--- run_id: 15 date: "2026-03-11 10:01:44 UTC" question: "spy wheel strategy" total_hypotheses: 62 confirmed: 0 rejected: 0 pending: 62 rounds: 5 --- # 研究迴圈報告 — Run #15 > 研究主題：spy wheel strategy > 產生時間：2026-03-11 10:01:44 UTC ## 執行摘要 本次研究迴圈共進行 **5** 輪，產生 **62** 個假說。 最高信心假說（confidence=0.9700）： > 閃崩/黑天鵝事件中，5% OTM CSP的權利金（$1.50-$3.00/股）完全無法抵消$25-$50+的股價暴跌，風險收益比為1:15至1:30 ## 各輪摘要 | 輪次 | Agent 數 | 假說數 | 持續時間 | Token 用量 | |------|----------|--------|----------|------------| | R1 | 3 | 12 | — | — | | R2 | 3 | 13 | — | — | | R3 | 3 | 11 | — | — | | R4 | 2 | 13 | — | — | | R5 | 2 | 13 | — | — | ## 假說清單 ### Round 1 - ⏳ **[spy wheel strategy]** (confidence=0.7000) Selling 3% OTM CSPs weekly (30-45 DTE) on SPY leads to ~20% assignment rate with avg 4.8% drawdown at assignment - ⏳ **[spy wheel strategy]** (confidence=0.7000) Selling 5% OTM CSPs weekly (30-45 DTE) on SPY leads to ~13% assignment rate with avg 4.7% drawdown at assignment - ⏳ **[spy wheel strategy]** (confidence=0.7000) Selling 7% OTM CSPs weekly (30-45 DTE) on SPY leads to ~6% assignment rate with avg 6.7% drawdown at assignment - ⏳ **[spy wheel strategy]** (confidence=0.7000) Selling 10% OTM CSPs weekly (30-45 DTE) on SPY leads to ~3% assignment rate with avg 6.7% drawdown at assignment - ⏳ **[spy wheel strategy]** (confidence=0.7500) SPY buy-and-hold (2020-2024) returned ~12% annualized, making wheel strategy's ~10% target return underperform during bull markets - ⏳ **[spy wheel strategy]** (confidence=0.8000) Wheel strategy CSP collateral opportunity cost is material: $50K collateral at 5% T-bill = $2,500/yr guaranteed vs variable premium income with assignment risk - ⏳ **[spy wheel strategy]** (confidence=0.6500) Frequent assignment/calling in wheel creates short-term capital gains taxed at ordinary income rates (up to 37%), reducing after-tax returns by 15-20% vs long-term B&H - ⏳ **[spy wheel strategy]** (confidence=0.7200) Weekly CSP (7-DTE) at 3% OTM generates higher annualized premium yield than 30-45 DTE monthlies, but with worse Sharpe ratio due to gamma clustering during expiration week. Optimal DTE for SPY wheel is 30-DTE (not 45-DTE) when accounting for capital efficiency. - ⏳ **[spy wheel strategy]** (confidence=0.6800) Selling CSPs at 2-3% OTM (vs Run 12's 5% OTM) doubles premium income while keeping assignment rate below 10%, improving annualized return from 9.72% to ~14-16% on the SPY wheel. - ⏳ **[spy wheel strategy]** (confidence=0.6000) When assigned on CSP, selling ATM (0% OTM) covered calls maximizes premium extraction and accelerates capital recycling, outperforming 2-5% OTM CCs by >2% annualized despite more frequent call-away events. - ⏳ **[spy wheel strategy]** (confidence=0.5500) VIX-conditional entry (sell CSPs only when VIX > 20, or 2x notional when VIX > 25) improves Sharpe ratio by >0.15 vs fixed-schedule selling, capturing the volatility risk premium more efficiently. - ⏳ **[spy wheel strategy]** (confidence=0.5000) A mechanical roll rule (roll CSP down-and-out when put goes ITM by >1% with >7 DTE remaining) reduces max drawdown by 15-25% vs hold-to-expiration, with <2% drag on annualized return. ### Round 2 - ⏳ **[strike_selection]** (confidence=0.9200) Wheel strategy (5% OTM, 30-45 DTE) has a positive expected value per cycle but massively underperforms buy-hold SPY - ⏳ **[strike_selection]** (confidence=0.8800) Assignment events are heavily clustered in bear markets, creating catastrophic tail risk that premiums cannot compensate - ⏳ **[strike_selection]** (confidence=0.8500) Kelly Criterion indicates a marginal positive edge for 5% OTM wheel, but optimal bet size is tiny (2.5% of capital) - ⏳ **[strike_selection]** (confidence=0.9500) The wheel strategy has an inherent structural disadvantage: it is short gamma (concave payoff) while SPY is a trending asset - ⏳ **[strike_selection]** (confidence=0.9000) All wheel variants (3-7% OTM, weekly to monthly) underperform SPY buy-hold; the aggressive/weekly variants add more risk for marginally more return - ⏳ **[strike_selection]** (confidence=0.8200) Buy-write (own SPY + sell covered calls) dominates the wheel strategy in all market conditions - ⏳ **[strike_selection]** (confidence=0.7800) T-bills provide better risk-adjusted returns than the wheel strategy when accounting for tail risk - ⏳ **[strike_selection]** (confidence=0.9300) Strike selection is NOT the primary lever for improving the wheel — the strategy has a fundamental structural problem regardless of strike - ⏳ **[strike_selection]** (confidence=0.7200) Bimodal OTM strategy outperforms fixed 5% OTM: sell 7-10% OTM puts (Sharpe 1.578) in normal/low-vol environments and 0-2% OTM puts (Sharpe 1.384, 33.7% ann ROC) during high-vol spikes (VIX>25), capturing the two Sharpe peaks while avoiding the 3-7% valley. - ⏳ **[strike_selection]** (confidence=0.6800) Delta-based strike selection (targeting 15-20 delta using premium-as-proxy) produces more consistent returns than fixed OTM% because it automatically adjusts strike distance with volatility, selling closer in low-vol (higher premium capture) and farther in high-vol (more protection). - ⏳ **[strike_selection]** (confidence=0.6500) ATM covered calls (0-1% OTM) at 22-30 DTE maximize recovery speed after assignment, reducing average holding period from 60+ days to 30 days, because the 0.20% premium per cycle compounds to offset the assignment gap faster than waiting for price recovery alone. - ⏳ **[strike_selection]** (confidence=0.8500) The R15 baseline wheel dramatically underestimates true returns due to a capital utilization bug. The actual monthly 5% OTM, 30-45 DTE CSP wheel returns ~20% annualized on deployed capital (cash-secured at strike), not 1.66%. R15 likely divides P&L by total account value ($100k) while deploying only... - ⏳ **[strike_selection]** (confidence=0.5500) A weekly/bi-weekly rotation at 3-5% OTM outperforms monthly at 5% OTM on absolute return (but not risk-adjusted) because weekly cycles capture higher annualized theta decay despite lower per-trade premium. The 508 weekly expirations show 2.07% avg ROC per cycle * ~26 cycles/year = ~54% gross, but as... ### Round 3 - ⏳ **[structural_viability]** (confidence=0.9500) The Wheel strategy has a structural impossibility that cannot be solved by parameter optimization: it underperforms buy-hold in bull markets (51.7% of months), underperforms cash in bear markets (31.0%), and only wins in sideways markets (17.2%). This makes it mathematically inferior to a simple 60/... - ⏳ **[return_comparison]** (confidence=0.9200) Wheel annualized return is ~5.3% (per-contract, per-trade average), approximately 1/3 of SPY buy-hold's ~15% average annual return over the same period. Even the best-case year (2020: 96.7% win rate) still underperforms buy-hold (13.9% SPY return vs ~10% Wheel estimate). - ⏳ **[theta_decay_reality]** (confidence=0.8800) Theta decay for SPY OTM puts is highly non-monotonic and unreliable as an income source. The textbook smooth decay curve does NOT materialize in practice. Median put retains 36% of value at 15 DTE (expected ~50%), but then INCREASES to 46% at 1 DTE and 233% at 0 DTE due to gamma risk and bid-ask dyn... - ⏳ **[economic_equivalence]** (confidence=0.9000) The Wheel's premium collection is