Financial Analysis

Financial Analysis — Hedge-Fund-Grade Equity Deep Dive

Disclaimer — read before use. This skill produces educational research output, not financial advice. Verdicts, entry zones, and position sizes are analytical exercises generated by a language model from third-party data that may be wrong, stale, or incomplete. Nothing here is a recommendation to buy or sell any security. Verify primary filings before acting on anything, and consult a licensed adviser where appropriate. Use at your own risk.

Overview

A complete bottom-up equity analysis workflow. One run produces two files:

Framework arithmetic (WACC, F-Score, Z-Score variants, Schilit thresholds, Porter scoring) lives in references/frameworks.md.

Completeness is the point. Every section in both templates is mandatory. If data is unavailable, write the section anyway and state the limitation — never silently omit it.

Exit Discipline — the stance this skill takes

This skill encodes a pyramid-on-winners / structural-sell-only philosophy. If you prefer trimming on valuation or price targets, this skill is not for you — that is a legitimate style; it just isn't this one.

1. Trims and exits are triggered ONLY by named structural changes to the thesis — pre-registered invalidators firing, not price behavior.
2. Price-target hits, position-size growth, parabolic patterns, mean-reversion expected-value math, and "valuation looks stretched" are NOT exit triggers. A position growing large via thesis playout is the desired outcome of pyramiding, not a violation.
3. Initial-entry size caps apply to entries, not to positions that grew.
4. DO NOT GENERATE "trim at $X" / "scale to N% by $Y" / stop-ladder exit schedules. If you find yourself writing a price-based exit rule, stop and re-frame it as a structural-sell trigger.
5. Mean-reversion analytics (reverse-DCF reads, negative-EV math) MAY be included as context, flagged informational — not a decision rule.

The one bounded exception is the 3-key Valuation Backstop defined in the deep template — a partial risk-management de-risk requiring extreme valuation AND a mania cluster AND an oversized position simultaneously. Absent any one key, valuation does nothing.

Structural-Sell Decision Rule (memorize)

Signals are staged: S1 peak narrative (informational), S2 supply/cost breakdown, S3 demand/customer breakdown, S4 technical (timing only).

S2 and S3 are weighted symmetrically: a single signal of either type is a watch tier; concurrence escalates. Valuation amplifier: while the name's valuation tier (MGDV block, deep template §3) is 🟡 Rich or 🔴 Extreme, amplify any structural action one notch — single S2/S3 → TRIM 25–50%; S2+S3 concurrent → full exit. Valuation never fires alone; it only amplifies a real structural signal. Exception — binary per-name triggers: the named L1-<TICKER>-N triggers you write in the deep template are pre-escalated encodings of this matrix (persistence and concurrence are built into their definitions, e.g. "ASP falls >15% QoQ for 2 consecutive quarters"). A CONFIRMED binary trigger fires the action stated in its own definition without being re-scored against the single-signal=WATCH row. The matrix governs raw signals not yet codified into a name's trigger set.

Workflow

Step 0 — Orientation

1. Identify ticker, exchange, and the investment theme it belongs to.
2. Ask the user for (or confirm) the output directory; default ./analysis/.
3. If <output>/<TICKER>.md already exists, this is an update — read it first.
4. Confirm the IRR hurdle for reverse-DCF: default 15%; use a higher hurdle (e.g. 20%) for emerging-market listings where currency drag and country risk premium justify it. State the hurdle used in the output.

Step 1 — Data collection (before any writing)

Pull quotes, profile, 5y annual + 8q quarterly statements, key TTM metrics, analyst targets, insider trades, and RSI/SMA50/SMA200 — the full list with provider endpoints is in references/frameworks.md.

Dependency: this skill needs a market-data source (FMP MCP tools, Alpha Vantage, or equivalent). If none is connected, STOP and tell the user which data is needed. With user consent you may proceed from user-supplied figures or filings — label every such number (user-supplied, unverified). Never invent figures.

Step 2 — Write TICKER.md

Follow references/qualitative-template.md exactly.

Step 3 — Write TICKER_deep.md

Follow references/deep-template.md exactly.

Step 4 — Quality gates

Formatting gates:

• Both files at correct paths; every template section present
• DCF shows bear/base/bull with arithmetic, not just outputs
• Piotroski: all 9 signals individually documented
• Schilit: DSO, accrual ratio, OCF/NI explicitly computed
• Porter: all 5 forces individually scored
• Entry zone PRICE RANGES identical between the two files (labels differ by design: §6 "Aggressive Entry" = near-current; verdict "Deep-Value Accumulation" = deepest)

Safety gates (these are the ones that matter):

• Thesis Roadmap: structural catalyst named in ONE sentence, or explicitly marked "DIFFUSE" with conviction capped at MEDIUM — never left vague
• Thesis Roadmap: 2–4 sector triggers present, and NONE of them worded as an exit rule — disconfirming triggers feed WATCH/refresh only
• ≥4 named structural-sell triggers, each stage-tagged S2 or S3
• At least 1 trigger is S3 demand-side (not only supply-side)
• MGDV block populated and valuation tier assigned
• Zero price-based exit rules anywhere in either file — grep your own output for "trim at $", "stop at $", "scale out" before declaring done
• Disclaimer note present at the top of BOTH output files

Red Flags — STOP and re-frame

If you catch yourself writing any of these, you are violating the discipline:

• "Trim at $X" / "take profits above $Y" / "stop-loss at $Z" as a rule
• "Position is now N% of the book, consider trimming" — size ≠ signal
• "Stock hit the bull-case DCF, de-risk" — a DCF target is not an exit
• "Parabolic move, mean-reversion is likely" — that's S4-only, do nothing
• "Priced for perfection per reverse-DCF" — informational context only

Common Mistakes

• Writing files before data collection finishes — partial data produces anchored, wrong numbers. Collect everything first.
• Applying the original Altman Z to non-manufacturers — use the Z″ variant per frameworks.md; the original mis-scores asset-light and financial names.
• Stale ERP — look up the current Damodaran US equity risk premium at analysis time; do not reuse a remembered constant.
• Skipping the reverse-DCF when price > bull case — that is precisely when it is mandatory.
• Copying analyst price targets into the MGDV — MGDV is fundamentals-only by construction; targets chase price in manias and would disarm the backstop.