Selling & Mistakes

Phase 9 · Week 73 · ~20 hours

Where it sits. Sell a compounder at 3× and you can forfeit about 49 times the gain you just banked. That arithmetic is why the exit half of a round trip needs a week of its own. Third module of the portfolio phase, and the one that closes the loop nobody else teaches. The entire industry (every book, every course, every research report) is about buying: what to buy, when to buy, how to value it before you buy. The other half of every round trip is at least as hard and far less taught: when to let go, and how to learn from the times you were wrong. M9.01 taught you to size a position (risk as probability × magnitude of impairment; fractional Kelly; the hidden-correlation trap). M9.02 taught you to build a portfolio (hurdle rates, opportunity cost, cash as a position, and your written concentration policy). Both of those are entry-and-construction disciplines. Selling is the exit discipline, and it is where the behavioral demons you catalogued in M6.04 do their most expensive work: loss aversion, the disposition effect, anchoring to cost, endowment, sunk cost. By the time you are deciding whether to sell you already own the thing, and ownership warps judgment. What follows gives you five legitimate reasons to sell (and a decision test for each), the discipline to run winners that most people clip too early, a tax-aware framework for the discretionary sells, and a taxonomy that turns every mistake into a permanent upgrade to your checklist. It feeds directly into M9.04, where selling becomes one station in the full written process.

Suggested hour budget. Core content: 8h · worked examples (the compounder-sold-early computation, the reverse-DCF margin-of-danger sell, the after-tax switch, the mistake classification): 3h · practice set (the sell/hold/trim case bank): 4.5h · applied mini-project (write pre-committed sell criteria for a real holding): 2.5h · flashcards + mastery check: 2h.

App link. This module feeds the Portfolio & Risk branch and, specifically, three instruments. First, the sell-decision case engine: a scenario in (position, price move, new facts, your original written thesis), a decision out (sell / hold / trim / add), scored against the five-reason framework and the fresh-money test, not against whether the price later went up, because you are graded on reasoning, never on outcome. Second, the disposition-effect and tribe drill: rapid vignettes where you tag the behavior (Rabbit, Assassin, Hunter, Raider, Connoisseur) and the bias driving it, the sell-side twin of the Phase 2 red-flag drills. Third, the decision-journal and calibration plumbing shared with M9.04 and the M6.04 calibration tracker. Every simulated sell requires a written reason keyed to one of the five, and the season-end review scores whether your sells (not just your buys) were process-consistent. The portfolio simulator's seeded cycles (dot-com 2000, 2008, 2020) are where "sell in the panic or hold on thesis?" stops being theoretical.

Learning objectives

By the end you can:

1. Explain why selling is structurally harder than buying: you own the asset (endowment, loss aversion, sunk cost, and the disposition effect all now bind), the cost basis contaminates the decision as an anchor, and the sell decision is chronically under-instrumented because almost everyone journals buys and no one journals sells.
2. State and apply the five legitimate reasons to sell, each with its one-question decision test: (1) the thesis is broken; (2) you recognized a mistake; (3) the price is far above intrinsic value; (4) a materially better opportunity exists; (5) the position has grown beyond your risk tolerance. Then correctly classify which two are tax-blind (non-negotiable) and which three are tax-modulated (discretionary).
3. Distinguish a broken thesis from a fallen price using the written thesis in your decision journal as the arbiter, separating a load-bearing pillar actually breaking from thesis-noise (a weak quarter) or a thesis merely playing out slowly. Distinguish too a mistake (wrong at purchase) from bad luck (right process, bad outcome).
4. Compute the cost of selling a compounder too early, the 3× versus 100× arithmetic, and explain why "no one ever went broke taking a profit" is one of the most expensive clichés in investing, using the positive skew of equity returns (Bessembinder) and Munger's first rule of compounding.
5. Diagnose the disposition effect operationally through Freeman-Shor's five tribes (The Art of Execution): Rabbits and Raiders as the losing archetypes, Assassins, Hunters, and Connoisseurs as the winning ones. Explain why execution after the buy, not hit rate, separated his elite managers.
6. Pre-commit sell criteria at the moment of purchase, writing the thesis-break triggers, the overvaluation zone, and the maximum-size/trim rule into the buy note before you own the position. Argue the case for thesis-based versus mechanical price-stop selling for a fundamental investor.
7. Run a tax-aware sell decision across India (LTCG/STCG, the 12-month line) and the US (long- versus short-term, the 1-year line), computing the after-tax hurdle a switch must clear and explaining deferred tax as an interest-free loan that rewards patience.
8. Build and use a mistake taxonomy (commission versus omission; analytical versus behavioral versus bad-luck) and feed each learnable mistake through the M6.05 root-cause loop into a permanent checklist line, while correctly refusing to "learn" from genuine bad luck.

Prerequisites & connections

Builds on. M6.04 (Behavioral Finance) is the indispensable prerequisite: the disposition effect (riding losers, cutting winners), the fresh-money test ("ignoring what I paid, would I buy this at today's price?"), and the biases that bind hardest at the sell decision (anchoring to cost, loss aversion, sunk cost, endowment) are all taught there and operationalized here. This module is §4.8 of M6.04 (the behavioral edge) turned into a specific exit discipline. M6.05 (Checklist & Temperament) gives you the decision journal and the mistake-to-checklist loop; the journal is the arbiter of the "is the thesis broken?" question, and the mistake taxonomy you build here is the sorting stage that feeds that loop. M6.03 (Mental Models) supplies opportunity cost / hurdle rates (sell reason 4), process-over-outcome (the mistake-versus-bad-luck distinction), and inversion. M9.01 (Risk & Position Sizing) defines risk as probability × magnitude of impairment and gives the sizing discipline behind sell reason 5 (trim to your cap). M9.02 (Portfolio Construction) supplies your written concentration policy (the maximum single-position weight that reason 5 enforces) and the opportunity-cost hurdle that reason 4 tests against. M3 (Valuation) is the machinery for reason 3: intrinsic value, margin of safety, and the reverse DCF (M3.05) that reads what a price implies. M6.01/M6.02 supply the philosophical spine: Graham's margin of safety (whose reverse is the margin of danger), Fisher's three reasons to sell, Terry Smith's "do nothing," and the quality-compounder school that makes running winners rational.

Feeds forward. M9.04 (The Process + Phase Capstone) folds selling into the single running loop of checklist → journal → quarterly review → calibration, and requires a simulator season in which your sells are journaled and scored, not just your buys; this module authors the sell station of that process. Phase 10 (the two capstone deep dives, M10.01–M10.02): every institutional-grade thesis you write there must end with an explicit sell-criteria section (business, why it is a good business, the variant perception, valuation and expected return, and the pre-registered sell triggers), exactly the artifact §4.10 teaches you to write. The decision journal and calibration threads that run to the end of the program are fed here by the discipline of recording every sell reason and later scoring it. Competency C9 in the master map, "an investor's temperament and process", is substantially certified by this module and completed in M9.04.

The one-sentence version. Buying is where you place the bet; selling is where you either protect the fortune or throw it away, and the difference between the two is a small set of pre-committed, thesis-based, tax-aware rules that override the ownership-warped instinct to cut your flowers and water your weeds.

Core teaching content

4.4 Sell reason 1: the thesis is broken

This is the primary, cleanest reason to sell, and the one most often confused with its impostor: the price fell. A broken thesis means that the fundamental case you wrote down when you bought, the specific load-bearing things that had to be true for this to be a good investment, has factually stopped being true. Not "the stock is down"; the reasons you owned it are gone. The moat you bought is being competed away (share is bleeding structurally, not for one quarter); the growth runway you underwrote has closed (the category is saturating, the regulatory door shut); the unit economics you modeled have inverted (contribution margin has gone structurally negative); capital allocation has turned abusive (the promoter is expropriating minorities; cash is being torched on empire-building at returns below the cost of capital); the balance sheet you thought was safe has become a solvency question. When a pillar like that breaks, the business is no longer the one you analyzed, and the correct response is to exit at whatever price, up or down, because the tax and the loss are both irrelevant next to owning something whose reason-to-exist has evaporated.

The decision test, and the journal as arbiter. The single question is: has a load-bearing pillar that I wrote down at purchase actually, factually broken? The word "wrote down" is doing all the work. You cannot answer this honestly from memory, because memory is corrupted by exactly the biases now operating on you. Confirmation bias will reinterpret bad news as noise to protect the position; hindsight will rewrite what you "always knew"; the narrative you construct now, under the influence of ownership, is untrustworthy. The decision journal from M6.05 is the arbiter precisely because it is a fossil of your pre-outcome reasoning, un-editable, written before the position could warp you. You go back to the buy note, read the three-to-five things you said had to be true and the disconfirming evidence you pre-registered as thesis-breakers (§4.10), and you check reality against the written record. Did that break? Not "do I feel nervous?" but: did the specific pillar you named break? This is the whole reason the program has made you journal every thesis since week 53: so that at the sell decision you have an honest witness.

Three things a broken thesis is not, each a trap:

• It is not thesis-noise. A weak quarter, a temporary margin dip from input-cost inflation, a demand air-pocket, a macro wobble, a one-off litigation charge: these are the ordinary volatility of owning a business, and reacting to them is how Raiders and panic-sellers are made. A pillar is load-bearing; a soft quarter usually is not. The test: does this news change the five-year owner-earnings power of the business, or just this year's reported number?
• It is not "the thesis is playing out slowly." A thesis that is intact but taking longer than you hoped is a time problem, not a break, and often the Hunter's opportunity rather than the Rabbit's trap. Impatience dressed as discipline is still impatience. Distinguish "it isn't working yet" from "it won't work."
• It is not the price falling with the thesis intact. This is the impostor that fools everyone. A price that falls while every pillar holds is Mr. Market's mood, and it is a buy signal (Hunter), not a sell signal. The fresh-money test cuts straight through: if the thesis holds, you would buy more at the lower price, so of course you do not sell.

Worked example (India, illustrative, modeled on the FMCG/consumer archetype). Suppose two years ago you bought a mid-cap packaged-foods company on a three-pillar thesis, written in your journal: (P1) a widening distribution moat, with direct reach expanding faster than peers; (P2) mid-single-digit volume growth with premiumization lifting realizations; (P3) very high ROCE from an asset-light, negative-working-capital model, funding growth internally. The stock is now down 35%. Is the thesis broken? Go to the journal and test each pillar against the filings. Scenario A, noise: the fall came after two quarters of margin compression driven by a spike in palm-oil and packaging costs, but volumes are still growing mid-single-digits, direct reach is still expanding, and ROCE is intact; management is absorbing input costs temporarily rather than losing share. No pillar broke. This is thesis-noise plus Mr. Market, and the fresh-money test says add (Hunter), not sell. Scenario B, break: the fall came because a wave of direct-to-consumer and regional challengers has taken structural share for six straight quarters, volumes are now declining despite price cuts, the company is raising advertising spend just to stand still (ROCE falling), and distribution reach has stopped expanding. P1 and P2 have broken and P3 is going. The business is no longer the compounder you underwrote. Exit, regardless of the 35% loss, and do not average down (that would be the Rabbit funding a broken thesis). Same price move, opposite decision, and the journal is what tells them apart.

