Department of Economics Working Paper Series

Transcription

1 Department of Economcs Workng Paper Seres Increasng Dervatves Market Actvty n Emergng Markets and Exchange Rate Exposure Uluc Aysun Unversty of Connectcut Melane Guld Mount Holyoke College Workng Paper R March 2008, revsed October Mansfeld Road, Unt 1063 Storrs, CT Phone: (860) Fax: (860) Ths workng paper s ndexed on RePEc,

2 Abstract Usng frm level data, we report a sgnfcant fall n the exchange rate exposure of emergng market frms over the past 10 years, and relate ths to hgher dervatves market partcpaton. Our methodology follows a three stage approach. Frst, we measure and report foregn exchange exposures for each year usng the popularzed extenson of the Adler-Dumas (1984) model. Next, we use an ndrect approach to estmate the dervatves market partcpaton at the frm level. Fnally, we nvestgate the mplcatons of the level of dervatve market actvty on a frm s foregn exchange exposure. Our results show that foregn exchange exposure s negatvely related to dervatves usage, and support the hedgng explanaton of the exchange rate exposure puzzle. Journal of Economc Lterature Classfcaton: G15; G32; F31 Keywords: Exchange rate exposure; dervatves; emergng markets We are grateful to Thomas Schneewes and Raj Gupta at the CISDM of UMass and Esen Onur for allowng us to access the data used n ths paper. We would lke to thank Charles Dale and the partcpants of the 2007 WEAI and EEA conferences, and the Colloquum seres at Unversty of Connectcut, Stamford.

3 I. Introducton Followng the Asan and Long Term Captal Management (LTCM) crses, emergng market economes experenced sharp decreases n economc actvty. In contrast, durng the recent credt market related fnancal turbulence these countres have shown more reslency. Improved foregn exchange rsk management n these fnancal markets has been argued to be a major determnant of the muted effect. 1 Another recent development n these economes, whch we focus on n ths paper, s the ncrease n dervatves market actvty followng the fnancal crses of the late 1990 s. 2 The fgures n Table 1 llustrate how dramatcally the aggregate dervatves market volume (DMV) ncreased n emergng market economes and reveal that t has grown much faster when compared to advanced economes. For example, over the perod , growth rates of daly dervatve transacton averages n emergng markets were percent but only 29.9 percent for advanced economes. The remarkable growth n DMV underscores the mportance of dentfyng the role dervatves play n a frm s ablty to manage rsk. Furthermore, the rapd transton to more flexble exchange rate regmes n emergng markets and the growth of global fnancal ntegraton hghlghts the mportance of analyzng the effects of hgher dervatves market actvty on foregn exchange exposure n these economes. Unfortunately, data on frm level partcpaton n dervatves market s not wdely avalable n emergng market economes and therefore t s not possble to analyze the effects of dervatves market partcpaton drectly. Instead, we develop an ndrect method of measurng partcpaton 1 See for example the Aprl 2008 ssues of IMF s GFSR and WEO. 2 See for example: Fnancal Polcy Forum, Specal Polcy Bref, 15 (2004); Futures Industry Assocaton Annual Survey (2003) and Bank of Internatonal Settlements (BIS) (2007). These reports ndcate that some of the emergng market dervatves exchanges (Mexco s Merder, Brazllan BM&F, Korean Stock Exchange) are now ranked among the top 10 dervatves exchanges based on the number of contracts traded. See also studes such as Fratzscher (2006) and Basu (2006). 1

4 n dervatve markets and use ths to gauge whether growth n frms dervatves market partcpaton (DMP) lowers foregn exchange rate exposure n emergng markets. Our methodology s composed of a three stage approach. Frst, we measure and report exchange rate exposures for each year usng the popularzed extenson of the Adler-Dumas (AD) (1984) model. Next, and most mportantly, we use an ndrect approach to estmate dervatves market partcpaton at the frm level. Fnally, we nvestgate the mplcatons of the level of dervatves market actvty on a frm s exchange rate exposure. The lterature offers several explanatons for why a frm would choose to use dervatves. Among these are the desre to reduce cash flow volatlty, operatonal strateges to reduce the effect on frm value, undernvestment, fnancal dstress, tax ncentves and manageral ncentves. 3 The objectve of ths paper s to assess the effects of usng dervatves on exchange rate exposure, not to determne why a frm hedges. The relatonshp between a frm s DMP and rsk management has been studed for advanced economes. Allayanns and Ofek (2001) were the frst to use notonal, contnuous dervatves data at the frm level to study the effects on exposure. 4 They fnd that frms n the Unted States use dervatves to hedge and not to speculate, and that the level of foregn sales and trade s an mportant determnant of the decson to hedge. They obtan these results usng the two stage regresson process of Cragg (1971). Others (Hageln and Pramborg, 2004; Muller and Verschoor, 2006; Pantzals, Smkns and Laux, 2001) also have used a smlar approach to study other regons and dfferent hedgng procedures. Unfortunately, t s mpossble to apply these methods to emergng market economes snce frm level data on DMP s not avalable. The key 3 See Bartram, Brown, and Fehle (2007) for a dscusson of the undernvestment, fnancal dstress, tax ncentve and manageral ncentve theores of dervatve usage. 4 Pror to Allayanns and Ofek (2001), most studes used ether bnary data to ndcate whether frms use dervatves or gathered ths nformaton from surveys as n Smkns and Laux (1997). 2

