Understand herding and recency bias — Stock-Market notes

Overview

Two of the most powerful behavioral biases that drive market anomalies and create systematic mispricing: herding (following the crowd) and recency bias (overweighting recent events). These biases explain bubles, crashes, momentum effects, and why markets overshoot fundamental values.

Core Concepts

The Mechanics: Why These Biases Exist

Evolutionary Psychology of Herding

WHY does herding exist?

On the savanna, following the crowd when everyone runs = survival
Social learning is efficient: copying successful others saves time
Safety in numbers reduces individual predation risk
Humans evolved as social creatures where group cohesion > individual optimization

HOW does it manifest in markets?

Information asymetry: You observe others' trades but not their reasoning
Uncertainty: When fundamentals are unclear, social proof becomes evidence
Feedback loops: Herding → price moves → validates herding → more herding

WHAT breaks the herd?

Fundamentals eventually reassert (earnings, cash flows)
Contrarian traders with deep pockets (value investors)
External shocks that change the narrative

Cognitive Basis of Recency Bias

WHY does recency bias exist?

Memory accessibility: Recent events are easier to retrieve (availability heuristic)
Adaptive recency: In changing environments, recent data IS more relevant
Emotional vividness: Recent losses/gains still sting/thrill, anchoring expectations
Working memory limits: Can't hold full historical distribution, so use recent sample

HOW does it distort decisions?

True probability distribution: Based on decades of data
Perceived distribution: Dominated by last6-12 months
→ Systematic forecast errors
→ Over/under reaction to recent performance

Mathematical Framework

Formula Herding Intensity Measure

We can quantify herding using the Cross-Sectional Absolute Deviation (CSAD) model:

$\text{CSAD}_t = \frac{1}{N}\sum_{i=1}^{N}|R_{i,t} - R_{m,t}|$ Where:

$R_{i,t}$ = return of stock $i$ at time $t$
$R_{m,t}$ = market return at time $t$
$N$ = number of stocks Under rational asset pricing: $\text{CSAD}_t = \alpha + \gamma_1|R_{m,t}| + \gamma_2 R_{m,t}^2 + \varepsilon_t$ Expected: $\gamma_2 = 0$ (linear relationship) With herding: $\gamma_2 < 0$ (return dispersion decreases non-linearly during extreme moves)

WHY THIS WORKS:

During normal times: stocks move based on individual fundamentals → high dispersion
During herding: stocks move together regardless of fundamentals → low dispersion
Non-linear $\gamma_2 < 0$ captures this collapse of dispersion at extremes

DERIVATION:

Start with rational pricing: $R_{i,t} = \alpha_i + \beta_i R_{m,t} + \varepsilon_{i,t}$

Take absolute deviations from market: $|R_{i,t} - R_{m,t}| = |\alpha_i + (\beta_i - 1)R_{m,t} + \varepsilon_{i,t}|$

Average across stocks: $\text{CSAD}_t \approx a + b|R_{m,t}| + \text{noise}$

But with herding, when $|R_{m,t}|$ is large, investors ignore their $\beta_i$ and just follow the market: $\beta_i^{herd} \to 1\text{ for all } i$

This REDUCES dispersion, creating the negative quadratic term.

Formula Recency-Weighted Expectations

A recency-biased investor forms expectations using exponential decay weights:

$E_t^{biased}[R_{t+1}] = \sum_{s=0}^{\infty} w_s R_{t-s} \quad \text{where} \quad w_s = \frac{\lambda^s}{\sum_{k=0}^{\infty}\lambda^k} = (1-\lambda)\lambda^s$

With recency bias: $\lambda < 1$ (high decay, recent observations dominate) Rational: $\lambda \to 1$ (equal weights on all history)

WHAT THIS MEANS:

$\lambda = 0$ : Only today matters (pure recency)
$\lambda = 1$ : All history equally weighted (rational)
Typical biased: $\lambda \approx 0.6-0.8$ (last year matters2-5x more than 3 years ago)

WHY THE EXPONENTIAL FORM?

