The 16th Quantitative Finance Conference

The 16th Quantitative Finance Conference

Revisit The 16th Quantitative Finance Conference, originally presented: Monday 16th November – Friday 20th November 2020. 


The Importance of Soft Skills in a Quantitative World
by Jeff Scott: Founder and CEO, Section 810 Communications, LLC,

Quantitative finance often requires substantial “hard skills”–programming, machine learning, data science, AI.  But what about the “soft skills” required to collaborate with others, obtain buy-in for ideas, and lead cross-functional teams?

This fast-paced workshop will highlight the importance of these skills within the industry.  Attendees will learn:

  • What the DISC personality assessment tool is and how can it improve self-awareness
  • Ways to improve communication skills to increase your influence
  • How to obtain support for your ideas
  • Steps to build effective business relationships
  • How to work with “difficult” people

Quantum Computing and Quantum Machine Learning: Quant Finance Perspective
by Alexei Kondratyev: Managing Director, Head of Data Analytics, Standard Chartered Bank

  • Computation: Classical versus quantum logic gates
  • Why quantum computing is more powerful
  • Quantum Neural Network (QNN)
  • Born Machine and Boltzmann Machine
  • Quantum machine learning in practice

Market Generators for Long Time Horizons
by Alexander Sokol: Executive Chairman and Head of Quant Research, CompatibL

Market generators are machine learning algorithms for generating realistic samples of market data when historical time series has insufficient length or gaps. While most of the recent research on market generators focused on daily time horizons, the problem of generating realistic market data samples for time horizons from 1 year to 30 years and longer has multiple applications including limit management, insurance (economic scenario generation) and macro investing.

In this presentation, we describe a family of market generators that use machine learning to generate market scenarios with accurate probability distribution over long time horizons from limited time series.

Tackling Nonlinear High Dimensional Problems in Finance with Low Rank Tensor Techniques and Deep Neural Networks
by Kathrin Glau: Lecturer in Financial Mathematics, Queen Mary University of London


Accelerate your AI Modelling with IPUs
by Alexander Tsyplikhin: Senior AI Engineer, Graphcore

In the finance sector, the potential for innovation with advanced machine intelligence is significant. But often, new and complex models are not being fully leveraged due to latency issues and compute restraints. Enter the IPU – a completely new processing architecture designed for machine intelligence, capable of running advanced financial models up to 26x faster. Graphcore’s Alex Tsyplikhin explains how the IPU’s unique architecture can power such incredible breakthroughs – and what this means for the future of finance and trading.

What you’ll learn:

  • How the IPU is able to achieve faster financial model accelerations than other hardware available on the market
  • How to use IPUs for financial modelling training and inference
  • Insights into advanced models, use cases and IPU benchmarks


Algorithmic Differentiation for Machine Learning. Beyond Backpropagation
by Uwe Naumann: Professor of Computer Science, RWTH Aachen University

  • Neural networks as surrogate models
  • Reduction of the size of trained neural networks through pruning
    • a) Interval adjoint significance analysis
    • b) Results
  • Reduction of the cost of differentiation of neural networks
    • a) Generalized Jacobian matrix chain products
    • b) Results
  • Outlook
    • a) AD mission planning
    • b) Adaptive sampling and adaptive surrogates

Deep Pricing Theory and Practice by Youssef Elouerkhaoui: Managing Director, Global Head of Credit and Commodities Quantitative Analysis, Citi


Chebyshev Tenors + Machine Learning: applications to Dynamic Sensitivities, Counterparty Risk on exotics and FRTB
by Ignacio Ruiz: Founder & CEO & Mariano Zeron: Head of Research and Development, MoCaX Intelligence

The computation of risk metrics poses a huge computational challenge to banks. Many different techniques have been proposed in the last few years to try and tackle the problem. Chebyshev Tensors are a very powerful function approximator, but suffer from the curse of dimensionality, hence their use, being useful in many cases, is still somewhat constrained.

