I'm running a multiple regression model in Stata for my economics thesis and keep getting significant heteroskedasticity with the Breusch-Pagan test, even after trying robust standard errors. What's the most reliable way to correct this without losing interpretability of my coefficients?

I'm analyzing panel data on household income and education levels across different regions from 2000 to 2020. When I use OLS with fixed effects, the Breusch-Pagan test shows p < 0.01 for heteroskedasticity. I've applied robust standard errors (vce(robust) in Stata), but the confidence intervals widen so much that some key variables become insignificant. I've also tried log-transforming variables, but it didn't fully resolve the issue. I'm worried about model misspecification and need a practical solution to proceed with my analysis.