You are on page 1of 48

The Society for Financial Studies

International Asset Allocation under Regime Switching, Skew, and Kurtosis Preferences Author(s): Massimo Guidolin and Allan Timmermann Reviewed work(s): Source: The Review of Financial Studies, Vol. 21, No. 2 (Mar., 2008), pp. 889-935 Published by: Oxford University Press. Sponsor: The Society for Financial Studies. Stable URL: http://www.jstor.org/stable/40056838 . Accessed: 13/02/2012 16:04
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Oxford University Press and The Society for Financial Studies are collaborating with JSTOR to digitize, preserve and extend access to The Review of Financial Studies.

http://www.jstor.org

International Asset Allocation under Regime Switching, Skew, and Kurtosis Preferences
Massimo Guidolin ManchesterBusiness School and FederalReserve Bank of St. Louis Allan Timmermann Universityof California,San Diego, and CREATES
This paper investigates the international asset allocation effects of time-variations in higherorder moments of stock returns such as skewness and kurtosis. In the context of a fourmoment International Capital Asset Pricing Model (ICAPM) specification that relates stock returns in five regions to returns on a global market portfolio and allows for time-varying prices of covariance, co-skewness, and co-kurtosis risk, we find evidence of distinct bull and bear regimes. Ignoring such regimes, an unhedged US investor's optimal portfolio is strongly diversified internationally. The presence of regimes in the return distribution leads to a substantial increase in the investor's optimal holdings of US stocks, as does the introduction of skewness and kurtosis preferences. (/LG12,F30,C32)

Despite the increased integrationof internationalcapital markets, investors continueto holdequityportfoliosthatarelargelydominatedby domesticassets. According to Thomas, Warnock,and Wongswan(2006), by the end of 2003 US investorsheld only 14%of theirequityportfoliosin foreign stocks at a time This when such stocks accountedfor 54% of the world marketcapitalization.1 Calculationsreportedby Lewis (1999) suggest evidence is poorly understood: thata US investorwith mean-variancepreferencesshouldhold upwardsof 40% in foreign stocks or, equivalently,only 60% in US stocks. Potential explanationsfor the home bias include barriersto international investmentand transactioncosts (Stulz, 1981; Black, 1990; and Chaieb and
We thankthe editor,Cam Harvey,and an anonymousreferee for making many valuable suggestions. We also thankKarimAbadir,Ines Chaieb,CharlesEngel (a discussant),Serguey Sarkissian(a discussant),Lucio Sarno, at FabioTrojani,Giorgio Valente,and Mike Wickens,as well as seminarparticipants CatholicUniversityMilan, in FinancialManagement AnnualConferencein Milan(July Collegio CarloAlbertoFoundation Turin,European FinancialMarketsand 2005), HEC Parisand the Hong Kong MonetaryAuthorityConferenceon "International School of Management at the Macroeconomy" (July 2006). ImperialCollege TanakaBusiness School, Krannert Purdue,ManchesterBusiness School, Universityof Lund, Universityof Washington,Universityof York(UK), the ThirdBiennialMcGill Conferenceon GlobalAsset Management (June2007), WarwickBusiness School, and the WorldCongressof the EconometricSociety in London also providedhelpful comments. All errorsremain our own. Allan Timmermann acknowledgesresearchsupportfrom CREATES,funded by the Danish National Research Foundation.Address correspondenceto Massimo Guidolin, University of Manchester,Manchester Business School, MBS CrawfordHouse, Booth StreetEast, ManchesterM13 9PL, UK; telephone:+44-(0)161306-6406; fax: +44-(0)161-275-4023; e-mail: Massimo.Guidolin@mbs.ac.uk. 1 Similarhome biases in aggregateequity portfoliosare presentin othercountries;see Frenchand Poterba(1991) and Tesarand Werner(1994). The Author2008. Publishedby Oxford UniversityPress on behalf of The Society for FinancialStudies. All rightsreserved.For permissions,please e-mail:journals.permissions@oxfordjournals.org. Advance Access publicationFebruary 2008 doi:10.1093/rfs/hhn006 20,

The Reviewof Financial Studies!v21 n2 2008

Errunza, 2007); hedging demand for stocks that have low correlations with domestic state variables such as inflation risk or nontraded assets (Adler and Dumas, 1983; Serrat, 2001); information asymmetries and higher estimation uncertainty for foreign than domestic stocks (Brennan and Cao, 1997; Guidolin, 2005); and political or corporate governance risks related to investor protection (Dahlquist et al., 2004).2 As pointed out by Lewis ( 1999) and Karolyi and Stulz (2002), the first of these explanations is weakened by the fact that barriers to international investment have come down significantly over the last 30 years and by the large size of gross investment flows. Yet there is little evidence that US investors' holdings of foreign stocks have been increasing over the last decade, where this share has fluctuated around 10-15%. The second explanation is weakened by the magnitude by which foreign stocks should be correlated more strongly with domestic risk factors as compared to domestic stocks. In fact, correlations with deviations from purchasing power parity can exacerbate the home bias puzzle (Cooper and Kaplanis, 1994), as can the strong positive correlation between domestic stock returns and returns on human capital (Baxter and Jermann, 1997). It is also not clear that estimation uncertainty provides a robust explanation, as it affects domestic as well as foreign stocks. Finally, political risk seems to apply more to emerging and developing financial markets and is a less obvious explanation of investors' limited diversification among stable developed economies. Observations such as these lead Lewis (1999, p. 589) to conclude that "Two decades of research on equity home bias have yet to provide a definitive answer as to why domestic investors do not invest more heavily in foreign assets." In this paper we address whether a combination of investor preferences that put weight on the skewness and kurtosis of portfolio returns along with time-variations in international investment opportunities captured by regime switches can help explain the home bias and, if so, why US stocks may be more attractive to domestic investors than previously thought. Our analysis generalizes the standard International Capital Asset Pricing Model (ICAPM) specification that assumes mean-variance preferences over a time-invariant distribution of stock returns in two significant ways. First, we allow investor preferences to depend not only on the first two moments of returns but also on higher moments such as skewness and kurtosis. The motivation for the generalization to higher moments arises from studies such as Harvey and Siddique (2000); Dittmar (2002); and, subsequently, Smith (2007), which, in the context of three- and four-moment ICAPM specifications for the cross section of US stock returns, have found that higher-order moments add considerable explanatory power and have first-order effects on equilibrium expected returns. In addition, Harvey et al. (2004) have found that international asset holdings can
2 Behavioral or have been proposed(e.g., explanations(e.g., "patriotism" a generic preferencefor "familiarity") CovalandMoskowitz, 1999;andMorseandShive, 2003). UppalandWang(2003) providetheoreticalfoundations based on ambiguityaversion.

890

International Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences

be quite different under third-moment preferences compared to the standard mean-variance case.3 Second, we model returns by means of a four-moment ICAPM with regimes that capture time-variations in the risk premia, volatility, correlations, skewness, and kurtosis (as well as co-skewness and co-kurtosis) of local equity return indices and the world market portfolio. Studies such as Ang and Bekaert (2002) have shown that regime-switching models can successfully capture the asymmetric correlations found in international equity returns during volatile and stable markets, while Das and Uppal (2004) report that simultaneously occurring jumps that capture large declines in most international markets can affect international diversification. We go further than these studies and allow both the exposure of local markets to global risk factors and the world price of covariance, co-skewness, and co-kurtosis risk to vary across regimes. Both higher-order preferences and regimes turn out to play important roles in US investors' international asset allocation and thus help explain the home bias. Regimes in the distribution of international equity returns generate skewness and kurtosis and therefore affect the asset allocation of a mean-variance investor differently from that of an investor whose objectives depend on higher moments of returns. This is significant since the single-state model is severely misspecifled and fails to capture basic features of international stock market returns. Our estimates suggest that a US mean- variance investor with access to the US, UK, European, Japanese, and Pacific stock markets should hold only 30% in domestic stocks. The presence of bull and bear states raises this investor's weight on US stocks to 50%. Introducing both skewness and kurtosis preferences and bull and bear states further increases the weight on US stocks to 70% of the equity portfolio, much closer to what is observed empirically. To gain intuition for these findings, note that markets that have positive coskewness with the global market portfolio are desirable to risk-averse investors since they tend to have higher expected returns during volatile periods. Similarly, low co-kurtosis with global market returns means that local returns tend to be higher when world market returns are skewed to the left (i.e., during global bear markets), and is thus attractive since it decreases the overall portfolio risk. This turns out to be important because US stocks have attractive co-skewness and co-kurtosis properties. The US portfolio has a co-skewness of -0.05 and a co-kurtosis of 3.40 with the global market portfolio. In comparison, a fully diversified ICAPM portfolio has lower skewness (-0.50) and higher kurtosis (4.51). Moreover, the US portfolio also has better co-skewness and co-kurtosis properties than most of the other equity markets included in our analysis. Previous studies have found that foreign stocks form an important part of US investors' optimal portfolio holdings under mean- variance preferences.
3 Harvey et al. (2004) propose a Bayesian frameworkfor portfolio choice based on Taylor expansions of an of skewed underlyingexpected utility function.They assume thatthe distribution asset returnsis a multivariate normal. In their application to an internationaldiversificationproblem, they find that under third-moment preferences,roughly50% of the equity portfolioshould be investedin US stocks.

891

The Reviewof Financial Studies/ v2J n2 2008

However, the intuition above suggests that a US investor who dislikes negative co-skewness and high co-kurtosis with returns on the world market portfolio will put more weight on domestic stocks. The question then becomes how investors trade off between the mean, variance, skewness, and kurtosis properties of local market returns in an international portfolio context. Our paper addresses this issue when higher-order moments are modeled endogenously as part of an asset pricing model with regime shifts. The contributions of our paper are as follows. First, we develop a flexible regime-switching model that captures time-variations in the covariance, co-skewness, and co-kurtosis risk of international stock markets with regard to the world equity portfolio. We find evidence of two regimes in the distribution of international stock returns. The first regime is a bear state with low ex post mean returns and high volatility related to uncertainty spurred by market crashes, uncertain economic prospects during recessions, or uncertainty about monetary policy. The second regime is a bull state that is associated with less volatile returns and more attractive investment opportunities. Variations in the skewness and kurtosis of the world market portfolio are linked to uncertainty induced by shifts between such states. For example, the uncertainty surrounding a switch from a bull to a bear state takes the form of an increased probability of large negative returns (high kurtosis and large negative skewness). Second, we build on and generalize Harvey's (1991) findings of timevariations in the world price of covariance risk to cover variations in the world price of co-skewness and co-kurtosis risk. We find that co-skewness and co-kurtosis risk are economically important components of the overall risk premium with magnitudes comparable to the covariance risk premium. This finding has substantial asset allocation implications and is an important difference between our study and that of Ang and Bekaert (2002), who cannot reject that expected returns are identical across different states, in part due to large estimation errors. By estimating a constrained asset pricing model, which nests the two- and three-moment conditional ICAPM as special cases, we manage to identify significant variations in expected returns across the two states. Third, we analyze the international portfolio implications of time-varying higher-order moments. As in Harvey and Siddique (2000) and Dittmar (2002), our approach approximates the unknown marginal utility function by means of a Taylor series expansion of the utility function. Moreover, our analysis decomposes the effect of regimes and higher-order moments on portfolio weights. We find that US stocks are more attractive than is reflected in the standard meanvariance case due to their relatively high co-skewness and low co-kurtosis with the global market portfolio. Compared to other stock markets, US stocks tend to perform better when global markets are volatile or skewed to the left (i.e., during global bear markets). This gives rise to another important difference to the analysis in Ang and Bekaert (2002), who conclude that the presence of a bear state with highly volatile and strongly correlated returns does not negate the economic gains from international diversification. We show that the

892

International Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences

of relativelygood performance US stocks in the bear state can in fact help exthe higherallocationto the US marketthanin the benchmark plain single-state model. Intuitionfor this resultcomes from the highermarginalutility of additionalpayoffs duringglobal bearmarkets,which meansthatstock marketswith in to good performance these states tend to be attractive risk-averseinvestors. of The fourthand final contribution our paperis to develop a new, tractable to approach optimalasset allocationthatis bothconvenientto use andoffersnew When coupled with a utility specificationthat incorporates skewness insights. and kurtosis preferences, the otherwise complicated numerical problem of optimalasset allocationin the presenceof regime-switchingis reducedto that of solving for the roots of a low-orderpolynomial.While paperssuch as Ang and Bekaert(2002) use numericalmethodsto solve bi- or tri-variate portfolio problems,our paperemploys a moment-basedutility specificationthat offers advantagesboth computationallyand in terms of the economic intuitionfor how results change relative to the case with mean-variancepreferences.The to abilityof ourapproach solve theportfolioselectionproblemin thepresenceof since gains frominternational assets is important asset allocation multiplerisky can be quite sensitive to the numberof includedassets. The plan of the paperis as follows. Section 1 describesthe returnprocess in the contextof an ICAPMextendedto accountfor higher-order moments,timereturns,and regime-switchingand reportsempirical results for this varying model. Section 2 sets up the optimal asset allocation problemfor an investor with a polynomial utility function over terminal wealth when asset returns follow a regime-switchingprocess. Section 3 describes the solution to the optimalasset allocationproblem,while Section 4 reportsa rangeof robustness checks. Section 5 concludes. Appendicesprovidetechnicaldetails. 1. A Four-Moment ICAPM with Regime-Switching in Asset Returns Our assumptionsabout the returnprocess build on extensive work in asset stochasticdiscount factor model for (gross) pricingbased on the no-arbitrage asset returns,R't+l:

=\ E[Rit+xmt+x\Tt] / = !,...,/.

(1)

availableat time Here, [-1^] is the conditionalexpectationgiven information t^Tt, and mt+\ is the investor's intertemporal marginalrate of substitution between currentand futureconsumptionor- underrestrictionsestablishedby Brownand Gibbons(1985)- currentand futurewealth. The two-momentICAPM follows from this setup when the pricing kernel mt+\ is linear in the returnson an aggregatewealth portfolio. Harvey(1991) shows that, in a globally integratedmarket,differencesacross countryportfolios' expected returnsshould be driven by their conditionalcovarianceswith

893

The Reviewof Financial Studies/ v21 n2 2008

returns on a world market portfolio, R_x :

= E[R\+X\rt]-R{

^E^I^LCovK,,^^,].

