Q&A

How much should I start with?

The question that mattered more than the literal answer ("now, with what you have") is why starting now matters more than starting with a lot.

The first ¥500K of personal investing isn't really about returns. It is about discovering how you respond emotionally when a ¥100K position falls to ¥70K in three weeks. Most people predict their own reaction wrong. The investor who learns at 25, with ¥500K at stake, that they panic-sell at -30% is far better off than the investor who learns the same thing at 45 with ¥50M at stake. The fee for the lesson scales with the position size — pay it small and pay it early.

So the recommendation in two layers:

Layer 1 — the index core. Open an account, automate a monthly contribution to a low-cost index fund inside a NISA wrapper. eMAXIS Slim All-Country or S&P 500 are the standard low-fee choices. The amount doesn't matter much — what matters is the habit. ¥30,000/month is fine. ¥300,000/month is also fine. Set it up and forget it for ten years.

Layer 2 — the learning capital. Carve off a small allocation (5–20% of the investment portfolio, never more than what could be lost without changing your life) for individual positions you actively research. This is where real learning happens. Start with one company you understand from your work, your hobby, or your household — somewhere you have a genuine information edge that ChatGPT and sell-side analysts don't. Write a one-page thesis. Buy a starter position. Re-read the thesis quarterly.

The mistake is waiting until you have "enough" or until you "know enough." Both are infinite-recursion problems. The actual education happens through owning real positions; the cost of starting late is unrecoverable.

Should I buy individual stocks or just an index fund?

Default is index. Switch some allocation into individual positions only when there is a specific gap thesis — a credible reason to believe a specific company is mispriced, with an articulable view that differs from the market. Without a gap thesis, holding individual stocks is speculation, not investing — and the long-term statistical result is worse than the index.

The honest split for someone working a non-investing job: 80–95% index, 5–20% active. The active sleeve is for learning and for the occasional high-conviction idea, not for trying to beat the market with diversified amateur stock-picking. Diversified amateur stock-picking is just an expensive way to recreate the index.

There is also a behavioural argument. Owning individual companies is more interesting than owning the index, and "interesting" buys attention. Attention to one or two companies you know well is what builds investing skill over time. Attention spread across 30 names is just noise — neither educational nor effective.

If the active sleeve outperforms the index over a five-year window across multiple ideas, scale it up cautiously. If it doesn't, scale it back and treat the time as tuition. The framework's role: the same Quality / Price / Two-engines model on the Frameworks page applies to both buckets. The index is a passive bet on aggregate market quality. The active sleeve is an active bet on specific quality-vs-price gaps. The mindset is one, the execution is two.

Should I focus on Japan, the US, or one specific industry?

The right answer is wherever you have informational advantage. For most attendees that means Japan markets first — because of language access, work proximity, daily-life familiarity, and the simple fact that Japanese companies are under-covered by global investors who don't read Japanese filings. A Japan-based investor reading 有価証券報告書 directly, talking to suppliers and customers in Japanese, observing consumer behaviour in person, has an edge that no New York analyst can replicate.

This doesn't mean exclude the US — the US market is deeper, more liquid, has better disclosure, and most of the canonical case studies are American. It just means start where the edge is, then expand.

Industry focus follows the same logic. Industries you understand from your job (banking, consulting, tech, manufacturing) are your circle of competence. Industries you observe as a consumer (retail, food, hospitality, entertainment) are an adjacent circle. Industries you have no exposure to (biotech, mining, defense) are outside the circle — they can be learned, but you start behind sell-side analysts who do that full-time. Don't fight unfair fights.

Within Japan specifically, three categories stand out for retail observation: consumer brands with international expansion stories (FAST RETAILING / Uniqlo, Nintendo, Sony), industrial companies with global moats (Keyence, Shin-Etsu Chemical, Tokyo Electron, Daikin), and structural-reform stories (the post-Buffett interest in trading companies — Itochu, Mitsui, Mitsubishi, Sumitomo, Marubeni). All three are reasonable starting hunting grounds for a Japan-based investor with limited capital and time.

