The Deep Dive, Days 1–2

Phase 8 · Week 67 · ~20 hours

Where it sits. A teardown gives you an opinion. A deep dive gives you conviction. M8.01 taught you to assemble the source stack and skim a 10-K or Indian annual report in the right order in 45–60 minutes; M8.02 compressed the whole system into the 1–2 hour teardown: twelve questions, an eight-line output, a variant question to carry away. The first two days of the three-day deep dive are the part where you stop pattern-matching and start rebuilding the business from primary filings with your own hands. Day 1 is history and the machine: you read five-to-ten years of reports oldest-to-newest (watching, like a time-traveller, how each year's promises aged against the next year's results), you build the ten-year spreadsheet history that every other analysis leans on, you compute incremental ROIC, the truest compounder test, and you read every note in the accounts. Day 2 is industry, competitors, and a model: you map the industry with Mauboussin's supplier-firm-customer chain and locate the profit pool, you run Greenwald's market-share-stability test, you benchmark two or three peers on identical KPIs, you build a 3-statement-lite model driven by three-to-five assumptions into base/bull/bear, and you run a reverse DCF to decode what the price is silently assuming and judge whether that story is too optimistic or too pessimistic. That last hour is where edge actually lives. Day 3 (M8.04) is the qualitative edge, run on the same company you carry through these two days: scuttlebutt, incentives, the earnings-call time-series, the steelmanned bear, and the written thesis. M8.05 then certifies the whole system under a stopwatch.

Suggested hour budget. Core content: 8h · worked examples: 2h · practice set: 4h · applied mini-project (the Day-1/2 workbook): 4h · flashcards + mastery check: 2h. In real deep dives this is two working days, ~14–16 focused hours; the module compresses the method so the days run clean.

App link. Days 1–2 power the Rapid-Analysis System branch's deep-dive workbench: the ten-year history builder (paste filing lines, the grid computes margins, ROIC, ROIIC, and flags the Ind AS 116 and tax-cut breaks), the share-stability calculator (shared with M4.05), the peer-benchmark normalizer, the 3-statement-lite sandbox (three drivers in, base/bull/bear out), and the reverse-DCF decoder (price in; implied growth, duration, and required return out, wired to the M3.05 engine). The mini-project workbook is the app's Day-1/2 case template.

Learning objectives

By the end of these two days you can:

1. Read 5–10 years of a company's reports oldest-to-newest and run the management-credibility "time-travel" test, extracting a scored track record of promises made versus results delivered, before you know how the story turned out.
2. Build a ten-year spreadsheet history from primary filings using the standard row template (segment revenue, the margin stack, ROIC/ROCE, invested capital, CFO, capex, FCF, net debt, share count, dividends + buybacks), and apply the required normalizations (the Ind AS 116 lease break, the FY20 India tax-rate break, exceptionals, re-segmentation, and acquisitions) so the series is comparable end to end.
3. Compute incremental ROIC (ΔNOPAT ÷ Δinvested capital) over multi-year windows, reconcile it to growth via g = RR × ROIC, handle a mid-window acquisition, and read it as the compounder test that headline ROIC hides.
4. Read every major note (revenue recognition, related-party, leases, contingencies, segments, ESOP) and state precisely what an analyst extracts from each and which red flag each can hide.
5. Map an industry with Mauboussin's supplier→firm→customer chain, locate the profit pool link by link, and run Greenwald's market-share-stability test on a real share table against the ~2%/yr threshold, cross-checked against incumbent returns.
6. Benchmark a company against 2–3 peers on identical, normalized KPIs and read the comparison as a story, not a scoreboard.
7. Build a 3-statement-lite model driven by three-to-five assumptions and generate internally-consistent base/bull/bear cases: fewer rows than M3.08's full build, the same discipline.
8. Run a reverse DCF to extract the growth, duration, and required return the current price implies, translate the implied numbers into countable operational units, and judge, against base rates, whether the market's story is too optimistic or too pessimistic, ending with the single variant question to resolve on Day 3.

Prerequisites & connections

Builds on. This is a synthesis: it assumes almost the whole program and puts it on a stopwatch-adjacent workbench.

• M8.01 & M8.02: the source stack, fast filing navigation, and the twelve teardown questions. The deep dive is the teardown with the scaffolding removed and the numbers rebuilt by hand: teardown question 7 (the economics) becomes Day 1's spreadsheet; questions 5–6 (competition, moat) become Day 2's industry map and share table; question 11 (what's priced in) becomes Day 2's reverse DCF.
• M1.02–M1.10. You cannot build a history you cannot read; the notes-reading section (§4.7) is the M1.10 filing-anatomy skill aimed at extraction.
• M2.03 is the backbone. NOPAT, the operating/financing split of invested capital, headline ROIC, and incremental ROIC were built there by hand (the DMart worked example). The deep dive fills the "incremental-ROIC row for every company you ever study" that M2.03 promised you would. M2.04 (FCF), M2.05 (benchmarking discipline: time-series, cross-sectional, common-size, normalize-before-comparing) are used wholesale on Day 1 and in §4.11.
• M3.03 (WACC, the hurdle every ROIC and the reverse DCF is measured against: ~12% post-tax INR, ~8–9% USD as of mid-2026), M3.04–M3.05 (FCFF forecasting, terminal value, and the reverse DCF; §4.13 is that machine run for real), M3.07 (capital allocation: the five uses of cash you read off the history), M3.08 (the full 3-statement model; §4.12's lite model is its faster cousin, same articulation, three drivers instead of thirty).
• M4.03 (Porter's five forces, the value chain, the profit pool) and M4.05 (Greenwald's share-stability test, Mauboussin's Measuring-the-Moat four-pass protocol, the CAP, and Damodaran's narrative-and-numbers). §4.8–§4.10 are those modules under a two-day clock, run on the carried company.

Feeds forward. M8.04 takes the same company (we carry Asian Paints) into Day 3: the variant question these two days produce is does dealer-and-painter captivity survive Birla Opus's ₹10,000-crore assault?, and that is exactly what Day 3's scuttlebutt (dealer channel checks), incentive reading (the promoter families' alignment), earnings-call time-series, and steelmanned bear are built to resolve. M8.05 certifies the two days under time pressure. Phase 10's two capstone deep dives (M10.01 US, M10.02 India) are the same two days run unaided. Competencies C3 (understand a business model cold), C5 (value three ways; the reverse DCF is the third way), and C10 (the signature skill) all route through here.

The one-sentence version. Day 1 rebuilds the machine from a decade of filings and asks is it a good business and did management tell the truth about it?; Day 2 places the machine in its industry, benchmarks it, models it forward on a handful of honest drivers, and then inverts the price to ask the only question that pays: is the future the market has already bought more or less demanding than the evidence supports?

Core teaching content

4.1 What the two days produce, and the workbook that holds it

A deep dive is not "reading more." It is producing a small number of durable artifacts that together answer the five questions (§2.1 of the map) at a depth no teardown reaches. By the end of Day 2 your deep-dive workbook (the reusable template in artifacts/playbook-2-3-day-deep-dive.md) has these tabs filled:

1. History: the ten-year spreadsheet (§4.3–4.5), with normalizations logged.
2. Incremental ROIC: the compounder table (§4.6).
3. Notes extract: one line of extracted signal per major note (§4.7).
4. Industry map: supplier→firm→customer chain, profit pool, five-forces scores (§4.8–4.9).
5. Share stability: the league table and the instability metric (§4.10).
6. Peer benchmark: 2–3 peers on identical normalized KPIs (§4.11).
7. Model: 3-statement-lite, base/bull/bear (§4.12).
8. Reverse DCF: the implied story, the three-lever diagnosis, and the verdict (§4.13).
9. The variant question: the one thing that decides the thesis, handed to Day 3.

The discipline that makes this fast: you do it by hand once, then you have a template forever. The workbook does not change shape company to company; only the numbers do. After thirty deep dives you fill it in your sleep, which is the entire point of Phase 8: to turn a two-day marathon into a repeatable craft.

The carried specimen. We run all of Days 1–2 end-to-end on Asian Paints (India), the decorative-paints leader you already met as the "structurally blessed industry" in M4.03 and the share-stability exemplar in M4.05. It is the ideal deep-dive teacher: a genuine multi-decade compounder with very high returns on capital, a clean balance sheet, a legible business, and, as of 2024, a live existential test, the entry of Grasim's Birla Opus with a committed ~₹10,000 crore. That last fact is what turns this from a hagiography into an analysis: the whole of these two days bends toward one question the market is being forced to re-price in real time.

We use Sherwin-Williams (US) as the recurring cross-market parallel: same product, opposite moat architecture (company-owned stores and a pro-contractor lock instead of a 1.5-lakh-dealer network). Seeing the same industry solved two structural ways is how you learn to see structure at all.

All Asian Paints and Sherwin-Williams figures here are illustrative, rounded, and modeled on ~FY24/FY25 scale for teaching. They are order-of-magnitude correct but are NOT filing-exact. Pull the actual consolidated annual reports (asianpaints.com investor relations; screener.in) and 10-Ks (SEC EDGAR) and rebuild. The point is the method, and doing it yourself is the skill being certified. Nothing here is a recommendation; both companies are teaching specimens.

4.2 Reading 5–10 years oldest-to-newest: the time-travel credibility test

Everyone reads the latest annual report. The deep-dive move is to read the old ones first, in order, forward through time. It is the single highest-signal qualitative exercise in the whole three days, because it is the only way to judge management honestly.

Here is the logic. If you read the FY24 report first, you already know how the story ended, and hindsight poisons your judgment of everything management ever said. You will forgive the failed promises you can see paid off and condemn the prudent caution that looks timid in retrospect. But if you read FY15's letter and guidance first, form an expectation, and only then turn to FY16's results, you are doing what the market did in real time: you are grading the forecast before you see the outcome. Do this for ten years and you have built something no screener can give you: a calibrated read of management's credibility, promise by promise.

The mechanics. Open the chairman's letter / MD&A of the oldest report in your window. Extract every forward statement into a two-column log:

Then advance one year and fill the last two columns from the next report's results, before reading that year's new promises. Score each: kept / partly kept / missed / quietly dropped. The "quietly dropped" column is the most valuable. Managements announce ambitions loudly and abandon them silently, and the analyst who tracks the graveyard of dropped targets knows more than the analyst who reads only live ones.

What you are looking for (the four tells):

• Guidance calibration. Do stated targets (revenue growth, margin, capacity, market share, a new segment's break-even) come true at roughly the promised size and date? Chronic over-promising is a governance signal that outlives any single CEO.
• Ownership of misses. When a year goes wrong, does the letter name the miss and the cause, or does it discover that "macro headwinds" and "unprecedented input inflation" arrived precisely when results did? Straight talk in a bad year is the cheapest, rarest credibility signal there is.
• Consistency of the story. Does the strategic narrative hold across a decade, or does it lurch: "we are a focused paints company" (FY16) → "we are a home-décor ecosystem" (FY22) → "we are refocusing on the core" (FY25)? Strategy tourism shows up here years before it shows up in the numbers.
• Capital-allocation promises vs deeds. "We will maintain capital discipline" is words; the §4.5 history is deeds. Read them against each other.

Asian Paints, the time-travel read (illustrative synthesis). Walk FY15→FY25 and a coherent, largely-honest arc appears: a decade of decorative-paints leadership defended and extended, margins guided conservatively and generally delivered, capacity added roughly on the schedule promised (the Mysuru, Visakhapatnam and later plants came up broadly when and where the letters said they would), dividends raised steadily. The credibility ledger is strong, which is why the stock earned its premium. But the honest reader also logs the tells that matter now. The home-décor / "Beautiful Homes" ecosystem ambition (kitchens, baths, décor, services) was sold from ~FY18 as the next growth engine and has consumed capital and management attention while remaining a small, thin-margin drag: a promise still "partly kept" a half-decade on. And the FY23–FY25 letters shift in tone from serene market leadership to explicit acknowledgment of "heightened competitive intensity," the euphemism whose arrival you can date precisely to Birla Opus. None of this is damning; all of it is evidence you carry forward. The time-travel test's output is not a verdict on management. It is a scored, dated ledger that makes the Day-3 management assessment (M8.04) an audit rather than an impression.

