Equal Profits, Unequal One-Year Persistence
Overview
This completed study asks whether equal current profit has different one-year persistence depending on its accounting composition. It was independently reconstructed, corrected, executed, and documented with explicit stopping rules for claims not supported by the data.
Research Question
Holding current profit before tax constant, is other profit less persistent than reported business-operation profit at the one-year horizon?
Economic or Technical Motivation
Aggregate earnings can combine operating components with items that reverse more quickly. Testing the components separately can reveal a timing pattern that a single profit number conceals.
Data
The primary common sample contains 7,355 firm-years and 1,197 firms. Accounting validation covers 12,074 usable observations in 2016–2024: 99.925% reconcile PBT with business-operation profit plus other profit within 1%.
Methodology
The analysis uses firm and year fixed effects, exact common-sample intersections, actual reported component comparisons, wild-cluster bootstrap inference, Holm adjustment, multi-horizon models, tail and influential-firm sensitivity, System GMM diagnostics, and FE-aware expanding-window forecasting.
Main Findings
Conditional other profit predicts lower next-year PBT (-0.3001, SE 0.1153, p=0.0094) and net income (-0.2687, SE 0.1063, p=0.0116). Both survive Holm adjustment and wild-cluster bootstrap inference. Reported business-operation profit is more persistent than other profit in raw and common-trimmed earnings models. The association is approximately zero at years two and three, indicating a one-year reversal rather than medium-term underperformance.
Robustness, Validation, or Model Assessment
The result survives multiple-inference correction and sequential exclusion of firms with the largest other-profit observations, but its magnitude and precision change under alternative tail treatments. The net-income GMM specification passes the reported AR(2) and Hansen diagnostics; it remains secondary and is not treated as causal identification.
Tools and Technologies
Stata 17, Python, firm fixed effects, wild-cluster bootstrap, Holm correction, System GMM diagnostics, and expanding-window validation are directly evidenced.
Limitations
The result is predictive and tail-sensitive, not causal. It does not support claims about fraud, manipulation, operating cash flow, or a general medium-term persistence mechanism.
Deliverables
- One-command Stata pipeline and complete logs
- Corrected results and decision reports
- Accounting and common-sample audits
- Robustness tables and six figures