Abissi DIBI

As part of the course Machine Learning for Portfolio Management taught by M. Sylvain Champonnois, I developed a research project exploring new hypotheses on the predictability of the VIX. The central question was: Can we forecast VIX movements using machine learning techniques?

To test this, I built predictive models and backtested them to validate their performance. The dataset combined both traditional and alternative sources of information, including:

Macroeconomic factors

Interest rate variables

Sentiment analysis indicators

This multi-faceted data approach aimed to capture a richer view of market dynamics and improve predictive accuracy. The full implementation and results of this project are available on my GitHub repository.

As part of the Enseignement d'Approfondissement project at École Polytechnique, under the supervision of M. Claudio Fontana, I developed and numerically simulated two variants of a pairs trading strategy, analyzing their performance both in the presence and absence of regime shifts.

The spread between assets was modeled using an Ornstein–Uhlenbeck process, and the simulations focused on evaluating key performance metrics such as:

• PnL (Profit and Loss)

• Sharpe Ratio

• Maximum Drawdown

This work provided insights into how regime changes affect the robustness and risk-return profile of pairs trading strategies.

The full implementation and results of this project are available on my GitHub repository.

As part of the Enseignement d’Approfondissement course at École Polytechnique, under the supervision of M. Mathieu Rosenbaum, I studied and replicated the seminal work of Almgren and Chriss, Optimal Execution of Portfolio Transactions with Market Impact.

The project focused on analyzing the theoretical framework of optimal trade execution in the presence of both temporary and permanent market impact, and implementing numerical simulations to replicate and validate the results of the article.

The full implementation and results of this project are available on my GitHub repository.

As part of the Enseignement d’Approfondissement course at École Polytechnique, I studied academic papers addressing the estimation of realized volatility in the presence of market microstructure noise. In particular, I focused on the seminal article “A Tale of Two Time Scales: Determining Integrated Volatility with Noisy High-Frequency Data” by Lan Zhang, Per A. Mykland, and Yacine Aït-Sahalia.

The project involved implementing and testing the convergence of the estimators proposed in the paper, assessing their performance in realistic high-frequency data settings.

VOLATILITY ESTIMATION UNDER ENDOGENOUS
MICROSTRUCTURE NOISE
By Christian Y. Robert and Mathieu Rosenbaum