a near-exact offset of assignment losses on average, making it economically equivalent to a low-yield money market with occasional large drawdowns. Total premium kept: $140,505. Total assignment losses: $54,378. The apparent net gain of $86K is an artifact of analyz... - ⏳ **[alternative_strategies]** (confidence=0.8500) RECOMMENDATION: Pivot Rounds 4-5 away from Wheel/credit spreads entirely. The most promising next strategies are: (1) Volatility-conditioned entry: only sell premium when VIX > 25, dramatically improving win rates; (2) Buy-write/covered call on SPY: captures upside + premium, proven to outperform in... - ⏳ **[final_verdict]** (confidence=0.9300) FINAL VERDICT: Wheel Strategy scores 2/10 with 0.93 confidence. It is a structurally flawed strategy for SPY that cannot be fixed by parameter tuning. Recommend IMMEDIATE PIVOT in Rounds 4-5 to VIX-conditioned strategies or buy-write. - ⏳ **[expiration_choice]** (confidence=0.8200) CSP leg should use 45-60 DTE entries (not 7-14) because annualized yield is 4x higher (107% vs 33% ann ROC) and bid-ask friction is 20x lower (4% vs 80% of premium). The sqrt(t) premium scaling means longer DTE captures more vol premium per unit of risk. - ⏳ **[expiration_choice]** (confidence=0.7800) Covered Call leg should use 7-14 DTE (not 30-45) because premium-per-day is 5x higher ($3.68/day vs $0.73/day for 3% OTM calls) and theta decay accelerates dramatically near expiration for near-ATM options. This asymmetry (long DTE for CSP, short DTE for CC) creates an optimal hybrid Wheel. - ⏳ **[expiration_choice]** (confidence=0.8500) Short-DTE CSPs (7-14 day) have catastrophic gamma risk that wipes out months of premium in a single move. A >50% overnight premium spike occurs 14% of the time for 3-10 DTE puts vs only 5% for 46-65 DTE puts. The max spike is 3526% for short DTE vs 567% for longer DTE. - ⏳ **[expiration_choice]** (confidence=0.7500) The Wheel's fundamental problem is capital utilization inefficiency, not DTE choice. With 5% OTM CSPs, assignment rate is near-zero (0-2.6%) across ALL DTE buckets and ALL years including 2022. The Wheel only generates meaningful stock-phase returns during bear markets when assignments cluster, but ... - ⏳ **[expiration_choice]** (confidence=0.7000) Optimal hybrid DTE strategy: sell 45-60 DTE CSPs (max premium efficiency, min spread friction), manage at 21 DTE (roll or close for ~60% of max profit), and if assigned sell 7-10 DTE covered calls (max theta decay). This asymmetric DTE approach should yield 2-3x the premium of any single-DTE approac... ### Round 4 - ⏳ **[position_sizing]** (confidence=0.9500) Kelly criterion (f*=0.025) makes the Wheel strategy structurally unviable: it prescribes 0.038 contracts per $100K, meaning even the minimum 1-contract position is 26x over Kelly-optimal sizing - ⏳ **[position_sizing]** (confidence=0.9200) Risk-adjusted metrics provide the ONLY argument for the Wheel, but it is weak: Sharpe is identical to SPY (0.432 vs 0.431), and the Calmar advantage (0.891 vs 0.514) comes entirely from lower drawdown, not higher return - ⏳ **[position_sizing]** (confidence=0.9300) Account size analysis reveals the Wheel is NEVER capital-efficient: at $50K you can't run it, at $100K it's 66% concentrated, at $500K+ Kelly says use 1 contract leaving 87%+ idle - ⏳ **[position_sizing]** (confidence=0.8800) The $57K idle capital opportunity cost is devastating: a second uncorrelated strategy on idle cash could theoretically help, but finding one is the REAL research problem, not optimizing the Wheel - ⏳ **[final_assessment]** (confidence=0.9400) FINAL VERDICT — SPY Wheel