5 nnovaton of our paper s our utlzaton of a method to ndrectly measure the effect of DMP on foregn exchange exposure that can be used when dervatves data at the frm level s mssng. Few papers n the lterature on foregn exchange exposure analyze emergng markets. 5 Ths s most lkely a result of the lmted data avalablty for these countres. For ths paper, we constructed a data set comprsed of frm-level observatons spannng the years for sx large emergng market economes and add to the scarce body of work related to exchange rate exposure n emergng market economes. Our results show that foregn exchange exposure s negatvely related to a frm s DMP. Ths declne s especally salent n sectors wth more foregn trade, for bgger frms, and over longer tme perods. Results are robust to usng dfferent exchange rates, usng non-lnear and GARCH-SVAR models to measure exchange rate exposure, to employng overlappng observatons, and to dfferng return horzons. The man advantage of a developed foregn exchange dervatves market s the ablty of both foregn nvestors and domestc busnesses to hedge currency rsk. The man drawbacks are potental ncrease n speculatve postons n the absence of strct regulatons, and vulnerablty to hgher volatlty and counter party rsks. One mplcaton of our man result s that n general, the postve effects of usng foregn exchange dervatves outwegh the negatve effects. We fnd that as emergng economy dervatves markets develop, foregn exchange exposure coeffcents decrease and converge to the low levels observed n advanced economes. Ths s nterestng n the context of the exchange rate exposure puzzle as we vew the declne n exposure n response to an ncrease n a frm s DMP as gvng dynamc support for the hedgng explanaton. 6 5 See Chue and Cook (2008); Domnguez and Tesar (2001); Kho and Stulz (2000); Parsley and Popper (2008). 6 Bartram and Bodnar (2007) state that the falure to fnd exposure, despte the effect of unantcpated exchange rate changes on the value of corporatons (exchange rate exposure puzzle) s due to hedgng, and not emprcal methodology or sample selecton related shortcomngs. 3

6 The remander of the paper s organzed as follows: We dscuss the data set n Secton II. In Secton III, we post a model and report results regardng the evoluton of foregn currency exposure over the perod. In Secton IV, we present our ndrect method of estmatng a frm s DMP then we estmate the relatonshp between exchange rate exposure and DMP n Secton V. We perform senstvty analyss and check the robustness of our results n Secton VI and then conclude wth Secton VII. II. Data We compled our data from the Datastream database. Our data set ncludes frms lsted n the stock exchanges of Brazl, Chle, Israel, Korea, Mexco, Turkey, and conssts of daly observatons spannng the perod The number of frms for each country, the breakdown by sector and the stock exchanges are reported n Table 2. The data set conssts of stock prces, sector afflaton, and total assets at the frm level, a composte stock market ndex, and rsk free nterest rates for dfferent maturtes. Data avalablty was the man drver of our choce of countres. A majorty of the lterature usng advanced economy data gauges the sgnfcance of openness on exposure by ncludng the foregn sales varable n regressons. We were only able to fnd ths data for a few frms n Chle and Israel. To consder all sx countres, we classfed the frms nto three categores: hgh trade, low trade and fnancal. Manufacturng frms are classfed nto hgh trade and low trade groups usng data from the Unted Natons Common database. Classfcaton s based on the degree of trade openness at the sector level, and our proxy for ths varable s exports dvded by producton. 7 Data for Israel was avalable only at the weekly frequency. 4

7 We nclude the fnancal sector n our analyss. Although most of the lterature does not nclude the fnancal sector due to ts market-makng property n dervatves and foregn exchange markets, we do so for three reasons. Frst, there s evdence that ths sector has started usng dervatves nstruments to a greater extent for varous reasons. Among these s the desre to hedge foregn currency exposure and to ncrease yeld through hgher leverage (whle becomng more vulnerable to exchange rate volatlty). 8 Table 3 reports the ncrease n domestc fnancal nsttutons drect usage of foregn currency dervatves. Notce that the growth n dervatves market transactons for emergng market countres fnancal sectors are sgnfcantly hgher than country averages reported n Table 1. Second, the fragltes n emergng market bank balance sheets were argued to be one of the man determnants of the severty of the Asan and LTCM fnancal crses. Thrd, the recent fnancal turbulence n credt markets has shown that, despte the market-makng property, off balance sheet exposures of banks can affect ther balance sheets through ther contngent credt lnes to conduts. There s also evdence that foregn nvestors not only the domestc fnancal sector - play a major market-maker role for structured products traded n emergng markets. 9 Therefore, t s mportant to determne the nature of the relatonshp between dervatve usage and foregn exchange rsk n the fnancal sector. Snce there s no consensus n theory as to whch exchange rate s sutable for countres, we frst use Local Currency/US Dollar exchange rates and then examne Local Currency/other currency and trade-weghted exchange rates n our senstvty analyss (Secton VI). Daly aggregate DMV data could only be gathered for Chle, Israel, Korea and Turkey. Ths data s avalable at monthly, quarterly, and annual frequences for Brazl and Mexco. 8 For evdence on usng dervatves to ncrease leverage see IMF, Global Fnancal Stablty Report, October, Table 3 shows that foregn nvestor share of dervatves market transactons s sgnfcantly above 70% and has been ncreasng over the past 6 years. Foregn partcpaton n advanced economes s much hgher partally reflectng fnancal openness. 5

8 However, the number of observatons was nsuffcent n these two countres to apply our ndrect approach to measurng a frm s DMP. Therefore, we show what happens to foregn exchange exposure n all 6 countres, but can only say how a frm s DMP affects exposure for 4 countres. The contents and the sources of the data set are shown n Table A.1 of Appendx A. III. Evoluton of Exchange Rate Exposure In ths secton, we present two ways to measure foregn currency exposure. Frst we examne the relatonshp between the percent change n the exchange rate and a frm s stock return. Second, we examne the relatonshp between exchange rate volatlty (proxed usng the coeffcent of varaton) and a frm s stock return. In the next secton, we present our method of ndrectly estmatng the level of a frm s DMP. Then we use our estmates of exchange rate exposure and estmates of a frm s DMP to analyze the effects of DMP on exchange rate exposure n Secton V. Exposure to Changes n Exchange Rates. To measure foregn currency exposure, we follow the common practce n the lterature and utlze the followng extenson of the AD model: where ( Rmt R ft ) + ERt et R + t R ft = β 0 + β1 β 2 R s the rate of return on frm s common stock, R s the return on the value weghted t mt (1) stock market ndex, ERt s the percent change n the foregn exchange rate, and R ft represents the rsk free rate of return. The regresson model measures the dosyncratc effects of exchange rate volatlty on a frm s stock return. The coeffcent of nterest s β. If t s sgnfcant, we classfy the frm as havng a sgnfcant exchange rate exposure. The market ndex s ncluded to account for the changes n economy specfc factors that are common to every frm. Ths ncludes for example an expansonary monetary polcy that would nflate stock prces and 2 6