In this presentation we show how by incorporating Machine Learning flavoured techniques to the building of Chebyshev Tensors, the curse of dimensionality can be side-stepped, and hence a substantial range of pricing and sensitivity functions can be approximated with high accuracy and fast computation, to accelerate risk calculations substantially. In this way several risk calculations can be enhanced by this method. We illustrate the power of this technique through numerical results obtained in the calculation of Dynamic Sensitivities and its immediate side product, Dynamic Initial Margin. We also touch on applications for Counterparty Risk on exotics and IMA-FRTB.

  • Introduction to the pricing problem in risk calculations
  • Mathematical properties of Chebyshev Tensors
  • Tensor Extension Algorithm: borrow ideas from Machine Learning to overcome the curse of dimensionality
  • Numerical examples
    • Dynamic Sensitivities
    • Dynamic Initial Margin
    • Counterparty Risk on exotics and IMA-FRTB

Deep Adjoint XVA
by Andrew Green: Managing Director and XVA Lead Quant, Scotiabank


Differential Machine Learning: the Shape of Things to Come
by Antoine Savine: Chief Quantitative Analyst, Danske Bank & Brian Norsk Huge: Chief Quantitative Analyst, Danske Markets

Brian Huge and Antoine Savine will present their paper ‘Differential Machine Learning: the Shape of Things to Come’ published in the October 2020 issue of Risk Magazine, where they combine automatic adjoint differentiation (or AAD) with modern machine learning to learn fast, accurate pricing and risk approximations for arbitrary Derivatives transactions or trading books, in the context of arbitrary stochastic models, effectively resolving computation bottlenecks of risk reports and capital regulations. The presentation explains the main ideas of the paper and concludes with a demo in a production environment.


Adding Optionality 
by Peter Carr: Professor and Dept. Chair of FRE Tandon, New York University


Risk-neutral pricing of European vanilla options amounts to a selection of the risk-neutral PDF of the underlying. When the price of the underlying security is real-valued, the most common choice of PDF is normal. We motivate the use of a logistic distribution as an alternative on both theoretical and empirical grounds. We provide supporting continuous-time and discrete-time martingales and use them to value some path-dependent options. In particular, we show how to price both an American call on a dividend-paying security and a naturally arising Bermudan-style option in closed form.

  • Why be normal?
  • Entropic motivations for logistic distribution
  • Supporting martingales
  • Closed-form pricing of American and Bermudan options

Local Volatility in Multi Dimensions 
by Jesper Andreasen (Kwant Daddy): Global Head Of Quantitative Research, Saxo Bank

  • Multi asset arbitrage
  • Minimal multi asset model and arbitrage
  • Discrete minimal model — pricing and calibration
  • Application to foreign exchange, equities and interest rates

Black Basket Analytics for Mid-Curves and Spread-Options
by Alexandre Antonov: Chief Analyst, Danske Bank


The Joint S&P 500/VIX Smile Calibration Puzzle Solved With Continuous Stochastic Volatility Models
by Julien Guyon: Senior Quant, Bloomberg L.P.


General Stochasic Volatility - New models, classical (PDE, MC) and modern numerical methods (NN)
by Jörg Kienitz: Partner, Quaternion Risk Management


In this talk we introduce General Stochastic Volatility models, especially new SABR/ZABR type models. We outline methods for approximating option prices using Finite Difference methods and/or approximation formulas. We place the research into the existing context and show numerical examples. Furthermore, we consider another approach based on neural networks. Pricing and calibration is considered. In particular we introduce the CV (control variates) method for our General SV models. Finally, we consider generative methods in this set-up.


KVA x 2
by Matthias Arnsdorf: Global Head of Counterparty Credit Risk Quantitative Research, J.P. Morgan

In this talk we try to answer two questions:

  • Should KVA be part of derivatives valuation? If so, what are its properties?
  • Does a change in capital requirements have an impact on shareholder value and hence lead to a KVA?

