Behavioral Finance

Behavioral Finance Applications

Overview

Translate behavioral-finance theory into quantifiable trading signals and risk-control rules. Core assumption: market participants systematically deviate from rational decision-making, and these biases can be predicted and exploited.

Applicable scenarios:

• Behavioral interpretation and parameter optimization for momentum / reversal strategies
• Contrarian signals when market sentiment becomes extreme
• Debiasing mechanisms in portfolio construction
• Capturing behavior patterns specific to retail-driven China A-share markets

Core Concepts

Overreaction and Underreaction

Underreaction → momentum effect:

Copy code
Mechanism: anchoring bias + conservatism
  Investors anchor on old information and update insufficiently to new information
  After an earnings beat, the stock price digests it gradually rather than all at once
China A-share evidence:
  - Earnings-guidance beats still produce 3-5% excess return over the following 20 days
  - After analyst rating upgrades, momentum often persists for 1-3 months
Quant signal:
  SUE (standardized unexpected earnings) > 2σ -> buy and hold for 60 days
  Top 10% 20-day return -> continue holding for 20 days (China A-share momentum cycles are shorter)

Overreaction → reversal effect:

Copy code
Mechanism: representativeness heuristic + availability bias
  Investors extrapolate recent trends too aggressively and ignore mean reversion
  Panic / euphoria drives reactions beyond what fundamentals support
China A-share evidence:
  - Rebounds after consecutive limit-downs (after 3 limit-downs, the average 20-day rebound is 8%)
  - Big annual losers often earn 5-10% excess return the next year
Quant signal:
  Bottom 10% of 250-day return -> buy and hold for 250 days
  RSI(5) < 10 -> short-term rebound signal (5-10 days)

Key distinction:

Cognitive Bias Checklist

Individual decision biases:

Group behavior biases:

Investor Sentiment Cycle

Copy code
Fear -> Caution -> Optimism -> Excitement -> Euphoria -> Denial -> Panic -> Fear
  |        |        |        |        |       |        |
 Bottom   Recovery  Mid-uptrend  Pre-top   Top   Early selloff  Pre-bottom

Quant sentiment indicators:
  1. Closed-end fund discount: discount > 15% -> extreme fear -> buy signal
  2. Margin-financing growth: monthly growth > 20% -> euphoria -> reduce position
  3. New account openings: weekly openings > 2x average -> overheated market
  4. Turnover ratio: All-A daily turnover > 3% -> euphoric; < 0.5% -> deeply depressed
  5. Number of limit-up stocks: > 100 -> euphoric; < 10 -> weak

Analysis Framework

1. Disposition-Effect Signal

Principle: investors tend to sell winners and hold losers. Once winning positions are largely cleared, selling pressure eases; when trapped holders are deeply underwater, selling pressure can also ease.

Copy code
China A-share application:
  Compute the profit ratio in the chip-distribution structure:
  - Profit ratio > 90% and shrinking volume -> winners are reluctant to sell -> may continue rising
  - Profit ratio > 90% and expanding volume -> winners are exiting -> topping signal
  - Profit ratio < 10% and shrinking volume -> low willingness to cut losses -> bottom stabilization
  - Profit ratio < 10% and expanding volume -> panic selling -> short-term oversold

Quant implementation:
  capital_gain_overhang = (current_price - avg_cost) / avg_cost
  where avg_cost is approximated by 60-day VWAP
  CGO > 0.2 -> strong unrealized gains, watch for disposition-effect selling pressure
  CGO < -0.3 -> deeply trapped holders, selling pressure may actually ease

2. Composite Sentiment Indicator

pythonCopy code
# Multi-dimensional sentiment score (0-100, 50 = neutral)
sentiment_components = {
    'turnover_ratio': normalize(all_a_turnover, historical_percentile),      # weight 25%
    'margin_growth': normalize(monthly_margin_growth, historical_percentile), # weight 25%
    'new_high_ratio': normalize(new_high_ratio, historical_percentile),       # weight 20%
    'limit_up_count': normalize(limit_up_count, historical_percentile),       # weight 15%
    'fund_discount': normalize(closed_end_fund_discount, historical_percentile), # weight 15% (inverse)
}

sentiment_score = weighted_sum(components)
# > 80: extreme greed -> cut exposure below 60%
# 60-80: optimistic -> maintain normal exposure
# 40-60: neutral -> keep exposure unchanged
# 20-40: pessimistic -> add gradually
# < 20: extreme fear -> increase exposure above 80%