4.6 Sell reason 3: the price is far above intrinsic value

Reasons 1 and 2 are about the business being wrong. Reason 3 is about the price being wrong, specifically too high. It is the exact mirror image of the margin of safety that governed your purchase (M6.01, M3). You buy when price sits well below your intrinsic-value range (a margin of safety); symmetry suggests you sell when price sits well above it (a margin of danger). When the market is paying you far more than the business is worth, the risk-reward has inverted: your downside (a re-rating back toward value) now exceeds your upside, and the rational owner takes the market's generous offer. This is Mr. Market serving you on the other side: the euphoric mood that overpays is your cue, just as the despairing mood that underpays was your cue to buy.

The decision test: read what the price implies. Do not eyeball the multiple; reverse it. Run the reverse DCF from M3.05: solve for the growth, margin, and duration of excess returns that today's price requires, and ask whether that implied story is one you would underwrite with fresh money. The precise test for a sell is stronger than "it looks expensive": does today's price imply a forward return below my hurdle rate even under my bull case for the business? If even a generous, everything-goes-right continuation of the business cannot deliver your required return from this price, because the starting multiple is so high that it must compress faster than earnings can grow into it, then the margin of danger is real and you trim or exit. If, instead, the business can plausibly grow into the multiple and still clear your hurdle, it is not a reason-3 sell, however high the P/E looks.

Two caveats that make reason 3 the most abused of the five. Fisher (whose three reasons to sell were essentially our reasons 2, 1, and 4) deliberately excluded "it went up / it looks overvalued" from his routine sell reasons, and he was largely right to be suspicious, for two reasons:

• The compounder caveat. For a genuine compounder, a business reinvesting at a high ROIC, intrinsic value is a fast-moving target that grows every year at roughly the ROIC. A stock that looks 20% overvalued today can be undervalued in eighteen months without the price moving at all, because value grew into it. Selling a real compounder because the trailing P/E is high is the Raider's error (§4.9) and the most expensive mistake in the exit discipline. Reserve reason-3 selling on a compounder for extreme, mania-level overvaluation, where the price implies a return below your hurdle even under the bull case, as the test above demands. Not for the ordinary richness that quality always carries. Munger: never interrupt compounding unnecessarily.
• The tax caveat. Reason 3 is tax-modulated (§4.11). Selling a long-held winner triggers a tax bill that both reduces the capital you redeploy and forfeits the interest-free compounding of the deferred tax. A position that is "15% overvalued" frequently does not clear the after-tax-plus-transaction-cost-plus-reinvestment-risk hurdle of actually switching out. Overvaluation has to be large to justify paying tax to exit a good business.

The practical resolution: reason 3 fires cleanly for average or mediocre businesses priced richly (where value is not compounding fast enough to bail out the multiple, so you sell closer to fair value) and for any business at genuinely extreme, bull-case-defying valuations; it fires reluctantly and rarely for genuine compounders at merely-rich prices, where the default is to hold and let value catch up (or, at most, trim the oversized bit under reason 5).

Worked example (US/consumer-compounder archetype, illustrative). You own a high-quality consumer franchise that has re-rated to 70× earnings. Bull case: it can grow earnings ~14% a year for a decade. Reverse the multiple with a simple ten-year frame (full arithmetic in Worked Example 2). Earnings at 14% for ten years grow to 3.71×. But a business growing 14% cannot sustain a 70× multiple forever; a defensible terminal multiple for that growth-and-quality is ~40×. So the price factor over ten years is 3.71 × (40/70) = 2.12×, a 7.8% annual return, below a 12% hurdle, and that is under the bull case. The multiple must compress faster than even great execution can grow into it: margin of danger confirmed → trim or exit (modulated by tax). Now contrast the same business, same 14% growth, bought at 45×: price factor = 3.71 × (40/45) = 3.30×, a 12.7% annual return, at/above hurdle, so hold. Identical business and identical operating performance; the only difference is the entry multiple, which is the whole of reason 3. The lesson: reason 3 is not "sell because it's a great company that got expensive". It is "sell because, at this price, even the bull case can't pay you your hurdle."

4.7 Sell reason 4: a materially better opportunity (opportunity cost)

Reason 4 is the one that turns a portfolio from a static collection into a living thing, and it is also the one most easily abused into churn. Every position you hold is, implicitly, a statement that this is a better use of the capital than any alternative available to you (M9.02's hurdle-rate logic; Munger's insistence that every decision be measured against your next-best alternative). When a genuinely better idea appears, one with a materially higher expected risk-adjusted return, holding the incumbent has an opportunity cost, and selling A to buy B is rational. This is how a disciplined investor upgrades quality over time: capital continuously migrates toward the best available risk-adjusted return.

The decision test: is B's forward risk-adjusted return higher than A's by more than the sum of (i) the tax and transaction costs of switching, (ii) a conviction discount for knowing B less well than A, and (iii) a margin for the general truth that action is more often wrong than inaction? The phrase carrying the weight is materially better. A marginal edge (B looks like it might return 11% versus A's 10%) does not clear the bar, because three real frictions eat marginal edges alive:

• Frictional cost. Selling A realizes tax (§4.11) and pays transaction costs; the capital that actually arrives in B is smaller than the capital that left A. If A carries a large embedded gain, the switch can start with a 10–13% capital handicap (Worked Example 3), so B must out-compound A by that much just to break even, before it can add a rupee of value.
• The conviction discount ("the devil you know"). You have owned and studied A; you know its risks, its management, its footnotes, its behavior in a bad quarter. B is comparatively unknown, and unknown risks are the ones that hurt. B's estimated return is measured with more error than A's, so it deserves a discount: a margin of safety on your own conviction, not just on price. Switching from a well-understood holding to a less-understood one on a thin estimated edge is a classic way to trade a known good for an unknown bad.
• The churn / overconfidence tax. Activity feels like productivity and scratches the itch to do something, but trading is inversely related to returns (Barber–Odean, M6.04): the more you trade, the more you pay in costs, taxes, and mistimed switches, and the more you expose yourself to overconfidence. Most proposed switches are the overconfidence tax in disguise. The default answer to "should I switch?" is no, overridden only by a wide, durable gap.

The synthesis: reason 4 is real but should fire rarely, and only for a wide, durable, risk-adjusted edge that survives tax, the conviction discount, and your own itch to act. It is the legitimate engine of portfolio upgrading and, simultaneously, the favorite rationalization of the compulsive trader; the frictions above are what separate the two.

Worked example (cross-market, illustrative). You hold A: a fairly-valued quality compounder you know deeply, bought years ago, now carrying a large long-term gain, with an estimated forward return of ~10% a year. B appears: a comparable-quality business at a more attractive price, estimated forward return ~13%, but you have studied it for only a few weeks. Should you switch? First the tax handicap (India, A up 4× from cost, LTCG 12.5%): selling A leaves ~90.6% of the market value to redeploy (Worked Example 3), so B starts ~10.3% behind, roughly a 2%/year drag over a five-year horizon. That alone nearly erases B's 3-point estimated edge (13% − 10% = 3 points, minus ~2 points of tax drag = ~1 point). Now apply the conviction discount: your 13% estimate for a business you have known for three weeks deserves a haircut relative to your 10% estimate for one you have owned for years. After tax and conviction discount, the edge is inside the noise. Do not switch; keep studying B, and if it stays cheap and your conviction rises, revisit. Had B instead offered ~18% versus A's 10%, the 8-point edge would survive both frictions comfortably. That is the wide gap reason 4 exists for.

4.9 The great error: selling compounders too early

Now the most expensive mistake of all, expensive enough to deserve its own section: clipping winners, selling a compounder long before its thesis breaks, its value is exceeded, a better idea appears, or it breaches your size cap. It is a violation of none of the five reasons, which is exactly why it is so insidious: it feels responsible. It is the Raider's whole personality, and it hides behind the most comforting cliché in investing: "no one ever went broke taking a profit."

That cliché is one of the most costly sentences an investor can believe, and here is the arithmetic that proves it. Take a business compounding intrinsic value, and over time price, at ~26% a year (a rough doubling every three years; the realized rate of a genuine multi-decade winner). You buy ₹1,00,000 of it at ₹100 a share (1,000 shares). Five years in it has tripled; the cliché fires ("no one went broke taking a profit") and you sell your 1,000 shares for ₹3,00,000, banking a ₹2,00,000 gain. A win. Except the business keeps compounding. Twenty years from purchase, at 26% a year, the shares are worth ₹100 × 1.26²⁰ ≈ ₹10,170 each, and the position would have been worth ₹1.02 crore. By "taking a profit" at 3×, you captured 2.9% of the eventual value and forfeited roughly ₹98.7 lakh, about 49 times the ₹2 lakh gain you were so pleased to bank. You did not go broke. You merely gave away a fortune, which the cliché conveniently never mentions. And the tax made it worse: you paid LTCG on the ₹2 lakh gain for the privilege of amputating your best idea, and now you must find another 33-bagger to turn that ₹3 lakh back into ₹1 crore. Except the 33-bagger was the thing you just sold.

Why the whole game lives in the few big winners. This is not a cherry-picked anecdote; it is the mathematical structure of equity returns. Hendrik Bessembinder's study Do Stocks Outperform Treasury Bills? (2018) examined ~26,000 US stocks from 1926–2016 and found that the majority, about 58%, failed to beat one-month Treasury bills over their lifetimes, and that the entire net dollar wealth creation of the US stock market above T-bills was attributable to the best-performing ~4% of firms (illustrative of Bessembinder 2018). Equity returns are violently positively skewed: a tiny number of enormous, long-duration winners generate all the wealth, and everything else nets to roughly nothing. The investing implication is stark and directly about selling: if you are skilled or lucky enough to own one of the few stocks that drives everything, the only mistake that truly matters is selling it too early. Clipping your winners at 3× guarantees you can never own a 30-bagger, which guarantees you forfeit the skew, which guarantees mediocrity no matter how good your buying is. The Connoisseur's patience is not a personality trait; it is the rational response to the shape of the return distribution.

The masters say it plainly. Munger: "The first rule of compounding: never interrupt it unnecessarily." Peter Lynch: "Selling your winners and holding your losers is like cutting the flowers and watering the weeds." Philip Fisher held wonderful companies for decades and taught that the best time to sell one is "almost never." Terry Smith's entire three-step process ends in "do nothing" (M6.02). Nick Sleep held a concentrated book of scale-economies-shared compounders for years without trimming. None of this means "never sell". It means do not sell a compounder for any reason other than the five, and be especially skeptical of reason 3 (overvaluation) on a business whose value is compounding into the price. The discipline is asymmetric: be an Assassin with mistakes and broken theses, and a Connoisseur, verging on inert, with intact compounders.

