Finstreet: ML-Driven Financial Backtesting & Analysis System

An advanced Python-based framework for quantitative financial analysis, strategy backtesting, and machine learning-driven trading signal generation.

Project Overview

This project implements a production-style algorithmic trading pipeline that converts raw historical market data into actionable buy/sell signals, evaluates strategy robustness through realistic backtesting, and emphasizes risk management and reproducibility.

The system is built with a strong focus on:

• Methodological rigor
• Realistic trading constraints
• Explainable ML-driven decision making

Performance Summary

Metric	Value
Initial Capital	₹100,000
Final Capital	₹100,264
Total Return	₹264
Total Return (%)	0.26%
Sharpe Ratio	1.60
Max Drawdown (%)	0.39%
Win Rate (%)	45.8%
Total Trades	24
Winning Trades	11
Losing Trades	13
Profit Factor	1.40

Key Takeaways

• Low drawdown indicates strong risk management
• Profit factor > 1 confirms statistical edge
• Strategy prioritizes capital preservation over aggressive returns

---

End-to-End Trading Pipeline

[ Data Acquisition & Preprocessing ]
(CSV Data Loader, Cleaning, Validation)

↓

[ Feature Engineering & ML Modeling ]
(Indicators, Lags, Prediction)

↓

[ Signal Generation & Risk Management ]
(BUY / SELL / HOLD, Position Size, SL / TP)

↓

[ Strategy Execution & Backtesting ]
(Paper Trades, PnL Tracking)

↓

[ Performance Evaluation & Analysis ]
(Returns, Drawdown, Sharpe)

Features

Machine Learning Ensemble

• XGBoost + LightGBM ensemble model (30/70 weighted)
• Direct training on full dataset for maximum signal strength
• Probability-based signal generation with confidence thresholds

Technical Indicators

Category	Indicators
Momentum	RSI, MACD, MACD Histogram, Rate of Change (ROC)
Trend	SMA (10/20), EMA, ADX, DI+/DI-, Trend Score
Volatility	ATR, Bollinger Bands (BB%), Realized Volatility
Volume	OBV (On-Balance Volume), Volume Ratio, Volume Z-Score
Statistical	Z-Scores, Kaufman Efficiency Ratio (KER)
Pattern	Candlestick Body Analysis, Upper/Lower Wicks

Advanced Signal Features

• Momentum Confluence - Combines RSI, MACD, SMA, and DI signals
• Breakout Detection - 20-day high/low breakout signals
• Trend Scoring - Multi-factor trend strength assessment
• Triple-Barrier Labeling - Profit target, stop-loss, and time-based labels

Risk Management

• ATR-based dynamic stop-loss (1.5x ATR)
• ATR-based take-profit levels (3.0x ATR)
• Fixed position sizing (4% per trade)
• Confidence threshold filtering (>55%)

Backtesting Engine

• Chronological walk-forward simulation
• Commission and slippage modeling
• Equity curve and drawdown tracking
• Comprehensive performance metrics (Sharpe, Profit Factor, Win Rate)

---

Quick Start

Follow these steps to get Finstreet up and running on your local machine.

Prerequisites

• Python 3.9+ is required.
  • You can download it from python.org.
  • It's recommended to use a virtual environment.

Installation

1. Clone the repository

   git clone https://github.com/kingslayer35/finstreet_final.git
   cd finstreet_final

2. Create and activate a virtual environment (recommended)

   python -m venv venv
   # On Windows
   .\venv\Scripts\activate
   # On macOS/Linux
   source venv/bin/activate

3. Install dependencies

   pip install -r requirements.txt

4. Run the pipeline

   python main.py
   python charts.py
   python results.py

---

Results

Cumulative Returns

Trade Performance

Profit and Loss Distribution

Trade Execution

Trade Scatter Analysis

---

Forward Predictions (Jan 1-8, 2026)

Date	Signal	Direction	Confidence	Position Size	Stop Loss	Take Profit
2026-01-01	HOLD	UP	50.69%	0.00%	0.0	0.0
2026-01-02	HOLD	UP	50.69%	0.00%	0.0	0.0
2026-01-03	HOLD	UP	50.69%	0.00%	0.0	0.0
2026-01-06	HOLD	UP	50.69%	0.00%	0.0	0.0
2026-01-07	HOLD	UP	50.69%	0.00%	0.0	0.0
2026-01-08	HOLD	UP	50.69%	0.00%	0.0	0.0

---

Project Structure

finstreet_final/
├── main.py             # Primary entry point for the trading pipeline
├── config.py           # Centralized configuration settings
├── requirements.txt    # Python dependencies
│
├── ml_ensemble.py      # XGBoost + LightGBM ensemble model
├── indicators.py       # Technical analysis indicators (30+ features)
├── labels.py           # Triple-barrier labeling for ML targets
├── signals.py          # Trade signal generation with risk parameters
├── backtester.py       # Core backtesting engine
├── data_loader.py      # Data fetching and preprocessing
├── charts.py           # Visualization generation (6 chart types)
├── results.py          # Strategy results and metrics
│
├── data/               # Historical market data (CSV)
├── models/             # Saved ML models (.pkl)
└── output/             # Generated reports and charts
    └── figures/        # Visualization images

---

Evaluation Criteria Coverage

Criterion	Weight	Implementation
Strategy Performance	40%	Net P&L, Max Drawdown, Sharpe >1.5, Profit Factor
Predictive Signal Quality	20%	Directional accuracy, signal stability, confidence thresholds
Modeling & Code Quality	15%	Clean modular code, no data leakage, reproducible results
Feature Engineering	15%	30+ technical indicators, pattern recognition, confluence signals

---

License

This project is for educational and evaluation purposes.