Understand algorithmic trading fundamentals — Hinglish notes

6.1.1 · Stock-Market › Algorithmic & Quant Trading

What is Algorithmic Trading?

Ye kyun exist karta hai: Human traders ko teen fundamental limitations face karni padti hain:

Speed: Milliseconds matter karte hain jab opportunities aati aur jaati hain
Consistency: Emotions optimal strategy se deviation karate hain
Scale: Hazaaron securities ko simultaneously manually monitor karna impossible hai

Ye kaise kaam karta hai: Aap ek trading hypothesis ko precise rules mein translate karte ho → in rules ko code karte ho → program markets monitor karta hai → jab conditions match hoti hain, wo automatically trades execute karta hai.

The Three Pillars of Algo Trading

1. Signal Generation (The "Why Trade" Decision)

Kya hai: Woh logic jo trading opportunities identify karta hai.

Kaise banta hai: Aap conditions define karte ho jo true honi chahiye:

Technical signals: Moving average crossover, RSI oversold/overbought, range se breakout
Fundamental signals: P/E ratio industry average se below, earnings surprise >10%
Statistical signals: Mean se 2 standard deviations ka price deviation (mean reversion)
Market microstructure: Order book imbalance, bid-ask spread anomalies

Ye kyun matter karta hai: Valid signal ke bina, aap sirf randomly trades execute kar rahe ho. Signal aapka edge hai — aapki hypothesis ki kaunsa price pattern future movement predict karta hai.

Worked example Signal Example: Moving Average Crossover

Strategy: Buy karo jab 50-day MA, 200-day MA ke upar cross kare (bullish signal), sell karo jab neeche cross kare.

Logic:

Short-term average > long-term average → recent momentum upward hai → trend bullish ho raha hai
Ye "the trend is your friend" ko systematically capture karta hai Ye step kyun?

50-day MA calculate karo: $M A_{50} = \frac{1}{50} \sum_{i = 0}^{49} P_{t - i}$ (last 50 closes ka average)
200-day MA calculate karo: $M A_{200} = \frac{1}{200} \sum_{i = 0}^{199} P_{t - i}$
Check karo ki $M A_{50, t} > M A_{200, t}$ AND $M A_{50, t - 1} \leq M A_{200, t - 1}$ (crossover abhi hua hai)
Agar true hai → BUY signal generate karo "Abhi hua" check kyun? Iske bina, aapko har us din BUY signal milta jab 50-day, 200-day ke upar hota. Aap sirf transition par act karna chahte ho.

2. Risk Management (The "How Much" Decision)

Kya hai: Woh rules jo position size determine karte hain aur kab exit karna hai (stop-loss, take-profit).

Ye critical kyun hai: Ek profitable signal bhi aapka account destroy kar sakta hai agar aap per trade bahut zyada risk karo. Risk management losing streaks ke dauran survival ensure karta hai.

Kaise: Aap parameters trade se pehle set karte ho:

$Position Size = \frac{Risk Capital per Trade}{Risk per Share}$

Jahan:

Risk Capital per Trade = Account Size × Risk % (typically account ka 1-2%)
Risk per Share = Entry Price - Stop Loss Price

Derivation:

Agar aapka account $100, 000 hai aur aap per trade 1% risk karte ho \to Risk Capital =$ 1,000
Agar aap $50 p a r b u y k a r t e h o w i t h s t o p - l oss$ 48 par → Risk per Share = $2
Position Size = $1, 000/$ 2 = 500 shares
Ye formula kyun? Ye guarantee karta hai ki stop out hone par, aap exactly $1, 000 l ose k a r t e h o (500 s ha r es \times$ 2 loss per share), jo aapka predetermined 1% risk hai.

3. Execution (The "How to Trade" Decision)

Kya hai: Actually orders place karne ka tarika jisse costs aur market impact minimize ho.

Kyun matter karta hai: Poor execution profits ko erode karta hai. $50 s t oc k p a r$ 0.10 ka spread 0.2% slippage hai — ye 50 baar karo aur aapne execution costs mein hi 10% lose kar diya.

