Firm Survival, Performance, and the Exchange Rate*

Transcription

1 Department of Economcs Dscusson Paper Frm Survval, Performance, and the Exchange Rate* Jen Baggs Faculty of Busness Unversty of Vctora Vctora, BC V8W 2Y2 Eugene Beauleu Department of Economcs Unversty of Calgary Calgary, AB T2N 1N4 and Loretta Fung Department of Economcs Unversty of Alberta Edmonton, AB T6G 2E1 Ths paper can be downloaded wthout charge from

2 Frm Survval, Performance, and the Exchange Rate 1 Jen Baggs, Eugene Beauleu + and Loretta Fung February 27, 2007 Abstract: Large real exchange rate movements may affect the real economy n a manner comparable to sgnfcant epsodes of tarff changes. Ths paper conducts a frm level analyss of the mpact of exchange rate movements on frm survval, sales and ndustry entry rates. Our emprcal analyss explots a detaled set of Canadan frm level data from 1986 to 1997 durng whch the Canadan dollar experenced an approxmately 30% apprecaton n the frst sx years and a 30% deprecaton n the later sx years. These recprocal movements n the Canadan dollar offer an excellent opportunty to nvestgate the consequences of large exchange rate movements for domestc frms. We fnd that survval, sales and entry are negatvely assocated wth apprecatons n the Canadan dollar. The mpact on survval s less pronounced for more productve frms. We fnd that over our sample perod, the magntude of the mpact of exchange rate changes on frm survval, sales and entry were consderably larger than the effect of CUSTA mandated tarff changes. JEL Classfcaton: F1; Key words: Exchange-Rate, Monopolstc Competton, Frm Turnover, Productvty 1 We are extremely grateful to Statstcs Canada, n partcular the BLMA dvson, for provdng us wth the data used n ths paper. Ths paper has benefted from comments by Bob Fey, Danny Leung, John Murray, Herb Emery, Francsco Gonzalez, Scott Taylor, Robert Feenstra, Deborah Swenson, Stuart Landon, Sébastan LaRochelle-Côté and Constance Smth. We thank Stephen Yeaple for very helpful comments and acknowledge comments from partcpants at the 2006 EIIT Conference and at the Bank of Canada. Fnancal supports from the Unversty of Alberta Faculty of Arts Support for the Advancement of Scholarshp Fund are gratefully acknowledged. Any errors and omssons are our own. Faculty of Busness, Unversty of Vctora. emal: jenbaggs@uvc.ca + Department of Economcs, Unversty of Calgary. emal: beauleu@ucalgary.ca Department of Economcs, Unversty of Alberta. emal: loretta.fung@ualberta.ca 1

3 1 Introducton Large real exchange rate movements may affect the sales and the survval (entry and ext) of frms comparable to the effects of sgnfcant epsodes of tarff changes as ponted out by Feenstra (1989). There are a number of studes focusng on the mpact of trade lberalzaton on frm and ndustry performance. Recent lterature has shown that trade lberalzaton can affect aggregate productvty growth by changng the turnover and producton decsons of heterogeneous frms (plants). However, despte ther proportonally larger fluctuatons, the effects of large real exchange rate movements are rarely examned. Where exchange rates are consdered, ther mplcatons are most often analyzed at the country or ndustry level, not at the frm level. Ths paper conducts a frm level analyss of the mpact of exchange rate movements on frm survval, sales and ndustry entry rates. Our emprcal analyss explots a detaled set of Canadan frm level data coverng a twelve year perod from 1986 to 1997 durng whch the nomnal blateral Canadan exchange rate vs-à-vs the US dollar apprecated approxmately 30% over the frst sx years and deprecated 30% durng the later sx years. These recprocal fluctuatons n the Canadan dollar offer an excellent opportunty to nvestgate the consequences of large exchange rate movements for domestc frms. Moreover, the Canadan economy s an excellent lens through whch to nvestgate the mplcatons of large exchange rates movements for domestc frms. In Canada, vrtually every manufacturng frm s ether export-orented or mport-competng, and many fall nto both categores. The manufacturng sector exports about 40% of ts total producton to the Unted States alone and about 35% of manufacturng output consumed n Canada s mported from the Unted States. The large Canadan exchange rate movements durng ths perod provde a superb natural experment to examne the mpact of such large exchange rate movements on busness actvty. Although t s possble that large movements n the real exchange rate may have had mportant consequences for the Canadan economy, there s lttle emprcal evdence of ths. Harrs (2001) argues that the large deprecaton of the Canadan dollar n the 1990s may have contrbuted to the wdenng productvty gap between Canada and the US by rasng the prce of nvestment goods, by wdenng the nnovaton gap, and by slowng down the creatve destructon. Yan (2002) fnds that real exchange rate movements precede movements n the US-Canada productvty dfferental by one to two years, ndcatng the possblty that exchange rate movements may affect relatve productvty growth. Yet the mechansm of how exchange rate movements affect frm sales and survval and ndustry productvty s not well understood. Indeed, t s surprsng that researchers have focused consderable effort on examnng the mpacts of much smaller tarff changes durng ths perod rather than on changes n the real exchange rate. Even hstorc trade agreements, such as the Canada-US Free Trade Agreement (CUSTA), tend to produce rather modest tarff changes that are orders of magntude smaller than 2

