Dear Statlisters,
I am estimating a demand system with 4 goods, usingthe aidsills command, using Stata 14.

The estimations run smooth when I run the demand for the whole sample (sample period of seven years, weekly observations of sales, monthly or yearly observations of sociodemographic variables).

However, when I try to estimate the model for each year, separately, it only "works" for some of the years. For a majority of the years, the output consists of an empty table (i.e., no coefficient estimates, nor standard errors) along with the text: "Warning:Variance matrix is non-symmetric or highly singular".

I have seen threads on this, for example;
https://www.statalist.org/forums/for...ighly-singular
https://www.stata.com/statalist/arch.../msg00980.html

where the problem is mainly stated as a "sparse indicator problem".

However, in my data, there are no indicator variables (1/0) as such, but I do have variables with very little variation, though. For example, with year-wise estimations, income will only vary across regions (6 regions), but not across time, since I only have access to yearly income for each region, and not on weekly or monthly level, which is the level of the sales data.

My question is thus: Can the problem of "to little variation" be a problem with the aidsills here, even if there are no indicator variables, but continuous variables with slight variation?

Best regards,
Hanna