Types of execution algorithms:

TWAP (Time-Weighted Average Price):
- Bade order ko equal chunks mein split karo aur time ke saath evenly spread karo
- Use case: Jab aap urgency signal nahi karna chahte aur time risk tolerate kar sakte ho
VWAP (Volume-Weighted Average Price):
- Order ko din bhar expected volume ke proportional split karo
- Jab volume high ho (liquid) toh aggressive, jab low ho toh passive
- Use case: Din ke average execution price ke against benchmark karo
Implementation Shortfall (Arrival Price):
- Shuru mein aggressively execute karo, phir slow down karo
- Position mein nahi hone ka opportunity cost minimize karo
- Use case: Jab strong signal ho aur immediate exposure chahiye

The Algo Trading Lifecycle

1. Research → 2. Backtest → 3. Paper Trade → 4. Live Trade (small) → 5. Scale 6. Monitor
     ↑                                                                              ↓
     └────────────────────── Feedback Loop ──────────────────────┘

1. Research (Hypothesis Formation):

Market pattern ya inefficiency observe karo
Testable hypothesis formulate karo: "Agar X hota hai, toh Y, Z% probability ke saath follow karta hai"

2. Backtest (Historical Validation):

Algorithm ko historical data par run karo
Check karo: Kya ye actually kaam karta hai? Win rate, Sharpe ratio, max drawdown kya hai?
Critical: Look-ahead bias (future info use karna) aur survivorship bias (delisted stocks ignore karna) se bacho

3. Paper Trade (Real-time Simulation):

Algorithm ko live data ke saath lekin fake money se run karo
Test karo: Kya ye live market conditions handle karta hai? Kya fills realistic hain?

4. Small Live Trade:

Minimal capital ke saath deploy karo
Execution, slippage, aur latency verify karo ki expectations se match ho

5. Scale Up:

Gradually position sizes badhao
Capacity constraints ke liye monitor karo (aapke orders market move karne lagte hain)

6. Continuous Monitoring:

Daily performance metrics track karo
Regime changes ke liye watch karo (algorithm kaam karna band kar deta hai)

Common mistake Common Mistake: Overfitting in Backtests

Mistake: Aapka backtest 80% win rate aur 3.0 Sharpe ratio dikhata hai. Aap live jaate ho aur immediately paise lose karne lagte ho.

Kyun sahi lagta hai: "Backtest prove karta hai ki kaam karta hai! Data jhooth nahi bolta."

Actually kya hua: Aapne historical data par curve-fitted kar liya. Aapke algorithm mein 15 parameters hain jo past noise ke liye perfectly tuned hain, signal ke liye nahi. Ye bilkul waisa hai jaise test answers memorize karna subject seekhne ki bajaye — naye questions par fail ho jaata hai.

Example:

Aap moving average periods ke 1000 variations test karte ho (5-day ke saath 23-day, 7-day ke saath 31-day, etc.)
Aap paate ho ki 13-day ke saath 47-day MA historically best kaam karta hai (pure chance)
Ye specific combination future mein kabhi repeat nahi hota kyunki ye noise tha Fix:

Out-of-sample testing: Data ko training (yahan backtest karo) aur validation (yahan test karo) mein split karo. Algorithm development ke dauran kabhi validation data nahi dekhta.
Walk-forward analysis: Time ke through aage badhte raho, periodically re-optimize karo, next unseen period par test karo.
Simplicity: Kam parameters = kam overfitting risk. 2-parameter system, 20-parameter se zyada robust hai.
Economic rationale: Kya aap explain kar sakte ho kyun edge exist karta hai data dekhe bina? Agar nahi, toh shayad spurious hai.

Steel-man the mistake: Parameters optimize karna smart lagta hai — aap best settings chahte ho. Problem ye hai ki "meaningful pattern ka best fit" aur "random noise ka best fit" mein distinguish karna mushkil hai. Out-of-sample validation ke bina, aap fark nahi bata sakte.

Common mistake Common Mistake: Ignoring Transaction Costs

Mistake: Strategy backtest mein $50,000 profit dikhati hai, lekin live trading mein paise lose hote hain.

Kyun sahi lagta hai: "Mera signal profitable hai, yehi matter karta hai."

Kya galat hua: Aapne din mein 200 baar trade kiya. Har trade ka cost:

Commission: $1 p er t r a d e \times 200 =$ 200/day
Spread: $0.05 p er s ha r e \times 100 s ha r es \times 200 =$ 1,000/day
Slippage: $0.02 p er s ha r e \times 100 s ha r es \times 200 =$ 400/day
Total: $1, 600/ d a y cos t s =$ 32,000/month

Aapka $50, 000 ba c k t es tp r o f i t z er ocos t s a ss u m e k a r t a t ha . R e a l i t y : -$ 32,000/month.

