--- run_id: 20 date: "2026-03-11 12:19:05 UTC" question: "112 strategy for spy or spx or qqq, find the best setup and let me know the return." total_hypotheses: 45 confirmed: 0 rejected: 0 pending: 45 rounds: 5 --- # 研究迴圈報告 — Run #20 > 研究主題：112 strategy for spy or spx or qqq, find the best setup and let me know the return. > 產生時間：2026-03-11 12:19:05 UTC ## 執行摘要 本次研究迴圈共進行 **5** 輪，產生 **45** 個假說。 最高信心假說（confidence=1.0000）： > 0-DTE ATM_straddle strangle/straddle selling ## 各輪摘要 | 輪次 | Agent 數 | 假說數 | 持續時間 | Token 用量 | |------|----------|--------|----------|------------| | R1 | 3 | 15 | — | — | | R2 | 3 | 13 | — | — | | R3 | 3 | 5 | — | — | | R4 | 3 | 7 | — | — | | R5 | 3 | 5 | — | — | ## 輪次評分（Keep/Discard Gate） | 輪次 | 綜合分數 | Sharpe | Drawdown | 樣本量 | 決策 | 改善% | |------|----------|--------|----------|--------|------|-------| | R1 | 0.2355 | 0.0000 | 0.5000 | 0.0000 | KEEP | — | | R2 | 0.2173 | 0.0000 | 0.5000 | 0.0000 | KEEP | -7.73% | | R3 | 0.1437 | 0.0000 | 0.5000 | 0.0000 | DISCARD | -38.98% | | R4 | 0.1906 | 0.0000 | 0.5000 | 0.0000 | DISCARD | -19.07% | | R5 | 0.2355 | 0.0000 | 0.5000 | 0.0000 | KEEP | 0.00% | ## 假說清單 ### Round 1 - ⏳ **[data_quality_straddle]** (confidence=0.8500) ATM straddles are FEASIBLE for 0-7 DTE options at 9:31 open — 100% of ATM put-call pairs (where both have price > 0) also have BOTH legs with volume > 0 - ⏳ **[data_quality_calendar]** (confidence=0.7000) Calendar spreads are FEASIBLE — 659+ instances of same-strike, same-timestamp, 2+ liquid expirations found in just one month (Jan 2023) - ⏳ **[data_quality_overall]** (confidence=0.9500) 88.4% of ALL option bars at 9:31 near ATM have ZERO volume — the same data quality issue from prior runs persists across ALL strategy types - ⏳ **[data_quality_0dte]** (confidence=0.9000) 0DTE options have the BEST liquidity of any DTE bucket — 56-58% of ATM-range bars have volume > 0 vs ~12-16% for longer-dated - ⏳ **[data_quality_covered_call]** (confidence=0.7500) Covered call (OTM call selling) has MARGINAL feasibility — only 12-16% of OTM call bars have volume at 9:31 - ⏳ **[execution_risk]** (confidence=0.9000) VWAP-close spread is a CRITICAL slippage risk — p95 divergence is 15%, median 0.55% - ⏳ **[data_quality_tombstone]** (confidence=0.9900) 16:00 tombstone bars are 96.5% zero-volume, 98.8% flat (H=L=C) — must be excluded from all analysis - ⏳ **[data_quality_intraday]** (confidence=0.8500) Intraday volume follows U-shape — 9:31 (14.1%) and 15:59 (13.0%) are most liquid, midday (6.6-8.5%) least liquid - ⏳ **[strategy_feasibility_negative]** (confidence=0.9500) STRATEGIES THAT WILL DEFINITELY FAIL: (1) Any strategy using underlying price from 'underlying' column — it contains 'SPX' string, not a price. (2) Gamma scalping — requires continuous delta hedging with sub-minute precision, impossible with 1-min OHLCV. (3) Iron condor with 30+ DTE — liquidity drop... - ⏳ **[strategy_recommendation]** (confidence=0.8000) RECOMMENDED STRATEGY FAMILIES ranked by data feasibility: #1 0DTE ATM straddle/strangle (best liquidity), #2 1-3 DTE vertical spreads near ATM, #3 Calendar spreads 0-7 DTE, #4 Short-dated covered calls 0-3 DTE - ⏳ **[None]** (confidence=0.4500) Buying ATM straddles 1-2 days before known high-volatility events (FOMC, CPI, NFP) on 2-3 DTE options captures realized vol exceeding implied vol. Prior runs focused on SELLING premium; this tests BUYING premium when the vol risk premium inverts. - ⏳ **[None]** (confidence=0.3500) Buy 5-delta OTM options (calls for upside, puts for downside) when SPX breaks above/below the first 30-minute range, holding until 15:30 close. 