(2)

Here, both equity returns R't+Xand the conditionally risk-free return R{ are expressed in the same currency (e.g., US dollars). The two-moment ICAPM in Equation (2) can be extended to account for higher-order terms such as Cov[rtJ+1, (RY+x)2\Ft\ and Cov[rtJ+1, (R^)3]^] that track the conditional co-skewness or co-kurtosis between the aggregate (world) portfolio and local portfolio returns. Such terms arise in a nonlinear model for the pricing kernel that depends on higher-order powers of returns on the world market portfolio. Consistent with this, and building on Harvey and Siddique (2000) and Dittmar (2002), suppose that the pricing kernel can be approximated through a third-order Taylor series expansion of the marginal utility of returns on aggregate wealth: mt+x = got + guR^x + 82t{R^x)2 + gyt(R?+i)\ (3)

where gjt = Uj+l /U1 is the ratio of derivatives of the utility function (/(1) = U' is the first derivative, etc.) evaluated at current wealth. Assuming positive marginal utility (U1 > 0), risk aversion (U" < 0), decreasing absolute risk aversion ([/"' > 0), and decreasing absolute prudence (U'm < 0), it follows that gXt < 0, g2t > 0 and g$t < 0.4 Negative exponential utility satisfies such restrictions and the same applies to constant relative risk aversion preferences. More generally, Scott and Horvath (1980) have shown that a strictly risk-averse individual who always prefers more to less and consistently (i.e., for all wealth levels) likes skewness will necessarily dislike kurtosis. Combining Equation (1) with the cubic pricing kernel in Equation (3) and assuming that a conditionally risk-free asset exists, we get a four-moment asset pricing model:

[*;+. \x] - r{ = YirCov(*;+1, \rt) + Y2,cov(*;+1, <, ,) + V3,Cov(/?;+l,1)V<)>

2|;f,) (4)

where yyr = -gjt^f 0 = 1, 2, 3), so yXt > 0, y2, < 0, and 73, > 0. This means that covariance and co-kurtosis risk earn positive risk premia, while co-skewness risk earns a negative risk premium, since an asset with a high return during times when returns on the world portfolio are highly volatile is desirable to risk-averse investors. The positive premium on co-kurtosis risk suggests that the standard ICAPM covariance premium carries over to "large"
4 Vanden (2006) argues that investors'preferencefor positively skewed portfolio returnsmay have far-reaching so implicationsfor the stochasticdiscount factor,to the point of makingoptions nonredundant that(powersof) theirreturnsenterthe expressionfor the equilibriumpricingkernel.

894

Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences International

returns.Co-skewnessearns a negativerisk premiumsince an asset with a high returnduringtimes when the world portfolio is highly volatile is desirableto risk-averseinvestors. By imposing restrictionson Equation(4), it is possible to obtain a variety of asset pricing models from the literatureas special cases. If 73, = 0 at all times,thenEquation(4) reducesto HarveyandSiddique's(2000) three-moment framework which only covarianceandco-skewnessarepriced.If y2t = Y3/= in then Equation(4) becomes a time-varying,conditionalICAPM: 0,

E[R',+l\r,] R{ = y,Cov(*,'+1, <,|.F,) = P,([<,|.F,] - R{), (5)


where both the risk premiumand the exposureto risk (measuredby the conditional beta) are time-varying. asset pricingmodels, there Despite its ability to nest a numberof important are good reasons to be skeptical about the exact validity of the four-moment model in Equation(4). On theoreticalgrounds,a reason for the failure of the context is that it requiresthe world ICAPMto hold exactly in an international market portfolio to be perfectly correlatedwith world consumption (Stulz, BekaertandHarvey(1995) show thatlimitedinternational 1981). Furthermore, meansthattermssuch as Var[/?J+1 will affectthe \Tt] capitalmarketintegration risk premium.On empiricalgrounds,conditionalICAPM specificationshave been tested extensively for internationalstock portfolios and been found to have significantlimitations.Harvey(1991) reportsthat not all of the dynamic behaviorof countryreturnsis capturedby a two-momentmodel and interprets the this as evidence of eitherincompletemarketintegration, existence of other sources of risk, or model misspecification.The four-momentICAPM priced also ignoresthe presenceof persistent"regimes"documentedfor asset returns in paperssuch as Engel and Hamilton(1990); Gray (1996); Perez-Quirosand Timmermann (2000); Whitelaw (2001); Ang and Chen (2002); and Guidolin and Timmermann (2006). 1.1 Regime switches in To allow for conditionaltime-variations the return processandthe possibility of misspecificationbiases, we extend the four-momentICAPM as follows. on First,consistentwith Equations(3) and(4), we assumethatreturns the world marketportfoliodependnot only on the conditionalvariance,Varf/?,^ |^v], but also on theconditionalskewness,Sk[R+x andkurtosis,K [Rf+X \Tt], \Tt], of this to without imposing Furthermore, obtain a flexible representation portfolio.5 the too much structure, price of risk associatedwith these momentsis allowed to depend on a latent state variable,S,+i, that is assumedto follow a Markov process but is otherwise not restricted.In turn this state dependence carries over to the price of the risk factors appearingin the equationsfor returnson
5 Conditional skewness and kurtosis are defined as Sk[R^+l\T,] = E[(R^+l EiR^lJ7,))3]^,] and

respectively. K[R^x\Tt]s [(**, - E(R?+l\Tt))4\Ttl

895

The Reviewof Financial Studies!v21 n2 2008

the individual stock market portfolios, denoted by Yi,s,+, (covariance risk), Y2,s,+,(co-skewness risk), and Y3,s,+i(co-kurtosis risk). Finally, consistent with empirical evidence in the literature (Harvey, 1989; Ferson and Harvey, 1991), we allow for predictability of returns on the world market portfolio through a vector of instruments, z,+i, assumed to follow some autoregressive process. Defining excess returns on the/ individual country portfolios, x\+x = Rlt+l - R{ , our model is: /?/"(/ = 1, ..., /), and the world portfolio, x^x = R^_x + *;+i = <*s,+1 Yi.s#+ICov[jc;+lf^,|^] + Y2,sr+ICov[^lf i)Vr]

+ Y3,5f+1cov[x;+1, 03|^]

+ bLz< + ii+i

+Ywl+I*[^i|Jrr]+b^+1^+Tij!;1
Z/+1 = + Mr,5f+I BrS,+.zr+ if+i(6) Consistent with the restrictions implied by the four-moment ICAPM, the risk premia Yy,s,+i 0 = 1,2,3) are common across the individual assets and the world market portfolio. However, we allow for asset-specific intercepts als that capture other types of misspecification. The innovations ~ N(0, --ilr+i = Wt+X T),7+1 aSt+l) can have a state-dependent tiJ^j (vf+i)'] covariance matrix capturing periods of high and low volatility. The predictor variables, zt+\ follow a first-order autoregressive process with state-dependent parameters Bz5f+1, reflecting the persistence in commonly used predictor variables. To complete the model, we assume that the state variable St+\ follows a K -state Markov process with transition probability matrix P: P[i, j] = Pr(s,+1 = j\st = i) = Pij, i,y = l,...f K. (7)

Our model can thus be viewed as a time-varying version of the multi-beta latent variable model of Ferson (1990), where both risk premia and the amount of risk depend on a latent state variable. Country-specific returns in the asset pricing model in Equation (6) depend on their covariances, co-skewness, and co-kurtosis with returns on the world portfolio. Estimating the skewness and kurtosis of asset returns is difficult (Harvey and Siddique, 2000). However, our model allows us to obtain precise conditional estimates in a flexible manner, as it captures skewness and kurtosis as a function of the mean, variance, and persistence parameters of the underlying states. Such model-based estimates are typically determined with considerably more accuracy than estimates of the third and fourth moments obtained directly from realized returns, which tend to be very sensitive to outliers. Moreover, as we show in Appendix A, when the world price of covariance, co-skewness, and

896

Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences International

co-kurtosis risk is identical across all markets, the model implies a tight set of restrictions across asset returns. To gain intuition for the asset pricing model in Equation (6), consider the special case with a single-state where the price of risk is constant and- because the innovations f|,+1 ~ N(0, it) are drawn from a time-invariant distributionthe higher-order moment terms Cov[jcJ+1,(jc^,)2!^], Cov[jcJ+1,(x^)3^]* Sk[x^_x|J>], and K[x^{ \Tt] are constant and hence do not explain variations in returns: *;+1 = a'"+ YiCov[*;+1, x, \Tt] + b'z, + <+ <, =avv+YiVar[<1|jrr]+bH/zr + i|f+1. + rirH;i, (8)

i,+i =|iz+Bzzf

This is an extended version of the ICAPM in which instruments (z,) are allowed to predict returns and alphas are not restricted to be zero ex ante. When the restrictions a' = a^ = 0 and bl = bw = 0 are imposed on all return equations, Equation (8) simplifies to the standard ICAPM, which sets

Yu = E[xy+l\ft]/Vai[x^{\Ftlso:

Elxt+\ K'J v vw \Ti

Elxt+\ K'J = PtElxt+\ Kd-

There are several advantages to modeling returns according to the general specification in Equation (6). Conditional on knowing the state of the next period, St+\, the return distribution is Gaussian. However, since future states are not known in advance, the return distribution is a mixture of normals with weights reflecting the current state probabilities. Such mixtures of normals provide a flexible representation that can be used to approximate many distributions. They can accommodate mild serial correlation in returns- documented for returns on the world market portfolio by Harvey (1991) - and volatility clustering since they allow the first and second moments to vary as a function of the underlying state probabilities (Timmermann, 2000). Finally, multivariate regime-switching models allow return correlations across markets to vary with the underlying regime, consistent with the evidence of asymmetric correlations in Longin and Solnik (2001) and Ang and Chen (2002). 1.2 Data In addition to the world market portfolio, our analysis incorporates the largest international stock markets, namely the United States, Japan, the United Kingdom, the Pacific region (ex-Japan), and continental Europe. More markets could be included, but parameter estimation errors are likely to become increasingly important in such cases, so we do not go beyond five equity portfolios in addition to the world market portfolio.6
6 At the end of 2005, these markets represented roughly97% of the worldequity marketcapitalization.

897

The Reviewof Financial Studies/ v2I n2 2008

Ifablel Summary statistics for international stock returns Portfolio Mean SD Skewness Kurtosis Jarque-Bera Ljung-Box Ljung-Box (squares) 2.4714 11.888* 0.4998 12.560* 19.845** 1.9827 1084.5**

MSCI United States MSCIJapan MSCI Pacificex-Japan MSCI Europeex-UK MSCI United Kingdom MSCIWorld US 1-monthT-bills

0.5415 0.3733 0.3892 0.4158 0.7503 0.4560 0.4906

4.4825 6.4830 7.0538 5.0578 6.1898 5.1740 0.2517

-0.7084 0.0700 -2.2723 -0.5672 0.7587 -0.8711 0.8319

5.9138 3.5044 22.297 4.6124 10.316 6.9133 3.9949

162.71** 4.2475 5655.6** 60.246** 865.27** 282.88** 58.250**

1.8775 6.5087 2.7472 5.9087 4.1915 2.3197 1248.2**

This table reportssample statistics for six international MSCI portfolios.Returnsare monthly,denominatedin US dollars, include dividends and are in excess of the 1-monthUS T-bill rate. The sample period is 1975:012005:12. Jarque-Berais a test for normalitybased on the skewness and kurtosis.Ljung-Box and Ljung-Box serial correlationin returnsand squaredreturns,respectively. squaresdenote tests for fourth-order * Denotes ** significanceat the 5% level. Denotes significanceat the 1%level.

Following common practice,we considerreturnsfrom the perspectiveof an unhedged US investorand examine excess returnsin US dollars on Morgan (MSCI) indices.7The risk-freerate is measured Stanley CapitalInternational the 30-day US Treasurybill rate providedby the Center for Research in by Security Prices. Our data are monthly and cover the sample period 1975:012005:12, a totalof 372 observations. and Returnsarecontinuouslycompounded for dividendsand othernoncashpaymentsto shareholders. number A adjusted of studieshave documentedthe leadingrole of US monetarypolicy andthe US interestrateas a predictorof returnsacross international equity markets,so we include the shortUS T-bill rate as a predictorvariable.8Again, our framework allows morevariablesto be includedat the cost of havingto estimateadditional parameters. Table 1 reports summarystatistics for the internationalstock returns,the worldmarketportfolio,andthe US T-billrate.Meanreturns positive andlie are in a rangebetween 0.37% and 0.75% per month.Returnvolatilitiesvary from 4% to 7% per month. Comparingthe performanceacross stock markets,US stock returns characterized a fairlyhigh meanandlow volatility.Returns are by in all but one market(Japan)are strongly nonnormal,skewed, and fat-tailed, such features.While suggestingthata flexible model is requiredto incorporate the US T-billrateis highly persistent,thereis little evidenceof serialcorrelation in stock returns.However,many of the returnseries display strongevidence of time-varyingvolatility. L3 Empirical results PanelA of Table2 reportsparameter estimatesfor the benchmark single-state, two-momentICAPM in Equation(8). Alphas are positive in five regions and economically large but imprecisely estimated and statistically insignificant.
7 This is consistent with other authors' hold large and liquid foreign findings that US investorspredominantly stocks, such as those thatdominatethe MSCI indices (Thomas,Wamock,and Wongswan,2006). 8 See Obstfeldand of Rogoff ( 1995) for the microfoundations such models andKim (2001) for empiricalevidence.

