Probability Weighted Range Of Outcomes

Prompt

add this as a memory called "Probability Weighted Range of Outcomes":

Point-in-Time Probability-Weighted Equity Outcome Analysis

Purpose
Use this as a reusable investment-analysis process for public equities. The objective is to estimate the forward distribution of outcomes from a specific decision date, not to explain history after the fact.
Core question:
> Standing at the decision date, with only the information available then, what were the plausible future outcomes, what probability should be assigned to each, what were the upside/downside fat tails, and was the expected value attractive?
This framework is especially useful when a stock has large optionality, thematic exposure, possible narrative change, structural inflection, or a misunderstood debate.

Core rules
- Point-in-time only: stand at the decision date and estimate forward outcomes from that point.
- No hindsight: exclude information that was not available at the decision date unless explicitly doing a hindsight/post-mortem version.
- Historical data is evidence, not the object: use prior data only to infer market expectations, base rates, valuation ranges, estimate revisions, positioning, cyclicality, or probabilities.
- Do not jump to probabilities: first identify the market debate, the key axes of uncertainty, and the plausible fat-tail events.
- Do not invent data: if critical data is unavailable, stop, ask for it, or produce only a clearly labelled hypothesis map.
- No expected value without sufficient data: only produce EV when the LLM has enough information to support scenario returns and probabilities.
- Always include tails: consider at least one plausible upside fat-tail and one plausible downside fat-tail, even if they ultimately get low probability.

Required setup
Before analysis, define:
Item | Required
Company / ticker | Yes
Decision date | Yes
Entry price at decision date | Yes
Forecast horizon | Yes
Investment question / core debate | Yes
Information set allowed | Yes
Information explicitly excluded | Yes
Opening framing:
> This is a point-in-time forward-looking analysis from [decision date]. The objective is to estimate the distribution of future outcomes from the entry price of [price] over [horizon]. Historical data will be used only as evidence to inform forward probabilities.

Step 1 — Data access gate
First classify available data. Do not proceed to full EV if critical data is missing.
Data area | Available? | Quality | Needed? | Comment
Company filings / transcripts
Historical price data
Consensus estimates / revisions
Valuation multiples
Peer / sector data
News / thematic research
Options / implied volatility
Positioning / short interest / flows
Industry / alternative data
Decision after data gate:
- proceed to full EV analysis
- proceed with caveats
- stop and request more data
- produce only a qualitative scenario map

Step 2 — Market setup: what was priced?
Infer what the market likely believed at the decision date.
Assess, where available:
- recent price move
- valuation versus history and peers
- consensus expectations and revisions
- management guidance
- prevailing investor narrative
- sell-side debate
- peer read-across
- positioning / short interest
- options-implied move or volatility
Output:
Market belief | Evidence | Confidence
Distinguish known facts, plausible interpretations, and unknowns.

Step 3 — Identify key axes of uncertainty
Identify the 3–7 variables most likely to drive future stock outcomes. Do not create scenarios yet.
Axis | Why it matters | Bull interpretation | Bear interpretation | Data needed | Data available? | Confidence
Examples of axes:
- revenue growth / product adoption
- pricing / margin structure
- competitive position
- customer concentration
- regulation / geopolitics
- capital structure / liquidity
- estimate revision path
- multiple re-rating or de-rating
- structural thematic exposure

Step 4 — Broad fat-tail research pass
Before assigning probabilities, search for non-linear outcomes.
Upside fat-tail categories:
- TAM expansion
- category shift
- architecture shift
- platform transition
- product breakthrough
- customer adoption inflection
- competitor failure
- pricing power
- supply unlock
- regulatory change that helps the company
- strategic partnership or M&A
- hidden asset value
- reflexive positioning / flows
- multiple regime change
Downside fat-tail categories:
- technological obsolescence
- product failure
- customer loss
- regulatory/geopolitical shock
- margin collapse
- supply chain constraint
- financing or liquidity stress
- accounting issue
- cyclical downturn
- competitive price war
- management credibility break
- crowded positioning unwind
- multiple compression
Output:
Fat-tail candidate | Upside / downside | Mechanism | Evidence as of date | What would need to be true | Keep / drop | Confidence

Step 5 — Prioritise scenario drivers
Select the main drivers that will define the scenario tree.
Usually include:
- 3–5 core axes
- 1–2 upside fat-tail candidates
- 1–2 downside fat-tail candidates
Output:
Included driver | Reason
Excluded driver | Reason

Step 6 — Research each selected driver individually
For each included driver, assess:
Field | Answer
What needs to be true?
Evidence supporting it
Evidence against it
What market likely priced
What was underappreciated
Base-rate or historical analogy
Data gaps
Probability implication

Step 7 — Build the scenario tree
Only after the research pass, create scenarios.
Typical scenario set:
- downside fat-tail
- bear case
- base case
- bull case
- upside fat-tail
For each scenario, define:
- what happens
- why it happens
- revenue / EPS / FCF implication, if available
- valuation multiple implication, if available
- terminal share price or return
- probability
- probability-weighted return contribution
- key justification

Step 8 — Estimate terminal price / return
Terminal price must be tied to available evidence, such as:
- earnings / revenue / FCF outcome
- valuation multiple
- peer-relative valuation
- historical trading range
- sum-of-the-parts
- balance-sheet downside
- options-implied move, if available
If valuation data is unavailable, state that the terminal price is assumption-driven and low confidence.

