Error using mvregress : Covariance is not positive-definite. AND Error in varm/estimate : sigma(solve,solve), logL] = mvregress(D, R, 'covtype' , 'full' , 'varformat', 'beta' , ...

5 次查看（过去 30 天）

Spyridon Kariofylis 2020-7-9

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/562481-error-using-mvregress-covariance-is-not-positive-definite-and-error-in-varm-estimate-sigma-solv

Hello there

I am trying to calculate the betas and alphas from the CAPM model, for 62 stocks, using a var model but i get this error message. I tested the model for less stocks and i found out that it works for 32 or less. I tried to exclude the 32th or 33th stock but it didnt make any differance. i also checked if there are any negative values at the cov matrix but there were not. Any suggestions?

X = GSPC-rf;
[T,n] = size(Y1); %Y1(36x62) = matrix with monthly returns of 62 stocks for 36 months
Mdl = varm(n,0);
[EstMdl,EstCoeffSEMdl] = estimate(Mdl,Y1,'X',X);
results = summarize(EstMdl);
% looking at variance - covariance matrix
R = cov(Y1);
len=length(R);
for i=1:len
    if R(i,i)<0
        disp(i)
        
    end
end