9 deprecate the currency concurrently. The foregn exchange rate s measured n Local Currency/US Dollar. We also measure the exchange rate as Local Currency per the Japanese Yen, or Brtsh Pound, or Euro/DM or trade-weghted exchange rates n our senstvty analyss. In contrast to some studes, we dd not nclude exchange rates smultaneously due to the hgh degree of seral correlaton we found among advanced economy currency-local currency exchange rates. When estmatng equaton (1), we use General Method of Moments (GMM). Results are provded n Table 4. The numbers represent the percentage of frms wth sgnfcant exchange rate coeffcents ( ˆ β s) at the 5% level, measured usng robust standard 2 errors. The rows correspond to the dfferent frequences used to measure percent change n stock prces and exchange rates. Top panel of Table 4 shows the proportons of frms that had sgnfcant coeffcents throughout the whole sample perod. We can see that the proportons fall as frequency decreases, possbly due to the lower power assocated wth fewer observatons. In order to overcome the lack of power and make sgnfcance more comparable across dfferent frequences, we use overlappng observatons smlar to Domnguez and Tesar (2006) and Bodnar and Wong (2003). We calculate returns usng daly, weekly, monthly, quarterly, sem-annual, and annual horzons. In dong so, we use daly overlappng observatons for weekly return horzons, and weekly overlappng observatons for monthly, quarterly, sem-annual and annual frequences. 10 We correct for seral correlaton stemmng from the usage of overlappng observatons by employng the Newey and West (1987) method, and use robust standard errors. Results are reported n the bottom panel of Table 4. Beyond the weekly horzon, our fgures are 10 For detaled dscusson on the benefts of usng overlappng observatons see Rchardson and Smth (1991). In addton to weekly frequency, we tred dfferent horzons for overlappng observatons and obtaned smlar results. 7

10 very large when compared to some studes that use advanced economy data. 11 Consstent wth a majorty of the research, we fnd that the proporton of frms that have sgnfcant exposure ncreases wth the return horzon. 12 To observe the evoluton of exchange rate exposure, we repeat our experment for each year n our sample and plot the proporton of frms wth sgnfcant β2 coeffcents n Fgures 1 and Wth the excepton of daly returns, the pattern n the data ndcates a declnng proporton of exposure. Ths pattern s more evdent n regressons wth overlappng observatons beyond the weekly frequency, and Chle, Mexco, and Israel seem to have the most unform declnes n the exposure proportons. Furthermore, the two-proporton z-test results shown n Table A.2 of Appendx A mply that the declne n exposure s sgnfcant. Fracton of frms wth sgnfcant exchange rate coeffcents s not a good ndcator of exposure f the magntude of the coeffcents s economcally nsgnfcant. Hence, we also measure and report the average absolute value of the exchange rate coeffcents. The numbers reported n Table 5 reveal that the sze of exposure has decreased parallel to the declne n the proporton of frms wth sgnfcant exposure. Furthermore, although the magntudes of the exchange rate coeffcents were sgnfcantly hgher than advanced economy coeffcents at the start of the sample perod, they converged to these levels towards the end. 14 The decrease n exposure parallels the development of local bond markets n these economes followng the crss perods of the late 1990 s and early 2000 s. Snce these fnancal 11 Some of the exposure proportons n the lterature for the US are: Joron (1990), 5.2%; Walsh (1994), 5.6%; Prasad and Rajan (1995), 15.0%; Dukas, Fatem and Tavakkol (1996), 5%-8.3%. For other countres: Prasad and Rajan (1995), 4% (JPN), 5.9% (GBR), 16.7 % (DEU); Dodge, Grffn and Wllamson (2006), 8.2% (18 countres). 12 See for example, Domnguez and Tesar (2001, 2006); Bodnar and Wong (2003); Chen and Chow (1998); Chow, Lee and Solt (1997); Jongen, Muller and Verschoor (2007). 13 We omtted perods correspondng to exchange rate fluctuatons exceedng 3 standard devatons n a gven year. Includng these perods generated neglgble dfferences. Tables are avalable upon request. 14 For the U.S., Joron (1990)estmates maxmum exposure to be 0.56 over the perod Domnguez and Tesar (2006) estmate t to be over the perod n 6 non-us advanced economes. Hageln and Pramborg (2002) estmate t to be 0.52 over the perod n Sweden. 8

11 nstruments ncrease the avalablty of foregn currency hedges, the drop n foregn currency exposures can partally be attrbuted to greater degree of hedgng. However, over the same perod these countres have also reduced ther foregn currency share of total debt, thereby elmnatng some of ther exposure to exchange rate fluctuatons. 15 Therefore, t s mportant to dentfy the unque role that foregn exchange hedgng (a frm ncreasng DMP) plays n decreasng foregn exchange exposure as we do n ths paper. Exposure to Exchange Rate Volatlty The evoluton of foregn currency exposure estmated wth equaton (1) assumes that markets have nformaton on frms net foregn assets poston as well as the maturty of ther assets and labltes. However, due to lags n data or uncertanty about off-balance sheet tems, the drecton of foregn currency exposure may not be avalable. Nevertheless, the stock of a frm wth sgnfcant absolute exposure would be traded at a dscount (premum) f exchange rate volatlty ncreased (decreased). Furthermore, several studes have mpled the possblty of a nonlnear relatonshp between exchange rates and stock returns. Stulz (2003) shows that f cash flows are a nonlnear functon of exchange rates, exposures would also be nonlnear. Shftng of producton actvtes to dfferent locatons n response to exchange rate movements s another source of convexty outlned n Kogut and Kulatlaka (1994) and Ware and Wnter (1988). Other explanatons nclude the absence of nonlnear hedgng strateges by corporatons as n Bodnar and Gebhart (1999) default rsk as n Stulz (2003) and prcng to market as n Knetter (1994). There s also evdence that as emergng market countres allow ther currences to float, the resultng hgher exchange rate volatlty ncreases the demand for exchange rate hedgng 15 Source: IMF (2007). Foregn currency share of total debt has dropped approxmately 16% n Brazl between 2002 and 2005, 33% n Indonesa between 1998 and 2005, 17% n Korea between 1997 and 2005, 29% n Mexco between 1998 and 2005, 24% n Thaland between 1997 and 2005, and 25% n Turkey between 2001 and