Step 9 — Probability-weighted outcome table
Final table:
Scenario | Description | Probability | Terminal price | Return from entry | Probability-weighted return | Key justification | Confidence
Probabilities must sum to 100%.
Calculate:
Metric | Value
Entry price
Expected terminal price
Expected return
Probability of positive return
Probability of losing money
Probability of >50% upside
Probability of >50% downside
Median scenario return
Formula:
Expected return = Σ (Probability × Scenario return)
Expected terminal price = Σ (Probability × Scenario terminal price)

Step 10 — Sensitivity analysis
Test whether the conclusion depends on fragile assumptions.
At minimum test:
- upside fat-tail probability
- upside terminal price
- downside severity
- base-case return
- time horizon
Output:
Sensitivity | Change tested | EV impact | Interpretation

Step 11 — Investment interpretation
Convert the distribution into a decision.
Decision item | Answer
Is the stock positive expected value?
Is the skew attractive?
What type of position does this support? | Core / starter / option-like / avoid
What would make you add?
What would make you trim or exit?
Top thesis monitors
Biggest uncertainty
Possible interpretations:
- high EV, high uncertainty, right-tail skew → small/medium option-like position, add on evidence
- high EV, high confidence → larger position possible
- positive EV but severe left tail → position sizing must be conservative
- flat EV but huge right tail → option-like only
- negative EV despite exciting story → avoid unless price changes

Final memo structure
Use this structure for the written output:
1. Executive summary
2. Decision setup
3. What the market was pricing
4. Key axes of uncertainty
5. Fat-tail research pass
6. Targeted evidence by driver
7. Scenario table
8. Expected value
9. Sensitivities
10. Investment interpretation
11. What would change the view

Important lesson from AMD case
The mistake to avoid is assigning zero probability to a plausible right-tail event just because it feels extreme.
In AMD's case, the key debate may not have been only quarterly data center GPU performance. The more valuable point-in-time question was whether agentic AI could make AMD's CPU franchise strategically more important and trigger a broader AI infrastructure re-rating.
That kind of outcome may deserve only a modest probability, but if the payoff is +100% to +200%, it must be included in the distribution.

Time horizon discipline
The timeframe is not a minor input — it defines the distribution. A next-quarter outcome, a 3-year compounding outcome, and a 10-year structural outcome are different analyses.
Before assigning probabilities, explicitly choose the forecast horizon and make sure every scenario is calibrated to that same horizon.
Common horizons:
Horizon | Typical use | What matters most
Next quarter / next print | Earnings setup, guidance risk, near-term catalyst | Estimate revisions, positioning, guidance, inventory, demand checks
6–12 months | Near-term thesis validation | Revenue trajectory, margin path, product launches, customer adoption, multiple change
2–3 years | Business-model or cycle outcome | Structural growth, market share, normalized margins, competitive position, capital allocation
5–10 years | Long-duration compounder or secular shift | TAM, moat durability, reinvestment runway, disruption risk, terminal economics
Rules:
- Do not mix horizons inside one EV table unless explicitly building a multi-horizon analysis.
- Scenario terminal prices/returns must match the chosen horizon.
- A fat-tail event may have different probabilities at different horizons; state that clearly.
- If the thesis requires a long horizon but the market catalyst is near-term, separate the two.
- When uncertain, produce separate 12-month and 3-year distributions rather than forcing one blended answer.
The setup table must include forecast horizon, and the final memo should state whether the expected value is short-term, medium-term, or long-term.

Missing timeframe rule
If the user does not specify a forecast horizon, the LLM must pause and ask for one before producing a probability-weighted EV.
Use this exact clarification:
> What timeframe should I use for the outcome distribution — next quarter / next earnings event, 6–12 months, 2–3 years, or 5–10 years?
If helpful, briefly explain the trade-off:
- Next quarter / next event: best for catalyst and earnings setup analysis.
- 6–12 months: best for near-term thesis validation and estimate revisions.
- 2–3 years: best for business-model, market-share, and cycle outcomes.
- 5–10 years: best for secular compounders, disruption, or long-duration TAM shifts.
Hard rule:
- Do not produce a final scenario probability table or expected value until the horizon is specified.
- If the user refuses to choose, default to showing separate 12-month and 3-year distributions, clearly labelled as illustrative.