M2MO: Modélisation Aléatoire, Finance et Data Science

Master en statistique, probabilités et finance - Université Paris 7 - Paris Diderot


Quant analysis

Lecturer: S. Crépey
Period: Term 3
Hourly volume: 3 hours per week

This course deals with recent regulatory quant modeling issues and with a panoply of numerical optimization problems that are the bread and butter of quants’ activity.

Part I: XVA Analysis 

Since 2008, investment banks compute various X-valuation adjustments (XVAs) to assess counterparty risk and its funding and capital implications. XVAs deeply affect the derivative pricing task by making it global (portfolio-wide), nonlinear, and entity dependent. A proper financial understanding of even the first generation XVAs (CVA, DVA, and FVA, where C sits for credit, D for debt, and F for funding) points out to the distinction between firm and shareholder valuation, which mathematically goes to enlargement of filtration and singular probability change. Second generation XVAs involve not only conditional expectations (i.e. prices), but also conditional risk measures (value-at-risk and expected shortfall) for MVA (margin valuation adjustment) and KVA (capital valuation adjustment) computations. The course will provide an up-to-date covering of the XVA universe and of the embedded risk measure issues, from the triple angle of finance (wealth transfers), stochastic analysis (enlargement of filtration and backward SDEs), and computations (nested Monte Carlo and nonparametric regressions).   


0 Probabilistic credit risk prerequisites 

1 The XVA cost-of-capital approach in a static setup

2 The XVA cost-of-capital approach in continuous time

3 Dynamic default times modeling

4 XVA metrics for bilateral trade portfolios

5 XVA metrics for centrally cleared portfolios

6 XVA nested Monte Carlo computational strategies

7 XVA high-dimensional regression computational strategies

8 Uncertainty Quantification with XVAs in view 

Part II: Calibration, Training, and Other Optimization Problems

The implementation of approaches and models is conditioned by the solution of a panoply of numerical optimization tasks, such as: calibration of pricing models; training of supervised and unsupervised learning models for proxy pricing, pricing and hedging, or anomaly detection; collateral and capital compression schemes. 


0 Prerequisites in optimization: (possibly stochastic) gradient descents, handling constraints, derivative-free methods 

1 The ill-posed inverse calibration problem 

2 Calibration of the local volatility, old and new (Dupire and Gatheral formulas, SVI parameterization, Tikhonov regularization, Gaussian processes vs. neural net metamodeling)

3 Calibration of multi-curve interest rate models

4 Calibration of multi-name credit models

5 Proxy pricing by Gaussian process regression  

6 Pricing and hedging by neural network regression and quantile regression schemes  

7 Nowcasting (completion of  tensors such as interest rate curves, volatility surfaces,..)  and anomaly detection by auto-encoder compression / completion schemes

8 XVA compression by genetic algorithms

Prior knowledge:  Stochastic analysis, mathematical finance, numerical finance at MSc level.  Some knowledge of corporate finance or programming skills in python and/or C++  are also useful but can be acquired “on the job”.

Assessment:  Programming projects to be delivered as python jupyter notebooks running under Google Colab