market-analysis

name

market-analysis

description

Use when exploring a market, sector, theme, or macro regime top-down with source-cited research — e.g. "analyze the AI semis market", "what's happening in European defense stocks", "explore the lithium thesis", "give me a view on US small caps", "is the consumer rolling over", "map out the Mag 7 vs the rest". Triggers on phrases like "analyze the X market", "explore X theme", "what's the state of X sector", "give me a top-down view", "market analysis", "sector deep-dive", "macro view on X", "X thesis". Produces a structured report with inline footnote citations to primary sources (filings, central-bank releases, official statistics, exchange data, reputable financial press). Distinct from `stock-eval` (single-name fundamentals) and `research-bundle` (single-name multi-skill chain) — this skill is top-down only and always cites sources.

Market Analysis

Top-down, source-cited exploration of markets, sectors, themes, and macro regimes. Every substantive claim earns an inline footnote with URL and access date. Output is structured for a portfolio-level investment view, not single-name diligence.

Overview

This skill answers questions like:

• "What's actually happening in the AI semis market right now?"
• "Is the lithium thesis still intact?"
• "Are US small caps a buy here?"
• "Map the European defense complex."
• "Macro setup for 2026 H2?"

Core principle: the answer is only as good as the sources. Never claim a number or a fact without citing where it came from. The reader should be able to click every footnote and verify.

This skill is explicitly top-down. For single-name diligence use stock-eval, fundamental-analysis, dcf-valuation, or chain them via research-bundle. This skill sits a level above — it sizes the pond before you pick the fish.

When to use

Use when the user asks about:

• A market, sector, sub-sector, or thematic basket (semis, defense, lithium, GLP-1, regen ag, datacenters, EM ex-China…)
• A macro regime or rotation question (rates path, USD, oil, recession odds, breadth)
• A flow or positioning question (CTA exposure, retail vs institutional, short interest at sector level)
• A narrative check ("is the AI capex cycle real", "is the consumer rolling over")

Do NOT use when:

• The question is about a single ticker → use stock-eval or research-bundle
• The question is about a portfolio the user holds → use portfolio-review
• The question is about valuing a specific company → use dcf-valuation

Source-citation contract

Every numeric claim, every named source, every dated event MUST have an inline footnote.

Format: [^n] immediately after the claim, with footnotes collected at the end:

US Q1 2026 GDP came in at +2.1% annualized, below consensus of +2.4%[^1].

[^1]: BEA, "Gross Domestic Product, First Quarter 2026 (Advance Estimate)",
      bea.gov/news/2026/gross-domestic-product-first-quarter-2026, accessed 2026-05-15.

Source priorities (in order of preference):

1. Primary: filings (10-K/10-Q/8-K), central-bank releases (FOMC, ECB, BoE statements), official statistics (BEA, BLS, Eurostat, ONS, BIS), exchange data (NYSE, Nasdaq, LSE, CME)
2. Industry primary: trade-body data (SEMI, IEA, USGS, World Steel), company investor decks, conference transcripts
3. Reputable financial press: FT, WSJ, Bloomberg, Reuters, The Economist — use when primary not yet available, never as sole citation for a contested claim
4. Last resort: analyst notes, sell-side research summaries — must disclose

NEVER cite: forum posts, anonymous Substacks, X/Twitter threads (unless quoting a primary source's verified account), AI-generated summaries.

Workflow

Step 1 — Define the scope precisely

Before researching, restate the question in one sentence and confirm with the user. Vague briefs produce sprawling reports. E.g.:

• ❌ "AI market analysis"
• ✅ "Top-down view on US-listed AI semiconductor pure-plays (NVDA, AMD, AVGO, MRVL, ALAB, CRDO) for a 12-month horizon"

Step 2 — Map the analytical frame

Choose 3–5 of these lenses based on scope:

Lens	What it answers	Key data to pull
Market sizing	How big, growing how fast	TAM/SAM from primary sources, CAGR, unit vs $
Supply/demand	What drives the cycle	Capacity, inventory, utilization, lead times
Pricing power	Margins durable?	ASP trends, input costs, competitive intensity
Capital flows	Where is money going	Capex announcements, M&A, VC funding
Positioning	Who owns it already	Short interest, ETF flows, 13F changes
Macro overlay	What kills the thesis	Rates sensitivity, FX exposure, regulatory risk
Narrative state	What is consensus	Sell-side ratings distribution, sentiment indices
Catalysts	What moves it next 6m	Earnings, policy events, product cycles

Step 3 — Source-gather in parallel

Use firecrawl:firecrawl-search (NOT WebSearch) — it returns full-page markdown, which is necessary for accurate quotation. Search in parallel, not sequentially:

- Central-bank statements + dot plots
- Latest sector ETF flows (XLK, SMH, ITA, LIT, etc.)
- Latest 10-Q/10-K snippets for top 3-5 names
- Trade-body data releases
- 2 reputable press syntheses for cross-check

Pull primary sources first. Only consult press as cross-check or for very recent (last 7 days) events not yet in primary data.

Step 4 — Write the structured report

Standard structure:

# <Market/Sector/Theme> — Top-Down View

**Scope:** <one-sentence restatement>
**Horizon:** <e.g. 12 months>
**Bottom line:** <2-3 sentences with the actual view>

## 1. State of play
<What's true today, with footnotes>

## 2. The bull case
<3-5 bullet points, each footnoted>

## 3. The bear case
<3-5 bullet points, each footnoted>

## 4. Positioning & flows
<Who owns it, how crowded, ETF / 13F / short-interest data>

## 5. Catalysts (next 6m)
<Dated upcoming events: earnings, policy meetings, product cycles>

## 6. What would change my mind
<Concrete tripwires — number/event-level, not vibes>

---

## Sources
[^1]: ...
[^2]: ...

Step 5 — Self-check before delivering

Before returning to the user:

• Every numeric claim has a footnote
• Every footnote URL is real (not hallucinated) — verify by running firecrawl-scrape on at least 2 sampled URLs
• Access dates are present on all footnotes
• No claim relies solely on press for a number that has a primary source
• "Bottom line" is an actual view, not a hedge
• "What would change my mind" is concrete (numbers/events), not vibes

Common mistakes

Mistake	Why bad	Fix
Using WebSearch instead of firecrawl-search	Snippets aren't enough to quote accurately, may misattribute	Always use firecrawl:firecrawl-search for source-bound work
Citing AI-generated summaries	Hallucination risk, not verifiable	Cite the primary the summary was built from
Bullet lists without footnotes	Reader can't verify, low trust	Every substantive bullet earns a [^n]
"Consensus expects X" without source	Whose consensus?	Cite the specific FactSet/Bloomberg/IBES table or skip
Hedged bottom line ("could go either way")	Useless	Pick a side or explicitly say "no view, here's why"
Stale data without flagging	Misleading	If using >90-day-old data, note it in the footnote

Cross-references

• For single-name follow-up after market view: chain to stock-eval or research-bundle
• For valuation work on names that emerge: dcf-valuation, stock-valuation
• For positioning detail: short-interest, institutional-ownership, insider-trading
• For macro context: economics-analysis
• For peer comparison: competitor-analysis
• For the actual web research: firecrawl:firecrawl-search (always), firecrawl:firecrawl-scrape (verification)