Fix: Backtest mein realistic costs include karo. Model:

Entry_price = Bid_price - Slippage
Exit_price = Ask_price + Slippage
Cost_per_trade = Commission + (Spread × Shares)

Rule of thumb: Agar profit per trade < 3× transaction costs, toh strategy viable nahi hai.

Key Performance Metrics

Formula Sharpe Ratio

$S ha r p e = \frac{R _{p} - R _{f}}{σ _{p}}$

Jahan:

$\overline{R_{p}}$ = average portfolio return
$R_{f}$ = risk-free rate (e.g., T-bill yield)
$σ_{p}$ = portfolio returns ka standard deviation

Iska matlab: Return per unit of risk. 2.0 ka Sharpe matlab hai aapne har 1% volatility ke liye 2% excess return earn kiya.

Ye formula kyun?

Aap excess return measure karna chahte ho (risk-free ke upar), total return nahi
Aap volatility ko penalize karna chahte ho — smooth 10% choppy 10% se better hai
σ se divide karne se normalize hota hai: 10% volatility ke saath 20% return (Sharpe=2), 20% volatility ke saath 30% return (Sharpe=1.5) se better hai Interpretation:

Sharpe < 1.0: Poor (risk adjust karne ke baad barely risk-free se better)
Sharpe 1.0-2.0: Good
Sharpe > 2.0: Excellent (practice mein rare, backtest mein hoga toh aksar overfitting ka sign hai)

Types of Algorithmic Trading Strategies

Trend Following:
- Assumption: Prices momentum exhibit karti hain — jo upar jaata hai, upar jaata rehta hai
- Signal: Moving average crossover, range se breakout
- Best kaam karta hai: Trending markets mein (post-breakout news, commodity cycles)
Mean Reversion:
- Assumption: Prices ek mean ke around oscillate karti hain — extreme moves reverse hote hain
- Signal: Price moving average se >2 standard deviations deviate ho jaaye
- Best kaam karta hai: Range-bound, choppy markets mein (low volatility regimes)
Statistical Arbitrage:
- Assumption: Related assets saath move karne chahiye — deviations temporary hote hain
- Signal: Cointegrated pairs diverge ho jaayein (e.g., Coke vs. Pepsi)
- Example: Agar Coke/Pepsi spread wide ho jaaye, Coke short karo, Pepsi long karo, spread narrow hone par profit lo
Market Making:
- Assumption: Aap liquidity provide karke bid-ask spread se profit karte ho
- Method: Dono sides par limit orders place karo, dono fill hone par spread earn karo
- Risk: Inventory risk (market aapke against move karne par aap long/short ho jaate ho)
High-Frequency Trading (HFT):
- Assumption: Speed advantage tiny inefficiencies capture karne deta hai
- Method: Servers ko exchange ke paas co-locate karo, microsecond latency ke liye optimize karo
- Retail traders ke liye accessible nahi hai (infrastructure mein millions chahiye)

Worked example Mean Reversion Strategy: Bollinger Band Bounce

Hypothesis: Jab price lower Bollinger Band touch kare, toh wo oversold hai aur bounce karegi.

Setup:

Middle Band = 20-day SMA
Upper Band = Middle + (2 × 20-day Std Dev)
Lower Band = Middle - (2 × 20-day Std Dev) Signal Logic:

IF Price touches Lower Band:
   AND RSI < 30 (confirming oversold)
   AND Volume > Average Volume (confirming interest)
THEN Buy
Exit:
   IF Price reaches Middle Band (take profit)
   OR Price breaks below Lower Band by >1% (stop loss—trend might be starting)

Har condition kyun?

Lower Band touch: Price mean se 2 std devs below hai (statistically extreme)
RSI < 30: Confirm karta hai ki momentum exhausted hai (oversold)
Volume > avg: Ensure karta hai ki low-liquidity outlier nahi hai (slippage reduce karta hai)
Middle par exit: Mean par return achieve ho gaya (50% retracement)
Lower ke neeche stop: Agar support break ho, mean reversion fail — loss cut karo

Expected performance: Chhote winners (mean reversion distance), kabhi kabhi bade losers (jab trend shuru ho). Win rate 60-70%, Risk:Reward ~1:1.

Recall Ek 12-Saal Ke Bachche Ko Samjhao

Socho tum ek video game khel rahe ho jahan tumhe girte hue coins pakadne hain. Pehle tum manually khelte ho — basket apne haath se move karte ho. Tum kaafi acche ho, lekin thak jaate ho, blink karte ho aur coins miss ho jaate hain, aur screen ka sirf ek hissa dekh sakte ho.