The hardest thing in investing is not finding a great business. It is continuing to own one through the years of volatility, boredom, and "surely it's expensive now" that separate a 3-bagger from a 100-bagger.

4.10 Pre-committing sell criteria at purchase: the sell checklist

Everything above shares a structural weakness if left to the moment: the sell decision is made while you own the position, when endowment, loss aversion, and the disposition effect are all pulling. The fix is the same one M6.04 §4.8 prescribed for every behavioral error: pre-commitment. You make the sell decision, as far as possible, before you own the thing, in the cold, neutral state of the buy analysis, and you write it into the buy note. Ulysses lashed himself to the mast before the Sirens sang; you write your sell triggers before the position can warp you.

Concretely, every buy note in your decision journal must end with a sell-criteria section, and Phase 10's institutional theses require exactly this. It has four parts:

1. The thesis, as testable pillars. The three-to-five things that must remain true for this to stay a good investment (the moat, the growth runway, the unit economics, the capital allocation, the balance-sheet safety). These are what reason 1 tests against later. Write them as falsifiable statements, not adjectives: "direct distribution reach grows faster than the top three peers," not "great distribution."
2. The pre-registered thesis-break triggers. The specific, observable disconfirming evidence that would signal a pillar breaking: "sell if volume growth is negative for three consecutive quarters," "sell if net-interest-margin compression coincides with accelerating loan growth (the M5.01 warning)," "sell if the promoter pledges more than X% of holding," "sell if share is lost to challengers for four straight quarters." Pre-registering these is what lets you tell a real break from confirmation-biased noise later. You decided, in advance and unemotionally, what would count.
3. The overvaluation zone. A rough price or multiple at which the margin of danger (reason 3) would fire even on your bull case, with the reverse-DCF-implied return falling below your hurdle. Not a precise sell price (valuation is a range), but a zone that prompts a re-underwrite.
4. The maximum size and trim rule. The position's cap under your concentration policy (reason 5), stated in advance so a big gain does not tempt you to rationalize an oversized weight.

Thesis-based triggers versus mechanical price stops: the honest debate. There are two schools on pre-commitment, and an evenhanded module presents both. The mechanical-stop school (the Assassins, and most traders) sets a hard price rule ("sell any position down 25% from cost") and executes it without discretion; its virtue is that it guarantees the left tail is capped and removes all in-the-moment judgment (and thus all in-the-moment bias). Its vice, for a fundamental investor, is severe: a price stop keyed to your cost basis sells you out of exactly the intact-thesis drawdowns that are Mr. Market's gifts (the Hunter's buy), converts temporary volatility into permanent loss, and re-anchors the whole decision to the cost basis you were trying to delete. The thesis-based school (most concentrated business-owners) pre-registers fundamental triggers (part 2 above) and treats price moves as a prompt to re-examine the thesis, not as an automatic sell. A 30% drop is a signal to re-read the filings and re-run the fresh-money test, whose answer might be "add," not "sell." Its virtue is that it sells for the right reason (the business), not the wrong one (the price); its vice is that it demands honest judgment in the moment, which is exactly what bias corrupts, so it only works if the triggers were genuinely pre-registered and the journal is honestly consulted. The program's stance, consistent with a concentrated fundamental approach: use thesis-based triggers as the primary discipline, with price moves as prompts-to-review rather than auto-sells, while acknowledging that a disciplined price/time stop (the Assassin's rule) is a legitimate and often superior tool for the loser/mistake case, where the danger is precisely that you will not act. Cut mistakes mechanically; hold and re-underwrite intact-thesis drawdowns judgmentally.

4.11 Tax-aware selling

Reasons 3, 4, and 5 are discretionary, and their honest cost-benefit has to include tax, because tax converts a "worth it" pre-tax switch into a "not worth it" after-tax one more often than beginners expect. Teach the economics once, then the two jurisdictions.

The economics: deferred tax is an interest-free loan. When you hold a position with an unrealized gain, the tax you would owe on selling is money you have not yet paid, and it stays fully invested, compounding for you, until you sell. Selling early hands that money to the government now instead of letting it work for years; it is, in Buffett's framing, as if the government owns a silent, interest-free stake in your unrealized gain that becomes yours to compound for as long as you defer. This is a first-order reason great investors are reluctant sellers of winners: patience is behavioral discipline and tax-efficient compounding at once. Every discretionary sale voluntarily terminates that interest-free loan.

The after-tax switching hurdle. Because selling realizes tax, the capital that redeploys into idea B is smaller than the market value of A, so B must out-earn A by the tax handicap just to break even. Quantify it: if A has quadrupled (cost 100, value 400, gain 300) and long-term tax is 12.5% (India), you pay 37.5 and redeploy 362.5, a handicap of 400/362.5 − 1 ≈ 10.3%. B must generate ~10.3% more terminal wealth than A would have, before it adds anything, roughly 2%/year of extra return over a five-year horizon. That is why reason 4 demands a material edge (§4.7).

India (as of mid-2026, check against the current Finance Act and incometaxindia.gov.in; rates were revised in the July 2024 Budget). For STT-paid listed equity shares and equity-oriented mutual funds: the holding-period line is 12 months. Short-term gains (held ≤ 12 months, Section 111A) are taxed at 20%; long-term gains (held > 12 months, Section 112A) at 12.5% on aggregate long-term gains exceeding a ₹1.25 lakh annual exemption, without indexation. The two consequences for selling: the 12-month line creates a strong reason to cross it before selling a winner where the thesis allows (paying 12.5% instead of 20%), and the modest 12.5% long-term rate makes India's after-tax switching hurdle lower than the US's, but still real.

United States (as of mid-2026, check against IRS Topic 409 and current brackets). The holding-period line is 1 year. Short-term gains (held ≤ 1 year) are taxed as ordinary income (federal brackets up to 37%), while long-term gains (held > 1 year) are taxed at 0% / 15% / 20% depending on taxable income, plus a possible 3.8% Net Investment Income Tax, plus state tax where applicable. The ordinary-versus-long-term gap is enormous. A high earner selling a short-term winner can pay ~40%+ versus ~20% long-term, so crossing the 1-year line is worth far more in the US than in India (Worked Example 3 shows a short-term switch starting ~38% behind versus ~13% long-term). Note also the wash-sale rule: a realized loss is disallowed if you buy the same or a substantially identical security within 30 days before or after the sale, which is relevant to reason-2 loss-taking and tax-loss harvesting. India has no identical wash-sale rule but has other anti-avoidance provisions; near year-end, tax-loss harvesting (realizing a loss to offset gains while retaining or re-entering the position) is a legitimate tax action, distinct from the five investment reasons, and only sensible if you would hold the position anyway.

The non-negotiable exception. Tax modulates reasons 3, 4, and 5 only. It does not touch reasons 1 and 2: when the thesis is broken or you have recognized a mistake, you sell regardless of tax, because the impairment risk of holding a broken position dwarfs any tax saving, and "I don't want to pay the tax" is the tax tail wagging the risk dog. Indeed, a reason-2 sale often comes with a tax silver lining: a realized loss to harvest against gains elsewhere.

4.12 Building your mistake taxonomy

Selling and mistakes belong in one module because the sell decision is where most mistakes are finally recognized, and recognition is worthless unless it is converted into learning. The conversion engine is the mistake-to-checklist loop from M6.05 (isolate a real mistake → find the root cause, not the symptom → write one pointed question that would have caught it → keep it only if it generalizes). But before a mistake can enter that loop it has to be classified, because the classification tells you whether there is anything to learn at all and what kind of fix it needs. That classification is your mistake taxonomy, and you build it along two axes.

Axis 1: commission versus omission.

• Errors of commission are mistakes of doing: a bad buy, a bad sell, averaging into a loser, over-sizing from overconfidence. They are visible and painful because they leave a mark on the P&L, and so they get all the attention.
• Errors of omission are mistakes of not doing: the great business you understood and didn't buy; the compounder you sold at 3× and never repurchased; the fat pitch you let go by. Buffett and Munger have repeatedly said their largest mistakes are omissions, the failure to buy more of what they understood, precisely because these leave no line on the P&L and so escape scrutiny entirely. A disciplined taxonomy forces omissions into the record ("did not act on a thesis I had correctly formed"), because the invisible mistakes are usually the most expensive ones, and clipping a winner too early (§4.9) is an error of omission, the failure to keep owning, masquerading as a prudent commission.

Axis 2: analytical versus behavioral versus bad-luck.

• Analytical. The analysis was wrong: you misjudged the moat, mis-modeled unit economics, missed a red flag, overpaid on a flawed valuation. Fix: a better analytical checklist line (a question that would have surfaced what you missed).
• Behavioral. The analysis was right but you behaved wrongly: you panic-sold at the bottom, anchored to cost, rode a loser (Rabbit), clipped a winner (Raider), chased, or over-sized from overconfidence. Fix: a structural defense (a pre-commitment, a journal prompt, an environment change; M6.04 §4.8), not more analysis, because more analysis was never the problem.
• Bad-luck. The analysis and the behavior were sound, and an irreducibly unforeseeable event caused the loss (a genuine black swan, a fraud with no detectable footprints, an exogenous shock no diligence could have priced). Fix: none, and this is the crucial output of the taxonomy. Genuine bad luck must not generate a checklist line, because there was nothing to catch; "learning" from it is the resulting/outcome-bias trap (M6.03) that breeds superstition and overfits your process to the last war. The taxonomy's most valuable service is granting you permission not to learn from true bad luck, while demanding rigorous honesty that the "bad luck" verdict is earned, since most losses that feel like bad luck are analytical or behavioral misses in disguise.

The neglected category: sell mistakes. Because the sell decision is under-instrumented (§4.1), the taxonomy must explicitly include mistakes of selling, which most investors never record: sold a compounder too early (Raider: behavioral, omission); sold in a panic at the bottom (behavioral, commission); sold on a high trailing P/E despite intact compounding (analytical + behavioral); failed to trim an oversized winner that then blew up (analytical, omission of risk management); averaged down into a broken thesis (behavioral, commission). Journaling your sells with the same discipline as your buys is what surfaces this whole neglected class, and it is why M9.04 requires every sell to be recorded with its reason and later scored.

How it feeds forward. Each learnable mistake, analytical or behavioral, commission or omission, runs through the M6.05 loop to its root cause and yields at most one pointed, general checklist line; bad-luck losses are logged and closed with no line. Over years this taxonomy is what turns your decision journal from a diary into a compounding instrument: the same scar never cuts twice, and your checklist grows one earned line at a time (M6.05's "short on purpose; a killer list of six vetoes you actually run beats a hundred you admire and skip").

Common mistakes & how experts think differently

Anchoring the sell to the cost basis. "I'll sell when it's back to what I paid," "I never sell at a loss," "it's up 40%, I should book something." Every one of these keys the forward decision to a backward, irrelevant number. Experts delete the cost basis and run the fresh-money test: the only question is whether they would buy this business, at this price, at this weight, today. Up or down is not an input to the decision.