You will finish this in 90–120 minutes for a legible company using investor presentations and MD&As (you do not re-read whole reports; you hunt the forward statements and the following year's scorecard).

It is worth every minute: it is the only part of the deep dive that a competitor with the same spreadsheet cannot replicate without doing the same reading.

4.9 Porter's five forces, at deep-dive speed

You do not re-teach yourself Porter on Day 2; you score him from the evidence you already gathered, in about ten minutes (M4.03 under a clock). Indian decorative paints, scored 1 (benign) to 5 (brutal):

Greenwald's collapse (M4.05, and the second named scaffold; see §4.10): of these five, one decides everything, the threat of entry. Rivalry, buyer power, and supplier power only determine how a protected pool is split; whether the pool stays protected at all is the entry question. So the deep dive's entire competitive analysis funnels to a single test: are the barriers real, and are they holding? That test has a clean empirical instrument.

4.10 The market-share-stability test: Greenwald's barriers detector, run live

This is Greenwald's Competition Demystified scaffold (the third named framework), and it is the cleanest empirical instrument in strategy: if barriers to entry exist, entrants fail and incumbents keep their customers, so market shares barely move; if barriers are absent, share is bought and lost continuously. You do not have to argue the moat. The industry has been running the experiment for decades and publishing the results. The procedure (M4.05 §4.5): define the relevant market, build a share table over 8–15 years, compute average absolute share change per firm per year, apply the thresholds, and cross-check against incumbent returns.

The share table (Indian organized decorative paints, illustrative %; source your own from Indigo Paints' DRHP industry chapter, company presentations, and CRISIL/industry notes):

Average absolute movement, top four firms, FY16→FY24 (8 years): |0| + |2| + |2| + |0| = 4 points ÷ 4 firms = 1.0 ÷ 8 years = 0.125 points per firm per year, an order of magnitude inside the ~0.5-points/firm/year "strong barriers" line, and far below the ~2-points/year "weak barriers" threshold. Cross-check with returns (Greenwald step 5): stable shares must coincide with above-hurdle returns to confirm barriers rather than a low-return stalemate, and they do (Asian Paints ROIC ~40%, Berger ~20%+, all well above the ~12% hurdle, §4.11). Verdict: powerful, real barriers, nameable with M4.05's vocabulary as scale-economies-in-distribution interlocked with dealer/painter captivity.

The one riser, Berger (+2 in a decade), is the superior competitor pattern: a strong operator gaining slowly in a high-barrier market, not evidence of weak barriers.

Now the live experiment, and why this specimen was chosen. The table above is a backward-looking instrument: it proves the barriers existed. Whether they hold against a determined giant is a forward question, and in 2024 that question stopped being hypothetical. Birla Opus (Grasim/Aditya Birla) entered with a committed ~₹10,000 crore, six plants, and an explicit strategy of buying dealer shelf-space and share (free tinting machines, aggressive dealer margins, an attempt to purchase exactly the captivity the incumbents built over decades). Industry margins compressed within a year (FY25, visible in §4.5). The deep-dive analyst's job is not to know how this ends. It is to watch the FY25–FY28 share tables the way a cardiologist watches an ECG, and to recognize that the reverse DCF (§4.13) is pricing the market's bet on the outcome. Greenwald's test has handed you the exact tripwire: if the top-four instability metric jumps above ~2 points/firm/year, or if any entrant reaches ~5% share, the barrier is breaking and the CAP the price is paying for is too long.

That tripwire is the thesis, and it is what Day 3's dealer channel checks (M8.04) go to verify on the ground.

4.11 Benchmarking against peers on identical KPIs

No ratio means anything alone (M2.05). You benchmark the carried company against 2–3 close peers on identical, normalized KPIs: same consolidation basis, same lease treatment, same normalized tax, and ideally segment-matched (comparing Asian Paints' decorative-heavy mix to Kansai's auto/industrial-heavy mix without adjusting is a category error). Asian Paints vs its two closest listed decorative peers (illustrative, ~FY24 scale):

Read the table as a story, not a scoreboard. (1) Asian Paints' superiority is real and structural, and it is in the numbers: the best margins and roughly double the ROIC of a good #2, which is exactly what the scale-in-distribution + captivity moat predicts (twice the dealer reach spreads the same brand and logistics cost over more volume, per M4.05's local-density arithmetic). (2) But the incremental returns have converged: Berger's recent ROIIC matches Asian Paints' ~18%, and Berger is the share gainer, so the sharper competitor to watch is not the new entrant alone but the disciplined #2 that has been quietly closing the gap for a decade. (3) Kansai's structurally lower returns are not bad management; they are mix: a heavier auto/industrial exposure (lower-margin, OEM-buyer-power business), which is why segment-matching matters and why you never rank paint companies on blended margins without decomposing them.

The benchmark confirms the Day-1 verdict (great business) and refines the risk (two credible attackers on the pool, not one).

US contrast, in one line: Sherwin-Williams solves the same industry with the opposite architecture: ~4,000+ company-owned stores and a pro-contractor lock instead of a third-party dealer network, giving it even tighter control of the counter and pro relationship (higher structural margins than distributed rival PPG), which is why the US pool concentrates in Sherwin the way the Indian pool concentrates in Asian Paints: same profit pool, two different moats built to hold it. (Worked Example 1 rebuilds Sherwin's incremental ROIC, including the Valspar-goodwill normalization the acquisition forces.)

4.12 The 3-statement-lite model: three-to-five drivers into base/bull/bear

Now you build the machine that turns your judgment into a value. This is not M3.08's full 30-row, revolver-and-circularity model, which is a Phase 3 / capstone artifact. The deep-dive model is lite: the same articulation and the same discipline (M3.08's inputs/calcs/outputs separation, no hardcodes, the g = RR × ROIC consistency, TV constraints), but driven by three-to-five assumptions and carrying only the rows that change the answer. The craft is choosing the right three-to-five drivers, the ones the value is actually sensitive to, and admitting everything else is detail.

Asian Paints' value drivers (the 3–5 that matter):

1. Revenue growth, decomposed into volume (repaint demand × penetration × share) and price/mix. The master driver, and the one Birla Opus most directly attacks (share).
2. EBIT margin, a function of input costs (TiO2/crude, largely out of the firm's control) and competitive intensity (dealer-margin and A&P spend, entirely a function of the Birla Opus fight). This is the second Birla Opus lever.
3. Sales-to-capital / incremental capital intensity: how much reinvestment each rupee of new revenue needs (drives FCF and ties growth to ROIC via RR = g/ROIC).
4. Tax, fixed at the normalized 25.2%.
5. The fade: WACC (~11.5% INR, M3.03), terminal growth (≤ ~5–6%), and terminal ROIC (the CAP made numeric).

Everything else either nets out or is small enough to hold flat: exact working-capital days, the international and industrial segments' wobble, the home-décor drag. That is the lite model's bargain: fewer rows, honestly chosen, beat thirty rows falsely precise. You then flex the two live drivers (growth and margin, both hostages to the competitive outcome) into three internally-consistent scenarios:

• Bear. The barrier partly breaks: Birla Opus and Berger take share, a multi-year price/margin war. Revenue growth ~6%, EBIT margin settles ~15%, ROIC fades faster.
• Base. The barrier mostly holds at a cost: some permanent margin give-up to defend share, growth normalizes. Revenue growth ~9–10%, EBIT margin ~17%.
• Bull. The entrant fails like entrants before it (M4.05's telecom-vs-paints contrast: buying captivity is far harder than buying capacity): the industry grows, Asian Paints defends share and margin. Revenue growth ~12%, EBIT margin ~19%.

Each scenario runs the same lite machine to a value per share (fully tabulated in Worked Example 2). The output is not a point. It is a range with named scenarios, and a map from the one uncertainty that matters (does the barrier hold?) to the value. That mapping is what makes the next step, reading the price, devastating rather than academic.

4.13 The reverse DCF: decoding the price, where edge lives

Here is the hour that pays for the whole deep dive. Everything so far runs the machine forward, assumptions in and value out, and the forward run has a fatal social weakness: "my model says ₹1,900, the market says ₹2,500, therefore the market is wrong" is arrogance dressed as analysis. So flip the machine (M3.05 §4.12, expectations investing, Rappaport & Mauboussin): the price is not a mistake to correct; it is a message to decode. Hold the machine fixed, make the market price the output you must reproduce, and solve for the assumptions that hit it. The question stops being "what do I think it's worth?" (about your crystal ball) and becomes "is the future the price requires more, or less, demanding than the evidence supports?" (about base rates, which you can actually answer).

Edge lives in that gap.

Set-up (illustrative, on the FY2024-25-scale figures above; not a current quote). Price ~₹2,500; ~95.9 cr shares → market cap ~₹2,40,000 cr; net cash ~₹4,200 cr → EV ~₹2,35,800 cr. Normalized mid-cycle NOPAT ~₹4,500 cr (between FY24's 4,974 and FY25's 3,777) → EV/NOPAT ≈ 52×. WACC 11.5%; defensible ROIC ~25–30%.

Step 1: the 60-second steady-state inversion (the teardown-speed read; M3.05's quick tool). Invert EV/NOPAT₁ = (1 − g/ROIC)/(WACC − g):

(1 − g/0.30)/(0.115 − g) = 52
1 − 3.333g = 5.98 − 52g
48.67g = 4.98  →  g ≈ 10.2%  — forever

The price, read as a steady state, implies ~10% perpetual growth at 30% ROIC. But ~10% is roughly nominal GDP forever, a growth rate no company can sustain in perpetuity without becoming the entire economy (the M3.05 cap: terminal g ≤ ~5–6% for INR). So the price is not a steady state; it necessarily embeds a long super-normal phase. In one breath: 52× cannot be a perpetuity under any defensible parameter, so the price is a bet on duration, a long and undisturbed competitive-advantage period, not on next year's growth.

Step 2: translate the duration bet into countable units. To reproduce ₹2,35,800 cr EV with a defensible terminal (g_T = 5%, ROIC_T = 18%, WACC 11.5%), the explicit high-growth stage must run at roughly mid-to-high-teens revenue growth for ~a decade-plus before fading. Convert to something you can count: ~13–15% for ten years takes revenue from ~₹34,000 cr to ~₹1,20,000–1,35,000 cr by FY35, roughly 3.5–4× today, on a ~₹18% EBIT margin held throughout, while a ₹10,000-crore competitor and a share-gaining #2 press on the pool. That is a testable claim, not a vibe.

Step 3: the alternative single-lever readings (the three-lever diagnosis; M3.05 §4.12). A demanding price is explained by exactly one of three levers, or by a blend, and naming which is the whole skill:

• (i) Growth/margins beyond base rates: the ~13–15%-for-a-decade read above. The market sees an operational future your base case doesn't.
• (ii) Duration: grant the delivered ~10–12% growth but for 20+ years rather than ten: a very long CAP. This, not near-term growth, is the usual content of India's premium-multiple regime, and it is a claim about the moat, which §4.10 just taught you to audit.
• (iii) Required return: hold growth at a realistic long-run ~8% and solve for r: EV = FCFF₁/(r − g) → 235,800 = (4,500 × 1.08 × 0.68)/(r − 0.08) → r ≈ 9.4%, below the 11.5% WACC. The marginal buyer is accepting a ~9% return for a bond-like quality franchise (common late in bull markets and in crowded "quality" baskets, and it mean-reverts when rates or sentiment turn).