Captures memory decay neurologically
Mathematically convenient (recursive updating)
Fits observed behavior in experiments

DERIVATION FROM BAYESIAN UPDATING GONE WRONG:

Rational Bayesian: $P(\theta|D_1, D_2, .., D_T) \propto P(\theta) \prod_{t=1}^{T} P(D_t|\theta)$ All data treated equally.

Recency-biased Bayesian: $P(\theta|D_1, .., D_T) \propto P(\theta) \prod_{t=1}^{T} P(D_t|\theta)^{\lambda^{T-t}}$ Recent data gets higher exponent → higher weight.

For Gaussian returns, this leads to recency-weighted sample mean above.

Figure — Understand herding and recency bias

Worked Examples

Worked example Example 1: Detecting Herding in the2021 Meme Stock Frenzy

Setup: GameStop (GME) rallies from $20 to$ 483 in January 2021. We want to test if herding occurred. Data: Daily returns for GME and 100 similar small-cap stocks.

Step 1: Calculate CSAD

Day 1: GME +135%, Market +2%
Stock A: +8%, Stock B: +3%, Stock C: +12%, ... (100 stocks)
CSAD₁ = (1/100)[|8-2| + |3-2| + |12-2| + ... ] = 8.5%

WHY THIS STEP? CSAD measures how much individual stocks deviate from the crowd. High CSAD = stocks moving independently.

Step 2: Normal period comparison

Normal day: Market +1%, CSAD = 3.2%
GME frenzy: Market +2%, CSAD = 1.8%

WHY LOWER? During herding, ALL stocks move with GME regardless of fundamentals. Individual differences collapse.

Step 3: Regression test

CSAD_t = α + γ₁|R_m,t| + γ₂R²_m,t
Normal period: γ₂ = 0.02 (slight positive)
Frenzy period: γ₂ = -0.15 (strongly negative)

WHY NEGATIVE? The negative quadratic term means: "the bigger the market move, the MORE stocks move together" = herding signature.

Conclusion: Strong evidence of herding. Retail investors piled into meme stocks as a group, ignoring individual fundamentals.

Worked example Example 2: Recency Bias in Tech Stock Expectations (2022 Crash)

Setup: After 2020-2021 tech boom (+40% annual returns), investors extrapolate. Then 2022 hits with -30% crash. How did recency bias distort forecasts?

Step 1: Calculate recency-weighted expected return Historical data:

2022: -30%
2021: +40%
2020: +45%
2019: +30%
2018: -5%

Rational expectation (equal weights):

E[R] = (1/5)[-30 + 40 + 45 + 30 - 5] = 16% average

Recency-biased (λ=0.7) in early 2022 (before crash):

w₀ = (1-0.7) × 0.7⁰ = 0.30
w₁ = (1-0.7) × 0.7¹ = 0.21
w₂ = (1-0.7) × 0.7² = 0.147
w₃ = (1-0.7) × 0.7³ = 0.103
w₄ = (1-0.7) × 0.7⁴ = 0.072
...
E^biased[R] = 0.30(45%) + 0.21(40%) + 0.147(30%) + 0.103(-5%) + ... 
           ≈ 38%

WHY THIS STEP? Recent boom years get 2-3x weight of older data, inflating expectations.

Step 2: Post-crash recency bias (late 2022) Now the -30% is most recent:

E^biased[R] = 0.30(-30%) + 0.21(40%) + 0.147(45%) + ...
           ≈ 5%

WHY? Same bias, opposite direction! Recent crash now dominates, creating excessive pessimism.

Step 3: Compare to rational

Rational: 16% (unchanged, based on long-term average)
Biased pre-crash: 38% → Over-valuation
Biased post-crash: 5% → Under-valuation

Trading implication: Recency bias creates predictable swings around fundamental value. Contrarian strategy: fade the recency-driven moves.

Worked example Example 3: Herding + Recency Interaction in Bitcoin2017 Bubble

Setup: Bitcoin rises from $1,000 (Jan 2017) to$ 19,000 (Dec 2017), then crashes to $3,000 (2018). Both biases amplify each other.

