Data:
23-24 March 2022

Venue:
Hotel Bristol Warsaw /
CEETA Live

Machine Learning for Quantitative Finance

Course Leader:

Joerg Kienitz

The goal of this workshop is to provide a detailed overview of Statistical/Machine Learning techniques
applied to Quantitative Finance. We offer insights into the latest techniques of using such methods.
Tackling topics that arise in derivatives pricing, calibration and hedging but also from time series
management.
This includes sophisticated modeling approaches for the Q-quant setting and nowcasting, imputation or
anomaly detection for the P-quant.
We give a thorough theoretical introduction but illustrate the concepts with concrete examples. Live
demonstrations of the computational methods round up this course.
We cover Artificial Neural Networks, Gaussian Process Regression, Gaussian Mixtures and kernel methods
and their applications.
We also explain how to set up the methods in Python using various libraries such as Numpy, SciKit Learn,
Keras, Tensorflow, or PyTorch. Our chosen examples are directly linked to relevant practical applications
from Quantitative Finance and can be explored further after the course since all the material is available
either as Python code or Jupyter notebooks.
This workshop covers the fundamentals and it illustrates the application of state-of-the-art machine
learning applications for Mathematical and Quantitative Finance. We wish to bring you to the next level
with our presented material.

w w w . c e e t a . p l


 

 
 

JOERG KIENITZ COURSE HIGHLIGHTS

Joerg is a finance professional for over 20 years and This workshop covers the latest techniques for
held numerous positions as Head of Quantitative mastering Statistical Learning and Machine Learning
Analytics. He owns the finciraptor website methods including Neural Networks or Gaussian Process
(finciraptor.de). As a quantitative finance professional, Regression and apply those to Quantitative Finance and
he is primarily involved in Machine Learning Time Series analysis. Theoretical underpinnings are
applications in quantitative finance, consulting on giving and explained. The material is illustrated with
model validation, model development and model many relevant examples from Quantitative Finance.
implementation across all asset classes. Furthermore, Especially we cover:
Jörg acts as a trainer for public and onsite classes. • Overview of some Statistical Learning / Machine
As an academic Jörg lectures at the universities of Learning techniques
Wuppertal (BUW) and Cape Town (UCT) where • Implementation and examples from Quantitative
he is an Assistant Professor, respectively and Adjunct Finance using Python/Jupyter
Associate Professor. His research topics include • Neural Networks – Architectures, Mechanics and
advanced financial modeling, numerical methods, Applications
stochastics and machine/statistical learning. Further to • Generative Methods applied to Quantitative
the academic engagements he speaks on well-known Finance
Quantitative Finance conferences including Quant • Gaussian Process Regression in the context of
Minds or WBS Quant Conference. pricing and time series analysis
Jörg holds a Ph.D. in stochastic analysis and • Gaussian Mixtures for Local Stochastic Volatility or
probability theory and authored several papers in Bermudans in multiple dimensions
well-known journals and four books “Monte Carlo • Pricing and calibration for (rough) Stochastic
Object Oriented Frameworks in C++” Volatility models such as SABR or Heston or rough
(with Daniel J. Duffy) “Financial Modelling” (with Daniel models like the rough Bergomi model
Wetterau), “Interest Rate Derivatives Explained I” and • Term structure models and Bermudan/American
“Interest Rate Derivatives Explained II” (with Peter options
Caspers) with Wiley and Palgrave/Springer. His • Time Series forecasting/nowcasting and
LinkedIn and SSRN author page can be accessed via imputation using Gaussian Process Regression

COURSE METHODOLOGY
https://de.linkedin.com/in/j%C3%B6rg-kienitz-8b160a

• Presentation and examples (Jupyter

notebooks, Python Codes)

https://papers.ssrn.com/sol3/cf_dev/
AbsByAuth.cfm?per_id=744396

        


 

 
 

The course starts at 9 AM and ends approx. 5 PM. During the

Contents of Workshop: sessions two 30 min breaks (10.30-11.00 & 15.00-15.30) and one
hour break (12.30-13.30) is scheduled each day.

