An Empirical Study of Country Risk Adjustments to Market Multiples Valuation in Emerging Markets: the case for Russia

Transcription

1 ЖУРНАЛ "КОРПОРАТИВНЫЕ ФИНАНСЫ" An Emprcal Study of Country Rsk Adjustments to Market Multples Valuaton n Emergng Markets: the case for Russa Ivashkovskaya I. 17, Kuznetsov I. 18 Valuaton n emergng markets s always a challenge. The exstence of soveregn rsk and captal market segmentaton as well as small tradng volumes and narrow domestc captal market make t dffcult to dentfy peer companes for market multples valuaton wthout cross- border comparables. Ths paper nvestgates the practcal mplementaton of market multples valuaton n emergng markets when the analyst should nvolve peer companes from developed markets. Companes wth comparable operatonal parameters bear dfferent values on dfferent fnancal markets. The problem of unavodable dfference among natonal stock markets exsts, that s why methods of cross-border multples correctons are called for. We address cross-border correctons procedures for adjustng multples to a soveregn rsk to fnd out the role and the extent of these type of adjustments n valuaton. We are usng the samples of Russan and US companes to test three dfferent adjustments technques: the soveregn spread, the relatve market coeffcents and the regresson approach. 1. Introducton Ths paper nvestgates the practcal mplementaton of multples valuaton n emergng markets. The man purposes of ths work are: frst, to fnd out whether crossborder correcton procedures are necessary or not; and, second, to buld a relatve valuaton methodology whch wll have broad practcal applcaton. The market multples valuaton approach s consdered to be the most complex valuaton methodology, whch mples that the value of the common shareholder equty s determned by the prce at whch t can be sold on the stock exchange. In other words, the farest estmate of a company value can be expressed n terms of the sellng prce of a smlar busness recorded by the market. Hence, the meanng of the term smlar becomes crucal n establshng the correct set of market comparables. The majorty of emprcal and fundamental researches showed that the company value s determned by ts operatng and fnancal characterstcs, market poston, and the perspectve of ts further development. Therefore, comparable companes should have smlar relatonshps between ther prces and key operatonal characterstcs such as proftablty, payout rato, growth rate, return and etc.. Those factors play a sgnfcant role n explanng the return from the pont of vew of a prospectve nvestor. However, n case of peer companes lack wthn the border of one emergng market analysts do often look for comparables from developed markets. Such approach should be carefully reformulated takng nto account the fact that the allure of an emergng market s prmarly determned by dfferent market mperfectons lke entry barrers, nformaton asymmetry, economc and poltcal rsk, government regulatons and etc.. Such factors affect the country and other frm s specfc rsks. That s why nvestors perceptons of the value of the same asset traded on dfferent markets dffer. Smlarly companes wth comparable operatonal parameters carry dfferent values on dfferent fnancal markets. The problem of unavodable dfference among natonal stock markets 17 Professor of Fnance, Corporate Fnance Center, Head of Corporate Fnance Department, Hgher School of Economcs, Moscow, Russa vashkovskaya@yandex.ru 18 Corporate Fnance Center, Hgher School of Economcs van.kuznetsov@ru.ey.com

2 ЖУРНАЛ "КОРПОРАТИВНЫЕ ФИНАНСЫ" exsts, that s why methods of multples correctons are called for. On the other hand, adjustment procedures are usually ranked among the last ones to perform, despte the fact that they account for the major part of company value. Ths paper addresses the ssue of buldng the most relable correcton methodology accountng for country rsk, and fndng a sutable multples valuaton methodology, whch wll take nto account cross-border effects on comparable companes. To do so we have analyzed two markets: the Russan one, as an emergng market, and the U.S. one, as a developed market, over the perod from 2002 to The remander of ths paper s organzed as follows: nto the frst secton we revew the relevant lterature on emprcal assessment of dfferent steps n conductng a multples valuaton methodology. The second secton provdes data and descrptve statstcs. In the thrd secton we state possble reasons for dscrepancy between Russan and Amercan market multples values and argue n favor of adjustments. In the fourth secton we descrbe and derve some major adjustments for country-rsk. In the ffth secton we present our methodology for country-rsks correctons, whereas n the last secton we assess ther effcacy. 2. Lterature Revew The emprcal fnance lterature on comparables s focused on several major ssues. Among the most popular lnes of dscusson one can fnd the analyss of relatve weak and strong ponts of dfferent valuaton ratos. Baker M.R., Ruback R. [Baker M. R., Ruback R., 1999], Km M., Rtter J.R. [Km, M., Rtter J.R., 1999] show some advantages of ratos based on Earnngs before Interest, Taxes, Deprecaton and Amortzaton (EBITDA) over Enterprse Value-to-Sales (EV/S) or Enterprse Value-to- EBIT (EV/EBIT). Lu J., Nssm D. and Thomas J. [Lu J., Nssm D., Thomas J., 2002] argue that despte the type of the ndustry Prce-to-Earnngs (P/E) rato better fts valuatons for many dfferent ndustres. Le E., Le H. J. [Le E., Le H. J., 2002] show the types of ratos that better descrbe the valuaton of the companes whch dffer n sze. The research on fundamental factors affectng partcular valuaton multples has a long hstory and s stll under way. Zarown P. [Zarown P., 1990], Allen A.C., Cho Y., [Allen A.C., Cho Y., 1999] found out the crtcal nfluence of earnngs growth rate n the forecasts on P/E, whle hstorcal growth rates and rsks are not as much sgnfcant. Bhojraj S., C.M.C. Lee [Bhojraj S., C.M.C. Lee, 2002], Damodaran [Damodaran, 2004] underlne another powerful set of ndependent varables: operatng proft margn, busness rsk and growth measured by standard devaton of operatng ncome and expected earnngs growth rate. Boatsman J., Baskn E. [Boatsman J., Baskn E., 1981], Alford A.W. [Alford A.W., 1992], Bhojraj S., C.M.C. Lee [Bhojraj S., C.M.C. Lee, 2002], Herrmann V., Rchter F. [Herrmann V., Rchter F., 2003] proved that the fundamental factors affectng the ratos are more mportant for a comparable frm selecton than the standard ndustry classfcaton codes. Fnally, there s a collecton of papers on the dscusson of adjustments for unsystematc rsks (marketablty dscounts, sze and control premums). The factors affectng valuaton ratos n emergng markets have been tested by Oppong A. [Oppong A., 1993], Erb C.B., Harvey C.R., and Vskanta T.E. [Erb C.B., Harvey C.R., and Vskanta T.E., 1996], Nkbakht E., Polat C. [Nkbakht E., Polat C., 1998], Perero [Perero, 2002]. Damodaran [Damodaran, 2004] summarzed hs regresson analyss wth the concluson about the nfluence of real GDP growth rates, low nterest rates as ndependent varables to explan hgher P/E values. Accordng to Ramcharran H. [Ramcharran H., 2002], P/E n 21 emergng markets s explaned by economc growth and credt rsk whch he found to be statstcally sgnfcant factors. Thus, t s clear that multples drawn from dfferent market envronments depct dfferent

