate (i.e. > -). The expanded range o pass-through between 0 and + in the enhanced BDM model allows or an appreciation in the exporter s currency (dlns > 0) to result in a decrease in price. 4.3 Enhanced Model Analysis and Comparative Statics Figure plots the easible space o exchange rate exposures as a unction o product substitutability (ρ). The shaded area above δ=+.0 represents the exchange rate exposures allowed by the BDM model or various values o ρ. The shaded area below +.0 represents the additional range o exchange rate exposures or global irms that compete with other global irms in the enhanced version o the model. Given that prior empirical work documents that most irms have low exchange rate exposures, this additional range has substantial empirical relevance. The act that the range o theoretical exposures expands as product substitutability increases reinorces the notion that product market competition can have a signiicant eect on irms exchange rate exposures. A key aspect o the enhanced model is the ability to consider the exposure o irms with both oreign and domestic sales. Figure 3 plots maximum and minimum exposure as a unction o oreign sales percentage. Again we note that the range o possible exposures is signiicantly greater than in the BDM model even or the case when oreign sales equal 00%. This result derives rom the ability o both irms in the enhanced model to have costs in both domestic and 6

17 oreign currency. This dierence is relected in μ. More speciically, the second term o δ captures the impact o the exchange rate on the share o total oreign expenditures accruing to the exporting irm. Unlike BDM, this term can be zero and in some cases negative. The sign o this second term will depend on the magnitude o μ. Recall that ( ) ( ) μ = γ γ γ γ, where γ and (-γ ) represent the raction o marginal costs due to oreign currency based inputs or irm and irm, respectively. The smaller the raction o oreign cost o irm and the larger the raction o oreign cost or irm, the larger is the value o μ. For example, i γ = (- γ ) = 0.7, then μ equals In contrast, i γ = (-γ ) = 0., then μ equals 0.8. Finally, μ equals zero when γ = γ. The third term o δ captures the impact o the oreign exchange rate on the domestic currency proit margin o the exporter. Like the second term, the sign and magnitude o the third term will depend on μ. I μ is negative (positive), then the third term is negative (positive). As noted already, μ can be zero, and thus, the third term will also be zero i γ = γ. Since the model is symmetric, similar arguments apply to μ d, the relative costs in oreign currency or irm s domestic market. Altogether these results show that or a global irm both the sign and the magnitude o the exchange rate elasticity depends on the sign and magnitude o the relative costs in oreign currency, μ j. Figure 3 also plots the actual exchange rate exposures as a unction o oreign sales percentage or the global automakers discussed in Section 3. It is evident rom the scatter-plot that the actual exposures o all global automakers easily plot inside the allowed region or the enhanced model. However, it is also the case that most o the actual exposures are below the minimum values allowed or by the BDM model (even ater making an adjustment or the act that the BDM model assumes all sales are in the oreign country). To obtain an intuitive eel or the relative importance o the various model inputs, we examine exposure values or a range o parameter values. Figure 4 shows plots o exposure (on the vertical axis) as a unction o the model input parameters where we hold all parameters (besides the one o interest) ixed at values close to the averages o sample irms we examine in the next 7

18 section. Panel A shows exposure or values o the percentage o oreign sales (φ ) ranging rom 0% to 00%. As expected, exposure values increase rapidly with the level o oreign sales. The exposure or a irm with no oreign sales is close to zero (0.03) whereas a pure exporter will have an exposure o around.. Panel B shows the sizeable eect o oreign costs (γ ) on exposure. As oreign costs increase, the natural hedge rom oreign operations reduces exposure signiicantly rom about 0.8 or a irm with no oreign costs to about 0.3 or a irm with only oreign costs. Panels C and D show the eects on exposure o competitors cost structures in both the domestic and oreign markets. Panel C reveals that as competitors costs in the domestic market (γ d ) are denominated more in local currency, exposure declines though the eect is relatively small (since or the base case most exposure comes rom oreign operations). Panel D reveals the greater sensitivity to the currency o competitors costs in the oreign market (γ ). Speciically, as competitors costs switch rom all domestic to all oreign currency, the irm s exposure increases rom 0.4 to Panel E o Figure 4 shows that as product substitutability increases so does exposure. The clear non-linear relationship highlights the importance o product market competition in determining overall exposure, and in particular, that irms in very competitive industries ace quite high relative exposures. Finally, Panel F plots the relationship between exposure and the degree o import competition. The graph reveals the relatively low sensitivity to the percentage o competition that comes rom imports in the oreign market (λ ). For the ull range o possible values, exposure only varies by about 0.. A similar, but even weaker, relationship holds or the percentage o competition that comes rom imports in the domestic market (λ d ) so we do not plot it here. An important part o overall exchange rate exposure or global irms is determined by how much o exchange rate exposure can be passed through to customers via price changes. In the enhanced BDM model, the relationship between product market characteristics and passthrough can be quite complex. For example, pass-through can decrease or increase with increases in product substitutability. To see this, recall that the impact o product substitutability depends on the values o μ and o μ d. For example, consider Speciically, or the base case we set φ =0.5, γ =0.3, γ d =0.6, γ =0.6, ρ =ρ d =0.7, λ =0.3, and λ d =0.7. 8