Step 4: judge against base rates, and deliver the verdict. Not "does the story feel plausible" (every story feels plausible), but "of all companies that ever stood here, what fraction delivered this?" Indian nominal GDP runs ~10–11%; the paints market has historically grown ~1.5× GDP but is maturing; and the >13%-revenue-CAGR-for-a-decade club, for a company already the ₹35,000-crore leader, is a short and famous list (build the base rate yourself: export 10-year revenue CAGRs for the NSE-500 from screener.in and count; Mauboussin's Base Rate Book does the US equivalent). Whichever lever you read, the price assumes the moat holds, essentially undisturbed, for a very long time, and/or that buyers keep accepting a low return. These are precisely the two things the live event threatens: Birla Opus threatens the CAP (lever ii and the margin half of lever i), and a higher-rate / more-competitive world threatens the required return (lever iii).

What the method shows on these figures: the implied story and the live risk sit on the same lever. The price pays for a long, undisturbed CAP at the exact moment the CAP is being attacked by the best-funded entrant in the industry's history, so the question to research is whether the moat outlasts that attack. That is the finding, and it is evenhanded: the reverse DCF does not say whether to own the shares. It tells you exactly what an owner is betting on, that dealer-and-painter captivity defeats ₹10,000 crore the way it defeated every prior challenger, and exactly what someone who stays out is betting on, that this challenger is different. It converts a shouting match ("wonderful company!" / "absurd multiple!") into a single measurable proposition. It is the method's reading of illustrative figures, not a view on the stock today.

And it isolates the one variant question the price hinges on: does the captivity hold? Everything on Day 3 is built to resolve exactly that: dealer channel checks, the incentive read, the earnings-call time-series, the steelmanned bear.

Process notes that keep the tool honest (M3.05): run the reverse DCF before you fall in love (it vaccinates you against plug-to-price); re-run it on every big move ("what expectation just re-priced?" beats "why is it down 9%?"); and remember its usual, humbling output is "fairly priced — pass," which is the tool telling you where you have no edge. Asian Paints is interesting precisely because it is not that usual case: a live, datable event is forcing the market's long-CAP assumption to be re-underwritten in real time, and you now hold the instrument that reads what is priced.

4.14 Damodaran's narrative + numbers, and the Day 1–2 synthesis

The last named scaffold ties the two days into one object. Damodaran's Narrative and Numbers insists that a valuation is neither a story nor a spreadsheet but a disciplined marriage: every strategic claim must become a number in the model, and every number in the model must trace back to a claim you can defend. A story with no numbers is a fairy tale (the promoter's investor deck); numbers with no story are false precision (the analyst's 30-tab model that no one believes). The deep dive is this marriage enacted.

Watch it close on Asian Paints. The narrative you built over two days, "a dominant, high-return decorative-paints franchise whose scale-in-distribution and dealer/painter captivity have frozen share and compounded capital at ~40% for a decade, now facing a maturing core, a sub-scale décor option, and the first credibly-funded assault on its moat in a generation", maps clause by clause to numbers: "dominant, high-return" → ROIC ~37% vs ~12% hurdle and the §4.11 peer gap; "frozen share" → the 0.125-points/firm/year stability metric; "compounded at ~40%" → Window-1 ROIIC; "maturing core, first real assault" → the halving to ~18% incremental ROIC, the FY25 revenue and margin dip, and the bear/base/bull margin fork; "moat under attack" → the reverse DCF's finding that the price pays for a long CAP the entrant threatens. The reverse DCF is the market's competing narrative decoded into numbers: a story of an undisturbed 20-year CAP and/or a 9% required return. Your job across the three days is to decide whose narrative the evidence supports.

Damodaran's rule keeps both sides honest: you may not tell the bull story without putting the 13% decade growth on the page where a base rate can shame it, and you may not tell the bear story without pricing the possibility that captivity holds.

What you can now say, after Days 1–2 (the five questions, answered to deep-dive depth):

1. What is it and how does it make money? A decorative-paints manufacturer that collects the industry's profit pool by controlling brand, distribution density, and the tinting counter (§4.5, §4.8).
2. Is it a good business? Emphatically: ~37% ROIC, ~25-point spread, net cash, decade-frozen share (§4.5, §4.6, §4.10), but with incremental returns compressing and a live threat to the moat.
3. Is it run and capitalized well? Provisionally yes: no dilution, disciplined payout, credible history (§4.2, §4.5), to be confirmed against incentives on Day 3.
4. What can kill it? Not death, but a permanent CAP-shortening if Birla Opus buys through the captivity; the halving incremental ROIC is the early tremor (§4.6, §4.10).
5. What is priced, and is it beatable? A long, undisturbed CAP and/or a bond-proxy return, priced at the moment a live entrant attacks the CAP; the work is testing whether the moat outlasts the attack (§4.13).

And you hold one variant question that a spreadsheet cannot answer: does dealer-and-painter captivity survive a ₹10,000-crore assault? That is the perfect handoff to Day 3 (M8.04), which leaves the filings for the field: channel checks with dealers, the promoter families' alignment and incentives, four-to-eight earnings calls read as a time-series through the entry, the steelmanned bear, and the one-page thesis. Days 1–2 have done their job. They have turned a famous name into a precise, falsifiable bet.

Common mistakes & how experts think differently

1. Reading newest-to-oldest, and killing the only honest management test. Beginners open the latest report because it's most "relevant." But once you know how the story ended, hindsight forgives every promise that happened to pay off and condemns every prudent caution. Experts read oldest-to-newest precisely to grade the forecast before seeing the outcome (§4.2). It is the only way to build a credibility ledger that isn't contaminated by knowing the answer.

1. Not defusing the Indian ten-year landmines: Ind AS 116 and the FY20 tax cut. The single most common way a well-meaning Indian history build lies: EBITDA margin "improves" in FY20 (it's just leases moving on-balance-sheet) and post-tax ROIC "steps up" in FY20 (it's just the 34.9%→25.2% tax cut). Experts normalize before they admire, holding one tax rate across the decade and one lease treatment, so the trend they see is operations, not accounting policy (§4.4, §4.6). "You will otherwise credit the finance minister's ROIC to the paint chemist."

1. Stopping at headline ROIC. Headline ROIC (40%) is the return on all capital, most of it old. The compounder question is what new capital earns. Experts always compute the two-window incremental-ROIC contrast (§4.6), and it is exactly where Asian Paints tells its real story (~40% legacy, ~18% recent). The screener will never show you this; you compute ΔNOPAT/ΔIC yourself or you miss the whole plot.

1. Single-year incremental ROIC, and un-lagged acquisition years. One year of ΔNOPAT ÷ ΔIC is noise (a capex year reads as "value destruction" while the plant is unfinished); a mid-window acquisition makes it meaningless (you charge organic ΔNOPAT against acquired capital; Worked Example 1). Experts use 3–5-year windows, lag where growth is fast, and evaluate acquisitions separately as capital-allocation decisions on their purchase price (M3.07).

1. Building thirty rows when five drivers decide the answer. False precision is not rigor. The deep-dive-lite model's craft (§4.12) is identifying the three-to-five drivers the value is actually sensitive to (for Asian Paints: growth, margin, capital intensity, and the fade) and holding everything else flat, honestly. A thirty-tab model no one can audit is less trustworthy than five drivers you can defend.

1. Running share-stability on the wrong market. Greenwald's test is meaningless on "global retail" when the fixed costs are regional, or on "paints" when the economics live in decorative specifically. Experts define the arena where the fixed costs actually sit before measuring share (§4.10, M4.05 §4.4), and they always add the profitability cross-check, because frozen shares at cost-of-capital returns are a stalemate, not a moat.

1. Benchmarking on un-normalized KPIs. Comparing Asian Paints' decorative-heavy margins to Kansai's auto/industrial-heavy margins, or a lease-capitalizing peer to one that isn't, or standalone to consolidated, produces a scoreboard that lies. Experts normalize consolidation, leases, tax, and segment mix before ranking (§4.11, M2.05), and they read the table as a story ("who is gaining and why") rather than a league table.

1. Treating the reverse DCF as a forward DCF (plug-to-price), or skipping it entirely. The two failure modes are opposite. Some quietly tune a forward model until it blesses the price (the M3.05 "plug-to-price" sin). Experts freeze assumptions before seeing the output, and if they want to start from price they do it honestly, as a reverse DCF. Others never invert at all and end up shouting "overvalued!" without knowing what is priced. The expert's reflex: let the market speak first (§4.13). Decode the implied growth, duration, and required return before the forward model hardens into a position, because knowing exactly what you're disagreeing with, in units, is the difference between a variant perception and a mood.

1. Confusing sell-in with end-demand, and skipping the notes. Revenue booked on dispatch to dealers is sell-in, not consumption; in a competitive shock it can be dealer-shelf-filling that will reverse (§4.7). And the related-party and contingency notes are where the value gets siphoned and the losses hide. The analyst who read the summary instead of the notes is the one surprised by the fraud everyone else's checklist caught.

1. Falling in love with a wonderful business and forgetting that price is a separate question. Asian Paints is a wonderful business — and that fact tells you nothing about whether ₹2,500 is a good price. Experts hold the two thoughts separately: quality is question 2 of the five; price-versus-quality is question 5, and a great business at a demanding price pricing an undisturbed CAP into a live moat-attack is exactly the setup where the crowd's admiration and the analyst's edge diverge.

Worked examples

Worked Example 1: Incremental ROIC across an acquisition: Sherwin-Williams & Valspar (US)

The problem the acquisition creates. In 2017 Sherwin-Williams bought Valspar for ~$11.3bn, largely debt-funded, adding ~$10bn+ of goodwill and acquired intangibles. This is the set-piece for why you never compute incremental ROIC blindly across a deal. Illustrative figures, ~FY16/FY18/FY24 scale; pull the 10-Ks from EDGAR.

US$m (illustrative)            FY16      FY18       FY24
Revenue                       11,856    17,534     23,100
EBIT                           1,755     2,244      3,696
NOPAT (normalized @ 21% tax)   1,386*    1,773      2,920
Invested capital, TOTAL        4,000    16,500     19,500
Invested capital, EX-goodwill  3,500     5,000      6,500

*FY16 NOPAT normalized to 21% for comparability, though the actual 2016 US rate was 35% (the TCJA break; normalize US series exactly as you normalize Indian ones for the FY20 cut).

Move 1: the naive cross-acquisition ROIIC, and why it's garbage.

FY16 → FY18 (spanning the deal):
ΔNOPAT = 1,773 − 1,386 = 387
ΔIC(total) = 16,500 − 4,000 = 12,500
naive ROIIC = 387 / 12,500 = 3.1%   ← "value destruction!"

It is not value destruction. The ΔIC is ~$12.5bn of acquired capital (Valspar's purchase price and goodwill), while ΔNOPAT reflects only a partial-year, integration-depressed Valspar. You have charged a barely-digested acquisition's first-year earnings against its entire purchase price. The number is meaningless. You cannot measure a compounder across the year it makes a transformational acquisition.

Move 2: evaluate the acquisition as an acquisition (M3.07), not as ROIIC. The right question for the $11.3bn is a capital-allocation question: what steady after-tax cash return did Valspar's earnings + synergies represent on the price paid? If integrated Valspar contributed on the order of ~$900m–1,000m of steady NOPAT plus synergies on ~$11.3bn, that is a ~8–9% initial cash return, growing: a fair, not spectacular, use of a large slug of capital, financed with cheap debt. Judge the deal on that; do not let it contaminate the operating-returns series.