3 ЖУРНАЛ "КОРПОРАТИВНЫЕ ФИНАНСЫ" fundamentally justfed values. That s why, methods for reducng such a dscrepancy level n multples values by usng adjustment procedures are called for. However, despte many researchers acknowledge ths fact, ts justfcaton remans an open ssue. 3. Data and descrptve statstcs In ths secton we gve a detaled statstcal analyss of the multples dstrbutons propertes n order to depct the cross-border dfference n multples values. We have chosen the Russan market as the emergng one and the U.S. as the developed. Our analyss s based on the annual data from 2001 to 2004 for both markets. For the U.S. market we have gathered the annual data from the Compustat Database provded by the FEDC: Database Envronment. We selected all publc traded companes domcled n the Unted States at the NYSE, NASDAQ, and AMEX. Our sample companes matched the followng crtera: the book value should be a postve measure, and the frm s profts should be postve for at least three out of four years. Ths restrcton was mposed n order to assure a comparson wth the Russan sample, whch exhbts such characterstcs. We excluded companes wth unavalable Global Industry Classfcaton Standard Code (GICS) 19, or not classfable. All tems from the Balance Sheet and from the Income Statement were taken at the end of years respectvely. The share prces were taken four months after the fscal year end, that s the closng prce at the end of Aprl for years 2002 to Ths common practce s often used by many researchers, snce the four month lag ensures that the company has publshed ts fnancal statements, and that they were avalable to nvestors, thus reflected n the share prces. The fnal sample conssts of 15,104 frm-years observatons. For the Russan market we took data for all publcly traded frms on the Russan Trade System (RTS), RTS Board, Moscow Interbank Currency Exchange (MICEX), whose fnancal statements are reported n Internatonal Accountng Standards (IAS), or n Unted States Generally Accepted Accountng Prncples (US GAAP). Ths was done n order to ensure comparablty between data, because t s practcally mpossble to compare frms under the US GAAP and the Russan Accountng Prncples, due to large dfference n several key areas, such as deprecaton accountng, good-wll amortzaton, consoldated accountng etc. The same data were taken from the Balance Sheets and the Income Statement as for the Amercan sample. We used the RosBusnessConsultng database for fndng requested tems, as well as the companes statements posted on ther webstes. The majorty of the frms ddn t report negatve earnngs; however a lmted number of companes reported losses, but not more than for one of the four consdered years. The fnal sample consttutes 255 frm-years observaton. However, t was necessary for the relatve Russan frms to be smlar to the US ndustry classfcaton code. That s why we have lnked the Russan companes to the 8-dgt Sub-ndustry GICS code, snce Russan companes do not have an approved classfcaton scheme. The share prces were taken n the same way as for the Amercan Market. For these two markets we computed the followng multples: P/B rato, defned as the market value of equty (total shares outstandng multpled by the share prce) dvded by the book value of equty (or common shareholders equty, thus excluded preferred stocks). P/E rato, defned as the market value of equty dvded by the net ncome. EV/S, defned as the rato of enterprse value to sales, where the enterprse value was defned as the market value of equty plus total debt (Long-term debt plus the 19 The GICS classfcaton methodology can be found at:

4 ЖУРНАЛ "КОРПОРАТИВНЫЕ ФИНАНСЫ" current porton of the long-term debt) plus preferred stock outstandng mnus Cash and equvalents. Accordng to these defntons all multples wth undefned values were deleted from the sample; however, some negatve multples of unproftable companes, whch matched the crtera descrbed above, were not deleted. Fnally, for the U.S. market we have 12,709 P/B observatons, 10,401 P/E observatons, and 13,281 EV/S observatons. Each multple was wnsorzed for each year at 1% and 99%. For the Russan market we computed 172 P/B multples, 152 P/E multples and 177 EV/S multples. Table #1 reports the descrptve statstcs by countres for years Table #1: Descrptve Statstcs for Multples U.S. Market Std. dev. Kurtoss Skewness Multple # obs. Mean Medan P/E 10, P/B 12, EV/S 13, Russan Market Std. Multple # obs. Mean Medan dev. Kurtoss Skewness P/E P/B EV/S n Medan Std. dev. Kurtoss Skewness Ths table shows the comparson of multples dstrbuton across two countres. The U.S. market statstcs s qute smlar to precedent researches. For example, Maug E., Dttmann I. [Maug E., Dttmann I., 2006] report smlar results the medan for P/E equals , for P/B , for EV/S -1.63, takng nto account that they have analyzed a broader sample from year 1994 to As usual all dstrbutons are characterzed by a hgh degree of skewness. The means substantally exceed the medans, especally for the P/E multple, whch s true for the two markets. Ths s due to the outlers property of skewng the representatve mean toward the rght tal. That s why, t s not advsable to use the arthmetc mean as an averagng procedure. The varablty of the P/E multple s the bggest due to ts dependency on proft. In order to reduce the estmates nterval, t s advsable to use the P/B and EV/S multples, whch have the lowest standard devaton. The Kurtoss coeffcent shows us, that all dstrbutons have tals fatter than the normal dstrbuton. From ths vewpont the EV/S seems to be a closer approxmaton, whch s also confrmed by the lowest skewness coeffcent. For the Russan market the same relatonshps hold. The P/E multple has the broadest varance, the mean exceeds the medan, whch means that the multples are not unformly dstrbuted. The standard devaton for the P/E multple s the hghest, for the other multples t s much lower. The standard devaton of the P/B and the EV/S multples for the Russan market s pretty smlar to those of the U.S., takng nto account that the number of observatons dffers sgnfcantly. Thus, t s clear, that the multples values exhbt smlar dstrbuton patterns across the two markets, whch confrms the possblty of makng a comparatve analyss between these two markets. However, t s mportant to pont out, that table #1 depcts

