The Schools II: Fisher to Pabrai

Phase 6 · Week 53 · ~20 hours

Where it sits. Second module of the philosophy phase, and the second of the two "schools" modules. The first one traced the trunk of the value tree: Graham's deep value (intrinsic value, Mr. Market, margin of safety, net-nets) and its evolution through Buffett and Munger into quality-at-a-fair-price (the four filters, owner earnings, pricing power). Here are the rest of the great schools, the branches that grew off that trunk and, in two cases, off entirely different roots. You meet Fisher's growth-and-scuttlebutt; Lynch's GARP and his six-category taxonomy; Greenblatt's special situations and the Magic Formula; Klarman's absolute-value discipline and the courage to hold cash; Marks's cycles, risk, and second-level thinking; Pabrai's Dhandho asymmetry; and the modern quality-compounder school (Akre, Terry Smith, Nick Sleep). The week closes on the debate that divides all of them, concentration versus diversification, presented so you can take a reasoned side rather than inherit one. These are not nine unrelated ideologies to memorize. They are nine lenses and temperaments, each strongest on a different kind of business and a different kind of investor, and the point of studying all of them is to build a toolkit and to find the corner of it that fits your own mind.

Suggested hour budget. Core content: 8h · worked examples (incl. the Magic-Formula ranking): 2.5h · practice set (match-the-school + apply-the-lens + the ranking exercise): 4.5h · applied mini-project (the Schools Panel): 3h · flashcards + mastery check: 2h.

App link. The work here feeds the Philosophy & Behavior branch: the match-the-school deck (the gate for this week, Schools Exam Part B, ≥85%: a description in, the school out) and the apply-the-lens case drill (one company shown four ways: what would Lynch/Greenblatt/Klarman/Smith each ask?). It also feeds the Magic-Formula ranking sandbox (a universe of EBIT/EV and ROC figures in, the combined rank out, wired to the ROIC engine and the multiples engine), and a Dhandho asymmetry calculator that shares its plumbing with the position-sizing tools you'll meet in Phase 9.

Learning objectives

By the end of the week you can:

1. State each school's core tenets, master text, and edge-window for Fisher (growth + scuttlebutt), Lynch (GARP), Greenblatt (special situations + the Magic Formula), Klarman (absolute value), Marks (cycles + risk), Pabrai (Dhandho asymmetry), and the quality-compounders (Akre, Smith, Sleep), in two or three sentences each, and name when the school works and when it fails.
2. Run Fisher's scuttlebutt method: turn the fifteen points into a concrete qualitative-research plan (whom to call, what to read, what to listen for) on a real company, and connect it to the Phase 8 deep dive's Day 3.
3. Classify any company into Lynch's six categories (slow grower, stalwart, fast grower, cyclical, turnaround, asset play) and derive the category-specific question that decides the thesis. Compute and interpret a PEG (and the dividend-adjusted PEG), knowing exactly what PEG can and cannot see.
4. Execute the Magic Formula end-to-end: define earnings yield (EBIT/EV) and return on capital (EBIT/tangible capital), rank a universe, combine the ranks, and state the honest evidence on the strategy: why it worked, why it lags for years at a time, and how much is just value × quality.
5. Distinguish absolute-value from relative-value investing (Klarman), explain why holding cash can be the disciplined output of a bottom-up process rather than a market call, and say what "margin of safety" adds beyond a cheap multiple.
6. Apply Marks's second-level thinking, his definition of risk as permanent loss, and the pendulum/cycle model to a real market episode, and locate today's market on the cycle without pretending to forecast it (the bridge into Phase 7).
7. Build a Dhandho ("heads I win, tails I don't lose much") asymmetric bet: lay out the outcome tree, compute the expected value, verify the downside is survivable, and connect the sizing logic to Kelly and to Phase 9.
8. Argue the concentration-versus-diversification question from both sides, Munger/Pabrai/Sleep versus Graham/Schloss, using Kelly, hidden correlation, and ruin-avoidance, and state the specific factors that set the right answer for a given investor.

Prerequisites & connections

Builds on. M6.01: everything here is "Schools II," so the vocabulary of intrinsic value, Mr. Market, margin of safety, owner earnings, and the four filters is assumed. Fisher and the quality-compounders are the growth half of the synthesis Buffett made ("85% Graham, 15% Fisher"), and the missing 15% and its descendants arrive this week. M4.04 to M4.05: the moat lenses and the ROIC-persistence test. The quality-compounder school you meet this week is moat analysis worn as a portfolio philosophy, and Sleep's "scale economies shared" is a specific flywheel you already met in M4. M2.03: ROIC/ROCE built by hand and the ~12% (INR) / ~8 to 9% (USD) hurdle (as of mid-2026, check against your own WACC builds). The Magic Formula's "return on capital" leg and Akre's "compounding machine" are that number, systematized. M3.05 to M3.06: reverse DCF, the multiples and their driver algebra, and the India premium-multiple discipline. Lynch's PEG, the Magic Formula's earnings yield and the quality school's "don't overpay" all live inside that machinery. M3.10: special situations and the SOTP/holdco discount. Greenblatt's spin-offs are where those discounts are born and closed.

Feeds forward. M6.03, mental models and decision science: Pabrai's Dhandho and the concentration debate hand directly to Kelly, fractional Kelly, expected value, and ergodicity/ruin, which M6.03 makes rigorous. M6.04, behavioral finance: Marks's pendulum and three-stages model is the sentiment cycle M6.04 explains bias-by-bias, and every school here is, at bottom, a pre-commitment device against a specific bias. M6.05, the checklist capstone: the schools you resonate with become lines in your personal checklist. M8.04: the deep dive's Day 3 is Fisher's scuttlebutt run under a workbook, and the fifteen points are the skeleton of that day. Phase 7: Marks's cycles are the investor's-eye view of the macro machine Phase 7 builds mechanically. M9.01 to M9.02: risk as permanent loss (Marks, Klarman), position sizing (Pabrai/Kelly), hidden correlation, and the concentration policy you'll commit to in writing are the back half of this week's tour, turned into a portfolio process. Competency C9 in the master map ("an investor's temperament and process") begins here, and the schools are where temperament gets its vocabulary.

The short version. There's no single correct way to invest. There's a small set of internally coherent approaches, and each one works when its owner has the temperament and the edge it demands. It fails when it's applied to the wrong business or the wrong person, or in the wrong market weather. Your job this week is to learn all of them well enough to borrow from each, and to recognize which one is actually yours.

Core teaching content

4.1 Schools, not sects: the map before the tour

A beginner meets the great investors as a set of clashing commandments. Graham says buy statistical bargains and diversify widely; Munger says buy wonderful businesses and concentrate. Lynch says invest in what you see at the mall; Klarman says wait in cash until the math is undeniable. Fisher says hold forever; Greenblatt says rank, buy the basket, and rotate every year. Confronted with the contradictions, the beginner does one of two unhelpful things: picks a guru and treats the others as heretics, or concludes the whole thing is arbitrary.

Both are wrong, and you'll want to be inoculated against both before the tour starts.

Here's the idea that organizes the whole tour. Every durable school is a different answer to the same two questions. Where does mispricing come from? And what edge lets me exploit it without being ruined? Each answer is correct for a specific mix of business type and market condition, matched to an investor's temperament. So the schools aren't really in conflict. Each one specializes, and they sort more cleanly than you'd expect:

• Where the mispricing lives. Graham and Schloss hunt statistical cheapness, because the market ignores dull, small, ugly balance sheets. Fisher and the quality-compounders hunt underestimated durability. The market systematically underpays for how long a great business stays great, the duration lever from your reverse-DCF work. Lynch hunts the retail information gap: you see the trend on the ground before the analyst models it. Greenblatt hunts structural or forced selling. Think of the spin-offs and restructurings dumped by holders who can't or won't keep them. Klarman hunts forced selling and complexity: distress, illiquidity, the ugly situations institutions avoid. Marks hunts the cycle, since sentiment overshoots value in both directions. Pabrai hunts asymmetry, situations where the downside is capped and the upside isn't.
• What edge each demands. Graham and Schloss need breadth and a steady hand. You own enough cheap things that the group works even as individuals fail. Fisher and the compounders need judgment about the future and the patience to hold through it. Lynch needs legwork, and honesty about where your circle of competence ends. Greenblatt needs a willingness to do unglamorous plumbing in overlooked corners. Klarman needs the temperament to sit in cash and look wrong for years. Marks needs the emotional control to be greedy and fearful at the times that feel wrong. Pabrai needs the courage to bet big, and rarely.