Strategy: 3/10, CONDITIONAL NO-GO. The Wheel is a mathematically valid but practically useless strategy for retail accounts under $500K. It earns a thin positive edge that is entirely consumed by opportunity cost and capital inefficiency. - ⏳ **[methodology_bias]** (confidence=0.9200) R1 backtest period (2020-2026) is severely biased AGAINST the Wheel due to historically anomalous bull market: SPY returned 13.2% annualized (geometric 11.9%) with 66.7% positive months. The 2022 bear was the ONLY year where the Wheel could outperform. In any sideways/flat decade (2000-2010 style), ... - ⏳ **[return_calculation_error]** (confidence=0.8800) R1 return calculations contain a CRITICAL capital utilization error. Using $100K denominator when only ~$42K is deployed (5% OTM CSP on ~$420 SPY) inflates the denominator by 2.3x. The remaining $58K can earn T-bill rates (3-5%). Corrected Wheel return is 5.7-8.5% annualized, NOT 1.66-2.48%. - ⏳ **[missing_income_stream]** (confidence=0.8500) R1 backtest likely OMITS covered call income during stock holding periods. Our data shows weekly CC premiums average 0.68% of spot (ATM-2% OTM, 7-14 DTE), annualizing to 23.5%. Even collecting half this during assignment periods (~4 weeks × $300/week = $1,200 per assignment) adds significant return. - ⏳ **[risk_adjusted_comparison]** (confidence=0.9000) The risk-auditor's 2/10 rating uses retrospective SPY buy-hold as benchmark — this is textbook hindsight bias. No investor knew in 2020 that SPY would return 12%+ annually for 6 years. On a RISK-ADJUSTED basis (Sharpe, Sortino, Calmar), the Wheel significantly outperforms SPY due to dramatically low... - ⏳ **[theta_decay_reliability]** (confidence=0.9500) R3's claim that '25% of options double in value day before expiration' is GROSSLY overstated. Actual data shows only 3.9% of puts with meaningful premium doubled near expiration. 62.8% of puts DECREASED in value (normal theta decay). The claim likely sampled only deep OTM or low-premium options wher... - ⏳ **[regime_classification_error]** (confidence=0.8200) The '17.2% sideways months' statistic from R3 uses too narrow a definition. Using ±3% (reasonable for monthly returns), 50.7% of months are 'sideways'. The Wheel also profits in mildly bullish months (up to ~5% where CSP expires worthless). Total 'Wheel-favorable' months = sideways + mildly bullish ... - ⏳ **[premium_yield_verification]** (confidence=0.8700) The annualized put premium surface shows 8.9% average yield for 5% OTM 30-45 DTE puts — dramatically higher than R1's reported 1.66%. This confirms the capital utilization bug: R1 is dividing ~$3,700/yr premium by $100K instead of by ~$42K deployed capital. The strategy is NOT returning 1.66%; the C... - ⏳ **[assignment_reframe]** (confidence=0.8000) The 2022 assignment clustering (33% of assignments in bear market) is actually a FEATURE, not a bug. In 2022, SPY dropped 19.7%, but the Wheel investor was assigned at 5% OTM strikes (buying at a discount) and immediately sold covered calls at elevated IV. The Wheel's 2022 return likely BEAT SPY's -... ### Round 5 - ⏳ **[return_comparison]** (confidence=0.1500) SPY Wheel strategy generates competitive risk-adjusted returns compared to SPY buy-hold - ⏳ **[income_generation]** (confidence=0.7800) The Wheel is a viable fixed-income alternative with 3-5% alpha over T-bills - ⏳ **[parameter_optimization]** (confidence=0.7200) 5% OTM with 30-45 DTE is the optimal CSP configuration for the Wheel - ⏳ **[capital_efficiency]** (confidence=0.8500) Capital utilization is the primary return driver, not premium selection - ⏳ **[strategy_comparison]** (confidence=0.7000) Multi-asset momentum is the most promising alternative strategy for future research - ⏳ **[risk_analysis]** (confidence=0.8000) Assignment risk is manageable but concentrated in high-volatility regimes - ⏳ **[final_verdict]** (confidence=0.8200) The Wheel strategy should receive a final score of 4/10 with CONDITIONAL NO-GO - ⏳ **[tail_risk_stress_test]** (confidence=0.9500) COVID壓力測試：2020年2月14日賣出5% OTM CSP（行權價$321, 到期3/20）在市場崩盤中遭受災難性虧損，每股淨虧$63.53（19.8%），若持有至3月23日底部則虧損$97.50/股（30.4%） - ⏳ **[bear_market_resilience]** (confidence=0.8800) 2022年熊市期間，月度5% OTM CSP在12個月中被指派5次（41.7%），但總淨收益仍為正（+$32.01/股），顯示Wheel策略在漸進式下跌中有一定韌性 - ⏳ **[black_swan_exposure]** (confidence=0.9700) 閃崩/黑天鵝事件中，5% OTM CSP的權利金（$1.50-$3.00/股）完全無法抵消$25-$50+的股價暴跌，風險收益比為1:15至1:30 - ⏳ **[tail_risk_metrics]** (confidence=0.9200) Wheel策略的月度回報分佈呈嚴重負偏態（偏度-3.78），VaR(95%)=-2.51%，CVaR(95%)=-2.83%，最大回撤-6.35%需29個月恢復 - ⏳ **[final_risk_score]** (confidence=0.9000) SPY Wheel策略在$100K帳戶中的尾部風險「有條件可接受」——前提是實施嚴格停損規則和倉位控制，裸Wheel不可接受。最終尾部風險管理評分：5/10 - ⏳ **[risk_mitigation_rules]** (confidence=0.7500) 建議的停損規則體系：四層防禦機制可將最大虧損從30%降至約12-15%，但代價是年化收益降低約30-40% ## 錯誤日誌摘要 - **backtest_crash**: 3 次 最近錯誤： - [backtest_crash] Script wheel_dte_optimizer.py failed (rc=1): Traceback (most recent call last): File "/home/francis/project/backtest/r - [backtest_crash] Script wheel_strike_optimizer_r2.py failed (rc=1): no stderr - [backtest_crash] Script spy_wheel_r1.py completed but no output at research/results/run-015/spy_wheel_r1.json ## 方法評分 | Agent | 方法 | 成功率 | 提出數 | 確認數 | 樣本數 | |-------|------|--------|--------|--------|--------| | risk-auditor | spy wheel strategy | 0.00% | 7 | 0 | 7 | | portfolio-mgr | risk_analysis | 0.00% | 1 | 0 | 1 | | risk-auditor | strike_selection | 0.00% | 8 | 0 | 8 | | researcher | strike_selection | 0.00% | 5 | 0 | 5 | | risk-auditor | structural_viability | 0.00% | 1 | 0 | 1 | | risk-auditor | return_comparison | 0.00% | 1 | 0 | 1 | | risk-auditor | theta_decay_reality | 0.00% | 1 | 0 | 1 | | risk-auditor | economic_equivalence | 0.00% | 1 | 0 | 1 | | risk-auditor | alternative_strategies | 0.00% | 1 | 0 | 1 | | researcher | spy wheel strategy | 0.00% | 5 | 0 | 5 | | researcher | expiration_choice | 0.00% | 5 | 0 | 5 | | risk-auditor | position_sizing | 0.00% | 4 | 0 | 4 | | risk-auditor | final_assessment | 0.00% | 1 | 0 | 1 | | devil | methodology_bias | 0.00% | 1 | 0 | 1 | | devil | return_calculation_error | 0.00% | 1 | 0 | 1 | | devil | missing_income_stream | 0.00% | 1 | 0 | 1 | | devil | risk_adjusted_comparison | 0.00% | 1 | 0 | 1 | | devil | theta_decay_reliability | 0.00% | 1 | 0 | 1 | | devil | regime_classification_error | 0.00% | 1 | 0 | 1 | | devil | premium_yield_verification | 0.00% | 1 | 0 | 1 | | devil | assignment_reframe | 0.00% | 1 | 0 | 1 | | portfolio-mgr | return_comparison | 0.00% | 1 | 0 | 1 | | portfolio-mgr | income_generation | 0.00% | 1 | 0 | 1 | | portfolio-mgr | parameter_optimization | 0.00% | 1 | 0 | 1 | | portfolio-mgr | capital_efficiency | 0.00% | 1 | 0 | 1 | | portfolio-mgr | strategy_comparison | 0.00% | 1 | 0 | 1 | | risk-auditor | risk_mitigation_rules | 0.00% | 1 | 0 | 1 | | portfolio-mgr | final_verdict | 0.00% | 1 | 0 | 1 | | risk-auditor | tail_risk_stress_test | 0.00% | 1 | 0 | 1 | | risk-auditor | bear_market_resilience | 0.00% | 1 | 0 | 1 | | risk-auditor | black_swan_exposure | 0.00% | 1 | 0 | 1 | | risk-auditor | tail_risk_metrics | 0.00% | 1 | 0 | 1 | | risk-auditor | final_risk_score | 0.00% | 1 | 0 | 1 | | risk-auditor | final_verdict | 0.00% | 1 | 0 | 1 | ## 建議與後續行動 - 62 個假說仍待驗證，可增加輪次或手動檢視。 - 最常見錯誤類型為 **backtest_crash**（3 次），建議優先修復。