Ab socho tum rules ka ek set likhte ho: "Agar top-left mein coin aaye, basket wahaan le jao. Agar ek saath do coins girein, pehle bada pakdo. Agar lal coin aaye, ignore karo — ye fake hai."

Tum computer ko in exact rules follow karne ke liye program karte ho. Ab computer tumhare liye khelta hai, kabhi thakta nahi, kabhi blink nahi karta, aur poori screen ek saath dekh sakta hai. Wo tumse kahin zyada coins pakad leta hai.

Yehi algorithmic trading hai. Tum figure out karte ho ki stocks kab buy aur sell karne ke rules kya hain (ye "game"), unhein code likhte ho, aur computer ko automatically execute karne dete ho. Ye faster hai, more consistent hai, aur hazaaron stocks watch kar sakta hai jab tum so rahe ho. Tumhara kaam acche rules likhna hai (rules jo actually coins pakdein, random movements nahi).

Connections

6.1.02-Learn-basic-backtesting – Real paise risk karne se pehle algorithm validate kaise karein
6.1.03-Quantitative-analysis-indicators – Technical indicators jo signals ke roop mein use hote hain
6.2.01-Risk-management-in-algo-trading – Position sizing aur drawdown management
3.3.01-Types-of-market-orders – Execution algorithms ke liye order types samajhna
5.2.01-Technical-vs-fundamental-analysis – Signal types jo algorithms ko feed karte hain
4.1.02-Market-volatility-measurement – Volatility algo performance ko kaise affect karti hai

#flashcards/stock-market

Algorithmic trading kya hai? :: Computer programs ka use jisse predefined rules, mathematical models, ya statistical patterns ke basis par automatically trades execute ho sakein, bina execution ke dauran human intervention ke.

Har trading algorithm ko kaunse teen core decisions lene hote hain?

1) Signal Generation (kyun trade karein), 2) Risk Management (kitna), 3) Execution (kaise trade karein).

Moving average crossover signal kya hota hai?

Ek buy signal jo tab generate hota hai jab short-term MA, long-term MA ke upar cross kare (bullish), ya sell signal jab neeche cross kare (bearish). Jaise 50-day ka 200-day cross karna.

Algorithmic trading mein position size kaise calculate karte hain?

Position Size = (Risk Capital per Trade) / (Risk per Share), jahan Risk per Share = Entry Price - Stop Loss Price. Ye ensure karta hai ki stop out hone par aap ek predetermined % lose karo.

VWAP kya hai aur iska use kyun karte hain?

Volume-Weighted Average Price: (Price × Volume) ka sum total volume se divide kiya hua. Execution benchmark ke roop mein use hota hai ye ensure karne ke liye ki aap us din ke average trader se zyada nahi pay kar rahe.

Backtesting mein overfitting kya hoti hai?

Jab aapka algorithm historical noise fit karne ke liye tuned ho na ki real patterns ke liye. Backtest mein accha perform karta hai lekin live trading mein fail ho jaata hai kyunki usne past data memorize kar liya bajaaye generalizable rules seekhne ke.

Sharpe ratio kya hai?

(Average Return - Risk-Free Rate) / Standard Deviation. Ye risk per unit of return measure karta hai. Zyada better hai; >2.0 excellent hai.

Maximum drawdown (MDD) kya hai?

Ek period ke dauran account value mein sabse bada peak-to-trough decline. Recovery se pehle worst-case loss dikhata hai jo tumhe endure karni padegi.

Mean reversion kya hai?

Trading strategy jo is assumption par based hai ki prices ek mean value ke around oscillate karti hain. Jab price mean se bahut door deviate ho jaaye (e.g., >2 standard deviations), toh shayad wapas revert karegi.

Out-of-sample testing ka kya purpose hai?

Ye validate karne ke liye ki aapka algorithm un data par kaam karta hai jo usne development ke dauran kabhi nahi dekha, historical patterns par overfitting ko prevent karne ke liye.

Backtests mein transaction costs kyun include karne chahiye?

Kyunki commissions, spreads, aur slippage high-frequency strategies ke profits erase kar sakte hain. Ek backtest jo profit dikhata hai woh real costs ke baad actually paise lose kar sakta hai.

ATR ka matlab kya hai aur iska use kaise hota hai?

Average True Range: volatility measure karta hai. Stop-losses set karne ke liye use hota hai (e.g., Entry -2×ATR) jo normal price fluctuation account kare bina premature exit ke.