Read the two lists together and the "contradictions" turn into a division of labor. Diversify if your edge is statistical cheapness with unreliable individual outcomes. Concentrate if your edge is deep understanding of a few durable businesses. Hold forever if you bought a compounding machine, and rotate every year if you bought a systematic factor basket.

The commandments only clash when you rip them out of the approach that makes each one rational.

Three cautions before the tour. First, the schools overlap more than their marketing admits. Margin of safety runs through all of them. Klarman got it from Graham. So did Pabrai and Marks. "Return on capital" drives Greenblatt's formula and Akre's stool and Smith's screen. Second, a school is a temperament as much as a technique. Klarman's cash rule is useless to someone who can't bear to look wrong, and Pabrai's big bets are ruinous to someone who can't hold through a 40% drawdown without selling. The best investors didn't pick their school from a menu. They found the one their own mind could actually carry out under pressure.

Third, and this runs through all of Phase 6, the technical skill of Phases 1 to 5 is necessary but not sufficient. Each school on the tour is a device for surviving contact with a live market and with your own psychology. The analysis tells you what a business is worth. The school tells you how to behave so that knowing it does you any good.

Now the tour. For each school you'll get the core tenets, the master text, when it works and when it fails, and one worked application, with Indian and US examples taking turns.

4.10 Synthesis: nine lenses, one toolkit

Step back from the tour and the nine schools resolve into one toolkit, organized around the five questions that have shaped the whole course so far. Each school is a specialist instrument for a different situation:

• A great, durable business with a long runway that you understand, where the market underrates how long it will last, is a job for Fisher and the compounders: Akre, Smith, Sleep. Hold for years, remember the multiple is a duration bet, and watch the moat's falsifiers.
• A growing business you spotted early and can understand, at a sane price, is Lynch territory. Classify the category first and use PEG as a flag. Do the work, and let winners run.
• Structural mispricing from forced selling or a corporate event belongs to Greenblatt's special situations, or to Klarman for distress and complexity. Find the forced seller and demand the catalyst.
• Cheap-and-good across a whole universe, run mechanically to beat your own emotions, is Greenblatt's Magic Formula. Diversify the basket and hold through the lean years.
• Klarman and Marks are the ones to reach for when nothing clears the bar and the pendulum is at euphoria. Raise standards and hold cash. Wait, and read the temperature.
• A rare capped-downside, uncapped-upside situation is Pabrai's. Size it fractional-Kelly big, and verify the floor is real.
• Any decision, always, gets Marks's second-level thinking (what's priced in?) and margin of safety (what if I'm wrong?). Every school shares those two.

Running through all of it is the Phase 6 thesis. The analysis of Phases 1 to 5 tells you what a business is worth. The schools tell you how to behave: which mispricings to hunt, how long to hold, when to do nothing, when to bet big and when to sit in cash, so that your analysis actually earns something. Nobody runs all nine at once. The best investors pick the two or three that fit their edge and temperament, and run those relentlessly.

For the rest of Phase 6, notice which lenses feel like yours, the ones you'd actually use when the market is screaming. That self-knowledge goes into your personal checklist at the end of the phase, and a durable process starts there.

Common mistakes & how experts think differently

1. Reading "invest in what you know" as "buy what you like." It's the most expensive misreading in the retail world. Liking a product gives you a lead, and a lead isn't a thesis. A beloved product inside a competitive, low-return, over-valued business is a bad investment. An expert treats a ground-level observation as the start of the work and never the end: read the filings, check unit economics and the balance sheet, buy only at a GARP price. They also know that whole sectors (banks, biotech, semis) offer the amateur no consumer edge at all.

2. Mislabeling Lynch's category and applying the wrong thesis. This looks like buying a cyclical at its lowest P/E, which comes at peak earnings right before the fall. Or selling a fast grower after a 40% gain and cutting off the ten-bagger. Or holding a stalwart forever and expecting fast-grower returns. The expert states the category first, because it decides whether a low P/E is a bargain or a trap, and whether a 50% gain is a sell signal or a reason to sit still.

3. Using PEG as a valuation instead of a screen. PEG is blind to how durable the growth is, to return on capital, and to balance-sheet risk. A 25%-grower with a two-year runway is dear at a PEG of 0.5. One with a twenty-year runway and 40% incremental ROIC can be cheap at a PEG of 2, because you're paying for duration. The expert uses PEG to rank and flag, then values the business properly and asks what the price implies about how long the growth lasts.

4. Treating the Magic Formula (or any backtest) as a promise. The headline ~30% CAGR is a debated, in-sample figure. The live strategy lags the market for years at a time, and much of its edge is just value × quality, which factor funds also capture. The expert runs it as a screen with fixed rules. They hand-check the top names for value traps and shenanigans, since a high EBIT/EV often hides a melting cyclical or an accounting problem, and they diversify the basket. Above all, they have the temperament to keep running it through the lean stretches, and that's the only reason the edge survives.

5. Cloning the ticker instead of the reasoning. If you copy a famous investor's holdings from a 13F or a bulk-deal disclosure without working out why, you can't tell when the thesis breaks. You're guaranteed to sell at the bottom, because the conviction was never yours. The expert treats cloning as sourcing just like screens and the mall, then does the independent work that lets them hold through a drawdown.

6. Confusing relative value with absolute value. "It's cheaper than its peers" doesn't mean "it's cheap." A whole sector can be overpriced together, and relative-value buyers ride it down. The expert, Klarman, demands an absolute discount to a conservative estimate of intrinsic value and won't own anything that lacks one. That rule is what keeps this kind of investor out of bubbles.

7. Mistaking risk for volatility, and feeling safest at the top. Academics call a wobbly price "risky." Marks and Klarman save the word for the probability of permanent loss, which is highest when prices are extended and everyone is confident, and lowest after a crash when everyone is terrified. The expert flips the crowd's instinct. They get cautious as euphoria and prices rise, because the price being paid is creating the risk, and they get bold as fear and cheapness peak.

8. First-level thinking: buying "good companies" without asking what's priced in. "Great business, therefore buy" ignores that the greatness may already be fully in the price, which leaves you with an overpriced stock in a wonderful company. The expert thinks at the second level. What does the price already assume, and is my view of the future defensibly different? That's the reverse-DCF habit turned into a reflex.

9. Letting "do nothing" and "hold forever" become alibis for not selling a broken thesis. Fisher's "sell almost never" and Smith's "do nothing" apply to winners whose economics are intact. They never apply to a former compounder whose moat has eroded. The expert separates conviction, which means holding through price volatility while the thesis holds, from stubbornness, which means holding through a deteriorating thesis out of loyalty or anchoring. They also commit in advance to the falsifiers that would force a sale.

10. "Quality at any price." This is the most seductive modern error. Compounders reward patience, so investors conclude any multiple is fine. They forget that a wonderful business bought at 60× can return nothing for a decade while the multiple halves and earnings double. The expert keeps three dials separate, as the moat work taught: the business's quality, how long its advantage lasts, and the price paid for both. To them "don't overpay" is a real constraint, and far more than a footnote.

11. Treating the schools as mutually exclusive religions. Beginners pick a guru and dismiss the rest. Experts see nine specialized instruments and choose the two or three that fit the business, the market weather and, most of all, their own temperament. They borrow margin of safety and second-level thinking everywhere, since every school shares those two.

12. Adopting a concentration policy that doesn't match your real edge or temperament. Concentrate without doing the work, or without the stomach for drawdowns, and conviction turns into ruin. Diversify so broadly that you own your 50th-best idea and your edge turns into the index minus costs. Own 30 names that are all one factor and you've got a concentrated bet in a diversified costume. The expert sets the dial by how reliable their edge is and by the survival constraint. They count independent bets instead of tickers, and they never let a single name's worst case sink the portfolio.

Worked examples

Worked Example 1: The Magic Formula, ranked end-to-end (synthetic 8-name universe)

All figures are synthetic and internally consistent. You're learning the mechanics here, and you'll run them on real screener.in or stockanalysis.com data. Units are currency-neutral. A real run would exclude financials and utilities, so assume these eight are all industrial or consumer names above the market-cap floor.

The first factor is earnings yield = EBIT / EV, which measures cheapness. The second is return on capital = EBIT / tangible capital employed, which measures quality. Tangible capital = net working capital + net fixed assets. Here's the universe with both ratios worked out:

Step 1: rank on earnings yield (1 = cheapest/highest yield): E (24.0) → 1; A (21.0) → 2; H (18.0) → 3; C (16.0) → 4; F (15.0) → 5; B (10.0) → 6; D (8.0) → 7; G (7.5) → 8.

