Mean Reversion

Regime rules (Faber 2007, QuantifiedStrategies):

• SPY > 200 DMA + VIX < 25 → Full position
• SPY > 200 DMA + VIX 25-35 → Half position
• SPY < 200 DMA + VIX < 35 → Quarter position, require 3+ confirmations
• VIX > 35 → PAUSE. No entries. (All 3 historical 200 WMA failures occurred here.)

Step 3: Evaluate Against Research

Which WMA matters?

Do NOT assume 200 WMA. The script checks 40/100/150/200 WMA and reports bounce rates for each. Use whichever the stock historically respects most. Example: META respects the 100 WMA (84% bounce rate), not the 200 WMA.

Confirmation signals (ranked by backtested evidence):

Require 1+ for moderate setup, 2+ for strong. 0 confirmations = WAIT for better entry.

Academic basis for the trade:

• Avramov et al. (2021): MAD effect → 9% annualized alpha from stocks far below long-term MA. t-stats 4.29-6.04. Stronger on long side. Survives transaction costs.
• Poterba & Summers (1988): Negative autocorrelation at 2+ year horizons → prices revert. 200 WMA (3.85 years) sits in the optimal reversion timeframe.
• Grimes (2015): The MA is NOT magic support — it's a proxy for "deeply oversold relative to long-term trend." The edge is general mean-reversion at extremes.
• Build Alpha (2022): Mean-reversion is HARMED by tight stops. 8 stop variants tested across 3 strategies. Use wide stops (5-8% below WMA) or time-based exits.

Quality filter (Novy-Marx 2013, JFE):

For mega-cap mean-reversion (>$100B), add a profitability screen:

• Gross profit / total assets > sector median → quality name, full conviction
• Below median → the stock may be cheap for fundamental reasons, not temporary oversold. Reduce size or skip.
• This separates "MSFT temporarily oversold at 200 WMA" from a name whose business is deteriorating.
• Search for recent earnings: positive revenue growth + positive FCF = quality. Declining margins + rising debt = skip.

Universe tier adjustments:

Stop-loss framework:

Critical: mean-reversion gets MORE attractive as price drops further below the WMA (within reason). Don't stop out too early.

Exit targets:

• Trim 1/3 at 20 EMA (first resistance)
• Trim 1/3 at 50 DMA
• Trail final 1/3 with 3-week trailing stop
• Alternative: RSI(2) > 70 for quick profit-taking (Connors R3: 90% win rate, PF 7.17)

Step 4: Search for Fundamental Context

Search the web for:

• Why is this stock declining? Technical correction vs fundamental crisis.
• Recent earnings, guidance, analyst changes.
• The 2022 META lesson: mean-reversion FAILS during fundamental crises. Distinguish.

Also check neighboring repos for relevant patterns:

• ../iran-oil-statarb/src/models/ols.py — Rolling regression methodology for signal generation
• ../quant-trading/Bollinger Bands Pattern Recognition backtest.py — Mean-reversion pattern detection
• ../quant-trading/RSI Pattern Recognition backtest.py — RSI confirmation patterns

Step 5: Output Format

State clearly:

1. Setup quality: STRONG (2+ confirms, >70% bounce rate) / MODERATE (>60% bounce rate) / WEAK / NONE
2. Which WMA and its historical bounce rate with sample size
3. Confirmations active — list each with its current value vs threshold
4. Regime — full/half/quarter/paused with VIX and SPY levels
5. Stop-loss price with method and rationale
6. Exit targets with prices and % from entry
7. Bull scenario with median forward return from historical touches
8. Bear scenario — what if the WMA breaks? Where is the next support?
9. Statistical basis — cite the paper or backtest behind each recommendation