4 exchange rate movements. Drven by the FTA mandated tarff reductons, from 1986 to 1998 the average Canadan tarff on US manufacturng mports fell from 6.9 percent to zero, whle the average Amercan tarff on Canadan manufacturng mports declned from 3.9 percent to zero. 2 Durng the same tme frame, the Canadan dollar fluctuated between $0.63 US and $0.89 US, wth both apprecatons and deprecatons near 30%. Curre and Harrson (1997) argue that the small tarff changes mandated by the Canada-US Free Trade Agreement make t an unlkely place to examne tarff changes. Nonetheless, Trefler (2004) fnds large effects from these relatvely small tarff changes. 3 Trefler (2004) fnds that the FTA led to a 12 percent employment loss for the most mport-competng ndustres and led to a 14 percent ncrease n productvty at the plant level among the most export-orented ndustres. If these relatvely smaller tarff reductons had large effects on ndustry productvty, t s reasonable to expect that the much larger exchange rate fluctuatons may also have had substantal consequences. It s mportant to note, however, that the fluctuatons of exchange rates are temporary and the elmnaton of tarffs through CUSTA was a permanent polcy change to whch frms had to adjust on a permanent bass. Due to the temporary nature of exchange rate fluctuatons, t mght well be that frms do not react to changes n exchange rates. Therefore, a frm s response to large exchange rate movements s the mportant emprcal queston that we address n ths paper. Most of the studes examnng tarff changes nclude the exchange rate as a control varable. In the ndustry-level research on the effects of the Canada-US free trade agreement, Gaston and Trefler (1997) and Beauleu (2000) fnd an exchange effect n the labour market, whle Head and Res (1999) fnd an effect on the number of plants but not on output per plant. Some ndustry-level studes have focused on exchange rates. Bodnar and Gentry (1993) examne the mpact of the exchange rate on average ndustry stock market returns n Canada, Japan and the US and fnd evdence that the exchange rate affects ndustry returns n all three countres. Campbell and Lapham (2004) examne the mpact of the exchange rate on the number of frms (the extensve margn) and the average sze of frms (the ntensve margn) n four retal trade sectors located n border countes. The exchange rate affects the number of one day vsts from Canada and they fnd that the exchange rate affects both the number of frms and the average sze of the frms n three retal ndustres: food gas statons and restaurants. They fnd that employment, but not the number of frms, s affected n the drnkng ndustry where entry s restrcted through regulaton. Although these ndustry-level studes provde some mportant nsghts, they are not able to capture the heterogeneous nature of frms. The dfferent response of heterogeneous frms to exchange rate movements s mportant because frm behavour and market share reallocaton 2 Beauleu (2000), Head and Res (1999). 3 Trefler (2004, p. 872) argues that the 1988 tarff was ndeed hgh based on the observaton that tarffs n excess of 10 percent exsted n 25 percent of ndustres that were characterzed by low wages, low captallabour ratos and low proft margns. 3

5 among frms may lead to productvty growth at the ndustry level. 4 At the frm level, whle several studes 5 have examned the consequences of trade polcy for frm productvty, survval and growth, few studes have conducted a systematc analyss of the consequences of large exchange rate movements n otherwse stable economes. 6 A recent excepton s Domnguez and Tesar (2006) who examne the relatonshp between the exchange rate and the value of publcly traded frms n eght countres, most of whch are stable, ndustralzed natons. They fnd that the profts of a sgnfcant fracton of frms are affected by the exchange rate, but that there s consderable heterogenety n ther exposure. Ths agan underles the mportance of frm level analyss. Ths paper ncorporates the productvty aspects of Meltz (2003) and Meltz and Ottavano (2005) nto Fung s (2006) theoretcal model of exchange rates and frm dynamcs to develop a set of emprcal predctons concernng the mplcatons of exchange rate movements for frm survval, entry and sales. We emprcally examne these hypotheses usng Canadan mcro-level data. Our fndngs suggest that the real effectve exchange rate had substantal and sgnfcant mplcatons for the Canadan manufacturng ndustry. Both frm sales and the probablty that Canadan frms survve from one perod to the next are found to be negatvely related to apprecatons n the Canadan dollar and to hgher levels of the Canadan dollar. These effects are smaller for frms wth hgher levels of labour productvty. Deprecatons n the Canadan dollar are assocated wth mproved survval and sales, whle ndustry entry rates are negatvely related to apprecatons, and postvely related to deprecatons of the Canadan dollar. Further, we contrast the effect of exchange rates to the effect of tarff reductons and fnd that, durng our sample perod, the effect of the exchange rate on survval, entry and sales was consderably larger than the effect of CUSTA mandated tarff changes. Ths paper contrbutes to a growng lterature n nternatonal economcs that provdes compellng evdence that t s mportant to consder frm heterogenety when examnng the mpact of nternatonal shocks on domestc economes. Ths lterature has generally focused prmarly on tarffs and has gnored the role of exchange rate movements n understandng frm-level outcomes n the face of nternatonal shocks. Untl now the emprcal work that does examne the mpact of exchange rates on frm performance has focused on developng countres and nvolved very large exchange devaluatons n relatvely unstable economes. As far as we know, there have been no 4 Several studes of trade lberalzaton, such as Pavcnk (2002) and Trefler (2004), dscuss the effects that trade lberalzaton may have on ndustry-level productvty growth by affectng the entry, ext, producton, and market share reallocaton among frms that are heterogeneous n productvty. 5 See for example: Baldwn and Gu (2003, 2004), Trefler (2004), Baggs (2005) and Baggs and Brander (2006). 6 Forbes (2002) examnes the mpact of currency devaluatons on the growth of frm output and operatng profts across 51 countres (some of them have experenced currency crses). She fnds that frms n countres that have experenced devaluatons have hgher output growth and hgher operatng proft growth. 4