Step 2: rank on return on capital (1 = highest quality): F (60.0) → 1; H (50.0) → 2; B (48.0) → 3; A (40.0) → 4; D (30.0) → 5; C (20.0) → 6; E (12.0) → 7; G (9.0) → 8.

Step 3: sum the two ranks (lowest combined = best):

Step 4: read the result, and learn what the formula is doing. The winner is H (combined 5). It isn't the cheapest name (that's E) or the highest-quality one (that's F). It's the best balance of cheap and good, and this one table carries the Magic Formula's whole lesson:

• E is the cheapest name, with an earnings yield of 24%, but it finishes 4th, dragged down by weak quality (ROC 12%, rank 7). The formula won't reward cheapness alone. E is a potential value trap, the kind of statistically cheap but mediocre business a naive low-P/E screen would buy.
• F is the highest-quality name, with ROC of 60%, but it only ties for 2nd, because you're paying more for it (earnings yield 15%, rank 5). The formula refuses to overpay for quality, and that's what separates it from "quality at any price."
• G is worst on both counts (combined 16), expensive and low-return. It isn't even a value trap. It's just a bad business at a bad price, and an easy pass.
• D is the "expensive good-ish growth story" (earnings yield rank 7). The formula declines it despite a decent ROC, which makes it a systematic antidote to paying up for a narrative.

Step 5: how you'd actually use this. In a real run you buy a basket of the top names. Greenblatt suggests ~20 to 30 from a universe of hundreds, so the tie between A and F doesn't matter. You'd buy H, A, F, E and more. Then, remembering what the evidence says, you wouldn't trust the ranks blindly. You'd hand-check the top names with your forensic tools. Is H's high EBIT/EV hiding a melting cyclical at peak earnings, a one-time gain or an accounting shenanigan? You'd confirm you're not loading up on one factor or sector by accident. And you'd do the hard part, which is committing to hold the basket through the two- or three-year stretches when it lags.

The ranking takes 20 minutes of arithmetic. The edge is having the temperament to keep running it.

Worked Example 2: Lynch: category first, then PEG (US and India)

Part A: the US stalwart (PEG in practice). A high-quality US consumer name trades at a P/E of 22, grows EPS ~11% and yields 2.5%. These are illustrative, so pull the real figures. Lynch's first move is to classify. It's mature and high-quality with low-double-digit growth, and it resists recessions, which makes it a stalwart. The stalwart question is "is the price reasonable, and is the growth story intact?" So compute the PEG.

• Simple PEG = P/E ÷ growth = 22 ÷ 11 = 2.0. On Lynch's scale, where fairly priced ≈ 1.0, that reads expensive.
• Dividend-adjusted PEG = (growth + yield) ÷ P/E = (11 + 2.5) ÷ 22 = 0.61. Lynch credits the yield here, since it's part of your return. On his dividend-adjusted scale (>2 very attractive, ~1 fair, <1 poor), that also reads poor, in other words expensive.

Both flavors agree: it's a wonderful business at a demanding price. For a stalwart, the move is to wait for a better entry. A 30 to 50% gain from a lower price is the realistic prize here, and a ten-bagger isn't on the table. Compare a fairly valued case, P/E 15 and growth 15% for a PEG of 1.0, which is Lynch's "the P/E of a fairly priced company equals its growth rate." Then compare an attractive one: P/E 14, growth 22%, yield 1%, giving a simple PEG of 0.64 and a dividend-adjusted (22+1)/14 = 1.64, which is good.

Part B: the India fast grower (PEG's blind spot). An Indian consumer or specialty-finance fast grower trades at a P/E of 45 and grows EPS ~25%, with no dividend (illustrative). Classify it: small or mid-sized, aggressive, 20%+ growth. That's a fast grower, the home of ten-baggers. PEG = 45 ÷ 25 = 1.8, "expensive" by the mechanical rule. A PEG robot goes wrong right here, because PEG can't see how durable the growth is or what return on capital it earns. Suppose this business can reinvest at 40% incremental ROIC for another 15 to 20 years, behind a real, gated moat. Then the market is rationally paying for duration, and a PEG above 1 can be justified. India's premium-multiple regime is partly this effect. If the 25% growth has only two or three years left, though, the stock is dear even at a PEG of 0.5.

So use PEG to flag and rank, then answer the fast-grower question properly. Is the growth profitable and self-financed? Is it repeatable, with room left to roll out? Then reverse the multiple to see how many years of growth the price actually demands.

Worked Example 3: Pabrai: the Dhandho asymmetry, and the Kelly bridge to sizing

Part A: is it a "heads I win, tails I don't lose much" bet? A distressed-but-surviving situation trades at ₹100. Here's your outcome tree over ~3 years. It's synthetic, and in real work you'd build these probabilities from proper downside and upside analysis instead of guessing:

Expected value = 0.60 × 250 + 0.30 × 120 + 0.10 × 40 = 150 + 36 + 4 = ₹190, a +90% expected return. Now the two Dhandho tests. (1) Is the expected value well above the price? ₹190 against ₹100, so yes, decisively. (2) Is the downside survivable? The worst realistic case is −60%. It would hurt, but it wouldn't wipe you out, and it carries only a 10% probability. Both tests pass, so this is a Dhandho bet.

It's the shape of the payoff that makes it one, far more than the headline upside. A coin flip paying +100% or −100% at even odds has the same "150% upside" flavor. But it has zero expected value and a ruinous tail. Pabrai passes on that all day and pulls the trigger on this one.

Part B: how big? The Kelly bridge (a first look at sizing). "Few bets, big bets" needs a sizing rule, and Kelly is the growth-optimal one. For a clean binary illustration, take a bet that wins +100% with probability 0.60 and loses 100% with probability 0.40. Kelly's fraction is f* = (b·p − q) / b, with net odds b = 1, p = 0.60, q = 0.40:

f* = (1 × 0.60 − 0.40) / 1 = 0.20, so full Kelly says stake 20% of capital.

That small number teaches three lessons. The concentration debate and Phase 9 both lean on all of them.

1. Kelly sizes by edge. A bigger edge earns a bigger bet, and that's the mathematical case for concentration.
2. Full Kelly is too aggressive for real investing. It assumes you know p = 0.60 exactly, which you don't. It also tolerates savage drawdowns. So practitioners bet fractional Kelly: half-Kelly here = 10%, quarter-Kelly = 5%.
3. Dhandho's capped downside is what makes big bets survivable. Part A's real situation, with its bounded −60% worst case, is the kind of shape where you can size up without risking ruin. A bet with a total-loss tail, like the binary above, has to be sized far smaller however favorable the odds, because ruin ends the compounding game for good. That's ergodicity again.

Pabrai's "bet big" means fractional-Kelly big on capped-downside bets. It never means full-Kelly big on anything, and it never means big at all when the floor is uncertain.

Worked Example 4: The reinvestment runway: identical ROIC, opposite fortunes (India, ₹)

Synthetic. The arithmetic is the sustainable-growth identity g = ROIC × retention ratio, the engine under Akre's third leg and under the whole quality-compounder school.

Two Indian consumer franchises are identical on Akre's first two legs. Each earns a 25% ROIC on ₹100 crore of starting equity, and each is honestly run and debt-free. They differ only on the third leg, the reinvestment runway:

• Vriddhi Foods is early in its national rollout and can plough 100% of earnings back into new stores and capacity at the same 25%.
• Sthir Foods has saturated its niche and can reinvest only 20% of earnings at 25%, paying the other 80% out as dividends.

Step 1: the growth each can actually compound. Earnings grow at g = ROIC × retention ratio. Vriddhi: 25% × 1.00 = 25%/yr. Sthir: 25% × 0.20 = 5%/yr. The quality is the same and the runway isn't, and that gives a five-fold gap in growth between two businesses that look equally excellent on a one-year ROIC screen.

Step 2: fifteen years of compounding. Starting from ₹100 crore of earnings each:

• Vriddhi: 100 × 1.25¹⁵ ≈ ₹2,842 crore, earnings ~28.4× today's.
• Sthir: 100 × 1.05¹⁵ ≈ ₹208 crore, earnings ~2.08× today's.

After fifteen years Vriddhi earns ~13.7× as much as Sthir (2,842 ÷ 208), from an identical starting point and an identical return on capital. The entire difference is the runway.