5 ЖУРНАЛ "КОРПОРАТИВНЫЕ ФИНАНСЫ" the aggregated statstcs for 4 years merged, both for the Amercan and the Russan markets. Apart from ths, t s necessary to understand how ther propertes evolved over tme. For ths purpose we perform a comparatve analyss of Russan and Amercan multples dstrbutons. Fgure #1 depcts the dynamcal change of the P/E and P/B multples dstrbuton for the Amercan market. Fgure #1: Dstrbuton of the P/E and P/B multples for the U.S. market for years Number of companes >72 As we can see from the fgure, the dstrbuton of the multples underwent sgnfcant changes for years 2002 to On the whole, we observe a stable asymmetry; however, n year 2002 the arthmetc mean and the medan were shfted to the left. For example, the number of companes, havng the P/E multple lower than 18 and P/B lower than 2, was sgnfcantly hgher as compared to the subsequent years. For example, n year 2003 we observed the hghest shft to the rght, whch pushed many companes from the left tal to the rght. However, n year 2004 (depcted by a sold lne) there was a small backward shft. But, on the other hand, we also observe an ncrease n the multples values for companes lyng wthn the standard devaton range from the medan multple value. Thus we can wtness a general ncrease of the multples values for the last three years. The dstrbuton of EV/S multple depcted n appendx #1 s characterzed by lower varatons. However, smlar tendences could be observed. Thus, we can state that wthn the tme perod of a sgnfcant swft of all multples dstrbutons took place, wth a subsequent ncrease n the medan values and an observable redstrbuton of companes from the left tal to the rght. Unfortunately, the conducton of a smlar dynamcal analyss for the Russan market s hampered, due to the lack of observatons. However, we can make some mportant observatons on aggregated data. Fgure #2: Dstrbuton of the P/E and the P/B multples for the Russan market for years P/E P/B `02 `03 `04 Number of companes P/E > P/B >6

6 ЖУРНАЛ "КОРПОРАТИВНЫЕ ФИНАНСЫ" At a closer look at the P/B and P/E multples dstrbutons we found qute smlar propertes, as for the U.S. peers. The dstrbutons are also skewed to the left, but the densty of the P/E dstrbuton s thn compared to the P/B multple. The dstrbuton of the EV/S multple (Appendx #2) s more symmetrcal compared to the others. The skewness coeffcent supports these specfc fndngs. Ths s probably due to the fact that we ddn t exclude companes wth negatve earnngs, whch lead to a creaton of a more symmetrcal dstrbuton. Another nterestng feature: the densty of the dstrbutons for these multples dffers sgnfcantly from the Amercan peers. If for the latter we observe a smooth growth toward the medan value, for the Russan ones a sharp jump,.e. the speed of reachng the medan value s much hgher. After a detaled comparatve analyss of the multples dstrbutons propertes we found one mportant dstnctve feature: as compared to the US dstrbutons, the Russan dstrbutons are characterzed by a stronger left-sded shft. Moreover, a strkng observaton arses from the fact that the means and the medans for all the multples are lower for the Russan market n comparson wth the US (see table #1). Thus, we have an obvous llustraton that, on the whole, t s naccurate to use the Amercan multples as t may lead to the overvaluaton of Russan companes. 4. Cross-border correctons ratonale At frst consderaton the performed analyss reveals that the Russan companes are traded at dscounted multples due to the avalablty of hdden effects, whch dstort ther fundamental values. Such effects could be a drect result of dfferent economcal states of affars, fnancal, legal and regulatng nsttutons. It s also related to the overall country rsk level, whch s born by the neffectveness or nstablty of the State machnery, by the weak protectablty of nvestors rghts, and by a hgh degree of nformaton asymmetry. These country rsks become rrelevant f the nvestor has an opportunty to globally dversfy ts portfolo. However, f nvestors are constraned from enterng or extng specfc country markets, they may fnd themselves solated, or segmented, and come to bear country-related rsk [Perero, 2002, p.108]. It s obvous that such factors have a straghtforward nfluence on the whole busness envronment, and more precsely on the nvestors percepton of assets values. Thus, by takng two dentcal companes, functonng on two dfferent markets, we can assume that nvestors wll value hgher those, whch wll operate on the stable one, because ts future development prospect wll be brghter. At the same tme the companes fundamentals can reach smlar levels. Several emprcal studes have clearly shown that ths effect on stock return s frequently more szable, than the ndustry effect. In other words, stock performance seems to be much more tghtly lnked to the local volatlty of the economy, than to the fluctuatons and trends of the correspondng ndustry at the nternatonal level [Perero, 2002, p.109]. Ths problem can sometmes be extrapolated to the comparson of two developed markets; however, the dfference n the magntude of the above descrbed effects wll be smaller. It s known, that companes havng smlar expected growth rate of Net Income can have dfferent P/E multples. For example, Amercan and Great Brtan companes usually have a smaller P/E multple, as compared to the German and Japanese ones. These dfferences are well llustrated by dfferent researches analyzng the markets multples relatonshps through dfferent countres [Ramcharran H., 2002], [Damodaran, 2004], whch show drect relatonshp between rsk ncrease and lower market multples values. Thus, gnorng the country rsk n valuaton can lead to sgnfcant overvaluaton of companes operatng on emergng markets, snce ther comparable counterparts wll have completely dfferent values due to a more stable economc state of affars. Thus, the