12 nstruments as suggested by Turner (2002). Therefore, we feel that t s mportant to nclude nonlnearty n the model to gauge the effect of dervatve usage on vulnerablty to volatlty. To measure nonlnear exposure, we extend the AD model by ncludng uncondtonal exchange rate volatlty as an addtonal ndependent varable. We measure exchange rate volatlty usng the coeffcent of varaton whch provdes a unt free proxy for dsperson. More specfcally, we dvde the standard devaton of the exchange rate n a gven perod by the mean value of ths varable and estmate the followng regresson model. t ft er ( Rmt R ft ) + β 2 ERt + β σ t ut R R = + er t β 0 + β1 3 whereσ s the coeffcent of varaton and s our exchange rate volatlty proxy. In ths regresson, we are nterested n the coeffcents on both the exchange rate measure ( β ) and the exchange rate volatlty measure ( β ). To calculate returns we use weekly and monthly horzons wth daly and weekly overlappng observatons. 3 Results from estmatng equaton (2) are reported n Table 6 and show that our estmates of exposure ncreases sgnfcantly when exchange rate volatlty s ncluded n the model. The proporton of frms wth sgnfcant coeffcent estmates for the percent change n the exchange rate and the exchange rate volatlty are reported n the frst and second rows, respectvely. Proportons of frms wth at least one sgnfcant coeffcent are reported n the thrd row. Wth the excepton of Turkey, the declne n exposure was more gradual when we ncluded exchange rate volatlty. Ths pattern s evdent when we compare Fgures 2 and 3. Consstent wth the majorty of the lterature we fnd that the sgn of the exchange rate coeffcent s evenly splt between postve and negatve. However, the sgn of exchange rate volatlty was negatve n most of the cases (mplyng that n general an ncrease n exchange rate volatlty has decreased share values). 2 (2) 10

13 Havng found some evdence supportng a decreasng senstvty to exchange rate movements both n terms of fracton of frms and magntudes, we nvestgate f ths s due to the developments n the dervatves markets. IV. Measurng a Frm s DMP Indrectly Ths secton presents the procedure we use to ndrectly measure a frm s DMP. In the absence of dervatves market actvty at the frm level, t s not possble to measure the effect of dervatves usage on a frm s exchange rate exposure drectly. Nevertheless, we can use a formulaton smlar to that employed n Secton III to estmate the level of DMP at the frm level. More specfcally, we replace the exchange rate varable n equaton (1) wth the aggregate DMV, and estmate the followng: ( Rmt R ft ) + DMVt wt R t R ft = γ 0 + γ 1 γ 2 + (3) where DMV s the dervatves market volume varable dscussed n Secton II and n Appendx t Table A.1. Ths equaton captures the dosyncratc effects of dervatves market volume on a frm s stock return. 16 Hence, we use the coeffcent ˆ2 γ as a proxy for DMP. The ntuton s as follows: On one hand, f a domestc frm takes advantage of the greater depth n the dervatves markets to lower ts exchange rate assocated rsks, demand for ts stock would ncrease, and would be sgnfcant and negatve. On the other hand, f a frm uses these nstruments for hgher leverage to ncrease return, value of ˆ2 γ wll be sgnfcant and postve. For ether case, the absolute ˆ2 γ would measure the senstvty of stock returns to the avalablty of dervatves nstruments and thereby ndcate the degree of DMP. ˆ2 γ 16 We excluded the market premum, mt ft results smlar to those we report n ths paper. R R from equaton (3) to estmate the senstvty to t DMV and found 11

14 Our methodology s smlar n essence to several approaches already taken n the lterature. Senstvty of stock returns to macroeconomc varables has been the subject of numerous studes, especally after Fama (1981) and Chen, Roll and Ross (1986). These studes fnd that macroeconomc varables such as ndustral producton, money supply, nterest rates, nflaton rates have explanatory power over stock returns. 17 Some of these studes also menton that local macroeconomc varables n emergng markets have hgher explanatory power over stock returns compared to advanced economes snce they are partally segmented from global captal markets. The macroeconomc varable we consder here s the dervatves market volume. Related to our analyss, there s a plethora of research that examnes the relatonshp between fnancal development and stock returns. Whle most of these are at the country level, there are few studes that examne stock market returns (e.g. Dellas and Hess, 2005). The unavalablty of dervatves data at the frm level n our data set prevents us from testng the ablty of ˆ2 γ to proxy DMP. However, there s substantal frm level evdence as n Allayanns and Ofek (2001) and Muller and Verschoor (2005) lnkng the amount of foregn sales to the degree of dervatves usage. In lght of ths evdence, we test the performance of our model and measure the proporton of frms wth sgnfcant ˆ2 γ coeffcents n 6 manufacturng sectors. The orderng of the sectors s determned by the average rato of exports to producton between 1995 and 2005 usng Unted Natons Common Database such that sector 1 has the hghest and sector 6 has the lowest exports to producton ratos. 18 Subsectors were combned to generate 6 sectors so that there were at least 20 frms classfed under each sector. We report results from equaton (3) n Table 7. Each cell n the table contans the percentage of frms wthn the correspondng sector n a country that have a sgnfcant ˆ2 γ 17 See for example Schwert (1990); He and Ng (1994); Blson, Bralsford and Hooper (2001); Ibrahm (1999). 18 We have also used mports to producton and exports plus mports to producton ratos and found smlar results. 12

