The Expectations Game — How I Actually Pick Stocks
Prior work. Framework paper attached. Everything below is the compressed version — the mental models, the picks, and the rules that make the whole thing survivable over a decade.
The reframe that changes everything.Most people ask "is this stock cheap?" I think that's the wrong first question. The better question — the one that actually determines whether you make money — is: what does the world look like in 5-10 years, and which companies become obvious winners if that world happens?
Start with the world, not the ticker. Everything else — multiples, entry points, position sizing — flows from the answer to that. Get the world right and the company right, and the multiple resolves itself. Get seduced by a screen and you spend your career mean-reverting into value traps.
Valuation is not a number. Valuation is a story about future cash flow.The market is constantly asking whether your story is better than everyone else's. Alpha is the gap between what the market prices in and what actually happens. That's the whole game.
The single most important equation in equity returns. Stock Return = Actual Growth − Priced-In Growth ± Multiple Re-rating
Three moving parts. The market is always pricing in some growth path — that path is embedded in the current multiple. Buying a stock is implicitly betting that actual growth will exceed the priced-in path, or that the multiple itself will expand, or both. If actual growth merely matches consensus, your return is roughly the earnings yield — 2.5% on a 40× stock. Not enough.
The multiple is not a valuation. The multiple is the market's forecast, encoded in a price. Your job is not to compare multiples across companies. Your job is to figure out which company's forecast is most likely wrong in your favor.
Why the traditional value playbook stopped working.The dominant framework of the last 60 years was built for an economy that no longer exists — physical assets, cyclical industries, linear growth curves, interchangeable competitors. In that world, the "cheap" bucket usually contained temporarily out-of-favor good businesses that would eventually re-rate.
That world is gone. Today's economy is dominated by winner-take-most platforms, network effects, IP moats, and non-linear compounding. In this new regime: • The "cheap" bucket increasingly contains structurally impaired businesses that will not re-rate • The "expensive" bucket increasingly contains businesses whose future cash flow is so much bigger than what the market prices today that even 50-70× understates their eventual worth
Traditional value investors keep waiting for the mean to revert. It doesn't. There is no mean to revert to — the industry structure changed.
The valuation toolkit you should actually understand.Every serious analysis triangulates across at least two of these. Every method has a blind spot — the discipline is knowing which method to reach for when.
P/E (trailing). Fast. Works on stable mature businesses with normalized earnings. Lies at cyclical peaks (E is inflated), at turnaround troughs (E is depressed), and on hyper-growth (E doesn't reflect the future). Two companies at 15× P/E are not equivalent — one might be a mature 5%-grower priced correctly, the other a 20%-grower structurally mispriced by peer-anchoring.
P/E (forward). Better but still hostage to consensus estimates that get revised toward whatever management guides. Consensus is systematically anchored to last quarter.
DCF. Theoretically correct — value equals the sum of future cash flows discounted to today. Practically fragile: a 1% change in WACC swings fair value 15-25%, and any DCF where terminal value is >60% of total value is a story model, not a valuation.
The DCF trick that actually matters is reverse DCF. Take the current stock price and back-solve the growth rate, margin trajectory, and terminal value that consensus is implicitly assuming. That tells you exactly what you have to disagree with to buy the stock. That's the whole game.
P/S and EV/Sales. Useful for early-stage companies. Blind to margin — 10× P/S on a 30% gross margin business is not remotely equivalent to 10× on an 80% gross margin business.
P/B. Financials, REITs, insurance. Useless for asset-light software and IP-heavy businesses whose book value is essentially zero.
EV/EBITDA. Capital-structure-adjusted. Blind to capex intensity — makes telecoms, semis fabs, and utilities look artificially cheap because it ignores the massive reinvestment they need to survive.
PEG. Only as good as the G. You are outsourcing the hardest question — the real growth rate — to consensus, which is systematically wrong on both ends of the distribution.
FCF Yield. Ignores reinvestment opportunity. A low-yield business reinvesting at 30% ROIC is worth more than a high-yield business reinvesting at 5%.
Asset-based methods (NAV, book, replacement cost). Downside-floor discipline. Sets a floor, not a target price.
