SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsa-share-strategy-allmarket-multi-swing-defensive

a-share-strategy-allmarket-multi-swing-defensive

Agents & models

View repository

Updated 10-16 08:38

About this skill

Whole-market multi-swing defensive · Decision signals

This skill core does three things:Stock selection scope、Buy-side signals、Sell-side signals; and optionally uses realtime_quotes.py to batch-fetch for specified codescurrent-price-type snapshots(based on MarketDataProvider.get_quote, Tencent quotes + minute K aggregation). What is output isdecision reference(structured list or JSON),does not includehistorical backtest matching,does not includeautomatic order placement; for real orders, please use a broker or separately connect a-share-paper-trading and other execution channels. The strategy signal is positioned asconditionally valid: under changes in market structure and trading costs, signal strength may decay and requires continuous review.

Capability boundaries

Do Do not do
Take the top N from whole-market high-liquidity stocks (including ChiNext and STAR Market) to form the day's stock pool Minute-level backtesting, mixed backtesting, return curves
Use daily line trend_pullback to mark entry/exit conditions Guarantee returns or replace investment advisory
Provide two types of "buy reference" lists (see below) Place orders directly to exchanges or paper trading services
Optional: read a position list file and mark "strategy exit" symbols Save live positions for you (unless you create your own file)
Batch-fetch current prices, change percentages, limit-up/limit-down reference prices, etc. for multiple stocks (realtime_quotes.py） Replace a-share-data 's heavy interfaces such as whole-market real-time sectors and capital flows

Strategy logic (and parameters)

Default parameters are in scripts/strategy_lab/strategy_params.py(fast/slow moving averages, pullback magnitude, RSI range and exit RSI, etc.). Signal calculation is in scripts/strategy_lab/strategies.py 's trend_pullback. The script additionally applies two layers of filtering:

  • Cost filter: estimate round-trip cost according to --roundtrip-cost-bps , and only edge_after_cost > 0 candidates enter the final buy list.
  • Robustness filter: calculate within the parameter neighborhood of ROBUSTNESS_PARAM_GRID computed within the parameter neighborhood entry_consensus_ratio, and candidates below --entry-consensus-min will be filtered out.

Buy reference (two groups, please distinguish the semantics):

  • from_previous_day_close: the previous closed daily line satisfies entry. Consistent with the habit of "signal after the previous day's close, execute on the current day",closer to an ex-ante plan。
  • from_last_close: the latest daily line also satisfies entry, leaning towardpattern display; if it repeats with the previous day, please avoid double counting.

Within the list, sort by strategy-internal score(moving-average strength) in descending order; by defaulteach buy list retains at most 5 stocks（strategy_params.MAX_BUY_CANDIDATES), consistent with "positions to watch simultaneously should not be too many". When you need more or all, add --max-buys 0 to indicate no truncation, or --max-buys 10 , etc.

Sell reference: For --holdings codes in the file, ifthe latest daily barsatisfies exit(conditions within the rules such as breaking the slow line or RSI being too high), then it is listed as a sell reference. File format: one code per line, comment lines allowed (# at the start).

Risk control reference (documentation and JSON fields only): REFERENCE_INTRADAY_STOP_PCT represents the historicalintraday stop-loss percentage referenceconsistent with the strategy document; this scriptdoes notmonitor intraday stop-loss for you; you need to set it yourself in your order placement software.

Environment and dependencies

pip install akshare pandas numpy requests

Run

SKILL_DIR="<本 skill 绝对路径>"
python3 "$SKILL_DIR/scripts/daily_decisions.py" --json

Common parameters:

  • --top-n: stock pool size, default 120
  • --max-buys: maximum number of entries to keep on the buy side after sorting, default 5; pass 0 for no truncation
  • --holdings: path to the holdings code file
  • --workers: number of concurrent daily bar fetches
  • --json: output a JSON file for programmatic consumption (the JSON contains the pre-truncation count *_total）
  • --roundtrip-cost-bps: round-trip transaction cost estimate (in bps), default 45
  • --entry-consensus-min: parameter neighborhood entry consistency threshold, default 0.67
  • --disable-robust-check: disable parameter neighborhood consistency filtering (use only when debugging)

Example:

python3 "$SKILL_DIR/scripts/daily_decisions.py" --top-n 120 --holdings "$HOME/my_holdings.txt"

The JSON output will contain both the raw candidates and the filtered candidates:

  • from_previous_day_close_raw / from_last_close_raw: raw candidates that satisfy the signal only
  • from_previous_day_close / from_last_close: candidates after cost and robustness filtering
  • Each candidate is accompanied by entry_consensus_ratio、edge_after_cost、cost_filter_passed、consensus_filter_passed
  • todo_confirm_items: record of the currently adopted conventions (roundtrip=45bps, consensus=0.67, RSI interval offset)

Real-time quote snapshot (enhanced)

Fora known list of codes, batch-fetch quotes (concurrently), which can be used together with the codes output by daily_decisions :

python3 "$SKILL_DIR/scripts/realtime_quotes.py" 600519 000001 601318 --json
python3 "$SKILL_DIR/scripts/realtime_quotes.py" -f "$HOME/my_holdings.txt" --workers 10
  • --json: output unified JSON (including quotes and per-item details）
  • --intraday: additionally fetchthe lastminute K (--intraday-freq default 5m), to facilitate checking time alignment with the daily snapshot
  • Data comes from paper_trading/market_data.py: compared with full-market depth quotes, this is alightweight snapshot, and intraday prices update with the latest minute K

Script layout

Path Purpose
scripts/daily_decisions.py Entry point: fetch pool, compute signals, print or --json
scripts/realtime_quotes.py batch current price snapshots, optionally with the last minute K attached
scripts/paper_trading/market_data.py Quotes and get_all_market_universe
scripts/strategy_lab/strategies.py trend_pullback
scripts/strategy_lab/indicators.py moving averages, RSI
scripts/strategy_lab/strategy_params.py Default parameters and strategy name

Integration with the execution layer

If you want to put signals into a paper trading account, you can combine and use in the Agent a-share-paper-trading: first read this skill's output, then call the paper trading CLI or HTTP API to place orders; this skill does notdepend on the paper trading process.