Country Allocaton and Mutual Fund Returns By Dr. Lela Heckman, Senor Managng Drector and Dr. John Mulln, Managng Drector Bear Stearns Asset Management Bear Stearns Actve Country Equty Executve Summary We examned the actual performance of 56 nternatonal and global mutual funds durng 2001 to 2005. Our conclusons are: Tme seres analyss suggests that, on a month-to-month bass, country allocaton explans 88% of the varaton of fund returns and 21% of the varaton of fund excess returns. Cross-sectonally, and over longer holdng perods, country allocaton explans a substantal (34%-50%) share of excess return varance. Together, sze and country allocaton explan as much as 60% of the cross-sectonal varaton of excess returns. In terms of performance contrbuton, country allocaton appears to be a postve contrbutor to fund performance, whle stock selecton appears to add no value to nternatonal and global equty funds. Ths paper provdes new evdence on the relatve mportance of country allocaton for nternatonal equty portfolos. Many studes have used factor analyss to address the ssue (see, for example, Heston and Rouwenhorst (1995). Ths lterature provdes nterestng nsghts, but n our vew suffers from a key problem: The analyss s based mplctly on the assumpton of perfect foresght. In contrast, ths paper adopts a dfferent framework and attempts to measure the hstorcal mportance of country allocaton for actual nternatonal equty mutual funds. To do so, we use a return decomposton methodology analogous to that used by Brnson, Hood, and Beebower (1989), who examned the mportance of asset allocaton strategy relatve to nstrument selecton n balanced portfolos (portfolos contanng stocks, bonds, and cash). We apply ths methodology to data provded by Emergng Portfolo.com 1 on the actual mutual fund country allocatons of 56 nternatonal and global mutual funds durng 2001-2005. We fnd that, durng the perod covered by our study, 34%-50% of the varance of excess returns across nternatonal and global equty mutual funds s explaned by ther country allocaton strateges. Combned, country allocaton and style selecton explan up to 60% of the cross-sectonal varaton. Fnally, we fnd that country allocaton margnally added to excess returns, whle decsons other than country allocaton (ncludng stock selecton) modestly subtracted from excess returns durng the perod of analyss. 1 Emergng Portfolo Fund Research, Inc. (EPFR) tracks equty and bond fund flows, cross border captal flows, country and sector allocatons, and company holdngs data from ts unverse of 10,000 nternatonal, emergng markets and US funds wth $5 trllon n assets. EPFR data comes drectly from funds or ther admnstrators and ncludes funds regstered n the major domcles of North Amerca, Europe, Asa and other offshore domcles. 1
The Factor Analytc Approach Many analysts have assessed the relatve mportance of country allocaton usng factor analyss. In ther semnal paper, Industry and Country Effects n Internatonal Stock Returns, Heston and Rouwenhorst (1995) compared the relatve mportance of country and sector factors for developed markets. 2 They found that the dsperson of countryfactor returns was greater than that of sector-factor returns whch they nterpreted as evdence that country factors domnate sector factors. The ntuton behnd ths nterpretaton can be grasped by lookng at an extreme example. Suppose that all countres have the same sector composton. If ths s the case, a country s factor return equals the excess return of the country portfolo. Smlarly, a sector s factor return equals the excess return of the global sector portfolo. If the varance of returns across countres s large relatve to the varance across sectors, portfolo returns can be better enhanced by shftng $1 from the worst performng country to the best than by shftng $1 from the worst performng sector to the best. Ergo, country factors domnate. An mportant crtcsm of the foregong analyss s that t presumes perfect knowledge. The exstence of varance s only an opportunty f t can be predcted. In a very mportant sense, the whole lterature on country effects versus sector effects has been off base. The queston should not be, Where s the varance greatest? But, rather, Where s the predctable varance greatest? An Alternatve Framework for Assessng Country Allocaton Ths paper assesses the mportance of country allocaton n practce by analyzng the performances of actual nternatonal and global equty mutual funds from January 2001 to August 2005. Usng data on fund returns (gross of fund expenses) and country allocaton weghts, we decompose returns nto country allocaton and non-country allocaton components. We defne the return to country allocaton as the hypothetcal return that would be acheved by a fund n a month, f the fund began the month wth the country weghts reported by Emergng Portfolo.com and nvested passvely wthn each country accordng to MSCI wthn-country stock weghts, thereby achevng the MSCI country ndex return n each country. We assume no md-month portfolo adjustments. Our estmate of the country allocaton return ncludes a deducton for the estmated transactons costs assocated wth mplementng an actve country, passve stockselecton strategy. 2 A country factor return say the Australan factor return can be nterpreted as the excess return (n ths secton, excess return s defned as the return n excess of the MSCI world return) on an Australan portfolo wth the same sector composton as the world portfolo. A sector factor return say the Energy factor return s the excess return on a world Energy portfolo wth the same geographcal composton as the MSCI world return. 2
