Asset Pricing with Real Estate and Skewed Returns: Work
Transcription
Asset Pricing with Real Estate and Skewed Returns: Work
1 Asset Pricing with Real Estate and Skewed Returns: Work in Progress Benoît Carmichael : Université Laval Alain Coën : Université du Québec à Montréal (UQAM) © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 2 Stylized Facts ² Since Samuelson (1970) and Kraus and Litzenberger (1976), skewness and coskewness are seen as potentially important determinants of equilibrium asset returns. ² Problem: It is difficult to build a model linking analytically skewness premiums to deep structural parameters governing preference and the distributions of stocks returns. ² Wealth should be decomposed into: financial and real estate assets. ² A CAPM with real estate including idiosyncratic coskewness. © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 3 Stylized Facts ² Solution: Harvey and Siddique (2000, among others) : explicit expressions for the risk premiums in skewed environment. ² Advantage: their premiums are distribution free (no specific distributional assumption). ² Drawbacks: difficult to isolate the influence of skewness on asset returns. ² Second and third moments are tied by a common set of structural parameters. ² Risk premiums are reduced form expressions of deeper structural parameters determining returns. ² No specific distribution: empirical analysis cannot be made with fully efficient estimation methods. A need for a more powerful method: Azzalini (1985)’s Skew-Normal distribution © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 4 Our contribution ² A natural departure point: Azzalini (1985)’s SkewNormal (SN) distribution in a financial economics framework. ² Main motivations: -1- The standard normal distribution is a special case of the SN distribution. -2- The skewness and coskewness of joint stochastic skewnormal variates have explicit expressions. ² Analytic expressions for the skewness premiums: ² Coskewness and idiosyncratic coskewness. -3- The financial returns and real estate returns, as joint skewnormal variates have explicit expressions for their skewness and coskewness. © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 5 Theoretical Framework: The Skew Normal Distribution ² The Skew-Normal distribution: Azzalini(1985) The density function for a standardized univariate skewed z variate. The skew-normal (SN()) distribution shares many formal properties with the normal distribution. © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 6 Theoretical Framework: The Skew Normal Distribution © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 7 Theoretical Framework: The Skew Normal Distribution ² skew-normal variable y location Scale parameter The three moments of y © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Reliance on the deeper structural parameters, and Montreal May 2012 8 Theoretical Framework: The Skew Normal Distribution ² The skew-normal distribution in multivariate environments: Azzalini and Dalla Valla (1996) The density The first three moments Reliance on the deeper structural parameters, © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 9 Theoretical Framework: The Skew Normal Distribution ² Explicitit expression of the coskewness Analytic solutions should help to elucidate intricate role of skewness and coskewness in asset pricing. © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 10 Theoretical Framework: A Model of Asset Returns ² Here we develop the theoretical restrictions imposed by the skewnormal distribution on asset returns. ² The starting point is the Euler equation of optimal portfolio diversification. ² Harvey and Siddique assume that the marginal rate of substitution is a quadratic function of the market return. ² We adopt Brown and Gibbons (1985)’s assumption: the marginal rate of substitution is a power function of the gross return on wealth. © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 11 Theoretical Framework: A Model of Asset Returns ² Brown and Gibbons (1985)’s assumption is that optimal consumption is proportional to wealth. ² Decomposition of wealth between financial wealth and real estate wealth Mt +1 = b 1- q - d q RF ,t +1 RE ,t +1 ² Euler equation: 1 = b.Et © 2012, Carmichael, Coën 1- q - d q RF ,t +1 RE ,t +1 ACFAS_ Real_Estate_Finance ×Ri ,t +1 Montreal May 2012 12 Theoretical Framework: A Model of Asset Returns 1 = b .E t R qF ,t + 1 - d 1- q R E ,t + 1 ×R i ,t + 1 ijkHL B HL F Euler equation: xi- d 1= b e .2 F 2 2 2 1- q xE- d q xF+xi+ d wE - d2 q wE2 + 1 d2 q2 wE2 +d2 q r FE wE wF- d2 q2 r FE wE wF+ 1 d2 q2 wF2 - d r iE wE wi+d q r iE wE wi- d q r iF wF wi+ wi 2 2 2 2 mzi wi - dqmzF wF - d 1 - q mzEwE y N.B.: for aggregated wealth: © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 y z { 13 Theoretical Framework: A Model of Asset Returns ² We obtain the main equation , the key expression determining expected log returns in a skew-normal environment: @ DJ HHLH L L BBHL F N E ri = mzi wi + 1 2u- d2 wE2 + 2d2 qwE2 - d2 q2 wE2 + 2d ErE - mzEwE - 2dq ErE- mzEwE 2 - 2d2 qr FEwEwF + 2d2 q2 r FEwEwF - d2 q2 wF2 + 2dq ErF - mzF wF + 2dr iEwEwi - 2dqr iEwEwi + 2dqr iF wF wi - wi2 - 2Log 2F d - 1+q mzE wE- d q mzF wF+mzi wi y ² It holds for all returns, including the safe return: HL rf = 1 2u- d2wE2+2d2qwE2- d2q2wE2+2d ErE- mzEwE - 2dq ErE- mzEwE - 2d2qr FEwEwF 2 +2d2q2r FEwEwF- d2q2wF2+ 2dq ErF- mzFwF - 2Log 2F © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance d -1+q mzEwE-dqmzFwF y Montreal May 2012 14 Theoretical Framework: A Model of Asset Returns ² To highlight the contribution of skewness we use a third order Taylor expansion Approximate real interest rate: for aggregated wealth © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 15 Theoretical Framework: A Model of Asset Returns ² Equations of the expected log return for the risky asset and the risk-free asset together with (15) ( lead the following expression for the aggregated wealth coskewness © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Idiosync.coskewness Montreal May 2012 16 Theoretical Framework: A Model of Asset Returns The idiosyncratic coskewness is a «volatility» risk. Systemic variance risk skewness risk The idiosyncratic coskewness of asset i is one of the Elements determining the systemic variance of asset i © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 17 Empirical Framework: The Log likelihood function ² We report maximum likelihood estimates of constrained and unconstrained version of the model. ² The empirical specification is based on the asset pricing equation. ² Expected and actual skew-normal log returns are linked by the following relation: © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 18 Empirical Framework: The Log likelihood function ² We complete the empirical specification with the assumption that the expected market return is linear function of forecasting variables known at time t. We use the same assumption for the FTSE NAREIT Real Estate index ² With skew-normal error, the empirical market log return equation becomes (we obtain the same decompositon for NAREIT US index) : © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 19 Empirical Framework: The Log likelihood function ² The log likelihood function of a sample of T observations is: ² (25) differs from the log likelihood of the normal distribution because of the third term. ² If asset returns are symmetric, ’is a zero vector and (25) reduces to the Gaussian log likelihood. © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 20 Empirical Framework: The Data ² Estimates are based on monthly portfolio returns from Center for Research in Security Prices (CRSP) dataset. ² The returns are the monthly returns of NYSE/AMEX decile portfolios ranked by market equity: CRSP Stock File Indices. ² The market return is proxied by the CRSP ValueWeighted Index of the S&P 500 Universe. ² The data cover the period from1971:12 to 2007:12. ² Real returns are obtained by deflating nominal returns with the rate of inflation, measured by the CRSP CPI index. ² For real estate, we use as a proxy FTSE NAREIT US Real Estate index. © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 21 Empirical Framework: The Data ² The forecasting variables are: The inflation rate; The log market return; The log return on human capital; The log price/earnings ratio The spread between the log gross returns on long (10 years) and short (30 days) bonds. ² The log return of the FTSE NAREIT US index. ² These variables are all lagged one period in the forecasting equations. ² ² ² ² ² © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 22 Empirical Framework: The Data ² Sample summary statistics r1 r2 r3 r4 r5 r6 r7 r8 r9 r10 r. market r. real est. Mean S.E. Min Max Skewness Kurtosis 0,01535186 0,07615109 ‐0,306655 0,534829 1,51198746 9,38540382 0,01264907 0,06385139 ‐0,300949 0,461993 0,70894223 9,07202395 0,01211059 0,05696148 ‐0,297883 0,397882 0,2833992 8,35731014 0,01181936 0,05533863 ‐0,293397 0,371884 0,13740836 7,60921824 0,01158394 0,05409892 ‐0,290756 0,335461 ‐0,11845072 6,35869416 0,01251645 0,0527319 ‐0,280157 0,317161 ‐0,32913639 5,67816433 0,01220951 0,05084083 ‐0,272037 0,250385 ‐0,54233391 4,23193045 0,01140755 0,04802515 ‐0,262605 0,247008 ‐0,39286354 4,04437753 0,01146519 0,04752104 ‐0,248313 0,216687 ‐0,36420239 3,15606022 0,00970279 0,04180037 ‐0,199221 0,176525 ‐0,28313561 2,26606031 0,01011776 0,04247369 ‐0,218138 0,165042 ‐0,40349802 2,66626066 1,10158608 4,03515098 ‐15,2393736 14,0847253 ‐0,40257948 1,78906511 © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 23 Empirical Framework: The Method and Expected Results ² We report unconstrained equation maximum likelihood estimates. ² These estimates are called unconstrained because the theoretical restrictions on equilibrium expected log returns imposed by skew-normal distribution are not taken into account. ² This procedure allows to focus on the coskewness of joint log returns in an environment that is not contamined by potentially false elements of the asset pricing model. ² Equations (22) and (24) are jointly estimated taking into account all cross-equation restrictions. We report constrained maximum likelihood estimates. © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 24 Empirical Framework: The Results ² We report implied unconstrained estimates of factor loadings. ² We expect coefficients of skewness (for financial and real estate returns) to be highly significant in all cases. ² Coskewness and idiosyncratic coskewness should have an important role in the explanation of the cross-sectional variations of returns. ² To recover information about the premiums related to volatily and skewness, we integrate the theoretical restrictions imposed by (22) in the estimation procedure. ² We can easily compute implied risk premiums and associated factor loadings of the constrained and unconstrained estimates (Covariance and Skewness risks). © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 25 Conclusions Main results: - Modelling with Azzalini (1985)’s skew-normal distribution taking into account real estate in the wealth function. - Explicit and analytic expressions of the skewness premiums: the roles of Coskewness and Idiosyncratic coskewness in an asset pricing with real estate. - Maximum likelihood estimates reveal very small statistically significant skewness premiums. - With Azzalini (1985)s’ distribution, the skewness market premium and the skewness real estate premium extend the CAPM focusing on idiosyncratic risk. © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012 26 Conclusions ² Following Poterba (1984, 1991), Mankiw and Weil (1989), Case and Shiller (1988), Ayuso and Restoy (2006), extension of an equilibrium asset pricing approach in a real estate context where returns are skewed. ² A CAPM with real estate including idiosyncratic coskewness. © 2012, Carmichael, Coën ACFAS_ Real_Estate_Finance Montreal May 2012