3.CAPM
-
不存在无风险证券时的资产定价问题
-
两个边界组合的协方差关系
已知两个市场组合 P P P、 Q Q Q,两者收益率的协方差
(62) C o v ( r P ,   r Q ) = C o v ( X P T   R ,   X Q T   R ) = X P T   V   X Q = X Q T   V   X P Cov\left(r_P,\,r_Q\right)=Cov\left(\textbf X_P^T\,\textbf R,\,\textbf X_Q^T\,\textbf R\right)=\textbf X_P^T\,\textbf V\,\textbf X_Q=\textbf X_Q^T\,\textbf V\,\textbf X_P\tag{62} Cov(rP,rQ)=Cov(XPTR,XQTR)=XPTVXQ=XQTVXP(62)
由   ( 25 )   \,\left(25\right)\, (25),我们知道边界组合的投资比例与期望收益率满足
X P = E ( r P ) C − B D V − 1 R − E ( r P ) B − A D V − 1 e = C   V − 1 R − B   V − 1 e D E ( r P ) − B   V − 1 R − A   V − 1 e D \begin{aligned}\textbf X_P&=\frac{E\left(r_P\right)C-B}D\textbf V^{-1}\textbf R-\frac{E\left(r_P\right)B-A}D\textbf V^{-1}\textbf e\\&=\frac{C\,\textbf V^{-1}\textbf R-B\,\textbf V^{-1}\textbf e}DE\left(r_P\right)-\frac{B\,\textbf V^{-1}\textbf R-A\,\textbf V^{-1}\textbf e}D\end{aligned} XP=DE(rP)C−BV−1R−DE(rP)B−AV−1e=DCV−1R−BV−1eE(rP)−DBV−1R−AV−1e
X Q = C   V − 1 R − B   V − 1 e D E ( r Q ) − B   V − 1 R − A   V − 1 e D \textbf X_Q=\frac{C\,\textbf V^{-1}\textbf R-B\,\textbf V^{-1}\textbf e}DE\left(r_Q\right)-\frac{B\,\textbf V^{-1}\textbf R-A\,\textbf V^{-1}\textbf e}D XQ=DCV−1R−BV−1eE(rQ)−DBV−1R−AV−1e
故两者的协方差
(63) C o v ( r P ,   r Q ) = X Q T   V   X P = C   X Q T R − B   X Q T e D E ( r P ) − B   X Q T R − A   X Q T e D = C ⋅ E ( r Q ) − B D E ( r P ) − B ⋅ E ( r Q ) − A D = C D [ ( E ( r Q ) − B C ) ( E ( r P ) − B C ) ] + 1 C \begin{aligned}Cov\left(r_P,\,r_Q\right)=\textbf X_Q^T\,\textbf V\,\textbf X_P&=\frac{C\,\textbf X_Q^T\textbf R-B\,\textbf X_Q^T\textbf e}DE\left(r_P\right)-\frac{B\,\textbf X_Q^T\textbf R-A\,\textbf X_Q^T\textbf e}D\\&=\frac{C\cdot E\left(r_Q\right)-B}DE\left(r_P\right)-\frac{B\cdot E\left(r_Q\right)-A}D\\&=\frac CD\left[\left(E\left(r_Q\right)-\frac BC\right)\left(E\left(r_P\right)-\frac BC\right)\right]+\frac 1C\end{aligned}\tag{63} Cov(rP,rQ)=XQTVXP=DCXQTR−BXQTeE(rP)−DBXQTR−AXQTe=DC⋅E(rQ)−BE(rP)−DB⋅E(rQ)−A=DC[(E(rQ)−CB)(E(rP)−CB)]+C1(63) -
零方差组合
已知组合   P   \,P\, P是一个边界组合(非   M V P   \,MVP\, MVP),组合   Q   \,Q\, Q为在边界组合曲线上,与组合   P   \,P\, P协方差为零的组合。
我们记这个组合为   Z C ( P )   \,Z_{C\left(P\right)}\, ZC(P),为关于   P   \,P\, P的零方差组合。事实上,在推导   ( 63 )   \,\left(63\right)\, (63)时,我们并没有用到组合   Q   \,Q\, Q为边界组合的条件。所以,对于所有市场组合,满足与组合   P   \,P\, P协方差为零的组合的集合为过组合   Z C ( P )   \,Z_{C\left(P\right)\,} ZC(P)的一条等收益直线。
容易得出   Z C ( P )   \,Z_{C\left(P\right)\,} ZC(P)的坐标为
( B ( r P ) − A C E ( r P ) − B ,   C E ( r P ) 2 − 2 B E ( r P ) + A C E ( r P ) − B ) \left(\frac{B\left(r_P\right)-A}{CE\left(r_P\right)-B},\,\frac{\sqrt{CE\left(r_P\right)^2-2BE\left(r_P\right)+A}}{CE\left(r_P\right)-B}\right) ⎝⎛CE(rP)−BB(rP)−A,CE(rP)−BCE(rP)2−2BE(rP)+A⎠⎞
我们对以上结论进行推广。一般意义下,设组合   Q   \,Q\, Q为任意一个组合,已知   P   \,P\, P是一个边界组合(非   M V P   \,MVP\, MVP),我们依然有   ( 63 )   \,\left(63\right)\, (63)的结论。设