In exploring these questions we find:

  • There are two distinct KVA quantities that can be defined.
  • KVA1 is a result of market incompleteness and captures the value of unpriced risk. This should be considered integral to an assets value.
  • KVA1 is not sensitive to changes in capital levels but is risk based.
  • In many cases of interest, KVA1 has implicitly already been incorporated in valuation and no additional adjustment is required.
  • KVA2 is compensation for shareholder losses due to changes in a firms leverage. This represents a “Transfer of Wealth” between shareholders and creditors similar to FVA.
  • KVA2 is proportional to marginal changes in capital levels. However,  the effective cost of this capital is the firms junior funding rate and not the return-on-equity.

AI-Accelerated Derivatives Models - from R&D to Production
by Ryan Ferguson: Founder & CEO, Riskfuel


KVA Under IMM and Advanced Approaches
by Justin Chan: Head of Product Management, Risk, FIS

The two largest components of Capital Valuation Adjustment (KVA) are the costs of Counterparty Credit Risk (CCR) and CVA capital. For a bank using the most advanced capital models – Internal Models Method for CCR and the incoming SA-CVA capital –an accurate KVA involves forward simulating expected exposures (EE) over the lifetime of the portfolio – potentially a Monte Carlo in a Monte Carlo. We present a practical regression-based solution.

  • Simulating EE: from regulatory stressed real-world measure to market implied measure
  • A comparative study of regression vs brute force nested Monte Carlo
  • SA-CVA: extending from simulating forward EE to simulating forward CVA sensitivities

Frameworks for Model Risk Management of AI
by Jos Gheerardyn: Co-Founder and CEO,


  • Model risk components 
    • Overview of market practice
    • Technological evolutions
  • Adapting for AI
    • Typical ML model dependencies
    • Frameworks 
      • designing AI-safety 
      • assessment list for trustworthy AI
      • quantitative tests
    • Design considerations
    • Limitations

In the presentation webinar, Jos Gheerardyn will give an overview of the available frameworks to manage model risk of AI. He will discuss in-depth how to evolve existing frameworks and will as well review more recent initiatives such as the recently published assessment list for trustworthy AI. 

New HPC Paradigm for Object Oriented Languages
by Dmitri Goloubentsev: Head of Automatic Adjoint Differentiation, Matlogica & Mahesh Bhat: Principal Engineer, Intel Corporation

30 Minute Talk: In the talk we present the joint results of Intel and MatLogica for 2 key XVA benchmarks yielding x1000 performance improvements on XEON Scalable CPUs. We demonstrate how MatLogica library works, in simple terms, and provide an idea of integration complexity using our recent use cases with QuantLib and ORE.

Quantifi Leverages Intel® Hardware to Accelerate XVA Calculations
by Rama Krishna Nagamalla: Senior Software Developer, Quantifi & Mahesh Bhat: Principal Engineer, Intel Corporation

30 Minute Talk: Calculating XVAs requires Microservices Architecture using distributed computing  for achieving scalability, reliability and resilience. Loading data, transferring and saving results can introduce I/O issues causing significant degradation in overall system performance. This presentation explains how Quantifi, achieved end-to-end acceleration using Intel’s CPU and Optane™ Persistent Memory technology solutions.


Consistent Multivariate Modelling of Swaptions and CMS Derivatives
by Elias Daboussi: Quantitative Analyst, Bank of America Merrill Lynch

Aspects discussed:

  • From annuity measures to a common measure
  • Joint Simulation of Swap rates
  • Results and implications

Improve CVA Proxy Hedging Efficiency – Model Enhancements during a Volatile Time
by Shengyao Zhu: Senior Quantitative Analyst, XVA Trading Desk, Nordea


Interest Rates Benchmark Reform and Options Markets
by Vladimir Piterbarg: MD, Head of Quantitative Analytics and Quantitative Development, NatWest Markets

We examine the impact of interest rates benchmark reform and upcoming Libor transition on options markets. We address various modelling challenges the transition brings. We specifically focus on the impact of the clearing houses’ discounting switch on swaptions, and the consequences of Libor transition on Libor-in-arrears swaps, caps, and range accruals as typical representatives of a very wide range of Libor derivatives.

Local Gaussian Approximation for Modeling Collateralized Exposure
by Michael Pykhtin: Manager, Quantitative Risk, U.S. Federal Reserve Board