Equaton 1 : R F = R A + λ The return equaton can be easly recast to reflect the composton of a funds performance relatve to ts benchmark: Equaton 2 : Where: F B A B ( R R ) = ( R R ) + λ R F = Fund return, gross of expenses R A F = Fund country allocaton return = w * r B c λ = R B Fund return due to everythng other than country allocaton = Fund benchmark return = w B r B w F = Fund weght for country w B = r B = Benchmark weght Benchmark return for country for country c = Estmated country allocaton transactons cost Ths paper s strategy s to assess the relatve mportance of country allocaton by lookng at Equatons 1 and 2 n three ways: Tme-seres analyss: What percent of the average fund s month-to-month return varance s explaned by the country allocaton return? Cross-sectonal analyss: What percent of the return varance across funds can be explaned by country allocaton returns? Performance analyss: What are the contrbutons of country allocaton versus other nvestment decsons for the average fund s excess return? 3
Dscusson of Data Set Ths paper analyses mutual fund returns gross of the expenses reflected n fund expense ratos. In other words, we examne the sum of the net fund return to the nvestor plus fund expenses. By examnng gross returns, we restrct our focus to portfolo management performance. To calculate country allocaton returns, the study uses Emergng Portfolo.com s database of the monthly country allocaton weghts of 56 nternatonal and global equty mutual funds. For the average fund, the database provdes 37.5 months of data coverng varous spans durng January 2001 August 2005. The average fund s assets under management averaged $2.3 bllon durng the perod. Net fund returns were collected from Factset and Bloomberg wth the gudance of tckers provded by Emergng Portfolo.com. Expense ratos were obtaned from Bloomberg. Benchmark returns were provded by Morgan Stanley Captal Internatonal (MSCI). Usng fund country allocaton weghts and MSCI country benchmark returns, we were able to calculate gross allocaton returns. These fgures can be nterpreted as the returns funds would have acheved wth ther country allocaton choces had they remaned passve stock selectors and nvested n each country s benchmark ndex wthout transactons costs. In cases where an allocaton share was assgned to a resdual category (for example, other Europe ), we used the MSCI All-Country World Index (ACWI) as a proxy n the calculaton. Ths may produce a modest amount of nose n our estmates and therefore decrease the estmated explanatory power of country allocaton for fund returns. We then estmated the transactons costs assocated wth country allocaton for each fund by drectly measurng allocaton transactons volumes and chargng 50 bass ponts for each developed market transacton and 150 bass ponts for each emergng market transacton. 3 We beleve that our estmates are conservatve, both n terms of the estmated cost per transacton and n terms of the volume of transactons. The volume estmates are probably overstated, because t appears that n some cases hgh turnover estmates resulted from categorzaton shfts (for example, n one case a large shft of weght appeared to occur from other Europe n one month to specfc European countres n the next month). By subtractng these estmated costs from the gross allocaton returns, we arrved at our estmates for country allocaton returns. Across the 56 funds n the data set, turnover averaged 48% annualzed, and transactons costs averaged 0.59% annualzed (Fgure 1). 3 Transactons for each country were estmated as the absolute value of the dfference between the current weght and the lagged weght adjusted by the country s relatve return (one plus the country return dvded by one plus the fund return). 4
Fgure 1 Fund Turnover and Transactons Costs Annualzed Turnover Annualzed Cost Average 48% 0.59% Standard Devaton 16% 0.22% Mnmum 23% 0.23% Maxmum 103% 1.36% Source: Emergng Portfolo.com, Bear Stearns estmates. Excess returns were calculated usng MSCI EAFE as the benchmark for mutual funds wth nternatonal mandates and MSCI All-Country World for stocks wth global mandates. These ndexes were used as benchmarks for all funds, ncludng those wth value, growth, small cap, or large cap mandates. Tme Seres Analyss Our tme-seres analyss of Equaton 1 conssts of a regresson of each fund s monthly returns aganst the fund s country allocaton returns. Intutvely, we expect ths regresson to produce a good ft or R-Squared whch n ths case measures the percent of fund return varance that can be explaned by country allocaton return varance. The reason we expect a good ft s that country equty returns tend to be correlated wth one another. Consequently, f t s a good month for equtes globally, most equty funds are gong to perform well, regardless of country allocaton. Because of ths, the hgh observed tme seres R-squared of 88% does not n tself make a strong statement n favor of the mportance of country allocaton (see Fgure 2). Brnson, Hood, and Beebower (1986) performed an analogous return decomposton n ther study of the mportance of asset allocaton vs. nstrument selecton for balanced portfolos (consstng of stocks, bonds, cash, and other nstruments). Ther tme-seres analyss ncluded tme-seres regressons of fund returns aganst benchmark returns and aganst asset allocaton returns (see the frst two rows of Fgure 2). 