(64) β P Q ≜ C o v ( r P , r Q ) C o v ( r P , r P ) \beta_{PQ}\triangleq\frac{Cov\left(r_P,r_Q\right)}{Cov\left(r_P,r_P\right)}\tag{64} βPQ≜Cov(rP,rP)Cov(rP,rQ)(64)
有
(65) E ( r Q ) = E ( r Z C ( P ) ) + β P Q [ E ( r P ) − E ( r Z C ( P ) ) ] E\left(r_Q\right)=E\left(r_{Z_{C\left(P\right)}}\right)+\beta_{PQ}\left[E\left(r_P\right)-E\left(r_{Z_{C\left(P\right)}}\right)\right]\tag{65} E(rQ)=E(rZC(P))+βPQ[E(rP)−E(rZC(P))](65)
证明是很容易的。由   ( 63 )   \,\left(63\right)\, (63)我们可以看出:
E ( r Q ) − B C = E ( r Q ) − E ( r Z C ( P ) ) + E ( r Z C ( P ) ) − B C = E ( r Q ) − E ( r Z C ( P ) ) − 1 C C D ( E ( r P ) − B C ) \begin{aligned} E\left(r_Q\right)-\frac BC&=E\left(r_Q\right)-E\left(r_{Z_{C\left(P\right)}}\right)+E\left(r_{Z_{C\left(P\right)}}\right)-\frac BC\\&=E\left(r_Q\right)-E\left(r_{Z_{C\left(P\right)}}\right)-\frac{\frac 1C}{\frac CD\left(E\left(r_P\right)-\frac BC\right)} \end{aligned} E(rQ)−CB=E(rQ)−E(rZC(P))+E(rZC(P))−CB=E(rQ)−E(rZC(P))−DC(E(rP)−CB)C1
同理
E ( r P ) − B C = E ( r P ) − E ( r Z C ( P ) ) − 1 C C D ( E ( r P ) − B C ) E\left(r_P\right)-\frac BC=E\left(r_P\right)-E\left(r_{Z_{C\left(P\right)}}\right)-\frac{\frac 1C}{\frac CD\left(E\left(r_P\right)-\frac BC\right)} E(rP)−CB=E(rP)−E(rZC(P))−DC(E(rP)−CB)C1
于是有
( E ( r Q ) − B C ) ( E ( r P ) − B C ) = [ E ( r Q ) − E ( r Z C ( P ) ) ] ( E ( r P ) − B C ) − 1 C C D \left(E\left(r_Q\right)-\frac BC\right)\left(E\left(r_P\right)-\frac BC\right) =\left[E\left(r_Q\right)-E\left(r_{Z_{C\left(P\right)}}\right)\right]\left(E\left(r_P\right)-\frac BC\right)-\frac{\frac 1C}{\frac CD} (E(rQ)−CB)(E(rP)−CB)=[E(rQ)−E(rZC(P))](E(rP)−CB)−DCC1 ( E ( r P ) − B C ) ( E ( r P ) − B C ) = [ E ( r P ) − E ( r Z C ( P ) ) ] ( E ( r P ) − B C ) − 1 C C D \left(E\left(r_P\right)-\frac BC\right)\left(E\left(r_P\right)-\frac BC\right) =\left[E\left(r_P\right)-E\left(r_{Z_{C\left(P\right)}}\right)\right]\left(E\left(r_P\right)-\frac BC\right)-\frac{\frac 1C}{\frac CD} (E(rP)−CB)(E(rP)−CB)=[E(rP)−E(rZC(P))](E(rP)−CB)−DCC1 ⟹ β P Q = C o v ( r P , r Q ) C o v ( r P , r P ) = C D [ ( E ( r Q ) − B C ) ( E ( r P ) − B C ) ] + 1 C C D [ ( E ( r P ) − B C ) ( E ( r P ) − B C ) ] + 1 C = C D [ E ( r Q ) − E ( r Z C ( P ) ) ] ( E ( r P ) − B C ) − 1 C + 1 C C D [ E ( r P ) − E ( r Z C ( P ) ) ] ( E ( r P ) − B C ) − 1 C + 1 C = E ( r Q ) − E ( r Z C ( P ) ) E ( r P ) − E ( r Z C ( P ) ) \begin{aligned}\Longrightarrow\beta_{PQ} &=\frac{Cov\left(r_P,r_Q\right)}{Cov\left(r_P,r_P\right)}=\frac{\frac CD\left[\left(E\left(r_Q\right)-\frac BC\right)\left(E\left(r_P\right)-\frac BC\right)\right]+\frac 1C}{\frac CD\left[\left(E\left(r_P\right)-\frac BC\right)\left(E\left(r_P\right)-\frac BC\right)\right]+\frac 1C}\\ &=\frac{\frac CD \left[E\left(r_Q\right)-E\left(r_{Z_{C\left(P\right)}}\right)\right]\left(E\left(r_P\right)-\frac