I have balanced panel data on N=10 industries over T=20 years. For each (industry, year) i have a) the # of patents filings and b) R&D expenditures. I want to estimate the following model:
However, using the xtserial program rejects the null hypothesis of no serial correlation. Hence, it seems that the only option available to me is to use feasible generalized least squares (xtxtgls), with the panel(correlated) option (to allow for cross section correlation) eventually with corr(ar1) or corr(psar1) (to allow for within panel autocorrelation). Am I right on my diagnostic ?
In particular, can/should I use a standard OLS but with a clustering of the standard errors by industry instead ?
Many thanks