Lesson Learned: Pyramid Buying During Fear Destroyed $96 (Dec 15, 2025)

ID: ll_020 Date: December 15, 2025 Severity: CRITICAL Category: Strategy, Risk Management, Behavioral Finance, Data Integrity Impact: Portfolio lost $96 in 5 days, 0% win rate on fear-based trades

Executive Summary

The “buy the dip” strategy during “Extreme Fear” conditions was catching falling knives with 0% win rate. The system was programmed to BUY MORE (1.5-2x) when the market was falling, which is the opposite of what works for retail traders. Additionally, misleading backtest metrics (“62.2% win rate, 2.18 Sharpe”) were hardcoded in hooks despite never existing in any actual data.

The Problem

What Happened (Dec 13-15, 2025)

Date	Fear/Greed Index	Action	Result
Dec 13	21 (Extreme Fear)	BUY 1.5x ETHUSD	Loss
Dec 13	21 (Extreme Fear)	BUY 1.5x BTCUSD	Loss
Dec 14	16 (Extreme Fear)	BUY 1.5x ETHUSD	Loss
Dec 14	16 (Extreme Fear)	BUY 1.5x BTCUSD	Loss

Portfolio Impact: +$17.49 → -$96.19 (lost $113.68 in 5 days)

Root Causes

Contrarian Timing Doesn’t Work for Retail
- Fear Greed Index at 21 was NOT the bottom
- “Extreme Fear” can persist for weeks in downtrends
- Research shows trend following beats contrarian for retail (97% of day traders lose)

Size Multiplier Amplified Losses

# BEFORE (WRONG):
if value <= EXTREME_FEAR_THRESHOLD:
    size_multiplier = 1.5  # Buy 50% MORE during fear

Buy The Dip Logic Caught Falling Knives

# BEFORE (WRONG):
if btc_change <= -2.0:
    multiplier = min(2.0, 1.0 + abs(btc_change) / 10)  # Up to 2x during dips

Misleading Metrics in Hooks
- Hooks showed: “62.2% win rate, 2.18 Sharpe”
- Reality: 0/13 scenarios pass, Sharpe -7 to -2086 (ALL NEGATIVE)
- The 2.18 Sharpe NEVER EXISTED in any backtest data

Evidence from Live Trades

Strategy	P/L	Win Rate	Status
Options (short)	+$327.82	100%	WORKED
Crypto (long)	-$0.43	0%	FAILED
Equities (long)	-$4.15	~40%	MARGINAL

Options selling was the actual edge. Crypto buying had zero edge.

The Fix

1. Disabled Fear Size Multiplier

File: src/utils/fear_greed_index.py

# AFTER (CORRECT):
if value <= self.EXTREME_FEAR_THRESHOLD:
    action = "HOLD"  # Changed from BUY
    size_multiplier = 1.0  # Changed from 1.5
    confidence = 0.3  # Low confidence - fear can persist
    reasoning = f"Extreme Fear ({value}) - WAITING for trend confirmation."

2. Reversed Buy The Dip Logic

File: src/strategies/crypto_strategy.py

# AFTER (CORRECT):
if btc_change <= -2.0:
    multiplier = 0.5  # REDUCE position during downtrend, don't increase
    logger.info(f"⚠️  BTC DIP: REDUCING position - DON'T CATCH FALLING KNIVES")

3. Honest Metrics in Hooks

Files: .claude/hooks/inject_trading_context.sh, load_trading_state.sh

# AFTER (CORRECT):
Win Rate: $WIN_RATE% (live) | Backtest: 0/13 scenarios pass (Sharpe all negative)
Backtest Reality: 0/13 pass | Sharpe: -7 to -2086 | NO EDGE YET

Prevention Rules: Tests and Guardrails

1. Backtest Metric Validation Test

def test_no_hardcoded_metrics():
    """Ensure hooks don't contain hardcoded performance claims."""
    with open('.claude/hooks/inject_trading_context.sh') as f:
        content = f.read()

    # These should never be hardcoded
    assert "2.18 Sharpe" not in content
    assert "62.2% win rate" not in content

    # Should reference actual data
    assert "latest_summary.json" in content or "dynamic" in content.lower()

2. Size Multiplier Sanity Check

def test_fear_multiplier_not_increasing():
    """Ensure fear conditions don't INCREASE position size."""
    from src.utils.fear_greed_index import FearGreedIndex

    fgi = FearGreedIndex()
    signal = fgi.get_trading_signal_for_value(20)  # Extreme fear

    # Should NOT increase position during fear
    assert signal["size_multiplier"] <= 1.0
    assert signal["action"] != "BUY"  # Should wait for confirmation

3. Strategy Win Rate Gate

def test_strategy_minimum_win_rate():
    """Block strategies with <45% win rate from production."""
    from data.backtests.latest_summary import get_aggregate_metrics

    metrics = get_aggregate_metrics()

    # If win rate falls below 45%, strategy should be quarantined
    assert metrics["min_win_rate"] >= 45.0, "Strategy win rate too low for production"

4. Live P/L Circuit Breaker

def test_pl_circuit_breaker():
    """Halt trading if P/L drops more than 1% in a week."""
    from data.performance_log import get_weekly_change

    weekly_change = get_weekly_change()

    if weekly_change < -1.0:
        raise CircuitBreakerTriggered(
            f"Weekly P/L {weekly_change}% exceeds -1% threshold. "
            "Trading halted for review."
        )

Behavioral Finance Lessons

Why Retail Traders Fail at Contrarian Timing

Fear can persist: FGI at 20 doesn’t mean bottom - can go to 10
Trend following wins: Research shows 12.5% success for swing traders vs 5% for day traders
Buy strength, not weakness: Buying during uptrends (+5%) has better odds than buying dips (-5%)
97% of day traders lose: The “buy the dip” mentality is why

The Correct Mental Model

WRONG: "Market is fearful, this is a buying opportunity!"
RIGHT: "Market is fearful, fear can persist. Wait for trend confirmation."

WRONG: "BTC is down 5%, buy more to average down!"
RIGHT: "BTC is down 5%, reduce exposure until trend reverses."

Checklist for Future Strategy Changes

Before deploying any strategy change:

Does it increase position size during adverse conditions? (RED FLAG)
Is it based on backtested data or wishful thinking?
What’s the win rate in the WORST scenario, not the best?
Does it match what research says works for retail traders?
Are metrics shown to users derived from actual data, not hardcoded?
Is there a circuit breaker if it starts losing?

ll_016_regime_pivot_safety_gates_dec12.md - Safety gate architecture
ll_011_facts_benchmark_factuality_ceiling.md - Fact-checking systems
ll_012_deep_research_safety_improvements_dec11.md - Research validation