BC\right)-\frac 1C+\frac 1C}{\frac CD \left[E\left(r_P\right)-E\left(r_{Z_{C\left(P\right)}}\right)\right]\left(E\left(r_P\right)-\frac BC\right)-\frac 1C+\frac 1C}\\ &=\frac{E\left(r_Q\right)-E\left(r_{Z_{C\left(P\right)}}\right)}{E\left(r_P\right)-E\left(r_{Z_{C\left(P\right)}}\right)} \end{aligned} ⟹βPQ=Cov(rP,rP)Cov(rP,rQ)=DC[(E(rP)−CB)(E(rP)−CB)]+C1DC[(E(rQ)−CB)(E(rP)−CB)]+C1=DC[E(rP)−E(rZC(P))](E(rP)−CB)−C1+C1DC[E(rQ)−E(rZC(P))](E(rP)−CB)−C1+C1=E(rP)−E(rZC(P))E(rQ)−E(rZC(P))
这与式   ( 65 )   \,\left(65\right)\, (65)是等价的。
- 存在无风险证券时的资产定价问题(一般意义下的资产定价)
\quad
P为一个已知有效前沿组合,则任意组合   Q   \,Q\, Q,有
(66) E ( r Q ) = r f + β P Q [ E ( r P ) − r f ] E\left(r_Q\right)=r_f+\beta_{PQ}\left[E\left(r_P\right)-r_f\right]\tag{66} E(rQ)=rf+βPQ[E(rP)−rf](66)
式 ( 66 ) \left(66\right) (66)是很显然的。由式   ( 45 )   \,\left(45\right)\, (45)我们知道
X P = E ( r P ) − r f H V − 1 ( R − r f ⋅ e ) \textbf X_P=\frac{E\left(r_P\right)-r_f}{H}\textbf V^{-1}\left(\textbf R-r_f\cdot\textbf e\right) XP=HE(rP)−rfV−1(R−rf⋅e)
故
C o v ( r P ,   r Q ) = X Q T V X P = E ( r P ) − r f H ( X Q T R − r f ⋅ X Q T e ) = ( E ( r P ) − r f ) ( E ( r Q ) − r f ) H \begin{aligned} Cov\left(r_P,\,r_Q\right)&=\textbf X_Q^T\textbf V\textbf X_P\\ &=\frac{E\left(r_P\right)-r_f}{H}\left(\textbf X_Q^T\textbf R-r_f\cdot\textbf X_Q^T\textbf e\right)\\ &=\frac{\left(E\left(r_P\right)-r_f\right)\left(E\left(r_Q\right)-r_f\right)}{H} \end{aligned} Cov(rP,rQ)=XQTVXP=HE(rP)−rf(XQTR−rf⋅XQTe)=H(E(rP)−rf)(E(rQ)−rf)
又由式   ( 46 )   \,\left(46\right)\, (46)
C o v ( r P ,   r P ) = σ P 2 = ( E ( r P ) − r f ) 2 H Cov\left(r_P,\,r_P\right)=\sigma_P^2=\frac{\left(E\left(r_P\right)-r_f\right)^2}{H} Cov(rP,rP)=σP2=H(E(rP)−rf)2
故
β P Q = C o v ( r P , r Q ) C o v ( r P , r P ) = ( E ( r P ) − r f ) ( E ( r Q ) − r f ) H ( E ( r P ) − r f ) 2 H = E ( r Q ) − r f E ( r P ) − r f \beta_{PQ}=\frac{Cov\left(r_P,r_Q\right)}{Cov\left(r_P,r_P\right)}=\frac{\frac{\left(E\left(r_P\right)-r_f\right)\left(E\left(r_Q\right)-r_f\right)}{H}}{\frac{\left(E\left(r_P\right)-r_f\right)^2}{H}}=\frac{E\left(r_Q\right)-r_f}{E\left(r_P\right)-r_f} βPQ=Cov(rP,rP)Cov(rP,rQ)=H(E(rP)−rf)2H(E(rP)−rf)(E(rQ)−rf)=E(rP)−rfE(rQ)−rf
即
E ( r Q ) = r f + β P Q [ E ( r P ) − r f ] E\left(r_Q\right)=r_f+\beta_{PQ}\left[E\left(r_P\right)-r_f\right] E(rQ)=rf+βPQ[E(rP)−rf]
- 资本资产定价模型(Captital Asset Pricing Model, CAPM)
- 基本假设
- 记切点代表的组合为市场组合   M   \,M\, M,为以各自的证券市值比为权重的一个组合。
证券1 | 证券2 | … \dots … | 证券k | … \dots … | 证券n | |
---|---|---|---|---|---|---|
价格 | P 1 P_1 P1 | P 2 P_2 P2 | ⋯ \cdots ⋯ | P k P_k Pk | … \dots … | P n P_n Pn |
发行量 | Q 1 Q_1 Q1 | Q 2 Q_2 Q2 | ⋯ \cdots ⋯ | Q k Q_k Qk | … \dots … | Q n Q_n Qn |
市值 | P 1 Q 1 P_1Q_1 P1Q1 | P 2 Q 2 P_2Q_2 P2Q2 | ⋯ \cdots ⋯ | P k Q k P_kQ_k PkQk | … \dots … | P n Q n P_nQ_n PnQn |
