ntaffzii Market Breadth Analyzer
Overview
A cap-weighted index (S&P 500, Nasdaq) can rise while most individual stocks fall — a handful of large names carry the average. Breadth measures actual participation: what fraction of a universe is trading above a given moving average. This skill computes that from a CSV the user already has (or pastes), so it works with no paid market-data API.
Not financial advice. This is a descriptive statistic over data you supply. A single breadth reading in isolation is a weak signal — see Known limitations.
When to use
- เช็คว่าตลาดกำลังขึ้นแบบกว้าง (broad) หรือขึ้นแค่ไม่กี่ตัว (narrow)
- Scoring a watchlist or index-constituent CSV for % above 50-day/200-day moving average
- Part of a daily/weekly market-regime check (pairs naturally with uptrend-analyzer and exposure-coach)
When NOT to use
- The user wants an actual buy/sell signal from one breadth reading — breadth informs market posture, it is not a trade trigger on its own
- No price/MA data is available at all and the user isn't willing to supply even a rough CSV — this skill has nothing to compute on
Core knowledge
Definition: breadth % = (number of tickers with price > moving average) / (total tickers with valid data) × 100. Most common MA choices: 50-day (shorter-term participation) and 200-day (longer-term trend participation).
Rough interpretation bands (starting heuristic, not calibrated — tune against your own universe's history):
| Breadth % | Reading |
|---|---|
| ≥ 70% | Broad participation |
| 50-69% | Moderate participation |
| 30-49% | Narrow participation |
| < 30% | Weak participation |
Data-quality handling: rows with missing or non-numeric price/MA values are excluded and reported, not silently treated as "below the MA" — folding bad data into the score in either direction would quietly bias the result.
Common mistakes
- Treating a single breadth reading as a standalone buy/sell signal instead of one input to overall market posture.
- Silently dropping rows with bad data instead of excluding-and-reporting them — this can shift the percentage without anyone noticing.
- Comparing breadth readings computed against different MAs (50-day vs. 200-day) as if they measure the same thing — always state which MA a given percentage refers to.
- Assuming
price == macounts as "above" — this implementation uses strict>, so an exact match counts as not-above; be explicit about that boundary when reporting results.
Code
breadth.py — no dependencies:
compute_breadth(rows, price_field="price", ma_field="ma200")→BreadthResult(total, above, below, excluded, pct_above)compute_breadth_from_csv(csv_text, ...)→ same, from raw CSV text (header row required)classify_breadth(pct_above)→ one of the four bands above
Run python3 breadth.py for the self-test.
Known limitations
- Interpretation bands are a generic starting heuristic — different universes (small-cap vs. large-cap, different sectors) behave differently at the same percentage; recalibrate against history before treating a threshold as meaningful.
- This skill does not fetch data. The user (or another tool/skill) must supply the CSV; accuracy of the output is only as good as the input.
- A rising breadth percentage and a falling index (or vice versa) is itself informative (a breadth divergence) — this skill reports the raw number, not divergence detection; that's a manual read on top of the output.