7 ЖУРНАЛ "КОРПОРАТИВНЫЕ ФИНАНСЫ" frst effect whch can explan the above stated dscrepancy between Russan and Amercan multples could be the country rsk. Apart from the country rsk, other hdden effects can sometmes nfluence the dscrepancy between two companes multples levels. In the work by B. Black [B. Black, 2001] an nterestng hypothess s ntroduced: the dfferences n value between western companes and companes from emergng markets (or precsely the Russan ones) are a consequence of dfferent Corporate Governance levels. The author departs from the pont that the corporate governance behavor of the Unted States frms that affects ther market value s scarce probably because the varaton n such frm s behavor s small. The mnmum qualty of Amercan corporate governance, set by law and by norms so wdely accepted that almost no publc frms depart from them, s qute hgh [B. Black, 2001, p.1]. In contrast, the dfference n the corporate governance level of Russan frms s very large, and has a measurable effect on the frm value. For ths purpose the author examned 21 frms from the Russan market n year 1999, and found out that the correlaton between the natural logarthm of the value rato (the rato of actual market captalzaton to the potental Western, computed by an ndependent Russan bank) strongly correlates wth the governance rankng (determned ndependently by a second Russan nvestment bank). Thus, corporate governance behavor has a powerful effect on market value n a country where legal and cultural constrants on corporate behavor are weak [B. Black, 2001, p.20]. That s why the second hdden effect can also have an nfluence on the dfference between the multples values of emergng and developed markets. Thus, we observe that the combnaton of the two above mentoned effects explans to a certan extent the observed dfference n market multples. In order to reduce ths overestmaton of Russan companes, or precsely n order to reduce to a common denomnator the Amercan multples values, we need to ntroduce some correcton measures, whch should be then ncorporated nto the valuaton algorthm. 5. Methods for multples cross-border adjustments The use of several rsk correcton procedures plays a key role nto the fnal company s value determnaton. Ths s especally true for unsystematc dscounts and premums whch are often characterzed by a hgh degree of subjectvty, whch naturally lead to a hgh degree of uncertanty of the valuaton outcome. Another problem arses wth the use of some adjustments already computed by large fnancal agences. The use of these numbers, wthout a clear understandng of ther computaton methodology could not be assumed satsfactory. We examne cross-border adjustments based on three dfferent approaches: the Spread of Soveregn Bonds Yeld, the Correcton Coeffcent (Market Multples Rato) and Multple Regresson Approach for Country and Frm Specfc Rsks Correctons. The Market Multples Rato, as well as the regresson methods have been already descrbed by Perero [Perero, 2002], Damodaran [Damodaran, 2004] wth the samples other than Russan companes. We add to the lterature the thrd approach - Spread of Soveregn Bonds Yeld method and examne the relatve strength of all three methods to determne the most approprate technques. Let s consder them n detal. Spread of Soveregn Bonds Yeld Ths method s often used by practcal apprasers, who beleve that the country rsk s essental n determnng the multples values. In fact, ths s a smple modfcaton of the rsk premum computed as an addton to the standard CAPM. The man dea of ths correcton s that there exsts a postve relatonshp between the yeld to maturty of a

8 ЖУРНАЛ "КОРПОРАТИВНЫЕ ФИНАНСЫ" government bond and ts default probablty. The rsker the bond s, the hgher return expects the nvestors. That s why the rsk-free rate can be consdered as a good proxy for market mperfectons or country rsk, or precsely the more the rsk of the busness envronment, the less the nvestors wllngness to pay for a certan company, whch wll be reflected n smaller multples values. The rato between the US and Russan soveregn bonds yeld could be seen as an adjustment for country rsks. As for the rsk free US bonds we chose the US Treasury Notes wth maturty perod equal to fve years at the end of Aprl For the Russan rsk free government bonds, we have chosen the Russan Federaton Mnfn s Eurobonds, wth maturty perod equal to fve years 20. Takng nto account that these two fnancal nstruments are denomnated n US dollars, the computed yeld to maturty reflect only the relatve country rsk effect, wthout ncludng other mportant frm-specfc rsks. Table #2 provdes the data on correspondng fnancal nstruments at year 2004 (for years 2003 and 2002 look nto the Appendx #2). Table #2: Fnancal nstruments at Name of the Fn. Instrument Yeld to Maturty (%) Maturty Term (years) Maturty Date Euro Euro Euro Euro Euro Euro US TN However, before comparng the yeld to maturty t s necessary to ensure that both nstruments have same maturty terms,.e. we need to fnd Russan Eurobonds wth a maturty term equal to 5 years on 29 Aprl For example, we can see from the table, that the best comparable nstrument would be the EURO-10, but the dfference stll exsts. Takng nto account the term structure of the nterest rates, t s worthwhle to assume that there s a cause-and-effect relatonshp between the expected spot rates and the maturty term of the Bonds, whch has to have an ncreasng character. Ths s explaned by the theory of unbased expectatons (the expected spot rate should be equal to the correspondng forward rate,.e. the expected ncrease n the annual spot rate s a cause of the ncrease of the yeld curve). Thus, by usng a regresson analyss we dentfed for each year a tme relatonshp between the yeld to maturty of the Eurobonds and ts maturty term. For year 2005 ths relatonshp s descrbed by the followng equaton: (1) YTM = ln T; R 2 = 91%. Consequently, by nsertng the correspondng maturty term of the US 5-years Treasury Notes we obtan a yeld to maturty equal to 5.4% for the Mnfn s Eurobond. In Appendx #5 other equatons can be found. For each year we computed a theoretcal yeld to maturty for 5-years Eurobonds. The rato between the US and the Russan yeld to maturty s the sought correcton coeffcent, whch only accounts for the country rsk. Table #6 presents the obtaned results. 20 The data for Mnfn s Eurobonds were taken from the monthly reports of the Alfa-bank (see at

