Understand strategy idea generation — Stock-Market notes

What Is Strategy Idea Generation?

WHY this definition? Trading isn't guessing—it's scientific method applied to markets. You need the "why" (hypothesis) because markets adapt; understanding the cause helps you know when the edge will decay.

The Strategy Idea Lifecycle

###1. Observation Sources

Worked example Where Ideas Come From

A. Academic Research (The "Factor Zoo"):

Example: Jegadeesh & Titman (1993) show momentum—stocks that outperformed in the past3-12 months continue outperforming.
Why it works: Behavioral bias (herding, under-reaction to news) + institutional friction (slow capital reallocation).
How to use: Screen for stocks with 6-month return > 20%, hold1 month, rebalance monthly.

B. Market Microstructure:

Example: On earnings announcement days, implied volatility typically drops 10-30% post-release (the "volatility crush").
Why: Option sellers overprice near-term uncertainty; realized move is usually less than implied.
How: Sell stradles 1 day before earnings, exit at open next day.

C. Domain Expertise:

Example: A supply-chain analyst notices semiconductor lead times spiking → hypothesis that chip stocks will beat earnings.
Why: Private information advantage from specialized knowledge.
How: Long semiconductor ETF when lead-time index crosses above 12-week moving average.

D. Data Mining (with caution):

Example: You notice stocks with tickers starting with "A" outperform on Mondays (0.2% avg return vs 0.05% market).
Why this is dangerous: Probably spurious correlation (random chance in sample). No economic mechanism.
Fix: Demand a plausible story + out-of-sample validation + multiple markets.

WHY these sources? Each has different risk profiles. Academic factors are well-known (crowded) but robust. Microstructure exploits structural features (durable). Domain expertise is high-alpha but hard to scale. Data mining without theory is curve-fitting.

2. Formulating the Hypothesis

Formula Hypothesis Structure

$\text{Hypothesis} = \underbrace{\text{Signal}}_{\text{What predicts?}} + \underbrace{\text{Mechanism}}_{\text{Why?}} + \underbrace{\text{Regime}}_{\text{When?}}$

Derivation from first principles:

Signal construction: Convert observation into quantifiable variable.
- Example: Momentum signal = $\frac{P_{\text{today}} - P_{\text{252 days ago}}}{P_{\text{252 days ago}}}$ (1-year return)
Economic mechanism: Link signal to expected return via rational/behavioral story.
- Momentum persists because information diffusion is slow (Hong & Stein 1999)—news propagates gradually through investor networks.
Regime conditions: Specify when the mechanism is strongest.
- Momentum works best in trending markets (low volatility), fails in mean-reverting crashes (2008-09). Mathematical representation: $\mathbb{E}[R_{t+1} \mid \text{Signal}_t] = \alpha + \beta \cdot \text{Signal}_t + \varepsilon_t$

$\alpha$ : Baseline return (should be ≈0 if market-neutral)
$\beta$ : Signal sensitivity—this is your edge
$\varepsilon_t$ : Noise (minimize with proper filtering)

WHY derive it? Because "buy when momentum > 0" is incomplete. You need to know:

Position size (proportional to signal strength? Binary?)
Holding period (momentum decays after 6-12 months)
Risk filters (avoid momentum in high-vol regimes)

3. Converting Hypothesis → Strategy

Worked example Worked Example: Mean Reversion Strategy

Step 1: Observation Intraday, stock prices often "overeact" to news, then partially reverse within hours.

Step 2: Hypothesis Liquidity provision creates temporary imbalances. When a large sell order hits the book, market makers lower bids → price dips below "fair value" → bargain hunters step in → price recovers.

Step 3: Signal Construction $\text{Signal}_t = \frac{P_{\text{VWAP}} - P_t}{P_{\text{VWAP}}}$

$P_t$ : Current price
$P_{\text{VWAP}}$ : Volume-weighted average price over last 30 minutes
Interpretation: If $\text{Signal}_t > 0.5\%$ , price is "cheap" relative to recent average → buy.

Step 4: Entry/Exit Rules

Entry: Buy when Signal > +0.5% (price 0.5% below VWAP)
Exit: Sell when price crosses VWAP (mean-reversion complete) OR stop-loss at -1% (trend continues)
Position size: $\frac{\text{Capital} \times 0.02}{\text{StopLoss}}$ (2% risk per trade)

Step 5: Why Each Rule?

Why 0.5% threshold? Smaller moves are noise; larger moves may be informed (real news, not just liquidity). Backtest showed 0.5% optimizes Sharpe ratio.
Why VWAP exit? Mean reversion is short-lived (hours). Waiting longer exposes you to overnight risk (news, gaps).
Why 1% stop? Limits loss if you're wrong (e.g., panic selling continues). 2:1 reward:risk (0.5% target / 1% stop ≈ 0.5, but win rate > 60% makes it positive expectancy).