How do I calculate a multiple?

A multiple is just one number divided by another. The two most useful ones:

P/E (price-to-earnings) = price per share ÷ earnings per share. If a stock trades at ¥3,000 and earned ¥150/share last year, its P/E is 20. Interpretation: the market is willing to pay 20 yen for every 1 yen of current annual profit. The inverse — the earnings yield — is 5%, which is what one share returns if profits were paid out in full.

EV/EBIT (enterprise value to operating profit) = (market cap + total debt − cash) ÷ operating profit. More technically correct than P/E for cross-company comparison because it neutralises the company's choice of capital structure. A company with a lot of debt looks cheap on P/E (lower share price relative to earnings, after debt servicing) but isn't actually cheaper; EV/EBIT normalises for this.

When to use which:

• Quick scan, single company → P/E is fine
• Comparing two companies in the same industry → EV/EBIT for fairness
• Comparing across industries with very different capital structures → EV/EBIT
• Highly-leveraged business (banks, REITs) → use specialised multiples (P/B, P/AFFO) instead

The number itself doesn't tell you whether something is cheap or expensive. A P/E of 10 can be a steal (a high-quality compounder mispriced) or a value trap (a melting ice cube). The interpretation requires the Quality framework on the Frameworks page — multiple plus quality plus growth runway, not multiple alone.

How do you actually find pricing power in less-obvious companies?

Pricing power is rarely advertised; companies that have it know better than to say so out loud (it attracts antitrust attention, regulators, and competitors). So it has to be read for indirectly. Four sources of signal:

1. Historical gross margins, stable or expanding through cycles. A company that holds 40% gross margin through a downturn — when its competitors are discounting to clear inventory — is signalling pricing power. A company whose margin contracts heavily in a downturn does not have it; the market controls the price, not the company.

2. Price increases over time, with stable volume. Read the management discussion in the annual report (Japan: 経営者による財政状態，経営成績及びキャッシュ・フローの状況の分析 section). If management consistently reports "price increased 3% / volume +2%" over multiple years, that is direct evidence. If price increases consistently come with volume declines, the company is choosing between price and growth — not the same thing.

3. Customer concentration data. A company whose top-10 customers are sticky over a decade has switching costs, which is one form of pricing power (the customer can't easily defect even at higher prices). A company whose top customers rotate every few years does not.

4. Direct conversation with customers. Philip Fisher's "scuttlebutt." Ask people who buy the product: would you switch if they raised prices 10%? If the answer is "probably not, it's not worth the hassle / the alternatives are worse / we're locked in" — that's pricing power expressed as a customer's revealed preference. Cross-check with reviews, industry forums, and salespeople for the competitor product.

For consumer brands, the easiest test is yourself. If you own Apple products, ask honestly whether you would switch to Samsung at 30% lower price. If the answer is no, the pricing power is real. If yes, it isn't.

How much financial analysis is sufficient?

The rule is materiality. Identify the three to five line items that materially move the thesis; study those carefully. Ignore everything else unless something breaks.

For Costco, those three to five line items are: membership renewal rate, membership-fee revenue trajectory, ticket size per visit, new club openings, gross margin (to verify the "we sell at near-cost" model is intact). Inventory days, accounts receivable days, depreciation policy — all could be perfectly fine or perfectly broken, but none of them materially changes the answer to "should I own this." So they don't get studied unless an earnings call mentions a problem there.

The trap that catches new investors is the opposite — reading the 10-K cover to cover, building a 50-line spreadsheet of every ratio, and at the end of two weeks of work having no clearer view than at hour one. Time spent on irrelevant detail is time stolen from the few items that actually matter, plus it produces a false sense of confidence ("I did all this work, so I must understand it").

