wiki-query
PApaperclipai/paperclip101k近 7 天新增 2,948 个 star
quant-lab
- Star
- 2,173
- 分类
- 科研
- 许可证
- MIT
- 最近提交
- 2026年10月12日
resources/builtin/marketplace/agents/1040b336306f/skills/quant-lab/SKILL.md
在 GitHub 上查看Quant Lab
Read with market-data for strategy design, backtesting, or experiment review.
Experiment card
Before calculation, state universe, data/adjustment, signal, rebalance timing, execution price assumption, sizing, cash treatment, costs, benchmark, sample, and invalidation rule. A close-derived signal may affect only the next return.
The compact core supports sma_cross, momentum, and rsi_reversion as
transparent baselines:
"$ORKAS_NODE" "$ORKAS_PC_DIR/bin/run-skill.cjs" market-data stock -- backtest --input history.json --strategy sma_cross --fast 20 --slow 60 --cost-bps 8 --slippage-bps 5 --include-series --output result.json
Use the baselines to test a claim, not to stretch the engine into an unsupported strategy. For a custom strategy, produce a precise experiment specification and use a user-authorized project framework when available.
Review gate
- Confirm point-in-time membership and fundamentals; disclose survivorship or restatement exposure when they cannot be reconstructed.
- Include transaction costs and a relevant passive benchmark. State whether dividends, borrow, market impact, taxes, and FX are absent.
- Prefer chronological train/validation/test or walk-forward evidence. Keep the final test untouched while tuning.
- Inspect sensitivity to neighboring parameters, later entry, higher cost, and subperiods. A fragile result is a finding, not something to optimize away.
- Report CAGR/return, volatility, Sharpe convention, max drawdown, turnover, exposure, trade count, and the exact signal-delay guard.
Never describe a backtest as expected future performance. Preserve the core's
result JSON; put supplementary interpretation in a separate dossier or file.
--include-series supplies signal date, applied position, turnover and gross,
cost and net returns from the same calculation as the summary. Use that ledger
for trade explanations rather than rebuilding it with another timing model.
--cutoff YYYY-MM-DD selects a requested historical price window in memory;
the result records exclusions without changing the input.
相似的 Skill
官方
clinical-case-report
NEnexu-io/open-design101k近 7 天新增 1,126 个 star
systematic-literature-review
BYbytedance/deer-flow84k近 7 天新增 368 个 star
academic-paper-review
BYbytedance/deer-flow84k近 7 天新增 368 个 star
directory-submissions
COcoreyhaines31/marketingskills54k近 7 天新增 958 个 star
academic-paper
IMImbad0202/academic-research-skills52k近 7 天新增 972 个 star