898

International Asset Allocation under Regime Switching, Skew, and Kurtosis Preferences

The model's failure to capture returns in Japan is consistent with the strong rejections for Japan in the two-moment ICAPM tests reported in Harvey (1991) and is perhaps to be expected in view of the gradual liberalization of financial markets in Japan during the 1980s and the analysis in Bekaert and Harvey (1995). The negative coefficients on the lagged T-bill rate are also consistent with existing literature. At 5.3, the estimated world price of covariance risk yi is positive and significant as expected. Next, we consider the asset pricing model with two states, estimates of which are shown in Panel B of Table 2. To reduce the number of parameters, we impose two sets of constraints on the general model in Equation (6). First, the regression coefficients on the lagged T-bill rate were found to be insignificant for all stock markets in the first state, and hence we impose that these coefficients are zero. In the second state, the coefficients on the T-bill rate are large and negative and most are significant. Second, we impose that the correlations (but not the variances) between country-specific innovations, Coit(t]}+1 , r\Jt+x are the same ), in the two states. This restriction is again supported by the data and does not imply that the correlations between country returns (Corr(jtJ+1,jc/+1)) are the same in the two states since state-dependence in both the alphas and the bls and bf+[ coefficients generates time-variations in return correlations.9 As we shall see below, the economic interpretation suggested by the estimates reported in Table 2 is that state one is a bear state where returns have low (ex post) means and high volatility, and are more strongly correlated across markets. Conversely, state two is associated with more attractive, less uncertain, and less correlated return prospects. Figure 1 shows that the two states are generally well identified with state probabilities near zero or one most of the time. The bear state occurred during the three-year period between 1979 and 1982 when the Fed changed its monetary policy and again during shorter spells in 1984, 1987, 1990/1991, and 2002. These periods coincide with global recessions (the early 1980s, 1990s, and 2002 recessions) and occasions with high return volatility, such as October 1987. Common to these episodes is the high degree of uncertainty about economic prospects and the associated high volatility of global equity returns. In fact, volatility is highest in the first state for all equity portfolios with the exception of the UK.10 The persistence of the first state (0.90) is lower than that of the second state (0.94) and so the average duration of the first state (10 months) is shorter than that of the second state (20 months). In steady-state, one-third and two-thirds of the time is spent in states one and two, respectively. Neither of the states

9 A likelihood ratio test of the restriction that correlations do not depend on the state, i.e., Cov(tjJ+1, n/+1) = , produces a p-value of 0. 1 1 and is not rejected. , ti/+ , ^ aJs Corr(Ti}+, 10 The finding for the UK is due to two outliers in January and February of 1975 with monthly excess returns of 44% and 23%. If excluded from the data, the volatility in the first state is highest also for the UK.

899

TheReviewof Financial Studies/ v2l n2 2008

S3 *S

oo o

o # ^ #

Q rn O O

^ O

Oo\

OO

. do

Oq . d o

On d

1
*^

I It
2 ' " I
O O0 ""*

8
I 00 ^g 00

1|
^ *~
<Js 00 ^ I

f-

? 1 I? I 11

!i|

g.S
o

B0^
soO'

^SoJnoo^nnvooB

SS
^ f

dw<N

oovor-^o

wnw(N

8o

cK7^

7-

vq2 d ?

vp

OOOOO

'

^ -

I
if
900

1 l^sijii'jisis

HI ahzis

International Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences

iS

^ 3

L *

Ei s | o *- <Gg

1
g
:.
.

t
I"

fil Hit
t
+

If
II

Ii2"*
r^l'

S-k

L r-

S =-

a I * _ ,t <1 ^ ^ + i| -^+

i 1 i-i 11*1 i
5^-S -

:1 *8 g- --g

5 ..

..

*I 3Si 5 ^ I 1 $ *t ^_

r it ^ *

i il i"

g^-gg

il

4lh
CU

^ il
+ ^ CL,

T3
^+ ?C

Mljf
, |K!h!
dfl Hi^i
" 8 '5 -I
>" C O S "

g t Hi fi
| as -ss^a s s i s^

>T.+

i ii r ? ? iinis
^ u U s * -,!>- *^" N-M~

1;S

. S! 5S

,S

gg"

S55?|8j.^

i Is

S2S2S

ill?
2

HHK

rtjllii
cxT 1 O..P g "^

-a 8.

^.!2og;

901

The Reviewof Financial Studies/ v21 n2 2008

Figure 1 Bear state probabilities.

identifies isolated "outliers" or jumps - a feature distinguishing our model from that proposed by Das and Uppal (2004). It is interesting to compare the alpha estimates for the single-state and twostate models. Alpha estimates are negative in state 1 but positive in state 2 for all portfolios. The alphas in the two states may appear to be quite large in economic terms.11 However, as they measure returns conditional on being in a particular state and the state is never known in advance, they are not directly comparable to the corresponding estimates from the single-state model. To account for this, we simulated 50,000 returns from the two-state model over a 12-month horizon, allowing for regime shifts and uncertainty about future states. Measured this way, the 12-month alphas starting from the first and second states are 0.06 and 0.70 for the United States, while those for Japan are -0.45 and 0.86. The world portfolio generates alphas of -0.13 and 0.70, starting from the first and second states, respectively. All other estimates of the alphas in the two regimes shrink toward zero. Hence, although the individual state alphas appear to be quite large conditional on knowing the true state, in many regards they imply weaker evidence of mispricing than the single-state model, which assumes that nonzero alphas are constant and constitute evidence of permanent model misspecification or mispricing. Figure 2 shows that, consistent with previous studies (Longin and Solnik, 1995, 2001; Karolyi and Stulz, 1999; and Ang and Bekaert, 2002), return correlations are higher in the bear state than in the full sample. Pairwise correlations between US stock returns and returns in Japan, Pacific ex-Japan, UK, and Europe in the bear (bull) states are 0.39 (0.27), 0.65 (0.47), 0.67 (0.48), and 0.59 (0.45) and are thus systematically higher in the bear state. This happens
1' Furthermore, the alphas in the two states are sufficientlyprecisely estimated that the hypothesis that they are equal to zero is very stronglyrejectedby a likelihoodratiotest.

902

International Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences

Figure 2 Correlations between world marketand regionalmarketreturnsperiods. Periods where the global bear state is most likely are shown in gray shades.

despite the fact that correlations between return innovations are identical in the two states. In part this is due to the higher volatility of the common world market return in the bear state. Furthermore, since mean returns are different in the two states, return correlations also depend on the extent of the covariation between these parameters. To help interpret the two states and gain intuition for what leads to changes in skewness and kurtosis, it is useful to consider the time-variation in the conditional moments of the world market portfolio. To this end, Figure 3 shows the mean, volatility, skewness, and kurtosis implied by our model estimates, computed using the results in Appendix A.I. Consistent with our interpretation of state 1 as a bear state, mean excess returns are lower in this state, while

903

The Reviewof Financial Studies/ v21 n2 2008

Figure 3 Mean excess returns,volatility, skewness, and kurtosisof the world marketportfolio implied by the two-state model (annualizedfigures).Periodswhere the global bearstate is most likely are shown in gray shades.

conversely the volatility of returns is much higher.12 Moreover, large changes in the conditional skewness and kurtosis turn out to be linked to regime switches. Preceding a shift from the bull to the bear state, the kurtosis of the world market portfolio rises while its skewness becomes large and negative and volatility is reduced. Uncertainty surrounding shifts from a bull to a bear state therefore takes the form of an increased probability of large negative returns. Once in the bear state, the kurtosis gets very low and the skewness close to zero, while world market volatility is much higher than normal. Hence the return distribution within the bear state is more dispersed, although closer to symmetric. Finally, when exiting from the bear to the bull state, the kurtosis again rises - reflecting the increased uncertainty associated with a regime shift- while volatility and skewness decline to their normal levels. These large variations in the volatility, skewness, and kurtosis of world market returns mean that our model is able to capture the correlated extremes across local markets found to be an important feature of stock returns in Harvey et al. (2004).

12 Notice that,consistentwith basic

intuition,the expected excess returnon the world portfoliois nevernegative.

904

International Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences

1.4 Time-variations in risk premia Furthereconomic intuitioncan be gained from studyingvariationsin the risk on premiain Table2. The premiumon covariancewith returns the worldmarket portfolio (yi) is positive in both states but, at 15.9, is much higher in the bull statethanin the bearstatefor which an estimateof 9.5 is obtained.The number reportedby Harvey (1991) for the subset of G7 countries is 11.5 and hence lies betweenthese two values. Consistentwith the largedifferencebetweenthe covariancerisk premiumin the bull and bear state that we find here, Harvey rejectsthatthe world price of risk is constant. A similarconclusionholdsfor theco-kurtosispremium(73), whichis positive and insignificantin the bear state but positive and significantin the bull state. After suitable scaling, the estimates of 73 can be compared to the price of covariancerisk, 72.13This yields a price of co-kurtosis risk of 1.7 and 12.3 in the bear and bull state, respectively,and a steady-stateaverage of 8.7. As expected, the co-skewness premium(72) is negative in both states, althoughit is only significant(and by far largest)in the bull state. When convertedto the same units as the covariancerisk premium,the estimates are -1.1 and -3.1 in the bearand bull state, respectively,while the steady-stateaverageis -2.4. Both the price of risk andthe quantityof risk arerequiredto show how much co-skewness risk and co-kurtosisrisk contributeto expected returnscompared estimates from Table 2, we find that to covariancerisk. Using the parameter covariancerisk (measuredrelative to the world marketportfolio) contributes roughlythe same amountto the risk premiumin all markets,namely between 2.7%and 3.3%per year.Co-skewnessrisk premiavarymore cross-sectionally, namely from 0.6% per year in Japanto 2.6% for Pacific stocks. Finally, coin between0.5%and 1%to expectedreturns annualized kurtosisriskcontributes terms.Forfourof the six portfolioswe studyhere (includingthe US andWorld portfolios),the combinedco-skewness and co-kurtosisrisk premiumis within 1%of the covariancerisk premium. We conclude from this analysis that the coefficients on covariance, coskewness, and co-kurtosisrisk have the expected signs and are economically meaningful:Investorsdislike risk in the form of highervolatility or fattertails the but like positively skewed returndistributions.Furthermore, co-skewness in and co-kurtosisrisk premiaappearto be important economic terms,as they are of the same orderof magnitudeas the covariancerisk premium.

13

Scaling is requiredfor meaningfulcomparisons.For instance, yi measures the covariancerisk premiumper unit of covariancerisk, Cov[jt,'+1jc,^,\F,], while y2 measuresthe co-skewness risk premiumwith referenceto , x.^?,] and Cov[jcJ+1, (jc,^,)2!^] are measuredin differentunits (coCov[a:/+,,(jc,^,)2|.F,]. Since Cov[jcJ+i, skewness involves squaredreturns),they cannotbe compareddirectly.Because the scale of Cov[jc,'+,jc^, \Ft] , is similarto VarU^ \Tt] and the scale of Cov[jcJ+1 , (jc^,)2|/",] is similar to |S*[jt,^i |/J]|, the transformation y2 = y2 x Var[jfr^I|/J]/|5ifc[jcl^,|/i]| leads to a new coefficient, y2, which is comparableto yi. Similarly, y3 = y3 x Var[^, \T,)/K[x^ \?t] can be comparedto y, .

905

The Reviewof Financial Studies!v21 n2 2008

1.5 Are two states needed? A questionthatnaturally arisesin the empiricalanalysisis whetherregimesare in the distribution international returns. answer To of stockmarket reallypresent we computedthe specificationtest suggestedby Davies (1977), whichvery this, strongly rejected the single-state specification.14Inspection of the residuals from the single-state model confirmedthat this model fails to captureeven the most basic propertiesof the international returnsdata while the residuals from the two-state model (standardized subtractingthe conditional mean by and dividing by the conditional standarddeviation) were much closer to the model assumptions. 2. Investor's Asset Allocation Problem We next turn to the investor's asset allocation problem. Consistent with the analysisin the previoussection, we assumethatinvestorpreferencesdependon momentsof returns allow regimesto affect the return and higher-order process. 2.1 Preferences over moments of the wealth distribution Supposethatthe investor'sutilityfunctionU(Wt+T\0) only dependson wealth at time t + T, Wt+r, and a set of shape parameters, where t is the current 6, time and T is the investmenthorizon. Consider an mth-orderTaylor series expansionof U aroundsome wealth level vt :
m -

I/W+7.;8)= T -Uin\vT;*)(Wt+T vTf + Sm,

(10)

where the remainder qm is of order o((Wt+r - vr)m) and U\vt\Q) = U(vt\Q). /(n)() denotes the nth derivativeof the utility function with respect to terminalwealth. Providedthat (i) the Taylorseries converges;(ii) the distribution wealthis uniquelydetermined its moments;and(iii) the order of by of sums andintegralscan be exchanged,the expansionin Equation(10) extends to the expected utility functional:
m .

Et[U(Wt+Tl9)] = T -Uin\vT;*)Et[(Wt+T - vT)n]+ E,[Sw], ^-^ n\

(11)

where Et[.] is shortfor E[.\Tt], For instance,Tsiang (1972) shows that these conditions are satisfiedfor negativeexponentialutility when asset returnsare drawnfrom a multivariate distributionfor which the first m centralmoments
14

Regime-switching models have parametersthat are unidentifiedunder the null hypothesis of a single state. Standardcritical values are thereforeinvalid in the hypothesis test. Details of the analysis are available upon request.

906

Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences International

exist. We thus have


m

E,[U{WI+T;%)] E,[Um(Wt+T;*)]= T -:U{"\vT;W,[(W,+T

(12) While the approximationimproves as m gets larger setting m = 2 or 3 is for likely to give accurateapproximations CARA utility, accordingto Tsiang - manyclasses of Von-Neumann (1972) Morgenstern expectedutilityfunctions a relativelysmall value of m and a functionof can be well approximated using the form15
m

tt>nl

- vT)n].

E,[Um(Wt+T;*)]= J2KnE,[(Wl+T - vT)n], with Ko> 0, and Knpositive (negative)if n is odd (even).

(13)

2.2 Solution to the asset allocation problem Wenextcharacterize solutionto the investor'sassetallocationproblemwhen the preferencesaredefinedover momentsof terminalwealthwhile, consistentwith follow a regime-switchingprocess.Following the analysisin Section 1, returns most papers on portfolio choice (e.g., Ang and Bekaert, 2002; and Das and that Uppal,2004), we assume a partialequilibriumframework treatsreturnsas exogenous. The investormaximizes expected utility by choosing among / risky assets whose continuouslycompoundedexcess returnsare given by the vector xj = = co^ (xj x? jc/y. Portfolio weights are collected in the vector co,1 (co,1 - coji/) is investedin a short-term bond, where interest-bearing ($1)'while (1 i/ is an / x 1 vector of ones. The portfolio selection problem solved by a buy-and-holdinvestorwith unit initial wealth then becomes max Et[U(Wt+T(<*t);9)]
0),

s.t. Wt+T(ut)= {(1 - (o;i,)expr) + a>;exp(R?+r)}, (14) is the vecwhere Rst+j= (x*+1+ r,*+1) (xst+2 r^2) + + (xj+r + + + r^T) tor of continuouslycompoundedequity returnsover the T -period investment
15 For ( power utility,Tsiang (1972) and Krausand Litzenberger 1976) provethatthe condition

= Pri\h\= \W,+T-VT\<E,[Wl+T]} \
is requiredfor the series (Et[Wt+T]y-y + 1- y _ 1 2 ]ry-iA2 + LEt[Wl+T]r*-2h3 6

+ -(-\)m-^(Et[wt+T]yy-m+'hm
to converge.This correspondsto imposing a bound on the amountof risk accepted by the investor.In general, convergenceis slower than in the exponentialutility case and dependson the investmenthorizon,T .

