I write a loop of cross-sectional regressions for calculating abnormal accruals. This program can be easily modified and replaced with\u00a0Jones, modified Jones, or Dechow and Dichev model.<\/p>\n
I add detailed comments in the program to help you prepare the input file.<\/p>\n
set more off\r\ncd \"D:\\Dropbox\"\r\n\r\nuse funda_full_regdata, clear\r\n\/\/This is the input file. Download required Compustat data and then\r\n\/\/compute all dependent variable and independent variable (e.g., total\r\n\/\/accruals scaled by total assets (ta), delta sales scaled by total\r\n\/\/asset (dsale), PP&E scaled by total assets (ppe).\r\n\r\nkeep if fyear>=2003 & fyear<=2014\r\n\r\n* Remove obs without required data\r\nforeach v in ta dsale ppe {\r\n drop if `v'==.\r\n}\r\n\r\n* Handle outliers for each industry year\r\nge trunc=0\r\nlocal vlist \"ta dsale ppe\"\r\negen groupid=group(fyear sic2)\r\nsum groupid, meanonly\r\nlocal num=r(max)\r\nforvalues i=1\/`num' {\r\n foreach v in `vlist' {\r\n _pctile `v' if groupid==`i', p(1 99)\r\n replace trunc=1 if groupid==`i' & (`v'<r(r1) | `v'>r(r2)) \/\/drop top and bottom 1%. You can do winsorize as well\r\n }\r\n}\r\ndrop if trunc==1\r\ndrop trunc groupid\r\n\r\n* Require at least 20 obs for each estimation\r\nbysort groupid: egen obsnum=count(datadate)\r\nkeep if obsnum>=20\r\ndrop groupid\r\negen groupid=group(fyear sic2)\r\nsum groupid, meanonly\r\nlocal num=r(max)\r\n\r\n* Run cross-sectional regressions\r\nge da=.\r\nge regobs=.\r\nge df=.\r\nge r2a=.\r\n\r\nforeach x in dsale ppe _cons {\r\n ge `x'_b=.\r\n ge `x'_se=.\r\n}\r\n\r\nforvalues i=1\/`num' {\r\n regress ta dsale ppe if groupid==`i'\r\n predict resid if groupid==`i', residual\r\n replace da=resid if groupid==`i'\r\n drop resid\r\n replace regobs=e(N) if groupid==`i'\r\n replace df=e(df_r) if groupid==`i'\r\n replace r2a=e(r2_a) if groupid==`i'\r\n foreach x in dsale _cons {\r\n replace `x'_b=_b[`x'] if groupid==`i'\r\n replace `x'_se=_se[`x'] if groupid==`i'\r\n }\r\n}\r\n\r\nsave da, replace\r\n\/\/This is the output file. It includes:\r\n\/\/DA - discretionary accruals (residuals)\r\n\/\/Regression coefficients such as dsale_b, ppe_b\r\n\/\/Standard error for each coefficient such as dsale_se, ppe_se\r\n\/\/Some regression stats such as number of obs (regobs), (degree of\r\n\/\/freedom (df), and adjusted R squared (r2a)<\/pre>\n","protected":false},"excerpt":{"rendered":"I write a loop of cross-sectional regressions for calculating abnormal accruals. This program can be easily modified and replaced with\u00a0Jones, modified Jones, or Dechow and Dichev model. I add detailed comments in the program to help you prepare the input … Continue reading