Move 3: organic incremental ROIC, post-integration (both years already contain Valspar):

FY18 → FY24 (no new large deal in the window):
ΔNOPAT = 2,920 − 1,773 = 1,147
ΔIC(total) = 19,500 − 16,500 = 3,000
organic ROIIC = 1,147 / 3,000 = 38%

Now the number means something: post-integration, Sherwin's growth (store rollout, pricing, share) needed only ~$3bn of incremental capital to add ~$1.15bn of NOPAT. The store-network model is genuinely capital-light, and its incremental economics are excellent (~38%).

Move 4: the honest headline, and the cross-company lesson. Sherwin's ROIC with goodwill in FY24 is 2,920 / 19,500 ≈ 15%, a good business (well above the ~8% US hurdle), but a long way from Asian Paints' ~37%. The gap is not that Asian Paints is a "better business" line-for-line; it is that Asian Paints built its growth organically at ~40% incremental returns, while Sherwin bought a chunk of its growth with a goodwill-heavy acquisition, and shareholders paid that goodwill, so it belongs in the denominator of the return on the money actually spent. Two superb paint franchises; the ROIC gap is an allocation story (organic compounding vs transformational M&A), and you only see it by computing ROIC both ways and evaluating the deal on its own. This is the incremental-ROIC discipline (§4.6) doing exactly the work headline ROIC cannot.

Worked Example 2: Asian Paints, the 3-statement-lite model into base/bull/bear

We run the §4.12 lite model to three values. Mechanics: NOPAT₀ = ₹4,500 cr (normalized, mid-cycle); a 10-year explicit stage at scenario growth g1 and scenario stage-ROIC (so reinvestment RR = g1/ROIC and FCFF = NOPAT × (1 − g1/ROIC)); then a terminal stage at g_T = 5%, ROIC_T = 18% (RR_T = 27.8%); WACC 11.5%; net cash ₹4,200 cr; 95.9 cr shares. (Illustrative: the point is the structure and the gap, not the rupee.)

Worked once (base, so you can reproduce the row): FCFF₁ = 4,500 × 1.09 × (1 − 0.09/0.25) = 4,905 × 0.64 = ₹3,139 cr; growing-annuity PV over 10 yrs at (g 9%, r 11.5%) = 3,139 × [1 − (1.09/1.115)¹⁰]/(0.115 − 0.09) = 3,139 × 8.10 = ₹25,430 cr. NOPAT₁₀ = 4,500 × 1.09¹⁰ = ₹10,650 cr; TV = [10,650 × 1.05 × 0.722]/(0.115 − 0.05) = 8,079/0.065 = ₹124,300 cr; PV(TV) = 124,300/1.115¹⁰ = 124,300 × 0.3367 = ₹41,850 cr. EV ≈ ₹67,300 cr; per share = (67,300 + 4,200)/95.9 ≈ ₹745.

The lesson, and why this hands off to the reverse DCF. Every defensible 10-year scenario, including the bull, lands far below the market price of ~₹2,500. The beginner concludes "3× overvalued." The expert recognizes the M3.05 phenomenon: a 10-year-explicit forward model structurally understates a genuine long-CAP franchise, because it hard-stops the excess-return period at year 10 and fades to a mature terminal, so for a premium compounder, the gap between your forward range and the price is not (mainly) a mispricing signal, it is a measurement of how much competitive-advantage-period duration the market is granting beyond your explicit window. That is why, for a franchise like this, the reverse DCF (§4.13) is the more honest instrument. It solves directly for the implied duration and required return instead of assuming a 10-year cutoff. The forward model's job here is narrower but vital: it maps the one live uncertainty (does the barrier hold?) to a value fork (₹630 bear vs ₹890 bull), so that when Day 3's channel checks move your probability on that fork, you know exactly what it's worth.

Worked Example 3: A reverse-DCF verdict on a premium compounder (synthetic, clean)

Specimen (synthetic, for a clean answer): SnackCo, an Indian packaged-foods leader. EV = ₹90,000 cr; normalized NOPAT = ₹2,000 cr (so EV/NOPAT = 45×); WACC 11%; defensible ROIC ~28%; and, from the qualitative facts you'd have gathered on Days 1–2, ~40% category share held within ±1 point for 15 years (share-stability metric ≈ 0.13/yr → strong, proven barriers) with ROIC persistently ~28%.

Step 1: steady-state inversion. (1 − g/0.28)/(0.11 − g) = 45 → 1 − 3.571g = 4.95 − 45g → 41.43g = 3.95 → g ≈ 9.5% forever. Above the ~5.5% cap, so the price embeds a long super-normal phase: a duration bet, as always for a 45× franchise.

Step 2: the three levers. (i) Growth: ~9.5%-perpetual-equivalent ≈ ~13% for a decade then fade → revenue ~3.4× in ten years. (ii) Duration: grant the delivered ~10% growth for ~15 years, a long CAP. (iii) Required return: hold long-run growth at 7% (RR = 25%, payout 0.75) → 90,000 = (2,000 × 1.07 × 0.75)/(r − 0.07) → r − 0.07 = 1,605/90,000 = 0.0178 → r ≈ 8.8%.

Step 3: judge against the evidence, and deliver the verdict. Here is the contrast with Asian Paints that makes the method sing. SnackCo's implied story is a long, undisturbed CAP, and SnackCo's evidence is 15 years of ±1-point share stability with 28% ROIC and no credible entrant on the horizon. The evidence supports a long CAP. So the reverse DCF's honest verdict is "demanding but roughly fairly priced for a proven long-CAP franchise — no clear edge; pass". The price implies ~13%-for-a-decade or a ~15-year CAP or an ~8.8% required return, all at the optimistic-but-defensible end for a franchise this durable. The only variant worth holding is on lever (iii): if the ~8.8% implied return reflects a crowded "quality" bid that mean-reverts when rates rise, the stock de-rates without the business changing, so the play is patience for a better price, not a short.

The two readings side by side: the whole point of the tool. Same 45–52× multiple, different work: for Asian Paints the price pays for an undisturbed long CAP at the exact moment the best-funded entrant in the industry's history is attacking it, so the open question is the moat's durability against that live risk and the research goes there; for SnackCo the price pays for a long CAP that 15 years of frozen share and no entrant actually support, so the implied story is defensible, there is no edge, and you pass. The reverse DCF doesn't hand you a verdict; it hands you the precise proposition the price embeds, so your real work, auditing the moat's durability against the live evidence, has a target. That is expectations investing, and it is where the edge lives.

Worked Example 4: A normalized ten-year history and the two-window compounder test (India ₹)

The whole of Day 1 in miniature. GripWell (synthetic), an Indian adhesives-and-sealants leader. You have pasted the raw filing lines; before you trust a single trend you must normalize, then compute the incremental-ROIC contrast that is the compounder's real verdict (§4.4, §4.6). Three representative columns of the ten-year build (₹ cr, consolidated, exceptionals already stripped):

Step 1: normalize the tax, or you credit the finance minister. Compute NOPAT at one rate across the decade (25.2%, the §4.4 rule), never each year's actual rate:

NOPAT (normalized @ 25.2%):  FY15 = 800 × 0.748   = 598.4
                             FY20 = 1,400 × 0.748 = 1,047.2
                             FY25 = 2,300 × 0.748 = 1,720.4
FY25 headline ROIC = 1,720.4 / 8,800 = 19.6%

Using FY15's actual 34.9% would read FY15 NOPAT as only 800 × 0.651 = 520.8, artificially low, so the FY15→FY20 step would fold the September-2019 tax cut into what looks like operating progress.

Step 2: Window 1, the golden years FY15→FY20 (normalized):

ΔNOPAT = 1,047.2 − 598.4 = 448.8
ΔIC    = 4,200 − 2,500   = 1,700
ROIIC (5-yr) = 448.8 / 1,700 = 26.4%

Run the same window on actual tax and it reads 526.4 / 1,700 = 31.0%, or 4.6 points of phantom incremental return that is pure tax policy. This is the single most common way an Indian ten-year build overstates a compounder (§4.6).

Step 3: Window 2, the recent years FY20→FY25 (normalized):

ΔNOPAT = 1,720.4 − 1,047.2 = 673.2
ΔIC    = 8,800 − 4,200     = 4,600
ROIIC (5-yr) = 673.2 / 4,600 = 14.6%

The whole story in one contrast: headline ROIC is still a comfortable ~20%, but incremental returns have fallen from ~26% to ~15%. The marginal rupee still clears the ~12% hurdle (value is still created at the margin), but the direction is unmistakable, and headline ROIC hid it completely.

Step 4: the growth check. Cumulative normalized NOPAT over FY21–FY25 ≈ ₹7,000 cr, so RR = 4,600 / 7,000 = 66%. Then RR × ROIIC = 0.66 × 14.6% = 9.6% vs actual NOPAT CAGR FY20→FY25 = (1,720.4 / 1,047.2)^(1/5) − 1 = 10.4%. The identity roughly closes (the ~0.8-point gap is timing and margin drift; it is a consistency check, not a law of single years). Read: a still-good business whose reinvestment economics are cooling: exactly the profile that should slow reinvestment and lift its payout, and exactly the fingerprint §4.5's Asian Paints history left when its FY25 payout leapt to 82%.

Worked Example 5: Peer benchmark: why the higher-margin retailer is the worse business (US $)

Peer-map ratios, read as a story not a scoreboard (§4.11). Two synthetic US retailers benchmarked on identical, normalized KPIs (~FY24, $m; tax normalized to the post-TCJA 21% for both):

Step 1: the naive read (rank by margin) is wrong. DeptCo keeps $9 of EBIT per $100 of sales to WarehouseCo's $4, more than double. A beginner ranks DeptCo the better business. Now compute what the master idea (M0.01 §4.6; DuPont, M2.03) actually demands: return on capital.

Step 2: ROIC = NOPAT margin × capital turnover.

WarehouseCo: NOPAT = 2,400 × 0.79 = 1,896 → ROIC = 1,896 / 6,000  = 31.6%
             = NOPAT margin 3.16% × capital turnover (60,000/6,000)  10.0×
DeptCo:      NOPAT = 1,800 × 0.79 = 1,422 → ROIC = 1,422 / 12,000 = 11.9%
             = NOPAT margin 7.11% × capital turnover (20,000/12,000)  1.67×

WarehouseCo earns ~2.7× the ROIC on less than half the margin, because it spins its capital six times as fast. Thin margins and fat returns are the same business (the DMart/Costco lesson from M0.01). Against a ~8–9% US hurdle, WarehouseCo's 31.6% is a wide spread; DeptCo's 11.9% barely clears it.

Step 3: the incremental read confirms it. Over the last five years (normalized NOPAT, $m):

WarehouseCo: ΔNOPAT = 1,896 − 1,200 = 696;  ΔIC = 6,000 − 3,600   = 2,400 → ROIIC = 29.0%
DeptCo:      ΔNOPAT = 1,422 − 1,300 = 122;  ΔIC = 12,000 − 10,500 = 1,500 → ROIIC =  8.1%

WarehouseCo reinvests new capital at ~29% (still compounding); DeptCo's marginal dollar earns ~8%, right at the hurdle, so its growth is value-neutral. The story: identical KPIs, decomposed correctly, invert the naive ranking: the fast-turn, thin-margin machine is the far better business and the better compounder, and the fat margin was a trap. That is exactly what benchmarking on normalized KPIs (§4.11), read through margin × turnover, is for.

Practice set

Work each fully before reading the solution (map rule 3). Synthetic mini-companies (clean numbers) are labelled; treat them as drills. Hurdles: ~12% post-tax INR, ~8–9% USD (as of mid-2026).

P1 (guided): spot the accounting break. Synthetic DecorCo (₹ cr), adopted Ind AS 116 in FY20:

Identify the year of the standard-driven discontinuity and state what you must do before comparing FY18 margins to FY21 margins.