组合 M M M | P 1 Q 1 ∑ P i Q i \frac{P_1Q_1}{\sum P_iQ_i} ∑PiQiP1Q1 | P 2 Q 2 ∑ P i Q i \frac{P_2Q_2}{\sum P_iQ_i} ∑PiQiP2Q2 | ⋯ \cdots ⋯ | P k Q k ∑ P i Q i \frac{P_kQ_k}{\sum P_iQ_i} ∑PiQiPkQk | … \dots … | P n Q n ∑ P i Q i \frac{P_nQ_n}{\sum P_iQ_i} ∑PiQiPnQn |
β \beta β系数 | ρ 1 M σ 1 σ M \rho_{1M}\frac{\sigma_1}{\sigma_M} ρ1MσMσ1 | ρ 2 M σ 2 σ M \rho_{2M}\frac{\sigma_2}{\sigma_M} ρ2MσMσ2 | ⋯ \cdots ⋯ | ρ k M σ k σ M \rho_{kM}\frac{\sigma_k}{\sigma_M} ρkMσMσk | … \dots … | ρ n M σ n σ M \rho_{nM}\frac{\sigma_n}{\sigma_M} ρnMσMσn |
组合 P P P | x 1 x_1 x1 | x 2 x_2 x2 | ⋯ \cdots ⋯ | x k x_k xk | … \dots … | x n x_n xn |
β i M = C o v ( r i , r M ) C o v ( r M , r M ) = C o v ( r i , r M ) σ M 2 = ρ i σ i σ M \quad\quad\beta_{iM}=\frac{Cov\left(r_i,r_M\right)}{Cov\left(r_M,r_M\right)}=\frac{Cov\left(r_i,r_M\right)}{\sigma_M^2}=\rho_i\frac{\sigma_i}{\sigma_M} βiM=Cov(rM,rM)Cov(ri,rM)=σM2Cov(ri,rM)=ρiσMσi
- 投资组合
 
P
 
\,P\,
P
X P = ( x 1 , x 2 , ⋯   , x n ) T \textbf X_P=\left(x_1,x_2,\cdots,x_n\right)^T XP=(x1,x2,⋯,xn)T β P = ∑ x i β i \beta_P=\sum x_i\beta_i βP=∑xiβi - 对证券
 
i
 
\,i\,
i有
(67) E ( r i ) = r f + β i [ E ( r M ) − r f ] E\left(r_i\right)=r_f+\beta_i\left[E\left(r_M\right)-r_f\right]\tag{67} E(ri)=rf+βi[E(rM)−rf](67) - 对投资组合
 
P
 
\,P\,
P有
(68) E ( r P ) = r f + β P [ E ( r M ) − r f ] E\left(r_P\right)=r_f+\beta_P\left[E\left(r_M\right)-r_f\right]\tag{68} E(rP)=rf+βP[E(rM)−rf](68)
这就是著名的资本资产定价模型。
其正确性我们在上一小节经证明过了,下面再给出一种新的证明思路:
- 构造   i   \,i\, i、   M   \,M\, M组合,其中组合   i   \,i\, i为单一证券   i   \,i\, i:
过   M   \,M\, M与点   F ( 0 ,   r f )   \,F\left(0,\,r_f\right)\, F(0,rf)的直线(即切线)斜率
k = E ( r M ) − r f σ M k=\frac{E\left(r_M\right)-r_f}{\sigma_M} k=σME(rM)−rf
对于两证券   i   \,i\, i、   M   \,M\, M,按照比例   ( x : 1 − x )   \,\left(x:1-x\right)\, (x:1−x)构成的组合   P   \,P\, P,有
r P = x ⋅ r i + ( 1 − x ) ⋅ r M r_P=x\cdot r_i+\left(1-x\right)\cdot r_M rP=x⋅ri+(1−x)⋅rM ⟹ E ( r P ) = x ⋅ E ( r i ) + ( 1 − x ) ⋅ E ( r M ) \Longrightarrow E\left(r_P\right)=x\cdot E\left(r_i\right)+\left(1-x\right)\cdot E\left(r_M\right) ⟹E(rP)=x⋅E(ri)+(1−x)⋅E(rM) σ P 2 = x 2 σ i 2 + ( 1 − x ) 2 σ M 2 + 2 x ( 1 − x ) σ i M \sigma_P^2=x^2\sigma_i^2+\left(1-x\right)^2\sigma_M^2+2x\left(1-x\right)\sigma_{iM} σP2=x2σi2+(1−x)2σM2+2x(1−x)σiM
在第二节中我们已经证明了,组合   P   \,P\, P的可行集为双曲线的一支,且经过点   M   \,M\, M。此曲线过   M   \,M\, M点的切线斜率为
k ′ = d E ( r P ) d σ P ∣ M = d E ( r P ) d x d σ P d x ∣ M = E ( r i ) − E ( r m ) [ x σ i 2 − ( 1 − x ) σ M 2 + ( 1 − 2 x ) σ i M ] / σ P ∣ M k'=\left.\frac{\textbf dE\left(r_P\right)}{\textbf d\sigma_P}\right|_M=\left.\frac{\frac{\textbf dE\left(r_P\right)}{\textbf dx}}{\frac{\textbf d\sigma_P}{\textbf dx}}\right|_M=\left.\frac{E\left(r_i\right)-E\left(r_m\right)}{\left[x\sigma_i^2-\left(1-x\right)\sigma_M^2+\left(1-2x\right)\sigma_{iM}\right]/\sigma_P}\right|_M k′=dσPdE(rP)∣∣∣∣M=dxdσPdxdE(rP)∣∣∣∣∣M=[xσi2−(1−x)σM2+(1−2x)σiM]/σPE(ri)−E(rm)∣∣∣∣M