The Feedback Loop: Phase 1: Initial rise ( $1,000 →$ 5,000)

Some early adopters buy (fundamentals-based)
Price rises, making news
Recency bias kicks in: "It's up 5x this year!"
More investors see recent gains, extrapolate to future

Phase 2: Herding accelerates ( $5,000 →$ 15,000)

Media coverage intensifies
Herding begins: "Everyone's talking about Bitcoin, I should buy"
Friends, family, coworkers all buying
FOMO dominates: Fear of missing out > fundamental analysis
Each buyer validates the next: "See, it's still going up!"

Phase 3: Mania ( $15,000 →$ 19,000)

Recency bias extreme: 6-month return = +400%
Investors with λ=0.6: E[R_annual] ≈ 300% (absurd)
Herding complete: Google Trends for "Bitcoin" at all-time high
Grandmothers asking how to buy Bitcoin = classic top signal

Phase 4: Crash ( $19,000 →$ 3,000)

Some whales sell, price drops20%
Recency flips negative: Recent return = -20%
Herding reverses: "Everyone's selling, I should too"
Margin calls force more selling
New recency-biased expectation: "It's going to zero"
Herding accelerates downside: -84% total

Quantifying the interaction:

Herding intensity (CSAD γ₂): 
 Normal: -0.05
 Bubble phase: -0.35 (7x stronger)
Recency weight (λ):
 Normal investor: 0.85
 During bubble: 0.55 (extreme recency)
Combined effect: 7× 1.5 ≈ 10x amplification

WHY THIS MATTERS: This is how all bubles work. Recency bias causes over-extrapolation. Herding amplifies the initial move. Feedback loop continues until fundamentals reassert or the greater fool runs out.

Common Mistakes

Common mistake Mistake 1: "Herding Is Always Irrational"

The wrong intuition: "Following the crowd is dumb. Smart investors think independently."

Why it feels right: We celebrate contrarians like Buffett. Media loves "I was right when everyone else was wrong" stories.

The truth: Herding can be RATIONAL in several situations:

Information cascades with asymmetric info: If others likely have private information, mimicking them is Bayesian-optimal
Career risk: Fund managers herd to avoid career-ending solo mistakes ("No one gets fired for buying IBM")
Coordination games: When value depends on others' actions (e.g., bank runs), herding is Nash equilibrium

How to think correctly:

Distinguish rational informational herding from emotional contagion
Ask: "Do I have information others lack?" If no, some herding is justified
The PROBLEM is herding that ignores fundamentals, not all herding

Example: If 5 expert investors you respect all buy a stock, following them (absent your own analysis) isn't necessarily irrational—they might know more than you.

Common mistake Mistake 2: "Recency Bias Means Recent Data Is Useless"

The wrong intuition: "Since recency bias exists, I should ignore recent events and only look at long-term averages."

Why it feels right: We're told think long-term, not chase recent performance. "This time is NOT different."

The truth: Recent data CAN be more informative when the world has structurally changed:

Regime shifts: Pre-2008 vs. post-2008 monetary policy
Technology disruption: Netflix in 2010vs. 2020
Company lifecycle: Startup vs. mature company

The bias is OVER-weighting recent data, not that recent data should be ignored.

The fix:

Ask: "Has the underlying process changed?"
If yes: Weight recent data more (but still not 100%)
If no: Use longer history to avoid recency bias
Use statistical tests for structural breaks

Example: Tesla's growth rate2018-2020 IS more relevant than 2012-2014 because production scaled and battery costs dropped. But extrapolating 2020 growth forever would be recency bias.

Common mistake Mistake 3: "I Can Exploit These Biases by Doing the Opposite"

The wrong intuition: "Everyone's herding into tech stocks, so I'll short them all. Contrarian = profit."

Why it feels right: Behavioral finance teaches biases create mispricings. Exploit biases → alpha!

