I have a wide panel with multiple decades of T.
In my habit formation model, my dependent variable is alcohol consumption, and the coefficients of interest are on past and future consumption. My problem is while I have alcohol prices/taxes by US states but my panel only classifies individuals by the four regions, giving measurement errors and biases on the coefficients of interest that I cannot sign--on top of the endogeneity of the key regressors along with serial correlation.
I'm aware that the prices/taxes would be a proxy for the true prices at which each individual consumer buys his/her alcohol anyway, but I'm worried this would exacerbate the mismeasurement too much--especially as I plan to use FE. Should I proceed by weighting the state-level prices/taxes to proxy the regions and finding instruments for the prices/taxes? Are there better solutions? Thanks very much in advance.