Solution. The ~3.5-point jump in EBITDA margin at FY20 coincides exactly with Ind AS 116 adoption: rent that sat in EBITDA as an expense moved below EBITDA (into depreciation of the ROU asset + lease interest), so margin rises with zero operating improvement (§4.4, the M1.08 trap). Before any cross-year comparison you must either (a) capitalize leases consistently across FY18–FY19 too (add imputed depreciation + interest to those years), or (b) compare on EBIT or EBITDAR, which the standard leaves largely undistorted. Never read the FY19→FY20 "margin improvement" as operating progress; check the lease note first.

P2 (guided): normalize a one-off. Synthetic MetalCo FY24: reported EBIT ₹800 cr includes a one-time gain of ₹120 cr on a land sale. Invested capital ₹5,000 cr; tax 25%. Compute clean ROIC and state how much the one-off flattered it.

Solution. Clean operating EBIT = 800 − 120 = ₹680 cr → NOPAT = 680 × 0.75 = ₹510 cr → clean ROIC = 510 / 5,000 = 10.2%. Naive ROIC (leaving the gain in) = 600 / 5,000 = 12.0%. The one-off flattered ROIC by 1.8 points, the difference between "clears the ~12% hurdle" and "misses it." Strip one-offs into their own row and compute returns on the recurring line (§4.4); and if MetalCo has a land-sale "exceptional" most years, it isn't exceptional.

P3 (guided): the FY20 tax landmine. Synthetic IndCo: FY18 EBIT ₹1,000 cr (actual tax 34.9%); FY22 EBIT ₹1,600 cr (actual tax 25.2%). Invested capital ₹3,000 cr (FY18) and ₹5,000 cr (FY22). (a) ROIC each year at actual tax. (b) ROIC each year normalized to 25.2%. (c) How much of the apparent change is the tax cut?

Solution. (a) FY18 NOPAT = 1,000 × 0.651 = 651 → ROIC = 651/3,000 = 21.7%; FY22 NOPAT = 1,600 × 0.748 = 1,197 → ROIC = 1,197/5,000 = 23.9%. At actual tax, ROIC rose +2.2 points. (b) Normalized at 25.2% both years: FY18 NOPAT = 1,000 × 0.748 = 748 → ROIC = 24.9%; FY22 = 23.9%. Normalized, ROIC fell −1.0 point. (c) So ~3.2 points of the "improvement" was the tax cut, not operations. Once you strip it, the operating return actually declined. This is exactly the §4.6 discipline: normalize the tax rate across the decade or you credit the finance minister's ROIC to the business.

P4 (guided): build invested capital and reconcile. Synthetic CleanCo balance sheet (₹ cr): inventory 300, receivables 200, payables 150, net fixed assets 600, goodwill 100, cash 250, investments 100, debt 400, equity 1,000. EBIT 240; tax 25%. Build invested capital both ways, reconcile, and compute ROIC (with and without goodwill).

Solution. Operating approach: NWC = 300 + 200 − 150 = 350; + net fixed 600 + goodwill 100 = ₹1,050 cr (₹950 cr ex-goodwill). Financing approach: debt 400 + equity 1,000 − non-operating assets (cash 250 + investments 100) = ₹1,050 cr. The two tie exactly, which is your built-in error check (M2.03). NOPAT = 240 × 0.75 = 180. ROIC = 180/1,050 = 17.1% (with goodwill), 180/950 = 18.9% (ex-goodwill). Both above the ~12% hurdle; the ex-goodwill figure shows the operating business's return on tangible capital, the with-goodwill figure the return on what shareholders actually paid.

P5 (independent): incremental ROIC + growth check. Synthetic GrowthCo: FY20 NOPAT ₹300 cr, invested capital ₹1,500 cr; FY24 NOPAT ₹520 cr, invested capital ₹2,900 cr; cumulative NOPAT over FY21–24 ≈ ₹1,720 cr. Compute 4-year ROIIC and the reinvestment rate, run the g = RR × ROIC check against the actual NOPAT CAGR, and give a one-line compounder verdict.

Solution. ΔNOPAT = 520 − 300 = 220; ΔIC = 2,900 − 1,500 = 1,400. ROIIC = 220/1,400 = 15.7%. RR = ΔIC/cumNOPAT = 1,400/1,720 = 81.4%. Growth check: RR × ROIIC = 0.814 × 15.7% = 12.8% vs actual NOPAT CAGR = (520/300)^(1/4) − 1 = 14.7%. The identity roughly closes (the ~2-point gap is timing/margin drift, a consistency check not a law). Verdict: GrowthCo reinvests ~81% of profit at ~16%, above the 12% hurdle, so value-creating, but the incremental return is only modestly above the hurdle; this is a heavy-reinvestment, decent-return compounder (most of the growth is the high RR, not a superstar ROIIC), very different from an Asian Paints that grew on ~40% incremental returns while paying most profit out.

P6 (independent): incremental ROIC across an acquisition. Synthetic BuyCo: FY22 NOPAT ₹200 cr, IC ₹1,000 cr. In FY23 it acquires a competitor for ₹900 cr (₹700 cr of it goodwill). FY24 NOPAT ₹330 cr, IC ₹2,050 cr. (a) Compute the naive FY22→FY24 ROIIC. (b) Say in one sentence why it's meaningless. (c) State the correct three-step handling.

Solution. (a) ΔNOPAT = 330 − 200 = 130; ΔIC = 2,050 − 1,000 = 1,050; naive ROIIC = 130/1,050 = 12.4%. (b) Meaningless because ~₹900 cr of the ΔIC is acquired capital, so you are charging BuyCo's organic profit growth against the price of a whole acquired company (plus a barely-digested first-year contribution): apples charged against oranges. (c) Correct handling (§4.6, Worked Example 1): (1) evaluate the acquisition separately as a capital-allocation decision, acquired steady NOPAT ÷ ₹900 cr price; (2) compute organic ROIIC by stripping the acquired revenue, NOPAT, and capital; (3) report ROIC with and without goodwill, remembering shareholders still paid the ₹700 cr goodwill.

P7 (independent): the share-stability test with the profitability cross-check. Synthetic WidgetCo industry, five firms, shares FY15 → FY23 (8 years): A 35→33, B 25→24, C 20→22, D 12→13, E 8→8. You also learn sector ROIC has hovered ≈ 11% (≈ the cost of capital) throughout. Compute the instability metric and deliver the Greenwald verdict.

Solution. Absolute movements: 2 + 1 + 2 + 1 + 0 = 6 points ÷ 5 firms = 1.2 ÷ 8 years = 0.15 points/firm/year, well under the ~0.5 "strong barriers" line, so shares are frozen. But the cross-check (Greenwald step 5, M4.05): sector ROIC ≈ WACC. Verdict: a stalemate, not a moat. Frozen shares with cost-of-capital returns describe an undifferentiated oligopoly that competes its own pool away (airlines pre-consolidation). Barriers to entry may exist, but they protect no economic profit, so there is nothing to buy. Stability is necessary but not sufficient; returns must show up too.

P8 (independent): read the note. A related-party note reads: "During FY24 the Company purchased raw materials of ₹450 cr (18% of total RM cost) from Promoter Holdings Pvt Ltd (controlled by the promoter family) and paid royalty of ₹120 cr (2.1% of sales) to the same entity." State what you extract and what you check.

Solution. Extract: two material related-party leakage channels: 18% of raw-material spend routed through a promoter entity, and a royalty equal to ~2.1% of sales / roughly a mid-teens % of EBIT flowing to promoters. Red flag: value siphoning: RM bought above market price, or a royalty for IP the promoter may not genuinely own, either of which quietly transfers profit from minority shareholders to the promoter (the M2.08 / ethics-thread failure mode). Checks: (1) is it arm's-length and independently certified; (2) the multi-year trend (a rising share is the tell); (3) is the royalty rate within SEBI LODR norms and shareholder-approved; (4) model EBIT with the royalty at zero and the RM at market to size the leakage; (5) carry the RPT map into the Day-3 incentive read (M8.04). The bar is not "are there RPTs" (there always are) but "material, arm's-length, stable?"

P9 (independent): normalize before benchmarking. Two peers: PeerA reports EBITDA margin 22% under IFRS 16 (leases capitalized, so lease cost sits below EBITDA); PeerB reports EBITDA margin 17% and still expenses its (large) operating leases inside EBITDA. PeerA's lease depreciation + interest ≈ 4% of sales; PeerB's rent expense ≈ 4% of sales. Are the margins comparable? Re-rank on a consistent basis.

Solution. Not comparable. PeerA's 22% excludes lease cost; PeerB's 17% includes it, a 4-point definitional gap masquerading as a performance gap. Put both on a rent-inside basis: PeerA ≈ 22% − 4% = 18% vs PeerB 17%. The true operating gap is ~1 point, not the 5 points the raw numbers showed. The naive comparison overstated PeerA's superiority roughly five-fold. (Cleanest fix in practice: compare EBITDAR or EBIT, both lease-consistent.) This is the §4.11 / M2.05 rule: normalize leases, tax, consolidation, and segment mix before you rank anything.

P10 (timed, 3 min): quick reverse-DCF inversion. A company trades at EV/NOPAT = 30×. WACC 10%, defensible ROIC 20%. Solve for the implied perpetual growth and give the one-line verdict against a ~5% cap.

Solution. (1 − g/0.20)/(0.10 − g) = 30 → 1 − 5g = 3 − 30g → 25g = 2 → g = 8% forever. Above the ~5% perpetuity cap, so 30× cannot be a steady state: the price embeds a long super-normal phase (a duration/CAP bet), not a claim about next year. Verdict in one breath: 30× is a bet on a long competitive-advantage period. Now go test whether the moat evidence supports it.

P11 (timed): the three-lever diagnosis. Synthetic PremiumCo: EV ₹60,000 cr, normalized NOPAT ₹1,500 cr (so 40×), WACC 11%, defensible ROIC 25%. (i) Steady-state implied g. (ii) The required-return lever, holding long-run g = 6%. (iii) Frame the verdict.

Solution. (i) (1 − g/0.25)/(0.11 − g) = 40 → 1 − 4g = 4.4 − 40g → 36g = 3.4 → g ≈ 9.4% forever (> cap → duration bet). (ii) Hold g = 6% (RR = 6/25 = 24%, payout 0.76): 60,000 = (1,500 × 1.06 × 0.76)/(r − 0.06) = 1,208/(r − 0.06) → r − 0.06 = 0.0201 → r ≈ 8.0%. (iii) The price implies either ~9.4%-perpetual-equivalent growth (≈ ~13% for a decade then fade) or a long ~15-year CAP or an ~8% required return. The reverse DCF's output is this proposition, not a verdict. The verdict comes from auditing the moat: if share stability + ROIC persistence prove a long CAP, "demanding but fairly priced — pass, no edge"; if the moat is contested (a live entrant, eroding share), "too optimistic — the price pays for a CAP the evidence no longer supports." Same multiple, opposite conclusions depending on the moat evidence (Worked Example 3).

P12 (independent, capstone recall): list the normalizations. You are handed a screener export of an Indian industrial's 8-year P&L and balance sheet (FY17–FY24). Before you trust the ROIC trend, list every normalization you would make and one line of why for each.

Solution. (1) Ind AS 116 lease break (FY20): capitalize leases across all years or use EBIT/EBITDAR, else FY20 EBITDA margin fakes an improvement. (2) FY20 tax cut: compute NOPAT at one normalized statutory rate (25.2%) across the decade, else post-tax ROIC steps up on tax policy. (3) Strip exceptionals each year (restructuring, asset sales, impairment, forex one-offs) into their own row, and flag any that recur. (4) Re-segment to the current definition so segment-revenue rows are continuous. (5) Handle acquisitions: ROIC with and without goodwill; split or lag incremental-ROIC windows around deals. (6) Consolidated, not standalone. (7) Matching rule: pull non-operating other/investment income out of EBIT and the matching non-operating assets (surplus cash, investments) out of invested capital. (8) Average vs year-end capital: use averages by default; time-weight or pro-forma large mid-year deals. (9) Split/bonus-adjust the per-share and share-count history. (10) Consistent capitalization policy: confirm R&D/software treatment didn't change mid-decade. Each normalization is a place a raw export silently lies; the log of them is your audit trail (§4.4).