Machine Learning and Quantitative Finance – Overview

• The first part of the course introduces the subject and lays the foundation for the use cases and the more
advanced considerations of this course.
o Introduction to the subject – Basics of Statistical Learning / Machine Learning
This introduces the concept of statistical learning/machine learning and discusses applications as well as
fundamental notions and introduce some of the ML slang.
o Supervised Learning for Classification, Regression
o Unsupervised Learning
o Selfsupervised Learning
o Reinforcement Learning
o Overfitting / Underfitting
o Train, validate, test
• Hyperparameters
• Bayesian vs Frequentist View
We consider two views on statistics and highlight the differences.
• Linear and Logistic Regression as our guiding examples
Before learning about some state-of-the-art techniques in statistical learning/machine learning we focus on an
illustrative example and discuss different angles of the problem by focusing on an introductory example. This
serves also as a guide for what we wish to achieve in this course. As an example, we consider the least
squares Monte Carlo method
o Quantitative Finance
We consider applications from Quantitative Finance, especially we focus on the following models as
examples: Black-Scholes, Heston, SABR, rough Bergomi, Hull-White Term Structure Model. This is an overview
and not a deep mathematical analysis of the models. We consider:
o Pricing and calibration
o Hedging
o Time Series Analysis (prediction, imputation, outliers/anomalies)

Introduction to Machine Learning Methods and their Implementation – An Overview

• We give an overview of ML methods with some illustrations in Python. We discuss some available packages
that are necessary to work on topics in Quantitative Finance.
• Clustering (k-means and Gaussian Mean Mixture) – the E-M algorithm
• Support Vector Machines
• Lasso and Ridge Regression
• Ensemble Learning - bagging, boosting, decision trees and random forrests
• Python packages
o Numpy and Scikit Learn
o Pandas
o Tensorflow, PyTorch, Keras
• Some examples:
o Prcing and Greeking (Black-Scholes and SABR)
o Monte Carlo (Bermudan Swaption Hull-White)
o Stochastic Local Volatility Models
o Xlwings – interacting with Excel

        


 

 
 

Artificial Neural Networks in Finance – introduction and examples

A hot topic is the application of Artificial Neural Networks to practical problems. We illustrate the theory and show
different architectures that can be applied in practice. Finally, we show applications to option pricing, calibration and
hedging. Furthermore, we give advice for further optimizing the methods by using e.g. feature engineering.

• Intro to Artificial Neural Networks (ANN)

This introduces the architecure of Neural Networks and their implementation. You’ll learn the math used by
Tensorflow/PyTorch and we give a visual proof that an ANN can virtually learn any function.
o Construction of ANNs
o ANN at work
o Math recap - (Backpropagation and Optimization, ie. stochastic gradient descent methods)
o Learning functions – some visual experiments
• Learning pricing functions
„learning to smile“ – application to volatility modelling
• It’s only an approximation! - We discuss interpolation and extrapolation
• Shallow vs Deep Learning and Preprocessing/Feature engineering – The CV method
• Different Types of Networks, their implementation and applications
o FNN - Feed Forward
o CNN - Convolutional
o RNN - Recursive
o LSTM – Long Short Term Memory
• Examples:
o Deep Pricing and calibration - ANN and CNN
o Deep Hedging – LSTM

Generative Modelling in Quantitative Finance – Introduction and Examples

Generative Modelling becomes increasingly popular. Many applications are known in image or video processing.
VAE and GAN and suggest some applications for practical application.

• Differential Machine Learning – The CV method revisted

We show how we could use the combination of two neural networks to enhance the standard learning by
incorporating derivatives (Greeks) into the learning process
• VAE – Variational Autoencoder
o Illustration of the Network Architecture
o Convolutional Layers, Pooling Layers and how they work
o Tuning the network architecture
o Applications: Implied Volatility Surface Generation, Anomaly Detection, Nowcasting
• GAN - Generative Adversarial Neural Networks
o Basic architecture of GANs
o The Generator and the Discriminator
o Linking the Generator and the Discriminator
o Stabilizing the training
o Applications: CorrGAN, QuantGAN

        


 

 
 

Gaussian Methods in Quantitative Finance

• Kernel Density Estimation

o Kernel density estimation and its application
o Local linear and Nadarya-Watson
o Basic examples

• Gaussian Mixture Models

o Mixture Models in finance – dos and donts
o GMM as a kernel density estimator
o GMM-DCKE
o Conditional expectation and pricing – American/Bermudan Options
o Stochastic Local Volatility – a new approach to calibration
o GMM as a generative method
o xVA/exposure

• Gaussian Process Regression and Kernel Methods (GPR)

Based on Bayesian statistics GPR is successfully applied in science. We show how it can be used and optimized
for Quantitative Finance applications.
o Multi-dimensional Gaussian distribution (marginal, conditional distribution, etc.)
o Intro to GPR and regression - how does it work?
o Covariance functions and kernels – The kernel trick
o Choosing kernels

Time Series Analysis

o Introduction
o Point estimates and distributional estimates
o Anomaly detection and Forecasting for time series
o Conditional expectation and pricing – American/Bermudan Options

The course starts at 9 AM and ends approx. 5 PM. During the

sessions two 30min breaks (10.30-11.00 & 15.00-15.30) and one
hour break (12.30-13.30) is scheduled each day.