Offer title : Postdoc M/W on the analysis and modeling of agent behavior in financial markets using neural networks or Hawkes processes (H/F)
Reference : UMR7534-EMMBAC-005
Number of position : 1
Workplace : PARIS 16
Date of publication : 17 July 2023
Type of Contract : FTC Scientist
Contract Period : 12 months
Expected date of employment : 1 October 2023
Proportion of work : Full time
Remuneration : Bewteen 2505 and 4000 euros gross per month
Desired level of education : Niveau 8 – (Doctorat)
Experience required : Indifferent
Section(s) CN : Science and Data
New mathematical techniques from process statistics or more generally from machine learning (or even deep learning) are gradually gaining ground in quantitative finance. The applications are very numerous and often respond to essential issues.
As part of this position, we propose to analyze and model the behavior of the various agents involved in the same financial market as well as their
interaction dynamics. To do this, we are fortunate to have access to an incredibly rich database. This is a data set provided by the company Euronext (which owns the main market of the euro zone) which includes a complete history over several years of all the orders sent on the order book of each of the 40 shares of the CAC40 as well as all the orders sent to the order book for future contracts on the CAC40 index itself. The richness of this database comes not only from the quantity of data it contains (nearly 1 TB) but, in fact and above all, from the fact that each order is labeled by an identification number (anonymized) of the agent which sent it. Thus the behavior of each agent can be replayed historically and therefore studied.
As part of this position, we therefore propose to analyze and model the behavior of the various agents involved in the same financial market as well as their
interaction dynamics. Two approaches can be followed depending on the profile of the candidate:
The first would be based on the use of Hawkes process while the second would be a form of generalization that incorporates strong non-linearity using a deep learning type formalism with auto supervision (with, for example, “transformers” type structures).
– Getting started with the Euronext database
– Literature review on the subject of agent behavior analysis from high frequency data
– Review of the use of the latest advances in statistical finance on Hawkes processes or on neural networks (self-supervised, transformer type or other)
– Implementation of algorithms following one of the 2 approaches
– Publication of a research paper on the subject
– Presentation of the results to the medical and finance/statistics/AI/machine learning communities
– Solid scientific expertise in statistics, data science, ML/AI (PhD degree level)
– Expertise in data engineering
– Programming expertise
The work will be done in Emmanuel Bacry’s team at Paris-Dauphine University in Paris. It will be done in close collaboration with J.F.Muzy, expert in statistical finance (DR CNRS, UMR SPE CNRS – University of Corsica) and other experts (business or scientific) in our network depending on the approaches studied.
Constraints and risks