Believing "no one ever went broke taking a profit." The most expensive cliché in the module. You do not go broke; you forfeit the fortune, and since a few big winners drive the entire portfolio (Bessembinder), clipping them is the one error that guarantees mediocrity. Experts run winners like Connoisseurs and reserve their selling energy for losers that are mistakes.

Confusing a fallen price with a broken thesis. The impostor that fools everyone. A price fall with the thesis intact is a buy (Hunter); only a broken pillar is a sell (reason 1). Experts consult the written journal, not their nerves, and can state exactly which pre-registered pillar broke, or admit that none did.

Averaging down into a mistake to "lower the average cost." The seductive logic that doubles a bet on a false premise and turns a small mistake into a large one: the Rabbit funding a broken thesis. Experts add only to intact, strengthening theses and are Assassins with recognized mistakes: cut fast, never reinforce.

Selling a compounder because the P/E looks high. Reason 3 abused. For a business compounding value at a high ROIC, "expensive" today is often "cheap" in eighteen months without the price moving. Experts apply reason 3 cleanly only to average businesses or genuinely mania-level prices, and demand that the price fail even the bull case before selling a real compounder, and then weigh the tax.

Letting the tax tail wag the risk dog, in both directions. Beginners either ignore tax entirely (churning away their after-tax returns on marginal switches) or let it paralyze them (refusing to sell a broken thesis "because of the tax bill"). Experts modulate reasons 3–5 by tax but sell reasons 1–2 tax-blind, and treat deferred tax on a winner as an interest-free loan worth protecting.

Naive mechanical rebalancing. Selling this year's winners to "rebalance" into this year's losers, on a calendar, can be the disposition effect institutionalized: cutting flowers to water weeds. Experts rebalance for risk (reason 5, trim the oversized) and thesis (reasons 1–2), not to force a position back to an arbitrary equal weight regardless of merit.

Not writing sell criteria at purchase. The root structural failure: making the most bias-compromised decision (the sell) with none of the apparatus (thesis, triggers, size cap) that the buy got. Experts pre-commit: every buy note ends with testable pillars, pre-registered break triggers, an overvaluation zone, and a size cap, decided before ownership can warp them.

Treating every loss as bad luck (or every loss as skill error). Over-attributing to luck kills learning; over-attributing to skill breeds superstition and overfitting. Experts classify honestly on the taxonomy, learn ruthlessly from analytical and behavioral misses, and grant themselves permission not to learn from genuine, earned bad luck.

The round-trip. Riding a winner all the way up and all the way back down because no overvaluation zone (reason 3) or size cap (reason 5) was ever set: the mirror image of clipping too early. Experts hold the asymmetric discipline: run intact compounders, but trim genuine mania (reason 3) and cap concentration (reason 5), so a great winner is neither amputated early nor round-tripped to nothing.

Worked examples

Worked Example 1: The cost of selling a compounder at 3× (full computation)

Setup. Buy ₹1,00,000 of a compounder at ₹100/share = 1,000 shares. It compounds price/value at 26%/year. You sell at 3× after ~5 years, invoking "no one ever went broke taking a profit." Quantify what the decision costs against a 20-year hold.

Step 1 — the proceeds at 3×. 1,000 shares × ₹300 = ₹3,00,000; realized gain ₹2,00,000.

Step 2 — after-tax proceeds. LTCG at 12.5% on the ₹2,00,000 gain (ignore the ₹1.25 lakh exemption for a clean per-position figure, or apply it: taxable ₹75,000 → tax ₹9,375). Un-exempted tax = ₹25,000 → net ≈ ₹2,75,000. You now need to redeploy this into something as good — but you just sold the best thing you had found.

Step 3 — the value forgone. Held 20 years at 26%: ₹100 × 1.26²⁰ = ₹100 × 101.7 = ₹10,170/share, so the 1,000 shares would be worth ₹1,01,72,000 (~₹1.02 crore).

Step 4 — the damage. Forfeited value = ₹1.02 crore − ₹3 lakh = ₹98.7 lakh, which is ≈ 49× the ₹2 lakh gain you banked. You captured 2.9% of the eventual value. To turn your ₹2.75 lakh after-tax proceeds back into ₹1 crore you would need to find a fresh ~37-bagger, i.e., replace the exact 33-bagger you just sold. Lesson: the compounder was the wealth event; "taking a profit" amputated it. The cliché is true (you didn't go broke) and ruinous (you gave away the fortune) at once.

Worked Example 2: Reverse-DCF margin of danger: when a high P/E is a sell and when it isn't

Setup. A high-quality franchise; bull-case earnings growth 14%/year for 10 years; hurdle 12%. Decide sell-versus-hold at two entry multiples, 70× and 45×, assuming a defensible terminal multiple of 40× for 14% growth of that quality.

Step 1 — earnings growth factor. 1.14¹⁰ = 3.707× over ten years (same for both cases; the business is identical).

Step 2 — price factor = earnings growth × multiple change.

• At 70×, exit 40×: 3.707 × (40/70) = 3.707 × 0.571 = 2.118× over 10 years → annual return = 2.118^(1/10) − 1 = 7.8%.
• At 45×, exit 40×: 3.707 × (40/45) = 3.707 × 0.889 = 3.295× over 10 years → 3.295^(1/10) − 1 = 12.7%.

Step 3 — decide against the 12% hurdle. At 70×, even the bull case yields 7.8%, below hurdle → trim/exit (margin of danger), modulated by tax. At 45×, the same business yields 12.7%, at/above hurdle → hold. Lesson: reason 3 is not "great company got expensive"; it is "at this price even the bull case can't pay my hurdle." Identical operating performance; the entry multiple is the entire sell decision, which is why you must reverse the price rather than eyeball the P/E, and why on a true compounder you demand the bull case fail before selling.

Worked Example 3: The after-tax switching hurdle (India and US)

Setup. You hold A (cost 100, market value 400, a 4× long-term gain) and are tempted to switch into B. Compute the capital handicap the switch starts with in three tax cases, and the annual edge B needs over a 5-year horizon.

India, long-term (12.5%). Tax = 12.5% × 300 = 37.5 → redeploy 362.5. Handicap = 400/362.5 − 1 = 10.3%; annualized over 5 years = 1.103^(1/5) − 1 = ~2.0%/year. B must beat A by ~2 points/year just to break even.

US, long-term (15%). Tax = 15% × 300 = 45 → redeploy 355. Handicap = 400/355 − 1 = 12.7%; ~2.4%/year over 5 years.

US, short-term (37% ordinary). Tax = 37% × 300 = 111 → redeploy 289. Handicap = 400/289 − 1 = 38.4%, a crippling ~6.7%/year over 5 years. Lesson: the after-tax hurdle is real and, in the US short-term case, decisive, which is why you (a) demand a material edge for any switch (reason 4), and (b) cross the 1-year / 12-month line before selling a winner whenever the thesis allows. Reasons 1–2 override all of this; reasons 3–5 must respect it.

Worked Example 4: Classifying five mistakes on the taxonomy

Classify each on both axes (commission/omission; analytical/behavioral/bad-luck) and state whether it generates a checklist line.

1. Bought an NBFC without reading the ALM gap; it defaulted. Commission · Analytical. Line: "Read the ALM/funding mix before the growth story for any lender." (Learnable.)
2. Sold a 3-bagger that became a 30-bagger; never repurchased. Omission · Behavioral (Raider). Line: "Do not clip an intact compounder; apply reasons 1–5 or hold." (Learnable, and the most expensive kind.)
3. Panic-sold the whole book near the March-2020 low; thesis intact. Commission · Behavioral. Line/structure: pre-commit "no selling in a drawdown without a written thesis-break; check prices less." (Learnable: structural fix, not analytical.)
4. Owned a company whose CFO ran a sophisticated fraud with no detectable footprints in the filings; it collapsed. Commission · Bad-luck (if honestly undetectable). No line, but first stress-test the "undetectable" claim (were there governance/related-party/auditor-turnover flags? If so, reclassify as analytical). (Not learnable if the verdict is earned.)
5. Understood a great compounder deeply, wrote the thesis, but never bought it. Omission · Behavioral/analytical (failure to act on correct analysis). Line: "When conviction is high and the pitch is fat, size in — omission is a real cost." (Learnable, invisible on the P&L.) Lesson: the taxonomy sorts the learnable (1,2,3,5) from the genuinely unlearnable (4), prevents both the "it was all bad luck" evasion and the "add a line for everything" overfit, and drags invisible omissions (2, 5) into the record where they can be fixed.

Worked Example 5: Selling an intact compounder to "diversify" (US $): the opportunity cost, tax included

Setup. You own $100,000 of a quality compounder (thesis intact, comfortably within your size cap) bought years ago at a $25,000 cost basis; it is compounding intrinsic value (and, over time, price) at ~18%/year. You are tempted to sell and move the money into a diversified index fund you expect to return ~8%/year "to be safe." US long-term rate 15% + 3.8% NIIT = 18.8%; horizon 15 years. What does the switch cost?

Step 1 — the capital that actually redeploys. Gain = 100,000 − 25,000 = $75,000; tax = 18.8% × 75,000 = $14,100; you redeploy only $85,900, not $100,000. The tax drag lands before the new investment earns a cent.

Step 2 — the two terminal wealths. Hold: 100,000 × 1.18¹⁵ = 100,000 × 11.974 = $1,197,375. Switch: 85,900 × 1.08¹⁵ = 85,900 × 3.172 = $272,489.

Step 3 — the damage. Forgone = 1,197,375 − 272,489 = $924,885, and holding ends 4.39× the switch. Two forces compounded against you: the $14,100 of tax that never got to work, and (the larger effect here) the 18% − 8% = 10-point return gap over fifteen years.

Step 4 — which reason fires? None of the five. Thesis intact (not 1/2); no reverse-DCF margin of danger (not 3); an 8% index is not a materially better risk-adjusted return than an ~18% compounder, so reason 4 fails badly; and the position is within your cap (not 5). Lesson. Selling an intact compounder "to diversify" is the Raider's error (§4.9) with a tax bill stapled on. You forfeit the positive skew (Bessembinder) and pay to do it. If genuine concentration risk is the worry, the disciplined tool is to trim to your cap (§4.8), which keeps the compounder compounding, not to exit into an average asset.

Worked Example 6: Expected value of holding through a drawdown vs selling (India ₹)

Setup. A position, after a sharp fall, is quoted at ₹8,00,000 (assume it sits near your cost, so selling triggers negligible tax; we isolate the investment decision). You have done the journal work but remain genuinely uncertain whether a load-bearing pillar has broken. Your honest, journal-tested assessment: 65% the thesis is intact and it recovers to fair value ₹15,00,000 over your horizon; 35% it is genuinely broken and drifts to a residual ₹3,00,000. Hold or sell?