Step 3: Akre's "nature of the return." Over a long hold, a full reinvestor's shareholder return approximates the business's ROIC, here ~25%, compounded internally and tax-deferred. Sthir mailed you 80% of a slow-growing stream as dividends, which you have to redeploy yourself at whatever the market then offers, and after tax. So Akre puts reinvestment at the centre of the stool. A great business that can't reinvest is a good income stock. A great business that can is the eighth wonder, and the difference doesn't show up in this year's ROIC at all.

The trap it disarms: "both earn 25% ROIC, so both are equally good compounders" is false. ROIC is one leg. The reinvestment runway, the reinvestment engine from your returns-on-capital work, is what turns a high return into a high growth rate.

Worked Example 5: Focus vs diversification, in numbers: Kelly, independent bets, and the ruin cap (US, $)

Synthetic. The concentration debate has a mathematical spine, and here it is, computed. It's anchored in $ for a US book, but the logic works in any currency.

Part A: Kelly sizes by edge (the case for concentration). A US investor has done deep work on one business. Over two years, they judge, it returns +60% if the thesis holds (probability 0.55) and −50% if it breaks (0.45). For a bet that gains a fraction W or loses a fraction L, which needn't be the whole stake, the growth-optimal Kelly fraction takes the general form f* = (p·W − q·L) / (W·L) = p/L − q/W:

f* = (0.55 × 0.60 − 0.45 × 0.50) / (0.60 × 0.50) = (0.330 − 0.225) / 0.30 = 0.35, so full Kelly says stake 35% of capital.

A real edge with a bounded loss earns a large position. That's the mathematical case for concentration, and it's why Buffett's "diversification is protection against ignorance" is more than bravado.

Part B: why you wouldn't actually stake 35%. Full Kelly assumes you know p = 0.55 exactly. You don't, and overestimating your edge turns full Kelly into a ruin machine. It also tolerates savage drawdowns. So you bet fractional Kelly, and quarter-Kelly here ≈ 8.75%.

Notice why the figure was even that large: the downside was capped at −50%. With a total-loss tail (leverage, fraud, a floor that gives way), the growth-optimal bet would collapse. Pabrai's "verify the floor" rule is what permits the big bet in the first place.

Part C: the number of names is not the number of bets. Now for the diversification half, in numbers. Take a $1,000,000 book of ten equal $100,000 positions, and suppose each name independently carries a 15% chance of a −70% idiosyncratic blow-up (fraud, a lost patent, a key-customer exit).

• The expected drag is the same either way. 10 × $100,000 × 0.15 × 0.70 = $105,000, or −10.5% expected, whether the risks are independent or shared, because expectation is linear.
• The tail is where they part ways. If the ten really are independent bets, the chance that all ten blow up together is 0.15¹⁰ ≈ 6 in a billion, which is never. The realistic bad case is one or two names down (expected number = 1.5), each costing 7% of the book. That's a bruise. There's an 80.3% chance, 1 − 0.85¹⁰, that at least one name blows up in a given window. That's routine, and survivable.
• Suppose instead all ten share one hidden factor, such as ten US high-multiple software names or ten positions all levered to the same rate or credit cycle. When that factor turns, all ten take the −70% hit at once. That's a −70% portfolio loss, $700,000 gone, with the whole 15% probability landing on that single catastrophic outcome. Ten names, and one bet held ten times.

The lesson, and the trap it disarms. Same names, same expected loss, opposite survival odds. The "diversified" ten-name book with hidden correlation carries a 15% chance of a −70% ruin-class drawdown. The independent ten-name book carries essentially none. "I own ten (or thirty) names, so I'm diversified" is the trap. Count independent risk factors instead of tickers. The IL&FS and DHFL contagion took the entire Indian NBFC complex down as one bet.

Here's the hard cap on concentration from the debate. Size every position, and every shared factor, so that neither being wrong nor the factor turning can ever be fatal. That's the ruin constraint, and you'll formalize it when you get to position sizing.

Practice set

Problems 1 to 4 are guided: the framework is applied for you, with scaffolding. Problems 5 to 9 are independent, and 10 to 12 are under exam conditions (12 minutes each, closed book). Currency is ₹ crore or currency-neutral units as stated, and the hurdle is 12% unless noted. All companies are synthetic unless named.

P1 (guided · match the school). Match each description to the school it best fits: Fisher, Lynch, Greenblatt (special situations), Greenblatt (Magic Formula), Klarman, Marks, Pabrai, quality-compounder. (a) "Rank the whole market on EBIT/EV and EBIT/tangible-capital, buy the top 30, rotate yearly." (b) "The subsidiary just got spun off and index funds are dumping it at any price." (c) "Nothing clears my conservative value bar, so I'm 40% cash and comfortable waiting." (d) "Talk to the customers and ex-employees, and especially to the competitor who most fears them." (e) "Buy a few businesses with high ROC and a long reinvestment runway; then do nothing."

Solution. (a) Greenblatt, Magic Formula, the two-factor mechanical rank. (b) Greenblatt, special situations: forced, index-driven selling in a spin-off. (c) Klarman, absolute value, with cash as the residual of high standards. (d) Fisher, scuttlebutt, including the "which rival do you most fear?" question. (e) Quality-compounder: Akre's stool, Smith's "do nothing," Sleep's hold. Marks would be the answer if the description were about reading the cycle's temperature.

P2 (guided · Lynch's six categories). Classify each company into one of Lynch's six categories and state the one question that decides the thesis. (a) A large FMCG cash cow growing 5% with a fat, safe dividend. (b) A steel producer whose profits swing with the construction cycle. (c) A small QSR chain growing 25% by opening stores. (d) A beaten-down airline that has just refinanced and changed management. (e) A holding company trading at half the market value of its listed stakes.

Solution. (a) Slow grower. The deciding question: is the dividend safe and generous enough to justify owning this over a bond? (b) Cyclical. Where are we in the cycle, and is the low trailing P/E a peak-earnings trap? Never value it on trailing P/E. (c) Fast grower. Is the growth profitable and self-financed, and is it repeatable? In other words, do new stores still clear the return hurdle? (d) Turnaround. Can it survive (balance sheet, liquidity), and is the fix real and underway? (e) Asset play. What are the stakes really worth, and what catalyst (a demerger or buyback) will close the discount?

P3 (guided · PEG). Company X has a P/E of 30, EPS growth of 20% and a dividend yield of 1.5%. (a) Compute the simple PEG and the dividend-adjusted PEG. (b) Interpret both on Lynch's scale. (c) State two things PEG can't see that would change your verdict.

Solution. (a) Simple PEG = 30 ÷ 20 = 1.5. Dividend-adjusted = (20 + 1.5) ÷ 30 = 21.5 ÷ 30 = 0.72. (b) A simple PEG of 1.5 reads somewhat expensive, against fair ≈ 1.0. The dividend-adjusted 0.72 reads poor, or expensive, on Lynch's scale (>2 attractive, <1 poor). Both say "not cheap for the growth." (c) PEG can't see two things. (i) How durable the 20% growth is. If it lasts 15+ years at high incremental ROIC, a PEG of 1.5 may be a bargain, because you're paying for duration. If it lasts two years, it's dear. (ii) Return on capital and balance-sheet risk. A 20%-grower that funds growth with debt or dilution, or earns below its cost of capital, is worth far less than the PEG suggests. PEG is a flag, and it can't do a valuation's job.

P4 (guided · Magic Formula head-to-head). Two names. Alpha: EBIT 90, EV 600, tangible capital 300. Beta: EBIT 75, EV 500, tangible capital 500. (a) Compute earnings yield and ROC for each. (b) Which ranks better on the combined Magic-Formula logic, and why?

Solution. (a) Alpha: EY = 90/600 = 15.0%, ROC = 90/300 = 30.0%. Beta: EY = 75/500 = 15.0%, ROC = 75/500 = 15.0%. (b) The two are equally cheap, both at a 15% earnings yield, so the ROC leg breaks the tie. Alpha ranks better because it earns twice the return on the capital it employs, 30% against 15%. With cheapness held equal, the Magic Formula buys the higher-quality business. That's "a good company at a bargain price," with Alpha the better company at the same price.

P5 (independent · Fisher scuttlebutt plan). You're researching a listed Indian specialty-chemicals maker that screens well, with ROCE of 22%, a clean balance sheet and rising exports. Design a concrete scuttlebutt plan. Name four sources you'd contact or read, the specific question you'd ask each, and which of Fisher's fifteen points each one addresses.