0DTE options provide extreme leverage for intraday trend-following. - ⏳ **[None]** (confidence=0.4000) When short-term IV (0-2 DTE) exceeds long-term IV (14-21 DTE) by >2 vol points (inverted term structure), sell the short-dated ATM straddle and buy the longer-dated ATM straddle. Profit from term structure normalization as short IV mean-reverts faster. - ⏳ **[None]** (confidence=0.5000) Sell 5% OTM puts with 14 DTE systematically, but ONLY when VIX-implied 14-day move > 2× realized 14-day move (i.e., when VRP is wide). Skip entries when VRP is narrow. This filters for the highest-EV subset of put-selling. - ⏳ **[None]** (confidence=0.4000) After days when implied volatility spikes >20% (proxy: ATM straddle price jumps >20% day-over-day), sell ATM strangles with 7 DTE. Vol spikes are overwhelmingly mean-reverting — realized vol over the next 7 days is typically much lower than the spike-day IV implies. ### Round 2 - ⏳ **[None]** (confidence=0.0000) 0DTE ATM strangle selling (with stop loss): sell 1% OTM put + 1% OTM call at 9:35, exit at 15:45 - ⏳ **[None]** (confidence=0.0000) 0DTE ATM strangle selling (no stop loss): sell 1% OTM put + 1% OTM call at 9:35, exit at 15:45 - ⏳ **[data_quality]** (confidence=0.8500) Sharpe 5.59 is inflated due to zero-volume VWAP prices being treated as tradeable - ⏳ **[execution_feasibility]** (confidence=0.8000) Average P&L of $2.17 per trade does not survive realistic slippage and transaction costs - ⏳ **[theoretical_plausibility]** (confidence=0.9000) Intraday straddle sell captures minimal theta relative to gamma risk, making Sharpe 5.59 implausible for real trading - ⏳ **[methodology]** (confidence=0.9500) No look-ahead bias detected in the script logic - ⏳ **[variant_comparison]** (confidence=0.7500) Hold-to-expiry variant (Sharpe 1.66) is more realistic but still needs volume filtering - ⏳ **[regime_sensitivity]** (confidence=0.7000) Strategy performance likely deteriorates significantly in high-volatility regimes - ⏳ **[dte_optimization]** (confidence=0.7500) Selling 0DTE ATM straddles at 9:31 and closing at 15:00 captures intraday theta decay more efficiently than 7-DTE straddles, because 0DTE options have the highest theta decay rate (gamma/theta ratio peaks on expiration day) AND the best liquidity in our dataset (56-58% of ATM bars have volume vs ~12... - ⏳ **[strike_selection]** (confidence=0.5500) Selling 7-DTE OTM strangles (call at ATM+25pts, put at ATM-25pts) instead of ATM straddles produces higher win rate (80%+) and lower max drawdown at the cost of lower total P&L, because the breakeven zone is wider. The Sharpe ratio may be comparable or better than straddle if the reduced frequency o... - ⏳ **[exit_timing]** (confidence=0.6000) The optimal intraday exit time for 7-DTE straddle selling is NOT 15:00 but closer to 12:00-13:00, because theta decay is roughly linear intraday while gamma risk (adverse underlying moves) accumulates over time. Exiting