P13 (guided): peer-map ratios, rank two peers by ROIC, not margin. Two synthetic Indian peers, ~FY24 (₹ cr). FastCo: revenue 12,000, NOPAT 720, invested capital 3,000. SlowCo: revenue 4,000, NOPAT 480, invested capital 6,000. Decompose each ROIC into NOPAT margin × capital turnover and rank against the ~12% hurdle.

Solution. FastCo: ROIC = 720 / 3,000 = 24.0% = NOPAT margin 6.0% (720/12,000) × turnover 4.0× (12,000/3,000). SlowCo: ROIC = 480 / 6,000 = 8.0% = NOPAT margin 12.0% × turnover 0.67× (4,000/6,000). SlowCo has double the margin yet earns a third of the return, because FastCo turns its capital six times as fast (ROIC = margin × turnover, the DuPont/M2.03 identity). FastCo clears the 12% hurdle by a wide spread; SlowCo (8%) destroys value despite the fatter margin. Rank peers by ROIC, never by margin. That is the single most common Day-2 benchmarking error (§4.11, and the DMart/Costco master idea from M0.01).

P14 (guided): locate the profit pool. A synthetic paints-like value chain, one representative year (₹ cr revenue and operating margin per link): raw-material suppliers 3,000 @ 12%; manufacturers 10,000 @ 18%; dealers/distribution 12,000 @ 4%; applicators/contractors 5,000 @ 8%. Compute operating profit per link and state where the pool sits.

Solution. Operating profit = revenue × margin: suppliers 360; manufacturers 1,800; dealers 480; applicators 400. Total chain operating profit = ₹3,040 cr, of which the manufacturer link keeps ₹1,800 cr ≈ 59%, the largest pool and the fattest margin, even though dealers carry the most revenue (₹12,000 cr) at a thin 4%. The pool sits squarely at the manufacturer link (§4.8): it controls brand, distribution density, and the tinting counter, sitting at the pinch-point between fragmented suppliers and a fragmented downstream. Trap: confusing the biggest-revenue link (dealers) with the biggest-profit link (manufacturers) — revenue is not profit.

P15 (independent): the two-window compounder test with tax normalization. Synthetic CompoundCo (₹ cr); normalize tax at 25.2% throughout. FY16 EBIT 500, IC 1,600; FY20 EBIT 850, IC 2,700; FY24 EBIT 1,250, IC 5,400. Cumulative normalized NOPAT FY21–FY24 ≈ ₹3,300 cr. Compute headline ROIC (FY24), both windows' incremental ROIC, the FY20→FY24 growth check, and give the compounder verdict.

Solution. NOPAT (× 0.748): FY16 = 374; FY20 = 635.8; FY24 = 935. Headline ROIC FY24 = 935 / 5,400 = 17.3%, which looks like a healthy compounder. Window 1 (FY16→FY20): ΔNOPAT 261.8 / ΔIC 1,100 = 23.8%. Window 2 (FY20→FY24): ΔNOPAT 299.2 / ΔIC 2,700 = 11.1%. Growth check (W2): RR = 2,700 / 3,300 = 81.8%; RR × ROIIC = 0.818 × 11.1% = 9.1% vs actual NOPAT CAGR (935/635.8)^(1/4) − 1 = 10.1%, so it roughly closes. Verdict: the headline 17% hides the story. Incremental returns have fallen from ~24% to ~11%, now below the ~12% hurdle, so the marginal rupee is no longer creating value even as reported ROIC stays high and the reinvestment rate is rising (~82%). A business plowing more and more capital into sub-hurdle returns is the §4.6 danger sign; it should slow reinvestment and return cash, not chase growth.

P16 (independent): cash-conversion quality. Synthetic LedgerCo FY24 (₹ cr): EBITDA 1,000; CFO 520; capex 300; PAT 600. Compute CFO/EBITDA and FCF/PAT, and say what the numbers flag and what you check next.

Solution. FCF = CFO − capex = 520 − 300 = ₹220 cr. CFO/EBITDA = 520 / 1,000 = 0.52; FCF/PAT = 220 / 600 = 0.37. Both are weak. Healthy franchises run CFO/EBITDA ~0.8–1.0 and FCF/PAT ~0.7–0.9 (Asian Paints ~0.8, §4.5). A persistent gap between accrual profit and cash is the M2.04 earliest quality red flag: earnings aren't converting to cash, typically a working-capital build (rising receivables/inventory), which in a channel business is the sell-in-vs-sell-through / channel-stuffing tell (§4.7). Checks: the multi-year trend (one year isn't proof), the working-capital lines (DIO/DSO/DPO and the cash-conversion cycle), and dealer-incentive/receivables commentary. Cash is the reality check on the income statement.

P17 (independent): read the segment note. Synthetic DiverseCo FY24 segment disclosure (₹ cr): Core (staples) revenue 8,000, segment EBIT 1,600; Growth (new) revenue 3,000, segment EBIT 90; International revenue 2,000, segment EBIT 120. State the real profit engine, quantify the drag, and give the one-sentence framing.

Solution. Segment margins: Core 1,600/8,000 = 20%; Growth 90/3,000 = 3%; International 120/2,000 = 6%. Total EBIT = ₹1,810 cr, of which Core is ~88% (1,600/1,810). The "Growth" segment is a quarter of revenue but under 5% of profit: a sub-scale, thin-margin drag consuming capital and management attention (the §4.5/§4.7 home-décor analog). Framing: "a great staples business with attached projects," not "a diversified growth company", a very different sentence, and the reverse DCF (§4.13) must price the high-return core, not the flattering blended average. The segment note is where the real engine and the cross-subsidy both hide (§4.7).

P18 (timed, 4 min): reverse-DCF inversion meets the live tripwire. Synthetic MoatCo (₹ cr): EV 75,000; normalized NOPAT 1,500 (so EV/NOPAT = 50×); WACC 11%; defensible ROIC 25%. Field fact: a well-funded new entrant arrived last year and the top-four share-instability metric has jumped to ~2.5 points/firm/year. (i) Solve the steady-state implied growth. (ii) Deliver the verdict, using both the perpetuity cap and the Greenwald tripwire.

Solution. (i) (1 − g/0.25)/(0.11 − g) = 50 → 1 − 4g = 5.5 − 50g → 46g = 4.5 → g ≈ 9.8% forever. (ii) 9.8% is far above the ~5–6% perpetuity cap, so 50× cannot be a steady state: the price is a bet on duration, a long undisturbed competitive-advantage period (§4.13). But the live evidence contradicts that bet: the instability metric has crossed the ~2-points/firm/year weak-barrier tripwire (§4.10), so the barrier is visibly breaking in real time. So the implied story (a long, undisturbed CAP) is too optimistic relative to the live risk. The reverse DCF names the bet; the share table says the bet is going the wrong way.

Applied mini-project: the Day-1/2 deep-dive workbook

Task. Run Days 1–2 of the deep dive end-to-end on ONE company, producing the nine-tab workbook (§4.1). Pick the one you will carry into M8.04's Day-3 mini-project. You are starting a real deep dive here and finishing it next module. Choose:

• Titan Company (India). Data: screener.in ten-year financials (revenue, segment mix, margins, ROCE, the FY20 tax and Ind AS 116 breaks to normalize); annual reports + investor presentations from titancompany.in (the Jewellery / Watches & Wearables / Eyewear / Emerging segment note, where you find the real profit engine; Tanishq store counts; studded-jewellery mix); the shareholding pattern (Tata-group + promoter holding); peers Kalyan Jewellers and Senco Gold (their DRHPs on SEBI's site carry sourced industry-share data); Damodaran's retail/luxury ROIC dataset. Reverse-DCF angle: Titan carries a premium multiple on the jewellery-formalization story; decode what growth/duration it implies against the organized-jewellery base rate.
• Costco (US). Data: the latest 10-K from EDGAR (warehouse counts, membership counts and renewal rates, which Costco publishes, membership-fee income, comparable sales, Item 1 Competition); the membership-fee ÷ operating-income arithmetic (where the profit really is); peers Walmart/Sam's Club and BJ's; the NRF Top-100 retailers list (free share data); Damodaran retail ROIC. Reverse-DCF angle: Costco's premium multiple prices a very long CAP; decode it and test it against the renewal-rate and warehouse-runway evidence.

Required workbook contents (this is the module gate; the app grades it against a key):

1. Time-travel management read: a promise→result ledger from ≥5 years of letters/MD&A, read oldest-to-newest, with a verdict on guidance calibration and one "quietly dropped" target.
2. Ten-year history spreadsheet, normalized, with the normalization log (Ind AS 116 / TCJA, tax, exceptionals, re-segmentation, any acquisition).
3. Incremental-ROIC table: two windows + the g = RR × ROIC check + one-line compounder verdict.
4. Notes extract: one line of signal each from revenue recognition, related-party, leases, contingencies, segments, ESOP.
5. Industry map: supplier→firm→customer chain with a profit-pool estimate; state where the pool sits and which way it's migrating.
6. Five forces, scored 1–5 from filing evidence (not recited).
7. Share-stability table: correctly-defined market, 3+ dates, the instability metric, and the profitability cross-check.
8. Peer benchmark: 2–3 peers on identical normalized KPIs, read as a story.
9. 3-statement-lite model: 3–5 drivers → base/bull/bear → a value range.
10. Reverse DCF: implied growth, duration, and required return; the three-lever diagnosis; the base-rate judgment.
11. The verdict + the single variant question carried to Day 3.

Scoring rubric (self-score honestly; pass ≥18/22, the map's ≥80% written-deliverable bar):

If you score below 18, the miss is almost always tab 2 (an un-normalized history poisons everything downstream) or tab 10 (a forward model dressed up as a reverse DCF). Restudy §4.4 and §4.13 respectively before your retake.

Case Lab: run case CL-8.3 (Asian Paints vs Birla Opus — the Two-Day Deep Dive) in the app's Cases tab for this module: the deep-dive workbench pre-loaded with the carried specimen: normalize the ten-year history (defuse the Ind AS 116 and FY20 tax breaks), compute the two-window incremental-ROIC contrast, benchmark the peers on identical KPIs, invert the price for the implied CAP, and finish on the single variant question you hand to Day 3, all graded against the key.