The one rule: pick the method that fits the business, not the business that fits the method. Applying DCF to hyper-growth, P/E to a cyclical peak, or P/B to software — every one is a common way to arrive at the wrong answer. Method selection is a decision, not a default.
The compound math that makes "expensive" actually https://t.co/YaiQ4fCkJP X trades at $500 with EPS of $10 — 50× forward. Consensus expects 15% EPS growth for 5 years then 8%. That's what's priced in.
What if actual growth is 35% CAGR for 5 years, then 20% for the next 5? Here's what happens to the same "expensive" stock:
Year 0 EPS: $10
Year 5 EPS at 35% CAGR: $45
Year 10 EPS at 20% CAGR from Y5: $112
Year 10 stock at 25× (normalized mature multiple): $2,800
$500 → $2,800 over 10 years = 5.6× return = 18.8% CAGR
Notice: the multiple compressed from 50× to 25×, and the stock still compounded at 19%/year. The multiple didn't matter because the earnings grew fast enough. That is what expensive-but-cheap actually looks like.
Duration is the alpha.Almost every professional investor operates on a 12-month horizon. Their bonus, their fund flows, their career — all locked to it. That creates a structural mispricing at longer horizons: the market is very good at pricing the next four quarters and very bad at pricing the next 40 quarters.
If you can hold a real thesis for 10 years while the market re-prices it every 90 days, you're exploiting a duration arbitrage most participants can't access — not because they don't see it, but because their structure won't let them own it. That's available alpha, and it's available specifically to individual investors who aren't on a bonus cycle. Patience is not a personality trait. It's a framework that has to be pre-committed.
The four value traps — cheap for a reason.Not every low P/E is an opportunity. Cheap for a reason is a real category:
Structural decline — end market is shrinking. Every year the fundamentals get slightly worse. Multiple never re-rates because terminal value keeps drifting down. Print media, legacy landline telecom, mall REITs.
Peak margin — margins are at a cyclical or structural high. When they revert, EPS falls faster than the market realizes. What looked like 10× becomes 25× on normalized earnings.
Cyclical top mistaken for value — automotive, chemicals, semis, and materials all trade at their lowest multiples at the top of their cycle. The right time to buy is when they look expensive on trough earnings.
Moat erosion — a structural competitor is eating market share. Absolute revenue may still grow, but share is declining. Once a moat starts eroding it rarely reverses.
Before buying any low-multiple stock, answer four questions in writing: (1) is the end market growing over 10 years, (2) are current margins normalized or cyclical peaks, (3) where is the industry in its cycle, (4) is market share stable, growing, or eroding. Any negative answer means the multiple isn't cheap — it's a warning sign.
The eight inevitable trends I actually own.The world in 2035 isn't knowable in detail. But directional inevitabilities are. Eight trends look locked in based on capital already committed, physical infrastructure already being built, and demand curves already visible. Every stock in my active book has to belong to one of these — or I don't touch it.
1. AI Compute — $NVDA · $GOOGL · $MSFT · $META The largest capex build in modern history. Hyperscalers committed hundreds of billions per year of infrastructure spend. Whether or not any specific application "works," the compute is being built. Value flows to whoever supplies it. GOOGL is my underappreciated pick here — first-party TPU silicon, Gemini catching up, priced like a mature ad platform.
2. Advanced Semiconductors — $TSM · $ASML · $AMAT · $LRCX · $ASX Every one of the other seven trends needs chips. Structural monopolies inside a growing pie. TSM is the fab layer of the whole AI trade — effectively a monopoly at leading edge. ASML is the sole EUV lithography supplier. ASX is under-the-radar advanced packaging exposure via CoWoS/HBM.
3. Energy Infrastructure — $VST · $CEG · $TLN · $NRG Electricity demand from AI, EVs, and industrial re-electrification growing after four decades of stagnation. Grid capacity under-built. Nuclear coming back. Natural gas peakers structural. Utilities entering their largest capex cycle in 50 years. VST/CEG/TLN own the nuclear + gas assets that hyperscalers are actively contracting for.
4. Robotics & Automation — $ISRG · $TSLA · $ABB Aging demographics + labor cost inflation + AI capability improvements = robots become economically viable across more categories every year. ISRG has an unusually clean moat — 20+ years and no competitor has closed the gap on Da Vinci. TSLA is where the humanoid / autonomous optionality lives (sized as asymmetric).