The truth: Biases can persist and worsen before correcting:

Limits to arbitrage: Shorting bubles can bankrupt you before you're right ("Markets can stay irrational longer than you can stay solvent" - Keynes)
Bias timing: When does herding END? No one knows precisely
You might be wrong: Maybe "herding" is actually informed buying

The fix:

Don't fight herding DURING the herd—that's catching a falling knife in reverse
Wait for early reversal signs (momentum breakdown, volume exhaustion)
Position size for being early (don't go all-in on contrarian)
Use options for defined risk
Combine with fundamental anchors: only fade herding when price >> value

Example: Shorting Tesla at $200 (2020) because of herding would have destroyed you as it went to$ 1,200 (split-adjusted). Better: wait for momentum break + valuation catalyst.

Active Recall Practice

Recall Feynman Technique: Explain to a 12-Year-Old

Herding: Imagine you're at school and everyone suddenly runs toward the playground. You don't know why, but you run too because "something must be happening!" That's herding. In the stock market, when lots of people buy a stock, others see the price going up and think "I should buy too!" But they're not buying because the company is good—they're buying because everyone else is. Eventually, when people realize there's no real reason to buy, everyone tries to sell at once, and the price crashes.

Recency Bias: Imagine you flip a coin 100 times. It lands heads 70 times and tails 30 times. But in the last 10 flips, it was heads 9 times. Now you have to bet on the next flip. Your brain screams "HEADS!" because that's what you just saw. But the coin is still 50/50! Recency bias means your brain pays too much attention to what JUST happened and forgets about the bigger picture. In stocks, if a company's price went up for the last 3 months, people think it'll go up forever, forgetting that over 10 years it goes up AND down.

Connections

Momentum and reversal effects - Herding creates momentum; recency bias explains why it overshoots
Market bubles and crashes - Both biases necessary for bubble formation and collapse
Behavioral portfolio theory - Explains why investors don't diversify (herd into popular stocks)
Limits to arbitrage - Why smart money can't always correct herding-driven mispricings
Information cascades - Theoretical foundation for rational herding
Prospect theory - Recency bias amplifies loss aversion (recent losses more painful)
Overconfidence bias - Interacts with recency: recent success → overconfidence → herding
Social proof and FOMO - Psychological mechanisms underlying herding
Contrarian investing strategies - Practical application: fade extreme herding + recency
Market efficiency debate - These biases are key evidence against strong-form efficiency

#flashcards/stock-market

What is herding behavior in financial markets? :: The tendency of investors to mic the actions of a larger group, buying because others buy or selling because others sell, independent of fundamental analysis. Creates informational cascades and emotional contagion.

What is recency bias?

The cognitive error of overweighting recent events and underweighting long-term base rates when forming expectations. Recent information is more vivid and easier to recall, distorting probability estimates.

How do we measure herding intensity quantitatively?

Using Cross-Sectional Absolute Deviation (CSAD). Regress CSAD_t = α + γ₁|R_m,t| + γ₂R²_m,t. If γ₂ < 0 (negative quadratic term), it indicates herding—return dispersion decreases non-linearly during extreme market moves.

What does γ₂ < 0 in CSAD regression tell us?

Herding is present. Negative γ₂ means that during large market moves, individual stock returns converge (low dispersion) rather than diverge, indicating investors are following the crowd rather than individual fundamentals.

How do we model recency-biased expectations mathematically?

E_t^biased[R_{t+1}] = Σ w_s R_{t-s} where w_s = (1-λ)λ^s. With recency bias, λ < 1 (typically 0.6-0.8), giving recent observations exponentially higher weight than distant ones.

What does λ = 0.7 mean in recency-weighted expectations?

Recent data gets higher weight with exponential decay. The most recent period gets 30% weight, one period ago gets 21%, two periods ago 14.7%, etc. Last year matters 2-3x more than three years ago.

Why is herding not always irrational?

(1) Information cascades: others may have private information, so mimicking is Bayesian-optimal. (2) Career risk: fund managers herd to avoid solo mistakes. (3) Coordination games: when value depends on others' actions, herding is Nash equilibrium.