9 ЖУРНАЛ "КОРПОРАТИВНЫЕ ФИНАНСЫ" Table #3: Computaton of Country Rsk Correctons for years Date of Yeld to Maturty Country Rsk Valuaton Euro-5 US TN-5 Correctons From the table we can see that wthn the observed tme perod the relatve Russan economy s rsk decreased as compared to the Amercan one, whch lead to an ncrease n the correcton value. Thus, n year 2005 the Russan multples are approach ther US peers, whch s drven by the mprovng state of affars of the Russan economy. In prncple ths s entrely correspondng to the observed stuaton. Let s turn ourselves to the next correcton procedure. Market Multples Rato The second method ntroduced by L. Perero n hs book Valuaton of Companes n Emergng Markets [Perero, 2002] conssts n computng the rato between two countres market multples. The dea s to fnd the relatve value of one country s multples n terms of the other (or how much 1 unt of the US multple s worth n the emergng country s unts). Thus we frst calculate the means across the two markets multples at a specfc date: (2) medan{ P / B } rato = medan{ P / B rus;2004 us;2004 Ths method depcts the relatve dfference between the nvestors perceptons of the two markets and n prncple descrbes the dfference of country rsks (frms, performng on a stable market, should have a hgher multples value). Bascally, t s pretty smlar to the last one, where we computed the theoretcal market multples rato. However, t dffers sgnfcantly snce t already ncorporates other frm-specfc rsks (such as the dfference n the Corporate Governance level). That s why such correcton should gve better results as compared to the prevous one. For each multple we computed the correspondng correcton at the end of Aprl Table #4 depcts the results. Table #4: Correctons based on the Market Multples Rato Method Valuaton Multples Correctons Date P/B P/E EV/S % 56% 59% % 68% 70% % 60% 78% As we can see from the table, on the whole, the correctons stay at smlar levels n comparson wth the Spread of the Soveregn Bonds Yeld Method. However, we can wtness some peculartes n the dynamcal context. If for the P/B multple we can wtness an ncrease n the correctons level (whch means a decrease of the cross-border rsks), for the P/E multple we observe a small drop n year 2005, and for the EV/S somethng completely dfferent, a total decrease. The nterpretaton of these results s very ntrcate, snce t s mpossble to dstngush t from other effects. However, we can assume, that for the EV/S multple the results could be worse as compared to the frst correcton method, snce t doesn t tral the general economc pattern. On the other hand, we observe qute reasonable estmates, as for the prevous method. But we should recall one crucal fndng. Takng nto account that we ddn t take nto consderaton all Russan companes, or precsely those who reported ther accountng n RAS (the Russan Accountng Standards) we expect to have lower medan values of reported correctons on }

10 ЖУРНАЛ "КОРПОРАТИВНЫЕ ФИНАНСЫ" the whole Russan sample. Ths s lnked to the fact, that those companes are manly mddle- and small-sze frms, whose busness doesn t have any nternatonal character, whch probably means an ncrease of companes lyng nto the left tal of the multples dstrbuton. Multple Regresson Approach for Country and Frm Specfc Rsks Correctons At frst sght the above proposed correctons are derved from the practcal valuaton vewpont. In realty they could be subject to serous theoretcal crtcsm. At the same tme the fundamentally determned regresson approach truly answers the queston, whether cross-border correctons are requred or not. The dea behnd the regresson method s qute smple: we run regresson of multples on ther fundamentals parameters and on a dummy varable of the companes home country. For the fundamentals we could use those, whch reflect the return, rsk and growth varables. The dstnctve feature of ths approach for country rsk adjustments mply that the correcton coeffcents are predetermned by the model nputs, thus automatcally layng down the magntude of dfferent hdden effects n the dummy coeffcents estmates. If the dummy varable s sgnfcant the coeffcent should show the premum n the multples value for companes from developed countres. The bg advantage of ths method s that t helps to objectvely answer the queston: are the cross-border effects really sgnfcant? If yes, then we assume that a combnaton of the above mentoned effects really nfluence the multples values. Takng nto account the restrcted dataset of Russan multples, we consdered the Chemcals, Telecommuncatons, Electrc Utltes, Machnery, Metals & Mnng, and Ol & Gas ndustres. Fnally, t was possble to form an acceptable number of Russan and US companes, for the purpose of cross-border dfferences dentfcaton. For each multple a sngle regresson was run, whch had ts own explanatory varables, as the fundamentals varables and the general dummy varables. However, takng nto account that the structural relatonshp between multples and ther fundamentals s tme dependent, the use of the regresson approach on the pooled data sample s naccurate. For ths reason we performed the analyss for each selected year. Let s look at ths n detal: P/B multple: Based on prevous researches, the current ROE was chosen as the man fundamental varable, also an addtonal varable was ntroduced the standard devaton of Net Income, whch s consdered to reflect the rsk profle of the company. Takng nto account that many prevous researches deny the lnear relatonshp between the P/B multple and the ROE, we used a log lnear specfcaton form, whch n fact gave better results. We also took logs on standard devaton of the Net Income due to ts hgh range; ths helped us to scale ts value. In order to dentfy the cross-border effects, a dummy varable for the company s country was ntroduced (for the US =1, for Russa =0). If the dfference n cross-border effects really had a sgnfcant mpact on the multples values apart from the sgnfcance of ths coeffcent, we expected to fnd a postve sgn,.e. for the Amercan company the premum value attrbutable to the absence of country rsks should be a postve addton to the multple value, and for the Russan one t should to be equal to 0. We also used a dummy coeffcent explanng ndustral belongng of the frm. Ths s related to the fact that each ndustry has t own specfc fundamentals values, whose dfference should not be gnored. Let s consder the estmated models for year 2004 (the results for other years can be found n Appendx #3).