6 other studes lookng at the mpact of real exchange rate changes on survval n developed countres wth a flexble exchange rate system. Secton 2 summarzes the theoretcal motvaton for ths study. Secton 3 descrbes the data employed and secton 4 presents the emprcal fndngs. The fnal secton offers concludng remarks. 2 Theoretcal Motvaton The extensve margn: Frm Survval and Frm Entry Fung s (2006) model bulds on the Krugman (1979) monopolstc competton model by ncludng an exchange rate varable. A summary of Fung s (2006) model s sketched n Appendx A and the basc features of the model are outlned here. In ths model, labour s assumed to be the only factor of producton, and as a result, an apprecaton of the domestc currency gves foregn frms a cost advantage n terms of domestc currency unts. Ths ntensfes the competton faced by domestc frms n both domestc and export markets, reducng the prce they can charge. 7 Consequently, domestc frms must reduce ther mark-up to reman compettve. For some frms, ths ncrease n competton and consequent fall n prce wll lead them to ext. Accordngly, an apprecaton n the domestc currency leads to a reducton n the number of domestc frms. Owng to the assumpton that all the frms are equally productve, ths model s unable to address the role of frm heterogenety n productvty. To nvestgate the effects of exchange rate movements on the turnover decson of heterogeneous frms, we combne Fung s (2006) predctons on the reacton of frms to exchange rate movements wth the recent models of nternatonal trade and heterogeneous frms n Meltz (2003) and Meltz and Ottavano (2005). In these two models, the ncrease n competton resultng from trade lberalzaton leads to the ext of low productvty frms and an expanson n the market share of hgher productvty frms. To the extent that the compettve pressure frms experence from lower trade barrers s analogous to the ncreased competton resultng from a domestc currency apprecaton, we mght expect smlar results when consderng the affect of exchange rates. Combnng the predctons of Fung (2006) wth those of Meltz (2003) and Meltz and Ottavano (2005), an apprecaton of the domestc currency ntensfes competton, causng domestc frms to reduce mark-up for survval; however, the less productve frms may not be able to afford the necessary reducton of mark-up, and therefore ext the market. 8 Note, however, that 7 In Fung s (2006) model, the demand functon s derved from a symmetrc translog expendture functon used by Bergn and Feenstra (2000, 2001) and Feenstra (2003). In ths model, the (postve) prce elastcty of demand s postvely related to the relatve prce of the good to ts competng goods. Compared to the proft-maxmzng condtons derved from CES utlty functon, the advantage of usng ths demand functon s that t shows a pro-compettve reducton of mark-ups when facng a reducton n the prce of competng goods. 8 Ths cause of frm closure s n lne wth Meltz and Ottavano (2005) but dffers from Meltz (2003). In Meltz s (2003) model, a CES utlty functon was used and t results n a constant prce elastcty of demand 5

7 when the domestc currency deprecates, domestc frms gan a cost advantage. In ths case, the mproved compettve poston of domestc frms not only attracts new frms nto busness but also reduces the probablty of falure for ncumbent frms (especally those wth lower productvty). In our emprcal analyss we examne whether an exchange rate apprecaton reduces the probablty of frm survval and reduces the entry rate of frms nto an ndustry as predcted by Fung (2006). Ths has become known n the lterature as the extensve margn. We also examne whether the effect of the exchange rate s dfferent for frms wth dfferent labour productvty. 9 The Intensve Margn: Total Sales In Fung (2006), an apprecaton of the domestc currency gves foregn frms a cost advantage n both domestc and foregn markets, drvng some domestc frms out of busness. For the sales of a survvng frm (to both domestc and foregn markets), the model shows two opposng effects of an exchange rate apprecaton. The total sales of ncumbent frms have become known as the ntensve margn. Whle the cost dsadvantage faced by domestc frms causes each of them to sell less, the ext of some frms leaves the survvng frms wth a larger market share. Consequently, from a theoretcal perspectve, the net effect of a currency apprecaton on a survvng frm s total sales depends on the drecton and relatve magntude of the changes n exports and domestc sales. 10 If the ext rate s low or the extng frms are substantally smaller than the ncumbent frms, the total sales of ncumbent frms would be expected to decrease when domestc currency apprecates. Under these condtons, a currency apprecaton causes both exports and domestc sales of a survvng frm to decrease and leads to a reducton n the frm s total sales. 11 Note that we expect analogous though opposte effects when the domestc currency deprecates. In ths case, the cost advantage faced by domestc frms encourages each of them to sell more but the entry of new frms may reduce ncumbent frms market shares, lowerng the quantty of sales. Theoretcally, the mplcaton of currency movements for domestc sales (and total sales) s ambguous, dependng on the ntensty of the two effects descrbed above, and becomes an open emprcal queston. Testable Hypotheses The theoretcal consderatons descrbed above yeld three hypotheses whch we wll explore emprcally: and constant mark-up. In ths model, real wage s bd up by the entry of new frms and the expanson of more productve frms, and t causes the ext of less productve frms. 9 Data lmtatons preclude us from measurng total factor productvty. 10 In the export market, t s more lkely that the effects of the cost dsadvantage for survvng frms may outwegh the effect of ncreased market shares, thus exports decrease. However, the effect on domestc sales may be ambguous, dependng on the relatve magntude of the cost dsadvantage and market share effects. 11 Conversely, a domestc currency apprecaton may cause an ncrease n the domestc sales of contnung frms when the ncreased market share effect outweghs the cost dsadvantage effect. Ths would occur f the ext rate s hgh. 6

8 1) An apprecaton of the domestc currency reduces a domestc frm s probablty of survval, a deprecaton ncreases the probablty of survval. 2) The negatve (postve) effect of a domestc currency apprecaton (deprecaton) on frm survval s less pronounced for more productve frms. 12 3) A domestc currency apprecaton (deprecaton) has an ambguous effect on a frm s total sales; however, f the ext (entry) rate s low or the extng (enterng) frms are substantally smaller, t s more lkely that an apprecaton (deprecaton) wll cause a reducton (ncrease) n total sales. 3 Data 3.1 Mcro data We have the good fortune of havng access to the T2-LEAP data, a detaled mcro-data set produced by Statstcs Canada. Ths data set s deal for testng our three hypotheses descrbed n the prevous secton. It was created by lnkng two underlyng sources of data: corporate tax nformaton from T2 tax forms, and the Longtudnal Employment Analyss Project (LEAP), whch obtans ts data from frm-specfc payroll nformaton fled wth the Canada Revenue Agency (CRA). Frm names are removed and replaced wth numercal dentfers to make the data set anonymous. T2-LEAP s a longtudnal dataset that provdes nformaton on every ncorporated Canadan establshment 13 that legally hres employees (and hence fles payroll nformaton wth the CRA) AND, n the same year, fles a T2 corporate ncome tax return. T2-LEAP covers the perod 1984 through 1998, and for the purposes of ths study data from are used. 14 It provdes annual frm-level data documentng the frm s employment level, proft, revenues, debt, equty, assets, locaton, and ndustry afflaton at the 3-dgt Standard Industral Classfcaton-Establshment (SIC-E) level. The dataset contans almost the entre Canadan prvate sector as measured by ether output or employment. Components of the economy that are omtted nclude non-ncorporated enterprses and corporatons that hred no employees. One advantage of the data s that t ncludes both publcly traded frms and (the more numerous) prvately held frms. These data provde an mportant complement to emprcal analyss based exclusvely on publcly traded corporatons. Whle T2-LEAP does nclude the very largest publcly held frms n Canada, more than half of the frms n our sample have fewer than ten 12 Meltz and Ottavano (2005) predct that openng up to trade causes less productve frms to ext and t can rase the aggregate productvty wthn an ndustry (known as the selecton effect). 13 An establshment s not necessarly equvalent to a frm as some large frms have more than one establshment, but the overwhelmng majorty of frms are sngle establshments and, correspondngly, the vast majorty of establshments correspond to ndependent frms. We wll use the term frm to represent the unts n the data set from now on. 14 The frst (1984) and last (1998) years are dropped because they are partal years only. We use 1985 but when the data are lagged we end up wth data from

