Main Conference Day 1: Thursday 28th September
08.00 – 09.00: Registration and Morning Welcome Coffee
Morning Stream Chair:
Ioana Boier:
09.00 – 09.45: Training and Taming LLMs for Quant Finance
Ioana Boier:
09.45 – 10.30: An Attention-like Mechanism for Model-Agnostic Pricing of Path-Dependent Options
Blanka Horvath:
Associate Professor in Mathematical and Computational Finance, University of Oxford
Zacharia Issa:
Associate at BNY Mellon | PhD Candidate, King’s College London
11.00 – 11.45: Customizing Large Language Models for Quant Finance Applications
Due to the highly specialized subject matter, the leading commercial (GPT-4) and open source (Llama 2, Code Llama) LLMs are unable to provide specialized comprehension and generation suitable for quant finance applications out of the box.
In this presentation, I will describe fine-tuning and prompt engineering techniques that convert stock LLMs into specialized tools that can assist with the following model governance functions, subject to final sign-off by human analysts:
- Validation of trade capture using LLM comprehension of trade confirmations
- Validation of model documentation using LLM comprehension of source code
- LLM generation of model documentation and release note drafts
Alexander Sokol:
Executive Chairman and Head of Quant Research, CompatibL
11.45 – 12.30: Documenting Models and Algorithms.
Summary: As a quant, one of the most impactful ways of mitigating model risk is by creating accurate and consistent documentation. In the present talk, I discuss how qualitative and quantitative techniques can be used to enhance the quality and effectiveness of documentation.
Agenda:
- Why documentation matters
- Challenges
- Architecture of documentation
- Technology
- Use cases
Jos Gheerardyn:
Co-founder and CEO, Yields.io
Afternoon Stream Chair:
Andrey Chirikhin:
Head of Structured Credit QA, Barclays Investment Bank
13.45 – 14.30: Generative Models for Time Series in Finance
Miquel Noguer Alonso:
Co – Founder and Chief Science Officer, Artificial Intelligence Finance Institute – AIFI
14.30 – 15.15: Autoencoders and Arbitrage Free Dynamics
- The mathematics of neural networks: approximation theorems.
- Training neural networks: Adjoint differentiation and stochastic gradient decent.
- Making and training your own neural networks in C++.
- Auto encoders and yield curve dynamics: how many factors?
- Arbitrage free dynamics and supporting stochastic processes.
- Extensions to other asset classes.
Jesper Andreasen:
Kwantfather! Global Head of Quantitative Research, Saxo Bank
15.45 – 16.30: AEN-VAR-AEN
Description: We show how apply traditional VAR time series methodology to the now famous data set of the IR swap rates, and how to naturally extend it by applying autoencoders both as autoregressors and dimension reduction technique for the residuals in the regular VAR setting.
Andrey Chirikhin:
Head of Structured Credit QA, Barclays Investment Bank
16.30 – 17.15: Ethical and Responsible Large Language Models: Challenges and Best Practices
Nicole Königstein:
Chief Data Scientist, Head of AI & Quant Research, Wyden Capital AG
17.15 – 18.00: QFC Panel
Large Language Models (LLM) in Quant Finance – A Gimmick or a Game Changer?
- Will LLLMs help us read derivatives term sheets, UCITS documentation, and securitization rules
- Will LLLMs help us document our models for the regulators
- Will LLLMs help us find bugs in our pricing models
- GPT-4 or LLAMA 2?
ML Models for Valuation, XVA, and Risk
- How to build trustworthy ML quant models that auditors and regulators will approve
- Due to the lack of sufficient training data, is ML in quant models truly learning or only providing a better way to interpolate
Alexander Sokol:
Executive Chairman and Head of Quant Research, CompatibL
Ioana Boier:
Vladimir Piterbarg:
MD, Head of Quantitative Analytics and Quantitative Development, NatWest Markets
Ignacio Ruiz:
Peter Jaeckel:
Independent financial mathematics and analytics consultant. OTC Analytics
Jos Gheerardyn:
Co-founder and CEO, Yields.io
Gala Dinner: Marina Restaurante
The main building of the Beach Club is home to the Marina Restaurant, a circular restaurant with spacious windows overlooking Las Arenas Playa. The terrace extends towards the sea, a space full of charm for those who want to have a meal “al fresco” with amazing sea views.
Marina Restaurante
Calle Marina Real Juan Carlos I
46011 Valencia
Spain
Tel: +34 961 150 007
Website
Directions from SH Valencia Palace to Marina Restaurante.