Solution (model · many valid). (1) Two customers or formulators abroad, or their procurement teams, reached through an expert network like Tegus or AlphaSense, or through industry contacts. Ask: "Is this supplier's molecule qualified into your product, and how hard would it be to switch?" That covers points 1 (runway and stickiness) and 5 and 6 (margin durability through switching cost). (2) A competing chemicals maker. Ask: "Which Indian supplier do you most fear on cost and quality, and why?" That's the Fisher master-question. It covers points 3 and 4 (R&D or process strength, and sales strength). (3) Glassdoor and ex-employees of the plants. Ask: "Is management honest; why did you leave; is attrition high?" That covers points 7 to 9 and 15 (labor/management/integrity). (4) The annual report's related-party and contingent-liability notes, plus the auditor's report (CARO), cross-checked with what the channel and distributors say. Ask: "Do the reported margins and receivables match what the channel says?" That covers point 10 along with points 14 and 15 (controls/candor/integrity), and it's the quality-of-earnings overlay from your filing and forensic work. The plan has one rule: seek the disconfirming voice (the rival who's winning, the ex-employee who left angry) and skip the company's friends. That's the bridge to "steelman the bear" in the deep dive.

P6 (independent · absolute vs relative value). An analyst pitches a bank at "0.9× book, versus the sector at 1.4×. Clearly cheap." Klarman would push back. (a) What's the distinction the pitch ignores? (b) What would make this an absolute-value buy instead of a relative-value trap? (c) Connect your answer to why Klarman holds cash.

Solution. (a) The pitch is relative ("cheaper than peers"), and that says nothing about whether the bank is cheap in absolute terms. The whole sector can be overpriced. Or this bank's book value can be overstated by under-provisioned bad loans, so 0.9× of an inflated book is no discount at all. (b) An absolute-value buy needs a conservative estimate of intrinsic value that shows a real discount. Adjust book for realistic credit losses and weigh sustainable ROE against the cost of equity, using the excess-return method. Then add a margin of safety wide enough to survive being wrong on asset quality, ideally with a catalyst. (c) If, after conservative adjustment, neither this bank nor its peers clears the absolute bar, Klarman owns none of them and holds the cash. Refusing to buy the "least overpriced" option is how his cash balance rises at frothy times, and how he avoids riding a whole sector down.

P7 (independent · Marks on second-level thinking and the cycle). A market shows profitless IPOs mobbed, credit spreads at record tights, "this time is different" in the headlines, bad news shrugged off, and your neighbors bragging about easy gains. (a) Which bull-market stage is this, and what is Marks's risk verdict? (b) Give the first-level and the second-level thought on a "great, beloved company" in this market. (c) What posture does Marks prescribe, and what does he explicitly not claim?

Solution. (a) Bull-market stage 3, where "everyone believes it will get better forever." Marks's verdict is that risk is at its maximum, because it feels lowest. Price has destroyed the margin of safety. (b) First-level: "It's a great, beloved company, so buy it." Second-level: "It's a great company, but everyone knows it. The price already embeds a flawless future, so expectations are too high and it's an overpriced stock. Reduce or avoid." (c) He prescribes a defensive lean: raise the bar, hold some cash, size down. Calibrate aggressiveness to where the cycle is. What he doesn't claim is that he can predict the timing of the turn. He's reading the temperature, which gives you the odds. He isn't forecasting the date. Markets can stay euphoric longer than you can stay solvent.

P8 (independent · Dhandho asymmetry). A stock trades at ₹80. Over ~2 years there's a 50% chance it reaches ₹200, a 35% chance of ₹90 and a 15% chance of ₹50. (a) Compute the expected value and expected return. (b) Is this a Dhandho bet? Apply both tests. (c) What single fact would you most need to verify before believing the asymmetry?

Solution. (a) EV = 0.50 × 200 + 0.35 × 90 + 0.15 × 50 = 100 + 31.5 + 7.5 = ₹139, a +73.75% expected return. (b) Test 1: is EV well above price? ₹139 against ₹80, so yes. Test 2: is the downside survivable? The worst case, ₹50, is −37.5%, a wound but not a wipeout, at only 15% probability. Both pass, so yes, it's a Dhandho bet: "heads I win, tails I don't lose much." (c) You need to know that the downside floor is real, that ₹50 is a true worst case and not an illusion masking leverage, fraud or terminal decline that could take it to zero. Dhandho's entire edge rests on the capped downside being true. Verify the balance sheet, the asset backing, and the accounting before you trust "tails I don't lose much." Pabrai's own worst losses came where the assumed floor gave way.

P9 (independent · the full Magic-Formula ranking). Rank this six-name universe by the Magic Formula. Compute both factors, rank each, combine, and state which you'd buy and which you'd avoid.

Solution. Compute earnings yield (EBIT/EV) and ROC (EBIT/capital):

EY ranks and ROC ranks, in order:

• EY: P(25)→1, T(20)→2, R(17.5)→3, U(15)→4, Q(10)→5, S(8)→6
• ROC: Q(60)→1, U(50)→2, T(40)→3, P(20)→4, R(17.5)→5, S(10)→6

Buy the top of the list, P and T (combined 5), then Q and U (combined 6). That's a basket of cheap-and-good. Avoid S (combined 12). It's expensive and low-return, a bad business at a bad price, and not even a value trap. Look at what the table teaches. Q is the highest-quality name (ROC 60%), but its rich price (EY rank 5) keeps it out of the top. P is the cheapest (EY 25%) and balanced enough on quality to top the list. T is the most balanced (ranks 2 and 3). The formula rewards the combination, and never one factor alone.

P10 (exam · 12 min · concentration and hidden correlation). An investor holds 25 stocks and calls the portfolio "well diversified." On inspection, 18 of them are Indian NBFCs and housing-finance companies. (a) Is this portfolio diversified? Explain using "independent bets." (b) State one condition under which a 10-stock portfolio would be less risky than this 25-stock one. (c) What does this imply for how you should count diversification?

Solution. (a) No. The 18 NBFC and HFC names share the same dominant risk factors: funding cost and liquidity (ALM), the credit cycle, and RBI regulation. So they'll rise and fall together when that factor turns, the way the IL&FS and DHFL contagion took the whole complex down at once. Eighteen names on one factor is roughly one bet held eighteen times. The portfolio has perhaps 8 independent bets, the 7 non-NBFCs plus the NBFC "bet," and not 25. (b) A 10-stock portfolio of uncorrelated businesses carries less factor risk than 25 names all levered to a single sector, because independent bets drive risk and ticker count doesn't. Think one bank, one FMCG, one IT-services exporter, one pharma, one cyclical, etc., across unrelated factors. (c) Count diversification by the number of independent risk factors and ignore the number of names. Then size every position, and every shared factor, so that being wrong on it can't permanently impair the portfolio.

P11 (exam · 12 min · apply the lens). A profitable, wide-moat US payments company trades at 35× earnings, grows ~15%, earns 40%+ ROIC, and has a long reinvestment runway. Give the verdict three schools would reach. (a) Terry Smith or the quality-compounder school. (b) A strict Graham deep-value investor. (c) Howard Marks. Then state the single question that reconciles them.

Solution. (a) Quality-compounder (Smith): it's a "good company" par excellence (high ROC, high margins, cash-generative, self-funded growth, long runway). Buy it, tolerate the high multiple because compounding overwhelms the entry price over a long hold, and then do nothing. (b) Graham deep-value: pass. At 35× earnings there's no margin of safety by any statistical measure. There's no asset protection, and the price bakes in years of flawless growth. Graham would call it speculation and refuse to call it investment. (c) Marks: ask the second-level question. Is the quality already fully in the price? A wonderful business can be a poor stock if expectations are too high, so the verdict depends on what the 35× multiple implies about duration and whether that's beatable. The reconciling question: how many years of the 15% growth and 40% ROIC does the 35× multiple actually require, and does the moat evidence support that many? That's the reverse-DCF or implied-CAP test from your valuation and moat work. That single number separates Smith's "fine price for quality" from Graham's "reckless," and it answers Marks's "what's priced in."

P12 (exam · 12 min · Kelly sizing and its limits). A bet wins +100% with probability 0.55 and loses 100% with probability 0.45. (a) Compute the full-Kelly fraction. (b) What would you actually stake, and why less than full Kelly? (c) Why does this bet's total-loss downside cap how large the position can be, regardless of the Kelly number, and how does a Dhandho bet differ?