15 coeffcent when we estmate equaton (3). The results are encouragng, and reveal that our proxy for DMP s not unreasonable. Notwthstandng a couple of exceptons, the proporton of frms wth sgnfcant ˆ2 γ coeffcents ncreases as the level of openness ncreases. Nevertheless, we should pont out that the experment provdes only a partal support that stock returns are senstve to DMP and s far from perfect. 19 V. Effects of DMP on Exchange Rate Exposure To measure the effects of DMP on exchange rate exposure we follow three steps: 1. We collect the ˆ β 2 and ˆβ 3 coeffcents estmated for each year n Secton III usng weekly overlappng observatons and a monthly return horzon and retan only the sgnfcant coeffcents We collect our proxes for DMP, ˆ2 γ s estmated for each year n Secton IV usng weekly overlappng observatons and a monthly return horzon and retan only the sgnfcant coeffcents We construct our panel data set usng the exposure and DMP coeffcents measured for each year and frm for a specfc country n steps 1 and 2. Next, we regress the exposure coeffcents on the market partcpaton proxy as follows: ˆ β + ν (4) t = λ + λ ˆ 2 0 1γ 2t t We use the absolute values of βˆ t and γˆ t n equaton (4). A negatve sgn of ˆλ 1 ndcates that foregn exchange exposure decreases when a frm partcpates more n the dervatves market. We also ncluded the foregn sales/total sales rato n equaton (4) to control for the degree of foregn operatons. Foregn sales data was only avalable for a few frms n Chle and Israel, and ts ncluson dd not change the results sgnfcantly for these countres. Pagan (1984) proves that n models wth generated regressors smlar to that n equaton (4), f the hypothess to be tested s λ = 1 0, the OLS estmator of the varance of ˆλ 1 s consstent, 19 Due to the lack of frm level data, we were unable to nclude foregn sales as an explanatory varable. 20 We have also estmated exposures usng levels and volatlty separately. Results were qualtatvely smlar. 21 Usng daly overlappng observatons and a weekly return horzon yelded smlar results. 13

16 and the asymptotc t-statstcs are vald. Furthermore, some studes have used maxmum lkelhood estmaton to deal wth the generated regressor problem (e.g. McAleer and Mackenze, 1991, 1994). Fnally, Arrelano and Bover (1995) propose a GMM strategy for panel data that s superor to OLS when the dependent varable s not statonary, there are unobserved frm specfc effects, and there s reverse causalty (when a frms stock return affects the return on the market ndex and/or the exchange rate). In our experments we have used all three of the methods mentoned above. Snce results were smlar, we report the output from only the GMM estmaton of equaton (4) n Table 8. Reported coeffcent values measure the percent change n exchange rate exposure n response to a one percent change n the senstvty to dervatves market volume. We make three observatons based on the results. Frst, all of the sectors wth low levels of foregn trade (low exports/producton ratos) have nsgnfcant coeffcents, and a majorty of the hgh trade sector coeffcents are sgnfcant and negatve. Hence, foregn currency exposure of frms classfed under sectors that trade heavly s more senstve to DMP. The nsgnfcance for low-trade frms can be partally explaned by the fewer number of sgnfcant exchange rate and DMV coeffcents n the frst and second stage regressons that gve us fewer observatons n the thrd stage. Ths result s n lne wth the evdence ndcatng hgher usage of dervatves by frms wth more foregn sales n Allayans and Ofek (2001), Nydahl (1999), Muller and Verschoor (2006). Our second fndng s that the fnancal sector partcpaton coeffcents are negatve, hgher than the hgh trade sector coeffcents n absolute value, and more sgnfcant. Ths observaton provdes support for the numbers reported n Table 3. Conversely, the hgher values of coeffcents mply that fnancal frms were more effcent n reducng ther foregn currency exposures by usng dervatves compared to non-fnancal frms. Furthermore, the negatve 14

17 coeffcents mply that dervatves market transactons of these frms were predomnantly due to hedgng motves rather than speculaton. Fnally, we fnd that the exposure to volatlty s less senstve to DMP. More specfcally, Table 8 shows that sgnfcant exchange rate volatlty exposure coeffcents are fewer n number and smaller n sze compared to the coeffcents of exchange rate levels. VI. Robustness In ths secton, we test the robustness of our fndngs to consderng dfferent exchange rates, model specfcatons, perod lengths and frm szes. Results are dsplayed n Appendx B. Usng Dfferent Exchange Rates -- EMAX In ths secton, we replcated our smulatons usng a dfferent measure of exchange rate constructed as follows: We ran separate regressons wth the exchange rate measured as Local Currency per Japanese Yen, or Brtsh Pound, or Euro/DM, or US Dollar, or a trade-weghted exchange rate. Usng the output of these regressons, for each frm we dentfed the currency that generates the maxmum exposure and referred to ths as EMAX. In our thrd stage regressons we used the ˆ β and ˆβ coeffcents correspondng to EMAX. Our experments usng 2 3 EMAX produced slghtly hgher senstvty of exposure to the partcpaton proxy for exchange rate levels compared to our man results n Table 8. However, the dfferences n exchange rate volatlty coeffcents estmated from regressons ncludng USD and EMAX were neglgble. The rest of Secton VI dscusses our senstvty experments usng EMAX. GARCH-SVAR The analyss so far has the followng caveat: Exchange rate fluctuatons, the change n the exchange rate, and the market return are hghly correlated. Ths n turn makes the coeffcent estmates unrelable. Therefore, t s mportant to dentfy the shocks to volatlty and to the 15

18 exchange rate that are ndependent of the shocks to other varables. In ths respect, we follow a three step procedure smlar to Jorda and Salyer (2003). Frst, we estmate a Vector Auto Regressve (VAR) model wth the followng orderng: [ ER ] R usng weekly returns and m R lags up to two perods. We assume that ndvdual stock returns do not affect macroeconomc varables. The other restrcton we need n order to obtan the orthogonalzed shocks s the ndependence of market returns from the change n exchange rates. To do ths, we use the resduals from a regresson of market returns on exchange rates, and consder the component of market returns that are not affected by exchange rates by constructon. Next, we ft a GARCH(1,1) model to resduals from the exchange rate equaton, and estmate the condtonal varances, er σˆ. Fnally we ncorporate ths condtonal exchange rate volatlty n a VAR model er wth the followng orderng: [ R ER R ] σˆ. m We are partcularly nterested n the effects of the exchange rate and exchange rate volatlty on a frm s stock returns. It s straghtforward to gauge ths effect usng decomposton of varance from our model. We measured the varance decompostons for forecast horzons 1 to 10 weeks ahead for each frm and constructed our two addtonal foregn exchange exposure proxes (SVAR proxes): percentage of varaton n stock returns explaned by exchange rates or by condtonal exchange rate volatlty. Despte slght dfferences, these two ratos were stable over dfferent forecast horzons. Therefore, we used the results from 3 perod ahead forecasts. Usng these proxes n our thrd stage regressons we fnd, smlar to our man results n Table 8, that coeffcents of the partcpaton proxy are negatve n general. The dsparty between fnancal, hgh trade and low trade sectors, however become much more apparent wth these alternatve proxes. Furthermore, these coeffcents were more sgnfcant compared to those obtaned usng the AD model proxes. Ths s partally due to the fact that every annual 16