907

The Reviewof Financial Studies!v21 n2 2008

= horizon while R^+T r*+1+ r*+2-\ hrf+r is the continuously compounded bond return.Accordingly, exp(RJ+r) is a vector of cumulatedreturns. Short-selling can be ruled out through the constraintcoj e [0, 1] for 1 = 1,2,...,/. For generality,we assumethe following process for a vectorof / + 1 excess returns (the last of which can be taken to representthe risky returnson a short-term bond, jcf+I= rf+T):16
p

+ X,+i = pL5/+| ^By,5,+IX,_y + 8,+i,

(15)

= where jLs/+I ([ilSt+l . . , H-',*1)' a vector of conditionalmeans in state S,+i is ,. is used to "fold in" all componentsof the mean in state St+\), B7ts/+1 (possibly a matrixof autoregressive coefficientsassociatedwith the yth lag in state S,+i , and e,+i = (fij+p . . . , e,7*/)'~ A^(0, fts,+1)is a vector of zero-meanreturninnovationswith state-dependent covariancematrix&st+l With / + 1 risky assets and K states, the wealth process becomes

= o>; <16> +rf+T)l Wt+T exp (x/+T + (1- ;i/)exp | [E^+tI


We next presenta simple andconvenientrecursiveprocedurefor evaluatingthe expected utility associatedwith a vector of portfolio weights, <o,,of relatively high dimension. Proposition 1. Under the regime -switchingreturnprocess in Equation(15) and m-momentpreferences in Equation (13j, the expected utility associated with the portfolio weights o>,is given by

= J2*n ^-^'^t'^CM^+t] E,[Um(Wt+T)}


n=0 j=0
=

E^E(-Dn^r(;)E(0
n=0 j=O VJ ' <=0 V '

x E,
16 This

((l-;i*)exp(rr/)y-'. (17) [(; exp(R?+r))']

equationis more convenientto use thanEquation(6) but is fully consistentwith the earliersetup if the last elements of the returnvector, r,+i, are used to capturethe predictorvariables,z,+i, which may themselves be asset returns.

908

International Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences

The nth momentof the cumulatedreturnon the riskyasset portfolio is

....,) r[(a,;exp(R?+r))"]= X(n,,n2, (f\A -


/i,=0 /=0 \/=l /

xM^i,...,*/), vWiere ,7=i /!,-= n, 0 < m < n (i = 1, . . . , /),


\(nx,n2, ...,/,) = -

(18>

n\

ni\n2l-nhl

-.

(19)

and M;+T(n\, ...,/) can be evaluatedrecursively,using Equation (B12) in AppendixB. Appendix B proves this result. The solution is in closed form in the sense thatit reducesthe expected utility calculationto a finite numberof steps, each of which can be solved by elementaryoperations. Given their recursive structure,these results are complex and difficult to analyze. AppendixB thereforeuses a simple two-statemodel to illustratethe resultwith a single riskyasset. Herewe use the setupof the model fromTable2 to provide intuitionfor Proposition1 in terms of the underlyingdeterminants of the optimalasset allocation: 1. The currentstate probabilities(tt,, 1 - nt) are particularlyimportantfor investors with a short horizon. Startingfrom the bear state, investment prospects are less favorablethan startingfrom the bull state since there is a higher chance of remaining in the initial state. Stock marketswith relatively good performancein the bear state (relative to other markets) when startingfrom this state. How far tt, is removed will thus be preferred fromzero or one reflectsinvestors'uncertainty aboutthe currentstate.The more uncertainthey are, the less aggressivethe resultingasset allocation. 2. State transition probabilitiespij affect the speed of mean reversiontoward set. the steady-stateinvestmentopportunity The closer the "stayer" probaare bilities p\ i , /?22 to one, the more persistentthe individualstates will be and hence the more the initial state matters.Conversely,if one state has a very low "stayer" probability,then this state is more likely to capturethe occasional outlieror jump in asset prices, as in Das and Uppal (2004). 3. Differences between mean parameters(jjli, p^) and varianceparameters 02) (<7i, across states are importantsince skewness can only arise in the model providedthatexpectedreturns differacrossstates, regime-switching i.e., |ii ^ |X2,while kurtosisis stronglyaffectedby differencesin variance in 2000). The greaterthe differences parameters the states (Timmermann, between expected returnsand volatilities in the bear and bull states, the largerthe role played by skewness and kurtosisrisk. Risk-averseinvestors preferto invest in countrieswith relativelygood performancein the bear

909

The Reviewof Financial Studies/ v21 n2 2008

4.

5.

state since these provide a hedge against the poor performance of the world market portfolio and since the marginal utility of payoffs is higher in this state. Investor preferences, as captured in part by m, are the number of higherorder moments that matter to the investor, in part by the weights assigned to the various moments, which we discuss further below. Going from m = 2 to m = 3 or m = 4, we move from mean- variance preferences to a setup where skewness and kurtosis matter as well. Moreover, as the weight in the utility function on skewness and kurtosis increases, investors become more sensitive to states with high volatility and higher probability of negative returns. This means that the significance of such states in determining the optimal asset allocation grows, as does the weight on countries with relatively attractive co-skewness and co-kurtosis properties. The investment horizon T plays a role in conjunction with the average duration of the states. The shorter the investment horizon and the more persistent the states, the more sensitive the investor's asset allocation will be with respect to the current state probability. As the investment horizon grows, the return distribution will converge to its "average" value, and so the asset allocation becomes less sensitive to the initial state and more sensitive to the steady-state probabilities.

It is useful to compare the solution method in Proposition 1 to existing alternatives. Classic results on optimal asset allocation have been derived for special cases, such as power utility with constant investment opportunities or under logarithmic utility (Merton, 1969; Samuelson, 1969). For general preferences, there is no closed-form solution to the problem in Equation (14), but given its economic importance it is not surprising that a variety of solution approaches have been suggested. Recent papers that solve the asset allocation problem under predictability of returns include Brennan, Schwarz, and Lagnado (1997); Brandt (1999); Campbell and Viceira (1999, 2001); and Ang and Bekaert (2002). These papers generally use approximate solutions or numerical techniques, such as quadrature (Ang and Bekaert, 2002) or Monte Carlo simulations (Detemple, Garcia, and Rindisbacher, 2003), to characterize optimal portfolio weights. Quadrature methods may not be very precise when the underlying asset return distributions are strongly nonnormal. They also have the problem that the number of quadrature points increases exponentially with the number of assets. Monte Carlo methods can be computationally expensive to use, as they rely on discretization of the state space and use grid methods.17 Although existing methods have clearly yielded important insights into the solution of the problem in Equation (14), they are therefore not particularly well suited to our analysis of international asset allocation, which involves a large number of portfolios.
17 In continuoustime, closed-form solutions can be obtainedunderless severe (1996). restrictions;see Kim and Omberg

910

Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences International

3. International Portfolio Holdings asset allocation under We next consider empiricallythe optimal international and four-momentpreferences.The weights on the first four regime-switching are to momentsof the wealthdistribution determined ensurethatourresultscan be comparedto those in the existing literature.Most studies on optimal asset allocationuse power utility,so we calibrateour coefficients to the benchmark:

= Wl~e u(wt+T;e) -^-iw e>o. i-

(20)

For a given coefficient of relativerisk aversion,6, the utility functionin Equain tion (20) serves as a guide in setting values of {kw}=0 the expansion in (13) but should otherwise not be viewed as an attemptto approxiEquation mate results under power utility. Expandingthe powers of (Wt+r - vt) and taking expectations,we obtain the following expression for the four-moment preferencefunction: + + Et[U\Wt+T;e)] = KO,r(0) KhT(d)Et[Wt+T] K2,r(0),[W,2+r] + + K3,r(0),[W,3+r] K4,T(fi)Ei[W?+Tl (21) where18

= + K0,r(0) 4"e[(i - er1 - 1 - ie - 0(0 d - o<o+ ixe + 2)],


e ci.HG) = zvt [6 + 60 + 30(0 + 1) + 0(0 + l)(0 + 2)] > 0, K2T(0)= -\Qv-(l+Q) [2 + 2(0 + 1) + (0 + l)(0 + 2)] < 0, (2+0)> 0, K3,r(6)= 6(0 + D(8 + 3)vf K4,r(e)= -^0(0 + l)(0 + 2)uf(3+e) < 0. (22)

Expected utility from final wealth increases in Et[Wt+r] and Et[W*+T],so higher expected returnsand more right-skeweddistributionslead to higher expected utility. Conversely,expected utility is a decreasing function of the Our second andfourthmomentsof the terminalwealthdistribution. benchmark resultsassumethat0 = 2, a coefficientof relativeriskaversioncompatiblewith much empiricalevidence.19 A solution to the optimalasset allocationproblemcan now easily be found from Proposition1 by solving a system of cubic equationsin <o,derivedfrom

18 The notation

makes it explicit that the coefficients of the fourth-order Taylor expansion depend on the Kn<r investmenthorizon throughthe coefficient vT, the point around which the approximationis calculated. We = Et[W,+T-\]. follow standard practiceand set vt

19 Based on the evidence in Ang and Bekaert(2002), who show thatthe optimalhome bias is an increasingfunction of the coefficientof relativerisk aversion,this is also a conservativechoice.

911

TheReviewof Financial Studies!v21 n2 2008

the first-order conditions: Vw,,[/4(W,+r;e)]|&, =0'. (23)

At the optimum,6, sets the gradient,V^tEt[UA{Wt+T\ equal to zero and 6)], 6)]. Et[U4(Wt+T\ producesa negativedefiniteHessian matrix,H&, 3.1 Empirical results As a benchmark,Table 3 first reportsequity allocations for the single-state model using a short 1-month and a longer 24-month horizon. Our empirical analysis considers returnson five equity portfolios and the world market.To arriveat total portfolio weights, we thereforereallocate the weight assigned to the world market using the regional marketcapitalizationsas weights.20 Since we are interestedin the home equity bias, we reportequity weights as percentagesof the total equity portfolioso they sum to unity.The allocationto the risk-freeasset (as a percentageof the total portfolio) is shown for interest deviationsfromthe mean.Whenthe T-bill ratesthatvaryby up to two standard rate is set at its sample mean of 5.9% per annum,at the 1-monthhorizononly 31% of the equity portfolio is invested in US stocks. Slightly less (29%) gets invested in US stocks at the 24-month horizon. Thus, in both low- and highinterest-rate environments,the fractionof the equity portfolioallocatedto US stocks remainsconsiderablyshort of the percentagestypically reportedin the These resultssupportearlierfindingsundermean-variance empiricalliterature. preferences(e.g., Lewis, 1999) andalso show thatthe home bias puzzle extends from the shortT-bill rate. to a setting with returnpredictability to Turning the two-statemodel, Table3 shows thatthe allocationto US stocks is much higher in the presence of regimes. This holds when startingfrom the steady-stateprobabilities(i.e., when the investor has imprecise information about the currentstate) as well as in the separatebull and bear states. Under US steady-stateprobabilitiesand assumingan averageshort-term interestrate, the 1-month allocation to US stocks is 70% of the total equity portfolio.This reflects an allocationof 75% in the bear state and an allocationof 60% in the bull state. asset pricingmodel These resultsshow thata four-moment regime-switching can substantiallyincrease the optimal weights on US stocks. Moreover,this finding is robust to the level of the short US interest rate. Varyingthis rate predominantlyaffects the allocation to the risk-free asset versus the overall equity portfolio but has little effect on the regionalcompositionof the equity portfolio.21
20 This introduces a very small approximation error, as the included stock markets account for only 97% of the world market. 21 The allocation to the short-term bond is much higher in the bear state than in the bull state. This happens because equity returns are small and volatile in the bear state and hence unattractive to risk-averse investors.

912

International Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences

Table 3 Optimal portfolio weights under the single-state and two-state models Mean- 2 x SD Mean- 1 x SD Mean Mean + 1 x SD Panel \:T = \ month Single-state benchmark 0.228 United States 0.261 Japan Pacific(ex-Japan) 0.348 0.130 United Kingdom 0.033 Europe(ex-UK) 0.082 UST-bills Bear state (x = 1) United States 0.697 0.121 Japan Pacific(ex-Japan) 0.061 0.030 United Kingdom 0.091 Europe(ex-UK) 0.668 UST-bills Steady-state probabilities (k = 0.33) 0.625 United States 0.125 Japan 0.063 Pacific (ex-Japan) 0.016 United Kingdom 0.172 Europe(ex-UK) UST-bills 0.357 Bull state (x = 0) 0.535 United States 0.198 Japan 0.116 Pacific (ex-Japan) United Kingdom 0.023 0.128 Europe(ex-UK) 0.142 UST-bills Single-state benchmark United States 0.310 0.310 Japan Pacific (ex-Japan) 0.230 0.149 United Kingdom 0.000 Europe(ex-UK) 0.132 UST-bills = 1) Bear state (x United States 0.595 0.139 Japan Pacific (ex-Japan) 0.089 0.127 United Kingdom 0.051 Europe(ex-UK) 0.208 UST-bills Steady-state probabilities (x = 0.33) 0.590 United States 0.108 Japan Pacific(ex-Japan) 0.048 0.084 United Kingdom 0.169 Europe(ex-UK) 0.168 UST-bills Bull state (x = 0) United States 0.565 0.087 Japan 0.054 Pacific(ex-Japan) 0.054 United Kingdom 0.239 Europe(ex-UK) 0.083 UST-bills 0.346 0.309 0.222 0.099 0.025 0.191 0.722 0.093 0.056 0.056 0.074 0.462 0.696 0.101 0.072 0.000 0.130 0.312 0.313 0.375 0.104 0.042 0.167 0.515 0.746 0.063 0.048 0.048 0.095 0.370 0.685 0.110 0.082 0.000 0.123 0.269 0.156 0.375 0.000 0.000 0.469 0.681 0.772 0.070 0.035 0.070 0.053 0.431 0.817 0.070 0.070 0.000 0.042 0.289 0.656 0.086 0.043 0.011 0.204 0.070

