Market Rally Prediction via Volatility Analysis

A practical approach to analysing market volatility using ensemble methods. This project helps spot potential market rallies by examining the relationship between realised and implied volatility.

Contents

• Overview
• Key Components
• Getting Started
• Under the Bonnet
• Quick Start
• Maths Behind the Magic

Overview

We've built this tool to combine several volatility measures into one robust indicator. By comparing our calculated realised volatility against market-implied volatility, we can spot potential market moves before they happen.

Key Components

Market Data Sources

• SPX: The S&P 500 Index (our main market benchmark)
• VIX: The CBOE Volatility Index (the market's "fear gauge")

We fetch this data through Finnhub's API, with built-in rate limiting and local caching to keep things running smoothly.

Volatility Calculations

We use several tried-and-tested volatility estimators:

• Garman-Klass
• Rogers-Satchell
• Parkinson
• Close-to-Close
• Yang-Zhang

Smart Weighting

Our system uses machine learning to work out the optimal mix of these estimators:

• Linear Regression
• Ridge Regression (for more stable results)
• LASSO (when we need to be selective)

Getting Started

Pop this in your terminal:

pip install -r requirements.txt

You'll need:

• An Alpha Vantage API key (set it as Finnhub_API_KEY in your environment)

Under the Bonnet

Data Handling

1. Live market data fetching (mindful of API limits)
2. Smart local caching
3. Rolling 21-day calculations
4. Automated quality checks

The Core Calculation

Here's how we calculate realised volatility:

$$ RV = \sum_{i=1}^{k} a_i \sigma_i $$

Where:

• $RV$ is our Realised Volatility
• $a_i$ are our clever ML-derived weights
• $\sigma_i$ are the individual volatility measures
• $k$ is how many measures we're using

Volatility Premium

$$ Volatility\ Premium = RV - IV $$

• RV = Our calculated Realised Volatility
• IV = Market-implied Volatility (from VIX)

Quick Start

Here's a simple example:

from volatility_system import DataFetcher, VolatilityEstimator

# Set things up
fetcher = DataFetcher()
estimator = VolatilityEstimator()

# Get our market data
spx_data = fetcher.fetch_data("SPX")
vix_data = fetcher.fetch_data("VIX")

# Work out the premium
vol_premium = estimator.calculate_premium(spx_data, vix_data)

Reading the Signals

1. When RV < IV (Negative Premium)

   • The market might be too worried
   • Often seen near market bottoms
   • Could be time to buy

2. When RV > IV (Positive Premium)

   • The market might be too complacent
   • Usually spotted near market tops
   • Worth being cautious

Maths Behind the Magic

Our Volatility Measures

1. Garman-Klass

   • Uses the day's full price range
   • Sharper than just looking at closes

2. Rogers-Satchell

   • Handles trending markets well
   • Stays accurate when prices drift

3. Parkinson

   • Focuses on price ranges
   • Brilliant for choppy markets

The Optimisation Bit

We fine-tune our weights by solving:

$$ \min_{a_i} \sum_{t=1}^{T} (RV_t - \sum_{i=1}^{k} a_i \sigma_{i,t})^2 $$

With these rules:

• All weights must add up to 1
• No negative weights allowed

Want to Help?

Found a bug? Got an idea? Open an issue or send us a pull request.

Licence

MIT