9 employees. 15 We are restrcted to book values of debt, equty, and assets. As noted above, the frequency of the data s annual. All fnancal data s converted to real (1986) Canadan dollars usng the Consumer Prce Index (CPI). T2-LEAP s deal for dentfyng entry/ext because t conssts of the unverse of Canadan frms. In the T2-LEAP data we can dentfy the year of brth (should the frm enter post 1984) and the year of death (should the frm ext pror to 1998). We use the crtera set out n Baggs (2005) for dentfyng entrants and exts. Specfcally, a panel of survvors and exters s constructed usng the T2-LEAP data as follows. The ntal populaton of frms we consder are those frms whch exsted n The sample wll be augmented n each consecutve year by removng extng frms and addng new frms. A frm s removed from the sample n year t f year t s the year n whch the frm fles ts last tax return (the T2SUF measure of ext) or f year t s the last year n whch the frm employs workers (the LEAP measure of ext). If a frm falls nto ether of these categores, t s counted as extng. A frm extng n year t wll be removed from the sample for all subsequent years. A frm enters our sample n the frst year t both employs workers and fles a tax return. 3.2 Exchange Rate Movements In order to estmate the effect of exchange rate movements on survval and entry, the T2- LEAP data was lnked to real exchange rate data at the 3-dgt SIC level. The experence of Canada s flexble exchange rate regme over the 1980s and 1990s was tumultuous. Throughout ths perod the Canadan exchange rate underwent dstnct epsodes of large deprecatons and apprecatons. The blateral Canada-US exchange rate deprecated n the early 1980s and reached a hstorc (at that tme) low of US$ 0.69 on February 4, The Canadan dollar apprecated throughout most of the late 1980s untl t peaked at US$ 0.89 on November 4, The exchange rate went through another prolonged deprecaton and reached the (new) hstorc low of US$ 0.63 on August 27, 1998, before begnnng a new cycle of apprecaton. 16 Fgure 1 presents the aggregate entry and ext rates for manufacturng on the rght axs and the nomnal Canada-US exchange rate on the left axs. The aggregate entry rate fell by almost half between 1986 and 1993, declnng from almost 8 percent n 1986 to just over 4 percent n Thereafter, entry ncreased to almost 7 percent and then declned back to around 6 percent n The aggregate ext rate ncreased throughout the exchange rate apprecaton from 1986 to 1991 and declned durng the deprecaton perod from 1992 to Whle the value of the Canadan dollar wth respect to the US dollar s of substantal mportance to the Canadan economy, the mpact of exchange rate movements s lkely to vary across ndustres due to dfferent exposure to nternatonal markets and dfferent compostons of 15 Frms n our sample range from 1 employee to 42,000 employees, the mean frm has 38 employees and the medan frm has just over 9 employees. 16 See Powell (2003). 8

10 major trade partners across ndustry. In order to measure the ndustry-specfc exchange rate effects, we construct ndustry-specfc trade-weghted real exchange rates (TWRERs) usng exchange rates, a prce ndex specfc to each country, and the volume of trade data between Canada and ts major trade partners. The detals of the methodology and data used to construct the trade weghted real exchange rate are presented n Appendx B. Generally, we constructed the traded weghted real exchange rate (TWRER) n three steps. Frst, we normalzed the blateral real exchange rates n order to avod the unt problem n aggregaton. 17 Second, we constructed ndustry specfc weghts based on the sum of exports and mports from 1985 to 1989 for each 3-dgt ndustry s ten largest tradng partners. Therefore, the shares and trade partners vary by ndustry, but not by year. Fnally, we constructed trade weghted real exchange rates weghted by the trade shares constructed n step two for each ndustry and year ( ). Fgure 1 Nomnal Canadan Exchange rate wth entry and ext rates of frms % % US$/CDN$ % 6.00% 5.00% US$/CDN$ %entrants %exts Year % 3.00% Source: The exchange rate s from Internatonal Fnancal Statstcs. The entry rate and ext rates were computed by the authors from T2-LEAP. 17 We follow Head and Res (1999) and Bernard and Jensen (2004) and normalze the real exchange rates for each country usng 1984 as the base year. The unt problem s that blateral exchange rates have dfferent unts and those wth large unts wll have a larger weght. For example n 1997 the US exchange rate was US $ per Japanese yen and was 1.64 US $ per Brtsh pound. It does not make sense to smply average across these dfferent unts to compute an effectve exchange rate. 9

