The Review of Financial Studies/v 13 n 1 2000 is an N(0,/vector conditional on time t-1 information and the model being well specified.For each test and most other tests below we use the generalized method of moments [Hansen(1982)]to test moment implications of a well-specified model,which are of the general form E[y以L-]=0, with V a vector stochastic process.The resulting test statistic has an asymptotic chi-square distribution with degrees of freedom equal to the dimension of V.The use of estimated residuals and the size of our sample may imply that the actual small sample distribution of the test statistics is no longer a chi-square distribution.Monte Carlo results in Bekaert and Harvey (1997)suggest that the small sample distribution of the tests may have more mass in the right tail so that we overreject at the asymptotic critical values. The conditional mean test,MEAN,sets -0 j=1,2,3;i=M,1,2,3. MEAN tests the serial correlation properties of the standardized re- siduals and is done for each portfolio separately and jointly for all portfolios. For the conditional variance tests,VAR,we introduce the variable qi=z-1 and we let Vi- Qiu'Qi-j j=1,2,3;i=M,1,2,3. Again the test is done separately for the different portfolios and jointly for all portfolios.Finally,to test the conditional covariance specifica- tion,consider the variable W:= i,M,-1i=1,2,3. OiM.t We let W Wa·W-j j=1,2,3 for each portfolio i and all portfolios jointly. 1.4.2 Testing the CAPM assumption.The MEAN test partially tests the CAPM assumption.If other risks are priced,the mean of the residual may not be zero.However,this test may not be powerful enough to detect particular deviations from the CAPM and we provide 16The Reiew of Financial Studies 13 n 1 2000 is an NŽ . 0, I vector conditional on time t  1 information and the model being well specified. For each test and most other tests below we use the generalized method of moments Hansen 1982 to test moment
Ž . implications of a well-specified model, which are of the general form EVI
 0, t t1 with Vt a vector stochastic process. The resulting test statistic has an asymptotic chi-square distribution with degrees of freedom equal to the dimension of V . The use of estimated residuals and the size of our t sample may imply that the actual small sample distribution of the test statistics is no longer a chi-square distribution. Monte Carlo results in Bekaert and Harvey 1997 suggest that the small sample distribution of Ž . the tests may have more mass in the right tail so that we overreject at the asymptotic critical values. The conditional mean test, MEAN, sets zit V  j  1,2,3; i  M,1,2,3. t z  z it itj MEAN tests the serial correlation properties of the standardized re￾siduals and is done for each portfolio separately and jointly for all portfolios. For the conditional variance tests, VAR, we introduce the variable q  z 2  1 and we let i,t i, t qit V  j  1,2,3; i  M,1,2,3. t qit it  q j Again the test is done separately for the different portfolios and jointly for all portfolios. Finally, to test the conditional covariance specifica￾tion, consider the variable i, t M , t W   1 i  1,2,3. i, t i M , t We let Wit V  j  1,2,3; t Wit it  W j for each portfolio i and all portfolios jointly. 1.4.2 Testing the CAPM assumption. The MEAN test partially tests the CAPM assumption. If other risks are priced, the mean of the residual may not be zero. However, this test may not be powerful enough to detect particular deviations from the CAPM and we provide 16