Morning Stream Chair:
Andrew McClelland:
Director, Quantitative Research, Numerix
09.00 – 09.45: Looking Beyond SA-CCR
Michael Pykhtin:
Manager, Quantitative Risk, U.S. Federal Reserve Board
09.45 – 10.30: Intraday TAA meets Systematic Global Macro
Blaž Žličar:
Quantitative Portfolio Mgmt & Machine Learning Research, Vice President, Deutsche Bank
11.00 – 11.45: Comparing AAD Techniques & Performance
In this presentation, Stephan Bosch and Dmitri Goloubentsev will present a review of Bump & Revalue, Tape-Based AAD and Code Generation AAD approaches for CVA calculations and cover the following:
- Performance gains of introducing AAD, and scalability for large portfolios in terms of speed and memory use;
- Comparative performance of traditional tape-based vs. code generation AAD approaches for stress testing protocols and analysis;
- How adopting AAD to the evaluation of higher-order sensitivities can deliver linear scalability with respect to the number of cross-gamma entries.
Dmitri Goloubentsev:
CTO, Head of Automatic Adjoint Differentiation, Matlogica
Stephan Bosch:
Quantitative Developer, ING
11.45 – 12.30: Projecting Exposures and Margin: Getting Risk Models and Pricing Models to Play Nice
- The problem of projecting exposures and margin requirements off a pricing model (continuous-time, arbitrage free etc.) is well understood
- We have a range of numerical techniques (e.g. LSMC) to do this efficiently when closed-form representations are not available
- But what if we are simulating risk scenarios off a dedicated risk model (discrete-time PCA, GARCH etc.)?
- Translating between the risk model and the pricing model on a per-scenario basis is complex, good candidate for fitting the inverse map via ML?
Andrew McClelland:
Director, Quantitative Research, Numerix
Afternoon Stream Chair:
Peter Jaeckel:
Independent financial mathematics and analytics consultant. OTC Analytics
13.45 – 14.30: Learning volatility – now with mean reversion
Brian Norsk Huge:
Head of Quant, Saxo Bank
14.30 – 15.15: Efficient Simulation of FRTB CVA Capital Costs
Sascha Geier:
Director, Head of Counterparty Risk and xVA Analytics, Commerzbank AG
15.45 – 16.30: Algorithmic Adjoint Differentiation (AAD) for tail risk, model risk and stress testing
- AAD is currently used in quant finance for fast greeks and valuation adjustments calculations.
- It turns out that many other computationally intensive problems related to risk measurement, stress testing and model risk can benefit from AAD precision and speed.
- In this presentation, we develop several such business cases and show how AAD can contribute to tail risk estimation, model risk measurement and stress testing in various framework such as climate risk.
Svetlana Borovkova:
Head of Quantitative Modelling, Probability & Partners. Associate Prof, Vrije Universiteit Amsterdam
16.30 – 17.15: Advanced Numerical Methods for Option Pricing
Leif Andersen:
Global Co-Head Of Quantitative Strategies Group, Bank of America
17.15 – 18.00: QFC Panel
Large Language Models (LLM) in Quant Finance – A Gimmick or a Game Changer?
- Will LLLMs help us read derivatives term sheets, UCITS documentation, and securitization rules
- Will LLLMs help us document our models for the regulators
- Will LLLMs help us find bugs in our pricing models
- GPT-4 or LLAMA 2?
ML Models for Valuation, XVA, and Risk
- How to build trustworthy ML quant models that auditors and regulators will approve
- Due to the lack of sufficient training data, is ML in quant models truly learning or only providing a better way to interpolate
Alexander Sokol:
Executive Chairman and Head of Quant Research, CompatibL
Ioana Boier:
Vladimir Piterbarg:
MD, Head of Quantitative Analytics and Quantitative Development, NatWest Markets
Ignacio Ruiz:
Peter Jaeckel:
Independent financial mathematics and analytics consultant. OTC Analytics
Jos Gheerardyn:
Co-founder and CEO, Yields.io
Gala Dinner: Marina Restaurante
The main building of the Beach Club is home to the Marina Restaurant, a circular restaurant with spacious windows overlooking Las Arenas Playa. The terrace extends towards the sea, a space full of charm for those who want to have a meal “al fresco” with amazing sea views.
Marina Restaurante
Calle Marina Real Juan Carlos I
46011 Valencia
Spain
Tel: +34 961 150 007
Website
Directions from SH Valencia Palace to Marina Restaurante.