而在点   M   \,M\, M处,应取   x = 0   \,x=0\, x=0有
E ( r P ) = E ( r M ) E\left(r_P\right)=E\left(r_M\right) E(rP)=E(rM) σ P 2 = σ M 2 \sigma_P^2=\sigma_M^2 σP2=σM2
则
k ′ = E ( r i ) − E ( r m ) ( − σ M 2 + σ i M ) / σ M = E ( r i ) − E ( r m ) ( − σ M 2 + β i σ M 2 ) / σ M = E ( r i ) − E ( r m ) ( β i − 1 ) σ M k'=\frac{E\left(r_i\right)-E\left(r_m\right)}{\left(-\sigma_M^2+\sigma_{iM}\right)/\sigma_M}=\frac{E\left(r_i\right)-E\left(r_m\right)}{\left(-\sigma_M^2+\beta_i\sigma_M^2\right)/\sigma_M}=\frac{E\left(r_i\right)-E\left(r_m\right)}{\left(\beta_i-1\right)\sigma_M} k′=(−σM2+σiM)/σME(ri)−E(rm)=(−σM2+βiσM2)/σME(ri)−E(rm)=(βi−1)σME(ri)−E(rm)
若式   ( 67 )   \,\left(67\right)\, (67)成立,则有
k ′ = r f + β i [ E ( r M ) − r f ] − E ( r M ) ( β i − 1 ) σ M = ( β i − 1 ) ( E ( r M ) − r f ) ( β i − 1 ) σ M = E ( r M ) − r f σ M = k k'=\frac{r_f+\beta_i\left[E\left(r_M\right)-r_f\right]-E\left(r_M\right)}{\left(\beta_i-1\right)\sigma_M}=\frac{\left(\beta_i-1\right)\left(E\left(r_M\right)-r_f\right)}{\left(\beta_i-1\right)\sigma_M}=\frac{E\left(r_M\right)-r_f}{\sigma_M}=k k′=(βi−1)σMrf+βi[E(rM)−rf]−E(rM)=(βi−1)σM(βi−1)(E(rM)−rf)=σME(rM)−rf=k
因此,若要证明式   ( 67 )   \,\left(67\right)\, (67),我么只需要验证此曲线过   M   \,M\, M点的切线与   M F   \,MF\, MF重合,而这是显然的。
4.CML & SML
- 资本市场线 CML
在第二节的最后部分,我们提出了资本市场线。
σ P = 1 H ( E ( r P ) − r f ) \sigma_P=\frac 1{\sqrt H}\left(E\left(r_P\right)-r_f\right) σP=H1(E(rP)−rf)
资本市场线是允许无风险借贷情况下的线性有效集,其反映的是有效组合的预期收益率和标准差之间的关系。当组合落在   R f   \,R_f\, Rf与   M   \,M\, M之间时,我们将一定比例的钱存入银行得到无风险收益   R f   \,R_f\, Rf的回报,剩余的钱用于投资   M   \,M\, M;当组合落在   M   \,M\, M之外时,我们从银行以无风险收益   R f   \,R_f\, Rf借贷,并将所有的钱用于投资组合   M   \,M\, M. - 证券市场线 SML
E ( r P ) = r f + β P [ E ( r M ) − r f ] E\left(r_P\right)=r_f+\beta_P\left[E\left(r_M\right)-r_f\right] E(rP)=rf+βP[E(rM)−rf]
证券市场线反映了单个证券与市场组合的协方差和其预期收益率之间的均衡关系。 - CML 与 SML 的关系
- 只有最优投资组合才落在资本市场线上,其他组合和证券则落在资本市场线的下方 只有最优投资组合才落在资本市场线上,其他组合和证券则落在资本市场线的下方
- 无论是有效组合还是非有效组合,它们都落在证券市场线上
5.APT
-
单因子模型
市场中有 n n n个证券,我们用一个因子 F F F对每一个证券 i i i的收益率进行最小二乘回归:
r i = a i + b i   F + ε i r_i=a_i+b_i\,F+\varepsilon_i ri=ai+biF+εi其中拟合残差项 ε i \varepsilon_i εi满足:
C o v ( F ,   ε i ) = 0 Cov\left(F,\,\varepsilon_i\right)=0 Cov(F,εi)=0 E ( ε i ) = 0 E\left(\varepsilon_i\right)=0 E(εi)=0 σ ε i > 0 \sigma_{\varepsilon_i}>0 σεi>0 C o v ( ε i ,   ε j ) = 0 Cov\left(\varepsilon_i,\,\varepsilon_j\right)=0 Cov(εi,εj)=0
容易看出
E ( r i ) = a i + b i E ( F ) E\left(r_i\right)=a_i+b_iE\left(F\right) E(ri)=ai+biE(F) σ i 2 = b i 2 σ F 2 + σ ε i \sigma_i^2=b_i^2\sigma_F^2+\sigma_{\varepsilon_i} σi2=bi2σF2+σεi
对于市场组合 P P P有