Step 1 — the expected value of each choice. EV(sell) = ₹8,00,000 (cash today). EV(hold) = 0.65 × 15,00,000 + 0.35 × 3,00,000 = 9,75,000 + 1,05,000 = ₹10,80,000. EV(hold) exceeds EV(sell) by ₹2,80,000 → hold, on this assessment. Note what did not enter: the size of the fall itself (orthogonality, §4.1). Only the forward probabilities and payoffs did.

Step 2 — the break-even probability. Solve p × 15 + (1 − p) × 3 = 8 (₹ lakh): 12p + 3 = 8 → break-even p = 5/12 ≈ 41.7%. Hold whenever your journal-tested probability that the thesis is intact exceeds ~42%; below that, selling now beats holding on expected value.

Step 3 — the ruin adjustment (why EV alone is not enough). Suppose the "broken" branch is not ₹3,00,000 but ≈ 0: a fraud or a funding run in which the equity goes to zero (M5.02; M9.01's magnitude-of-impairment). Re-solve p × 15 + (1 − p) × 0 = 8 → break-even p = 8/15 ≈ 53.3%. The ruinous downside lifts the intact-probability you need from 41.7% to 53.3%, because you cannot average your way out of a zero. Risk is probability × magnitude, not probability alone.

Lesson. An EV frame disciplines the genuinely uncertain hold/sell, on two conditions. First, the probability must be journal-tested against the written pillars, never a rosy number you invent to avoid crystallizing a loss (loss aversion and confirmation bias will supply one on request). Second, when a branch is a ruinous zero, weight the magnitude, not just the average. And once a pillar is confirmed broken, this stops being an EV call at all: it is reason 1, a tax-blind exit (§4.4, §4.11).

Practice set

Work each fully (decide and justify against the five reasons and the fresh-money test) before reading the solution. Ramps from guided to independent to timed.

Problem 1 (guided). You bought a consumer-staples compounder at 40× earnings three years ago on a thesis of durable brand + distribution moat and steady volume growth. It is now up 120% and trades at 55×. Volumes are still growing mid-single-digits; every pillar holds. A friend says "book your profits — no one ever went broke taking a profit." Sell, hold, or trim?

Solution. Hold. Test against the five: thesis intact (not reason 1); no mistake (not reason 2); at 55× with the thesis compounding, run a rough reverse check: if value is compounding ~13–15%/year, a merely-rich (not mania) multiple does not clearly fail the bull case, so reason 3 does not fire cleanly; no better opportunity named (reason 4); assume within your size cap (reason 5). No reason fires. The friend's cliché is §4.9's trap: selling here is the Raider clipping a flower. The fresh-money test ("would I buy this business at 55× today?") might argue against adding, but "don't add" is not "sell." Default: do nothing.

Problem 2 (guided). A mid-cap you own falls 45% in two months. You feel sick watching it. Walk through the correct decision procedure (not the answer, the procedure).

Solution. The price move is not the decision (orthogonality principle). Procedure: (1) open the decision journal and re-read the written thesis pillars and pre-registered break triggers; (2) test each pillar against the latest filings/results: did a load-bearing pillar factually break, or is this thesis-noise/Mr.-Market? (3) if a pillar broke → reason 1, exit regardless of the loss; if you now see you were wrong at purchase → reason 2, exit, do not average down; (4) if all pillars hold → run the fresh-money test: would you buy at the lower price? If yes, this is a Hunter's add, not a sell. The feeling of sickness is loss aversion and is not evidence; the journal is the arbiter.

Problem 3 (guided). Compute the disposition ratio for an investor holding 8 winners and 4 losers who sells 6 winners and 1 loser, and say what it reveals.

Solution. PGR = 6/8 = 0.75; PLR = 1/4 = 0.25; ratio = 3.0. A neutral investor would show PGR ≈ PLR (ratio ≈ 1). At 3.0 the investor is realizing gains at three times the rate of losses, a textbook disposition effect: cutting winners (Raider) and riding losers (Rabbit), the exact inversion of the correct discipline.

Problem 4 (independent). You hold A (fairly valued quality compounder you know deeply, ~10% expected forward return, large long-term gain). B appears at ~12.5% expected return but you have studied it three weeks. India LTCG 12.5%; A is up 4×. Should you switch? Show the tax hurdle.

Solution. Tax handicap (WE3): selling A redeploys ~90.6% of value, so B starts ~10.3% behind ≈ ~2.0%/year over 5 years. B's raw edge is 12.5% − 10% = 2.5 points; minus ~2 points of tax drag leaves ~0.5 point, before a conviction discount for knowing B three weeks versus A for years. After the discount the edge is inside the noise. Do not switch. Reason 4 requires a material edge that survives tax and the conviction discount; this is marginal, and switching here is likely the churn/overconfidence tax. Keep studying B.

Problem 5 (independent). Your portfolio is ₹2 crore. Position C, bought at 6% weight, has 6-bagged while the rest was flat; C is now worth ₹72 lakh. Your concentration cap is 18%. What do you do, exactly, and what is the tax (C cost ₹12 lakh; India LTCG 12.5%, ignore exemption)?

Solution. C weight = 72/200 = 36%, over the 18% cap → reason 5: trim, do not exit. Target C = 18% × ₹2 crore = ₹36 lakh. Trim = 72 − 36 = ₹36 lakh sold, ₹36 lakh kept (still 18%, still compounding). Cost basis fraction = 12/72 = 1/6, so gain fraction on any sale = 5/6. Gain on the trim = (5/6) × 36 = ₹30 lakh → LTCG 12.5% = ₹3.75 lakh tax. You capped single-name risk at policy weight and kept the Connoisseur's core position working. (Note: exiting C entirely would be the Raider's error dressed as prudence.)

Problem 6 (independent). A bank you own reports, for the third straight quarter, rising SMA-2 balances and slippages while net interest margin compresses and loan growth accelerates, precisely the pattern your buy note pre-registered as a thesis-break trigger. The stock is down only 8% and a broker just reiterated "Buy." Decision?

Solution. Sell (reason 1: thesis broken). Your own pre-registered trigger has fired: the M5.01 warning combination (asset-quality deterioration + NIM compression + aggressive growth) is a load-bearing pillar (asset quality/underwriting) breaking, not noise. The small price drop and the broker's "Buy" are irrelevant: reason 1 is tax-blind and price-blind; the market has not yet re-rated, which if anything means you can exit closer to your entry. This is exactly the case pre-commitment exists for: you decided, in the cold state, that this pattern ends the thesis, so you act now rather than rationalizing (confirmation bias would happily reinterpret three quarters of deterioration as "temporary").

Problem 7 (independent). Classify these three mistakes on both taxonomy axes and state the fix (or none): (a) sold your entire portfolio at the 2008 low, theses intact, because you couldn't stand the losses; (b) bought a "cheap" cyclical at peak trailing earnings, misreading peak-cycle margins as normal; (c) held a fraud that had auditor resignations and large related-party loans you didn't investigate.

Solution. (a) Commission · Behavioral → structural fix (pre-commit: no drawdown selling without a written thesis-break; watch prices less). (b) Commission · Analytical → checklist line ("normalize cyclicals to mid-cycle earnings; never value a cyclical on trailing P/E," M5.09). (c) Commission · Analytical, not bad luck: auditor resignation and related-party loans are detectable red flags (M2, M8.04), so the "undetectable fraud" defense fails → checklist line ("treat auditor turnover + related-party lending as a veto until explained"). Only genuinely footprint-free fraud would be bad-luck/no-line.

Problem 8 (independent). Explain, with the arithmetic, why crossing the 1-year holding line matters far more for a US investor than crossing the 12-month line matters for an Indian investor, for a position with a 300 gain on a 100 cost.

Solution. US: short-term at 37% ordinary → tax 111, redeploy 289, handicap 38.4%; long-term at 15% → tax 45, redeploy 355, handicap 12.7%. Crossing the line saves ~25.7 points of handicap. India: short-term at 20% → tax 60, redeploy 340, handicap 17.6%; long-term at 12.5% → tax 37.5, redeploy 362.5, handicap 10.3%. Crossing the line saves ~7.3 points. The US saving (~26 pts) dwarfs the Indian (~7 pts) because the US short-vs-long rate gap (37% vs 15%) is far wider than India's (20% vs 12.5%). Lesson: in the US, holding a winner a few extra days to cross one year can be worth a large, decisive amount; in India it matters but less.

Problem 9 (timed, 8 min). A founder-led US software compounder you bought at 12× sales is now at 32× sales; growth is decelerating from 35% to a still-strong 22%, net revenue retention is slipping from 128% toward 115% (still healthy), and Rule of 40 has fallen from 55 to 44 (still above 40). Nothing is broken, but the price implies ~30% growth for a decade. You have a large short-term gain (held 10 months). Decide.

Solution. Layered. Reason 1/2: no pillar broken and no purchase mistake. The business is executing, merely maturing (thesis playing out, decelerating from a high base, all KPIs still healthy). Reason 3: at 32× sales pricing in ~30% growth for a decade when the business is now growing 22% and decelerating, the reverse-DCF likely implies a forward return below hurdle even on a reasonable bull case → margin of danger is plausibly firing. Tax modulation: the gain is short-term (10 months), so selling now is maximally tax-inefficient (ordinary rates); the ~2 months to long-term status is a strong reason to wait if the thesis allows. Reasonable decision: begin trimming toward your size cap and plan to reduce further after crossing the 1-year line, rather than a full immediate exit. Reason 3 fires (trim), but the short-term tax hurdle argues for staging the sale across the holding-period line. Justify the trim by the margin of danger, the stage by the tax. (Full-hold or full-sell are defensible if argued against the five reasons; an un-argued sell "because it's up a lot" is not.)

Problem 10 (timed, 10 min). Write, from scratch, the four-part pre-committed sell-criteria section you would attach to a buy note for a hypothetical Indian FMCG compounder you are purchasing at 45× earnings. Then state which two of your triggers are tax-blind.

Solution (model). (1) Testable pillars: (P1) direct distribution reach grows faster than the top-3 peers; (P2) volume growth ≥ mid-single-digits with premiumization lifting realization; (P3) ROCE stays > 30% on an asset-light, negative-working-capital model; (P4) no promoter-governance or related-party deterioration. (2) Pre-registered break triggers: sell if volume declines for 3 consecutive quarters despite price cuts (P2 broken); sell if direct reach stops expanding while a challenger takes structural share for 4 quarters (P1 broken); sell if ROCE falls below ~20% on a debt-funded acquisition spree (P3 broken); sell if promoter pledges > 10% or a material related-party transaction appears (P4 broken). (3) Overvaluation zone: re-underwrite if the multiple exceeds ~70× such that the reverse-DCF-implied return falls below my 12% hurdle even on the bull case (reason 3, tax-modulated). (4) Size cap: trim back to 18% of the portfolio if drift takes it higher (reason 5). Tax-blind triggers: the pillar-break triggers in (2), which are reasons 1/2 and I act on them regardless of tax; (3) and (4) are tax-modulated.