11 Fgure 2 presents the trade weghted real exchange rate from 1984 to Gven the large weght on trade wth the Unted States, t s not surprsng that the overall pattern s smlar to the blateral Canada/US exchange rate presented n Fgure 1: apprecaton from 1987 to 1991 followed by a deprecaton. Note that the exchange rate varable s based on the average value for the year. The frm data, as descrbed above, are based on observatons over the year. Fgure 2 Trade Weghted Real Exchange Rate Trade Weghted Real Exchange Rate = Year In addton to the exchange rate data, data on tarff changes for the relevant perod were lnked to the T2-LEAP data. Canadan and US tarffs were translated, followng Head and Res (1999), to 3-dgt SIC codes for manufacturng frms. US tarffs are compled usng the 93-ndustry classfcaton provded n Table A2.1 of the Canada-US Free Trade Agreement: An Economc Assessment (Government of Canada, Department of Fnance, 1988). Canadan tarffs are compled from Lester and Morehen (1987). The US shpments data that are used as a control for ndustryspecfc busness cycles are from the NBER-CES Manufacturng Industry Database by Bartelsman and Gray (1996) and the US Bureau of Census. 18 For each year, each frm s assocated wth the mport and export tarff for ts 3-dgt SIC code. Most frms are confned to a sngle 3-dgt ndustry. However, some span more than one 3-dgt ndustry n whch case Statstcs Canada selects the most mportant 3-dgt code for that establshment. Our data set has over 376,809 frm-year observatons. Table 1 contans descrptve statstcs for three samples of frms. The frst column s comprsed of statstcs for the complete sample of 18 The authors thank Sébastan LaRochelle-Côté for provdng the US employment data and concordance. 10

12 376,809 frm-year observatons. The second column contans only survvng frm-years and the thrd contans frms extng at some pont between 1986 and A total of 21,835 manufacturng frms, or 6 percent of total exted over ths perod. Column 4 ndcates whether the mean value of each varable s sgnfcantly dfferent for the populaton of survvors as compared to exters, at the 5% level. We fnd that on average, extng frms are sgnfcantly younger and smaller than ther survvng counterparts. They are also subject to larger tarff decreases, hgher trade weghted real exchange rates and operate n more concentrated ndustres wth lower rates of growth. In the next secton we turn our attenton to a more formalzed model to test the mplcatons of exchange rate movements for frm survval. The TWRER s statstcally dfferent for survvors and exters but the dfference s very small. The dfference n mean TWRER between exters and survvors s Wth respect to other varables the dfferences are also relatvely small but are generally statstcally sgnfcant. On average exters are younger by 0.54 years; the Canadan tarff declned by 0.67 for exters and 0.61 for survvors; contrary to expectons, the US tarff declned more for exters than survvors whch s not what we mght expect; 19 the concentraton rato s only very slghtly hgher for exters a dfference of Table 1 Descrptve Statstcs All Frms Survvors Exts Number of Observatons Survvors and Exts sgnfcantly dfferent at the 5% level? Mean TWR exchange rate Yes Mean Enterprse Age (truncated at 14) Yes Mean Number of Employees Yes Mean Sales $ $ $ Yes Mean Sales per Worker $ $ $ No Mean Canadan Tarff Reducton (n year t) 0.61% 0.61% 0.67% Yes Mean US Tarff Reducton (n year t) 0.28% 0.28% 0.31% Yes Mean Annual Industry Sales Growth 4.4% 4.5% 3.7% Yes Mean four frm concentraton rato Yes 19 Larger tarff reductons for exters as compared to survvors may result from the recprocal reductons of Canadan and US tarffs. 11

13 4 Emprcal Analyss In ths secton, we employ frm-level data to emprcally examne the three hypotheses derved n secton 2. We begn by examnng how the exchange rate affects frm survval usng the estmatng equaton: CAN US P( survveft ) = β 1 + β 2ERt + λ1δtarfft + λ2δtarfft + γx ft 1 + θyt 1 + τ t + ε ft, (1) where ndexes ndustres, f ndexes frms, P(survve ft ) s a 0-1 dummy ndcatng whether enterprse f was stll alve at the end of year t, y t-1 a vector of ndustry level control varables, and estmaton s by probt. ER t s the ndustry-specfc trade-weghted real exchange rate 20 CAN, tarff t and tarff US t are the changes of Canadan and US tarff rates at the ndustry level, 21 x ft-1 s a vector of frm characterstcs (ncludng age, sze and productvty), and τ t s a tme trend. Our ntal results based on Equaton (1) are presented n Table 2. In all specfcatons, the coeffcent on the trade weghted real exchange rate s negatve and sgnfcant, ndcatng that an apprecaton of the Canadan dollar decreases the probablty of survval for Canadan frms. Ths s consstent wth our frst hypothess and wth the trade lberalzaton lterature where the ncrease n competton resultng from domestc tarff reductons tends to decrease the probablty of survval of domestc frms. 22 The standard errors reported are robust standard errors wth adjustment for clusterng at the frm level. Columns 1 and 3 look at the effect of changes n the exchange rate from one year to the next whle columns 2, 4 and 5 examne the effect of the level of the exchange rate. The results suggest that the survval of Canadan frms s adversely affected both by upward movements n the value of the Canadan dollar, as well as the hgher values themselves. In the former case, we mght expect that apprecatons, especally those that are large or unexpected, mght adversely affect frms as they adjust to new compettve condtons. The later case, however, suggests that the negatve effect of currency apprecatons s not just the adjustment to change, but that hgher values of the Canadan dollar reduce the survval rate of frms. We control for the busness cycle and macroeconomc shocks usng the real value of US shpments by ndustry, real nterest rates and a tme trend. 23 It s possble that some macro shocks, nterest rates for example, may be correlated wth real exchange rates and may affect frm survval n smlar ways. Therefore, US shpments and the real nterest rate are ncluded as controls for 20 Here, the trade weghted real exchange rate s rescaled to be 1 n the base year. In the regresson analyss we consder both the level of trade weghted real exchange rate and the change from year t-1 to year t. 21 The Canada and US tarffs are n decmals. 22 See for example Baggs (2005) and Pavcnk (2002). 23 Though unreported, specfcatons excludng these controls also fnd a negatve and sgnfcant coeffcent of smlar magntude for the trade-weghted real exchange rate (or change n trade-weghted real exchange rate). 12

