Machine Learning and Data Sciences for Financial Markets (10am-noon)

The last 10 years have seen an increase of the use of machine learning and of new kinds of data in financial markets. Following the book co-edited with Agostino Capponi (Machine Learning and Data Sciences for Financial Markets), I will start by explaining how these techniques and new sources of information can improve the traditional "quantitative finance" approaches.

After a general introduction, the session will be structured in 3 main parts: 

- customisation of services to users of the financial ecosystem; this will cover Robo-advisors and recommendation systems

- improvement of risk management; including off-line reinforcement learning to solve HJBs

- better connection with the real economy, that will introduce the use of alternative data (texts, satellite images, geolocation, etc).

Brainstorming: Machine Learning and Data Sciences in financial markets (2pm-4pm)

The discussion will be open on related topics:

• Why AI is a General Purpose Technology? and what are the consequences?
• The secondary innovations of AI that should be useful to market participants and users of the financial system: what can of problems can be improved using AI?
• Ethics of AI for Financial Markets (Inconclusive evidence; Inscrutable evidence; Misguided evidence; Unfair outcome; Transformative effects; Traceability) and the implications in terms of financial stability.

Reinforcement Learning in Financial Markets: from general considerations to market microstructure. (10am-noon)

Reinforcement Learning (RL) recently got some traction following the success in sandbox games like Go. I will start by reminding the basics of RL, and especially Q-learning, with a focus on its apparent similarity with stochastic control (and Hamilton-Jacobi-Bellman equations) and on the way it can be seen as a special case of stochastic algorithms.

I will then focus on the example of market microstructure because a lot of data are available in this domain.

Starting with the apparent opposition between price formation and price discovery, I will navigate through the data to extract stylised facts on price dynamics and the way buying or selling follows apparently modify them. I will try to reconcile orderbook time scales with the time scale of metaorders. I then be guided by the question: what are the price moves really caused by the trading flow of an isolated market participant?

After collecting all these evidences, I will setup different stochastic control frameworks to optimise execution and show how to solve them.

Last but not least, in an attempt to come back to the stylised facts, I will detail a Mean Filed Game approaches to model liquidity and show how to optimally interact in such a context.

Brainstorming: Reinforcement Learning. (June 12, 2pm-4pm)

The discussion will be opened on RL in general, and its potential applications in financial markets.

For instance

• Off-line vs On-line reinforcement learning: when to be accurate?
• The true outcome of RL in finance: average value of a decision or one controlled trajectory?
• The role of synthetic data in RL: what can we expect?
• The impact of dimension.

Using Alternative data for financial decisions (10am-noon)

During this session I will focus on alternative data. I will first try to classify them a way that makes sense for users, before underlying the kind of modern techniques that can be used to exploit them. That for, I will make the difference between recorded data and the information they contains, in an attempt to follow the way information propagates. In other terms: I will try to develop a causal reasoning and it apply to the way alternative data deliver information on the health of the economy.

I will then focus on three examples : satellite images, texts and shipping. And I will conclude on biases of data, coming back to some traditional questions of ethics of AI.

References: 

• Part III of Capponi, Agostino, and Charles-Albert Lehalle, eds. Machine Learning and Data Sciences for Financial Markets: A Guide to Contemporary Practices. Cambridge University Press, 2023.
• Li, Mengda, and Charles-Albert Lehalle. "Mathematics of Embeddings: Spillover of Polarities over Financial Texts." Reviews In Modern Quantitative Finance (2024): 151-188.
• Kearns, Michael, and Aaron Roth. The ethical algorithm: The science of socially aware algorithm design. Oxford University Press, 2019.

Brainstorming: Biases and causality. (2pm-4pm)

This last brainstorming will be open on two topics.

• How to identify biases of datasets for financial decisions?
• What can prevent us to correct a dataset from biases?
• Statistical procedures to identify causality in a dataset: stationarity, heteroskedasticity and covariate shift.
• Machine Learning and causality: is there a hope?