3. Behavioral Optimization of Momentum Strategies

Traditional momentum (sorting by past 12-month returns) is unstable in China A-shares. A behavioral-finance perspective suggests the following optimizations:

Copy code
Optimization 1: Separate sentiment momentum from fundamental momentum
  Sentiment momentum = part of recent price rise with no fundamental support -> short-term reversal
  Fundamental momentum = price rise consistent with earnings revisions -> can persist
  Trade: buy stocks with "strong fundamental momentum + weak sentiment momentum"

Optimization 2: Attention-weighted momentum
  High-attention retail names reverse faster
  Indicator: if abnormal turnover > 3x average, cut momentum holding period by 50%
  Example: if a normal momentum basket holds for 60 days, high-attention names hold only 30 days

Optimization 3: Combine cross-sectional momentum and time-series momentum
  Cross-sectional: relative strength (top 20% in return ranking)
  Time-series: absolute trend (price > MA60)
  Both satisfied -> strong signal; only one satisfied -> half position

4. Contrarian Trading Signals

Copy code
Extreme-fear buy conditions (at least 3 items):
  □ Shanghai Composite RSI(5) < 15
  □ All-A daily turnover < 0.5%
  □ Weekly margin-financing decline > 5%
  □ Limit-up count < 10 and limit-down count > 50
  □ Closed-end fund discount > 15%

Extreme-greed sell conditions (at least 3 items):
  □ Shanghai Composite RSI(5) > 90
  □ All-A daily turnover > 3%
  □ Weekly margin-financing growth > 10%
  □ Limit-up count > 150
  □ Weekly increase in new account openings > 100%

Output Format

Behavioral-finance analysis report:

Copy code
=== Market Sentiment Diagnosis ===
Date: 2026-03-28
Sentiment score: 72/100 (optimistic bias)
Current phase: transition from optimism to excitement

=== Behavioral-Bias Signals ===
Overreaction detection: 127 stocks rose > 15% in the past 5 days -> 65% probability of short-term reversal
Disposition effect: winner-clearing ratio is low (35%) -> overhead selling pressure remains
Herding effect: sector correlation 0.85 -> severe follow-the-leader behavior, divergence likely soon

=== Strategy Recommendations ===
Momentum strategy: shorten holding period from 60 days to 30 days (market attention is elevated)
Contrarian signal: not triggered (sentiment is not yet extreme)
Position suggestion: maintain 70% exposure, and prioritize names with "strong fundamental momentum + weak sentiment momentum"

=== Debiasing Checklist ===
□ Are you overconfident because of recent profits? -> check position concentration
□ Are you anchored to your entry price? -> re-evaluate using current PE/PB
□ Are you ignoring bearish information? -> force yourself to read bearish research reports

Notes

1. High retail participation in China A-shares: behavioral-bias signals are more pronounced than in US equities, but sector rotation is also faster, so momentum windows should be shorter
2. Lag in sentiment indicators: margin-financing balance is released T+1, and new account openings are weekly, so they are not suitable for intraday trading
3. Structural changes: after 2019, foreign capital and quant participation rose, so the effectiveness of traditional behavioral-finance signals may have weakened
4. Behavioral factors correlate with traditional factors: disposition-effect factors correlate about 0.3-0.5 with momentum, so control collinearity
5. Overfitting risk: behavioral stories are easy to explain after the fact, so out-of-sample validation is mandatory
6. Extreme sentiment is rare: extreme fear / greed appears only 2-3 times per year, so strategy capacity is limited

Dependencies

bashCopy code
pip install pandas numpy scipy