14 Table 2 Frm Survval Dependent Varable: =1 f frm survves to end of year t, 0 otherwse, Estmaton Method: Probt Trade Weghted Exchange Rate (ncrease=apprecaton of C$), year t-year t ** (0.095) ** (0.439) Trade Weghted Exchange Rate (ncrease=apprecaton of C$), level: year t ** (0.049) ** (0.219) ** (0.221) Reducton n Canadan Tarffs =ctar t-1 - ctar t ** (1.301) ** (1.300) ** (1.301) ** (1.308) ** (1.459) Reducton n US Tarffs =utar t-1 - utar t 5.986* (2.906) 7.592** (2.941) (2.855) (2.907) 7.326** (2.431) Labour Productvty =ln(sales t-1 /alu t-1 ) 0.072** (0.005) 0.092^ (0.048) 0.099* (0.046) Exchange Rate*Productvty = Trade Weghted ER*Labour productvty (0.102) Exchange Rate*Productvty =Trade Weghted ER*Labour productvty 0.172** (0.050) 0.193* (0.087) Assets =ln(assets t-1 ) Age =age n years Industry Sales Growth % growth n sales from yr t-1 to yr t, by 3-dgt sc Industral Concentraton 3-dgt CR4 (year t-1) Exchange Rate* IEE = Trade Weghted Exchange Rate* IEE Exchange Rate*IEE =Trade Weghted Exchange Rate*IEE US Shpments =ln(real US shpments by 3-dgt ndustry) Real Interest Rate =Canadan prme rate n year t 0.122** (0.019) (0.005) 0.122** (0.020) 0.016** (0.005) 0.059** (0.003) 0.015** (0.001) 0.098** (0.027) ** (0.028) (0.146) 0.109** (0.020) (0.005) 0.059** (0.003) 0.015** (0.001) 0.101** (0.028) ** (0.027) (0.014) 0.112** (0.021) 0.018** (0.005) 0.052** (0.003) 0.002** (0.001) 0.111** (0.028) ** (0.030) ** (0.006) 0.116** (0.021) 0.018** (0.005) Tme Trend =1 f year=87, =11 f year=97 YES YES YES YES YES Industry Fxed Effects two-dgt SIC dummy varables YES YES YES YES NO Log Lkelhood Observatons **= sgnfcant at 1%, * = sgnfcant at 5%, ^=sgnfcant at 10%. Robust standard errors (corrected for clusterng at the frm level) are n parentheses. 13

15 macroeconomc shocks and ndustry-specfc busness cycles. We also nclude ndustry fxed effects and ndustry sales growth. 24 Consstent wth Baggs (2005), our basc specfcatons n columns 1 and 2 fnd a sgnfcant postve effect of US tarff reductons on the survval of Canadan frms, and all fve specfcatons fnd a sgnfcant negatve effect of Canadan tarff reductons for survval. Columns 3 through 5 buld on our basc specfcaton, addng both frm and ndustry level control varables and nteractons between the exchange rate and these controls. In addton to ther drect effect on survval, t s possble that a number of factors temper the way exchange rates affect frms. The theoretcal work descrbed n secton 2 suggests that frm heterogenety wll be of partcular mportance n terms of a frm s productvty level and the extent to whch that frm exports ts product or reles on mported nputs to producton. We examne the mtgatng effect of labour productvty by nteractng t wth the exchange rate. Ideally we would want to use a measure of total factor productvty but the only data avalable s labour productvty. 25 Consstent wth our second hypothess, ths nteracton term s postve (and sgnfcant when the level of the exchange rate s used). Ths suggests that the negatve effect of an apprecaton s less pronounced for more productve frms. Accordngly, the demse of less productve frms ncreases the average productvty wthn an ndustry and may contrbute to a productvty mprovement at the ndustry level. Ths s analogous to the trade lberalzaton nduced ndustry level productvty mprovement descrbed n Meltz and Ottavano (2005). If frms manly use domestc nputs n producton, but export ther products, an apprecaton of the Canadan dollar puts that frm at a dstnct cost dsadvantage. However, a frm usng more mported nputs wll be at less of a dsadvantage (and may even beneft f lttle of ther producton s exported) as a result of an apprecaton. We explore the possblty that exchange rates may affect frm survval to dfferng degrees dependng on the afflated ndustry s relance on exports and mported nputs usng the Campa and Goldberg s (1995) ndex of effectve exposure (IEE). The ndex conssts of two elements: exports-domestc producton rato and the rato of mported nputs. If the later exceeds the former (negatve value of IEE), the cost advantage nduced by currency apprecaton outweghs the reduced export revenue. Please refer to Appendx B for the detals about the constructon of ths ndex. We nteract the IEE wth the trade weghted real exchange rate to nvestgate the mtgatng effect effectve exposure mght have on the survval response of frms to changes n the exchange rate. The coeffcent of the IEE-exchange rate nteracton term s expected to be negatve. In column 3 where the IEE s nteracted wth the change n the exchange rate, the nteracton s nsgnfcant (and n fact postve). However, when consderng the level of the exchange rate n 24 To prevent an endogenety problem, n the frm-level regressons, the ndustry sales are defned as the sum of sales of all the frms n ths ndustry except frm f. 25 Labour productvty s defned as sales per worker. 14