19 observaton for each frm s ncluded n the thrd stage unlke n our man model. Hence, the larger number of observatons causes a drop n the standard errors. We also fnd, consstent wth our man results that the senstvty of exposure to DMP s larger n Korea compared to others. To elmnate the effect of power n ths dsparty, we followed two methods. Frst, we consdered only annual data wth varance ratos over certan values so as to equate the number of observatons n SVAR and AD proxes. Second, we consdered all the observatons n the SVAR proxes. Results were smlar. Overall, the negatve effect of DMP on exchange rate exposure was robust to a more precse method of measurng the effect of exchange rate fluctuatons on stock returns. Dfferent Perod Lengths So far we used exposures and dervatve market partcpaton coeffcents measured annually n the thrd stage. Alternatvely, we extend the tme perod to three years. Ths modfcaton ncreases the number of observatons and therefore the power of our estmaton results n the frst and second stage. However, the power of the results n the thrd stage decreases as we use lower frequences. Our smulatons produced a sharp ncrease n the sze and sgnfcance of the coeffcents. Furthermore, exchange rate volatlty that was nsgnfcant n regressons usng annual data was more sgnfcant and larger n sze when the perod length was ncreased. These results are not surprsng snce the growth n the dervatve market volume varable s much more apparent over a three year perod. Sze Classfcaton A majorty of the studes measurng exchange rate exposure for dfferent sze frms fnds that larger frms are more exposed to exchange rate fluctuatons than smaller frms. 22 Usng data 22.See Parsley and Popper (2002); Kho and Stulz (2000); Joron (1990); Nance, Smth and Smthson (1993). 17

20 for Chle the only country wth sze data, we test how foregn exchange exposure senstvty to DMV s related to a frm s sze. We dvde the frms nto three equal sze groups, and mplement our experment for each of them usng weekly overlappng observatons and a monthly return horzon. 23 Our smulaton results showed that larger frms partcpate more n the dervatves markets. Ths result s consstent wth a number of studes that fnd evdence supportng the fxed start up costs n hedgng theory. 24 Second, we found that large frms reduced ther exposure to a greater extent over the sample perod compared to smaller frms. Fnally, the declne n exposure to exchange rate volatlty was bgger for smaller frms. The last observaton can possbly be nterpreted as the ablty of smaller frms to sgnal to the market that they are hedgng some of ther less transparent foregn exchange exposures. In contrast, bgger frms drecton of exposure was more transparent, and the declne n ther foregn exchange exposure was more apparent when exchange rate changes were used. VII. Concluson The rapd pace of fnancal ntegraton and nnovaton n emergng markets has mproved the effcency of fnancal ntermedaton, and enabled frms to dversfy rsks. However, these developments have also ntroduced challenges to dentfy the underlyng rsks n these markets. The negatve consequences of these rsks were evdent wth the recent credt market related fnancal turmol. Gven the complex and less transparent nature of off-balance sheet transactons, t s essental to develop methods to gauge these rsks. 23 Share of a frm s total assets n the ndustry total s averaged for the perod to classfy the frms. 24 See Geczy, Mnton and Schrand (1997); Froot, Scharfsten and Sten (1993); Allayanns and Ofek (2001); Mnton, Stulz and Wllamson (2008); Bartram, Brown and Fehle (2007). 18

21 In ths paper, we dentfy a declnng trend n emergng market frms exchange rate exposure usng varous measures. Next, we nvestgate whether the growng scale of dervatves market actvty n emergng market countres s partally responsble for ths observaton. Gven the absence of dervatves data at the frm level, we propose and mplement a method to ndrectly measure frm level dervatves market partcpaton, and measure the mplcatons for exchange rate exposure. Our focus on emergng markets s due to several reasons. Frst, these countres have recently been pursung more actve monetary polces and more flexble exchange rate regmes. Hence, understandng how frms cope wth exchange rate uncertantes s mportant for assessng the rsks to fnancal stablty n these markets. Our fndng that these economes are much more exposed to exchange rate fluctuatons justfes our choce. Second, together wth the low volatlty n the world, these countres have experenced large captal nflows n the last fve years. Ths flow of captal brought together wth t demand for more complex fnancal nstruments. In our opnon, t s crtcal to dentfy how domestc frms fare from ths growth n fnancal nnovaton. Furthermore, hgher captal ntegraton among emergng and advanced economes, and emergng markets growng share of world output underscore the emphass. Our paper fnds evdence that the proporton of frms wth exposure has decreased over tme and that ths declne s due to hgher level of actvty n dervatves markets. However, these fndngs do not mply that a sharp reversal of captal, especally n less lqud markets, wll not lead to a fnancal crss. References: Adler, M., and B. Dumas. Exposure to currency rsk: Defnton and measurement. Fnancal Management, 13, 1984,

