Research process

52-week-low list

The dollar store. Filtered for size and liquidity, the 52-week-low list is the most efficient sourcing tool for value investors. Most are cheap for good reasons; a few are cheap for bad ones (overreaction, forced selling, sector contagion). Your job is the second category. If you see onions on sale 50% off, you don't walk past assuming they're rotten — you check.

Spinoffs, post-bankruptcies, hated industries

Structural orphans. Spinoffs are often misunderstood because they're new tickers without coverage; post-bankruptcies are avoided by index funds and momentum players; hated industries are over-priced for fear. All three are places where price ≠ value because attention is structurally thin.

Great businesses waiting for a fair price

Watchlist. Identify the businesses you'd love to own, write the thesis, set the price, wait. Most will never reach your price; some will, occasionally suddenly (a quarter miss, a sector rotation, a macro panic). Patience converts a 30-stock universe into a 3-stock buying opportunity once or twice a decade.

Investor day presentations

Companies dump unusual detail at investor days — five-year plans, segment economics, capital-allocation priorities. Read every transcript from companies you're interested in; they reveal more than any sell-side note. Especially read investor days held when the company is in trouble — they have the most to explain.

Your circle of competence

The industries and business models you understand from work, life, or family — where you can read a 10-K and notice things sell-side analysts miss. Your circle is smaller than you think; staying inside it is most of the discipline.

Cross-industry shifts

Sell-side analysts cover vertical sectors; major shifts happen across them. EV → battery materials → mining; AI → datacenters → power generation → utilities. Whoever notices the second-order effect first gets the second-order opportunity.

Management commentaries

Sometimes the market doesn't believe what management says — and sometimes that's the market being wrong, not management lying. Track which CEOs have a track record of being literal about their forward statements; those companies are reliable sources of variant perception when the market disbelieves them.

Three filters

Circle of competence. Can you actually evaluate this business, or are you about to learn a new industry mid-thesis? If the latter — pass.

Balance sheet survivable? Debt-to-EBITDA, near-term maturities, covenant headroom. Too much debt means you need to be very precise on the valuation, and you barely know the company yet. Pass.

Margin of safety on a quick valuation? Greenwald's Step 1 (asset value) takes 30 minutes and produces a downside floor. If market price is well above that floor and quality isn't obvious, pass.

The discipline of saying no fast

The 5-minute kill exists because attention is the scarce resource. An idea pipeline of 100 candidates per year, all spending 20 hours each, is 2,000 hours — more than the year contains for anyone with another job. The kill ratio has to be brutal: 80–90% of candidates need to be passed in the first five minutes so the remaining 10–20% can get the focused depth they deserve.

Three ways the discipline breaks down:

• Confirmation bias on the way in. Reading the bullish material first builds attachment before the bear case is examined. Fix: read the bear case before the bull case on every new name.
• Sunk-cost on partial research. "I've already spent two hours on this, may as well finish." That's the loss talking. The two hours are gone regardless; the question is whether the next eighteen hours have positive expected value. Usually they don't.
• "But what if I'm wrong" hedging. Refusing to commit to a yes/no because either decision could turn out badly. The cost of holding everything as "maybe" is that nothing gets the focused depth. Force a decision; you can always revisit later if new information arrives.

The 5-minute kill template

Copy-paste into a notes file. Should take 5–10 minutes to fill in for any new name. If any answer is "I don't know," the answer to "should I continue" is no.

Date: ____
Ticker / Company: ____
Source (where I heard about this): ____

CIRCLE OF COMPETENCE
- What does this company do, in one sentence: ____
- Do I understand the industry well enough to evaluate it (yes/no)? ____
- If no — STOP.

BALANCE SHEET
- Net debt / EBITDA: ____
- Nearest debt maturity: ____ (year)
- Could this company survive a 30% revenue decline for 2 years? ____
- If shaky — STOP.

QUICK VALUATION
- Current EV/EBIT (or P/E): ____
- Greenwald Step 1 estimate (asset value floor): ____
- Margin of current price to asset floor: ____%
- If price >> floor and quality not obvious — STOP.

DECISION (yes / no / hold for re-check):
Reason in one sentence: ____

Lens 1 — Industry structure

Five Forces (Porter), Value Chain analysis, Capital Cycle (where in the cycle is the industry — overcapacity? undersupply?). The aim: understand the structural economics, not just this company's economics. A great business in a deteriorating industry is a different bet than a great business in an improving industry.

Lens 2 — Business economics

ROIC + reinvestment opportunities. The two numbers that determine compounder quality. Unit economics, segment-level margins, where the cash actually comes from. Distinguish the consolidated story from the underlying segment stories (Meta example: FoA vs Reality Labs).

Lens 3 — Management & capital allocation

Incentives (compensation structure), buyback history (at what multiples?), M&A track record (returns on acquisitions), culture, succession. The single best predictor of next-decade capital allocation is the last-decade capital allocation.

Sources to triangulate

Filings (10-K, 10-Q, proxy) — segment economics, MD&A trends, related-party transactions, compensation structure.