Mean -I-2 x SD

0.101 0.499 0.000 0.000 0.400 0.800 0.796 0.093 0.037 0.074 0.000 0.460 0.851 0.060 0.060 0.000 0.030 0.333 0.713 0.085 0.032 0.011 0.160 0.062

0.537 0.598 0.116 0.098 0.074 0.054 0.021 0.022 0.253 0.228 0.053 0.078 Panel B: T = 24 months 0.367 0.304 0.177 0.152 0.000 0.208 0.603 0.141 0.090 0.128 0.038 0.22 0.593 0.105 0.058 0.081 0.163 0.142 0.596 0.090 0.056 0.045 0.213 0.109 0.286 0.381 0.190 0.095 0.048 0.578 0.623 0.130 0.091 0.117 0.039 0.226 0.635 0.094 0.059 0.071 0.141 0.149 0.640 0.093 0.058 0.023 0.186 0.140

0.000 0.306 0.000 0.056 0.639 0.639 0.618 0.118 0.079 0.118 0.066 0.240 0.627 0.108 0.072 0.060 0.133 0.168 0.655 0.080 0.057 0.023 0.184 0.132

0.000 0.367 0.000 0.000 0.633 0.698 0.597 0.125 0.069 0.125 0.083 0.278 0.622 0.110 0.085 0.061 0.122 0.179 0.678 0.080 0.046 0.034 0.161 0.131

as Stock holdingsarereported a fractionof the totalequityportfolio(andthus sum to 1), while the T-billholdings are shown as a percentageof the total portfolio. Allocations are computedunderinterestrates that can deviate deviationsfrom theirmean. by up to two standard

913

TheReviewof Financial Studies!v2I n2 2008

While the next section explains these results in the context of higher-order moments, preliminaryintuitioncan be gained in terms of how well the stock portfolios performin the "bad"(bear) state. During bear markets,US stocks performrelativelybetterthanthe othermarketswith a higherSharperatiodue in part to higher mean returnsand in part to lower volatility. This turns out to be especially important here since states with poor returnstend to be more under four-momentpreferencesthan under mean-variance heavily weighted preferencesand helps explain why the two-state model, which distinguishes between returndistributions good and bad states,leads to higherallocations in to US stocks thanthe single-statemodel, which does not make this distinction. Of course,explanationsbasedonly on the firsttwo momentsmerely scratchthe surface of the issue here. We thereforenext turnto the effect of higher-order momentson international portfoliochoice. 3.2 Effects of higher moments to model, our four-moment Compared the benchmark regime-switchingmodel is able to significantly increase the allocation to US stocks. An economic understandingof the effect of skewness and kurtosis on the optimal asset allocationrequiresstudyingthe co-skewness and co-kurtosispropertiesat the portfoliolevel. To this end, define the conditionalco-skewness of the returnon marketi with the world marketas

v (T m_ 0/ WW f \) -

cov[*;+1,(^,)2|jr(,s,]

TT77-

(24)

{v4,;+1|^,5,](VarK,|^,S,])2}1/2
The co-skewness is normalizedby scaling by the appropriate powers of the volatilityof the respectiveportfolios.A securitythathas negativeco-skewness with the marketportfolio pays low returnswhen the world marketportfolio becomes highly volatile.To a risk-averseinvestor,this is an unattractive feature since global marketrisk rises in periodswith low returns.Conversely,positive co-skewness is desirable, as it means higher expected returnsduringvolatile periods. on Similarly,definethe co-kurtosisof the excess return asset i with the world portfolioas

Largepositive values are undesirable,as they mean that local returnsare low (high) when world marketreturnsare largely skewed to the left (right), thus increasingthe overallportfoliorisk. Table 4 reports estimates of these moments in the bull and bear states, as well as under steady-stateprobabilities.The latter gives a measure that is more directly comparableto the full-sample estimates listed in the final

914

International Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences

Ifable4 Estimates of co-skewness and co-kurtosis coefficients with world market portfolio Bear state Bull state Steady-state probabilities -0.128 3.408 0.016 3.303 -0.677 6.561 -0.339 5.230 -0.222 4.116 Data

United States Japan Pacificex-Japan United Kingdom Europeex-UK

Co-skewness Co-kurtosis Co-skewness Co-kurtosis Co-skewness Co-kurtosis Co-skewness Co-kurtosis Co-skewness Co-kurtosis

0.151 3.200 0.018 2.207 -0.161 4.522 -0.066 5.297 0.114 4.192

-0.127 3.434 -0.001 2.294 -0.567 5.782 -0.252 5.207 -0.167 4.095

-0.052 3.401 0.004 3.428 -0.535 6.704 -0.321 4.910 -0.227 4.113

This table reportssample co-skewness and co-kurtosiscoefficients for returnson individualmarketportfolios (i) versusthe world marketportfolio(W),

''W

covrtu^ia,]
{Vartjc/l^D'/^Var^l^]

Kiw .

CovW^3m

We also show the coefficients impliedby a two-stateregime-switchingmodel:

*Li = i,+, +vi,+1cov, [*;+!,*,+,] /= 1 + + y2s,+lCov, [*;+1,(^+1)3]+b\(+lr + ,;+1 [*;+1,(x?+l)2] yJ/+lCovf Var< Skew,[X&] + yj|+1Kt[#,] + *J+| r + r&x *Xi = ?+I + Vi/+1 Wi] + Yl,+1 5

The coefficientsare calculatedboth conditionalon the currentstate and understeady-stateprobabilities.

column. Comparing the values implied by the two-state model to the fullsample estimates, the model generally does a good job at matching the data. Interestingly, with the exception of Japan, US stocks have the lowest co-kurtosis and highest co-skewness coefficients in both the bear state and under steadystate probabilities. Moreover, Japanese stocks are unattractive due to their low mean returns over the sample period. As we shall see, these observations help explain why domestic stocks are more attractive to US investors with skewness and kurtosis preferences than in the mean-variance case. To address the effect of higher-order moments on the asset allocation, we next compute the optimal portfolio weights under mean- variance (m = 2) preferences: ,[/2W+r;0)] = KO,r(0)+ KltT(O)Eg[Wt+T] + K2T(e),[W,2+r], (26)

where Koj(Q)=v]rB[(l -0)"1 - 1 - 0], K1>r(e)=i;f e(l+0) > 0, and K2.r(G)= ^0i;^(1+e) < 0. We also consider optimal allocations under three-moment

915

The Reviewof Financial Studies!v21 n2 2008

preferences: + Et[U\Wt+T;e)] = K0,r(e)+ KXJ{)Et[Wt+T] K2,r(0), [W?+T] , +K3,r(9), [W?+T] (27)

= where now KO,r(0) v|"e[(l - 9)"1 - 1 - 10- 0(0 + 1)], Ki,r(9) = = vfe[l +0+ ^0(0+ 1)] > 0, K2,r(0)= -l0vf(1+0)(2 + 0) < 0, andK3,r(0) l0(0+l)vf(2+e) >0. Using steady-stateprobabilities,Table 5 shows that the allocation to US stocks as a portion of the overall equity portfolio is just above 50% under of both mean-variance and skewness preferences.The introduction two states on its own thus increases the allocation to US stocks from roughly 30% (as seen in Table 3) to 50%. This allocation rises furtherto 70% of the equity portfolio when we move to the case with skewness and kurtosispreferences. in Interestingly, the bear state the large increasein the allocationto US stocks due to introducing highermomentpreferencescomes fromthe skewness,while the kurtosisplays a similarrole in the bull state. The correlation, co-skewness, andco-kurtosisbetweenthe shortinterestrate and stock returnsalso affect asset allocations. At the 1-month horizon, the correlationbetween the risk-freerate and stock returnsis zero since the riskfree rate is known. Futureshort-termspot rates are stochastic,however.This mattersto buy-and-hold investorswith horizonsT > 2 monthswho effectively commit (1 - <o[i/) of their portfolio to roll over investmentsin T-bills T - 1 times at unknown future spot rates. We thereforecompute the co-skewness and co-kurtosisbetween the individualstock returnsand rolling 6-monthbond returns,assuming steady-stateprobabilitiesand setting the initial interestrate at its unconditional mean.The second-highestco-skewnesscoefficient(-0.06) was generated US stocksalong withthe secondlowest co-kurtosiscoefficient by (4.44). Only Japanesestocks turnout to be preferableto US stocks, although their conditional mean and variance propertiesmake them undesirableto a US investor.We conclude thatthe co-momentpropertiesof US stocks against rolling returnson shortUS T-billshelp us to explainthe high demandfor these stocks underthree-and four-moment preferences. 4. Robustness of Results To summarize resultsso far,we extendthe standard our model in two directions: First, by defining preferences over higher moments, such as skewness and kurtosis, and second, by allowing for the presence of bull and bear regimes trackingperiods with very differentmeans, variances,correlations,skewness, and kurtosis of stock returns.In this section, we consider the robustnessof our results with regard to alternativespecifications of investor preferences, estimationerrors,and dynamicportfoliochoice.

916

InternationalAsset Allocation underRegimeSwitching,Skew,and KurtosisPreferences

Table 5 Effects of preferences (m) on portfolio choice m US Japan Pacificex-Japan Bear state (71= 1) T= l m= 2 m= 3 m=4 0.661 0.841 0.746 0.778 0.721 0.653 0.594 0.550 0.623 0.143 0.079 0.063 0.032 0.088 0.153 0.000 0.163 0.130 0.036 0.016 0.048 0.016 0.000 0.056 0.058 0.013 0.091 0.036 0.063 0.048 0.000 0.162 0.056 0.125 0.000 0.095 0.175 0.029 0.083 0.348 0.075 0.039 0.441 0.373 0.370 0.369 0.320 0.282 0.309 0.201 0.226 UK Europeex-UK UST-bills

r=6 r=24
ro = 2 m= 3 m=4 m= 2 m= 3 m=4 T= \ m= 2 ro = 3 m=4 T= 6 m= 2 m= 3 m= 4 T = 24 m= 2 m= 3 m=4

0.000 0.200 0.117 Steady-state probabilities (tt = 0.33) 0.014 0.000 0.082 0.056 0.000 0.076 0.050 0.012 0.059 Bull state (71= 0) 0.000 0.113 0.000 0.000 0.169 0.051 0.013 0.247 0.071

0.536 0.521 0.685 0.500 0.532 0.646 0.525 0.519 0.635

0.014 0.366 0.110 0.000 0.286 0.127 0.000 0.210 0.094

0.435 0.000 0.123 0.444 0.013 0.101 0.413 0.012 0.141

0.310 0.289 0.269 0.282 0.231 0.209 0.198 0.190 0.149

r=6

m= 2 m= 3 m=4

0.262 0.131 0.598 0.189 0.215 0.632 0.427 0.422 0.640

0.299 0.446 0.098 0.232 0.398 0.092 0.012 0.241 0.093

0.020 0.002 0.054 0.042 0.000 0.057 0.049 0.012 0.058

0.181 0.408 0.022 0.147 0.366 0.023 0.049 0.301 0.023

0.238 0.013 0.228 0.389 0.022 0.195 0.463 0.024 0.186

0.000 0.000 0.078 0.048 0.069 0.125 0.180 0.171 0.140

m= 2 m= 3 m=4 T = 24 m= 2 m= 3 ro = 4

stocks as a function of the state probabilityfor three This table reportsthe optimal allocation to international choices of the orderof the preferencepolynomialm.m - 2 (mean-variance preferences),m = 3 (three-moment or or skewness preferences),and m = 4 (four-moment skewness and kurtosispreferences).T is the investment horizon. Stock holdings are reportedas a fractionof the total equity portfolio (and thus sum to 1), while the T-bill holdings are shown as a percentageof the total portfolio.

4.1 Preference specification We firstconsiderthe effect of changingthe coefficientof relativerisk aversion from 0 = 2 in the baseline scenarioto values of G= 5 (high) and 0=10 (very high). Ang andBekaert(2002) andDas andUppal(2004) foundthatchangesin risk aversionaffect theirconclusions on the importanceof eitherregime shifts resultsthat are availableupon request, or systemic (jump)risks. In unreported we foundthattherewas no monotonicrelationbetweenGandthe weight on US stocks, althoughthe allocationto US stocks tends to be greaterfor 0 = 10 than effect on the choice of T-bills versus for G= 2. Risk aversionhas a first-order

917

TheReviewof Financial Studies!v21 n2 2008

stocks but has far less of an effect on the compositionof the equity portfolio. choice Therefore,it does not seem thatour conclusions dependon a particular ofO. To makeourresultscomparable those reported the literature assume to in that we also comparedresultsunderfour-moment powerutility, preferencesto those under constant relative risk aversion. Differences between results computed underpower utility and four-momentpreferenceswere relativelyminor.22 In the bearstate,the allocationto US stocks was around2-4% lower underpower utility while conversely the allocation to UK stocks tended to be higher. In the more persistentbull state, allocations under the four-momentpreference specificationwere similarto those underconstantrelativerisk aversion. 4.2 Precision of portfolio weights Mean-variance portfolio weights are generally highly sensitive to the underestimates of mean returnsand covariances.Since such estimates often lying are impreciselyestimated,this means thatthe portfolio weights in turncan be Jones, 1999). As pointed out by Harveyet al. poorly determined(see Britten(2004), this could potentiallybe even more of a concernin a model with higher momentsdue to the difficultyof obtainingprecise estimatesof moments such as skewness and kurtosis.23 To addressthis concern, we compute standard errorbands for the portfolio weights underthe single-stateand two-statemodels using the fact that,in large estimates from a regime-switching samples, the distributionof the parameter model is

v/f(?-6)~N(0,Ve).

(28)

We set up the following simulationexperiment.In the <7th simulationwe drawa j0^a ~l vectorof parameters, , from N (8 , T Ve), whereVe is a consistentestimator 0 of Ve.Using this draw,8 , we solve for the associatedvectorof portfolioweights G>q repeatthis process Q times. Confidenceintervalsfor the optimalasset . We allocationd>,can then be derivedfromthe distribution &q, q = 1, 2, . . . , Q. of This approachis computationally intensive,as the asset allocationproblemin Equation(14) must be solved repeatedly,so we set the numberof simulations to Q = 2000. Results are reported in Table 6. Unsurprisingly,and consistent with the Jones (1999), the standard errorbandsarequitewide for the analysisin Brittenmodel. For example, at the 1-month horizon the 90% confidence single-state
22 A problemassociated with low-orderpolynomialutility functionalsis the difficultyof imposing restrictionson the derivatives(with respect to the moments of wealth) that apply globally. For example, nonsatiationcannot be imposed by restrictinga quadraticpolynomial to be monotonicallyincreasingand risk aversioncannot be imposedby restrictinga cubic polynomialto be globally concave (see Post and Levy, 2005; and Post, van Vliet, and Levy, 2007). It is thereforeimportant compareour resultsto those obtainedunderpower utility. to "Omega"in Cascon, Keating,and Shadwick (2003), which is used to capturesample information beyond point estimatesthroughthe cumulativedensity functionof returns.