WHY work through this? Generic ideas ("buy low, sell high") are useless. You need quantitative rules that a computer (or you) can execute identically every time.

Common Idea Generation Frameworks

A. Factor-Based Approach

Example: Value factor

Observation: Cheap stocks (low P/E, P/B) outperform expensive stocks by ~5% annually (Fama-French 1992).
Mechanism: Behavioral (overeaction to bad news → undervaluation) + risk-based (distressed firms have higher required returns).
Implementation: Long lowest quintile P/B, short highest quintile P/B, rebalance quarterly.

B. Event-Driven Approach

Worked example Earnings Surprise Strategy

Setup: Companies that beat earnings estimates by >10% continue outperforming for 1-2 months (post-earnings announcement drift, Ball & Brown 1968).

Why it works:

Anchoring bias: Analysts slowly revise forecasts upward → persistent underestimation.
Institutional constraints: Mutual funds update positions monthly, not instantly. Execution:

Screen: Stocks with $(EPS_{\text{actual}} - EPS_{\text{consensus}}) / |EPS_{\text{consensus}}| > 10\%$
Entry: Buy at market open day after earnings (avoid-hours volatility)
Exit: Sell after 30 days OR if surprise turns negative next quarter
Risk: Position size =5% of portfolio per stock (diversify across 20+ positions)

Why 30 days? Drift is strongest in first month; after that, information is fully priced. Holding longer increases risk without proportional reward.

C. Statistical Arbitrage (Pairs Trading)

Formula Cointegration-Based Pairs

Core idea: Two stocks with similar business models should move together long-term, even if they diverge short-term. $\text{Spread}_t = P_A - \beta \cdot P_B$

If Spread is stationary (mean-reverting), we have a pair.
When Spread > +2σ, short $A$ , long $B$ (expect convergence).
When Spread < -2σ, long $A$ , short $B$ .

Derivation:

Hypothesis: $P_A$ and $P_B$ are cointegrated if $P_A = \beta P_B + \varepsilon$ , where $\varepsilon$ is stationary.
Test: Run Engle-Granger cointegration test. If p-value < 0.05, reject null (no cointegration) → pair is valid.
Signal: $z\text{-score} = \frac{\text{Spread}_t - \mu_{\text{Spread}}}{\sigma_{\text{Spread}}}$ $z -score = \frac{Spread _{t} - μ _{Spread}}{σ _{Spread}}$
- Enter when $|z| > 2$ (2 standard deviations)
- Exit when $z$ crosses 0 (mean reversion)

Example: Coca-Cola (KO) vs Pepsi (PEP)

Both are beverage companies → similar revenue drivers (consumer spending, commodity costs).
Spread = $\text{KO} - 0.85 \times \text{PEP}$ (estimated β=0.85 from regression).
If Spread spikes to +$5 (KO expensive relative to PEP), short KO, long PEP.

WHY this works? Relative value is easier to predict than absolute direction. You're not betting on market going up/down—just that mispricing between similar assets will correct.

Validating Ideas (Pre-Backtest Sanity Checks)

Recall Explain to a 12-Year-Old

Imagine you're trying to win at video game, but instead of just playing randomly, you watch other players and notice patterns:

"When players pick the fire sword, they usually beat the ice boss faster."
"Players who save their power-up for the final level score higher." Strategy idea generation is doing this for the stock market. You look for patterns (like "stocks that go up a lot keep going up" or "companies that beat earnings keep going up"), figure out why the pattern happens (maybe people are slow to notice good news), and then make a rule: "I'll buy these stocks and sell after1 month."

But here's the trick: you can't just guess. You need to test your idea with past data (like replaying old game matches) to see if it actually worked. Most ideas fail! So you need lots of ideas, test them all, and only use the ones that worked consistently, not just got lucky once.

And just like in games, if everyone learns the same trick, it stops working (the game gets "patched"). So you always need new ideas.

Active Recall Checkpoints

Recall Self-Test Questions

What are the four components of a strategy hypothesis? (Observation, hypothesis/mechanism, implementation, falsifiability)
Why is "the why" crucial? (Understanding mechanism helps you know when edge decays; prevents curve-fitting)
What's the difference between rational and behavioral edges? (Rational = risk-based, decays faster as capital flows in; behavioral = persistent biases, more durable)
Walk through converting "stocks with high momentum outperform" into a complete strategy. (Define momentum metric → entry threshold → holding period → exit rule → position sizing → risk controls)
What is post-earnings announcement drift, and why does it exist? (Stocks that beat earnings keep going up for1-2 months due to analyst anchoring + slow institutional adjustment)

Mnemonic & Mental Models

Connections 6.1.01-Introduction-to-algorithmic-trading – Strategy generation is step 1 of the algo pipeline

6.1.04-Backtesting-strategies – Ideas must be validated with historical data
6.2.01-Risk-management-in-algo-trading – Position sizing depends on strategy risk profile
5.3.02-Market-efficiency-hypothesis – Idea generation exploits market inefficiencies
4.2.03-Behavioral-biases – Many strategies profit from investor psychology
3.4.01-Factor-investing – Factor models provide systematic idea frameworks

#flashcards/stock-market

What is strategy idea generation? :: The structured process of formulating testable hypotheses about market behavior that can be exploited for profit, transforming observations into concrete trading rules.

