Persistent Earnings–Cash Flow Misalignment
Overview
This completed study classifies annual firm observations into four earnings–cash-flow states and follows their transitions. The public result is descriptive: it documents prevalence, resolution, persistence, and loss-state movement without labeling negative cash realization as fraud or distress.
Research Question
How often does positive net income with negative operating cash flow resolve, persist, or transition to an earnings-loss state, and how does observed spell duration relate to those destinations?
Economic or Technical Motivation
Accrual earnings and operating cash flow can diverge for ordinary timing, growth, or working-capital reasons. A state-transition framework preserves this distinction while measuring whether repeated misalignment is empirically persistent.
Data
The validated panel contains 8,885 firm-years from Vietnamese listed nonfinancial firms during 2016–2024 and 7,488 consecutive transitions. Positive earnings with negative CFO accounts for 2,061 observations. The analysis identifies 1,399 spells and separately audits missing-CFO selection.
Methodology
The project constructs four states, measures spell duration, estimates registered multinomial transition models, tests first-order versus duration-augmented memory, runs CRE and FE checks, applies inverse-probability weighting for observed CFO selection, tests materiality thresholds, and performs expanding-window temporal validation.
Main Findings
Positive earnings with negative CFO represents 23.2% of valid firm-years. Among 1,803 next-year transitions from this state, 53.4% move to cash-realized profit, 36.7% persist, and 9.9% move to an earnings-loss state. In-sample predicted persistence rises from 34.6% at duration one to 61.0% at duration four or more.
Robustness, Validation, or Model Assessment
Materiality thresholds and IPW preserve the descriptive duration pattern. However, excluding left-censored spells removes the headline duration coefficient, and the duration model does not improve out-of-time log loss over simpler benchmarks. The portfolio therefore reports transition frequencies as the primary verified result, not a superior semi-Markov forecast.
Tools and Technologies
Stata 17, Python, multinomial transition models, spell analysis, inverse-probability weighting, materiality thresholds, and expanding-window evaluation are directly evidenced.
Limitations
Annual observations do not identify exact within-year transition timing. Long spells are sparse and left-censored, and the working-capital mechanism is not supported. The state definitions do not imply fraud, manipulation, or distress.
Deliverables
- Fully executed one-command analysis pipeline
- Nine figures and machine-readable result tables
- State, spell, selection, and calibration audits
- Final results report and decision memo