16 columns 4 and 5, the nteracton s negatve, though sgnfcant only n column 5 where ndustry fxed effects are not ncluded. Ths suggests that a frm s effectve exposure to exchange rates does not substantally mtgate the drect exchange rate effect, partcularly when controllng for other ndustry specfc factors. The domnance of the drect exchange rate effect s possbly related to a specfc attrbute of our Canadan data the Canadan manufacturng sector s hghly export orented, and the varance n exposure s not partcularly large amongst frms. Two-thrds of frms n our data have postve IEE s the exports/domestc producton rato exceeds the rato of mported nputs. Our results suggest that the drect effect of an exchange rate apprecaton s not substantally mtgated by effectve exposure. Ths s consstent wth Domnguez and Tesar (2006) who fnd a relatvely weak lnk between nternatonal actvty and exchange rate exposure. In addton to the nteractons reported here, we consdered a number of other possble varables whch mght temper the effect of exchange rates on frm survval (and sales below). These ncluded nteractng the trade weghted exchange rate wth Canadan and US tarff reductons and wth frm sze. In all cases, where productvty was also ncluded n the regresson, these nteractons were nsgnfcant. Ths hghlghts the relatve mportance of productvty n determnng the mplcatons of exchange rates for the survval of Canadan frms. Table 2 contans a number of control varables. Frm survval ncreases wth labour productvty, frm sze and age. At the ndustry level, a frm s more lkely to survve f ts ndustry s experencng hgher sales growth and f t operates n an ndustry where the four largest frms account for a smaller share of total sales. The real nterest rate s nsgnfcant n determnng survval when the change n the exchange rate s consdered, but postve and sgnfcant when the level s used. In addton, we also use the real shpments n the counterpart ndustres n the US to control for the ndustry-specfc busness cycles and fnd that hgher real US shpments n the ndustry ncrease a frm s probablty of survval. The next step n our analyss s to examne the mplcatons of exchange rates for entry. It s more dffcult to examne entry at the frm level because we observe whch frms dd enter, but we are unable to observe whch frms dd not enter, or the pool of possble entrants. Accordngly, lookng at the choce of entry at the frm level s emprcally ntractable. Instead, we analyze entry at the ndustry level usng an entry rate. The T2-LEAP data was aggregated to the ndustry level (3-dgt SIC) to estmate the followng equaton: Nt CAN US Entry = = β 1 + β 2ERt + λ1δtarfft + λ2δtarfft + θyt 1 + τ t + ε t, (2) I t where the entry rate s defned as the number of new frms n ndustry n year t (N t ) dvded by the number of ncumbent frms n ndustry n year t (I t ). In ths specfcaton, all control varables are at the ndustry level and the vector y t-1 ncludes such varables as the average ndustry proft rate, ndustry sales growth, average frm sze, and ndustral concentraton rato. We also 15

17 control for the busness cycle and macroeconomc shocks usng the real value of US shpments by ndustry as well as varables that only vary by tme such as the real nterest rates and a tme trend. The results of ths estmaton can be found n Table 3. In all four columns of Table 3, the coeffcent of the trade weghted real exchange rate s negatve and sgnfcant ndcatng that ndustry entry rates both fall as the Canadan dollar apprecates, and are lower for hgher levels of the Canadan dollar. Ths suggests that both a hgh level of the Canadan exchange rate and an apprecatng dollar dscourage entry nto the Canadan manufacturng sector. These results are consstent wth the smulaton results n Fung (2006) where an apprecaton reduces the number of frms,.e. the number of entrants s smaller than the number of extng frms; conversely, a deprecaton results n a net entry of frms. In other words, a domestc currency apprecaton dscourages frm entry whle deprecaton encourages frm entry. 26 The nsgnfcance of the majorty of control varables s somewhat puzzlng, though lkely attrbutable to the low degree of varaton n most of these varables wthn 3-dgt SIC ndustres. Interestngly, hghly concentrated ndustres have hgher entry rates, whle ndustres wth hgher US shpments have lower entry rates. Our results are consstent wth Head and Res (1999) who found that the deprecaton of the Canadan dollar ncreased the number of frms. Next we nvestgate the exchange rate effects on the total sales of survvng frms. Our thrd hypothess states that the net effect of currency apprecaton, for example, on an ncumbent frm s total sales s ambguous. If ext rates are hgh and survvng frms are able to substantally ncrease ther market share, despte lower prces, an apprecaton may allow survvng frms to ncrease ther total sales. If, however, the ext rate s not hgh or the extng frms are much smaller -- whch s more lkely to be the stuaton n the Canadan manufacturng sector 27 the ncreased foregn competton and lower prce level created by the apprecaton may nstead reduce survvng frm s total sales. We nvestgate the mplcatons of currency movements for frm sales by applyng the followng estmatng equaton to the sample of survvng frms: CAN US ln salesft = β 1 + β2ert + λ1δtarfft + λ2δtarfft + γx ft 1 + θyt 1 + τ t + ε ft, (3) where lnsales ft s the logarthm of total sales of frm f n real 1986 dollars. To avod an endogenety problem n estmatng equaton (3), where ndustry sales growth s used as a control varable, t s 26 In unreported regressons, we ntroduce trade-weghted exchange rate and IEE nteracton terms n the entry rate regressons. However, the coeffcents are nsgnfcant. The nsgnfcance may result from the hgh survval rate n the Canadan manufacturng sector, from the slower response of frm turnover to changes n the prce of mported nputs, and from the ncluson of ndustry fxed-effects. 27 In our data, approxmately 6% of frms ext n each year, and extng frms are about 30% smaller than survvors. 16

