Multivariate sequential data generator with long term, non-linear dependency

Financial data are known to be generated from complex distributions, often assumed to be changing over time. The aim of this project is to build a simulator of multivariate time series, parameterized as simply as possible by a user, with the freedom to build different models, to assess their general behavior and key properties. This will facilitate predictions of multiple sequences and their interactions, through a simple interface of possible configurations to the user, who could have a basic knowledge of statistics and limited knowledge of the mathematical details of the underlying models. More generally, the user will have to be able to define certain variables’ characteristics, as well as dependencies between variables and time dependencies. Standard econometric models will be used as well as recent machine learning techniques to introduce non-linearities and time-dependency over a large period of time.

Faculty Supervisor:

Fabian Bastin

Student:

Partner:

Caisse de dépôt et placement du Québec

Discipline:

Computer science

Sector:

Finance and Insurance

University:

Université de Montréal

Program: