stata-workflow

name	stata-workflow
description	Use when working with Stata .do files for empirical finance/accounting research. Covers panel data, reghdfe, esttab, merge workflows, variable construction, and debugging. Triggers - reghdfe, xtreg, xtset, esttab, eststo, merge, append, panel data, fixed effects, clustered standard errors.
allowed-tools	["Read","Bash","Glob","Grep"]

Stata Workflow for Empirical Finance/Accounting Research

Overview

Expert guidance on Stata programming for empirical research in finance and accounting. Prioritizes reproducibility, efficiency, and publication-quality output.

Panel Data Setup

Standard panel setup pattern:

use "data.dta", clear
xtset firmid yearq          // or: xtset gvkey fyear

Always verify panel structure:

• xtdescribe to check gaps
• duplicates report firmid yearq before xtset
• tsset for time-series operators (L. F. D. S.)

Regression Specifications

reghdfe (preferred for high-dimensional FE)

reghdfe depvar treatment controls, absorb(firmid yearq) cluster(firmid)

Key options:

• absorb(firmid yearq) — firm and time FE
• absorb(firmid industry#yearq) — firm FE + industry-by-year FE
• cluster(firmid) — cluster at firm level
• cluster(firmid yearq) — two-way clustering

Interaction terms

reghdfe y c.x1##c.x2 controls, absorb(fe) cluster(cl)
// Or manually:
gen x1_x2 = x1 * x2
reghdfe y x1 x2 x1_x2 controls, absorb(fe) cluster(cl)

Instrumental variables

ivreghdfe depvar controls (endogvar = instrument), absorb(fe) cluster(cl)
// First stage F-stat reported automatically

Output Formatting

esttab (preferred)

eststo clear
eststo: reghdfe y x1 controls1, absorb(firmid yearq) cluster(firmid)
eststo: reghdfe y x1 controls2, absorb(firmid yearq) cluster(firmid)
eststo: reghdfe y x1 controls3, absorb(firmid industry#yearq) cluster(firmid)

esttab using "table.tex", replace ///
    b(%9.3f) se(%9.3f) ///
    star(* 0.10 ** 0.05 *** 0.01) ///
    r2 N ///
    title("Main Results") ///
    mtitles("(1)" "(2)" "(3)") ///
    drop(_cons) ///
    booktabs ///
    label

Adding custom rows (FE indicators, etc.)

esttab using "table.tex", replace ///
    indicate("Firm FE = *firmid*" "Year FE = *yearq*") ///
    // OR manually:
    stats(N r2 firmfe yearfe, ///
        labels("Observations" "R-squared" "Firm FE" "Year FE") ///
        fmt(%9,0gc %9.3f 0 0))

Variable Construction

Winsorizing

// At 1st and 99th percentiles
winsor2 varlist, cuts(1 99) replace
// Or by group:
bysort yearq: winsor2 varlist, cuts(1 99) replace

Lags and leads (requires xtset)

gen lag_x = L.x          // one-period lag
gen lead_x = F.x         // one-period lead
gen delta_x = D.x        // first difference
gen lag2_x = L2.x        // two-period lag

Scaled variables

gen roa = ni / at                     // return on assets
gen bm = ceq / (csho * prcc_f)       // book-to-market
gen leverage = (dltt + dlc) / at      // leverage
gen size = ln(csho * prcc_f)          // log market cap

Merge Workflows

CRSP-Compustat merge pattern

// Start with Compustat
use "compustat_funda.dta", clear
// Merge linking table
merge m:1 gvkey using "ccmxpf_lnkhist.dta"
keep if _merge == 3
drop _merge
// Keep valid links
keep if linktype == "LU" | linktype == "LC"
keep if linkprim == "P" | linkprim == "C"
// Merge CRSP
merge 1:1 permno date using "crsp_msf.dta"

Always check merges

tab _merge
// Document merge rates in comments
// _merge==1: master only, _merge==2: using only, _merge==3: matched

Common Debugging

• r(111): variable not found — check spelling, check if data loaded
• r(198): variable already defined — drop or rename first
• r(459): not sorted — sort varlist or bysort varlist:
• r(2000): no observations — check if conditions, check merge
• type mismatch in merge: tostring/destring to align types
• singleton observations dropped: normal with reghdfe, document count

Best Practices

• Always use version 17 (or your version) at top of .do file
• Set set more off and clear at start (NOT clear all if called from master)
• Use compress before saving to reduce file size
• Use label variable for all constructed variables
• Log output: log using "filename.log", replace
• Use tempfile and tempvar for intermediate objects
• Comment liberally with // and /// for line continuation
• Use assert statements to verify data integrity