earlier reduces exposure to afternoon volatility (which includes power hour 15:0... - ⏳ **[data_integrity]** (confidence=0.8000) The Round 1 Sharpe of 5.59 is inflated because the backtest used VWAP prices from bars that may include zero-volume bars (the query filters vwap > 0 but does NOT filter volume > 0). Adding a strict volume > 0 filter on both legs will reduce the trade count and likely reduce Sharpe to 2.0-3.5, but th... - ⏳ **[risk_management]** (confidence=0.6500) Adding a stop-loss at 1.5x-2.0x initial premium collected eliminates the fat left tail of the P&L distribution, improving the Sharpe ratio for the hold-to-expiry variant (currently Sharpe 1.66, max DD $1,323) while sacrificing minimal total P&L. The intraday variant (Sharpe 5.59) may not benefit fro... ### Round 3 - ⏳ **[entry_timing]** (confidence=0.7000) Delaying 0DTE strangle entry from 9:35 to 10:00 or 10:30 improves risk-adjusted returns because: (1) opening auction volatility creates adverse fills — bid-ask spreads are widest in the first 5-10 minutes, (2) volume ramps up after 9:45 providing better liquidity and more reliable VWAP prices, (3) t... - ⏳ **[entry_timing]** (confidence=0.5500) 0DTE strangle selling has a significant day-of-week effect: Monday entries produce higher Sharpe than Friday entries because (1) weekend theta decay is priced into Monday open but often over-estimated by the market, (2) Friday options face event risk from weekend news creating a 'reverse weekend eff... - ⏳ **[entry_timing]** (confidence=0.5000) Entering 0DTE strangles on the day AFTER major volatility events (FOMC announcement, CPI release, NFP) produces higher risk-adjusted returns than typical days because: IV remains elevated from the prior day's shock but realized vol has already peaked and is declining (vol crush). This creates a temp... - ⏳ **[entry_timing]** (confidence=0.7500) Restricting 0DTE strangle entries to days when BOTH legs have volume >= 100 contracts at entry time materially improves Sharpe ratio because: (1) higher volume implies tighter effective spreads, reducing implicit slippage, (2) volume > 100 acts as a proxy for institutional participation — these days... - ⏳ **[entry_timing]** (confidence=0.6500) Selling 0DTE strangles ONLY when trailing 5-day realized volatility is below implied volatility (VRP > 0) improves Sharpe by 30-50% because: the volatility risk premium is time-varying and short premium strategies should only be deployed when VRP is positive (sellers are overcompensated). When reali... ### Round 4 - ⏳ **[None]** (confidence=1.0000) 0-DTE ATM_straddle strangle/straddle selling - ⏳ **[None]** (confidence=1.0000) 0-DTE OTM_0.5pct strangle/straddle selling - ⏳ **[None]** (confidence=1.0000) 0-DTE OTM_1pct strangle/straddle selling - ⏳ **[None]** (confidence=1.0000) 0-DTE OTM_2pct strangle/straddle selling - ⏳ **[None]** (confidence=1.0000) 0-DTE Asym_P2_C05 strangle/straddle selling - ⏳ **[None]** (confidence=1.0000) 0-DTE IC_1pct_50pt strangle/straddle selling - ⏳ **[None]** (confidence=1.0000) 0-DTE VolBased strangle/straddle selling ### Round 5 - ⏳ **[None]** (confidence=0.0000) Closing the 0.5% OTM 0-DTE short strangle when unrealized profit reaches 50%, 75%, or 90% of initial credit received will improve risk-adjusted returns