Stock Screener

Stock Screener

Prompt-first skill. Filter and rank by the rules below. scripts/screen.py is OPTIONAL (handles large universes deterministically). Output is a multi-ticker list, not a single card.

Role & objective

You screen a supplied universe of DSE stocks against filters (explicit, template, or parsed from natural language), then rank the survivors by a quality+momentum score.

When to use

"Show me cheap stocks", "value stocks under P/E 15", "dividend champions", "momentum names breaking 52-week highs", "profitable banks". Hand the top tickers to analyze_ticker.

Inputs you need

universe — array of {ticker, fundamentals{pe,pb,roe,debt_to_equity,dividend_yield, market_cap}, ohlcv[], sector}. Optional: filters, template, query, limit (25).

Method (follow in order)

1. Compose filters: template seeds → NL query adds → explicit filters override.
   • Templates: value (pe≤15, pb≤1.5, roe≥0.12), dividend_champions (div≥4%, roe≥0.12), momentum_leaders (rsi≥55, breakout, pos_52w≥0.7), oversold_quality (rsi≤35, roe≥0.15, de≤0.6), small_cap_growth (roe≥0.15).
   • NL parse: "cheap/value"→value; "dividend/income"→dividend; "momentum/breakout"→momentum; "oversold"→oversold; sector keywords; "p/e under N", "roe over N%", "52-week high", "volume surge"→rel_volume≥1.5.
2. Evaluate each stock against every active filter (fundamental + technical derived from ohlcv: RSI, MA50/200 cross, 52-week position, breakout vs 40-bar high, rel-volume). A stock passes only if it satisfies all active filters.

Scoring rubric (rank score)

rank = 0.5·(min(ROE,0.4)/0.4) + 0.3·pos_52w + 0.2·clamp((RSI−30)/50, 0, 1). Sort survivors descending; return the top limit.

Output

{ "skill": "stock-screener", "as_of": "..",
  "applied": { "template": null, "query": "..", "filters": {} },
  "count": 0, "results": [ { "ticker": "..", "passes": true, "score": 0.0,
    "matched_filters": {}, "key_metrics": { "pe": 0, "roe": 0, "rsi": 0, "pos_52w": 0 } } ],
  "reasoning": ["filters applied; N of M passed; top K returned"],
  "disclaimer": "Educational analysis only. Not financial advice." }

DSE pitfalls

• Stocks with missing fields fail a filter that needs them — don't pass unknowns; say which filter excluded them if asked.
• Thin ohlcv means technical filters can't evaluate — those stocks drop out of technical screens.
• US-style P/E≤15 is strict for some DSE sectors — adjust thresholds via explicit filters when relevant.

Optional precision helper

python3 scripts/screen.py --input universe.json --pretty

Returns the passing tickers ranked, with matched filters and key metrics.

Worked example

query "profitable banks with p/e under 15" → {sector:Bank, roe_min:0.10, pe_max:15}; 4 of 30 pass; ranked by ROE + 52-week position.

References

Templates & NL grammar: references/SCREENS.md. Output is educational analysis only, never financial advice.