《实证资产定价-present》由会员分享,可在线阅读,更多相关《实证资产定价-present(24页珍藏版)》请在金锄头文库上搜索。
1、Multifactor Explanations of Asset Pricing AnomaliesFama and French.ABSTRACT普通股的平均收益率与公司特征有关:大小、earnings/price,cash flow/price,book-to-equity,长期历史收益与短期历史收益等等有关。而这些是不能被CAPM所解释,故被称为市场异象。笔者发现,除了短期收益率,大部分市场异常在三因素模型中都消失了。.市场异常DeBondt and Thaler(1985):low long-term past returns tend to have higher future r
2、eturnsJegadeesh and Titman(1993):short-term returns tend to continue其他人发现平均股票收益率还与诸如:size(ME=P*No. of shares)、BE/ME、E/P、C/P and past sales growth有关。因为这些都不能用CAPM解释,故被称为异象。笔者认为,很多基于CAPM的股票收益率异象是相关联的。而且都可以用Fama and French的3因素模型来解释。3因素模型: 可以被三个因素解释1. 市场组合的超额收益2.小股票与大股票组合的收益率差:SMB3.高B/P与低B/P的组合的收益率差:HML.
3、三因素模型阐述模型表述回归模型.Relative distressBE/ME and slopes on HML are proxy for relative distress.Weak firms: low earnings, high BE/ME, positive slopes on HMLStrong firms: high earnings, low BE/ME, negative slopes on HMLChan and Chen(1991): covariation in returns related to relative distress which is not cap
4、tured by the market return and is compensated in average returns. Justify using HMLHuberman and Kandel(1987):covariation in returns on small stocks that is not captured by the market return and is compensated in average returns. Justify using SMB.FF(1993):3-因素模型较好的解释了基于size和BE/ME的组合收益率。FF(1994):使用3-
5、因素模型解释行业收益率。此处,FF要说明3-因素模型解释了基于E/P,C/P,和sales growth组合收益率。Strong firm: Low E/P, low C/P and high sales growth, negative slopes on HML(HML平均收益率大约是6%每年)imply lower expected returns。Weak firm: High E/P, High C/P, low sales growth, positive slopes on HML(relatively distressed),imply higher expected retu
6、rns.3因素模型也扑捉了长期收益率的回复效应。Low long-term past returns(losers) tend to have positive SMB and HML slopes(smaller and relatively distressed)and highter future average returns. Long-term winners tend to be strong stocks that have negative slopes on HML and low future returns. .3-factor的局限不能解释short-term ret
7、urns的延续。与long-term losers一样,low short-term past returns倾向于有正的HML loading。Short-term past winners load negatively on HML. 只有reversal能被解释,continuation不能被解释。不过不过3因素模型还是解释了大部分异象。因素模型还是解释了大部分异象。当然,反对声音也比较多的集中在当然,反对声音也比较多的集中在distress的的premium上面。上面。(survivor bias, data snooping, real but irrational).I. Tes
8、ts on the 25 FF Size-BE/ME portfoliosRf为月初观测到的one-month Treasury bill rate. 1963-1993年期间,每年的6月底,把NYSE、AMEX和Nasdaq的股票分为两组(大小,BS),基于该股票的ME到底是高于还是低于NYSE股票的ME中位数。把NYSE、AMEX和Nasdaq的股票分为三组(BE/ME,L30% M40% H30%),基于该股票的BE/ME在NYSE股票的相应位置。七月到明年6月的Value-weight 组合月收益率即可被求出。25size-BE/ME组合是通过类似的手段做出来的。BE/ME比率实际上用
9、了t-1年的BE,和t-1年的ME。他们没有使用负BE的公司和非常规common equity的数据。RM是所有的size-BE/ME组合股票的value-weight return,还加上被剔除的负BE股票的收益率。.表1Small stocks tend to have higher returns than big stocks high-book-market stocks have higher returns than low-BE/ME stocks.如果3因素模型描述了预期收益,那回归的节距项应该接近0。估计的节距项上看,小股票低BE/ME组合有大的负收益没有解释,大股票低BE
10、/ME组合有正收益没有解释。其余情况还是接近于0的。.LSV DecilesLSV1994:检验了基于BE/ME,E/P,C/P 和5年期销售排名构件的10分位组合。不过作者使用的时间仍然是6月,而不是像LSV那样使用4月。而且,作者只使用了NYSE的,包含了所有必要信息的股票1. Average return and BE/ME,E/P or C/P 有强正相关关系。2. Past sales growth is negatively related to future return。3. 表3表明,3因素模型反映了这些模式关系。回归的截距项普遍很小,R2都很大(解释力度都很大)。GRS检验
11、没有拒绝假设(3因素模型描述了平均收益率)。考虑到截距项的大小和GRS检验,3因素模型在LSV deciles上的表现要好于其在25 FF size-BE/ME组合上的表现的。表3:Higher-C/P produce larger slopes on SMB and HML. 用股价除会计变量得到的特征似乎与HML的回归系数相关。考虑到HML回归系数可以反映relative distress,笔者认为,low BE/ME,E/P and C/P是强势股的特征,而high BE/ME,E/P and C/P是 relatviely distressed 股票特征。.表2:Summary sta
