Alpha Factor Discovery

Team: Pranav Revuri & Krish Shah Track: FPD FinTech – Junior Analyst Project

---

What It Does

analyze.py runs a complete alpha-factor discovery pipeline:

1. Downloads 1 year of OHLCV price data via yfinance
2. Computes 5 technical indicators manually (no ta-lib):
   • RSI (14-day)
   • MACD (12/26/9)
   • SMA (20-day)
   • EMA (20-day)
   • Bollinger Bands (20-day, ±2σ)
   • Volume Ratio (today's volume / 20-day avg)
3. Trains a RandomForest to predict whether the price goes up or down 5 days later
4. Prints a full indicator summary table, model accuracy, and ranked feature importances
5. Generates charts saved as PNG files

---

Install

pip install -r requirements.txt

Required packages: yfinance, pandas, matplotlib, scikit-learn, numpy

---

Run

Single Ticker

python analyze.py AAPL

Output:

• Indicator summary table (last 10 rows)
• Model test accuracy and ranked feature importances
• AAPL_analysis.png chart file

Two Tickers (Comparison Mode)

python analyze.py AAPL MSFT

Output:

• Full analysis for both tickers
• Side-by-side indicator comparison table
• Model comparison (accuracy + top feature)
• comparison_rsi.png and comparison_importances.png

---

File Structure

alpha-factor/
├── analyze.py      # CLI entry point
├── data.py         # yfinance download & cleaning
├── indicators.py   # RSI, MACD, SMA, EMA, BB, VolumeRatio
├── model.py        # RandomForest training & prediction
├── charts.py       # Matplotlib visualisation
├── compare.py      # Side-by-side comparison logic
├── requirements.txt
└── README.md

---

Indicators

Indicator	Type	Description
RSI	Momentum	14-day avg gain/loss ratio — overbought >70, oversold <30
MACD	Trend	12-day EMA − 26-day EMA; signal line = 9-day EMA of MACD
SMA	Trend	Simple 20-day moving average
EMA	Trend	Exponential 20-day moving average (recent days weighted more)
Bollinger Bands	Volatility	20-day SMA ± 2 standard deviations
Volume Ratio	Volume	Today's volume ÷ 20-day average volume

---

Model

• Algorithm: RandomForestClassifier (100 trees, random_state=42)
• Target: 1 if price 5 days in the future > today's close; 0 otherwise
• Split: Chronological — first 80% train, last 20% test (no random shuffle)
• Features: All 5 indicators (RSI, MACD, Signal, Histogram, SMA, EMA, BB_Upper, BB_Middle, BB_Lower, VolumeRatio)