This repository contains the code and analysis for a comprehensive quantitative finance project focused on the Ethereum options market. The project demonstrates an end-to-end workflow: from ingesting and cleaning raw derivatives data to calibrating advanced financial models, identifying market anomalies, and quantifying model risk.
- Project Objective
- Data Sources
- Quantitative Methodology
- Results and Key Findings
- Conclusion
- How to Run
The goal of this project is to develop a robust framework for analyzing the Ethereum implied volatility (IV) surface to extract forward-looking risk signals and identify anomalous market structures. This is achieved through a multi-stage process that includes:
- Constructing a smooth, arbitrage-free IV surface from raw market data.
- Analyzing the term structures of volatility and skew to interpret market sentiment.
- Quantifying the Volatility Risk Premium (VRP) by comparing implied vs. realized volatility.
- Validating the model by analyzing option sensitivities (Greeks).
- Stress-testing a dynamic stochastic volatility model (Heston) against the observed market state.
The analysis relies on a rich, multimodal dataset:
- Derivatives Data: A high-frequency snapshot of the entire ETH options order book (
ETH_options_data-*.csv). - Historical Price Data: Daily OHLC data for ETH/USD (
ETH_day.csv) for calculating realized volatility. - Contextual Data: Time-series data for on-chain volume (
volume-24h.csv) and social media sentiment (reddit_posts_sentiment_formal.csv) were used in the initial exploratory phase to define the project's scope.
- SVI (Stochastic Volatility Inspired) Model: Used to parameterize and fit a smooth, arbitrage-free curve to the discrete IV quotes for each expiration. Calibration is performed via a bounded non-linear optimization (L-BFGS-B).
- Garman-Klass Volatility Estimator: A high-efficiency method used to calculate historical realized volatility from daily OHLC data.
- Heston Stochastic Volatility Model: A dynamic, SDE-based model calibrated to the market via Fourier inversion pricing and a global optimizer (SLSQP) to diagnose the market's underlying risk dynamics.
The SVI model was successfully calibrated to the raw market data, transforming noisy, discrete quotes into a continuous and theoretically sound surface.
Figure 1: SVI Smile Fit
The model demonstrates a high-fidelity fit to the market data for a single expiration, accurately capturing the "volatility smile" phenomenon.
Figure 2: 3D Implied Volatility Surface
By interpolating the fitted SVI parameters across all expirations, we construct the complete 3D implied volatility surface.
Deconstructing the 3D surface reveals key insights into the market's expectation of risk over time.
Figure 3: Term Structures of Volatility and Skew
- Volatility Backwardation (Left Plot): The term structure is steeply inverted, with short-term IV (>70%) far exceeding long-term IV (~64%). This is an anomalous market state signaling acute, near-term fear or uncertainty.
- Negative Skew (Right Plot): The SVI skew parameter (
ρ) is strongly negative, indicating high demand for downside protection (puts) and confirming a significant "crash fear" priced into the market.
The project's key finding comes from comparing the market's expectation of volatility (IV) with historical reality (RV). The analysis was contextualized to the period following the "Black Thursday" crash of 2020.
Figure 4: Volatility Risk Premium (VRP) Analysis
The analysis uncovered a profoundly negative Volatility Risk Premium averaging -42.26 percentage points. The realized volatility (green line) from the recent past was dramatically higher than the implied volatility (blue line) being priced for the future. This rare anomaly signifies a market regime of extreme complacency or disbelief, where options were trading at a deep discount to recently observed risk.
To validate the SVI model, its implied Vega was compared against the exchange's benchmark.
Figure 5: Vega Comparison
The results show a strong linear agreement, validating the SVI model's consistency. However, a slight, systematic bias was detected, quantitatively demonstrating the concept of model risk and its implications for practical hedging strategies.
A Heston stochastic volatility model was calibrated to the data. The optimizer pushed all key parameters to their absolute boundaries, and the final pricing error remained high. This served as a powerful diagnostic, concluding that the market state was so anomalous that it stressed the descriptive capacity of a standard dynamic volatility model.
This project successfully executed an end-to-end quantitative workflow, moving from raw data to sophisticated model calibration and conclusive, data-driven insights. It identified a multi-faceted anomaly in the Ethereum market, characterized by backwardation, negative skew, and a profoundly negative Volatility Risk Premium. The analysis confirms a rare market regime of post-crash complacency and demonstrates the limitations of standard financial models under stressed conditions.
- Environment Setup: The project was developed in a Paperspace environment with an A6000 GPU. Key libraries include
pandas,numpy,scipy, andmatplotlib. - Data: Place all dataset CSV files into the
/notebooks/datasets/directory. - Execution: The analysis is contained within a series of Jupyter Notebooks. Run the data preparation and
final_dfcreation cell first, followed by the main analysis cell which performs the SVI fitting and generates all results.