r P = ∑ x i r i = ( ∑ x i a i ) + ( ∑ x i b i )   F + ( ∑ x i ε i ) r_P=\sum x_ir_i=\left(\sum x_ia_i\right)+\left(\sum x_ib_i\right)\,F+\left(\sum x_i\varepsilon_i\right) rP=∑xiri=(∑xiai)+(∑xibi)F+(∑xiεi) ⟹ E ( r P ) = ( ∑ x i a i ) + ( ∑ x i b i )   F \Longrightarrow E\left(r_P\right)=\left(\sum x_ia_i\right)+\left(\sum x_ib_i\right)\,F ⟹E(rP)=(∑xiai)+(∑xibi)F σ P 2 = C o v ( r P ,   r P ) = ( x 1 x 2 ⋯ x n ) ( b 1 2 σ F 2 + σ ε 1 2 b 1 b 2 σ F 2 ⋯ b 1 b n σ F 2 b 2 b 1 σ F 2 b 2 2 σ F 2 + σ ε 2 2 ⋯ b 2 b n σ F 2 ⋮ ⋮ ⋱ ⋮ b n b 1 σ F 2 b n b 2 σ F 2 ⋯ b n 2 σ F 2 + σ ε n 2 ) ( x 1 x 2 ⋮ x n ) = ( x 1 x 2 ⋯ x n ) ( b 1 2 σ F 2 b 1 b 2 σ F 2 ⋯ b 1 b n σ F 2 b 2 b 1 σ F 2 b 2 2 σ F 2 ⋯ b 2 b n σ F 2 ⋮ ⋮ ⋱ ⋮ b n b 1 σ F 2 b n b 2 σ F 2 ⋯ b n 2 σ F 2 ) ( x 1 x 2 ⋮ x n ) + ( x 1 x 2 ⋯ x n ) ( σ ε 1 2 0 ⋯ 0 0 σ ε 2 2 ⋯ 0 ⋮ ⋮ ⋱ ⋮ 0 0 ⋯ σ ε n 2 ) ( x 1 x 2 ⋮ x n ) \begin{aligned}\sigma_P^2&=Cov\left(r_P,\,r_P\right)\\&=\left(\begin{matrix}x_1&x_2&\cdots&x_n\end{matrix}\right)\left(\begin{matrix}b_1^2\sigma_F^2+\sigma_{\varepsilon_1}^2&b_1b_2\sigma_F^2&\cdots&b_1b_n\sigma_F^2\\b_2b_1\sigma_F^2&b_2^2\sigma_F^2+\sigma_{\varepsilon_2}^2&\cdots&b_2b_n\sigma_F^2\\\vdots&\vdots&\ddots&\vdots&\\b_nb_1\sigma_F^2&b_nb_2\sigma_F^2&\cdots&b_n^2\sigma_F^2+\sigma_{\varepsilon_n}^2\end{matrix}\right)\left(\begin{matrix}x_1\\x_2\\\vdots\\x_n\end{matrix}\right)\\&=\left(\begin{matrix}x_1&x_2&\cdots&x_n\end{matrix}\right)\left(\begin{matrix}b_1^2\sigma_F^2&b_1b_2\sigma_F^2&\cdots&b_1b_n\sigma_F^2\\b_2b_1\sigma_F^2&b_2^2\sigma_F^2&\cdots&b_2b_n\sigma_F^2\\\vdots&\vdots&\ddots&\vdots&\\b_nb_1\sigma_F^2&b_nb_2\sigma_F^2&\cdots&b_n^2\sigma_F^2\end{matrix}\right)\left(\begin{matrix}x_1\\x_2\\\vdots\\x_n\end{matrix}\right)\\&\quad+\left(\begin{matrix}x_1&x_2&\cdots&x_n\end{matrix}\right)\left(\begin{matrix}\sigma_{\varepsilon_1}^2&0&\cdots&0\\0&\sigma_{\varepsilon_2}^2&\cdots&0\\\vdots&\vdots&\ddots&\vdots&\\0&0&\cdots&\sigma_{\varepsilon_n}^2\end{matrix}\right)\left(\begin{matrix}x_1\\x_2\\\vdots\\x_n\end{matrix}\right)\end{aligned} σP2=Cov(rP,rP)=(x1x2⋯xn)⎝⎜⎜⎜⎛b12σF2+σε12b2b1σF2⋮bnb1σF2b1b2σF2b22σF2+σε22⋮bnb2σF2⋯⋯⋱⋯b1bnσF2b2bnσF2⋮bn2σF2+σεn2⎠⎟⎟⎟⎞⎝⎜⎜⎜⎛x1x2⋮xn⎠⎟⎟⎟⎞=(x1x2⋯xn)⎝⎜⎜⎜⎛b12σF2b2b1σF2⋮bnb1σF2b1b2σF2b22σF2⋮bnb2σF2⋯⋯⋱⋯b1bnσF2b2bnσF2⋮bn2σF2⎠⎟⎟⎟⎞⎝⎜⎜⎜⎛x1x2⋮xn⎠⎟⎟⎟⎞+(x1x2⋯xn)⎝⎜⎜⎜⎛σε120⋮00σε22⋮0⋯⋯⋱⋯00⋮σεn2⎠⎟⎟⎟⎞⎝⎜⎜⎜⎛x1x2⋮xn⎠⎟⎟⎟⎞
记
a P = ∑ x i a i a_P=\sum x_ia_i aP=∑xiai b P = ∑ x i b i b_P=\sum x_ib_i bP=∑xibi σ ε P 2 = ∑ x i 2 σ ε i 2 \sigma_{\varepsilon_P}^2=\sum x_i^2\sigma_{\varepsilon_i}^2 σεP2=∑xi2σεi2
则有
E ( r P ) = a P + b P F E\left(r_P\right)=a_P+b_PF E(rP)=aP+bPF σ P 2 = b P 2 σ F 2 + σ ε P 2 \sigma_P^2=b_P^2\sigma_F^2+\sigma_{\varepsilon_P}^2 σP2=bP2σF2+σεP2 -
多因子模型
现在我们考虑有 k k k个因子 F 1 F_1 F1, F 2 F_2 F2, ⋯ \cdots ⋯, F n F_n Fn
和前面一样,我们也对收益率做最小二乘回归
r i = a i + b i 1 F 1 + b i 2 F 2 + ⋯ + b i k F k + ε i r_i=a_i+b_{i1}F_1+b_{i2}F_2+\cdots+b_{ik}F_k+\varepsilon_i ri=ai+bi1F1+bi2F2+⋯+bikFk+εi
其中拟合残差项 ε i \varepsilon_i εi和 F j F_j Fj满足:
