Inferenc on factor model with GMM

23 Views Asked by At

I'm considering inference of the factor loading in the classical factor analysis.

There is the model of classical factor analysis: suppose there are $T$ i.i.d samples from the data generating process $Y = \Lambda F + \varepsilon, Var(\varepsilon) = \Sigma$, which is diagonal, and $Var(F) = I_r$, where $r$ is a fixed known constant that is typically much smaller than the sample size $T$. Classical factor analysis solves $Var(Y) = \Lambda\Lambda^T + \Sigma$ to identify and estimate the parameters.

Now I wonder if we can use the generalized method of moments (GMM) to inference on the loading parameter $\Lambda$? In Anderson et al (1956), they say the theoretical form of the variance of $\Lambda$ is complicated, but perhaps numerically estimating the variance by GMM is possible. Does anyone know about something with this? Appreciate any directions or discussions!