Problem 11 (timed, 10 min). An investor says: "My rule is a hard 25% stop-loss on every position — if it's down 25% from my cost, I'm out, no exceptions. It caps my losses." Critique this for a fundamental, concentrated investor, and state when a stop is appropriate.

Solution. For a fundamental business-owner, a cost-based hard price stop is flawed on three counts: (i) it re-anchors the whole decision to the cost basis you should have deleted (§4.1); (ii) it sells you out of exactly the intact-thesis drawdowns that are Mr. Market's gifts: a 25% fall with every pillar holding is a Hunter's buy, and the stop forces the opposite, converting temporary volatility into permanent loss; (iii) it makes the price, not the thesis, the sell trigger, violating the orthogonality principle. The correct primary discipline is thesis-based triggers with price as a prompt-to-review (§4.10). However, a disciplined price/time stop is appropriate and often superior for the loser/mistake case (the Assassin), where the real danger is inaction (the Rabbit): if you know you tend to freeze on losers, a mechanical rule that forces a re-underwrite or an exit protects you from yourself. The nuance: mechanical stops for the behavioral risk of not-cutting-mistakes; judgmental thesis review for intact drawdowns. A blanket 25%-on-everything rule fails to make that distinction.

Problem 12 (timed, 10 min). You realize that over three years you have sold six of your ten best-performing positions after roughly doubling, and still hold four of your five worst, each down 40%+ with deteriorating fundamentals. Diagnose the pattern, name the tribes, quantify the behavior qualitatively, and prescribe two structural fixes.

Solution. Diagnosis: a severe disposition effect, cutting winners (you sold your best after a mere double) and riding losers (you hold your worst despite deteriorating fundamentals). Tribes: you are a Raider with winners (snatching small gains, forfeiting the skew, §4.9) and a Rabbit with losers (freezing on falling positions whose theses may be broken, reasons 1/2 unaddressed). The behavior is the exact inversion of the Assassin/Connoisseur discipline that works. Structural fixes (M6.04 §4.8 style): (1) pre-commit sell criteria at purchase (§4.10) so winners are held to thesis/valuation/size rules rather than clipped on a double, and losers face pre-registered break triggers that force the reason-1/2 decision; (2) journal and score every sell (feeding M9.04's calibration) so the pattern is visible and the taxonomy (§4.12) converts each into a checklist line, specifically "do not sell a winner absent one of the five reasons" and "re-underwrite any loser against its written thesis quarterly; cut mistakes as an Assassin." A trim-to-cap rule (reason 5) can also convert the urge to "take something off" a winner into a disciplined partial rather than a full amputation.

Problem 13 (guided). After a sharp fall, a position is quoted at ₹6,00,000 (near your cost, so tax is negligible). Journal-tested, you judge a 60% chance the thesis is intact (recovers to ₹12,00,000) and 40% it is broken (residual ₹2,00,000). (a) Compute EV(hold) and EV(sell) and decide. (b) Find the break-even intact-probability. (c) Recompute it if the broken branch is a zero, and say what discipline that enforces.

Solution. (a) EV(sell) = ₹6,00,000; EV(hold) = 0.60 × 12,00,000 + 0.40 × 2,00,000 = 7,20,000 + 80,000 = ₹8,00,000 → hold (higher by ₹2,00,000), the size of the fall itself being no part of the decision (orthogonality, §4.1). (b) 12p + 2(1 − p) = 6 → 10p = 4 → break-even p = 0.40: hold while your journal-tested intact-probability exceeds 40%. (c) With the broken branch ≈ 0: 12p = 6 → break-even p = 0.50. A ruinous downside lifts the required intact-probability from 40% to 50%, because risk is probability × magnitude (M9.01) and you cannot average your way out of a zero. And the probability must be the journal's, not one reverse-engineered to keep the position.

Problem 14 (guided). You hold $60,000 of an intact compounder (cost basis $20,000) growing ~16%/year, within your size cap. You consider selling to buy a diversified fund expected to return ~9%/year. US long-term tax 15% + 3.8% NIIT = 18.8%; horizon 12 years. (a) What capital redeploys after tax? (b) Terminal wealth on each path. (c) Which of the five reasons, if any, justifies the switch?

Solution. (a) Gain = 60,000 − 20,000 = 40,000; tax = 18.8% × 40,000 = $7,520; redeploy = $52,480. (b) Hold: 60,000 × 1.16¹² = 60,000 × 5.936 = $356,162; switch: 52,480 × 1.09¹² = 52,480 × 2.813 = $147,609 → forgoing $208,553. (c) None. Thesis intact (not 1/2); no margin of danger stated (not 3); a 9% fund is not a materially better risk-adjusted return than a 16% compounder, so reason 4 fails badly; the position is within cap (not 5). Selling here is the Raider's error (§4.9) plus a $7,520 tax drag; the honest tool for concentration worry is trim to the cap (§4.8), not exit into an average asset.

Problem 15 (independent). You buy ₹2,00,000 of a compounder at ₹200/share (1,000 shares) that compounds price and value at 22%/year. After 4 years you sell to "lock in the gain" and put the after-tax proceeds into an index returning 10%/year for the remaining 16 years (20-year total horizon). India LTCG 12.5%, ignore the exemption. (a) Value at year 4 and after-tax proceeds. (b) Terminal wealth of the redeploy-to-index path at year 20. (c) Terminal wealth had you simply held the compounder. (d) The amount forfeited.

Solution. (a) 1.22⁴ = 2.2153, so value = 200,000 × 2.2153 = ₹4,43,067 (₹443.07/share); gain = ₹2,43,067; tax = 12.5% × 2,43,067 = ₹30,383; after-tax = ₹4,12,684. (b) 1.10¹⁶ = 4.595 → 4,12,684 × 4.595 = ₹18,96,270. (c) 1.22²⁰ = 53.36 → 200,000 × 53.36 = ₹1,06,71,528 (~₹1.07 crore). (d) Forfeited = 1,06,71,528 − 18,96,270 = ₹87,75,258 (~₹87.8 lakh). Even redeploying into a solid 10% index (not idle cash), clipping the 22% compounder at year 4 gave away ~₹87.8 lakh, because two forces compound against you: the tax paid at year 4 (which stops working) and the 22% − 10% = 12-point rate gap over sixteen years. The compounder was the wealth event; §4.9's cliché hid the fortune.

Problem 16 (independent). Two investors each start with ₹10,00,000 in the same asset returning 15%/year pre-tax for 10 years. Investor H holds throughout and pays LTCG 12.5% once, at the end. Investor C sells and immediately rebuys the identical asset every year, paying 12.5% on each year's gain (assume all long-term, ignore the exemption). (a) Investor H's after-tax terminal wealth. (b) Investor C's after-tax terminal wealth. (c) The drag, and what causes it.

Solution. (a) 1.15¹⁰ = 4.0456 → pre-tax ₹40,45,558; gain ₹30,45,558; tax = 12.5% × 30,45,558 = ₹3,80,695; after-tax = ₹36,64,863. (b) Paying tax each year means C compounds at a net 15% × (1 − 0.125) = 13.125%: 10,00,000 × 1.13125¹⁰ = 10,00,000 × 3.4323 = ₹34,32,305 (already after-tax; each rebuy resets the basis). (c) Drag = 36,64,863 − 34,32,305 = ₹2,32,558 (~₹2.33 lakh), with identical assets and identical pre-tax returns. The only difference is when tax is paid: C terminates the interest-free deferred-tax loan (§4.11) every year, handing money to the government early instead of letting it compound. This is the pure tax cost of churn, before any mistimed switch, and it widens with the horizon.

Problem 17 (independent). You hold A (market value ₹500, cost ₹100, long-term) expected to return 11%/year; India LTCG 12.5%, ignore the exemption; horizon 6 years. Replacement B would be bought with the after-tax proceeds. (a) What redeploys after tax, and the one-time capital handicap. (b) The annual forward return B must earn just to break even with holding A. (c) Interpret.

Solution. (a) Tax = 12.5% × (500 − 100) = ₹50 → redeploy ₹450; one-time handicap = 500/450 − 1 = 11.1%. (b) Hold A = 500 × 1.11⁶ = 500 × 1.8704 = ₹935.21; B must satisfy 450 × (1 + R)⁶ = 935.21 → (1 + R)⁶ = 2.0782 → R = 12.97%/year, a ~2.0-point edge over A's 11% just to tie. (c) And that is before the conviction discount for knowing B less well and before transaction costs. A "1-point better" idea (12% vs 11%) does not clear a ~2-point tax hurdle — it loses. This is the arithmetic behind reason 4's demand for a material, durable edge (§4.7), and behind crossing the holding-period line before selling a winner.

Problem 18 (timed, 12 min). A US position: market value $84,000, cost basis $24,000, held 11 months (short-term); the thesis is intact and compounding ~17%/year. Your portfolio totals $280,000; your written concentration cap is 20%. Short-term rate 37%; long-term (reached in ~1 month) 18.8%. (a) Which reason fires, and the action? (b) The dollar trim to reach the cap. (c) Compare the tax of trimming that amount now versus after crossing the 1-year line. (d) Recommend, naming the tension.

Solution. (a) Weight = 84,000 / 280,000 = 30% > 20% cap → reason 5 (beyond risk tolerance); action = trim, not exit (thesis intact, so you keep the compounder working). Reasons 1/2 (intact), 3 (no margin of danger stated), and 4 (no better idea) do not fire. (b) Target = 20% × 280,000 = $56,000 → trim = 84,000 − 56,000 = $28,000, keep $56,000. (c) Gain fraction = 1 − 24,000/84,000 = 5/7, so the gain on the $28,000 trim = $20,000. Now (short-term 37%): tax = $7,400. After ~1 month (long-term 18.8%): tax = $3,760, a $3,640 saving. (d) Reason 5 is tax-modulated (discretionary): a 30% weight is above cap but not an imminent, binary threat, so staging the trim across the 1-year line (or trimming part now, the rest after) is defensible and nearly halves the tax. But if the single-name risk were acute (an imminent regulatory verdict, say) you trim now and pay the short-term tax, because you never let a $3,640 tax saving stop a genuine risk reduction. (A broken thesis or recognized mistake would be reason 1/2, tax-blind, and you would not wait at all.)

Applied mini-project

Task: Write the pre-committed sell criteria for a real holding candidate, then audit one real past sell.

Part A. Pre-commit (a real company). Choose one real company you could plausibly own (one Indian and one US option are encouraged across the program's alternation) and use its latest 10-K/annual report and a current price. Write the full four-part sell-criteria section (§4.10) as if attaching it to a buy note today: (1) three-to-five testable thesis pillars drawn from the actual business (moat, growth runway, unit economics, capital allocation, balance sheet); (2) the specific pre-registered break triggers for each pillar, stated as observable events you could verify in future filings; (3) the overvaluation zone from a rough reverse-DCF (what growth does today's price imply, and at what multiple would even the bull case fail your hurdle?); (4) the size cap and trim rule from your M9.02 concentration policy. Label which triggers are tax-blind (reasons 1–2) and which are tax-modulated (3–5).