The discipline: before reading anything, write down three to five specific questions the research is meant to answer. If the question can't be specified, don't start the research — you'll find what you're looking for in the noise, which is a hallmark of bad analysis. The questions evolve as understanding grows; the discipline of having them in writing prevents reading drift.

For a 5% position in a portfolio, 20 hours of focused research on the right line items is more valuable than 200 hours of generic 10-K reading. Position size determines depth — the next question below covers the other end of that spectrum.

Do PE shops over-analyze?

Yes and no, depending on context.

In a leveraged buyout, the analysis depth is necessary, not excessive. PE firms are concentrating a large fraction of a fund's position into a single deal, often with multiple turns of debt, and the only exit is via sale or IPO three to seven years later. There is no daily liquidity to correct an analytical mistake. Under those conditions, exhaustive operational, financial, legal, environmental, and management diligence is rational.

For public-equity investing the same depth would be a waste. A 5% position can be sized down or out tomorrow if new information arrives; daily liquidity is the safety valve that lets the analyst be roughly right rather than precisely right. Spending PE-level effort on a public position is misallocating the most expensive resource (time) against an asset class where the work doesn't pay back in incremental conviction.

The right framing: depth of analysis should match the irreversibility of the decision plus the position size. A reversible, small position deserves quick research. An irreversible, concentrated position deserves deep research. PE work is expensive because PE conditions demand it; public-equity work that mimics PE conditions is unnecessary expense.

The professional pattern: spend 20 hours of focused research to get to a yes/no on whether a 5% position makes sense. If yes, monitor quarterly with maybe one hour of work per quarter unless something breaks. The marginal hour of research beyond the first 20 produces dramatically less insight than the first hour did. Stop when that curve flattens.

When is the right time to sell?

The discipline: sell when the thesis breaks, not when the price moves. This is one of the hardest rules to internalise because the market constantly invites the opposite — the price screen rewards reaction.

Two clean reasons to sell:

1. The bull case is in the price. The market has come around to your view, the multiple has re-rated, and the upside-to-downside ratio that justified the original buy no longer exists. The opportunity is over. Sell and redeploy the capital to the next gap — even if the company itself is still a good business. Don't conflate "good company" with "good investment from here." A wonderful business at an over-priced multiple is a poor forward investment.

2. The bear case is materialising. The kill-thesis that was written at purchase (see Research → Stage 3) is happening in real time — pricing power eroding, ROIC dropping, management making destructive capital-allocation decisions, the moat narrowing. Sell even at a loss. The loss is the cost of the lesson; holding on hoping the thesis recovers is what turns a 30% loss into a 70% one.

Reasons that are not reasons to sell:

• Price went down. If the thesis is intact, a lower price is an opportunity to add, not a reason to leave. The market periodically offers your favourite businesses at discounts; that's a feature, not a bug.
• "I'm up a lot, let me take some off the table." This is psychology, not analysis. Either the thesis is still intact (hold) or it isn't (sell completely). Trimming for emotional comfort is a sign of weak thesis conviction.
• Generic macro fear. "The market feels expensive" is not a thesis-break for the specific name you own. Either the specific name's thesis has changed (sell) or it hasn't (hold). Macro moves the prices around your positions; it doesn't generally change the businesses underneath.
• "I need the money" for something unrelated. Plan liquidity needs separately from investing — selling forced by an unrelated cash need is paying for that need with whatever the market happens to offer that day, which can be a steep discount.

The selling decision is symmetric with the buying decision: both depend on the gap between intrinsic value and market price. When that gap closes (Reason 1) or inverts (Reason 2), sell. When it persists, hold.

How good is ChatGPT for investment analysis? Is it better than a human analyst?

Better than most people expect; worse than the hype implies. The honest assessment, used in practice:

Where it's excellent: the first five hours of any new name. ChatGPT, Claude, Gemini, or any modern frontier model is essentially a research analyst on tap — summarise a 10-K, build a Porter's Five Forces view, lay out the bull and bear cases, surface the major competitors and how they compare. This used to be an entry-level analyst's job. The model does it in 10 minutes for the cost of a coffee. The work that used to fill the first week of a new-coverage initiation is now a single thoughtful prompt.

