use productivity log using productivity.log, replace tsset time * generate ln of outcomes gen rwagel=ln(rwage) gen prodl=ln(productivity) * run a model of productivity * on just a time trend * see how well it fits reg prodl time * output predicted value predict prodl_pred *output predicted values to csv * file to graph in pretty graph outsheet time year prodl prodl_pred using predict_value.csv, comma reg rwagel time prodl estat dwatson * test for random walk * first, generate lags gen rwagel1=rwagel[_n-1] gen prodl1=prodl[_n-1] * regress values on lags * test that coef on lag equals1 reg rwagel time rwagel1 test rwagel1=1 reg prodl time prodl1 test prodl1=1 * now difference data gen drwagel=rwagel-rwagel1 gen dprodl=prodl-prodl1 * run regression of diff on diff reg drwagel time dprodl estat dwatson prais drwagel time dprodl, corc twostep log close