Morning Stream Chair:
Saeed Amen
09.00 – 09.45: Unlocking Efficiency in Risk Calculations through Machine Learning
The realm of risk calculations grapples with a recurring challenge – the portfolio revaluation dilemma.
From XVA and PFE to IMM, IMA-FRTB, VaR, and Stress-testing, the need for a high number of derivative valuations exacts a toll on both computational resources and operational speed.
Enter Machine Learning methods, a potent solution to this challenge. When harnessed effectively, Machine Learning methodologies hold the promise of significantly enhancing computational efficiency, all while maintaining accuracy.
In this presentation, Ignacio shares his insights and experience in this field. He will cover a spectrum of topics:
- Various categories of learning methods
- Navigating the intricate challenge of dimensionality
- Techniques for dimensionality reduction
- Unveiling the potency of Chebyshev Tensors
- Optimal use of Deep Neural Networks
- Tensors Train formats for optimal outcomes
- Chebyshev Sliders to fine-tune results
- Real-world numerical illustrations, including
-
- Vanilla Swaps
- Black-Scholes Options
- Barrier Options
- Bermuda Swaptions
- TARFs (Target Redemption Forwards)
- Autocallable Options
- Simulation of Sensitivities and SIMM (Standard Initial Margin Model)
Ignacio Ruiz:
09.45 – 10.30: Causal Discovery in Reinforcement Learning and its Applications in Finance
Topic and Presenter to be confirmed.
Leila Korbosli:
Quantitative Analyst, UBS
11.00 – 11.45: Forecasting Inflation with Machine Learning and alt data
Inflation has become a key topic in recent months. In our talk, we discuss how to approach the topic from a machine learning perspective, and also how to incorporate alternative data in the process. We will also discuss some specific use cases for inflation forecasting, in particular a systematic trading rule for FX.
Saeed Amen
11.45 – 12.30: An Inflation-Based Asset Allocation Strategy With Sentiment Data
Abstract: We develop a general approach to perform a multi-asset allocation strategy, which depends on inflation nowcasts enhanced with RavenPack sentiment analytics. Specifically, we build a systematic inflation hedging strategy that maximizes the multiperiod mean-variance criterion of the portfolio to invest in three financial assets (equity, commodity, bond), by exploiting inflation nowcasts (headline and core inflation). We test the robustness of the strategy via backtest, by investing in the US equity market, the US bond with 10 years maturity and six conventional commodity indexes. We also extend the proposed strategy to a volatility targeting approach, which searches for risk tolerance parameters of the risky assets given a risk preference of the investor in terms of volatility.
Anmar Al-Wakil:
Head of Quantitative Strategies, RavenPack
Afternoon Stream Chair:
Jason Charlesworth:
NAG and Founder of Zettamatics
13.45 – 14.30: Mathematical Finance in the Exascale Era
Historically quant analytics have largely ignored the hardware resulting in an algorithm Lottery. Many widely used algorithms and approaches are very poorly performing on modern hardware. Exacerbated by the rapidity of the banking business cycle opportunities for vastly faster quant analytics have been missed.
In this talk we will illustrate some of the problems and their solutions:
- Hardware specialisation: Superscalar/SIMD/GPU…
- AAD in calibration and risk
- Usable front-office Machine learning
Jason Charlesworth:
NAG and Founder of Zettamatics
14.30 – 15.15: Learning Market Data Anomalies
- Why anomaly data detection is crucial for market risk management
- Techniques: from simple statistics to isolation forest and (variational) autoencoders
- Comparing and reporting outputs
- Operational framework
Abstract
Everyday market risk managers are required to check a huge amount of market data, scenarios and positions used to compute market risk measures. Some data may present anomalous values because of a wide range of reasons, e.g. bugs in the related production processes, sudden and severe market movements, etc. Hence, it is crucial to integrate the daily data quality process with automatic and statistically robust tools able to smartly analyse all the available information and identify possible anomalies (the “needles in the haystack”). To this purpose we combined both simple statistics and machine learning algorithms to build an anomaly detection framework which is general enough to cover different asset classes and data dimensionalities (multiple curves, surfaces and cubes). The results are encouraging but they strongly depend on assumptions and parameters, keeping crucial the human supervision.
Marco Bianchetti:
Head of Internal Model Market Risk, Intesa Sanpaolo
Marco Scaringi:
Quantitative Analyst, Risk Management, Intesa Sanpaolo
15.45 – 16.30: Common Causal Manifolds for Portfolio Risk Management via Sensitivity Modelling with Market Data.
