The 8th London-Paris Bachelier Workshop
Synthetic Data Generation For Portfolios Optimisation: Traps and counter measures
The 8th London-Paris Bachelier Workshop will take place at IHP in Paris on 19 and 20 September 2024.
I will deliver a talk titled “Synthetic Data Generation For Portfolios Optimisation: Traps and counter measures” that is inpired from an on-going work with Rengim Cetingoz.
The abstract:
There are already a dozen of papers dedicated to generating synthetic data for portfolio optimization, ranging from risk-replicating methods for derivative pricing to those similar to Markowitz’s ones. The goal of this talk is to identify the main mathematical reasons of the complexity of this generative task : first the available sample is often small, and generating more data than what is available can provide a false sentiment of confort, second I will show why the standard distance used by GANs is not adequate for portfolio construction that involves hedging of large variance components (that is often the case). I will make my best to show where the memory should be located in the generate processes to provide meaningful results. Last but not least, I will discuss that generating more than one time series can be needed, even to estimate a L-statistic (like a VaR or ES) on a one dimensional index.
Finally, I will review the key features of the existing papers and present preliminary results from a work with Rengim Cetingoz on a new generative model designed to address many of these challenges.
More to come (including a paper) soon…
Back to top