Predicting material misstatements
Dechow F-Score
The Dechow F-Score is a 7-variable logistic regression model developed by Professor Patricia Dechow and colleagues at UC Berkeley in 2011. It estimates the probability that a company's financial statements contain material misstatements — not just earnings manipulation, but any accounting error or fraud that would be significant enough to restate.
What Is the F-Score?
While the Beneish M-Score focuses on earnings manipulation specifically, the Dechow F-Score casts a wider net. It was built by studying 494 companies that received SEC Accounting and Auditing Enforcement Releases (AAERs) between 1982 and 2005 — companies the SEC determined had materially misstated their financial reports. The model identifies patterns in financial data that distinguish misstating firms from non-misstating firms.
How It Differs from M-Score
- *M-Score detects intentional manipulation of earnings. F-Score detects material misstatements, which includes both intentional fraud and unintentional errors.
- *M-Score uses year-over-year index ratios (DSRI, GMI, etc.). F-Score uses change variables relative to average assets.
- *M-Score outputs a score where higher = more suspicious. F-Score outputs a predicted probability (0% to 100%).
- *M-Score directly affects our A-F grade (it's Check F1). F-Score is a parallel indicator that does NOT affect the grade.
The 7 Variables (Model 1)
Measures the change in working capital, non-current operating assets, and financial assets (the Richardson et al. 2005 decomposition). Captures a broader definition of accruals than simple net income minus cash flow. Higher accruals = higher misstatement risk.
Measures the change in accounts receivable scaled by average total assets. A large increase in receivables relative to assets is a common feature of misstating firms — it may indicate fictitious revenue.
Measures the change in inventory scaled by average total assets. Inventory build-up can indicate channel stuffing, obsolescence concealment, or capitalized costs that should have been expensed.
(Total Assets - PP&E - Cash) / Total Assets. Measures the proportion of assets that are "soft" — harder to verify, easier to manipulate. Includes goodwill, intangibles, deferred charges, and other items that rely heavily on management estimates.
Measures the percentage change in cash sales (revenue minus change in receivables). Declining cash sales while reported revenue grows is a strong warning signal.
Measures the year-over-year change in ROA. Declining profitability creates incentive to misstate — management under performance pressure is more likely to cut corners on accounting.
Binary variable: 1 if the company issued long-term debt or equity during the year, 0 otherwise. Companies raising capital have incentive to present favorable financials to investors and underwriters.
Model 1 Formula
PV = -7.893 + 0.790×rsst_acc + 2.518×ch_rec + 1.191×ch_inv + 1.979×soft_assets + 0.171×ch_cs - 0.932×ch_roa + 1.029×issuePV is the predicted value from the logistic regression. The predicted probability is calculated as: P = 1 / (1 + e^(-PV)). Note the large coefficient on ch_rec (2.518) and soft_assets (1.979) — receivables changes and soft asset ratios are the strongest predictors.
Probability Interpretation
The financial statements do not show patterns associated with misstatement.
Some concerning patterns are present. Worth investigating further.
The financial statements show patterns strongly associated with material misstatement in the academic study.
Why It's a Parallel Indicator
The F-Score does NOT affect our A through F grade. We include it as additional context for three reasons:
- 1.The model has a higher false positive rate than the M-Score, especially for companies with large intangible assets or recent acquisitions (high soft_assets).
- 2.The academic study's base rate of misstatement was approximately 0.6% of firm-years, making even a "high probability" result statistically uncommon in practice.
- 3.We already capture many of the same signals through our 18-point checks (AR vs Revenue, Inventory vs COGS, Accruals Ratio). Adding F-Score to the grade would double-count these signals.
Think of it as a second opinion from a different doctor. If the F-Score flags something that our 18 checks missed, it's worth investigating even if it doesn't change the grade.
Academic Source
Dechow, P.M., Ge, W., Larson, C.R., and Sloan, R.G. (2011). "Predicting Material Accounting Misstatements." Contemporary Accounting Research, 28(1), 17-82.
The paper presents three models of increasing complexity. We use Model 1, which relies only on financial statement data and does not require market data or discretionary accrual estimates.