23 See also the discussion of

918

International Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences

Table 6 Confidence bands for portfolio weights T= 1 month 5% Lower bound 95% Upper bound T = 6 months 5% Lower bound 95% Upper bound T = 24 months 5% Lower bound 95% Upper bound

Single-state model United States Japan Pacific (ex-Japan) United Kingdom Europe(ex-UK) UST-bills 0.024 0.088 0.000 0.000 0.000 0.418 0.379 0.469 0.217 0.314 0.320 0.861 0.000 0.075 0.000 0.000 0.000 0.423 0.371 0.451 0.240 0.331 0.307 0.897 0.000 0.137 0.002 0.000 0.000 0.475 0.350 0.404 0.259 0.414 0.272 0.937

Two-state model Bear regime (n = 1) 0.586 United States 0.037 Japan Pacific(ex-Japan) 0.033 United Kingdom 0.020 0.013 Europe(ex-UK) UST-bills 0.311 Steady-state probabilities (n = 033) 0.636 United States 0.090 Japan Pacific(ex-Japan) 0.070 0.004 United Kingdom 0.065 Europe(ex-UK) UST-bills 0.256 Bull regime (7i = 0) 0.484 United States 0.015 Japan 0.082 Pacific(ex-Japan) 0.000 United Kingdom 0.000 Europe(ex-UK) 0.090 UST-bills 0.834 0.095 0.065 0.081 0.220 0.381 0.727 0.127 0.093 0.015 0.225 0.289 0.744 0.145 0.118 0.047 0.407 0.226 0.438 0.045 0.000 0.000 0.000 0.201 0.606 0.107 0.052 0.029 0.064 0.209 0.578 0.073 0.079 0.000 0.000 0.124 0.964 0.270 0.114 0.150 0.382 0.440 0.665 0.134 0.067 0.059 0.207 0.216 0.674 0.106 0.098 0.027 0.308 0.142 0.416 0.029 0.006 0.012 0.000 0.138 0.504 0.060 0.038 0.037 0.114 0.047 0.518 0.060 0.069 0.000 0.000 0.050 0.845 0.196 0.137 0.184 0.201 0.343 0.691 0.108 0.067 0.083 0.267 0.194 0.736 0.113 0.109 0.036 0.317 0.230

The table reportssimulatedconfidence bands for optimal portfolio weights under either a two-state regimeswitching model or a single-state model. The weights are calculated assuming the 1-month US T-bill rate is set at its mean. The weight on the world marketportfolio is reallocatedto the five regional portfolios using theirrelativemarketcapitalizationsas of 2005: 12. T is the investmenthorizon.Stock holdings are reportedas a fractionof the total equity portfolio(and thus sum to 1), while the T-bill holdings are shown as a percentageof the total portfolio.

band for the weight on the US market in the equity portfolio goes from 2% to 38%- a width of 36%. The width of the confidence band is roughly similar at the 24-month horizon. In comparison, the confidence band for the US weight in the two-state model under steady-state probabilities only extends from 64% to 73%, a width of less than 10%. Even at longer investment horizons, the confidence bands remain quite narrow under the two-state model (e.g., from 50% to 69% under steady-state probabilities when T = 24 months). In fact, the standard error bands for the portfolio weights are generally narrower under the two-state model than under the single-state model. This suggests that the finding that a large part of the home bias can be explained by the US stock market portfolio's co-skewness and co-kurtosis properties in bull and bear states is fairly robust.

919

The Reviewof Financial Studies!v21 n2 2008

Intuitionfor these findings is as follows. First, the fact that the portfolio weights do not become less precise, even thoughwe accountfor skewness and kurtosis,is relatedto the way we compute these moments from a constrained two-stateasset pricingmodel. As can be seen from the time series in Figures2 and 3, these momentsare well behaved,withoutthe huge spikes and sampling variationstypically observedwhen such moments are estimateddirectlyfrom returnsdata using rolling or expandingdata windows. Second, the two-state model capturesmany propertiesof the returnsdata far betterthan the singlestate model and so reduces noise due to misspecification.Third, and related to this point, one effect of conditioning on states is to capture more of the return dynamics. This means that some of the parametersin the two-state model are more precisely estimatedthan in the single-statemodel. Again, this reduces the standarderrorbands on the portfolio weights underthe two-state model. An alternativeway to measurethe effect of parameter estimationerrorthat addressesits economic costs is to compute the investor'saverage(or directly expected) utility when the estimated parameters,as opposed to the true parameters,are used to guide the portfolio selection. To this end, Panel A of Table 7 reportsthe outcome of a Monte Carlo simulationwhere returnswere generatedfromthe two-statemodel in Table2. In these simulations,the parameter values were assumed to be unknownto the investorwho had to estimate these using a sample of the same length as the actual data before selecting the portfolio weights assuming either a 1-month or a 24-month investment horizon.For comparison,we also reportresultsfor alternatives,such as using the single-state model in Equation(8) or adoptingthe ICAPM weights (i.e., each region is purchasedin the proportion that it enters into the global market portfolio). Even afteraccountingfor the effect of parameter estimationerrors,the twostate model produces the highest certaintyequivalentreturnand the highest the averagewealth at both the 1-month and 24-monthhorizons. Furthermore, improvementsare meaningful in economic terms, suggesting an increase in the certaintyequivalentreturnof about 2% per annum.Since US stock holdings are considerablyhigherunderthe two-statemodel, the betterperformance of this model again indicates that parameter estimationerrordoes not diminish the ability of this model to explain home biases in US investors' equity holdings. 4.3 Out-of-sample portfolio selection Econometricmodels fitted to asset returnsmay produce good in-sample (or historical) fits and imply asset allocations that are quite different from the benchmarkICAPM portfolio. However,this is by no means a guaranteethat such models will lead to improvements "realtime"when used on futuredata. in This problemarises, for example, when the proposedmodel is misspecified.It could also be the resultof parameter estimationerror,as discussed above.

920

Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences International

Table 7 Out-of-sample portfolio performance T = 1 month Realized utility Mean SD CEV Annual ized returns Mean SD Mean

T = 24 months Realized utility SD CEV Annualized returns Mean SD

Two-stateRS VAR(l) ICAPM Equalweights Two-stateRS VAR(l) ICAPM Equalweights

-0.987 -0.992 -0.989 -0.991 -0.993 -0.995 -0.994 -0.994

0.021 0.017 0.011 0.015 0.029 0.022 0.039 0.031

16.42 9.89 14.22 11.95 8.22 6.18 7.44 7.63

Panel A: Simulated Data 16.77 7.28 -0.722 0.108 5.89 0.070 11.35 -0.799 14.03 4.16 -0.764 0.094 12.68 5.54 -0.802 0.066 Panel B: Actual Data 10.05 -0.849 8.73 8.72 7.62 -0.872 11.35 12.82 -0.850 10.03 10.74 -0.849 0.158 0.103 0.223 0.154

17.69 11.89 14.42 11.63 8.54 7.09 7.72 7.33

18.79 12.38 15.15 12.03 10.09 7.89 11.45 8.72

12.59 7.99 10.11 7.50 13.08 9.69 16.76 12.30

The table reports summary statistics for realized utility (using four-momentpreferences) and (annualized) portfolio returnsbased on the portfolio weights associated with the recursiveestimates of a two-state regimeswitching model, a single-state VAR(l) model, and a static ICAPM in which all international portfolios are to equity bought in proportion their weight in the world marketportfolio.Asset allocationsacross international = 1 month and T = 24 months. The weight on the marketsare calculated for two investmenthorizons, T worldmarketportfoliois reallocatedto the five regionalportfoliosusing theirrelativemarketcapitalization.SD denotes standarddeviations;CEV is the annualizedpercentagecertaintyequivalentof a given mean realized equity portfoliossuch thatthe utility."Equalweights"is a portfoliothatassigns equal weight to all international holdings in 1-month US T-bills match those from the two-state model. Panel A reportsportfolio performance from a simulationexperimentin which the data-generating process is the two-state regime-switchingmodel of Table2. Panel B uses actualMSCIreturnsdatafrom the sample period 1986:01-2005:12.

To address both concerns, we next explored how well the two-state model performs out-of-sample through the following recursive estimation and portfolio selection experiment. We first used data up to 1985:12 to estimate the parameters of the two-state model. Using these estimates, we computed the mean, variance, skewness, and kurtosis of returns and solved for the optimal portfolio weights at 1-month and 24-month horizons. This exercise was repeated the following month, using data up to 1986:1 to forecast returns and select the portfolio weights. Repeating this until the end of the sample (2005: 12) generated a sequence of realized returns from which realized utilities and certainty equivalent returns were computed.24 Since this experiment does not assume that the two-state model is the "true" model - realized returns are computed using actual data and not simulated returns- and since the sample (1986-2005) covered several bull and bear markets, this experiment provides an ideal way to test if the two-state model can add value over alternative approaches. Results are shown in Panel B in Table 7. Again, the two-state model came out ahead of the single-state model and ICAPM specifications in realized utility terms and for both investment horizons.25 For example, at the 1-month
24 In this once a year while the state probabilitieswere updatedeach experiment,we updatedall the parameters monthusing the Hamilton-Kimfilter(see Hamilton, 1990, for details). 25 An investment strategybased on the two-state model fails to produce the highest out-of-samplemean return, which is now associatedwith the ICAPM.However,the ICAPMportfolioweights also generatereturn volatilities

921

The Reviewof Financial Studies/ v21 n2 2008

horizon,the certaintyequivalencereturnof the two-statemodel was 2%higher than underthe single-statemodel while it exceeded that of the ICAPMby 80 basis points per annum.Results were very similarat the 24-monthhorizon. 4.4 Rebalancing To keep the analysis simple, so far we ignored the possibility of portfolio rebalancing.However,as noted in the literature,rebalancingopportunities give investorsincentives to exploit currentinformationmore aggressively.To explore the importanceof this point, we thereforeconsideredrebalancingusing two investmenthorizons (T = 6 and 24 months) and variousrebalancingfrequencies (cp= 1, 3, 6, 12 months). To save space, we simply summarizethe resultshere, while further details areavailableon request.Ouranalysis showed that rebalancingmattersmost when it occurs very frequently(i.e., when (pis small). Stock allocationsunderrebalancingare large and always exceed 60% of currentwealth. Startingfrom the bull state, the allocation under frequent rebalancing(cp= 1 and 3 months)differs significantlyfrom the buy-and-hold results as the investor attemptsto time the marketby shifting the portfolio towardPacific stocks and away from US and UK equities. However, startingfrom the bear state or assuming that the initial state is unknown(i.e., adoptingsteady-stateprobabilities),very frequentrebalancing (cp= 1 and 3 months) increasesthe allocationto US stocks for long horizons (7 = 24 months), while Japanesestocks also emerge as an attractiveinvestment. For all possible values of <p, implies an even greaterallocationto US this stocks thanunderthe buy-and-holdscenario.In fact, underfrequentrebalancing, a US investor with four-momentpreferencesand a long horizon should hold even more in US securities than underno rebalancing.For example, for T = 24, almost 100%of wealth goes into domestic securities,comprisingbetween 60% and 85% in stocks (only 8-12% of total wealth goes into foreign stocks). All told, regimeshiftscombinedwith preferences reflectaversionagainst that fat tails and negative skewness help explain the home bias under a range of assumptions about the rebalancingfrequency,especially when investors have little informationabout the current state (and thus adopt steady-state probabilities),which seems to be a plausibleassumption. 5. Conclusion The composition of US investors' equity portfolio into domestic and foreign stocks depends critically on how the distributionof global equity returnsis modeled and which preferencesinvestorsare assumed to have. Under meanvariancepreferencesand a single-state model for stock returns,we continue to find substantialgains to US investorsfrom international diversificationand
thatare 2-3% higherthanthe portfolioassociatedwith the two-statemodel. This helps explain why the two-state portfolioattainshigherrealizedutilities and certaintyequivalentreturns.

922

Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences International

thus confirmthe presence of a home bias puzzle. However,we arguethat the standardtwo-moment ICAPM has importantshortcomings since it ignores stock returns,and we presentempirical skewness and kurtosisin international evidence that such moments are associatedwith significantrisk premia.Once we accountfor US investors'dislike for negativeskewnessandfat-tailedreturn distributionsand incorporatethe strong evidence of persistentbull and bear states in global equity returns,we find a much larger allocation to domestic stocks. Intuition for this result comes from the attractive properties that US stocks have for an investorwho- besides being risk-averse preferspositively skewed (asymmetric)payoffs and dislikes fat tails (kurtosis).For example, US stocks have relativelyhigh co-skewness and low co-kurtosiswith respectto the of global marketportfolio.The performance US stocks certainlyworsens durstate. However,comparedto otherinternational the global "bear" markets, ing the US marketportfolio is relativelyless affected and offers betterinvestment when globalequitymarketsarehighly volatileor experiencelarge opportunities returns. negative Our empirical findings are consistent with and shed new light on recent from a theoreticalperspective.Mitton papersthatjustify underdiversification and Vorkink(2007) propose a model in which investors' skewness preferunderdiversified ences lead themto hold substantially portfoliosin equilibrium. shows thatseveralformsof rank-dependent Polkovnichenko (2005) preferences generatepreferencefor wealth skewness. For a range of plausibleparameteriof zations, this can lead to underdiversification the optimalportfolio. An interestingissue that goes beyond the analysis in the currentpaper is whetherour resultsextendto the home bias observedin investors'equity holdings in other countries. One may conjecture that because stock and bond marketsin the same economy are more likely to be "in phase"than are markets across national borders the finding that stock returnsin one country have attractiveco-moment propertieswith national short-termrates extends beyond our analysisfor the United States.This would contributeto explainthe evidence of a pervasivehome bias in stock holdings.26 international Appendix A. Conditional Moments and Estimation Procedure
Appendix A describes how we derive the conditional higher-order moments of stock returns and explains the econometric methodology used in estimating the asset pricing model in Equation (6).

