2019年系列讲座课件14.170ProgrammingforEconomists.ppt

* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * Monte Carlo in Stata, con’t set more off On UNIX, this will keep the buffer from “locking” set matsize 1000 Sets default matrix size matrix Bvals = J(`B, 1, 0) Creates `B’-by-1 matrix drop _all Unlike “clear”, this only drops the data (NOT matrices!)? quietly set obs 200 Suppresses output qui regress y x cons, nocons “qui” is abbreviation; nocons means constant not included matrix betas = e(b) e() stores the return values from regression; e(b) is betas matrix Bvals[`b,1] = betas[1,1] syntax to set matrix values qui testparm x performs a Wald test to see if “x” is statistically significant qui regress y x cons , robust nocons uses “robust” standard errors svmat Bvals writes out matrix as a data column Monte Carlo in Stata, con’t OLS “by hand” clear set obs 10 set seed 14170 gen x1 = invnorm(uniform()) gen x2 = invnorm(uniform()) gen y = 1 + x1 + x2 + 0.1 * invnorm(uniform()) gen cons = 1 mkmat x1 x2 cons, matrix(X) mkmat y, matrix(y) matrix list X matrix list y matrix beta_ols = invsym(X*X) * (X*y) matrix e_hat = y - X * beta_ols matrix V = (e_hat * e_hat) * invsym(X*X) / (rowsof(X) - colsof(X)) matrix beta_se = (vecdiag(V)) local rows = rowsof(V) forvalues i = 1/`rows { matrix beta_se[`i,1] = sqrt(beta_se[`i,1]) } matrix ols_results = [beta_ols, beta_se] matrix list ols_results reg y x1 x2 OLS “by hand” clear set obs 100000 set seed 14170 gen x1 = invnorm(uniform()) gen x2 = invnorm(uniform()) gen y = 1 + x1 + x2 + 0.1 * invnorm(uniform()) gen cons = 1 mkmat x1 x2 cons, matrix(X) mkmat y, matrix(y) matrix list X matrix list y matrix beta_ols = invsym(X*X) * (X*y) matrix e_hat = y - X * beta_ols matrix V = (e_hat * e_hat) * invsym(X*X) / (rowsof(X) - colsof(X)) matrix beta_se = (vecdiag(V)) local rows = rowsof(V) forvalues i = 1/`rows { matrix beta_se[`i,1] = sqrt(beta_se[`i,1]) } matrix ols_results = [beta_ols, beta_se] matrix list ols_results reg y x1 x