Where it's mediocre: synthesis from public information. The model is trained on the same public sources every other reader has access to — analyst reports, press releases, news coverage, 10-Ks. If "what does the market believe" is the question, the model can answer well. If "what does the market not yet believe" is the question — which is the entire game in active investing — the model can only guess from public material, which by definition is already priced in.

Where it's terrible: primary research. Conversations with former employees, suppliers, channel-checks with customers, attendance at investor days and conferences, building a relationship with management's IR — none of this is in the training data. The investor who relies on AI for synthesis but does primary research themselves has a real edge over the investor who relies on AI for both.

The practical workflow that actually works:

1. Give the model the 10-K and the framework — "apply the six-quality-dimensions framework from this methodology to this company."
2. Ask for the bear case explicitly — "what would a short seller argue here, in their strongest framing?"
3. Ask for the variant-perception attempt — "what might the market be wrong about?" Then test those hypotheses with your own research.
4. Have it summarise specific filings, transcripts, investor-day decks — the model is faster than you at extracting structured information.
5. Do not ask it "should I buy this?" — that question collapses the model into a cheerful, useless response. Ask for components of the decision, not the decision itself.

Diane's follow-up at the workshop ("you have to feed it the right information") is exactly right. Bad prompt → useless output. Good prompt with context → analyst-quality first pass. The skill is shifting from "can you read a 10-K" to "can you ask the right questions about a 10-K," and the latter is still a human skill.

The deeper structural shift: primary research is now the single most valuable activity, because it's the only research not commoditised. The old workflow was "spend a week reading public material, then decide." The new workflow is "spend a day having AI summarise public material, then spend the week on primary research." Same total time, dramatically better output. Amateur investors who get this right have a wider edge over professional investors than they did five years ago — because the professionals are using the same AI tools, and the primary-research differential is the same for both.

Why does the US market outperform Japan long-term?

The short answer: the United States allows businesses to fail and Japan, post-1990, structurally did not. The longer answer covers four reinforcing mechanisms.

1. Bank-loan reset failure (1990s). When Japan's asset bubble collapsed in 1990, regulators chose not to recognise loan losses on bank balance sheets. Banks were allowed to roll over non-performing loans for years rather than writing them down and recapitalising. Result: bank balance sheets remained impaired, new lending capacity collapsed, and the credit channel that fuels productive investment stayed broken for over a decade. The US, by contrast, forced bank recapitalisation (RTC in the early 90s, later TARP in 2008–09) — painful in the short run, productive over time.

2. Bankruptcy-as-shame culture. In Japan, individual and corporate bankruptcy carry social stigma that is harder to recover from than in the US. The Japanese entrepreneur who fails once is often locked out of a second attempt; the US entrepreneur who fails twice can raise their third venture. Across an economy, failed capital and failed founders stay stuck rather than being recycled into productive new ventures. Innovation cycles slow accordingly.

3. Cross-shareholding and weak shareholder governance. Japanese companies historically held large stakes in each other (the keiretsu structure), which meant shareholders weren't in a position to pressure management for better capital allocation. A bank that owns 5% of your company isn't going to tell you to fire 20% of your workforce. The result: companies that retained excess employees and excess cash, optimising for stability over returns. The US shareholder base has always been more activist, and the post-2000 rise of institutional shareholders accelerated the gap.

4. Employment-protection vs. labour mobility. Japan's stronger employment protections make it expensive to lay off workers, which means companies hire more cautiously and innovate more slowly. The US's at-will employment is brutal at the individual level but produces faster reallocation of labour into newly productive sectors. This shows up directly in productivity-growth statistics over multi-decade windows.

