Lieven posted on Thursday, October 14, 2004 - 3:13 am
I have 1056 variables (equity portfolios) with the number of observations ranging from 10 to 95 (monthly dollar returns). For some variables the covariance is missing because the observations may not overlap in time. These missing covariances can be set to zero. There are 74 factors. 1 world-factor: each variable can have an unrestricted loading on it; 39 country-factors: each porfolio belongs to a specific country and can have an unrestricted loading to only one country-factor, the rest zero; and 34 industry-factors: each portfolio (variable) belongs to a specific industry and can have an unrestricted loading on only one industry-factor, the rest zero. I have initial values, coming from a two-step regression methode, for the factors and the loadings. Our aim is twofold: (1) getting estimates for the loadings and the factors in one-shot (2) testing whether the 3 loadings per variable are equal. A reliable estimate of the fit of the model is less important. Is it possible to perform such an analysis?