22 Allayans, G., and E. Ofek. Exchange rate exposure, hedgng and the use of foregn currency dervatves. Journal of Internatonal Money and Fnance, 20, 2001, Arellano, M., and O. Bover. Another Look at the Instrumental-Varable Estmaton Components Models. Journal of Econometrcs, 68, 1995, Bartram, S.M., G.M. Bodnar. The exchange rate exposure puzzle Manageral Fnance, 33(9), 2007, Bartram, S.M., G.W. Brown, and F.R. Fehle. Internatonal Evdence on Fnancal Dervatve Usage. Fnancal Management. Forthcomng. Basu, S., and M. Bappadtya. Dervatves n Asa-Pacfc Markets. Journal of Emergng Market Fnance, 5, 2006, 207. Blson, C.M., T.J. Bralsford, and V.J. Hooper. Selectng macroeconomc varables as explanatory factors of emergng stock market returns. Pacfc-Basn Fnance Journal, 9, 2001, Bodnar, G.M., and G. Gebhardt. Dervatves Usage n Rsk Management by U.S. and German Non-Fnancal Frms." Journal of Internatonal Fnancal Management and Accountng, 30, Bodnar, G. M., and M.H.F. Wong. Estmatng exchange rate exposure some weghty ssues. Fnancal Management 32, 2003, Chen, N.F., R. Roll and S.A. Ross. Economc forces and the stock market. Journal of Busness, 59, 1986, Chen, H.L. and E.H. Chow. The determnants of foregn exchange rate exposure: evdence on Japanese frms. Pacfc-Basn Fnance Journal, 6, 1988, Chow, E.H., W.Y. Lee, and M.E. Solt. "The economc exposure of US multnatonal frms. The Journal of Fnancal Research 20, 1997, Chue, T.K., and D.E. Cook. Emergng Market Exchange-Rate Exposure. Journal of Bankng and Fnance, 32(7), 2008, Cragg, J. Some statstcal models for lmted dependent varable wth applcaton to the demand of durable goods. Econometrca, 39, 1971,

23 Dellas, H., and M. Hess. Fnancal Development and Stock Returns: A Cross-Country Analyss. Journal of Internatonal Money and Fnance, 24(6), 2005, Dodge, C., J. Grffn, and R. Wllamson. Measurng the economc mportance of exchange rate exposure. Journal of Emprcal Fnance, 4(5), 2006, Domnguez, K.M.E., and L.T. Tesar. A re-examnaton of exchange rate exposure. Amercan Economc Revew 91, 2001, Exchange Rate Exposure. Journal of Internatonal Economcs, 68, 2006, Dukas, S.P., A.M. Fatem, and A. Tavakkol. Foregn exchange rate exposure and the prcng of exchange rate rsk. Global Fnance Journal, 7(2), 1996, Fama, E.F. Stock returns, real actvty, nflaton and money. Amercan Economc Revew, 71(4), 1981, Fratzscher, O. Emergng Dervatve Market n Asa. Chapter for EAP Flagshp on Asan. Fnancal Market Development, The World Bank, Froot, K., D. Scharfsten, and J. Sten. Rsk management: coordnatng corporate nvestment and fnancng polces. Journal of Fnance, 48, 1993, Geczy, C., B. Mnton, and C., Schrand. Why frms use currency dervatves? Journal of Fnance 52, 1997, Hageln, N., and B. Pramborg. Hedgng Foregn Exchange Exposure: Rsk Reducton from Transacton and Translaton Hedgng. Journal of Internatonal Fnancal Management and Accountng, 15(1), 2004, He, J., and L.K. Ng. Economc Forces, Fundamental Varables, and Equty Returns. The Journal of Busness, 67(4), 1994, Ibrahm, M. Macroeconomc Varables and Stock Prces n Malaysa: An Emprcal Analyss. Asan Economc Journal 13(2), 1999, Internatonal Monetary Fund. The development of bond markets n emergng market countres. Note prepared for the G7 Deputes meetng, February, Jongen, R., A. Muller, and W.F.C, Verschoor. Usng Survey Data to Resolve the Exchange Rsk Exposure Puzzle: Evdence from U.S. Multnatonal Frms. workng paper EFA,

24 Jorda, O., and K. Salyer. The Response of Term Rates to Monetary Polcy Uncertanty. Revew of Economc Dynamcs, 64, 2003, Joron, P. The exchange rate exposure of U.S. multnatonals. Journal of Busness, 63(3), 1990, Kho, B., and R.M. Stulz. Banks, the IMF and the Asan Crss. Pacfc-Basn Fnance Journal, 8(2), 2000, Knetter, M.M. Is Export Prce Adjustment Asymmetrc? Evaluatng the Market Share and Marketng Bottlenecks Hypothess. Journal of Internatonal Money and Fnance, 13, 1994, Kogut, B., and N. Kulatlaka, N. Operatng flexblty, global manufacturng, and the opton value of a multnatonal network. Management Scence 40(1), 1994, Lel, U. Currency hedgng and corporate governance: a cross-country analyss. Internatonal Fnance Dscusson Papers 858, Board of Governors of the Federal Reserve System, McAleer, M., and C.R. McKenze. When are two step estmators effcent? Econometrc Revews, 10, 1991, On the effects of msspecfcaton errors n models wth generated regressors. Oxford Bulletn of Economcs and Statstcs, 56, 1994, Mnton, B.A., R.M. Stulz, and R.G. Wllamson. "How Much Do Banks Use Credt Dervatves to Hedge Loans?" Workng Paper Seres , 2008, Oho State Unversty. Muller, A., and W.F.C. Verschoor. The Impact of Corporate Dervatve Usage on Foregn Exchange Rsk Exposure. Workng Paper European Foregn Exchange Rsk Exposure. European Fnancal Management, 12(2), 2006, Nance, D.R., C. Smth, and C.W. Smthson. On the determnants of corporate hedgng. Journal of Fnance, 48, 1993, Newey, W.K., and K.D. West. A Smple, Postve Sem-Defnte, Heteroskedastcty and Autocorrelaton Consstent Covarance Matrx. Econometrca, 55(3), 1987, Nydahl, S. Exchange Rate Exposure, Foregn Involvement and Currency Hedgng of Frms: Some Swedsh Evdence. European Fnancal Management, 5, 1999,