A.I. Moments of returns


Letting yr+i = (xj+1 , x^+x, zj+1 )' be a vector of excess returns and predictor variables with inter)', cepts |is,+1 = (a^ | , . . . , a^+i , a^+i , Wzst+X we can collect the conditional moments of returns 26 We are

gratefulto an anonymousrefereefor pointingour attentionin this direction.

923

TheReviewof Financial Studies/ v2l n2 2008

and the world price of co-momentrisk in the matricesMs, and Ts,+, as follows: MS, =

1 (x?+tf (x?+])2 Cov[x,+1) |JF,]1 \ (\ rCov[x,+1,<,|jF,] Cov[x,+1, f,] I [ Var[^+1|^] K[xlx\T,} J <A Sk[x^,]
O(i'3 I),

vU+1y^, o-o" "v!.s,+1


T5,+1^ 0-0 Vl,s,+, Yl5,+1V2\+, .vU+1 y!s,+, y3\+1o--o_ ,

where 13 is a 3 x 1 vector of ones andJ is a matrixthatselects the co-momentsof excess returns: "I 1 10000001 1 1000000

J~

111000000 000111000 000000000

.000000000. We can then write the asset pricingmodel in Equation(6) more compactlyas yr+i = M-s,+I +Ms,vec(T5,+1) + B,,+1y, +i|,+I. (Al)

terms in state St+\ and also collects the coefficientsblS(( and Here, Bs,+1 capturesautoregressive that measure the impact of the lagged instrumentsz, on the risk premia. Finally, r\t+\ ~ b< j innovations. yV(8,fts/+1) is the vector of state-dependent To characterize momentsof returnson the world marketportfolioand its co-momentswith the local marketreturns,note thatmean returnscan be computedfrom
K K

y,+1 = [y,+, \Tt] = (jc;Pe*)iL* + (*;Pe*)A*y,,

(A2)

where n( is the vectorof stateprobabilities e* is a vectorof zeros with a one in the kthposition, and so (njPe*) is the ex ante probabilityof being in state k at time t + 1 given informationat time f , Tt, and jL*= |l* + Ms, vecCY.t). Because jl^ involves higher-order momentsof the world marketportfoliosuch as Ms, vec(T*) as well as higher-order co-momentsbetween individualportfolioreturnsand returnson the global marketportfolio,the (conditional)mean returnsE[yt+\ \ft] enterthe right-hand side of Equation (Al). For instance, computingCov[xr+i, jc^, \Tt] requiresknowledge of the first / elements of E[yt+\\Ft]- Below we explain the iterativeestimation procedureused to solve the associated nonlinearoptimizationproblem. The conditionalvariance,skewness, andkurtosisof returnson the worldmarketportfolio,jc^j , can now be computedas follows:

+(e';+1At -,+1)y,)] Var[^,|/;]= >;Pe*)[(|ir-e'/+1y,+l

924

Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences International

= + S*L+i|^<] I>iP*>[Oiy ~e/+iy/+i (e'/+1A* -a,+I)y/)3]


K

+ 3][>;Pe*)[(^ -e'l+1y,+, (e'#+1A* + -<*/+1)y,)

= xVarfn^l^,*]], a:
*[*,+i|*i] = EPe4K
if

-fi+i*'+| + (e'/+iA* -#+i)yi) ]


-e'/+1y,+i + (e'/+IA* -<*/+i)y,)
2

6^(ic;Pe^)[(^

xVar^,!^^^:]].

(A3)

Clearly the skewness and kurtosis of global equity returns are functions of the mean and variance parameters {1,*,..., /,*, A*, ft*}f=1, state probabilities, 11,, and mean VAR coefficients, *j = e'j Hk= 1(n't^ek )A* Hence, no new parameters are introduced to capture the higher moments of the return distribution. Similarly, the covariance between country returns x't+l and the world market return jc^j is
K

Cov[jc;+i, jc^,|^]

YS<**k)[{li,k -ejy/+i -h(^Ak-a,)y,) x (V* -e'/+iy<+i +(//+iA/ -#+i)y/)]


AT

+ (*;Pe*)Cov[y,+1, |5,+i= *]. nj^.,

(A4)

Given estimates of the parameters and state probabilities, Cov[jcJ+1, jc^J^, St] can easily be calculated. Finally, the co-skewness and co-kurtosis between local market returns and the world market return is

Cov[jc;+1, (<i)2|^/]

-e;y,+i ^CejA*-/)y,) Y^{*',**k)[{\Li.k x (plf -e'/+1y,+i +(ei+,A* -a/+i)yr) ]

+ (*;P^)I /.* - e{y,+i + (ejA*- a,)yr)

x Var^,1^, = *]] + 2E(";Pe*) |s,+i


x [(l^ - e'/+iy+l (e'/+l - ;+,)y,) + A*

^-

xCov[ti;+I,ri^l|s/+, =*]],

(A5)

925

TheReviewof Financial Studies/ v21 n2 2008

and
3 K

(*,) Cov[*;+1,

|^]

+ ^(7c;Pe^)[(pLl,Jte;y,+1 (ejA^ a^y,) x -a/+i)y,j (|lf -e'/+1y,+i +(e'/+1Ajt


K

+ 3 (K;Pe*)[(|I/,* - e,'y,+i+ (ejA*- a,)y,)


*=i

+ (pLJ^e'/+1y,+i (e'/+1A*a/+i)y,j
K

x Varf^, = *]]+ 3>;Pfe*) |s,+,


x [(plf - e'/+1y,+i (c'h.^ + a/+i)yr) (A6)

xCovfrij+pTi^lSf+i =*]].

Terms such as (,,* - eJy/+i)(pL^ - C/+iyr+i) show the deviations of the state-specific mean from the overall mean and do not arise in single-state models.

A.2. Estimation
to Definingr\S(+las a vectorof residualsin state St+\ , the contribution the log-likelihoodfunction conditionalon being in state St+\ at time t + 1 is given (up to a constant)by ln/?(y,+1|.F,, S,+\\\) (x-\ ln|ftsl+,l ^nif+1s,Vi 1|s'+i (A7)

where \ = {<|>5, variance (ft), and transitionprobability(P) ft5,P}^L, collects the mean (<|>), of parameters the model in Equation(Al). The expected value of the log-likelihoodemployed by \l+l in the / -I-1 iterationto satisfy the EM algorithmis maximizedby choosing the parameters (Hamilton, 1990, p. 51):

where [p(St+\ |y2, y3, . . . , yr; X')}^ ]=1 are the smoothed state probabilitiesfor each of the K states. Lettingy = [y^y'3 y^]' and i| a [r\\ r\'2... i\'K]',it is useful to rewritethe log-likelihood as 1 ^(yi,...,yr|8)cx-;rX>
5=1 ^ 5=1

T ^\ ^piSt+nX1)
r=2 7 /=2

- -1

K ^'s&s
5=1

a,"1)^

= --]^ln|aJ|J]p(5r+,;TL/)--i|'W-Ii|,
where "Zi "I Z2 . , r [eJeJ-Oy'il^Iiv [e;.e;.(g)y2](g)IN "

Z=

and Z,=

926

Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences International

w1-

"z^a-1 ... : .

o :

% = diag{p(52 = i, \l), p(si = i\ \l), . . . , p(sT = i, X')}. The EM updatingequationfor the transitionprobabilitiesis based on the smoothedstate probabilities and can be found in Equation(4.1) of Hamilton(1990, p. 51). Filteredstate probabilitiesare The first-order conditionsfor the mean and varianceparameters, and calculatedas a by-product. $ ft, are

>hy>--W-Z-
and

(A9)

= O 5 = 1,2,...,*,

(A10)

where e5 = [(y2 - ZJ2=/$)' (y3 - ZS3=ii)f . . . (yr - Z5r=/$)']' are the residuals in state s and W~ *is a functionof {65 }JLt . Equation(A9) impliesthat<j> is a GLS estimatoronce observations are replacedby their smoothedprobability-weighted counterparts: $/+1 =(Z/W-1Z)-1Z'W-1(ik(g)y). Similarly,Equation(A9) implies a covarianceestimator: (All)

ft, =

8*r^' . E,-ip(S,+i;V)

(A12)

$/+1 and {fi'+1}^Li must be solved for jointly, since bs enters the expression for the covariance matrix and also depends on $ , while the regime-dependentcovariance matrices {S^+1}^Lj enter Equation(All) via W~l. Hence, within each step of the EM algorithm,the equationpair (Al 1HA12) is iteratedupon until convergenceof the estimates $/+1 and {itls+l}f=1. Finally,notice thatEquation(A9) defines fj from %+i syl+i-jL5f+I -B5,+1y, = yr+J - |is,+1-M5/vec(Ts,+1) - B5,+1y,, so that E[yt+i \Ft, St] entersMs, vec(Y/), while MS(vec(T/) also affects E[yt+\ \Tt, St], creating a nonlinearsystem of simultaneousequations. For instance, computing Cov[x,+i,;t/v^1|.7r/, St] requiresknowledgeof the first / elements of E[yt+\ \Tt, St]. To make estimationpossible, within the (/ 4- l)th step of the EM algorithm,we use an iterativescheme by which Ms, vec(T/) is firstestimatedusing the values in E[yt+\ \Ft, St] fromthe previousoptimizationstep, [y,+i \Tt , St; $']. New values of E\yt+\ \Tt, St; $/+1] arethencomputedusing estimatesof Ms, vec(T/) thatemploy E[y,+i \Tt, St\ ft]- We then proceediterativelyuntil convergence.

B. Proof of Proposition 1
our AppendixB firstuses a simpleexampleto introduce solutiontechnique,thenderivesProposition terms in the returnprocess. 1 and shows how to extend the resultsto include autoregressive

927

The Reviewof Financial Studies!v21 n2 2008

To gain intuition,we first study the asset allocationproblemunderthe simplifying assumptionof a single risky asset (/ = 1), a risk-freeasset paying a constantreturn/*/, and a regime-switching process with two states: ~ xt+\ = M-s,+1 +as,+le,+i, e,+i N(0, 1), = k\St = k) = pkk, k = 1, 2. Pr(S,+1

B.I. Two-state example with a single risky asset

(Bl)

This specificationis consistentwith the ICAPManalysisin Section 1 since the conditionalmoment information from Equation(6) can be folded into {|xs,+1 os,+l } as describedin Section 1. , With a single risky asset (stocks) and initial wealth set at unity,the wealth process is Wt+T= {(1 - co,)exp(7>/) + go,exp(/?,+r)}, (B2)

where Rt+r is the continuouslycompoundedstock returnover the T periodsand co, is the stock holding. For a given value of oor,the only unknowncomponentin Equation(B2) is the cumulated return,exp(/?,+7). Underthe assumptionof two states, K = 2, the wthnoncentralmomentof the cumulatedreturnsis given by + rt+T))n\Tt] A*,+V= [(exp(r,+1 +
2

= J2
sl+T=i

E[(txp(rt+l+---+rt+T))n\St+T,Jrt]PT(St+T\Jrt)

= <lr+<lr>

<B3)

where rt = xt + r f . Using properties of the moment generating function of a log-normal random variable, each of these conditional moments Af^j (k = 1,2) satisfies the recursions M{^+t = [exp(/i(r,+1 + + rt+T-\))\St+T]

x E[exp(nrt+T)\St+T^t]PT(St+T\^t)

x exp I n\ik + - erf I, (=1,2), where we used the notation - k for the converse of state &, i.e., - k = 2 when k = 1 and vice versa. In more compact notation, we have

<li <li
where

=^<,")<)+p<1")<). =4")<) + p2")<>.

(B4)

^ = pn exp(n\n + yo? j,

= (n\n fi (1 - p22)exp + ya? V f P^ = p22expn^2+ yai I

^ = (1 " Pn)expU^2 + ya^ j,

Equation(B4) can be reducedto a set of second-orderdifferenceequations:

<+2 = (W + i^K+i

+ {liV? ~ tf'a?0)*,*" (i = 1,2).

(B5)

928

International Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences

momentsin the vector$]"+T = (MJ"+T Collectingthe two regime-dependent Equation MJ"^_T_l)f, (B5) can be writtenin companionform:

^ =p |p2*2Pi 2I J^-.-a00*^-.o
, x r fc(") _i_ ft(") fc(")o(") ft(n)fc(n) 1 , s

The elements of A(/l) only depend on the mean and varianceparametersof the two states (|x i , a}, |i2, aj) and the state transitions,(pi i , P22).Substitutingbackward,we get the ith conditional moment:

"it+T

-\A

Vir

Applying similarprinciplesat T = 1, 2 and letting n\t = Pr(S, = l|^v) the initial conditions used in determiningthe nth momentof cumulatedreturnsare as follows:

- P22))exp (n\n + Af{;|, = (icirpn + (1 iti/)(l ya?J

- iii/Kl M\%2 = pn (wi/pn +(1 P22exp(2njii +2a?) + (1 ~P22)(Wi,(l -Pll) + (1 -K\)P22)

x expf /i(|ii + M.2) +

+ y (cr? ai)J .

- pn) + (1 - 7ti)p22)exp f /1M.2 + . Af^i, = (iri/(l y ^2) M2tl2 = P22(wird P11)+ (1 Hi)/722)exp(2ni2 + n2oty + (1 - Pii)(7ti/Pii +(1 -*i/)U x exp Mjii + V2) + -P22)) (B6)

y (a?

+ ajj . )j

Finally,using Equation(B3) we get an equationfor the nth momentof the cumulatedreturn:

<r = <!r + <lr = i*ftr+^*2lr


= ei(A<)r<)+e^(A<)r<';), (B7)

where e, is a 2 x 1 vector of zeros except for unity in the 1th place. Having obtained the moments of the cumulatedreturnprocess, it is simple to compute the by expected utility for any mth orderpolynomial representation using Equation(13) in the main

929

TheReviewof Financial Studies/ v21 n2 2008

text and Equation(B2):

= f> tt[Um(Wt+Ti9)] ^(-ir^r

( y) <[<r]

=E^E(-"-^r(.)E(-)0)'<r
n=0 ;=0 V 7 7 =0 X 7

xai-co^expCrr/

))>"'.

(B8)

this The first-order condition is obtainedby differentiating equationwith respectto a>,

=0

y=0

X</ ^ *= 1 V

xexp(O-/)rr/)0-ycor)

= 0.

The solution takes the form of the roots of an m - 1 order polynomial in a),, which are easily obtained.The optimal solution for 00, correspondsto the root for which Equation(B8) has the highest value.

