Data Mining & Machine Learning in Finance

• Association rules
• Classification vs. regression problems
• Clustering analysis

• Overview of third party solutions (Tableau, QlikeTech etc.) for visualization of large sets of data. Case studies will be worked out using matplotlib-library and plotly (open-source online data-collaboration platform)
• Graphical databases: applying network theory on portfolio analysis and introduction to graphical databases

• OLS (ordinary least squares)
• Ridge regression
• Sparsity
• Lasso
• Elastic Net

Workshop: Working out the optimal hedge of a large real-world equity portfolio using futures. The portfolio has a global nature (100+ shares) but only a limited set of futures is available

• Principal component analysis of the term structure of interest rates and implied volatilities
• Principal component regression (PCR)
• Partial least squares (PLS)

Workshop: Using PCA to reduce dimensionality of a large data set of historical interest rate curves. The complex behaviour of this curve is spread over different maturities and this technique allows a risk manager to have a much better view of the dynamics of interest rate curves

• Kernel density estimation is an unsupervised learning procedure which leads to a simple family of procedures for non-parametric classification

Case study: Using kernels to derive probability distributions for financial data

• Naive Bayes classification: A straightforward and powerful technique to classify data

Case study: Working out a Bayes-predictor for a large data set containing different attributes of US banks. The Bayes classifier will be used to separate those banks that are likely to fail from those that are going to remain solvent

• Linear Discriminant Analysis (LDA)
• Logistic Regression

Case study: apply log-regression on a real-world dataset with high dimensionality

• Classification Trees: CART-modelling leads to easy-to-use practical decision trees
• The concept of decision trees will be extended with techniques such as Random Forest and Bagging

Case study: Concepts such as cost functions, impurity levels, tree pruning and cross validation will be handled in detail

• Support Vector Machines (SVM)
• K-Nearest Neighbour learning
• Logistic Regression

Case study: The classification methods (SVM, K-Nearest and CART) are going to be put at work on different technical indicators (RSI, MACD etc.) of large sets of real-world financial data. This will illustrate how these classifiers can be used to partition stocks in different buckets according to the strength of different attributes in a fast way

An introduction to Python - a powerful programming language. The applicability of Python in the domain of data analysis will be illustrated through practical examples with focus on machine learning using the 'scikit-learn' package. All examples will be covered in Jupyter-notebooks. Delegates will learn how to build custom reports in Python

Day Three

Natural Language Processing

Extracting real value from social media posts, images, email, PDFs and other sources of unstructured data is a big challenge for enterprises.

This section is devoted to the application of Natural Language Processing (NLP) to extract value from unstructured data. Several real-world examples of examining unstructured data in finance - including sentiment analysis of financial news - will be explored.

Workshop: Using the NLTK package of Python to:

• Explore and tokenize a text using Tf-Idf and Count Vectors
• Predict words in a text: building a word predictor starting from a text; writing a programme that can predict the word that follows a given word
• Understand the sentiment of a news item on a particular stock

Deep Learning

Deep Learning as a subfield of machine learning - Artificial Neural Networks (ANN) algorithms.

• Introduction to Deep Learning
• Forward propagation
• Word2vec approach
• Deeper networks and forward propagation
• Optimizing Neural Network with backward propagation

Case Study: Building a Deep Learning model with Python (with a focus on the Keras and Tensorflow packages)