25 Pagan, A. Econometrc Issues n the Analyss of Regressons wth Generated Regressors. Internatonal Economc Revew, 25, 1984, Pantzals, C., B.J. Smkns, and P.A. Laux. Operatonal hedges and the foregn exchange exposure of US multnatonal corporatons. Journal of Internatonal Busness Studes 32, 2001, Parsley, D.C., and H. Popper. Exchange Rate Pegs and Foregn Exchange Exposure n East and Southeast Asa. Journal of Internatonal Money and Fnance, 2008, Forthcomng. Prasad, A.M., and M. Rajan. The Role of Exchange and Interest Rsk n Equty Valuaton: A Comparatve Study of Internatonal Stock Markets. Journal of Economcs and Busness, 47(5), 1995, Rchardson, M., and T. Smth. Tests of Fnancal Models n the Presence of Overlappng Observatons. Revew of fnancal Studes 4, 1991, Schwert, G.W. Stock returns and real actvty: A century of evdence. Journal of Fnance, 45, 1990, Smkns, B., and P. Laux. Dervatves use and the exchange rate rsk of nvestng n large U.S. corporatons. Workng Paper, Case Western Reserve Unversty, Stulz, R.M. Rsk Management and Dervatves. Southwestern Publshng Co., Cncnnat, Turner, P., The Development of Bond markets n emergng economes: An overvew of polcy ssues. Bank of Internatonal Settlements, Papers no: 11, Walsh, E.J. Operatng ncome, exchange rate changes and the value of the frm: an emprcal analyss. Journal of Accountng, Audtng and Fnance, 9, 1994, Ware, R., and R. Wnter. Forward Markets, Currency Optons and the Hedgng of Foregn Exchange Rsk. Journal of Internatonal Economcs 25(3/4), 1988,

26 Table 1. Dervatves Market Volume (Daly Averages, Mllons, US Dollar) Brazl 1,881 1, Chle ,967 Colomba Czech Republc 2,998 1,245 1,429 3,631 Hungary ,141 4,658 Inda 1,290 1,848 3,457 24,015 Indonesa 1, ,355 1,357 Israel 414 2,274 4,812 Korea 1,046 3,950 10,269 17,819 Malaysa ,812 Mexco 2,397 4,186 4,543 10,795 Peru Phlppnes ,256 Poland 541 3,341 4,604 6,820 Russa ,153 16,190 Slovaka 497 1,459 3,246 Slovena South Afrca 5,206 7,858 8,032 10,568 Tawan 1,524 1,669 4,636 6,725 Thaland 2,228 1,315 1,979 4,931 Turkey 678 2,232 3,311 Emergng Countres Average 1,520 1,527 2,774 5,978 Industralzed Countres Average 50,640 42,757 62,995 81,831 Emergng Countres Total 23,271 32,967 55, ,717 Industralzed Countres Total 1,337,907 1,186,071 1,757,614 1,800,271 Source: BIS Trannual Survey (2007). OTC outrght forward, swap, optons and other foregn currency dervatves. Note: Datastream, the source of the data used to conduct the analyss n ths paper, does not contan daly nformaton at the frm-level from all of the countres lsted above for the perod consdered n ths paper. Suffcent data was avalable only for: Brazl, Chle, South Korea, Mexco, and Turkey. 24

27 Table 2. Number of Frms by Sector Brazl Chle Israel Korea Mexco Turkey Sao Paulo Stock Exchange (BOVESPA) Santago Stock Exchange (SSE) Tel Avv Stock Exchange (TASE) Korea Stock Exchange (KSE) Bolsa Mexcana de Valores (BMV) Istanbul Stock Exchange (ISE) Alumnum Auto Parts Automobles Banks Brewers Buldng Mat.& Fx Busness Support Svs Comm. Vehcles,Trucks Commodty Chemcals Contaners & Package Electrcal Equpment Electrcty Exploraton & Prod Farmng & Fshng Fxed Lne Telecom Food Products Food Retal,Wholesale Footwear Forestry Gas Dstrbuton Heavy Constructon Industral Machnery Investment Servces Paper Pharmaceutcals Ppelnes Real Estate Hold, Dev Specalty Chemcals Specalty Fnance Steel Tobacco Unclassfed Total Source: Authors computaton usng frm-level data from , compled from Datastream. 25

28 Table 3. Fnancal Sector s and Foregn Investor s Dervatve Usage % Foregn Domestc Fnancal % Foregn Domestc Fnancal % Foregn Emergng Advanced Percent Foregn represents the Percent Non-local nsttutons and dealers share of OTC outrght forward and swap transactons. Domestc Fnancal Columns show the percentage ncrease n outrght forward and swap transactons of the domestc fnancal sector over the past three years. Source BIS. Table 4. Percentage of Frms wth Sgnfcant Exchange Rate Exposure ( ) Brazl Chle Israel Korea Mexco Turkey Daly Weekly Monthly Quarterly Overlappng Observatons Daly Weekly Monthly Quarterly Sem Annual Annual The numbers represent the percentage of frms that have a sgnfcant β2 coeffcent at the 5% level n the followng regresson: R t R ft = β 0 + β1 ( Rmt R ft ) + β 2 ERt + et. Rows correspond to the dfferent horzons used. Local Currency/US Dollar exchange rate s used n the estmaton. In regressons usng overlappng observatons, daly overlappng observatons are used for the weekly return horzons and weekly overlappng observatons are used for monthly, quarterly, sem-annual, and annual return horzons. 26

29 Table 5. Average Exchange Rate Coeffcents Brazl Chle Israel Korea Mexco Turkey Fgures represent the smple average of the absolute value of ˆβ coeffcents estmated from the followng regresson: R t R ft = β 0 + β1 ( Rmt R ft ) + β 2 ERt + et. The coeffcents represent the percent change n stock returns n response to a one percent change n exchange rates. Weekly return horzons are used n estmaton. 2 Table 6. Percentage of Frms wth Sgnfcant Exposure ( ) Brazl Chle Israel Korea Mexco Turkey Weekly ER ER_Vol Ether Coeff Monthly ER ER_Vol Ether Coeff % of Neg ER % of Neg ERVol Exchange rate volatlty s proxed by the standard devaton of the US Dollar exchange rate dvded by ts mean value n the respectve perod. Weekly horzons wth daly overlappng observatons and monthly horzons wth weekly overlappng observatons are used to calculate the returns. Coeffcent rows report the % of frms wth a er sgnfcant β and/or 2 3 R R = + β R R + β ER + β σ + e β coeffcents (at 5%) estmated from: t ft β0 1 ( mt ft ) 2 t 3 t t 27