Part B. Audit a real sell. Take one real sale you (or a documented investor) made, ideally a winner sold or a loser held. Reconstruct: which of the five reasons, if any, actually justified it? Was it a Rabbit/Raider behavior in disguise? Classify any mistake on the taxonomy (both axes) and write the one checklist line it earns (or "bad luck — no line," if honestly earned). If a compounder was sold early, compute the §4.9 cost (proceeds vs. later value).

Deliverable. A two-page memo: Part A (the pre-commitment) and Part B (the audit + the earned checklist line). Where you use current tax rates, cite the source and date and mark "verify."

Scoring rubric (0/1/2 each; 12 = mastery):

Case Lab: run case CL-9.2 in the app's Cases tab for this module: a seeded sell/hold/trim scenario (position, price move, new facts, and your own original written thesis in; a decision out) scored against the five-reason framework and the fresh-money test, not against whether the price later went up.

Reading & resources

• Lee Freeman-Shor, The Art of Execution (2015). The anchor text and the source of the five tribes, the ~49% hit-rate finding, and the "execution beats selection" thesis. Short, concrete, evidence-based; read it whole for this module. [Paid] [Beginner→Intermediate]
• Philip Fisher, Common Stocks and Uncommon Profits (1958). The chapter "When to Sell — and When Not To" is the classic statement of the disciplined-holder's three reasons to sell (mistake; deterioration; a much better opportunity) and the skepticism about selling good companies on valuation alone. [Paid] [Intermediate]
• Howard Marks, Mastering the Market Cycle (2018) and the Oaktree memos (oaktreecapital.com/insights, free), especially "Selling Out" (Jan 2022), Marks's direct memo on the psychology and logic of selling, and "You Can't Predict. You Can Prepare." Read "Selling Out" alongside §4.9. [Book Paid / Memos Free] [Beginner→Intermediate]
• Hendrik Bessembinder, "Do Stocks Outperform Treasury Bills?" (Journal of Financial Economics, 2018). The positive-skew evidence behind §4.9 (most stocks lag T-bills; a tiny minority create all the wealth). Read the abstract and the summary tables; findable free as a working paper. [Free] [Intermediate]
• Daniel Kahneman, Thinking, Fast and Slow (2011). Part IV covers prospect theory, the disposition effect, and the endowment effect; the psychological foundation this module operationalizes. [Paid] [Intermediate]
• Peter Lynch, One Up on Wall Street. The chapters on when to sell and "the best time to sell"; the source of "cutting the flowers and watering the weeds" and the category-based sell logic. [Paid] [Beginner]
• Charlie Munger (via Poor Charlie's Almanack) and Terry Smith (Fundsmith annual letters and Investing for Growth, free excerpts). Munger on "never interrupt compounding unnecessarily"; Smith on "do nothing" and the discipline of not selling quality. [Book Paid / Letters Free] [Beginner→Intermediate]
• Tax primary sources (check the live rates): India, the Income Tax Department (incometaxindia.gov.in), Sections 196 and 198 of the Income-tax Act, 2025, which applies from tax year 2026-27 (Sections 111A and 112A of the 1961 Act govern earlier years), and the current Finance Act for LTCG/STCG on listed equity (rates revised July 2024). US, IRS Topic No. 409 (Capital Gains and Losses) and Publication 550, plus the current long-term brackets and the 3.8% NIIT; the wash-sale rule (Publication 550). [Free] [Intermediate]
• India ties. Sanjay Bakshi (BFBV / Fundoo Professor) on selling, compounders, and the temperament to hold; Gautam Baid, The Joys of Compounding, chapters on holding winners and the discipline of patience for the Indian investor. [Free / Paid] [Intermediate]
• Going deeper: Terrance Odean, "Are Investors Reluctant to Realize Their Losses?" (Journal of Finance, 1998; findable free as a working paper) [Free] [Intermediate]. The original large-sample measurement of the disposition effect (the rate of realizing gains running far above the rate of realizing losses) that §4.2's disposition ratio operationalizes. Read it for the evidence that the Rabbit-and-Raider pattern is pervasive and measured, not anecdotal.
• Going deeper: Nick Sleep & Qais Zakaria, the Nomad Investment Partnership letters (collected editions circulate free as a PDF) [Free] [Intermediate]. The clearest lived example of the Connoisseur discipline of §4.9: a concentrated book of "scale economies shared" compounders held for years without clipping. Read alongside the compounder caveat (§4.6) and Munger's "never interrupt compounding unnecessarily."

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 disposition effect and Freeman-Shor's five tribes, the five legitimate sell reasons and the fresh-money test, the cost of clipping compounders, pre-committed sell criteria, tax-aware selling in both jurisdictions, and the two-axis mistake taxonomy",
  "use": "The sell decision is compromised at the moment you make it, so the useful AI role is adversarial and it must operate on the written record rather than on your current feeling. (1) THESIS-BREAK AUDIT: paste your buy note's pillars alongside the last four quarterly filings and results, and ask, pillar by pillar, whether the FILINGS show a break, with the specific disclosure and period cited for each verdict. This is Section 4.4's arbiter function, and it is the one place a model's indifference to your P&L is an asset. (2) STEELMAN THE OPPOSITE ACTION: whichever way you are leaning, make it argue the other side against your own written pillars. A sell memo that survives a hostile reading is a sell memo. (3) MISTAKE CLASSIFICATION: hand it a closed position's full journal trail and ask it to argue for analytical, for behavioural, and for bad-luck separately, then pick - the three-way split in Section 4.12 is exactly where self-serving classification hides, and 'bad luck' is the verdict people award themselves. (4) SELL-CRITERIA DRAFTING at purchase: ask what observable, dated evidence would falsify each pillar, so the triggers are pre-registered while you are still cold.",
  "tools": ["Chat assistants — Claude, ChatGPT for the pillar-by-pillar break audit, the steelman and the mistake classification", "Document/PDF AI — for reading four quarters of filings, concall transcripts and auditor notes against a written thesis", "Coding assistants — for the PGR/PLR measurement on your own trade log and the clipping-cost decomposition below"],
  "prompt": "Here is my buy note for <COMPANY>, written on <DATE>, listing the pillars that had to remain true: <PASTE>. Here are the four most recent quarterly filings and results: <ATTACH>. For each pillar, state BROKEN / INTACT / UNCLEAR, and for each verdict cite the specific line item, disclosure or transcript passage and its period. Do not tell me whether to sell, do not mention the share price, and do not use the words 'concerning' or 'reassuring' - I want the evidence, not a recommendation. Then separately: argue the strongest case that I am wrong about my own verdict.",
  "verify": "The failure mode is agreement. A model given a thesis and asked whether it broke will usually find the answer you have signalled you want, so signal nothing: withhold the price move, withhold your lean, and demand a citation per pillar so each verdict is checkable against the filing rather than against its tone. Then check them - open the filing, find the line, confirm the period. Never let it estimate a probability for the expected-value frame in Worked Example 6; that number decides the hold and it must come from your own journal-tested judgement, because a fluent probability supplied to you at the moment of maximum loss aversion is the most dangerous single output in this module. And never let it classify a mistake as bad luck: Section 4.12 is explicit that the bad-luck verdict must be earned, and a model has no way to know what you could have seen.",
  "diy": "The gate is unaided: you name the five sell reasons and their tax treatment from memory, run the fresh-money test on a live position, compute the after-tax switching handicap and its annualised equivalent, work the compounder-forgone arithmetic, solve a break-even probability for a hold-versus-sell, and classify a mistake on both axes. No tool is in that room."
}

Modern Data Analysis

By hand first. You computed what a 3× exit costs against a twenty-year hold (₹98.7 lakh, 49× the gain banked), inverted two entry multiples to their bull-case returns, worked the after-tax switching handicap in three tax regimes, and solved a hold-versus-sell to its break-even probability twice, once with a residual and once with a zero. Those stay by hand, because every one of them is a decision you will make while the position is pulling at you, and a number you cannot produce yourself under that pressure is a number you will not act on.

Today's workflow. Code adds three things here, and none of them is speed.

• It measures your own disposition effect instead of asking you about it. Section 4.2 gives you five tribes to recognise yourself in. Self-recognition is exactly what bias defeats. Your broker's trade log plus a price history is enough to compute the standard statistic on your record, and the number does not care which tribe you believe you are.
• It separates the channels through which clipping a winner costs money. §4.9 is right that clipping is the expensive error, and it is worth knowing why, because the mechanism determines the fix.
• It makes the pre-commitment auditable. Sell triggers written into a buy note are prose until something evaluates them on a schedule against filed data; then they are a monitoring job that reports "pillar 3 has been UNCLEAR for two quarters."

Tools & sources (IN + US). pandas for joining your trade log to a price history (the whole PGR/PLR computation is one groupby once the join is right), numpy for the path simulations, matplotlib for the realised-versus-paper plot that makes your own disposition effect visible. India. Your broker's contract notes and the NSDL/CDSL consolidated account statement for the actual trade log including dates and cost basis; NSE/BSE bhavcopy for the price history to mark unsold positions; the annual report and quarterly results plus concall transcripts for the pillar audit; incometaxindia.gov.in for Sections 111A and 112A, the 12-month line and the ₹1.25 lakh exemption in §4.11; screener.in for the multiple history behind a reason-3 re-underwrite. US. Form 1099-B and your broker's realised/unrealised gain report for the trade log; EDGAR full-text search and the 10-Q/10-K for the pillar audit, with the 8-K stream as the trigger feed; IRS Topic 409 for the long-term brackets and Publication 550 for the wash-sale rule §4.11 names; FRED (DGS3MO) for the risk-free line in a Bessembinder-style comparison; Hendrik Bessembinder's papers (SSRN, free) for the skew evidence the module cites.