Reading & resources

• Mauboussin & Callahan, Measuring the Moat (Morgan Stanley Counterpoint Global, 2024 update): the four-pass industry→profitability→structure→firm workflow that §4.8 runs; the single highest-value free PDF for Day 2. [Free] [Intermediate]
• Rappaport & Mauboussin, Expectations Investing (revised ed., 2021): the reverse-DCF engine of §4.13, in full: reading price-implied expectations, the value triggers, and the discipline of judging the implied story against base rates. The book §4.13 compresses. [Paid] [Intermediate]
• Damodaran, Narrative and Numbers: the §4.14 scaffold: every story-claim becomes a number, every number traces to a claim. Pair with his blog (Musings on Markets) and the free fcffginzu / reverse-DCF spreadsheets at pages.stern.nyu.edu, which automate the §4.12–4.13 machine. [Book: Paid; spreadsheets/blog: Free] [Intermediate]
• Greenwald & Kahn, Competition Demystified, ch. 4: the market-share-stability test of §4.10 in the original; ch. 5 ("Big Where It Counts") for the local-scale logic behind the paints moat. [Paid] [Beginner→Intermediate]
• Mauboussin & Callahan, The Base Rate Book (Credit Suisse, 2016): the mean-reversion evidence your reverse-DCF verdict must answer to (§4.13, step 4); for India, build your own from screener.in NSE-500 exports. [Free] [Advanced]
• McKinsey, Valuation (8e, 2025), chs. on ROIC, growth, and value creation: the incremental-ROIC / g = RR × ROIC value-driver logic of §4.6 at book length. [Paid] [Intermediate]
• Penman, Financial Statement Analysis and Security Valuation, reformulation chapters: the deep discipline behind §4.4's normalizations (separating operating from financing, reformulating for comparability). [Paid] [Advanced]
• Primary data (build the workbook from these, always): screener.in (India ten-year financials) [Free]; stockanalysis.com / TIKR / Koyfin (US) [Free/Paid]; SEC EDGAR full-text search (10-K, DEF 14A) [Free]; BSE/NSE and company investor-relations pages [Free]; DRHP industry chapters on SEBI's site for sourced share data (Indigo Paints for paints; Kalyan/Senco for jewellery) [Free]; Damodaran data library (industry ROIC, margins, betas, updated each January) [Free].
• Company filings for this module's specimens: Asian Paints consolidated annual reports + investor presentations (asianpaints.com) [Free]; Sherwin-Williams 10-Ks (EDGAR) [Free]. Rebuild both histories yourself; the numbers here are illustrative.
• Podcasts as pre-loaded context (check every number against filings): Business Breakdowns (Asian Paints, Sherwin-Williams, Costco episodes) and Acquired (Costco, Walmart) [Free]; Marcellus Investment Managers' public letters on Asian Paints-class franchises, read as advocacy and audit each claim with §4.10's instruments. [Free]
• Going deeper: Mauboussin & Callahan, Calculating Return on Invested Capital (Counterpoint Global, free PDF) [Free] [Intermediate], the definitive treatment of the ROIC and incremental-ROIC construction §4.6 runs by hand, including the goodwill, operating-lease, and non-operating-asset adjustments that decide whether your denominator is honest. The ideal bench companion to Day 1's history build.
• Going deeper: Aswath Damodaran, The Dark Side of Valuation (3rd ed.) [Paid] [Advanced], on valuing franchises at competitive inflection points: exactly the problem of pricing a high-ROIC leader whose competitive-advantage period is being attacked (§4.13), extending the narrative-and-numbers discipline into the hard cases where the past is a poor guide to the future.

The Modern Analyst's Addendum

Everything above teaches this skill from first principles, by hand, and that is how you learn it, and the mastery check still tests it that way. This addendum shows how a working analyst amplifies the same skill today. It adds; it never replaces. (R1/R10)

AI-Augment this skill

{
  "skill": "The two-day workbook: the normalized ten-year history, the incremental-ROIC compounder test, the notes extraction map, the industry and share-stability work, peer benchmarking on identical KPIs, the 3-statement-lite model, and the reverse DCF that decodes the price",
  "use": "Sections 4.3 and 4.4 already state this module's position on tooling more sharply than any addendum could: the screener gets you 80% of the rows instantly, and you build it by hand anyway, because THE ACT OF BUILDING IS WHERE YOU FIND THE NORMALIZATIONS THE VENDOR SILENTLY SMOOTHED OVER. Everything below respects that. LEGITIMATE: (1) The as-reported series as a CHECK, never a substitute — pull the XBRL history after your hand build and DIFF it against your sheet, because every disagreement is either your arithmetic error or a normalization you made and the vendor did not, and reading that diff line by line is the fastest audit of your own build there is. (2) The oldest-to-newest read of section 4.2 — retrieval over the ten annual reports you have downloaded, with a page citation on every claim, so 'what did management guide for FY18, in their own words' is a locatable quotation rather than a memory. That is the promise ledger's raw material and it is the single most defensible AI pattern in the deep dive, because every claim is pinnable. (3) The notes extraction map of section 4.7 across a decade rather than one year: 'show me every occurrence of the related-party heading in these ten filings, with page numbers' is a search problem. (4) The reverse DCF of section 4.13 as a root-find, so you can invert for growth, for duration and for the required return in three lines instead of bracketing three times. BANNED, and each is a workbook row the certification grades: the five normalization DECISIONS of section 4.4 (which is which lease treatment, what tax rate across the decade, how to re-map a re-segmented history, whether to split the window at an acquisition); the choice of incremental-ROIC windows; step 4 of section 4.13, where the implied story is judged against base rates — a language model asked whether a growth path is plausible will produce a fluent, confident, unsourced frequency, and the only thing that answers the question is a distribution you counted; and the variant question the two days exist to produce.",
  "tools": ["The data.sec.gov XBRL endpoints — companyfacts for the as-reported ten-year series you diff your hand build against, frames for the peer benchmark of section 4.11", "screener.in (ten-year sheet, Excel export, concalls tab), Tijori, Trendlyne — the Indian equivalent, where the export button substitutes for the endpoint that does not exist", "Document/PDF AI with retrieval over your own downloaded set — ten annual reports and four to eight transcripts — used only with a page or note locator on every claim", "A notebook with scipy.optimize.brentq — the reverse DCF of section 4.13 inverted for growth, for the competitive-advantage period, and for the required return", "MCA21 — the AOC-4 and MGT-7 filings of the unlisted subsidiaries that explain a consolidation the listed filings do not"],
  "prompt": "Attached are <COMPANY>'s annual reports for FY16 through FY25. For each year, extract verbatim: (a) every forward-looking quantitative statement management made about revenue growth, margin or capacity, with the page number; (b) the exact wording of the segment-reporting basis note, with the page number. Present it as one row per year with the page number in its own column. Do not interpret, do not tell me whether a promise was kept, do not summarise across years, and do not fill a gap with a plausible sentence — write 'no such statement located, pages searched: <range>' instead. I am grading the promises myself.",
  "verify": "Prove any fetched or coded series against this module's own worked figures first: NOPAT = EBIT x 0.748 must give 1,922 / 2,207 / 2,768 / 3,553 / 4,974 / 3,777 across FY16-FY25, and ROIC on year-end invested capital must give 40.0% / 44.1% / 36.9% / 37.4% / 39.8% / 28.6%; Window 1 must give a 39.6% incremental ROIC on a 3,052 / 7,700 split, a ~30% reinvestment rate and a 11.9% growth check against a 12.6% actual NOPAT CAGR; Window 2 must give 17.7%; and the reverse DCF's steady-state inversion must return ~10.2% at 52x, with the required-return lever solving to ~9.4%. Then verify the thing the endpoint cannot: THAT THE SERIES MEANS ONE THING ACROSS TEN YEARS. A companyfacts payload returns what was tagged, tagged the way it was tagged at the time — so a re-segmentation appears as a concept that simply stops, a policy change appears as a value that steps, and the FY20 tax break appears as a genuine-looking jump in post-tax returns that belongs to the finance minister rather than the business. The Quant lens below prices exactly that error at 3.2 points of incremental ROIC. Diff your hand build against the as-reported series and read every disagreement; a diff you cannot explain is a normalization you have not yet found.",
  "diy": "The gate is unaided: given ten years of filings and nothing else, you build the normalized history by hand, defend each of the five normalizations out loud, compute incremental ROIC over windows you chose and justify the windows, extract the six notes, and invert the price to its implied story on paper. Nothing in this callout is admissible there, and M8.05 certifies the independent version of exactly this workbook."
}

Modern Data Analysis

By hand first. You built a normalized ten-year history and logged every normalization, computed the two incremental-ROIC windows that turn a 40% headline into an 18% marginal truth, extracted six notes for signal rather than completeness, and inverted a 52× EV/NOPAT into a claim about duration rather than growth. §4.3 already ruled on the tooling question: the vendor has 80% of the rows and you build it by hand anyway. So what follows is not permission to skip the build. It is what to do with the build once it exists.

Today's workflow. Get the as-reported series after your hand build and diff it. That inversion of the obvious order is the whole method, and it works because the two sheets disagree in exactly the places that matter. A companyfacts payload is a faithful record of what a filer tagged, tagged the way it was tagged at the time, which means it carries every break §4.4 tells you to defuse, undefused and invisible: a re-segmentation shows up as a concept that stops existing, a capitalization-policy change shows up as a value that steps, an acquisition shows up as invested capital that jumps without a note attached, and the FY20 tax cut shows up as post-tax returns that improve across the whole Indian market at once. Your hand build has views on all four. The diff is where those views become explicit, and a line you cannot explain is a normalization you have not found yet.

The peer benchmark of §4.11 deserves the same suspicion in a sharper form. The frames endpoint returns one concept across every filer for a period, which looks like the identical-KPI comparison §4.11 demands and is not: identical tags are not identical normalizations. Two filers can both tag OperatingIncomeLoss honestly while one capitalizes software the other expenses, one leases what the other owns, and one runs a September fiscal year against the other's March. The endpoint hands you a column that lines up perfectly and compares nothing. Use it to find the peers and pull their raw inputs; do the comparability work in your own sheet, where the adjustments are visible.

The Indian asymmetry, and where it actually costs you. At teardown scale the US–India gap is thirty seconds; screener.in has an export button. At deep-dive scale it is real, and it concentrates in one place: the unlisted subsidiaries. When a consolidation stops making sense (a JV whose economics you cannot see, a subsidiary carrying the margin, a step-acquisition that contaminates §4.6's windows), a US analyst has the parent's segment tagging and the acquiree's own historical filings on EDGAR, while an Indian analyst goes to MCA21 and pays per document for the AOC-4 financial statements and MGT-7 annual return of an entity that files nowhere else. That is not a worse workflow, it is a slower and more deliberate one, and it is often where the finding is: the entity nobody pulls is the entity nobody has checked.

Tools & sources (IN + US). US: data.sec.gov/api/xbrl/companyfacts/CIK##########.json for the as-reported decade you diff against; .../api/xbrl/companyconcept/CIK##########/us-gaap/<Tag>.json when you want one row's full history including the restated points; .../api/xbrl/frames/us-gaap/<Tag>/USD/<period>.json to find and populate the peer set; data.sec.gov/submissions/CIK##########.json for the filing history that dates an acquisition or an auditor change; EDGAR full-text search across the decade for §4.7's note headings; the DEF 14A for the incentive read Day 3 inherits; Damodaran's sector ROIC, margin and cost-of-capital tables at pages.stern.nyu.edu for §4.11's outside benchmark; FRED (DGS10, GDP) for the terminal-growth discipline §4.13 leans on. The SEC requires a declared User-Agent identifying you and publishes a fair-access rate policy; read the current policy, not a number. India: the consolidated annual reports themselves, oldest to newest, which §4.2 makes the highest-signal qualitative exercise in the two days; screener.in for the ten-year sheet, the Excel export and the concalls tab; BSE and NSE announcement feeds and filing archives for the dated events that anchor a time-travel read; SEBI LODR Reg 31 (shareholding and pledge), Reg 23 (half-yearly related-party disclosures) and Reg 30 (material events); MCA21 for the unlisted subsidiaries; RBI and MOSPI series for the nominal-GDP frame §4.13 Step 4 judges implied growth against; CCIL and FBIL G-sec yields for the risk-free rate inside the 11.5% WACC. Run any of it in your notebook.

