Giuseppe Nuti
and
Lluís Antoni Jiménez Rugama
June 8th
Title: Applying Explainable Bayesian Decision Trees to Trading
Speaker: Giuseppe Nuti and Lluís Antoni Jiménez Rugama
Date/Time: Tuesday, 6/8, 7pm CEST (10am PDT, 1pm EDT)
Abstract: In many trading applications, we need to balance explainability alongside the ability to expressly managing the statistical significance of the fitted parameters. As such, we have developed a Bayesian Decision Tree that is fully explainable and allows the user to actively input the a priori noise level expected in the data — trading data generally exhibits a very high noise to signal ratio, making techniques with a large number of parameters often impractical. After introducing the technique from a theoretical standpoint, we illustrate the suitability of the Bayesian Decision Tree via two typical trading applications: (a) estimating the probability of winning a Request For Quote (RFQ); and (b) optimizing a mean-reverting trading strategy.
Bio: Giuseppe Nuti heads a team focused on leveraging Machine Learning & AI for algorithmic execution and trading. The team works on a range of problems: from recommendation engines to optimal execution on behalf of UBS’s clients. The objective is to explore novel uses of ML in applications such as recommendations systems to match clients with specific UBS’s trading axes, anomaly detection for high-throughput, noisy systems, and optimal execution where venue micro-structure morphs order quality.
Prior to this role, Giuseppe was an algorithmic trader at UBS – New York – specialized in fixed income and foreign exchange. He has worked as a trader for over twenty years, initially in the interest-rates options and swaps market and in the European and US Government bond markets. He has experience working both within the primary dealer community and in the high-frequency environment (at KCG and Citadel.) At UBS, he has run the U.S. Rates Trading desk within FRC, with particular focus on electronic market-making.
Giuseppe holds a Ph.D. in Computer Science with particular focus on Markov Decision Processes applied to finance from University College London and an MSc in financial mathematics from City University, London. He is an adjunct professor at Cornell where he teaches a course on ML applied to trading in FX, Rates & Crypto. Previously, he has taught various courses, including Financial Computing at UCL and has supervised a number of Ph.D. students – both at UCL and CASS Business School. His research interests are in algorithmic trading and Bayesian formulation of standard Machine Learning techniques.
Bio: Lluís Antoni Jiménez Rugama is an algorithmic trading Director at UBS, working for the FX Central Risk Book and Equities Algo-execution teams. His main focus is to understand the market microstructure, generate alpha, and design new machine learning tools tailored to high frequency trading. During this time, he has also authored a manuscript about Bayesian decision trees published in the journal Frontiers in Applied Mathematics and Statistics and presented his machine learning research at the PyData Chicago Seminar.
Lluís Antoni holds a PhD in Applied Mathematics from Illinois Institute of Technology where he researched quasi-Monte Carlo methods for efficient high dimensional and infinite dimensional integration. During this time, he also collaborated with researchers at Fermilab to help measure the top quark mass. He published at several peer-reviewed journals and proceedings including the biannual Monte Carlo and Quasi-Monte Carlo conference. He also reviewed articles for the Journal of Complexity and Ian Sloan's 80th birthday festschrift. Prior to his PhD, he completed a master's degree in Financial Engineering at the Ecole des Ponts ParisTech, and two bachelor degrees in Mathematics and Civil Engineering from the Universitat Politècnica de Catalunya.