Transcripts & investor days — language drift, evasions, capital-allocation signals.

Competitors, suppliers, customers — three-side triangulation. Read competitor 10-Ks for the same industry; talk to suppliers about pricing power; check customer churn proxies.

Primary research — expert calls, scuttlebutt, AI-assisted synthesis. AI accelerates the secondary research; primary research (talking to people in the industry) is where the real edge lives.

The three questions

1. What does the market believe? Not just the price — the story behind the price. Read recent sell-side reports, financial-media coverage, the consensus EPS estimate trajectory.
2. What do I believe? Your independent view from the deep read.
3. Why am I right? Explicit articulation of where the market is wrong and what will cause it to change its mind.

If you can't answer all three, you don't have a thesis — you have an opinion. Pass or do more work.

The discipline behind the questions: separate facts from opinions, and label each. A research file should be readable as two columns — what is verifiable (revenue grew 12%, gross margin held at 40%, the CEO sold $30M of stock) and what is interpretation (the moat is widening, management is becoming more shareholder-friendly, the next product cycle will be the biggest yet). The interpretation column is where edge lives; the facts column is what supports or refutes it. Mixing the two — treating an opinion as if it were a fact — is how theses become unfalsifiable and how investors stop noticing when reality has changed.

The memo

Write the full thesis before buying. Prevents thesis creep after the fact (the post-hoc justification trap when something changes).

Memo structure:

• Thesis. One-paragraph elevator pitch + supporting evidence.
• Variant perception. Explicit market-view vs own-view comparison; the gap named.
• Catalysts. What will cause the market to update? Earnings prints, segment disclosures, capital-allocation announcements, sector rotations.
• Kill-thesis. Three years from today, this investment was a disaster. What happened? If you can't write this, you don't understand the risk.
• Position sizing rationale. Conviction × margin of safety / downside × quality.

The kill-thesis as discipline

Most theses fail because the writer couldn't (or didn't) articulate how it would fail. The kill-thesis exercise forces you to confront the bear case while you're still detached — before you own the position and become emotionally invested in not being wrong.

If your kill-thesis is something like "the multiple just doesn't expand," your edge isn't real — that's not a thesis-breaker, that's slow returns. A real kill-thesis names a specific quality factor (slide 20) that deteriorates, with a specific mechanism. "Pricing power erodes because [X new competitor] takes share with [Y product] starting in [Z quarter]."

Position sizing

Size = f(conviction, margin of safety, downside, quality).

• Conviction — how strong is your variant perception? Strong conviction lets you size larger.
• Margin of safety — how far below intrinsic value? Larger margin → larger size.
• Downside — what's the worst case? Larger downside → smaller size.
• Quality — high quality survives longer, lets you wait. Higher quality → larger maximum size.

These four pull in different directions. A high-conviction, high-margin-of-safety, low-downside, high-quality idea (rare) might earn a 10–15% portfolio weight. A typical alpha-generator earns 3–5%.

Monitoring

Quarterly review at minimum. Off-cycle review when something specific happens — earnings, M&A, major news. The discipline at each review: is the thesis still intact? Not "is the price still going up" — that's irrelevant. "Has anything changed in the kill-thesis territory?"

Selling

Sell when the thesis breaks, not when the price moves. Two kinds of thesis breaks:

• The bull case is in the price. The opportunity is gone; the market has caught up. Sell and redeploy.
• The bear case is materializing. The kill-thesis is happening. Sell even at a loss — the loss is the cost of the lesson, and the alternative is a bigger loss later.

What's NOT a reason to sell: "the price went down." If the thesis is intact, a lower price is an opportunity to add, not a reason to leave.

Memo template (full thesis, before buying)

Length: 1–3 pages. Written in plain prose, not bullets. Date it.

# [Company] — Investment Memo
Date: ____
Position size proposed: ____% of portfolio
Author's prior research hours: ____

## Thesis (one paragraph)
[The elevator pitch. What this business is, why it's mispriced,
what closes the gap. Should read as a coherent argument, not a list.]

## Quality assessment (six dimensions)
1. Pricing power: ____
2. Long-term growth runway: ____
3. Incremental ROIC: ____
4. Cash conversion: ____
5. Cyclicality: ____
6. Capital allocation: ____

## Valuation
- Greenwald asset value: ____
- Greenwald earnings power value: ____
- Greenwald growth value: ____
- My intrinsic value estimate: ____
- Current price: ____
- Margin of safety: ____%

## Variant perception
- What the market believes (cite sell-side / consensus): ____
- What I believe: ____
- Why I think I'm right: ____
- Why I might be wrong: ____

## Catalysts (what will close the gap)
- Specific events / dates that should update the market: ____

## Kill-thesis
"Three years from today, this investment lost 70%. What happened?"
[Specific, plausible bear scenario. If can't write it, don't buy.]

## Position sizing rationale
[Walk through Conviction × Margin of Safety × Downside × Quality
to justify the proposed size.]