B.2. General results


To derive the /2thmoment of the cumulatedreturnon the risky asset holdings in the generalcase with multiplerisky assets (/) and states (K), notice that

= , ,[(; exp(RJ+r))"]
|_n,=l

m= \

Mm. 112 ;)(^' x x J')

xexpfcr/J xexp/f>,'+Ij,

(B9)

wherethe powers0 < n{ < n (i = 1, . . . , /) satisfy the summing-upconstraint5Z/=i n* ~ n* anc* the coefficients X are given by

\(n\,n2,..

.,ni)=

n\\ri2\

-^ -

-. -nj\

represent the return on asset 1 in period t + x, r/+T = jcJ+t -I-r*+T. The sum in r/+T (/ = 1, ...,/) /+"~1 terms and ) requires solving for moments of the form Equation (B9) involves (

930

International Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences

\ "i T=l /

/ t \X=1

\nn / J

= , exp(^^^r;+t)l. \, = 1
X=l

(BIO)
/J

Equation(BIO) can be decomposedas follows:


K

M^)r(i^2,...^/)
where for A:= 1, ..., K,

X!Af^)+r(/i1,2,..
*=i

.,/),

(Bll)

exp M(k%T(nl,n2,...,nl)=Et ]K;+T =J (*, j |s,+r


x Pr(S,+r = *) Each of these termssatisfies the recursions
K

M[%T(ni,n2

"/) = ]T M<j!t+T-i^ ' n2

"/)/

x exp I J2 nirl+TI \st+T=k,ft\


K

pjk

= J^PjkMfr+T_l(nl,n2,...,nl)

\l = l

/=1 W=l

where (1a, is the mean returnof asset / in state k (inclusive of risk premiarelatedto covariance, = co-skewness, and co-kurtosis)and cr*,/a e'^k^u is the covariancebetween r/,+7 and rut+r in statek = 1, 2, . . . , K. This is a generalizationof the result in Equation(B4). Finally, using Equations (B9) and (BIO), we get an expression for the nth moment of the cumulatedreturn:
n n

E, [(<; (R,J+r))"]= exp


n\=Q

2>(/>i, 2
n/=0

/)
/). (B13)

x(a)?1 x-x(o;')Jl/>i

generalizationof Equation(B8): Expectedutility can now be evaluatedin a straightforward

= >(-i)"-'vrync,,[wV+r] E,ium(Wl+T;)) n=o ;=o

--t(-ir^(;)i:(0
/!=0 y=0 XJ ' i=0 X 7

x, [ (o>; (R?+r))'jai-^i^exp^r^))^"'. exp

931

The Reviewof Financial Studies/ v21 n2 2008

conditionthattakes the form of an InsertingEquation(B 13) into this expressiongives a first-order (m - l)th orderpolynomialin the portfolioweights. termsis straightforward. keep the notation To Generalizingthe resultsto includeautoregressive simple, suppose k = 2. Using Equation(15) in the main text, the /ith noncentralmomentsatisfies the recursions
/ P

= Mk"]+T Mk"]+T-i(n)PUexP I n&k+ n^btfEtin+T-j) J=l \ + Af^),+r_,(/2)(l -/>_*_*)

n2

ye*

I /

x exp I nfrk+ n^bj^Etirt+T-j] + - v\ I >=1 \ / or

n2 \

Mu+i= i<1")wi",)+p<;)M<",),
where now
/ P

$\n)= px\ exp I n\ix + nJ2bJAEdn+T-j] + yd? I , J=i \ I


( $\n)= (1 -/>22)exp I nvn +n^bjAEt[rt+T-j]+ P n2 \

n2

V
(

>=i
P

-o]
n2

/
\

I,

" = %2n) (1 PiOexp

H2 + ^^[rr+r-,-]

>=i

+ _a2
\

( = V>2} P22exp I n\L2+ n^2bj,2Et[rt+T-j] + ya^ I .

n2

Subjectto these changes, the earliermethodscan be used with the only differencethattermssuch as exp(/zpL* ^cr*) have to be replacedby +

+ exp I n\Lk nJ2bj,kEt[n+T-j] + y^

n2

The term YlP=\ bj,kEt[n+T-j] may be decomposedin the following way:

j=\

p p + 22bjjEt[rt+T-j] = Z[j>T) \Z[j>T)bj,kn+T-j I[j<T\bj,kEt[rt+T-j]y 2^


j=i

whereJ is an indicatorfunctionand Et[rt+\], . . . , Et[rt+T-\] can be evaluatedrecursively:

' ' ( \ ( E,[r,+i] = Jii, I (Ii +^fty,ir,_y J +(1 - Jti,) I fo + ^bjir,-]

J,

932

Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences International

E,[rt+2] = JiJPe, pLi+ >,,,,[r,+1] J=l \

+ (\ - npei) U2 + Y,bjjEdrt+\)\ ,

I ! I ,[r,+7-i] = n'tPT-lei pL+ J^bj,\Et[rt+T-i\


+ (1 - nJP7"^!) I pL2 ][>,,2r[r,+r-2] + I.

References PortfolioChoice and Corporation Finance:A Synthesis.Journal Adler, M., and B. Dumas. 1983. International of Finance 38:925-84. Asset Allocation with Regime Shifts. Reviewof Financial Studies Ang, A., and G. Bekaert.2002. International 15:1137-87. Ang, A., and J. Chen. 2002. AsymmetricCorrelationsof Equity Portfolios.Journal of Financial Economics 63:443-94. Diversification Puzzle Is WorseThanYouThink.American Baxter,M, and U. Jermann.1997. The International EconomicReview87: 170-80. Journalof Finance 50:403-44. Bekaert,G., and C. Harvey.1995. Time-VaryingWorldMarketIntegration. Black, F. 1990. Equilibrium ExchangeRate Hedging.Journalof Finance 45:899-908. Brandt,M. 1999. EstimatingPortfolio and ConsumptionChoice: A Conditional Euler EquationsApproach. Journalof Finance 54: 1609-45. PortfolioFlows. Journal of Finance 52:1851-80. Brennan,M, and H. Cao. 1997. International Brennan,M., E. Schwartz,and R. Lagnado. 1997. StrategicAsset Allocation.Journal of Economic Dynamics and Control21:1377-403. Variance EfficientPortfolioWeights.Journal BrittenJones, M. 1999. The SamplingErrorin Estimatesof Meanof Finance 54:655-71. Brown, D., and M. Gibbons. 1985. A Simple EconometricApproachfor Utility-Based Asset Pricing Models. Journalof Finance 40:359-8 1. Campbell,J., and L. Viceira. 1999. Consumptionand Portfolio Decisions When Expected ReturnsAre Time Varying.QuarterlyJournalof Economics 114:433-95. Bonds?AmericanEconomicReview91 :99-127. Campbell,J., andL. Viceira.2001. Who ShouldBuy Long-Term Cascon, A., C. Keating,and W. Shadwick. 2003. The Omega Function.UnpublishedWorkingPaper,Finance DevelopmentCentre,London. Asset Pricing underSegmentationand PPP Deviations.Journal Chaieb, I., and V. Errunza.2007. International of Financial Economics86:543-78. Cooper,I., and E. Kaplanis.1994. Home Bias in EquityPortfolios,InflationHedging,and International Capital MarketEquilibrium. Reviewof Financial Studies7:45-60.

933

TheReviewof Financial Studies!v21 n2 2008

and Coval,JM T. Moskowitz. 1999. Home Bias at Home:Local EquityPreferencein Domestic Portfolios.Journal of Finance 54:2045-74. Dahlquist, M., L. Pinkowitz, R. Stulz, and R. Williamson. 2003. CorporateGovernanceand the Home Bias. Journalof Financial and Quantitative Analysis 38:87-1 10. PortfolioChoice.Journalof Finance 59:2809-34. Das, S., and R. Uppal. 2004. Systemic Risk and International Davies, R. 1977. Hypothesis Testing When a Nuisance ParameterIs Present Only Under the Alternative. Biometrika64:247-54. 2003. A Monte CarloMethodfor OptimalPortfolios.Journal of Detemple,J., R. Garcia,and M. Rindisbacher. Finance 58:401-46. De Santis,G., and B. Gerard.1997. International Asset Pricingand PortfolioDiversificationwith Time-Varying Risk. Journalof Finance 52:1881-912. R. Dittmar, 2002. NonlinearPricingKernels,KurtosisPreference,andEvidencefromthe CrossSectionof Equity Returns.Journal of Finance 57:369-403. Engel, C, and J. Hamilton. 1990. Long Swings in the Dollar: Are They in the Data and Do MarketsKnow It? AmericanEconomicReview80:689-713. Ferson,W. 1990. Are the LatentVariablesin Time-Vary Expected ReturnsCompensationfor Consumption ing Risk?Journal of Finance 45:397-429. of Reviewof Financial Ferson,W, andC. Harvey.1993.The Risk andPredictability International EquityReturns. Studies6:527-66. and AmericanEconomic French,K., andJ. Poterba.1991. InvestorDiversification International EquityMarkets. Review81:222-26. of Gray,S. 1996. Modelingthe ConditionalDistribution InterestRates as Regime-SwitchingProcess.Journalof Financial Economics42:27-62. in GenerationsModel with Learning. Review Guidolin,M. 2005. Home Bias and High Turnover an Overlapping of InternationalEconomics 13:725-56. 2006. An Econometric Modelof NonlinearDynamicsin theJointDistribution Guidolin,M., andA. Timmermann. of Stock and Bond Returns.Journalof AppliedEconometrics21:1-22. Hamilton,J. 1990. Analysis of Time Series Subjectto Changesin Regime.Journalof Econometrics45:39-70. Harvey,C. 1989. Time-Vary ConditionalCovariancesin Tests of Asset PricingModels.Journalof Financial ing Economics24:289-317. Harvey,C. 1991. The WorldPrice of CovarianceRisk. Journal of Finance 46: 111-57. Harvey,C, J. Liechty,M. Liechty,and P. Mullen 2004. PortfolioSelection with HigherMoments.Mimeo, Duke University. Harvey,C, andA. Siddique.2000. ConditionalSkewnessin Asset PricingTests.Journalof Finance 55: 1263-95. of Karolyi,G., and R. Stulz. 1999. Why Do MarketsMove Together?An Investigation U.S.-JapanStock Return Comovements.Journalof Finance 54:95 1-86. Karolyi,G., and R. Stulz. 2003. Are FinancialAssets PricedLocally or Globally?in Handbookof the Economics the of Finance, vol. IB {FinancialMarketsand Asset Pricing). Elsevier,Amsterdam, Netherlands, 975-1020. pp. Transmissionof U.S. MonetaryPolicy Shocks: Evidence from VAR's.Journal of Kim, S. 2001. International MonetaryEconomics48:339-72. Kim, T, and E. Omberg. 1996. Dynamic Nonmyopic Portfolio Behavior.Review of Financial Studies 9:14161.

934

International Asset Allocation underRegimeSwitching,Skew,and KurtosisPreferences

1976. SkewnessPreferenceandthe Valuation Risky Assets. Journalof Finance of Kraus,A., andR. Litzenberger. 31:1085-100. Lewis, K. 1999. Trying to Explain Home Bias in Equities and Consumption.Journal of Economic Literature 37:571-608. in EquityReturnsConstant:1960-1990? Journal Longin,F., and B. Solnik. 1995. Is the Correlation International Moneyand Finance 14:3-26. of International of Structure International EquityMarketsDuringExtremelyVolatile Longin,F., andB. Solnik. 2001. Correlation Periods.Journalof Finance 56:649-76. Case. Reviewof Economicsand Statistics Merton,R. 1969. Lifetime PortfolioSelection:The Continuous-Time 51:247-57. Underdiversification the Preferencefor Skewness. Reviewof and Mitton,T, and K. Vorkink.2007. Equilibrium Financial Studies20: 1256-88. in Morse,A., andS. Shive. 2003. Patriotism YourPortfolio.Unpublished WorkingPaper,Universityof Michigan. Obstfeld,M., and K. Rogoff. 1995. ExchangeRate DynamicsRedux.Journalof Political Economy 103:624-60. 2000. Firm Size and Cyclical Variationsin Stock Returns.Journal of Perez-Quiros,G., and A. Timmermann. Finance 55:1229-62. A Preferences. V. Review Polkovnichenko, 2005. HouseholdPortfolioDiversification: Case for Rank-Dependent of Financial Studies 18:1467-502. Post, T., and H. Levy. 2005. Does Risk Seeking Drive Stock Prices?Reviewof Financial Studies 18:925-53. Post, T, H. Levy, and P. van Vliet. 2007. Risk Aversion and Skewness Preference.Journal of Banking and Finance, forthcoming. Review of Economics Samuelson,P. 1969. Lifetime Portfolio Selection by Dynamic Stochastic Programming. and Statistics5 1:239-46. Scott, R., and P. Horvath.1980. On the Directionof Preferencefor Momentsof Higher OrderThan Variance. Journalof Finance 35:915-19. PortfolioHoldings.Econometrica69:1467-89. Model of International Serrat,A. 2001. A DynamicEquilibrium Smith, D. 2007. ConditionalCoskewnessand Asset Pricing.Journalof EmpiricalFinance 14:91-1 19. Investment. Journalof Finance 36:923-34. Stulz, R. 1981. On Effects of Barriersto International and Tesar,L., and I. Werner.1994. International EquityTransactions U.S. PortfolioChoice, in J. Franked(ed.), The Internationalization EquityMarkets,Universityof Chicago Press, Chicago, IL, USA, pp. 185-216. of Journal of InternationalMoney and Finance Tesar,L., and I. Werner.1995. Home Bias and High Turnover. 14:467-92. of Portfolios.NBER Working Thomas,C, F. Warnock,and J. Wongswan.2006. The Performance International PaperNo. 12346. A. Timmermann, 2000. Momentsof MarkovSwitching Models.Journal of Econometrics96:75-1 11. Deviation Analysis, Skewness Preference,and the Tsiang, S. C. 1972. The Rationale of the Mean-Standard Demandfor Money.AmericanEconomicReview62:75-1 11. and Journalof Finance58:2465-86. Uppal,R., andT. Wang.2003. Model Misspecification Underdiversification. Vanden,J. 2006. OptionCoskewnessand CapitalAsset Pricing.Reviewof Financial Studies 19:1280-320. Whitelaw,R. 2001. Stock MarketRisk and Return:An EquilibriumApproach.Review of Financial Studies 13:521^8.

935

You might also like