C o v ( F j 1 ,   F j 2 ) = 0 Cov\left(F_{j_1},\,F_{j_2}\right)=0 Cov(Fj1,Fj2)=0 C o v ( F j ,   ε i ) = 0 Cov\left(F_j,\,\varepsilon_i\right)=0 Cov(Fj,εi)=0 E ( ε i ) = 0 E\left(\varepsilon_i\right)=0 E(εi)=0 σ ε i > 0 \sigma_{\varepsilon_i}>0 σεi>0 C o v ( ε i ,   ε j ) = 0 Cov\left(\varepsilon_i,\,\varepsilon_j\right)=0 Cov(εi,εj)=0
那么,市场组合P的期望收益与方差为
E ( r P ) = ∑ x i E ( r i ) = ∑ a i x i + F 1 ∑ b i 1 x i + ⋯ + F k ∑ b i k x i E\left(r_P\right)=\sum x_iE\left(r_i\right)=\sum a_ix_i+F_1\sum b_{i1}x_i+\cdots+F_k\sum b_{ik}x_i E(rP)=∑xiE(ri)=∑aixi+F1∑bi1xi+⋯+Fk∑bikxi σ P 2 = σ F 1 2 ( ∑ x i b 1 i ) 2 + ⋯ + σ F k 2 ( ∑ x i b k i ) 2 + ∑ x i 2 σ ε i 2 \sigma_P^2=\sigma_{F_1}^2\left(\sum x_ib_{1i}\right)^2+\cdots+\sigma_{F_k}^2\left(\sum x_ib_{ki}\right)^2+\sum x_i^2\sigma_{\varepsilon_i}^2 σP2=σF12(∑xib1i)2+⋯+σFk2(∑xibki)2+∑xi2σεi2 -
套利定价理论(Arbitrage Pricing Theory, APT)
\quad
在多因子模型的基础上,我们增加一些条件。首先有
r P = ∑ a i x i + F 1 ∑ b i 1 x i + ⋯ + F k ∑ b i k x i + ∑ x i σ ε i r_P=\sum a_ix_i+F_1\sum b_{i1}x_i+\cdots+F_k\sum b_{ik}x_i+\sum x_i\sigma_{\varepsilon_i} rP=∑aixi+F1∑bi1xi+⋯+Fk∑bikxi+∑xiσεi -
基本假设
\quad
a )   a)\, a)投资者不追加资金, 即套利组合属于自融资组合
∑ x i = 0 \sum x_i=0 ∑xi=0
b )   b)\, b)对任何因素的敏感度均为零,即套利组合没有风险(系统性风险与非系统性风险均无)- 无系统性风险
∑ x i σ ε i = 0 \sum x_i\sigma_{\varepsilon_i}=0 ∑xiσεi=0 - 无非系统性风险
∑ b i j x i = 0 j = 1 , 2 , ⋯   , k \sum b_{ij}x_i=0\quad j=1,2,\cdots,k ∑bijxi=0j=1,2,⋯,k
c )   c)\, c)预期收益率大于零(不等于零)
E ( r P ) = ∑ x i E ( r i ) > 0 E\left(r_P\right)=\sum x_iE\left(r_i\right)>0 E(rP)=∑xiE(ri)>0 - 无系统性风险
-
建模过程
m a x   E ( r P ) = X T R {\rm max}\,E\left(r_P\right)=\textbf X^T\textbf R maxE(rP)=XTR s . t . { X T e = 0 X T b 1 = 0 ⋮ X T b k = 0 s.t.\begin{cases}\begin{aligned}\textbf X^T\textbf e&=0\\\textbf X^T\textbf b_1&=0\\\vdots\\\textbf X^T\textbf b_k&=0\end{aligned}\end{cases} s.t.⎩⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎧XTeXTb1⋮XTbk=0=0=0
其中
R = ( E ( r 1 ) , E ( r 2 ) , ⋯   , E ( r n ) ) T \textbf R=\left(E\left(r_1\right),E\left(r_2\right),\cdots,E\left(r_n\right)\right)^T R=(E(r1),E(r2),⋯,E(rn))T b j = ( b 1 j , b 2 j , ⋯   , b n j ) T j = 1 , 2 , ⋯   , k \textbf b_j=\left(b_{1j},b_{2j},\cdots,b_{nj}\right)^T\quad j=1,2,\cdots,k bj=(b1j,b2j,⋯,bnj)Tj=1,2,⋯,k -
模型求解
\quad
令
L = X T R − λ 0 ( X T e ) − λ 1 ( X T b 1 ) − ⋯ − λ k ( X T b k ) L=\textbf X^T\textbf R-\lambda_0\left(\textbf X^T\textbf e\right)-\lambda_1\left(\textbf X^T\textbf b_1\right)-\cdots-\lambda_k\left(\textbf X^T\textbf b_k\right) L=XTR−λ0(XTe)−λ1(XTb1)−⋯−λk(XTbk)
则