Solution. (a) f* = (b·p − q)/b with net odds b = 1, p = 0.55, q = 0.45: f* = (0.55 − 0.45)/1 = 0.10, so full Kelly = 10% of capital. (b) I'd stake fractional Kelly, half (5%) or a quarter (2.5%). Full Kelly assumes I know p = 0.55 exactly. I don't, and overestimating edge turns full Kelly into a ruin machine. Full Kelly also tolerates brutal drawdowns most investors can't hold through. (c) The downside is a total loss (−100%), so even a modest position that goes wrong is gone for good. Make enough such bets and you'll eventually hit ruin, which ends compounding forever. That's ergodicity. A bet you can lose entirely has to be sized small however favorable the odds look. A Dhandho bet differs right here. Its downside is capped and survivable, say −40% instead of −100%, so the same conviction can be sized larger without risking ruin. That's why Pabrai hunts capped-downside asymmetry instead of fair-odds gambles, and why "verify the floor" decides everything.

P13 (guided · the reinvestment runway). Prakash Consumer (synthetic) earns a 22% ROIC. (a) If it retains 100% of earnings and reinvests them all at 22%, at what rate do its earnings compound, and what multiple of today's earnings will it reach in 12 years? (b) If instead it can reinvest only 30% (paying out 70%), what are the growth rate and the 12-year multiple? (c) State in one sentence what this shows about Akre's third leg.

Solution. (a) g = ROIC × retention = 22% × 1.00 = 22%, and in 12 years earnings reach 1.22¹² ≈ 10.9× today's. (b) g = 22% × 0.30 = 6.6%, and 1.066¹² ≈ 2.15×. (c) The same ROIC (22%) produces a 10.9× or a 2.15× outcome depending entirely on how much can be reinvested at that rate. The reinvestment runway sets the growth rate, and the return alone doesn't. That's the sustainable-growth identity, and Akre's most-forgotten leg.

P14 (guided · Kelly with a capped downside). A bet returns +50% if right (p = 0.55) and −40% if wrong (q = 0.45). (a) Using the general Kelly formula f* = (p·W − q·L)/(W·L) (with W, L the win and loss fractions), compute the growth-optimal fraction. (b) What would you actually stake? (c) Why does the −40% (not −100%) downside let you size larger than an even-money total-loss bet?

Solution. (a) f* = (0.55 × 0.50 − 0.45 × 0.40)/(0.50 × 0.40) = (0.275 − 0.180)/0.20 = 0.095/0.20 = 0.475, so full Kelly ≈ 47.5%. (b) Fractional Kelly: half ≈ 23.75% or a quarter ≈ 11.9%, because you don't know p exactly and full Kelly tolerates brutal drawdowns. (c) With the loss capped at 40% of the stake instead of 100%, each losing round leaves 60% of the position intact to compound. So the growth-optimal bet is far larger than for a wager that can vanish entirely. That's why Pabrai hunts capped-downside asymmetry, where "bet big" stays survivable. The trap: plugging the numbers into the total-loss form (b·p − q)/b would understate the bet, because that formula assumes you lose the whole stake when you lose.

P15 (independent · Dhandho asymmetry, US $). A US special-situation stock trades at $40. Over three years there's a 55% chance of $90, a 30% chance of $48 and a 15% chance of $20. (a) Compute the expected value and expected return. (b) Apply both Dhandho tests. (c) Name the single fact you must verify.

Solution. (a) EV = 0.55 × 90 + 0.30 × 48 + 0.15 × 20 = 49.5 + 14.4 + 3.0 = $66.9, a +67.25% expected return. (b) Test 1: is EV ($66.9) well above price ($40)? Yes. Test 2: is the downside survivable? The worst case, $20, is −50%, a wound but not a wipeout, at only 15% probability. Both pass, so it's a Dhandho bet. (c) That the $20 floor is real, and not an illusion masking leverage, fraud or terminal decline that could take it to zero. The whole asymmetry rests on the downside really being bounded, and Pabrai's own worst losses came where the assumed floor gave way.

P16 (independent · independent bets vs hidden correlation). A book holds eight equal-weighted positions, each independently carrying a 20% chance of a −60% blow-up. (a) Find the expected number of blow-ups, and the probability that all eight blow up at once if they're independent. (b) Suppose instead all eight share one factor that turns with probability 20%. What is the portfolio loss when it turns, and what's its probability? (c) Which book is "one bet," and what should you count instead of names?

Solution. (a) Expected blow-ups = 8 × 0.20 = 1.6. P(all eight, independent) = 0.20⁸ ≈ 0.0000026, about 3 in a million, which is never. There's also a 1 − 0.80⁸ ≈ 83.2% chance that at least one blows up. That's routine, and each one costs 7.5% of the book. (b) With one shared factor, all eight fall 60% together when it turns, so the portfolio loses 60%, and the whole 20% probability lands on that one catastrophic outcome. (c) The factor-sharing book is one bet held eight times, a 20% chance of a −60% drawdown. Count independent risk factors instead of tickers, and size each factor so it can't ruin you.

P17 (independent · compounding at high ROIC, US $). Vertex Systems (synthetic) reinvests all earnings at a 30% ROIC for ten years, while a broad index compounds at the ~12% hurdle. (a) Work out the multiple of today's value each reaches in ten years, per $1 invested. (b) How many times the index does the compounder deliver? (c) What one condition must hold for the 30% to be real, and which school insists on it?

Solution. (a) Vertex: 1.30¹⁰ ≈ 13.79×. Index: 1.12¹⁰ ≈ 3.11×. (b) 13.79 ÷ 3.11 ≈ 4.4× the index over the decade, which is the arithmetic behind "your long-run return approximates the business's ROIC." (c) The 30% has to be reinvestable at that rate for the full ten years (the runway) and durable (the moat must hold ROIC above the hurdle). The quality-compounder school (Akre/Smith/Sleep) insists on it, and it's Phase 4's moat-and-reinvestment test worn as a portfolio philosophy. The trap: assuming a high current ROIC will persist and reinvest for a decade without checking the moat's falsifiers. That's the "durability misjudgment" that sinks compounder theses.

P18 (exam · 12 min · the concentrated Dhandho bet, end to end). A focus investor finds a bet with the stock at ₹200. Over three years there's a 50% chance of ₹500, a 35% chance of ₹240 and a 15% chance of ₹120. (a) Compute the expected value and expected return, and state the worst-case loss. (b) For sizing, model it as a binary "+150% with p = 0.50, −40% with q = 0.50" and compute full Kelly with f* = (p·W − q·L)/(W·L). Give quarter-Kelly too. (c) The investor's other twelve holdings are all Indian NBFCs. He calls the book "diversified" and plans to make this a 20% position. Critique that in two sentences, naming the rule that actually caps the size.

Solution. (a) EV = 0.50 × 500 + 0.35 × 240 + 0.15 × 120 = 250 + 84 + 18 = ₹352, a +76% expected return. The worst case, ₹120, is −40%, which is survivable, at 15% probability. (b) f* = (0.50 × 1.50 − 0.50 × 0.40)/(1.50 × 0.40) = (0.75 − 0.20)/0.60 = 0.55/0.60 ≈ 0.917, so full Kelly ≈ 91.7%. That's absurdly large, and it's the signature of a big edge with a capped downside. Quarter-Kelly ≈ 22.9%. (c) Even quarter-Kelly's ~23% is too large. The reason isn't this bet's own math. It's that the other twelve names are one NBFC factor bet, so the book already carries enormous hidden concentration and has only a handful of independent bets, nowhere near thirteen. The cap is the ruin and independent-bets constraint: size so that no single name and no shared factor can permanently impair the portfolio. Here that means fixing the NBFC over-concentration first, and sizing the new bet well below what its standalone Kelly suggests.

Applied mini-project: the Schools Panel

Task. Pick ONE company you can get good data on. It can be one Indian or one US name from the starter bank (Asian Paints, DMart, Titan, HDFC Bank, Bajaj Finance; Costco, Visa, Adobe, Moody's, Copart), or another you understand. Write a three-page "Schools Panel" memo that looks at that one company through five different lenses, then gives your own synthesized view.

This is the core of the week. The same facts support different verdicts depending on the school, and knowing which lens you're using, and which one is actually yours, is what makes you an analyst. Without it you're just collecting opinions.