Mean Variance framework goal is a linear combination of assets (portfolio) with returns as a vector in a hyperplane in which portfolio constituents’ vectors returns, when projected I that hyperplane, have at best (optimal) opposite directions. In a way, a portfolio is a financial product in which constituents hedge each other up to a level (diversification). This setup while introducing the key concept of diversification has strong limitations known by any practitioner. Thinking of financial markets as a dynamical system consisting of all public and private information makes that hyperplane a useless abstraction. However, the concept of diversification allows to think in other, more realistic, manifolds, formed by data from the financial market’s dynamical system. If the markets are treated as a dynamical system, so are the constituents of a portfolio and the optimal portfolio itself. These systems can be modelled by PDEs and SPDEs, which most of the time are unknown but thanks to neural networks and automatic differentiation techniques their solution as well as their partial derivatives can be easily approximated from data. Just like in the Mean Variance setup, assets’ sensitivities can form manifolds in which these assets can be optimally projected for diversification. This time, diversification has a meaning defined by the PDE/SDE instead of an abstract hyperplane. Any manifold can be chosen by the portfolio manager depending on its preference, experience, or personal bias, differentiating from Bayesian methods in the structural non-linearity of PDE/SDE approximation. Our work focuses on causal manifolds. The infinite chain of events of causal dynamics in the markets makes it impossible to capture causal factors systematically, in contrast to what other practitioners believe, but physicist and engineers know this well, ending up with something that is too specific, abstract, and useless for general purposes. But if the focus is just on diversification, there is no need to care about causal dynamics, but about common causal dynamics of the portfolio constituents. For this, we have tools developed in the twentieth century, with which it is possible to approximate probabilistic common cause with correlation (Reichenbach Common Cause Principle, 1956). By finding the common cause optimal drivers of the portfolio constituents’ dynamics, we can approximate the sensitivities of the constituents with respect to them with neural networks, form the common causal manifold and find the optimal portfolio in terms of common causal diversification.
We will explain the methodology, including examples showing its impressive performance. Also, how portfolio managers can incorporate this toolkit as part of their investment process and in combination with other sources of alpha and beta. We will explain how by modelling sensitivities calibrated to market data allows modelling the trajectory of portfolio risk. Finally, we will show the connection with common causal factor models and some uses.
Alejandro Rodríguez Domínguez:
Head of Quantitative Research & Analysis, Miraltabank
16.30 – 17.15: Prediction and Imputation of financial data using collaborative filtering and Machine Learning
Abstract: Quant traders and data scientists regularly use automated ML & AI technologies to extract a variety of information from large datasets, e.g. sentiment from news data, or scoring methods for complex data sets like Supply Chain and ESG.
ML methods are also being used for imputation of financial data as well as prediction of asset prices. This talk will provide a brief overview of the following topics:
- The broad application of machine learning in finance: opportunities and challenges.
Machine Learning techniques for Imputation e.g. estimating granular Geographical Exposure of companies given partial & high-level disclosure from the company financial statement
- Collaborative filtering techniques for illiquid asset pricing, use of data driven methods to inform price movements in a target instrument from observations on related liquid instruments
Arun Verma:
17.15 – 18.00: QFC Panel
Large Language Models (LLM) in Quant Finance – A Gimmick or a Game Changer?
- Will LLLMs help us read derivatives term sheets, UCITS documentation, and securitization rules
- Will LLLMs help us document our models for the regulators
- Will LLLMs help us find bugs in our pricing models
- GPT-4 or LLAMA 2?
ML Models for Valuation, XVA, and Risk
- How to build trustworthy ML quant models that auditors and regulators will approve
- Due to the lack of sufficient training data, is ML in quant models truly learning or only providing a better way to interpolate
Alexander Sokol:
Executive Chairman and Head of Quant Research, CompatibL
Ioana Boier:
Vladimir Piterbarg:
MD, Head of Quantitative Analytics and Quantitative Development, NatWest Markets
Ignacio Ruiz:
Peter Jaeckel:
Independent financial mathematics and analytics consultant. OTC Analytics
Jos Gheerardyn:
Co-founder and CEO, Yields.io
Gala Dinner: Marina Restaurante
The main building of the Beach Club is home to the Marina Restaurant, a circular restaurant with spacious windows overlooking Las Arenas Playa. The terrace extends towards the sea, a space full of charm for those who want to have a meal “al fresco” with amazing sea views.
Marina Restaurante
Calle Marina Real Juan Carlos I
46011 Valencia
Spain
Tel: +34 961 150 007
Website
Directions from SH Valencia Palace to Marina Restaurante.
Download Event PDF