4 Ther results for balanced funds are qute smlar to our analogous fgures for nternatonal and global equty mutual funds. Tme-seres regressons based on Equaton 2 provde a more demandng test. For Equaton 2 to ft well, a hgh percent of a fund s excess return (relatve to ts benchmark) would have to be explaned by the country allocaton excess return. The mere fact that global equtes tend to be correlated wll not support a hgh R-Squared. The emprcal results suggest that about one-ffth of the average fund s excess return varance over tme can be explaned by country allocaton excess returns (Fgure 2). Brnson, et al (1989) dd not run a parallel tme-seres regresson for balanced funds, so we have no bass for comparson. 4 Instead of the phrase benchmark return, they used polcy return, whch they measured as the return a fund would have had n a gven month, f the fund held ts 10-year average weghtng n each asset class and each asset class had the same return as the asset class passve ndex. The asset allocaton return was defned as the return assumng the fund held ts current asset class weghtng and each asset class had the same return as the asset class passve ndex. 5
Fgure 2 Tme Seres Return R-Squares Equaton: Smple R-squared* Weghted R-squared** Correspondng Fgure from Brnson, et al. Equaton 0: Fund Return and Benchmark Return 73% 72% 93% Equaton 1: Fund Return and Allocaton Return 88% 89% 95% Equaton 2: Excess Fund Return and Excess Allocaton Return 21% 22%? * Funds wth less than 12 months of data (11 out of 56 funds) were excluded from the calculatons. ** Weghted by the number of months of data avalable for each fund. Cross-Sectonal Analyss The paper s cross-sectonal analyss examnes the varance of returns across the funds n our data base. The analyss dffers from the tme-seres work n one less-than-obvous respect: For each fund, the cross-sectonal analyss focuses exclusvely on cumulatve return performance. Ths presents a problem: The holdng perods for the varous funds do not lne up perfectly. Data for some funds span as many as 55 months, but less than twelve months for 11 of the 56 funds n the data base. Our soluton to ths problem s to focus on Equaton 2, whch s cast n terms of excess returns. We calculate both the fund excess return and the country allocaton excess return over the approprate holdng perod for each fund. In the cases where we run smple cross-sectonal regressons, we nclude only funds wth at least 12 months of data (45 observatons). We nclude all 56 funds, however, when we use weghted least squares, wth each fund s weght n the regresson equal to the number of months of data used n calculatng the fund s excess return. We began the cross-sectonal analyss by focusng on core nternatonal and global mandates; that s, we excluded funds wth value, growth, large cap, or small cap mandates. We found that country allocaton excess returns explan 34% of the varaton of fund excess returns (Fgure 3). The estmated slope of 0.87 s reasonably close to the level we would expect (unty). The near-zero ntercept suggests that fund managers nether helped nor hurt performance through stock selecton. The majorty of funds, however, fell below the 45 degree lne the Mendoza lne below whch fund returns due to everythng other than country allocaton (lambda) are negatve. 5 5 If you ht below.200 n major league baseball, you are known to be httng below the Mendoza lne. 6
Fgure 3 Country Allocaton and Fund Returns: Core Funds (Excludng Small Cap, Large Cap, Value, and Growth Mandates) 8.0% 6.0% 4.0% Fund Excess Returns 2.0% 0.0% -2.0% -4.0% y = 0.8723x - 0.0035 R 2 = 0.3427-6.0% -8.0% -8.0% -6.0% -4.0% -2.0% 0.0% 2.0% 4.0% 6.0% 8.0% Country Allocaton Excess Returns We subsequently expanded the unverse to nclude large cap, small cap, value, and growth mandates. We ran acrosssectonal regressons across the 45 funds for whch there were at least 12 months of data (Fgure 4). At 42%, the R-squared was hgher than for the core unverse, but so was the ntercept, at 1.7. 6 The hgh slope reflects the fact that funds wth small cap mandates not only exhbted hgh lambdas, but also were successful at country allocaton. The hgh lambdas, however, dd not stem from stock selecton per se, but rather from style bas. Although the small cap funds wth hgh lambdas produced cumulatve returns that exceeded the MSCI EAFE and ACWI ndexes, none of them outperformed the MSCI global small cap ndex. 6 Ths R-Squared s very close to the 39% cross-sectonal fndng of Ibottson and Kaplan (2000) n ther study of balanced funds. 7