18 Table 3 Entry Rate (by 3-dgt ndustry) Dependent Varable: number of entrants n ndustry n year t, dvded by ndustry ncumbents n year t Trade Weghted Exchange Rate (ncrease=apprecaton of C$), year t-year t * (0.032) * (0.033) Trade Weghted Exchange Rate (ncrease=apprecaton of C$), level: year t ** (0.021) ** (0.023) Reducton n Canadan Tarffs =ctar t-1 - ctar t (0.496) (0.495) (0.499) (0.498) Reducton n US Tarffs =utar t-1 - utar t (1.205) (1.218) (1.286) (1.297) Industry Sales Growth % growth n sales from yr t-1 to yr t, by 3-dgt sc (0.007) (0.007) Industral Concentraton 3-dgt CR4 (year t-1) 0.039** (0.012) 0.040** (0.010) Medan Industry Profts =ln(medan Profts) year t (0.001) (0.001) Medan Frm Sze =ln(medan Employees) year t-1 (by 3-dgt sc) (0.001) (0.001) US Shpments =ln(real US shpments by 3-dgt ndustry) ** (0.005) ** (0.005) ** (0.005) ** (0.005) Tme Trend =1 f year=87, =11 f year=97 YES YES YES YES Industry Fxed Effects two-dgt SIC dummy varables YES YES YES YES Estmaton Method OLS OLS OLS OLS R Observatons **= sgnfcant at 1%, * = sgnfcant at 5%, ^=sgnfcant at 10%. Robust standard errors are n parentheses. measured usng the total sales of the ndustry mnus the sales of frm f. The results are presented n Table 4. In columns 1, 2 and 4 estmaton s by OLS and n column 3 we used a cross sectonal fxed effects estmaton method to further explot the panel structure of our data. The coeffcents of the trade-weghted real exchange rate are negatve n all specfcatons and sgnfcant n columns 1, 2, and 3 but nsgnfcant when a labour productvty-exchange rate nteracton term s ncluded n column 4. Our results ndcate that an apprecaton of the Canadan dollar reduces the total sales of survvng frms. Ths suggests that the revenue lost due to foregn competton and 17

19 Table 4 Frm Sales Dependent Varable: ln(sales t ) ln(sales t ) ln(sales t ) ln(sales t ) Trade Weghted Exchange Rate (ncrease=apprecaton of C$), level: year t ** (0.057) ** (0.027) ** (0.015) (0.220) Reducton n Canadan Tarffs =ctar t-1 - ctar t ** (1.925) ** (0.733) ** (0.441) ** (0.708) Reducton n US Tarffs =utar t-1 - utar t ** (4.489) (1.666) 2.981** (0.961) (1.597) Labour Productvty =ln(sales t-1 /alu t-1 ) Exchange Rate*Productvty = Trade Weghted ER*Labour productvty Assets =ln(assets t-1 ) Age =age n years Industry Sales Growth % growth n sales from yr t-1 to yr t, by 3-dgt sc Industral Concentraton 3-dgt CR4 (year t-1) Exchange Rate*IEE =Trade Weghted Exchange Rate*IEE US Shpments =ln(real US shpments by 3-dgt ndustry) Real Interest Rate =Canadan prme rate n year t 0.279** (0.046) ** (0.002) 0.818** (0.002) 0.001** (0.0001) 0.124** (0.012) ** (0.022) ** (0.006) 0.061** (0.007) ** (0.001) 0.382** (0.002) 0.076** (0.007) ** (0.019) ** (0.003) 0.026** (0.013) ** (0.001) 0.286** (0.047) (0.036) 0.776** (0.002) (0.001) 0.108** (0.011) (0.021) ** (0.006) 0.242** (0.017) ** (0.001) Tme Trend =1 f year=87, =11 f year=97 YES YES YES YES Industry Fxed Effects two-dgt SIC dummy varables YES YES NO YES Frm Fxed Effects NO NO YES NO Estmaton Method OLS OLS Fxed OLS Effects R Observatons **= sgnfcant at 1%, * = sgnfcant at 5%, ^=sgnfcant at 10%. Robust standard errors (corrected for clusterng at the frm level) are n parentheses. 18

20 lower prces outweghs any possble gan n market share as a result of ext. Conversely, durng perods of deprecaton, ncumbent frms are able to expand ther sales, despte ncreased entry. Ths s consstent wth Forbes (2002) who fnds that frms n countres that have experenced devaluatons have hgher output and operatng proft growth. Smlar to the case of survval, we also nvestgate some heterogeneous aspects whch may nfluence the response of frm sales to exchange rate movements. In column 4 an exchange rate-labour productvty nteracton term s added. The coeffcent s negatve though nsgnfcant, mplyng that an apprecaton of Canadan dollar reduces an ncumbent frm s total sales but the adverse effect has lttle relaton wth the level of productvty. In other words, the exchange rate has lttle effect of reallocatng market shares between more productve and less productve ncumbent frms. Combned wth the results n Table 2, that the adverse effect on frm survval of a currency apprecaton s less pronounced for more productve frms, our results ndcate that a currency apprecaton may lead to productvty growth by hastenng the demse of less productve frms but not by the market share reallocaton from less productve to more productve ncumbent frms. 28 In columns 2, 3 and 4, we explore the possblty that the exchange rate may affect the total sales to dfferng degrees dependng on the afflated ndustry s relance on exports and mported nputs usng the Campa and Goldberg s (1995) ndex of effectve exposure (IEE). Lke the case of survval, we expect ths nteracton to be negatve. If the rato of mported nputs exceeds exported-domestc producton (negatve value of IEE), the cost advantage nduced by the currency apprecaton outweghs the reduced export revenue. In all specfcatons, the coeffcents of IEEexchange rate nteracton term are negatve and sgnfcant, consstent wth our predcton. Domestc apprecatons reduce sales, and ths effect s larger for frms who export relatvely more of ther fnal products than they mport ntermedate nputs. Lookng at the control varables, our results show that, lke domestc currency apprecatons, Canadan tarff reductons reduce an ncumbent frm s total sales. 29 US tarff reductons have a statstcally sgnfcant postve effect on frm sales equatons (columns 1 and 3) but have a statstcally nsgnfcant effect n columns 2 and 4. Unsurprsngly, older, larger frms have hgher levels of sales, though age s nsgnfcant when productvty s ncluded as a control. Frm sales are postvely assocated wth ndustry sales growth and US shpments, and negatvely assocated wth ndustry concentraton and the real nterest rate. We also examned the mpact on frm sales by re-estmatng equaton (3) usng the change n log sales as the dependent varable rather than the level of sales and replacng the level of the 28 Aw, Chen and Roberts (2001) show that the (weghted) productvty growth n an ndustry can be attrbutable to the replacement of less productve exts by more productve entrants, the productvty growth among contnuers, and market share reallocaton (ncludng between entrants and exts and among contnuers). 29 The adverse effect of apprecaton and Canadan tarff reductons are n lne wth Head and Res (1999) fndngs usng ndustry-level data. 19