⟹ { ∂ L ∂ X = ( ∂ L ∂ X 1 , ⋯   , ∂ L ∂ X n ) T = R − ( λ 0 e + λ 1 b 1 + ⋯ + λ k b k ) = 0 ∂ L ∂ λ 0 = − X T e = 0 ∂ L ∂ λ 1 = − X T b 1 = 0 ⋮ ∂ L ∂ λ k = − X T b k = 0 \Longrightarrow\left\{\begin{aligned}&\frac{\partial L}{\partial\textbf X}=\left(\frac{\partial L}{\partial\textbf X_1},\cdots,\frac{\partial L}{\partial\textbf X_n}\right)^T=\textbf R-\left(\lambda_0\textbf e+\lambda_1\textbf b_1+\cdots+\lambda_k\textbf b_k\right)=\textbf 0\\&\frac{\partial L}{\partial \lambda_0}=-\textbf X^T\textbf e=0\\&\frac{\partial L}{\partial \lambda_1}=-\textbf X^T\textbf b_1=0\\&\quad\quad\quad\vdots\\&\frac{\partial L}{\partial \lambda_k}=-\textbf X^T\textbf b_k=0\end{aligned}\right. ⟹⎩⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎧∂X∂L=(∂X1∂L,⋯,∂Xn∂L)T=R−(λ0e+λ1b1+⋯+λkbk)=0∂λ0∂L=−XTe=0∂λ1∂L=−XTb1=0⋮∂λk∂L=−XTbk=0
故有
R = ( E ( r 1 ) E ( r 1 ) ⋮ E ( r 1 ) ) = λ 0 ( 1 1 ⋮ 1 ) + λ 1 ( b 11 b 21 ⋮ b n 1 ) + ⋯ + λ k ( b 1 k b 2 k ⋮ b n k ) \textbf R=\left(\begin{matrix}E\left(r_1\right)\\E\left(r_1\right)\\\vdots\\E\left(r_1\right)\end{matrix}\right)=\lambda_0\left(\begin{matrix}1\\1\\\vdots\\1\end{matrix}\right)+\lambda_1\left(\begin{matrix}b_{11}\\b_{21}\\\vdots\\b_{n1}\end{matrix}\right)+\cdots+\lambda_k\left(\begin{matrix}b_{1k}\\b_{2k}\\\vdots\\b_{nk}\end{matrix}\right) R=⎝⎜⎜⎜⎛E(r1)E(r1)⋮E(r1)⎠⎟⎟⎟⎞=λ0⎝⎜⎜⎜⎛11⋮1⎠⎟⎟⎟⎞+λ1⎝⎜⎜⎜⎛b11b21⋮bn1⎠⎟⎟⎟⎞+⋯+λk⎝⎜⎜⎜⎛b1kb2k⋮bnk⎠⎟⎟⎟⎞
这里我们无法求解出 λ i \lambda_i λi,但可以发现
m a x   E ( r P ) = 0 {\rm max}\,E\left(r_P\right)=0 maxE(rP)=0
也就是说不存在套利组合,以上就是套利定价模型。
事实上,在对模型求解时,这里在提供一种方法
要满足
{ X T e = 0 X T b 1 = 0 ⋮ X T b k = 0 \begin{cases}\begin{aligned}\textbf X^T\textbf e&=0\\\textbf X^T\textbf b_1&=0\\\vdots\\\textbf X^T\textbf b_k&=0\end{aligned}\end{cases} ⎩⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎧XTeXTb1⋮XTbk=0=0=0
设
V 1 ‾ = S p a n { e , b 1 , ⋯   , b k } \overline{V_1}=Span\left\{\textbf e,\textbf b_1,\cdots,\textbf b_k\right\} V1=Span{e,b1,⋯,bk} V ‾ = V 1 ‾ ⊕ V 1 ‾ ⊥ \overline V=\overline{V_1}\oplus\overline{V_1}^{\bot} V=V1⊕V1⊥
则
X ∈ V 1 ‾ ⊥ ⇒ R ∈ V 1 ‾ \textbf X\in\overline{V_1}^{\bot}\Rightarrow\textbf R\in\overline{V_1} X∈V1⊥⇒R∈V1
则
R = λ 0 e + λ 1 b 1 + ⋯ + λ k b k \textbf R=\lambda_0\textbf e+\lambda_1\textbf b_1+\cdots+\lambda_k\textbf b_k R=λ0e+λ1b1+⋯+λkbk
6.CAPM 应用
- 资产估值
   r    \,\,r\,\, r:投资者需要的收益水平
D t D_t Dt:红利
  \,
关于股票投资的估值:
V ‾ = ∑ t = 1 ∞ D t ( 1 + r ) t \overline V=\sum_{t=1}^\infin\frac {D_t}{\left(1+r\right)^t} V=t=1∑∞(1+r)tDt
此处的   r   \,r\, r可用 CAPM 中
E ( r ) = r f + β [ E ( r M ) − r f ] E\left(r\right)=r_f+\beta\left[E\left(r_M\right)-r_f\right] E(r)=rf+β[E(rM)−rf]
代替。 - 单期收益率
r = P M − P 0 P 0 r=\frac{P_M-P_0}{P_0} r=P0PM−P0
其中   P M   \,P_M\, PM为期末价格,是随机的;   P M   \,P_M\, PM为期初价格,是确定的。则
E ( r ) = E ( P M ) − P 0 P 0 ⇒ P 0 = E ( P M ) 1 + E ( r ) E\left(r\right)=\frac{E\left(P_M\right)-P_0}{P_0}\Rightarrow P_0=\frac{E\left(P_M\right)}{1+E\left(r\right)} E(r)=P0E(PM)−P0⇒P0=1+E(r)E(PM)
这里给出   P M   \,P_M\, PM在不同策略下的概率分布,可用 CAPM 计算   E ( r )   \,E\left(r\right)\, E(r),进而算出期初价格。 - 公司理财
在对加权平均资本成本   W A C C   \,WACC\, WACC进行计算时,需要分别对股票及债券进行估值,而在股票估值中我们可以利用 CAPM 计算。