## Monitoring plan
[What I will check each quarter. What thesis-break triggers a sell.]

Quarterly review template

One per held position, every three months. Should take 30–60 minutes per position. Comparing answers across reviews is where the real signal lives.

# [Company] — Quarterly Review
Date: ____  |  Quarter: ____
Position size: ____% (started at: ____%)
Price now: ____  |  Price at memo: ____  |  Change: ____%

## What happened this quarter (factual)
- Earnings result vs. consensus: ____
- Management commentary highlights: ____
- Competitive developments: ____
- News / events affecting the thesis: ____

## Thesis status
- Is the original thesis (from memo) still intact? (yes/partial/no): ____
- Specifically what has changed vs. memo: ____
- Kill-thesis status — any signals firing? ____

## Action this quarter
- Hold / add / trim / exit: ____
- Reason in one sentence: ____
- If holding: anything to revisit next quarter? ____

## Honest self-assessment
- Am I rationalising? Anchoring on price? Defending out of ego? ____
- What would convince me to change my mind? ____

Sell decision template (when the thesis breaks)

Run this when the kill-thesis has fired or when the bull case is fully priced. Forces the decision to be made deliberately, not emotionally.

# [Company] — Sell Decision
Date: ____  |  Original purchase date: ____
Current price: ____  |  Cost basis: ____  |  P/L: ____

## Why selling
Reason category (mark one):
[ ] Bull case in the price — opportunity closed, redeploy
[ ] Bear case materialising — kill-thesis firing
[ ] Cluster risk reduction — too concentrated in factor X
[ ] Portfolio rebalance — better opportunity elsewhere

## Honest check — is this an emotional sell?
- Has the BUSINESS thesis broken, or just the PRICE? ____
- Would I be selling if the price had risen instead of fallen? ____
- Am I anchoring on cost basis? ____
- If I had no position, would I buy at today's price? ____

## Lessons (write before the sell, while it's vivid)
- What did I get right in the original analysis: ____
- What did I get wrong: ____
- What would I check next time before buying this profile: ____

## Action
- Sell all / sell half / sell quarter: ____
- Redeploying capital to: ____ (or hold cash)

Free, essential

• SEC EDGAR (sec.gov/edgar) — primary-source filings for any US-listed company. 10-K, 10-Q, 8-K, proxy. The original document, before any analyst's interpretation. Use the full-text search to find specific phrases across filings.
• EDINET (disclosure2.edinet-fsa.go.jp) — Japan's equivalent. 有価証券報告書, 四半期報告書, 株主総会招集通知 directly from the source.
• Company investor-relations pages — investor decks, earnings presentations, earnings-call audio, historical archives. Often better-organised than EDGAR/EDINET.
• StockAnalysis.com (stockanalysis.com) — clean, free financial statement summaries with 10-year history. Faster than building from filings for first-pass scanning.
• Macrotrends (macrotrends.net) — long-run historical multiples (P/E, EV/EBIT, P/B), ratios, and chart history. Useful for "is this name historically cheap or expensive."

Free, useful for deeper work

• Earnings-call transcripts — Seeking Alpha (free with registration, paywalled archive), Motley Fool Transcripts (free), or directly from the company's IR page (often a same-day audio link, transcript a few days later).
• Koyfin free tier (koyfin.com) — Bloomberg-style dashboards with charts, screens, and comparison views. Free tier covers most retail use cases.
• Finchat (finchat.io) — AI-powered financial-statement search and segment-level data. Free tier is generous.
• X / Twitter financial-analysis community — selectively. Following 20–30 sharp practitioners is better than reading sell-side notes. Treat as a source of pointers, not conclusions.

Paid, worth it if doing this seriously

• Tegus (or AlphaSense, or Stream.com) — expert-call transcripts. Primary-research-at-scale; reading 5–10 expert calls on a company is closer to spending time with industry insiders than any amount of public-source reading. ~$5–20K/yr for retail tiers depending on access level.
• Bloomberg or FactSet — institutional-grade tools. Realistically not needed for this style of investing; the budget is better spent on Tegus-class primary research.
• Quartr (quartr.com) — earnings-call audio + transcripts library, slick mobile UI. Useful if commute time is research time.

AI tools (changed the workflow)

Frontier LLMs (Claude, ChatGPT, Gemini) have changed what a single analyst can produce. The practical workflow:

• 10-K summarisation — paste the filing, ask for structured summary against the six-quality-dimensions framework. Saves 2–3 hours per filing.
• Bear case generation — ask explicitly: "what would a short seller argue here, in their strongest framing?" Better than relying on your own ability to argue against your own thesis.
• Cross-filing comparison — paste two competitors' filings, ask for differential analysis. Faster than reading both end-to-end.
• Transcript scanning — paste last 4 quarters of transcripts, ask for language drift / unfulfilled forecasts / topic-frequency changes. Patterns invisible to skimming.

What AI can't replace: primary research (expert calls, customer conversations, industry events), original variant perception (the model can only synthesise public material, which is already priced in), or the judgment call on whether to actually buy. See the Q&A AI section for the longer treatment.