19 ( ) ( ) μ = γ γ γ γ, where γ and (-γ ) represent the raction o marginal costs due to oreign currency based inputs or irm and irm, respectively. The smaller the raction o oreign cost o irm and the larger the raction o oreign cost or irm, the larger is the value o μ. Since the model is symmetric, similar arguments apply to μ d, the relative costs in oreign currency or irm s domestic market. Consequently, or a global irm, both the sign and the magnitude o the pass-through depend on the sign and magnitude o the relative costs in oreign currency, μ j. Market share also impacts the size o the pass-through. Considering η, the sign and magnitude o dη / dλ depend on the sign and magnitude o the relative costs in oreign currency, μ, and the magnitude o product substitutability such that dη d ρ μ λ =. Thus, higher market share increases pass-through when μ is negative (i.e. pass-through is more negative and thus larger). In this case, the larger is the raction o oreign cost o irm and the smaller is the raction o oreign cost or irm. I, on the other hand, μ is positive, higher market share decreases pass-through. Altogether, the results in this section show that or a global irm both the sign and the magnitude o the exchange rate elasticity (exposure) and pass-through depend on the sign and magnitude the relative costs in oreign currency, μ j. With the enhanced model in hand, we now turn to a more rigorous analysis o exchange rate exposure using a large sample o global corporations. 9

20 5 Estimation o the Enhanced Model 5. Data Our sample is comprised o,6 irms rom 6 countries. 3 The sample includes all manuacturing irms with accounting data or either the year 000 or 00 on the Thomson Analytics database, that have an annual report in English or the same year on the Global Reports database, and that have at least 5 non-missing weekly stock returns on Datastream during the year o the annual report. 4 Accounting data including inormation on oreign sales and oreign assets originate rom the Thomson Analytics database. To reduce the eect o data errors we exclude observations that all in the top and bottom one percentile or whose value exceeds ive standard deviations rom the median. 5 We collect data on oreign currency debt and derivatives use via an automated search o each irm s iscal year 000 or 00 annual report (see Bartram, Brown and Fehle, 004, or details). We create dichotomous variables or the use o oreign currency debt and oreign exchange rate derivatives usage. Firms are classiied into industries on the basis o 4-digit SIC codes. For the years 999, 000 and 00, data on industry competitiveness, production, exports and imports are collected rom various sources as ollows: We generate measures o import competition using the United Nations Industrial Development Organization (UNIDO) Industrial Statistics Database and the Structural Statistics or Industry and Services (SSIS) database o the OECD. Trade data are used to calculate an import penetration ratio by taking imports as a percent o imports plus domestic production. 6 We also obtain GDP data and calculate or each country and industry a measure o oreign import penetration as weighted averages o the import penetration variables o all oreign 3 The countries represented are Australia, Austria, Canada, Denmark, France, Germany, Indonesia, Ireland, Japan, Mexico, Netherlands, Norway, Singapore, Switzerland, United Kingdom, and the United States. In some parts o the analysis, we are able to expand the set o irms to a total o,34 non-inancial irms rom 9 additional countries (Argentina, Belgium, Brazil, Czech Republic, Finland, Greece, Hungary, India, Italy, Korea, New Zealand, Poland, South Arica, Spain, Sweden, Taiwan, Thailand, Turkey, and Venezuela). 4 Global reports ( is an online inormation provider o public company documents in portable document ormat (PDF). 5 In order to avoid the results being inluenced by the eect o the economic cycle, we use three-year averages o variables where this impact seems most relevant (e.g. gross proit margin). 6 Similar results are obtained by using a permutation o imports as a percent o production plus imports minus exports. However, this measure can have extreme values in countries with large exports in certain industries, so we do not use it in the primary analysis. 0

21 countries. Because the UNIDO data are only available by International Standard Industry Classiication (ISIC, Rev. 3), we calculate or each 4-digit SIC code the mean o the statistics or the corresponding ISIC. In addition, we calculate Herindahl indices using all WorldScope irms with sales data to measure competition at the industry and country level. For all o these measures, we use the value in the year prior to the irm observation. 7 All capital market data (i.e. the irms stock returns, stock index returns, interest rates, exchange rates) are rom Datastream. We create weekly return series to reduce microstructure eects such as bid-ask bounce. For each irm, we calculate stock returns in local currency, local currency returns o the corresponding Datastream national stock market index, and the percentage change in a trade-weighted oreign exchange rate index (in local currency relative to the basket o oreign currencies). All time series are limited to the year o the irm s annual report. We winzorize the return observations in the top and bottom 0.% in order to mitigate some obvious data errors. Consistent with prior research, we use these data to estimate augmented market model (time series) regressions that include returns on exchange rate indices. Speciically, we estimate or each irm R = α + β R + β R +ε, () jt j jm Mt jfx FXt jt where R jt is the stock return in excess o the risk-ree rate, R Mt is the return o the market index in excess o the risk-ree rate, and R FXt is the percentage change o the exchange rate index. The resulting coeicients on the exchange rate variable (β FX ) represent our estimates or actual (irmspeciic) exchange rate exposure. Table 3 reports summary statistics or the ull sample. The irst row reports actual estimated oreign exchange rate exposures obtained rom equation (). Exposures average 0.07 with a standard deviation o.945. The inner quartile runs rom to.058. Because the model we examine does not account or inancial leverage, we multiply the estimated exposure by the market value o equity divided by irm value to obtain estimates o unlevered oreign exchange rate exposure. These values, which are reported in the second row o Table 3, are very similar to the unadjusted exposure values though, o course, somewhat smaller in magnitude. 7 In a ew cases, industry data are not available or the prior year, in which case we use the value rom years prior.