None of this is a moral judgement. Japan's social model has produced one of the most stable, low-crime, high-quality-of-life societies in human history. The trade-off is real: capitalism with shock absorbers (Japan) produces fewer crashes and slower growth; capitalism with crashes (US) produces faster growth and more catastrophic losers. Whether one is "better" depends on what is being optimised.

What's changing. The past five years have seen a real shift in Japan — TSE pressure on companies trading below book value, the new NISA expansion increasing retail equity participation, governance reforms requiring independent directors, and structural pressure from activist investors (Palliser's TOTO thesis is the canonical recent example). The long stagnation may not be permanent. The next decade is more interesting for Japanese equities than the past three were.

Does the Japanese long-term-investing thesis (Shibasawa-style, 30+ year horizon) still work?

The thesis — that Japanese companies built to last centuries reward patient, long-term capital — has real merit but is structurally challenged in the current era.

The merit. Several Japanese companies are genuinely multi-generational compounders with quiet, durable moats. Shin-Etsu Chemical (silicon wafers, PVC), Keyence (industrial sensors, 50%+ operating margins for decades), Hoya (optics, semiconductor masks), Daikin (HVAC), and several of the Buffett-favoured trading houses (Itochu, Mitsui, Mitsubishi, Sumitomo, Marubeni) have produced compelling long-term returns when held through cycles. The "buy great Japanese businesses and hold forever" strategy works on the right names.

The challenge. Innovation cycles have accelerated globally. The pace at which a competitive advantage can be eroded is faster than it was in 1980 — software-enabled competitors scale globally in five years, AI-enabled disruption is rewriting industry boundaries every two years. A Japanese company on a 3–5 year decision cycle competing against an American or Chinese company on a 6-month decision cycle will lose share over time, even if its core moat is strong.

So the long-term thesis works if the company's moat is structural (geographic, regulatory, irreplicable specialised knowledge) rather than executional (better at decision-making than competitors). The Buffett trading-house bet works because the moats are structural — control of physical commodity flows, balance-sheet capacity, long-standing customer relationships. A Japanese consumer-software company facing global American competition is a different bet.

The practical adjustment for a Japan-based long-term investor:

• Favour companies with structural moats (specialised manufacturing, irreplicable IP, network effects in domestic markets that don't transfer easily) over execution-moat companies.
• Favour companies that have demonstrated willingness to evolve (TOTO pivoting into chip-equipment ceramics; Sony pivoting into IP/entertainment) over companies clinging to legacy business models.
• Discount any company whose moat is "we make the highest-quality product in a market where global competitors are catching up" — that moat is melting.

Shibasawa-era thinking ("seek harmony with society, build for centuries") is sound philosophy. The execution requires updating for an accelerated world. The long horizon stays; the assumption that the world will wait does not.

How much do you weight management quality in the investment thesis?

It depends on what kind of investment is being made.

For low-quality cyclical businesses, management matters less. A commodity oil company's returns are dictated by the oil price, the cost curve, and the capital cycle. Management can deploy capital well or poorly, but the structural returns are set by the industry. Don't over-weight CEO interviews when investing in cyclicals — the industry dynamics overwhelm management decisions.

For high-quality compounders, management matters enormously. The reason compounders work — the moat staying wide, the reinvestment compounding at high ROIC, the long runway being preserved — depends on management choosing to defend the core rather than chase distractions. Five things to study:

1. Capital-allocation track record. Look at the last 10 years of major capital decisions — buybacks at what multiples? Acquisitions at what return? Capex into high-ROIC projects or vanity? The capital-allocation history is the cleanest predictor of the next decade of capital allocation, because management style rarely changes.

2. Incentive structure. Compensation tied to short-term EPS produces short-term decisions. Compensation tied to long-term TSR or per-share metrics produces longer-term decisions. The proxy statement (株主総会招集通知) reveals this directly.

3. Founder-led vs. professional-CEO. Founder-led companies tend to defend the core when pressured (Zuckerberg at Meta, Huang at Nvidia, Yanai at FAST RETAILING). Professional CEOs tend to optimise for the next compensation cycle. Both can be right; the base rates are different.

