Quality of earnings analysis

The 10-step full workup. Budget ~3 hours per company. It's the protocol for capstone Part A, and your permanent template after that.

What are the steps in a quality of earnings analysis?

1. Pull and spread (30 min). Put five years of statements into your spread template. For the US, take 10-Ks from EDGAR. For India, take annual reports from the company or BSE. A screener.in export is fine for speed, but tie the totals back to the filing. Spread standalone and consolidated for Indian names, and note the gap.
2. Ratio panel (20 min). The Phase-2 panel: margin stack, DuPont, ROIC/ROCE, DSO/DIO/DPO and CCC, CFO/EBITDA, FCF/PAT, capex/D&A, dep/gross block, other income/PBT, leverage and coverage. Run it on a five-year trend plus two peers, as benchmarking taught you.
3. Residue pass (20 min). This is the 20-minute pass from the earnings-games residue table. Read the cash flow statement first, then the balance-sheet deltas against revenue, then the four notes (revenue recognition, RPT, provisions roll-forward, contingents). Finish with the Ctrl-F list: "related party," "pledge," "contingent," "restated," "resigned," "one-time," "capitalised," "guarantee".
4. Quant screens (20 min). Beneish M-Score (8 variables, flag > −1.78), Sloan accruals by both methods, Altman Z/Z′/Z″ (the right variant for the firm type, and skip it for financials), and Piotroski F. Compute them all. Don't eyeball. Note each screen's known failure modes beside it.
5. Sector overlay (10 min). Work out the working-capital sign the business model implies, and the sector's own flags. Businesses that collect in advance carry no DSO. Lenders get the ALM and divergence panel. Retailers get inventory and rebate questions.
6. Findings register (20 min). Every anomaly gets one row: finding → evidence (page-referenced to the filing) → family (one of the seven moves / CF game / meta) → severity (L1–L4) → the innocent explanation → the test that would tell the two apart.
7. Core-earnings bridge (20 min). Go from reported PAT to your estimate of sustainable owner earnings. Strip out one-offs and the carry from other income. Re-expense aggressive capitalisation. Normalise provisions and estimates. Haircut revenue that doesn't convince you. Show every adjustment as a line.
8. Management-credibility read (15 min). Set three years of chairman and MD letters, and one earnings call, against the numbers delivered. Were promises kept? Did the metrics switch when they got inconvenient? Watch for the meet-or-beat streak and metrics that migrate. Then the governance basics: board composition, audit-committee activity, auditor tenure, audit fees, pledge, how often RPTs come up.
9. Verdict memo (30 min). Two pages, Minto-style, conclusion first. Give the earnings-quality grade (A–F), the three findings that matter with their severity, the gap between core and reported in numbers, what would change your mind, and the tripwires you'll monitor. Use the language protocol throughout.
10. Calibration log (5 min). A journal entry: your confidence, the strongest innocent explanation you might be wrong about, and the date you'll re-check.

How do you spot earnings manipulation in a timed exam?

The timed-exam technique (capstone Part B; also the app's timed mode). You get ~6–8 minutes per vignette. (1) Read the numbers before the prose. Compute the one or two ratios the vignette lets you compute. (2) Find the residue. Which balance-sheet line is absorbing the story? (3) Name the family, and the test that would confirm it. (4) Assign severity. If the right call is "L1, plausible, innocent," say clean.

The exam scores false positives because the job does.

Under time pressure, people slip into narrative reasoning: "this feels like Manpasand". The fix is mechanical. Go ratio → residue → family → severity, every time, in that order.