11 ЖУРНАЛ "КОРПОРАТИВНЫЕ ФИНАНСЫ" (3) ln( P / B ) = ln( ROE ) 0.01ln( σ ) D D ; R NI country ndustry 2 = 0.39 (0.168) (0.036) (0.016) (0.097) (0.028) The statstcal relatonshp for the P/B multple s qute strong. The R 2 of the model reach about 40% for all 3 years. Apart from ths all coeffcents have the expected sgn: an ncrease of a company s ROE leads to an ncrease n the P/B multple value, the rsk ncrease, expressed by the standard devaton of the Net Income varable reduces the P/B multple. More mportant the coeffcent before the dummy varable s statstcally sgnfcant and has a postve sgn. Other thngs held constant, ths means, that for a US company, the natural logarthm of the P/B multple wll be 0.54 ponts hgher as compared to the Russan one. The same results hold for the other years, however, the ndustry dummy varable s not always statstcally sgnfcant. At the aggregated level the R 2 falls to 30%, but all conclusons are stll the same and the coeffcents estmates are statstcally sgnfcant. P/E multple: For the explanatory varable of the P/E multple we used the expected growth rate of the Net Income varable, and an addtonal rsk measure the standard devaton of the Net Income. As n the case of P/B multple, we assumed a log lnear specfcaton form. However, takng nto account, that t s mpossble to compute the logarthm of the negatve growth rate of the Net Income, we ddn t mpose a logarzaton on ths varable. The estmated regresson model for the P/E multple n the year 2003 looks lke: (4) ln( P / E ) = NI 0.02ln( σ ) D 0.01D ; R growth (0.131) (0.029) (0.014) (0.096) (0.024) As we can see all coeffcents estmates are sgnfcant, apart from the ndustral dummy varable. Moreover, the expected magntude of all coeffcents s preserved: wth an ncrease n the future Net Income growth rate ncreases the P/E multple s value, the rsk ncrease of the company leads to the reducton of ts value. Once more the country s dummy varable coeffcent s statstcally sgnfcant and has an expected sgn, thus the Amercan companes are traded wth a premum compared to the Russan ones, whch take the form of hgher multples value. However, for year 2004 we observe a statstcal nsgnfcance of the fundamental varable and an unexpected negatve sgn. Such results should not be surprsng, snce the projected value of the Net Income varable for year 2005, was taken as the future regresson-averaged value, whch n prncple s a poor substtute for the market s expectatons. On aggregated level, all coeffcents are statstcally sgnfcant, and the R 2 ncreases to 15%. Thus, we can state that the dfference n cross-border effects really affects the multples values. EV/S multple: For ths multple the Operatng Margn was chosen as the man fundamental varable, and for the rsk varable we choose the standard devaton of the EBIT. Compared to the prevous cases, we ddn t assume a log lnear specfcaton form, however n order to scale the standard devaton we used a logarzaton of ths varable,.e. that an addtonal contrbuton of the margnal ncrease of ths varable has a decreasng nfluence on the overall multple ncrease. For the EV/S multple we observe the most robust estmates. The R 2 of the model fluctuates around 40%, whereas all coeffcents (ncludng the ndustral dummy NI country ndustry 2 = 0.1

12 ЖУРНАЛ "КОРПОРАТИВНЫЕ ФИНАНСЫ" varable) for the three years are statstcally sgnfcant. For year 2004 the regresson model looks lke: (5) EV / S = OM ln( σ ) D 0.22D ; R (0.197) (0.777) (0.023) (0.136) (0.037) All varables, except for the company s rsk, have expected sgns. As to the prevous cases, the country dummy varable s stll sgnfcant, ths once more proves the necessty of usng some cross-border correctons. Thus, we descrbed n detal used country rsks correctons. However, n order to nterpret the effcacy and the necessty of such cross-border correctons, we should, frst, perform ther emprcal testng. Let s consder n detal the emprcal model for crossborder correctons. 6. The emprcal mode for cross-border correctons: underlyng prncples The most effectve way to determne the best used methods s an emprcal checkup based on a large database, whch means that we frst analyze dfferent companes valuaton methodologes, and then judge about ther relatve effcacy based on the devaton degree of projected company s value from the observed one by the market. Thus, we automatcally assume the effcent market hypothess, by agreeng that the market value of a company s a far estmate. In order to assess three country rsk correcton methodologes, we constructed two fundamentally dfferent valuaton algorthms: the algorthm of relatve correctons (where we focus on the effcency of the Spread of Soveregn Bonds Yeld, and Market Multples Rato methods), and the algorthm of the regresson approach for country and frm-specfc rsks correctons. Let s consder them n detal. Algorthm of relatve correctons effcency analyss Based on the recent emprcal researches, we observed that the best method of comparables companes selecton could be the method based on smlar fundamentals parameters dentfcatons, whch can n prncple gnore the companes ndustral classfcaton. On the one hand, the same ndustral belongng means that comparable companes are faced wth smlar ndustral rsks and comparable envronmental condtons; on the other hand, the majorty of companes have ther own exceptonal attrbutes, whose dentfcaton can gve better understandng of the current busness state, than the ndustral classfcaton. In vew of these reasonng, we take as a bass three companes selecton algorthms: Industral Classfcaton Method For the flter crtera based on ndustral belongng we used the 6-dgts GICS code (last degree of ndustral specfcaton). Thus Amercan companes, havng the same GICS code as the Russan company, were selected nto the comparables companes dataset. Such hgh degree of specfcaton ensures that companes should be subject to smlar ndustral factors, nfluencng the frm s value. Fundamentals Flter Method For ths method we chose for each multple a pre-specfed parameter: 1) for P/B multple current return on equty (ROE) 2) for P/E multple future devaton of net ncome from ts current state or smply the future growth rate of Net Income varable (NI growth), whch s a good substtute for ts expected value EBIT country ndustry 2 = 0.26