What's the key interaction between herding and recency bias in bubbles?

Positive feedback loop: (1) Recent price rise → recency bias inflates expectations. (2) High expectations → herding as more pile in. (3) Herding → higher prices. (4) Loop repeats until fundamentals reassert. Creates10x amplification in bubbles.

When is recent data actually MORE informative than long-term averages?

During structural regime shifts: monetary policy changes, technology disruption, company lifecycle transitions. The bias is OVER-weighting recent data, not that recent data is irrelevant. Use tests for structural breaks.

Why is "fade the herd" not always profitable?

(1) Limits to arbitrage: shorting bubbles can bankrupt you before correction. (2) Bias timing: herding can persist longer than expected. (3) You might be wrong: apparent herding might be informed buying. Markets can stay irrational longer than you can stay solvent.

What are the three types of herding?

(1) Informational cascades: "They know something I don't." (2) Reputational herding: "I won't get fired for doing what everyone does." (3) Emotional contagion: Fear and greed spread socially through networks.

How did herding manifest in the 2021 meme stock frenzy?

CSAD collapsed during extreme moves—γ₂ became strongly negative (-0.15 vs normal +0.02). All stocks moved together with GME regardless of individual fundamentals, as retail investors piled in as a group driven by social media.

What was the recency bias effect on tech expectations2021→2022?

Pre-crash: λ=0.7 weighting inflated expected returns to 38% (vs rational 16%), causing overvaluation. Post-crash: same bias flipped negative, deflating expectations to 5%, causing undervaluation. Same bias, opposite extremes.

Concept Map

Hinglish (regional understanding)

Intuition Hinglish mein samjho

Hinglish (regional understanding)

Intuition Hinglish mein samjho

Dekho yaar, ye do biases samajhna market mein bahut zaroori hai. Herding ka matlab hai bheed ko follow karna—jab sab log ek stock khareed rahe hote hain, toh tumhein bhi lagta hai "yaar inko kuch pata hoga jo mujhe nahi pata", aur tum bhi khareed lete ho, chahe fundamentals kuch bhi kahein. Ye FOMO (Fear Of Missing Out) ka game hai. Wahi doosri taraf recency bias hai—jahan tum recent events ko zyada importance dete ho. Jaise agar last 6 mahine ek stock upar gaya hai, toh tum sochte ho ye hamesha upar hi jayega, aur long-term data ko ignore kar dete ho. Restaurant wala example yaad rakho—doosron ke order dekhkar copy karna herding hai, aur kal ka special sunke plan change karna recency bias hai.

Ab ye biases matter kyun karte hain? Kyunki jab millions of investors ek saath yahi galti karte hain billions of dollars ke saath, toh market mein predictable patterns ban jaate hain—bubbles, crashes, momentum effects sab isi se aate hain. Prices apni real fundamental value se overshoot kar jaate hain. Evolution ne humein aisa banaya hai—savanna pe jab sab bhaagte the, follow karna survival tha. Lekin markets mein ye same instinct nuksaan karwa deta hai. Achhi baat ye hai ki agar tum in patterns ko samajh lo, toh inko exploit kar sakte ho ya at least avoid kar sakte ho.

Mathematically, herding ko measure karne ke liye CSAD (Cross-Sectional Absolute Deviation) model use hota hai. Simple intuition ye hai—normal time mein har stock apne individual fundamentals ke hisaab se move karta hai, toh returns mein dispersion (fark) zyada hota hai. Lekin jab herding hoti hai, toh saare stocks ek saath move karte hain regardless of fundamentals, aur dispersion kam ho jaati hai. Isiliye jab market extreme move karta hai aur $\gamma_2 < 0$ nikalta hai, toh ye herding ka signal hai. Yaad rakho—end mein fundamentals hamesha wapas aa jaate hain, isiliye bheed ke peeche andha-dhundh bhaagne se behtar hai apna analysis karna.

Test yourself — Factor & Behavioral Finance

Connections