Data to pull. For an Indian name, get the latest annual report and investor presentation from the company site, the screener.in ten-year page (for EBIT, EV, ROCE, growth, the multiple), and the shareholding and related-party disclosures. For a US name, get the latest 10-K + proxy from EDGAR, stockanalysis.com or TIKR for the ten-year series, and the company's investor page. For the Magic-Formula factors, compute EBIT/EV and EBIT/(net working capital + net fixed assets) by hand, then cross-check the idea against Greenblatt's free screen at magicformulainvesting.com (US only). For the quality-compounder lens, skim one Nomad letter from Sleep and one Fundsmith letter from Smith to calibrate the standard.

Required contents.

1. A one-line business summary, plus the ten-year ROIC/ROCE-vs-hurdle and growth series. That's the shared factual base every school argues from.
2. Five school verdicts, each a short paragraph stating (i) the question that school asks, (ii) its verdict on this company, and (iii) the one fact that most drives that verdict. Use five of these: Fisher (run three of the fifteen points + a scuttlebutt sketch), Lynch (classify the category + a PEG), Greenblatt (Magic-Formula factors computed + one special-situation angle if any), Klarman (absolute value + is there a margin of safety?), Marks (what's priced in / second-level view + cycle read), quality-compounder (Akre's three legs / Smith's three steps).
3. The disagreement, named. Identify the single question on which the schools most diverge for this company. It's almost always price and duration, or how reliable the future is. Then show how answering it reconciles them. The reverse-DCF or implied-duration test usually does that.
4. Your synthesized verdict and a candid note to yourself. Say which lens you find most persuasive for this company. Then, separately, say which school your temperament would actually let you carry out under a 40% drawdown. These often aren't the same answer, and if they differ, say so.

Scoring rubric (score yourself candidly, pass at ≥11/14):

Reading & resources

• Philip Fisher, Common Stocks and Uncommon Profits: ch. 2 ("What Scuttlebutt Can Do for You") and ch. 3 ("What to Buy: The Fifteen Points"); Part III on when to sell. The founding text of growth-quality investing and scuttlebutt. [Paid] [Intermediate]
• Peter Lynch, One Up on Wall Street: Part II ("Picking Winners"): the six categories, the two-minute drill, and the P/E-vs-growth (PEG) discipline; Beating the Street (1993) for dozens of worked category cases. The most readable investing book ever written. [Paid] [Beginner]
• Joel Greenblatt, You Can Be a Stock Market Genius: chs. 3–4 on spin-offs and restructurings (why structural inefficiency exists); the rest for the special-situations toolkit. [Paid] [Intermediate]
• Joel Greenblatt, The Little Book That Beats the Market: the Magic Formula, its logic, and Greenblatt's own honesty about the multi-year lag; run the free screen at magicformulainvesting.com (US universe). [Paid book / Free screen] [Beginner]
• Seth Klarman, Margin of Safety: out of print and expensive, but PDFs circulate widely; read for absolute value, the sources of margin of safety, and "the institutional imperative." If you read one thing, read the chapters on risk and on why to hold cash. [Hard to find] [Advanced]
• Howard Marks, The Most Important Thing (2011) and Mastering the Market Cycle (2018): second-level thinking, risk as permanent loss, the pendulum, the three stages. And the Oaktree memos (oaktreecapital.com/insights): free, decades deep, the best free resource here; start with "The Most Important Thing," "Risk," and "You Bet!" [Books Paid / Memos Free] [Beginner→Intermediate]
• Mohnish Pabrai, The Dhandho Investor: the asymmetry framework, the nine Dhandho principles, cloning, and Kelly-style sizing; his many lectures and interviews are free on YouTube (search his Google/university talks). [Book Paid / Talks Free] [Beginner]
• Nick Sleep & Qais Zakaria, the Nomad Investment Partnership letters: widely archived and free; read for "scale economies shared," destination analysis, and one of the finest demonstrations of long-horizon concentration ever written. Non-negotiable for the quality-compounder lens. [Free] [Intermediate]
• Terry Smith, Fundsmith Annual Letters and Annual Shareholder Meeting videos (fundsmith.co.uk): free; "buy good companies, don't overpay, do nothing," with a clear quality screen; collected in Investing for Growth (book). [Letters Free / Book Paid] [Beginner→Intermediate]
• Chuck Akre: his talks and interviews (search "Akre three-legged stool") and Akre Capital commentary; the clearest short statement of the reinvestment-runway idea. [Free] [Intermediate]
• Concentration vs diversification, primary sources: Buffett's shareholder letters (1993 on diversification-vs-ignorance; 1996), Munger's talks on "focus investing," and Walter Schloss's "16 rules" (all free online); Pulak Prasad, What I Learned About Investing from Darwin (2023) for the modern concentrated-quality case with an Indian lens. [Mostly Free; book Paid] [Intermediate]
• India ties: Gautam Baid, The Joys of Compounding (a superb synthesis of these schools for the Indian investor) [Paid] [Beginner→Intermediate]; Saurabh Mukherjea / Marcellus writings and Coffee Can Investing (the Indian consistent-compounder school) [Paid] [Beginner]; Sanjay Bakshi's BFBV lectures and blog (behavioral value investing, India) [Free] [Intermediate].
• William Green, Richer, Wiser, Happier: long-form profiles of many of these investors (Pabrai, Marks, Sleep, and others); excellent for hearing each school in its own voice. [Paid] [Beginner]
• Going deeper. William Poundstone, Fortune's Formula: the narrative history of the Kelly criterion (Shannon and Thorp, the blackjack table to Wall Street); the most enjoyable way to internalise why edge-based sizing and the ruin constraint matter, before M6.03/M9.01 make them rigorous. [Paid] [Intermediate]
• Going deeper. Michael Mauboussin, Measuring the Moat and The Base Rate Book (Counterpoint Global / Morgan Stanley, free PDFs). The empirical backbone under the quality-compounder lens: how ROIC persists (and mostly fades) and the base rates for high growth actually enduring, the antidote to the "durability misjudgment" that sinks compounder theses and to a PEG or a 60× multiple taken on faith. [Free] [Intermediate]

The Modern Analyst's Addendum

Everything above teaches this skill from first principles, by hand. 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 nine schools as specialized lenses, Lynch's six categories and PEG, Greenblatt's Magic Formula ranking and special situations, Klarman's absolute value, Marks's cycle read, Pabrai's Dhandho asymmetry and the Kelly bridge, the quality-compounder school, and the concentration-vs-diversification debate",
  "use": "Two legitimate jobs, and one hard ban. (1) THE RANKING PIPELINE: the Magic Formula is a deterministic calculation with a checkable right answer, which makes it a good candidate for a tool to draft and you to unit-test — build the earnings-yield/return-on-capital ranker, then verify it against two or three names by hand before trusting it on a real universe. (2) CLONING RESEARCH: turn a disclosed public filing (a 13F, an Indian bulk- or block-deal disclosure) into a structured 'who owns what, and what do they say the thesis is' comparison table, which you then go investigate — Pabrai's own discipline (§4.7) is to clone the reasoning, never the ticker. THE BAN: never ask a tool which school a company 'fits' or to write the Schools Panel verdict for you — the nine schools disagree with each other by design (§4.1), and the mini-project's whole point is holding that disagreement yourself and finding which lens is actually yours; a single fluent synthesized answer erases the exact tension you are there to resolve.",
  "tools": ["Chat assistants — Claude, ChatGPT for drafting the ranking pipeline's code and for structuring disclosed-holdings comparisons", "A notebook or spreadsheet with pandas — where the Magic Formula ranks and the Kelly/EV arithmetic actually run and get checked", "magicformulainvesting.com's free US screen, screener.in's ten-year data pages, and SEC EDGAR's 13F full-text search — the primary sources any ranking or cloning claim must trace to"],
  "prompt": "Write a function that takes a table of company, EBIT, enterprise value, and tangible capital employed; computes earnings yield (EBIT/EV) and return on capital (EBIT/tangible capital); ranks each ascending by rank position (1 = best); sums the two ranks; and returns the combined ranking sorted best-first, showing every intermediate rank so I can check it against my own by-hand calculation. Do not decide which names to buy or comment on the business quality of any of them — ranking mechanics only.",
  "verify": "Reproduce at least two names' earnings yield and return on capital by hand from the source filing before trusting the pipeline on the rest — a single sign error, or an enterprise value built from stale net debt, silently reorders the whole list with no error message. For any cloned position, read the primary disclosure yourself and reconstruct why before it earns a place on your list (§4.7) — an unverified clone is a ticker, not a thesis, and you will not know when to sell it.",
  "diy": "The gate is unaided: you compute earnings yield and return on capital and combine the ranks by hand for a small universe, classify a company into one of Lynch's six categories and compute its PEG, build a Dhandho expected-value table and both forms of the Kelly fraction with a calculator, and state which school's lens applies to a given situation and why — no tool in the room."
}

Modern Data Analysis

By hand first. You ranked an eight-name universe on earnings yield and return on capital, computed simple and dividend-adjusted PEG, built a Dhandho expected-value table and both forms of the Kelly fraction, and quantified independent-vs-correlated blow-up risk in a ten-name book.