by reducing tail-risk exposure in the final hours when gamma is highest. - ⏳ **[None]** (confidence=0.0000) Exiting the 0.5% OTM strangle at a fixed time before market close — particularly around 13:00-14:00 ET — will produce a superior Sharpe ratio by eliminating the high-gamma final hours while capturing the bulk of morning theta decay. - ⏳ **[None]** (confidence=0.0000) Implementing a trailing stop that exits when unrealized P&L drops by 30% / 50% / 75% from its intraday peak will protect gains from sudden reversals while allowing winning trades to run, producing better risk-adjusted returns than a fixed profit target. - ⏳ **[None]** (confidence=0.0000) When the underlying breaches one strike of the strangle (making that leg deep ITM), closing ONLY the losing leg while holding the winning leg to expiry will reduce max drawdown significantly compared to holding both legs, because the winning leg's remaining theta offsets some of the loss. - ⏳ **[None]** (confidence=0.0000) Exiting the entire strangle when the SPX moves more than 0.3% / 0.5% / 0.75% from entry level will act as a hard stop-loss that caps catastrophic losses on trending days, at the cost of some whipsaw exits on mean-reverting days. ## 負面結果（已排除的方向） - **R3** [expiration_choice] Delaying 0DTE strangle entry from 9:35 to 10:00 or 10:30 improves risk-adjusted returns because: (1) opening auction volatility creates adverse fills ... 失敗原因：Composite score 0.1437 below baseline 分數：0.1437，交易數：0 - **R4** [position_sizing] 0-DTE ATM_straddle strangle/straddle selling; 0-DTE OTM_0.5pct strangle/straddle selling; 0-DTE OTM_1pct strangle/straddle selling 失敗原因：Composite score 0.1906 below baseline 分數：0.1906，交易數：0 ## 方法評分 | Agent | 方法 | 成功率 | 提出數 | 確認數 | 樣本數 | |-------|------|--------|--------|--------|--------| | risk-auditor | data_quality_straddle | 0.00% | 1 | 0 | 1 | | researcher | exit_timing | 0.00% | 1 | 0 | 1 | | risk-auditor | data_quality_overall | 0.00% | 1 | 0 | 1 | | risk-auditor | data_quality_0dte | 0.00% | 1 | 0 | 1 | | risk-auditor | data_quality_covered_call | 0.00% | 1 | 0 | 1 | | risk-auditor | execution_risk | 0.00% | 1 | 0 | 1 | | risk-auditor | data_quality_tombstone | 0.00% | 1 | 0 | 1 | | risk-auditor | data_quality_calendar | 0.00% | 1 | 0 | 1 | | risk-auditor | strategy_feasibility_negative | 0.00% | 1 | 0 | 1 | | risk-auditor | strategy_recommendation | 0.00% | 1 | 0 | 1 | | researcher | general | 0.00% | 5 | 0 | 5 | | optimizer | general | 0.00% | 2 | 0 | 2 | | risk-auditor | data_quality | 0.00% | 1 | 0 | 1 | | risk-auditor | execution_feasibility | 0.00% | 1 | 0 | 1 | | risk-auditor | theoretical_plausibility | 0.00% | 1 | 0 | 1 | | risk-auditor | methodology | 0.00% | 1 | 0 | 1 | | risk-auditor | variant_comparison | 0.00% | 1 | 0 | 1 | | risk-auditor | regime_sensitivity | 0.00% | 1 | 0 | 1 | | researcher | dte_optimization | 0.00% | 1 | 0 | 1 | | researcher | strike_selection | 0.00% | 1 | 0 | 1 | | researcher | general | 0.00% | 5 | 0 | 5 | | researcher | data_integrity | 0.00% | 1 | 0 | 1 | | researcher | risk_management | 0.00% | 1 | 0 | 1 | | researcher | entry_timing | 0.00% | 5 | 0 | 5 | | researcher | general | 0.00% | 7 | 0 | 7 | | risk-auditor | data_quality_intraday | 0.00% | 1 | 0 | 1 | ## 建議與後續行動 - 45 個假說仍待驗證，可增加輪次或手動檢視。