4. Language drift over time. Read 10 years of earnings-call transcripts back-to-back. Did the management's claims age well? Did they hit their targets, or did the goalposts move every two years? Pattern recognition for evasion (changing metrics, vague forward statements, defensive answers to specific questions) is a learnable skill and a strong negative signal.

5. Cultural signals. Layoff patterns during downturns (does the CEO take a pay cut?), how senior departures are handled, how the company talks about competition. Cultures that respect the truth tend to make better long-term decisions than cultures that respect the hierarchy.

The cleanest single diagnostic: a great compounder run by a great capital allocator can compound for decades. A great compounder run by a mediocre capital allocator can still produce returns, but caps out. A great compounder run by a destructive capital allocator (overpaying for acquisitions, buying back stock at peaks, chasing fashionable adjacent markets) destroys value faster than the underlying business creates it.

For Japanese companies specifically, the recent governance reforms have started pressing for better disclosure of CEO succession plans, independent-director ratios, and shareholder engagement. This makes the "leadership quality" research easier than it was a decade ago. Read 株主総会招集通知 for proxy data; read 統合報告書 (integrated reports) for the management narrative.

Do you short stocks? Can you tell when a crisis is coming?

No short-selling. Two reasons:

1. Asymmetric payoff in the wrong direction. A long position has bounded loss (the position can go to zero, but no further) and unbounded upside (a great investment can compound 10×, 50×, 100×). A short position has bounded upside (the stock can go to zero, returning at most 100% on the short) and unbounded loss (the stock can go up arbitrarily — Tesla shorts learned this the hard way). The compounding math that works in favour of long-only investing works against short-selling.

2. Time is the enemy. Even when a short thesis is correct, "the market can stay irrational longer than you can stay solvent" (Keynes). Borrowing costs, margin calls, and short-squeeze risk all compound while you're waiting for the market to come around. Long-only investors who are wrong for two years just see flat returns; short-sellers who are wrong for two years can be wiped out.

The exceptions where shorting makes sense as a skill:

• Pure hedging — shorting an index against a long book to neutralise market risk, not to make money on the short. Different objective.
• Forensic short-sellers (Citron, Hindenburg, Muddy Waters) who identify outright fraud and have a specific catalyst (regulatory action, accounting restatement). This is a specialised craft with a very different research process than long investing — closer to investigative journalism than portfolio management.
• Pairs trades — long the high-quality name in an industry, short the low-quality competitor. The market exposure cancels out; the bet is on relative quality. Sometimes useful, but not the workshop's strategy.

On predicting crises. The honest answer is that nobody reliably does this. The investors who appear to have predicted past crises (Michael Burry, John Paulson, George Soros) were typically right about a specific mechanism, not "the market is going to crash." Burry was right about subprime mortgage quality, not about timing the broader crash. The investors who claim to predict crashes generally hide the dozens of times they were wrong.

The pragmatic posture: don't try to predict crises. Instead, prepare for them. Hold dry powder (cash) when valuations are stretched. Hold higher cash positions when the personal opportunity set looks thin (the workshop's track record shows ~29% average cash over four years). Maintain a written list of names you'd buy at specific prices if the market dropped 30%. When the crisis arrives — and one always arrives eventually — execute the list mechanically without trying to time the bottom.

The 2022 META drawdown is the case study (see the Frameworks page). The investor who tried to predict the drawdown was probably out of the stock by 2021. The investor who held cash, had META on the watch-list at a price below $150, and executed when the price arrived — that investor made money. Preparation beats prediction.

The Big Short (the more famous film with similar themes to Freefall) celebrates the prediction approach because it makes a better story. The actual track records of the Burry-style predictors over their full careers are far less impressive than the single famous trade suggests. Preparation, position-sizing discipline, and patience are the boring strategies that compound. Prediction is the entertaining strategy that mostly doesn't.