Keep the hand arithmetic. A ranking pipeline that returns a list with no visible ranks is a black box you can't audit. The warning about the Magic Formula's evidence exists because too many people trusted its headline number without ever looking inside it.

Today's workflow. Write the two-factor rank as a function over a real universe table, and have it return the ranks themselves instead of a single "score." Every name's earnings-yield rank and return-on-capital rank should be a visible output you can spot-check, the same way the first worked example does it by hand.

Tools & sources (IN + US). Use pandas for the ranking table and scipy.stats for the significance check below.

• India: screener.in's ten-year data pages for EBIT, enterprise value, and the balance-sheet lines that build tangible capital employed (net working capital + net fixed assets), plus the NSE/BSE bulk- and block-deal disclosures for the India side of a cloning exercise.
• US: magicformulainvesting.com's free screen for a ready-made comparison universe, SEC EDGAR's companyfacts for the raw EBIT and tangible-capital inputs, and EDGAR's 13F full-text search for disclosed institutional holdings.

# Magic Formula ranker as an auditable function, run on WE1's own 8-name universe -- R3
import pandas as pd
df = pd.DataFrame({
    "co":  list("ABCDEFGH"),
    "ebit": [105,150,80,200,60,120,90,108],
    "ev":   [500,1500,500,2500,250,800,1200,600],
    "cap":  [263,313,400,667,500,200,1000,216],
})
df["earnings_yield"] = df.ebit/df.ev
df["roc"] = df.ebit/df.cap
df["ey_rank"]  = df.earnings_yield.rank(ascending=False).astype(int)
df["roc_rank"] = df.roc.rank(ascending=False).astype(int)
df["combined"] = df.ey_rank + df.roc_rank
print(df.sort_values("combined")[["co","earnings_yield","roc","ey_rank","roc_rank","combined"]]
        .to_string(index=False, formatters={"earnings_yield":"{:.1%}".format, "roc":"{:.1%}".format}))

Verify. The pipeline reproduces the worked example's final order: H first (combined 5), A and F tied for second (combined 6), E fourth, and so on down to G last (combined 16), matching the table row for row. One cell deserves a flag, though it stays as printed. The table gives company A's return on capital as 40.0% and B's as 48.0%. Recomputed from the same stated inputs (EBIT ₹105 on tangible capital ₹263, and EBIT ₹150 on capital ₹313), the figures come to 39.9% and 47.9%. Those are cosmetic rounding slips, consistent with capital figures of 262.5 and 312.5 rounded to 263 and 313 for display. They change no rank and no conclusion in the worked example or in the related Magic Formula practice problems, and they're recorded here as a defect candidate instead of being corrected without notice.

Quantitative lens

The evidence section on the Magic Formula already warns that its ~30% headline is "legitimately debated" and that "much of it is just value × quality: known factors." Now make the mechanism behind that warning precise. One backtest can be wrong, sure. Searching many rule variants and reporting only the best one is worse. It manufactures apparent significance out of nothing, by construction, every time.

Best-of-200, pure noise, still "significant." Build 200 candidate trading strategies with different lookback windows, cutoffs or factor weightings, which is the ordinary process of "improving" a screen. Make every one of them worthless: five years of monthly returns (T=60) drawn from pure noise (mean zero, no edge whatsoever). Run a t-test against zero on each of the 200, and look only at whichever has the smallest p-value, the way a search for "the best variant" naturally does. Repeat that over 5,000 independent 200-strategy searches:

• In theory, the probability that at least one of the 200 noise strategies clears p < 0.05 by chance alone is 1 − (1 − 0.05)^200 = 99.996%, a near-certainty before a single real number is involved.
• The empirical rate across the 5,000 simulated searches is 100.0%. Every search turned up a "significant" strategy.
• On average, 10.0 of the 200 noise strategies individually cross p < 0.05 in any one search (200 × 0.05). Test 200 things at a 5%-per-test error rate and you get about 10 false alarms whether or not any of them has an edge.
• One concrete draw makes it vivid. In a single search, the best-of-200 noise "strategy" posted t = 3.66, p = 0.0005, a 23.7% annualized return and a Sharpe ratio of 1.65. That figure would headline a fund pitch, and it came from a random-number generator with zero underlying skill.

The fix, quantified. A Bonferroni correction tests each of the 200 at α = 0.05/200 = 0.00025 instead of 0.05, and that restores honesty. The empirical rate of the best-of-200 clearing the stricter bar falls to 5.18%, very close to the intended 5% family-wise error rate. It's the arithmetic version of what a real research process must do, and what a single backtest write-up never shows: report how many variants were tried, as well as the winner.

The toy has a real-world counterpart. McLean and Pontiff ("Does Academic Research Destroy Stock Return Predictability?", Journal of Finance, 2016) tracked a large set of published return-predicting characteristics. Their measured returns declined markedly out of sample and after publication. That's the same overfitting-and-crowding pattern this simulation isolates in miniature, on data where the only explanation available is chance and, later, arbitrage.

The simulation has limits. Its 200 "variants" are statistically independent. Real rule variants, such as a 6-month against a 12-month lookback or a 20-name against a 30-name cutoff, are highly correlated with each other because they mostly share the same underlying signal. So the effective number of independent tests in a real research process is smaller than 200, and the Bonferroni correction above is, if anything, conservative. The qualitative lesson survives regardless: searching and keeping the best inflates significance. The exact correction factor for a real search, though, needs the correlation structure as well as the count.

The simulation also says nothing about whether the actual Magic Formula has zero edge. The evidence section already grants it a real, if compressed, value-and-quality premium. All the simulation shows is that a headline backtest number, on its own, can't tell a real premium from the best of many noise draws. So you hand-check the top names, and you hold the strategy through its lean years instead of trusting the number alone.

Do it in code: before trusting any backtested rule (the Magic Formula, a Lynch category filter, a screen of your own), write down how many variants were searched to find it. Then either correct for that search (Bonferroni, or better, a pre-registered out-of-sample holdout) or treat the reported figure as an upper bound on what you should expect. See G1/QM1.02, G2/QD1.03 and G2/CS2.06.

Where this goes next: galaxy cross-links

• G1 · QM1.02 Hypothesis Testing & Regression: the formal machinery (p-values, multiple-comparison correction, false discovery rate) behind why a "best of 200" backtest needs a stricter bar than a single pre-registered test.
• G2 · QD1.03 Backtesting Engine: where walk-forward validation and an out-of-sample holdout turn the warning above into standing infrastructure rather than a one-time worry.
• G2 · CS2.06 Software Engineering & AI-Assisted Development: the ranking pipeline as tested, version-controlled code with unit tests against hand-computed names, not a notebook cell trusted on faith.

Teach it back & journal

Feynman prompt. Write a one-page explainer for a smart 15-year-old titled "There is no single 'right way' to invest: here are the main ways, and how to pick yours." Get three things across without jargon. First, the great investors disagree because each one is hunting a different kind of bargain: the great business hiding in plain sight, the thing everyone is dumping for silly reasons, or a whole market that's too greedy or too scared. Second, each way only works for a person with the right patience and nerve to do it. Third, there's one idea they all share: never pay so much that being wrong wipes you out, which is margin of safety. If your page needs the words "GARP," "asymmetry," or "second-level," you don't own it yet.

Journal prompt. Three entries. (1) Of the nine schools, which two did you feel an instinctive pull toward as you read? And be straight with yourself: is that because they fit your temperament, or because their recent returns have been good? (These are different reasons, and only the first is durable.) (2) Take the last stock you seriously considered (or own) and run it through three schools' questions: Lynch's category, Klarman's absolute-value bar, and Marks's "what's priced in?" Note whether the three agreed. Where they disagreed, which one were you actually using, and did you know that at the time? (3) Write, in one sentence each, the concentration policy you think you want and the one your temperament could actually survive in a 40% drawdown, and note the gap. You'll formalize the real answer when you build your portfolio policy.

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