Fgure 4 Country Allocaton and Fund Returns Small Cap, Value, and Growth Mandates Included 10.0% 8.0% 6.0% Small Cap Mandates 4.0% Fund Excess Returns 2.0% 0.0% -2.0% -4.0% -6.0% -8.0% y = 1.176x - 0.0017 R 2 = 0.4206-10.0% -10.0% -8.0% -6.0% -4.0% -2.0% 0.0% 2.0% 4.0% 6.0% 8.0% 10.0% Country Allocaton Excess Returns We mplemented weghted least squares to deal wth the potental problem that the database has more observatons for some funds than for others (and that, therefore, the results may be unduly swayed by funds wth relatvely short holdng perods). The weghted results suggest an even greater explanatory power 52 percent for country allocaton excess returns than n the un-weghted regresson (Fgure 5, top panel). As a fnal step, we added two dummy varables: a small cap dummy and a value-growth dummy. The small cap dummy varable equals zero for all fund wthout a specfc small cap mandate. For a fund wth a small cap mandate, the small-large cap dummy equals the relatve performance of global small cap equtes vs. the MSCI ACWI durng the fund s cumulatve return perod. For funds wth large cap mandates, the small cap dummy equals the negatve of the relatve return performance of small cap equtes. The value-growth dummy was constructed n an analogous manner. For funds wthout value or growth mandates, the value-growth dummy equals zero. For value funds, the value-growth dummy equals the relatve return of value vs. growth durng the fund s cumulatve return perod. For growth funds, the value-growth dummy equals the relatve return of growth vs. value durng the fund s cumulatve return perod. The small cap dummy had a great deal of explanatory power durng the perod much more than the value-growth dummy. In the weghted regresson, the combnaton of country allocaton excess return and sze mandate explaned about three-ffths of the varance of fund excess returns n the perod. These results suggest that, durng the sample perod, country allocaton was mportant, and the combnaton of country allocaton and style selecton was domnant. 8
Fgure 5 Fund Excess Return Regressons Unweghted Regresson Weghted Regresson Coeffcent T-Stat Coeffcent T-State Independent Varables: Constant 0% -0.4 0% -1.1 Allocaton Excess Return 118% 5.6 164% 7.6 R-Squared 42.1% 52.0% Probablty 0.0% 0.0% Note: Funds wth return hstores less than 12 months long are excluded from unweghted regressons. Weghted regressons are weghted by months of avalable data. Fund Excess Return Regressons Unweghted Regresson Weghted Regresson Coeffcent T-Stat Coeffcent T-State Independent Varables: Constant 0% -0.2 0% -0.4 Allocaton Excess Return 92% 4.4 136% 6.2 Small Cap Relatve Return Dummy 28% 3.3 22% 3.1 Value/Growth Return Dummy -4% -0.3 4% 0.3 R-Squared 54.1% 59.5% Probablty 0.0% 0.0% Note: Funds wth return hstores less than 12 months long are excluded from unweghted regressons. Weghted regressons are weghted by months of avalable data. Performance Analyss Do funds add value through country allocaton? Do they add value through actve stock selecton wthn countres? The answers are Yes to the frst queston and No to the second. The fund managers n our sample were able to get country allocaton rght not by a long shot, but at least margnally. The 0.5% average country allocaton excess return across 45 funds was statstcally sgnfcant, wth a t-statstc of 1.85. The average lambda 7 across 45 markets, however, was a statstcally nsgnfcant -0.1% (Fgure 6). The average lambda across funds wth small cap mandates was 4.0%. Ths excess return, however, has nothng to do wth stock selecton per se. To the contrary, the average mutual fund n the small cap category actually underperformed the MSCI small cap ndex by 2.8%. Consequently, the postve average lambda across small cap mutual funds was solely attrbutable to the fact that they nvested accordng to the small cap style, whch happened to outperform durng the perod. 7 Gross return attrbutable to everythng other than country allocaton. 9
Fgure 6 Decomposton of Gross Fund Excess Returns* 45 Funds** Core Funds*** Value Growth Large Cap Small Cap Gross Fund Excess Return 0.4% -0.1% 0.3% 0.2% -3.0% 6.8% Country Allocaton Excess Return 0.5% 0.3% 0.4% 0.6% -0.6% 2.8% Lambda**** -0.1% -0.4% 0.0% -0.4% -2.4% 4.0% * Excess returns are defned relatve to ether MSCI EAFE or ACWI, dependng on the fund's mandate. ** Includes all funds wth return hstores at least 12 months long. *** Excludes small cap, value, and growth mandates **** Gross return attrbutatble to everythng other than country allocaton References Brnson, Gary P., and L. Randolph Hood, and Glbert Beebower, 1986. Determnants of Portfolo Performance. Fnancal Analysts Journal (July/August); 39-44. Heston, Steven L., and K. Geert Rouwenhorst. 1995. Industry and Country Effects n Internatonal Stock Returns. Journal of Portfolo Management, vol.21, no.3 (Sprng):53-58. Ibbotson, Roger G., and Paul D. Kaplan, 2000. Does Asset Allocaton Polcy Explan 40, 90, or 100 percent of Performance? Fnancal Analysts Journal (January/February); 26-33. 10