# R3 — recomputed in-session against this module's own Asian Paints table (illustrative figures).
EBIT = {"FY16": 2570, "FY24": 6650}; IC = {"FY16": 4800, "FY24": 12500}
T_PRE, T_POST = 0.349, 0.252                       # 4.4 normalization 2: the Sept-2019 rate cut
print(f"rate cut alone lifts NOPAT {(1-T_POST)/(1-T_PRE)-1:+.1%}")            # +14.9%  (module: '~15%')
dIC = IC["FY24"] - IC["FY16"]
norm = (EBIT["FY24"]*(1-T_POST) - EBIT["FY16"]*(1-T_POST)) / dIC              # both years at 25.2%
act  = (EBIT["FY24"]*(1-T_POST) - EBIT["FY16"]*(1-T_PRE))  / dIC              # each year's own rate
print(f"Window-1 ROIIC normalized {norm:.1%} vs actual-tax {act:.1%} "
      f"-> overstated {100*(act-norm):.2f} pts = {(act-norm)/norm:.1%}")      # 39.6% / 42.9% / 3.24 / 8.2%

# The lag convention on Window 2 — capital spent this year earns NEXT year (4.6's refinement 1).
N = {"FY20": 2768, "FY25": 3777}; I = {"FY20": 7500, "FY24": 12500, "FY25": 13200}
dN, dI = N["FY25"]-N["FY20"], I["FY25"]-I["FY20"]
last = I["FY25"] - I["FY24"]
print(f"as published {dN/dI:.1%} | ex final-year capital ({last/dI:.1%} of dIC) {dN/(dI-last):.1%} "
      f"| average-IC {dN/(((I['FY25']+I['FY24'])/2)-I['FY20']):.1%}")         # 17.7% / 20.2% / 18.9%

# WHERE the FY25 collapse came from: ROIC = NOPAT margin x invested-capital turns.
R = {"FY24": 35500, "FY25": 33900}; NP = {"FY24": 4974, "FY25": 3777}
m = {y: NP[y]/R[y] for y in R}; t = {y: R[y]/I[y] for y in R}
tot = m["FY25"]*t["FY25"] - m["FY24"]*t["FY24"]
marg, turn = (m["FY25"]-m["FY24"])*t["FY24"], m["FY24"]*(t["FY25"]-t["FY24"])
print(f"margin {100*m['FY24']:.2f}->{100*m['FY25']:.2f}%, turns {t['FY24']:.3f}->{t['FY25']:.3f}")
print(f"ROIC {100*tot:+.2f} pts = margin {100*marg:+.2f} + turns {100*turn:+.2f} "
      f"+ interaction {100*(tot-marg-turn):+.2f}  ({marg/tot:.0%} / {turn/tot:.0%})")  # -11.18 = -8.15 -3.81 +0.78

# How often a multi-source reconciliation comes out clean, as a structure (not a measurement).
for k, q in [(12, .05), (12, .10), (20, .05), (20, .10)]:
    print(f"{k} line items at {q:.0%} per-item mismatch -> all agree {(1-q)**k:.1%}, expected {k*q:.1f}")
    #  54.0% / 28.2% / 35.8% / 12.2%

Verify. The engine has to reproduce the module's own build before it is allowed near a company: NOPAT of ₹1,922 / 2,207 / 2,768 / 3,553 / 4,974 / 3,777 crore from EBIT × 0.748, ROIC of 40.0 / 44.1 / 36.9 / 37.4 / 39.8 / 28.6% on year-end invested capital, EBIT margins of 17.0 / 17.4 / 18.3 / 16.3 / 18.7 / 14.9%, FCF/PAT of 0.70 / 0.39 / 0.57 / 0.82 / 0.83 / 0.79, Window 1's 39.6% incremental ROIC with a reinvestment rate of 30.1% and a growth identity of 11.9% against an actual NOPAT CAGR of 12.6%, Window 2's 17.7%, and §4.13's inversions at 10.2% implied perpetual growth and 9.4% implied required return. Every one of those reproduces exactly, which makes them a usable regression test. Then verify the two things the engine will not tell you. First, which year the tax break lands in: the cut took effect in September 2019, so an Indian March-year FY20 straddles it, and treating FY20 as fully post-cut, as this build does, is itself a normalization choice with a defensible alternative. Second, which invested-capital convention you used, because §4.3's own row template specifies average invested capital while §4.5's table computes on year-end. Both are legitimate, they differ materially, and a build that does not say which one it used cannot be compared to anything, including your own build of the same company next year.

Quantitative lens

Days 1–2 produce a number, incremental ROIC, that the module correctly calls the quantitative heart of the whole analysis. So the honest quantitative question is how much of that number is the business and how much is the bookkeeping. All figures below come from the block above, run on the illustrative Asian Paints table above; they are properties of that table and those conventions, not measurements of any company.

The tax normalization is worth 3.2 points of incremental ROIC, which is a fifth of the entire finding. §4.4 warns that computing NOPAT with each year's actual tax turns the September-2019 rate cut into a step-up that belongs to the finance minister. Price it: the cut alone lifts NOPAT by 14.9% on unchanged EBIT (the module says "~15%", and it is right), and running Window 1 with each year's own statutory rate gives an incremental ROIC of 42.9% against the normalized 39.6%. That is 3.24 percentage points, or 8.2% of the true figure, manufactured entirely by a change in tax law. It matters more than the size suggests because of where it lands: Window 1 straddles the break and Window 2 does not, so an un-normalized build flatters the golden years and leaves the recent years alone, which widens the very 40%-to-18% contrast the module builds its verdict on. The error does not just inflate a number. It inflates the story.

And the convention you never chose is worth another 2.5 points. §4.6 notes that capital spent this year earns next year, and that a lagged incremental ROIC "lifts the recent number a couple of points." Quantify it on the table's own figures: the final year's capital is 12.3% of Window 2's invested-capital increase, and excluding capital that has not yet had a year to earn moves the Window-2 incremental ROIC from 17.7% to 20.2%, a gain of +2.5 points, exactly the couple the module claims. Switch instead to averaging invested capital across the last two years and you get 18.9%. So the same data, honestly handled, supports a Window-2 answer anywhere from 17.7% to 20.2% depending on two conventions nobody argues about. This is why §4.6 tells you to state the answer as a range: "recent incremental returns in the high teens, down from around 40%", and it is a rule with a number behind it: a point estimate of incremental ROIC quoted to a decimal is a claim about your conventions dressed as a claim about the business. Every one of those readings still clears the 12% hurdle and every one is roughly half the legacy return, which is the finding; the decimal is not.

Decompose the FY25 collapse and it points Day 3 somewhere specific. ROIC is NOPAT margin times invested-capital turns, and both fell: the margin from 14.01% to 11.14%, the turns from 2.840 to 2.568. Split the −11.18-point fall and margin contributes −8.15 points and capital productivity −3.81, with an interaction giving back +0.78, so roughly 73% of the collapse is margin and a third is the capital base growing into a shrinking revenue line. That is a directional instruction for the qualitative day that follows. If the fall were mostly capital turns, Day 3 would be about the capex programme and the sub-scale home-décor build: was the money spent well, and how long until it earns? Because it is mostly margin, Day 3 is about price: discounting, dealer incentives, the mix shift, and whether the competitive event is showing up in realisation rather than in volume. §4.13 already isolates "does the captivity hold" as the variant question; this decomposition says the evidence for it is in the gross-margin bridge, not the capex line.

A three-source reconciliation is rarely clean, and that is arithmetic rather than misconduct. A deep dive checks the same figure across the filing, the transcript and the press release, and the ways they can legitimately differ are numerous: a period boundary, a consolidation scope, a segment definition, a pre- or post-exceptional basis, a like-for-like adjustment. Model it as a structure: check 12 line items with a 5% per-item chance that two sources define one of them differently and a fully clean reconciliation happens 54.0% of the time; at 10% it is 28.2%; across 20 items at 10% it is 12.2%, with two expected mismatches. Read that as permission and as warning at once. Finding a disagreement is the expected outcome and is not by itself evidence of anything. The mistake is treating each one as a scandal. Finding no disagreement across twenty items is the genuinely surprising result, and usually means you compared a number to itself.

Honest limits, and they run deeper here than in most modules. The Asian Paints table is explicitly illustrative, and the module says so twice, in bold, so every figure above is exact arithmetic on teaching specimens, not a measurement of a company; rebuild it from the filings and your decimals will differ. The per-item mismatch rates in the reconciliation model are assumed, not counted: I have no panel of filings scored for cross-source disagreement, so the correct use is to run it with your own estimate and notice how fast a "clean reconciliation" becomes unlikely, not to quote 54%. The lag adjustment assumes final-year capital earns nothing at all, which is a bound rather than an estimate; the truth sits between the published and the ex-final-year figure. The FY20 tax treatment is a choice, as flagged. And the largest limit is one the whole module already makes: incremental ROIC over a five- or eight-year window is a single ratio of two differences, so it inherits the noise of both endpoints. A depressed terminal year drags it down for reasons that have nothing to do with the marginal rupee's productivity, which is precisely FY25's situation. The number is a diagnostic that tells you where to look. It is not a verdict, and §4.13's price work, not §4.6's ratio, is where the two days actually resolve.

Do it in code: build the history by hand, then diff it against the as-reported series and explain every line of the diff; hold the tax rate constant across the decade before you compute anything incremental; report incremental ROIC as a range across conventions and say which conventions you ran; decompose any ROIC move into margin and turns before you go looking for a cause; and invert the price for growth and duration and required return, because §4.13's three-lever diagnosis is three root-finds on one function. See G2/DA1.03, G2/QD2.04 and G1/QM1.03.

Where this goes next: galaxy cross-links

• G2 · DA1.03 pandas II: Cleaning, Time Series & the Excel↔Python Bridge: the mechanics of the diff above: aligning a decade of as-reported facts onto one definition, handling concept names that stop and start, and building the normalization log as a column rather than a memory.
• G2 · QD2.04 Reproducible Research, Deployment & the Quant Stack: a two-day workbook whose every figure can be regenerated from its sources a year later, which is the difference between a deep dive you can update next quarter and one you have to redo.
• G1 · QM1.03 Time Series, Monte Carlo & Machine Learning: the reverse DCF of §4.13 as a root-find and then as a distribution: the three-lever diagnosis becomes three inversions of one function, and the base/bull/bear fork becomes a band.
• G3 · CN4.01 Reading an Industry Like a Strategist: the industry map and profit pool of §4.8–4.9 at consulting depth: the supplier-to-customer chain, where the pool sits, and why the five forces are a scoring instrument rather than an essay.
• G2 · CS2.03 Databases & SQL for Analysts: once you have deep-dived thirty companies, the workbooks stop being files and become a table you query: which of my names have incremental ROIC compressing toward the hurdle, and which normalizations did I log for each.

Teach it back & journal

Feynman prompt. Write a one-page explainer for a smart 15-year-old titled "Why can a wonderful, famous company still be a risky thing to buy?" Your page must smuggle in, without any jargon (no "moat," "CAP," "reverse DCF," "ROIC"): the idea that a great business is one whose edge lets it keep more of every rupee than its rivals; the idea that the important question is how many more years that edge will last; and the punchline, that the price you pay already contains someone's guess about how long the edge lasts, so buying "a great company" at a high price can mean betting the edge lasts twenty more years right when a giant new rival has shown up to shorten it. Use the town's one always-busy sweet shop, and a deep-pocketed new chain opening across the street, to carry the whole argument. If your page needs a formula, you don't own it yet.

Journal prompt. Three entries. (1) Take the company you most admire and compute its incremental ROIC over the last five years, not its headline ROIC. Did the marginal rupee's return surprise you, and in which direction? (2) Run the 60-second reverse-DCF inversion (§4.13, step 1) on one stock you currently think is "obviously expensive", and write down the perpetual growth or the required return it implies, and whether the number changed your conviction or merely confirmed your prejudice. (3) Name, in one sentence, the single variant question for that company, the one thing whose answer would settle the debate, and note honestly whether you could answer it from filings alone or whether it needs the field work of a Day 3.

This module's flashcards and mastery quiz are wired into the app: see the node's Quiz and Reviews.