13 ЖУРНАЛ "КОРПОРАТИВНЫЕ ФИНАНСЫ" ) for EV/S multple current operatng margn (OM). For selectng a potental comparable we assgn a specfc devaton nterval from the Russan company s fundamentals value, f the potental comparable company fundamental les wthn ths nterval we select t as a comparable company. For the ROE we assgned a 10% devaton nterval,.e. f the Russan company has a ROE=25%, the comparables companes are those, whose ROE s less than 35% and more than 15%. For the NI growth - 30%, snce ths varable has a very large dsperson both for the Russan and Amercan markets. For the OM - 10%. It s nterestng to know, that the ncrease n the nterval values leads to a sgnfcant reducton of the valuaton accuracy, whch s probably related to the fact, that we selected secondary companes. Combnaton of these two, whch assumes a prmary flterng on ndustral belongng, and a secondary - on smlar fundamentals. For the last method of comparables selecton (the combned method of the last two), an ndustry code flter s frst appled, and then contnued wth a selecton, based on a fundamental parameter. Takng nto account, that after the frst flter the obtaned dataset of comparables companes was sgnfcantly reduced, we ncreased the ntervals range n order to ensure that any companes would be dentfed n the P/B Set #3. For the ROE 25%, NI growth 60% and for OM 30%. It s nterestng to state, that compared to prevous researches our ntervals are sgnfcantly narrow. Herrmann V., Rchter F., [Herrmann V., Rchter F., 2003] use for the fundamentals flter values of 30% for ROE and NI growth, and for the combned method 50%. That s why we expect to have better results for ths method compared to the smple ndustry classfcaton, whereas the effcacy of the fundamentals method s dffcult to assess. For the valuaton purposes we used 3 common multples: P/B, P/E и EV/S, where the ntal dataset was the Amercan and Russan companes data, descrbed n the prevous secton. For the valuaton of 255 Russan companes-years wth the use of 15,104 Amercan companes-years, wth a combnaton of all possble valuaton methods reachng 27 (precse multple => precse selecton crtera => precse correcton procedure) we buld a smulated valuaton routne n the Ox edt core (based on Gauss programmng language), whch helped us to mx all possble combnatons and provde a summary statstcs, assessng ther overall performance. Scheme #1 depcts the general sequence of the programmed valuaton algorthm for the P/B multple, for a specfc year.

14 ЖУРНАЛ "КОРПОРАТИВНЫЕ ФИНАНСЫ" Scheme 1: Valuaton Algorthm for P/B Multple Russan Companes Dataset Company #1 Company #2 Amercan Companes Dataset Industry Code Flter Fundamentals Flter Industry Code Flter Multples Projected Value Computatons P/B Set #3 Fundamentals Flter P/B Set #1 P/B Set #2 P/B Set #3.1 Correctons W/O Correctons Multples Rato Sov. Bonds Spread W/O Correctons Multples Rato Sov. Bonds Spread W/O Correctons Multples Rato Sov. Bonds Spread Devaton of Projected Frm s Value from Realzed Valuaton Resduals Predcton Errors Dataset Analyss of Resduals Dstrbuton Summary Reteraton of the whole Algorthm for the Next Company The valuaton algorthm begns from takng the frst company from the Russan Companes Dataset (or more precsely ts fnancal and ndustral nformaton) and then performs a valuaton based on all possble combnatons of methods, only for ths frst Russan company. However, before performng a valuaton, we need to cast asde a certan amount of Amercan outlers. In order to get objectvely justfed values, we wnsorze the sample by 1-5% (dependng on the multple) for each year. Thus from the frst step the valuaton methodology goes nto three dfferent drectons. From the Amercan sample, the algorthm flters those companes whch satsfy the mposed crtera for each flter used. After ths step the algorthm computes the synthetc multple s value for each comparables selecton method used. For ths purpose we use the smple medan measure, as t s the most convenent way to deal wth the problem of reducng the nfluence of outlers on the overall frm s value. On the next stage, we use three dfferent correcton procedures for each multple: wthout any correcton, wth the Spread of Soveregn Bonds Yeld method, and wth the Market Multples Rato. We expect that wthout any correcton the valuaton accuracy wll be the worst, due to the avalablty of certan country and frm s specfc rsk dfferences between companes. After computng the correctorzed multple s value, we assess the overall accuracy of the used methods. For ths purpose we analyze the devaton of the projected frm s value from t s realzed on the Russan market. For each Russan company we computed the correspondng resduals (for each combnaton of methods used), whch are depcted by dotted arrows. Thus for company #1 we get 9 resduals (only for the P/B multple), whch are then regstered nto the valuaton s resduals database. After that the program automatcally returns to the frst step and repeats the whole algorthm for the next company. It s necessary to pont out, that for each multple the valuaton s performed at precse dates,.e. for year 2002 we use the Amercan Companes Dataset at year 2002 and so on. After we computed the resduals values for all Russan companes, the algorthm begns to analyze the dstrbuton of all resduals for each combnaton of methods (.e. for the P/B multple 9). On computed statstcs bass we draw conclusons about the

