A loop of cross-sectional regressions for calculating abnormal accruals in Stata

I write a loop of cross-sectional regressions for calculating abnormal accruals. This program can be easily modified and replaced with Jones, modified Jones, or Dechow and Dichev model.

I add detailed comments in the program to help you prepare the input file.

set more off
cd "D:\Dropbox"

use funda_full_regdata, clear
//This is the input file. Download required Compustat data and then
//compute all dependent variable and independent variable (e.g., total
//accruals scaled by total assets (ta), delta sales scaled by total
//asset (dsale), PP&E scaled by total assets (ppe).

keep if fyear>=2003 & fyear<=2014

* Remove obs without required data
foreach v in ta dsale ppe {
  drop if `v'==.
}

* Handle outliers for each industry year
ge trunc=0
local vlist "ta dsale ppe"
egen groupid=group(fyear sic2)
sum groupid, meanonly
local num=r(max)
forvalues i=1/`num' {
  foreach v in `vlist' {
    _pctile `v' if groupid==`i', p(1 99)
    replace trunc=1 if groupid==`i' & (`v'<r(r1) | `v'>r(r2)) //drop top and bottom 1%. You can do winsorize as well
  }
}
drop if trunc==1
drop trunc groupid

* Require at least 20 obs for each estimation
bysort groupid: egen obsnum=count(datadate)
keep if obsnum>=20
drop groupid
egen groupid=group(fyear sic2)
sum groupid, meanonly
local num=r(max)

* Run cross-sectional regressions
ge da=.
ge regobs=.
ge df=.
ge r2a=.

foreach x in dsale ppe _cons {
  ge `x'_b=.
  ge `x'_se=.
}

forvalues i=1/`num' {
  regress ta dsale ppe if groupid==`i'
  predict resid if groupid==`i', residual
  replace da=resid if groupid==`i'
  drop resid
  replace regobs=e(N) if groupid==`i'
  replace df=e(df_r) if groupid==`i'
  replace r2a=e(r2_a) if groupid==`i'
  foreach x in dsale _cons {
    replace `x'_b=_b[`x'] if groupid==`i'
    replace `x'_se=_se[`x'] if groupid==`i'
  }
}

save da, replace
//This is the output file. It includes:
//DA - discretionary accruals (residuals)
//Regression coefficients such as dsale_b, ppe_b
//Standard error for each coefficient such as dsale_se, ppe_se
//Some regression stats such as number of obs (regobs), (degree of
//freedom (df), and adjusted R squared (r2a)

set more off

cd "D:\Dropbox"

use funda_full_regdata, clear

//This is the input file. Download required Compustat data and then

//compute all dependent variable and independent variable (e.g., total

//accruals scaled by total assets (ta), delta sales scaled by total

//asset (dsale), PP&E scaled by total assets (ppe).

keep if fyear>=2003 & fyear<=2014

* Remove obs without required data

foreach v in ta dsale ppe {

drop if `v'==.

}

* Handle outliers for each industry year

ge trunc=0

local vlist "ta dsale ppe"

egen groupid=group(fyear sic2)

sum groupid, meanonly

local num=r(max)

forvalues i=1/`num' {

foreach v in `vlist' {

_pctile `v' if groupid==`i', p(1 99)

replace trunc=1 if groupid==`i' & (`v'<r(r1) | `v'>r(r2)) //drop top and bottom 1%. You can do winsorize as well

}

drop if trunc==1

drop trunc groupid

* Require at least 20 obs for each estimation

bysort groupid: egen obsnum=count(datadate)

keep if obsnum>=20

drop groupid

egen groupid=group(fyear sic2)

sum groupid, meanonly

local num=r(max)

* Run cross-sectional regressions

ge da=.

ge regobs=.

ge df=.

ge r2a=.

foreach x in dsale ppe _cons {

ge `x'_b=.

ge `x'_se=.

}

forvalues i=1/`num' {

regress ta dsale ppe if groupid==`i'

predict resid if groupid==`i', residual

replace da=resid if groupid==`i'

drop resid

replace regobs=e(N) if groupid==`i'

replace df=e(df_r) if groupid==`i'

replace r2a=e(r2_a) if groupid==`i'

foreach x in dsale _cons {

replace `x'_b=_b[`x'] if groupid==`i'

replace `x'_se=_se[`x'] if groupid==`i'

}

save da, replace

//This is the output file. It includes:

//DA - discretionary accruals (residuals)

//Regression coefficients such as dsale_b, ppe_b

//Standard error for each coefficient such as dsale_se, ppe_se

//Some regression stats such as number of obs (regobs), (degree of

//freedom (df), and adjusted R squared (r2a)

This entry was posted in Stata. Bookmark the permalink.

7 Responses to A loop of cross-sectional regressions for calculating abnormal accruals in Stata

Renee Adebile says:

October 10, 2016 at 6:50 pm

I am using your codes to calculate DA. However, I get ” Invalid Syntax”. Can you please help me figure out what I do wrong.
this the codes I used;
forvalues i=1/num' { regress ta x1 x2 x3 x4 x5 if groupid==i’
predict resid if groupid==i', residuals replace da=resid if groupid==i’
drop resid
replace regobs=e(N) if groupid==i' replace df=e(df_r) if groupid==i’
replace r2a=e(r2_a) if groupid==i' foreach x in x1 x2 x3 x4 x5 _cons { replacex’_b=_b[x'] if groupid==i’
replace x'_se=_se[x’] if groupid==`i’
}
}

- Kai Chen says:
  
  October 10, 2016 at 9:52 pm
  
  Your codes look good to me. If you can send me the entire codes, I may be able to pinpoint the error.
  
Renee Adebile says:

October 11, 2016 at 10:14 am

those are the entire codes I used:
* Remove obs without required data
foreach v in ta x1 x2 x3 x4 x5 {
drop if v'==. }
* Handle outliers for each industry year ge trunc=0 local vlist "ta x1 x2 x3 x4 x5" egen groupid=group(fyear sic_2) sum groupid, meanonly local num=r(max) forvalues i=1/num’ {
foreach v in vlist' { _pctilev’ if groupid==i', p(1 99) replace trunc=1 if groupid==i’ & (v'r(r2)) //drop top and bottom 1%. You can do winsorize as well } } drop if trunc==1 drop trunc groupid
* Require at least 20 obs for each estimation bysort groupid: egen obsnum=count(fyear) keep if obsnum>=20 drop groupid egen groupid=group(fyear sic_2) sum groupid, meanonly local num=r(max) * Run cross-sectional regressions ge da=. ge regobs=. ge df=. ge r2a=.
foreach x in x1 x2 x3 x4 x5 _cons { gex’_b=.
ge x'_se=. }
forvalues i=1/num’ {
regress ta x1 x2 x3 x4 x5 if groupid==i' predict resid if groupid==i’, residuals
replace da=resid if groupid==i' drop resid replace regobs=e(N) if groupid==i’
replace df=e(df_r) if groupid==i' replace r2a=e(r2_a) if groupid==i’
foreach x in x1 x2 x3 x4 x5 _cons {
replace x'_b=_b[x’] if groupid==i' replacex’_se=_se[x']if groupid==i’
}
}

save da, replace

Note that when I run all the entire codes, I get da regobs r2a x1_b for the first regression. and then I get ” invalid syntax”.

Suleman says:

August 30, 2017 at 5:02 pm

I am working on discretionary accrual model. I am having problem with calculating the Total accrual and PPE in Stata . I used this command :gen TAC =(NI-CFO)/lagged TA and Gross property plant and equipment using the following command :gen var name =PPE/lagged TA. How do I go about it please

- Kai Chen says:
  
  September 20, 2017 at 3:52 pm
  
  Do you mean you don’t know how to calculate lagged TA? If this is your problem, check the command tsset in Stata documentation. Stata is very powerful in calculating lagged value or changes in value.
  
Razoo says:

February 13, 2018 at 6:34 pm

Hello Kai,

Could you please provide us with the modified jones model’s code?

Thanks a lot,

Hui says:

November 22, 2018 at 6:35 pm

Hello Dr. Chen,

I just want to say thank for sharing the helpful codes and instructions. I just replicated Jones and Mod Jones by following your codes, and it really helps me to write it in a more efficient way.

Thank you once again.
Hui