# Overnight Edge Research — MU & Market Microstructure **Date:** August 22, 2026 **Trigger:** Prahlad asked about "buy at close, sell at open" strategies that supposedly "always make money" --- ## Key Finding: Overnight Premium EXISTS but is NOT Arbitrage ### MU (Micron) 5-Year Backtest Results **Period:** Aug 2021 – Aug 2026 (1,253 trading days) | Strategy | Final Value | CAGR | Sharpe | Max DD | Win Rate | |----------|-------------|------|--------|--------|----------| | **Overnight** (Buy 4PM, Sell 9:30AM) | $6,367 | **45.11%** | 1.18 | -53.69% | 52.5% | | **Intraday** (Buy 9:30AM, Sell 4PM) | $2,163 | 16.79% | 0.58 | -60.48% | 50.8% | | **Buy & Hold** (24H) | $13,774 | 69.47% | — | — | — | **🔑 Key Insight:** MU makes **46.2% of its total gains overnight** (only 17.5 hours), despite the market being closed. The remaining 53.8% comes from 6.5 hours of active trading. --- ## Annual Breakdown (MU) | Year | Overnight Return | Intraday Return | Total Return | |------|------------------|-----------------|--------------| | 2021 | +9.8% | +18.2% | +28.2% | | 2022 | **-63.9%** ❌ | +14.3% | -50.3% | | 2023 | +22.0% | +38.7% | +60.6% | | 2024 | **+67.2%** ✅ | -54.2% | +12.3% | | 2025 | +81.8% | +59.3% | +142.2% | | 2026 YTD | **+102.8%** ✅ | +44.0% | +148.2% | **Observation:** 2022 shows the overnight premium CAN go negative (-64%) in bear markets. Not risk-free. --- ## Why This Edge Exists From academic literature (QuantReturns.com, 2025): 1. **Overnight Mean Reversion (CO-OC):** - Stocks that gap up overnight tend to REVERSE intraday - Stocks that gap down overnight tend to BOUNCE intraday - Sharpe ratios: 2-4+ across equity futures, sector ETFs (Biotech: 4.44 Sharpe, Financials: 9.0 Sharpe) 2. **Overnight Momentum (Close → Open):** - Most S&P 500 gains since 1993 occur overnight - Retail can't trade after-hours → institutions drive price discovery - Earnings releases (pre-market/after-hours) create gaps 3. **Why MU specifically:** - Semiconductor sector → heavy after-hours news flow (Taiwan supply chain, AI demand, memory pricing) - High institutional ownership → overnight order flow imbalance - Earnings volatility → big overnight gaps --- ## Why Retail Doesn't Exploit This ### Practical Barriers (from QuantReturns caveat): 1. **Execution Timing:** - Strategy requires buying AT THE CLOSE (4:00:00 PM exactly) - And selling AT THE OPEN (9:30:00 AM exactly) - Retail orders at "market close" often fill seconds late → edge erodes - After-hours trading has WIDE spreads (0.5-2% on volatile stocks like MU) 2. **Transaction Costs:** - Daily round-trips = 252 trades/year - Even $0 commissions have SEC fees (~$0.23 per $1000 sold) - Bid-ask spread at close/open can be 0.1-0.3% on MU ($964 stock = $3-9 per share) - **Slippage kills the edge** unless you have institutional execution 3. **Overnight Risk:** - Can't exit if bad news hits 5PM-9AM (e.g., Taiwan earthquake, export ban) - Max drawdown -53% shows this isn't "always profitable" - 2022: -64% overnight loss in bear market 4. **Capital Efficiency:** - Holding overnight = capital locked 17.5 hours for 0.175% avg return - Intraday = 6.5 hours locked for 0.096% avg return - But you CAN'T compound both (mutually exclusive strategies) 5. **Psychological:** - Win rate only 52.5% → nearly half the days lose money - Drawdowns feel worse when you're "doing nothing" overnight vs active trading --- ## CO-OC Mean Reversion Strategy (Different from Momentum) The **QuantReturns strategy** is the INVERSE of what Prahlad described: - **Formation signal:** Overnight return (close → open) - **Trade:** If overnight return is NEGATIVE (gap down), GO LONG during next intraday session - **Trade:** If overnight return is POSITIVE (gap up), GO SHORT during next intraday session - **Hold:** Open → Close only (exit at 4PM) **This is NOT "buy at close, sell at open"** — it's "use overnight move to predict intraday reversal". ### Our MU Test of CO-OC Mean Reversion: - **CAGR:** -22.98% ❌ - **Final value:** $273 (from $1000) - **Sharpe:** -0.42 **Conclusion:** CO-OC mean reversion does NOT work on MU. The stock exhibits **overnight MOMENTUM** (gaps continue intraday), not reversion. --- ## What We DON'T Know Yet 1. **Does the overnight edge work on OTHER stocks?** - Test: SPY, QQQ, NVDA, TSLA, AAPL - Hypothesis: High-beta tech stocks (NVDA, TSLA) should show similar patterns 2. **Can we filter for WHEN the overnight edge is strongest?** - VIX regime (edge stronger in high-VIX environments?) - Earnings proximity (overnight gaps bigger near earnings?) - Sector rotation (overnight edge in momentum sectors?) 3. **Can we ACCESS better execution?** - Alpaca supports after-hours trading (4-8PM, 4-9:30AM) - But: spreads are WIDE, liquidity is THIN - MOC (Market-On-Close) orders → guarantee 4PM execution - MOO (Market-On-Open) orders → guarantee 9:30AM execution - **Actionable:** Test Alpaca MOC/MOO order types 4. **What about OPTIONS on overnight moves?** - Buy ATM call at close, sell at open → capture overnight gamma - Problem: options have WIDER spreads, theta decay overnight - Likely not profitable after costs --- ## Next Steps 1. **Backtest overnight edge on SPY/QQQ** (broader market) 2. **Test Alpaca MOC/MOO orders** (can we actually execute at close/open?) 3. **Build a filter:** Only trade overnight when VIX > 20 or earnings within 7 days 4. **Document in MEMORY.md:** This is a KNOWN edge but NOT exploitable by retail without execution infrastructure --- ## References - QuantReturns: "Overnight Mean-Reversion" (Oct 2025) — Sharpe 4+ strategies - QuantifiedStrategies: "200 Trading Strategies" — Overnight edge since 1993 - Our backtest: `/tmp/mu_overnight_momentum.py` (Aug 22, 2026) --- **Bottom Line:** The overnight premium is REAL but NOT arbitrage. MU gains 46% overnight, but: - Max drawdown -53% - Requires perfect execution (impossible for retail) - Transaction costs erode edge - Can't exit during overnight risk events **It's an EDGE, not a FREE LUNCH.**