# Two things the module cannot do on paper: measure YOUR disposition effect, and
# decompose what clipping actually costs. R3, recomputed in-session.
import numpy as np
rng = np.random.default_rng(9031)

# --- Odean's PGR/PLR, and a check that it is not just measuring market drift ---
def run(rule, drift, K=12, T=240, nsim=3000, sig=0.32/np.sqrt(12), seed=5):
    r = np.random.default_rng(seed); mu = drift/12 - 0.5*sig**2
    RG = RL = PG = PL = 0
    for _ in range(nsim):
        px = np.ones(K); cost = np.ones(K)
        for t in range(T):
            px *= np.exp(r.normal(mu, sig, K))
            sell = rule(px, cost, r)
            if not sell.any(): continue
            gain = px > cost
            RG += int((sell & gain).sum());  RL += int((sell & ~gain).sum())
            PG += int((~sell & gain).sum()); PL += int((~sell & ~gain).sum())
            px[sell] = 1.0; cost[sell] = 1.0          # replaced at a fresh basis
    return RG/(RG+PG), RL/(RL+PL)                     # PGR, PLR

free = lambda p, c, r: r.random(len(p)) < 0.02        # sells at random, ignores P&L
for drift in (-0.08, 0.0, 0.12):
    g, l = run(free, drift); print(f"{drift:+.0%}: ratio {g/l:.3f}")   # 1.000, 0.999, 1.005

# --- what does clipping a compounder cost, and through which channel? ---
T, N = 20, 100_000
mk = lambda g, s, seed: np.exp(np.random.default_rng(seed).normal(
        np.log(1+g) - 0.5*s**2, s, (N, T)))
A, B = mk(0.18, 0.28, 3), mk(0.10, 0.28, 4)           # the compounder; your next-best idea
hold = A.prod(1)
for tax, repl in ((0.0,'same'), (0.125,'same'), (0.0,'worse'), (0.125,'worse')):
    w = np.ones(N); basis = np.ones(N); src = A
    for t in range(T):
        w *= src[:, t]
        if (t+1) % 5 == 0:                            # clip every 5 years
            w -= tax*np.maximum(w-basis, 0); basis = w.copy()
            if repl == 'worse': src = B
    print(tax, repl, round(np.median(w)/np.median(hold)-1, 3))
# 0.0 same +0.000 | 0.125 same -0.198 | 0.0 worse -0.652 | 0.125 worse -0.703

Verify. Every worked figure in this module reproduces exactly. WE1: 1.26²⁰ = 101.721, so ₹100 becomes ₹10,172/share and the position ₹1,01,72,107; forfeited value ₹98,72,107, which is 49.4× the ₹2,00,000 banked, and the 3× exit captured 2.95% of the eventual value; the ₹2,75,000 after-tax proceeds would need a fresh 37.0-bagger to get back. WE2: 1.14¹⁰ = 3.7072; at 70× the price factor is 2.118 for 7.80%/yr, at 45× it is 3.295 for 12.67%/yr. WE3: handicaps of 10.3% / 12.7% / 38.4%, annualising over five years to 1.99% / 2.42% / 6.72%. WE5: 1.18¹⁵ = 11.974 giving $1,197,375 against 85,900 × 1.08¹⁵ = $272,489, a 4.394× ratio and $924,885 forgone. WE6: EV(hold) = ₹10,80,000, break-even 41.67%, and 53.33% when the broken branch is zero. §4.11's four-bagger switching handicap is 10.34%. §4.9's 26% compounding doubles in 3.00 years exactly.

Quantitative lens

The disposition effect is measurable, and the measure is built so that a rising market cannot fake it. This matters because most behavioural findings do not survive contact with a real dataset, and this one does. The standard statistic, from Odean's work on brokerage records, is a pair: the proportion of gains realised, PGR = realised gains ÷ (realised gains + paper gains), and the proportion of losses realised, PLR = realised losses ÷ (realised losses + paper losses), both counted only on days the account actually traded. The disposition effect is PGR > PLR. The obvious objection is that a bull market puts most positions in the money, so any seller will realise more gains than losses. The objection fails, because both ratios are conditioned on the available stock of gains and losses. Simulate it: a trader who sells 2% of positions per month chosen at random, with no reference to profit or loss, returns PGR/PLR of 1.000 in a −8%/yr market, 0.999 in a flat one and 1.005 in a +12%/yr one. Drift cancels out by construction. Give the same simulated trader a disposition rule (take profits readily above +25%, almost never cut a loser) and the ratio becomes 7.6, 10.5 and 16.3 in those same three markets. So a ratio meaningfully above 1 on your own trade log is a fact about you, not about the market you traded in. What that does not establish is the module's stronger claim about cost, which is a separate question and is treated next; and note that this simulation validates the statistic, it does not reproduce anyone's empirical estimate. For that, run it on your own log, or read Odean's paper.

Clipping a compounder costs nothing at all — unless one of two things is true, and then it costs everything. §4.9 is the module's most forceful section and it is worth knowing exactly where its force comes from. Simulate a genuine 18%/yr compounder over twenty years and a Raider who takes profits every five years, then vary two things: whether the sale is taxed, and whether the redeployment is into an equally good idea or into the next-best one. Clip every five years, pay no tax, and redeploy into an identical process: median terminal wealth 12.50× against holding's 12.50×, a difference of 0.0%. Clipping, in itself, costs precisely nothing. Now add India's 12.5% long-term rate at each clip, still redeploying into an identical idea: 10.02×, a 19.8% loss purely to the interest-free-loan effect §4.11 describes: four realisation events over twenty years, each one handing the taxman capital that would otherwise have compounded. Now remove the tax again but redeploy into a 10% idea rather than an 18% one, which is what actually happens when you sell your best holding: 4.35×, a 65.2% loss. Do both and you get 3.71×, down 70.3%. Read the decomposition: the replacement-quality channel is roughly three times the tax channel, and the pure act of selling contributes zero. So "no one ever went broke taking a profit" is not defeated by the observation that selling is bad. It is defeated by the observation that you cannot replace what you are selling, which is precisely the assumption WE5 makes explicit when it compares an 18% compounder to an 8% index. The instruction that follows is sharper than "hold winners": before clipping, name the specific thing you will buy with the proceeds and its honest expected return. If it is genuinely as good, the clip is free and the argument for it is behavioural comfort, which is a bad reason but a cheap one. If it is not as good, and by construction your best idea rarely has a peer, the clip costs the gap, compounded, and the tax is the smaller half of the bill.

The recovery function is exact, and it is where a permanent loss and a temporary drawdown genuinely part company. A permanent loss of fraction L requires a gain of L/(1−L) on what remains: 11.1% at −10%, 25.0% at −20%, 42.9% at −30%, 100.0% at −50%, 233.3% at −70%, 900.0% at −90%. That is exact arithmetic and the convexity is the whole reason §4.4 makes a broken thesis a tax-blind exit. But the same −50% costs nothing permanent if the thesis is intact and the price recovers. What it costs is time, and time is priceable: a −50% drawdown that fully round-trips over three years is a realised 0% CAGR across that window, which against a 12% alternative is 40.5% of capital in opportunity cost, and over ten years 210.6%. This is the module's orthogonality principle (§4.1) made quantitative in both directions: the price move alone tells you nothing about which case you are in, so the number to compute is not the drawdown but the answer to the pillar audit — and the two cases, identical on the screen, differ by the entire difference between a delay and a hole.

Honest limits, because two of the results above are easy to over-read. The clipping decomposition assumes the compounder actually compounds at 18% for twenty years, which is a hindsight-selected object: the whole difficulty of §4.9 is that you do not know at year five which of your holdings is the one, and a rule of "never clip" applied to the holdings that turn out not to be compounders has a cost this simulation does not price. The lognormal paths carry no thesis breaks; a real Connoisseur holding through a genuine reason-1 break loses far more than a Raider does. The PGR/PLR simulation validates that the statistic is drift-robust. It does not measure anyone's actual behaviour, and it says nothing about whether the disposition effect costs money on any particular record, only that it is detectable. On the skew that §4.9 leans on: a lognormal cross-section with an assumed 30% single-stock volatility and dispersed firm lifetimes mechanically produces 58% of firms below a T-bill line and the top 4% holding 82% of the positive excess wealth, close to the figures the module cites, which is exactly why it must be said clearly that this is a mechanism demonstration and not a confirmation: the parameters were chosen, not fitted, and the sensitivity proves it, since moving assumed volatility from 20% to 45% moves the below-T-bills share from 46.9% to 69.5%. The empirical claim belongs to Bessembinder's data; what the simulation shows is only that heavy positive skew is what compounding under dispersion does, so a finding of that shape should not be surprising and should not be quoted to a precision the underlying study does not support.

Do it in code: compute PGR and PLR on your own trade log before you decide which tribe you are, and re-compute annually; when a clip tempts you, price the replacement explicitly rather than assuming one exists; and run the pillar audit on a calendar rather than on a price move, so the review happens in the cold state where §4.10's pre-commitment was written. See G1/QM1.02, G2/DA1.02 and G3/AI0.06.

Where this goes next: galaxy cross-links

• G1 · QM1.02 Hypothesis Testing & Regression. What it takes to say a PGR/PLR gap on your own record is real rather than a small-sample artefact, and how a paired test on realised versus paper positions is constructed.
• G2 · DA1.02 Pandas Data Wrangling. The trade-log-to-price-history join that the entire disposition measurement rests on, including the corporate actions and partial fills that silently corrupt a cost basis.
• G2 · QD2.03 Performance Attribution in Code. Separating what your selling did to returns from what your buying did, which is the only way to know whether your exits are the leak this module says they usually are.
• G3 · AI0.06 Failure Modes & Guardrails. Sycophancy and confirmation as the specific failure mode that makes an AI thesis-break audit dangerous, and the prompt hygiene that reduces it.
• G3 · CN1.02 Hypothesis-Driven Problem Solving. Pre-registering the disconfirming evidence, which is §4.10's sell trigger written in the language of a falsifiable hypothesis rather than a hope.

Teach it back & journal

Feynman prompt. Write a one-page explainer for a smart 15-year-old titled "The hardest part of investing isn't buying — it's letting go." Without jargon, convey: that once you own something you get attached to it and start making bad decisions to protect your feelings (you hate admitting a loss, and you rush to lock in a small win); that there are really only five good reasons to sell (the thing you believed turned out to have changed, you realize you were wrong from the start, the price got crazy-high, you found something clearly better, or one winner grew so big it's scary), and that "the price went down" or "the price went up" are never, by themselves, reasons; that the biggest, most expensive mistake is selling a wonderful thing too early just because it went up a lot (do the math: a small win today can be a fortune given away tomorrow); and that the way to beat your own feelings is to write down, before you buy, exactly what would make you sell, like tying yourself to the mast before the sirens sing. If your page needs the words "disposition effect," "reverse-DCF," or "margin of danger," you don't own it yet.

Journal prompt. Four entries. (1) Look at every position you have ever sold (or, if you have none, three sells by an investor you follow). For each, which of the five reasons actually justified it, or was it a Rabbit/Raider behavior in disguise? Be honest; this is the calibration that makes the next sell better. (2) Take your single best-ever winner (real or a case) and compute the §4.9 cost if you had sold it at 3×: what would it be worth now, and what multiple of the banked gain did you forfeit or protect? Write one sentence on what that does to your view of "taking profits." (3) Pick one position (or watchlist candidate) and write its full four-part pre-committed sell criteria now, before you're in the moment: testable pillars, break triggers, overvaluation zone, size cap. (4) Classify your most painful investing mistake on both taxonomy axes, run it through the M6.05 root-cause loop, and write the one checklist line it earns — or, if it was genuinely earned bad luck, write why it earns no line, and notice how that feels.

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