15 ЖУРНАЛ "КОРПОРАТИВНЫЕ ФИНАНСЫ" relatve effcacy of the selecton of comparable companes crtera, and on the correcton procedure. It s noteworthy to pont out, that the regresson approach of country and frm specfc rsks correctons has a totally dfferent algorthm, that s why the analyss of ts effectveness s conducted separately (also for each multple). Let s now consder n detal the man assumptons and precondtons lad n each step of the above descrbed algorthm. Algorthm of the regresson approach n cross-border correctons We should take nto account the fact that the multple regresson approach has a qute dfferent algorthm compared to scheme #1. At the frst step we not only analyze the dataset the Russan companes, but the pooled sample n precse year, where the Amercan companes are dentfed wth a dummy varable. Takng nto account the outlers problem we wnsorzed the sample on 1 and 99%, n order to elmnate extreme values, whch were capable to dsturb the coeffcent s sgnfcance. A regresson was then run for each multple on the fundamental value and on the dummy coeffcent. After we got our coeffcents estmates for each multple, we bult a theoretcal valuaton model for valung only Russan frms. As n the prevous case, we then computed the projected frm s value and compared t wth the realzed one, thus gettng our valuaton resduals, whch were then ncluded nto the predctons errors dataset. Takng nto account that the multples regresson relatonshps suffer from tme varablty, the use of a regresson analyss on aggregated dataset for years s not reasonable. That s why the algorthm repeats the whole procedure for the next year. Thus, we descrbed n detal the general prncples and assumptons of the used emprcal smulated valuaton model. Let s consder now the fnal effcacy of the above descrbed correctons for cross-border effects. 7. Emprcal analyss of cross-border correctons effcency Effectveness of Relatve Correctons Procedures As we have seen from scheme #1, the algorthm ntally uses 3 dfferent selecton crtera for comparable companes. We frst, choose the two best out of the three selecton procedures, and then based on them, compare the relatve effcacy of the depcted correctons. For each multple we computed specfc effcency crtera, assessng the projected company's value wth the current realzed state. Thus, we use 4 crtera measurng correcton effcency: a smple arthmetc mean of the predctons errors, medan of the predctons errors, medan of the absolute predctons errors, root mean squared errors (RMSE). The advantage of usng the medan crteron over the smple average conssts n the fact that they effectvely solve the outlers problem, but the queston of countng the standard devaton s stll open. The advantage of the RMSE conssts of accountng for the bas, and the standard devaton of the errors at the same tme. Ths measure s effectvely used by Maug E., Dttmann I. [Maug E., Dttmann I., 2006]. Thus, ths varable s the most objectve n the overall effcacy assessment of the employed valuaton algorthm, however, at the same tme t gves no clear ndcaton whether we have under- or overvalued our companes. Table #5 depcts the obtaned results.

16 ЖУРНАЛ "КОРПОРАТИВНЫЕ ФИНАНСЫ" Table #5: Comparatve Effcacy of Comparable Companes Selecton Crtera Flter Method P/B Multple Mean Medan Abs. Medan RMSE Std. dev Industry Code 108% 57% 76% Fundamentals 103% 71% 79% Industry Code+Fundamentals 87% 47% 77% Flter Method P/E Multple Mean Medan Abs. Medan RMSE Std. dev Industry Code 62% 38% 68% Fundamentals 90% 41% 64% Industry Code+Fundamentals 59% 31% 63% EV/S Multple Flter Method Mean Medan Abs. Medan RMSE Std. dev Industry Code 67% 32% 63% Fundamentals 74% 38% 50% Industry Code+Fundamentals 65% 26% 56% Ths table focuses only on the effcacy of comparable companes selecton crtera, wthout mposng any correcton procedure. As we can see from the table, the smple arthmetc mean of the predctons errors s practcally always hgher, compared to the medan. The smple medan for each multple s hgher than 0, ndependently on the selecton crteron. In fact ths s another confrmaton that gnorng the cross-border effects wll lead to serous overvaluaton of the companes operatng on an emergng market. The depcted results prove objectvely ths fact. Thus, based on the smple medan, we can conclude that wth all possble selecton crtera the average predcton error (or overvaluaton) fluctuates nto the range of 30-60%, whch s not satsfactory. It s better to compare the relatve effcacy of the methods by combnng the absolute medan (or more precsely the medan of the absolute predctons errors) wth the RMSE. In ths table we marked n talcs and stressed the best possblty, the secondbest s wrtten n thck prnt. For the P/B multple the best results are showed by the combned method (ndustry code and fundamentals selecton crteron), the worst s the ndustry classfcaton method, whose RMSE s equal to 2.25, whch s also proved by the hghest dsperson. Thus, from the vewpont of the RMSE the preferences should be unambguously gven to the last two methods of comparables companes selecton. From the vewpont of the absolute medan, all methods gve practcally same bad results n average the ncorrect projected value devates from the realzed one by 76-79%. As a result we gve preference to the last two comparable companes selecton crtera. For the P/E multple the overvaluaton s slghtly smaller, however, stays at the level of 30-40%, whch s confrmed by the smple medan. On the other hand, the combned selecton method s unambguously the leader, whose RMSE and absolute medan reach 1.49 and 63% respectvely. As to the other two methods, t s dffcult to gve preference to any of them, snce the RMSE s hgher for the ndustral classfcaton method, but the absolute medan for the fundamentals method. On the other hand, such results are probably drven by the fact, that for year 2004 the fundamental of the P/E multple was the regressed projected value of Net Income growth rate for year 2005, whch sgnfcantly worsened the statstcs (remember the statstcal nsgnfcance of ths varable for year 2005 and ts ncorrect sgn showed n the last secton). That s why such large standard devaton s probably due to the neffcacy of ths projected value, rather than to the low self-descrptveness of the varable tself. In spte of ths fact, the effcacy of ths method stays at the same level as for the combned method: n average