5. Digital Finance — $MSTR · $COIN · $HOOD · $BTC / $IBIT Buyer base for crypto has permanently changed (RIAs, corporate treasuries, sovereigns, US state pensions). Beyond crypto, tokenization of real assets and digital-native brokerages are structural infrastructure. MSTR is BTC-per-share compounding when mNAV is below cycle average (separate framework paper).
6. Defense & Space — $LMT · $NOC · $RTX · $RKLB · $VELO Great-power competition + supply chain reshoring + space commercialization = multi-decade capex cycle. Every OECD government increasing defense spend as % of GDP. LMT/NOC/RTX are the durable primes with multi-decade programs. RKLB is the only vertically-integrated US-owned alternative to SpaceX at commercial scale. VELO is my current highest-conviction asymmetric bet.
7. Data Centers — $EQIX · $DLR · $VRT The physical container for AI compute. Real estate, power interconnect, cooling infrastructure — extremely hard to build, highly regulated, direct AI capex beneficiaries. EQIX + DLR are the two dominant global colocation platforms. VRT is the bottleneck below the bottleneck (power + cooling).
8. Memory & Storage — $MU · $SNDK HBM demand from AI, DRAM cycle upturn, NAND from data centers. Historically the most cyclical business in tech — the AI overlay changes the trough behavior. Structural demand floor keeps rising, so troughs get shallower and shorter. Both are core positions.
The 5-filter screening funnel.Trend membership is the gate — nothing gets considered without it. But it's not enough. Every candidate name passes through five filters:
Trend membership — one of the eight, real not stretched
Durable moat — IP, scale, network effects, or genuine cost advantage
Working financial engine — ROIC clearly > WACC at reinvestment scale
Expectations gap — my base case meaningfully more optimistic than consensus, and I can explain why consensus is wrong
Downside floor — bounded loss in a legitimate bear case
Most names die at Filter 1 (not a real trend) or Filter 4 (no gap — everyone already sees it). The names that survive all five are usually 20-40 total across all eight trends. Everything else is watchlist or "not for me."
Portfolio construction — the barbell.Even the best stock-picking framework fails if sizing is wrong.
60% Core Compounders — obvious winners of the eight trends. Held 5-10 years. Sized big but not concentrated. NVDA, TSM, GOOGL, MSFT, MU, LMT, EQIX.
20% Asymmetric Bets — higher-variance names with real thesis but wider outcome distribution. Sized to lose without breaking the book. RKLB, MSTR, VELO, HOOD, COIN.
20% Cash / Preservation — dry powder for drawdowns. Not a return bucket — a response bucket.
Concentration limits:
Max 30% per trend
Max 12% per name
Min 12 / max 25 names in the active book
These rules aren't tight enough to prevent every bad outcome. They're tight enough to prevent the fatal ones — the "one bad quarter wipes out the whole account" scenarios that end investing careers before they compound. The purpose of position limits is not to constrain the upside. It's to constrain how badly you can be wrong on any one name.
The two mistakes that cause most failure.The mistake is thinking the most obvious winner today will always be the best risk-reward tomorrow. The bigger mistake is forgetting that compounding only works if you survive the cycle. Both mistakes are made by traders anchoring to price instead of framework.
Closing rules.
Start with the world, not the ticker
Multiple is the market's forecast, not a valuation
High multiples can be cheap. Low multiples can be traps.
Pick the method that fits the business — never the reverse
Duration is the alpha
Trend membership is the gate; five filters are the screen
Position sizing constrains bad outcomes, not good ones
Trend break is the only real thesis break at portfolio level
Drawdowns = valuation events, not thesis events
Ten years from now, sit and count
If the trends compounded and the multiple math held, the portfolio worked. That's the whole game.
Failures are lessons, successes are experience. Everything public.
— @AdamLi1250240 Independent Research · Personal Portfolio · Educational framework only · Not financial advice
$NVDA $GOOGL $TSM $ASML $MU $SNDK $VST $CEG $ISRG $MSTR $COIN $RKLB $VELO $EQIX $DLR #Investing #Valuation #StockMarket #AI #Semis #Compounding #FutureFirst
Take in detail:
https://t.co/c5xx9p2ytX