12、tistics for simple monthly excess returns(in percent) on the LSV Equal-weight Deciles:7/63-12/93,366 monthsHigh sales rank, low future returns; low sales-rank, high future returns. 3因素模型捕捉到了这个模式,因为 low sales-rank的股票往往都是distressed stocks。.表33-factor time- series regressions for monthly excess returns
13、(in percent) on the LSV equal-weight deciles: 7/63-12/93,366 months截距项非常接近截距项非常接近截距项非常接近截距项非常接近0.0.尽管尽管尽管尽管sales-ranksales-rank产生了最大的产生了最大的产生了最大的产生了最大的GRSF-statistic(0.87)GRSF-statistic(0.87),但是,但是,但是,但是p-valuep-value却是却是却是却是0.56.0.56.笔者认笔者认笔者认笔者认为这是重要的问题,因为为这是重要的问题,因为为这是重要的问题,因为为这是重要的问题,因为sales-ran
14、ksales-rank是唯一不是价是唯一不是价是唯一不是价是唯一不是价格变量变化的指标。格变量变化的指标。格变量变化的指标。格变量变化的指标。Null: the regression Null: the regression intercepts for a set of tem intercepts for a set of tem portfolios are all 0.0. portfolios are all 0.0. p(GRS) is the p-value of p(GRS) is the p-value of GRS. GRS. 参见:参见:参见:参见:A test of t
15、he A test of the efficiency of a given efficiency of a given portfolio p11461147portfolio p11461147.LSV double-sort portfoliosLSV 认为:用两个会计指标对股票进行分类会更为准确的区分强势股和压力股,进而产生更大的平均收益spread。因为有股价的会计比率会倾向于相关。笔者于是按照sales-rank和BE/ME,E/P or C/P的方式进行了33的分类。表4也表明,sales-rank确实增加average return spread.表4summary stati
16、stics for excess returns(in percent) on the LSV equal-weight double-sort portfolios:7/63-12/93,366 monthsSales-rank高低.表53 factor regressions for monthly excess returns(in perent) on the LSV equal-weight double-sort portfolios:7/63-12/93,366 months 3 3因素模型较好的解释了因素模型较好的解释了因素模型较好的解释了因素模型较好的解释了LSV LSV d
17、ouble-sort double-sort 收益率问题。截收益率问题。截收益率问题。截收益率问题。截距项都不显著区别于距项都不显著区别于距项都不显著区别于距项都不显著区别于0.GRS0.GRS检检检检验也支持了验也支持了验也支持了验也支持了3 3因素回归模型截因素回归模型截因素回归模型截因素回归模型截距项为零的推断。最小的距项为零的推断。最小的距项为零的推断。最小的距项为零的推断。最小的p p值值值值为为为为0.284.0.284.Portfolios formed on past returnWhen portfolios are formed on long-term(35 years)
18、,past losers tend to be future winners(reversal).When portfolios are formed on short-term(upto one year), past losers tend to be future losers(continuation).表6average monthly excess returns(in percent) on equal-weight NYSE deciles formed monthly based on continuously compounded past returns低高.表73fac
19、tor regressions for monthly excess returns(in percent) on equal-weight NYSE portfolios formed on past returns: 7/63-12/93,366 months 显著显著不显著3因素模型不能较好的解释基于短期历史收益所建立的组合收益率延续情况。截距项显著不为0,且有符号。.Exploring 3-factor modelsMerton(1973): 提出了intertemporal 优化下的3-fund 定理。一般情况下,会使用value-weight market和另外两个投资者考虑对冲的
20、因素所构建的MMV组合(multifactor-minimum-variance:given expected returns and sensitivities to the state-variables, they have the smallest possible return variances). 但实际上,任意3个MMV 组合都可以。FF说:如果三因素模型能较好的描述平均收益问题,那M、S、B、H、L就比较接近MMV组合。进而笔者就检验两个推论:1. M、S、B、H、L中的任意3个应该提供平均收益的类似描述;2. 其中一个的excess return 可以被其他3个完美描述。下
21、图表就是去掉B的表现情况。因为B与M高度相关(S相对好些)。.第一个不懂Equation 1 is still legitimate 3-factor risk-return relation as long as the two components of SMB and the two components of HML are MMV. .表8regression to explain monthly excess returns(in percent) on M,S,L,H,SMB and HML:7/63-12/93,366 months不显著显著.表9summary of inte
22、rcepts from one-factor CAPM Excess-return regressions on different versions of the 3-factor ICAPM regressions: 7/63-12/93,366 months笔者说:我的3因素跟其他的MMV proxy基本上无差异。可是接下来,就是他们RM-RF,SMB和HML要更少的相关。.表10Average monthly Excess returns(in percent) and correlations of excess returns for MMV proxies: 7/63-12/93,366 months其中一个结论:Low、high领域内的个体更容易相关。连average excess return 都比较接近。问题2:MMV proxies constructed from the LSV sales-rank sort, or from long-term past returns ,cannot successfully replace L and H in tests of the 3-factor。在哪体现出来的?.