What are the three components of a complete hypothesis?

Signal (what predicts?), mechanism (why?), and regime (when?).

Why is understanding the "why" behind a strategy critical?

Because markets adapt—knowing the causal mechanism helps you recognize when the edge will decay and prevents overfitting to noise.

What is post-earnings announcement drift (PEAD)?

The tendency for stocks that beat earnings estimates to continue outperforming for 1-2 months due to anchoring bias and slow institutional adjustment.

In a pairs trading strategy, what does cointegration mean?

Two asset prices move together long-term (their spread is mean-reverting), even if they diverge short-term, allowing profitable bets on convergence.

What is the difference between a rational and behavioral edge?

Rational edges are risk-based (compensated by higher returns) and decay as capital flows in; behavioral edges exploit persistent investor biases and are more durable.

What is the z-score formula for pairs trading signals?

z = (Spread_t - μ_Spread) / σ_Spread, where entry typically occurs at |z| > 2 and exit at z = 0.

Why is data mining without economic theory dangerous?

It leads to spurious correlations (random chance in sample) with no causal mechanism, resulting in strategies that fail out-of-sample.

What is the momentum anomaly?

Stocks that outperformed in the past 3-12 months tend to continue outperforming, likely due to information diffusion being slow and herding behavior.

What is the capacity constraint in strategy design?

The maximum amount of capital a strategy can deploy before transaction costs and market impact erode profitability to zero.

What are the five key sources of strategy ideas?

Academic research, market microstructure, domain expertise, event studies, and data mining (with caution).

In the mean reversion example, why use VWAP as the benchmark?

VWAP represents the volume-weighted "fair value" over a recent period; deviations indicate temporary liquidity imbalances rather than informed moves.

What is the S.I.M.P.L.E. mnemonic for idea generation?

Source, Inefficiency, Mechanism, Parameters, Limitations, Evidence—a checklist for complete strategy formulation.

Why do momentum strategies fail in crash regimes?

In crashes, markets become mean-reverting (panic selling followed by relief rallies), violating the trending assumption momentum depends on.

What is alpha in the factor model equation?

The idiosyncratic return (the part not explained by known factors)—positive alpha represents your unique edge or skill.

Concept Map

Hinglish (regional understanding)

Intuition Hinglish mein samjho

Strategy idea generation matlab hai systematic tareka se trading ke liye profitable patterns dhoondhna. Dekho, market mein lakhs log trade karte hain, sab same information dekhte hain—Bloomberg, news charts. Toh profit kaise kamaye? Answer hai: tum ko aisi edge chahiye jo dusron ke pas nahi hai. Edge ka matlab hai information advantage ya better analysis.

Pehle observation karo—jaise academic research se pata chala ki momentum kaam karta hai (jo stocks pichle 6 mahine mein chade, woh age bhi badte hain). Phir hypothesis banao: kyun kaam karta hai? Kyunki log news ko slowly process karte hain, aur herding behavior hai (jab ek stock chalti hai toh sab jump karte hain). Phir concrete rules banao: "Agar stock 6-month return 20% se zyada hai, toh buy karo, 1 mahine bad sell karo." Yeh specific rule hai—no guessing, no emotions.

Sabse important chez hai falsifiability—matlab, agar galat ho toh pata chal jaye. Backtest karo (purane data pe test karo), dekhlo ki 2008 crash mein toh nahi fail hua, aur different markets mein bhi kaam karta hai ya nahi. Agar sirf ek time period mein kaam kare (lucky accident), toh strategy nahi, fluke hai.

Last mein yad rakho: 99% ideas fail karte hain. Isliye tumhe ek pipeline chahiye—har week5-10 naye ideas generate karo, test karo, reject karo joaam na karein. Survival of the fittest. Jo strategy bachti hai woh tumhare liye alpha (extra profit) generate karegi. Idea generation matlab hai creativity + discipline ka combo—imagine karo opportunities aur phir ruthlessly validate karo. Yeh quant trading ka sabse creative part hai!

Test yourself — Algorithmic & Quant Trading