AI Quantitative Trading

COURSE OBJECTIVE

Intensive course on artificial intelligence techniques applied to trading using two powerful programming languages: Python and R. The course aims to:

Show powerful Trading Algorithms programmed in Python and R within the Jupyter environment.

Present modern trading strategies among others: technical analysis, quantitative or fundamental trading and directional trading.

Show advanced machine learning and deep learning methodologies to predict price directions for stocks, commodities and cryptocurrencies.

Advanced supervised machine learning techniques such as ensemble learning, gradient tree boosting, random forest, support vector machine and unsupervised such as k-means are taught. Techniques are presented to validate the strategies and algorithms.

Advanced deep learning techniques are applied such as convolved neural networks for trading models, or the LSTM to modernize volatility and time series, the hybrid LSTM-CNN model for asset allocation.

An innovative module on reinforcement learning has been included to generate trading algorithms.

Present trading and forecasting strategies for stocks and cryptocurrencies, using Arima, Garch, VAR, VEC models and recurrent neural networks.

Show the use of machine learning and social networks to predict the price of shares, through sentiment analysis. As well as advanced Natural Language Processing models using bidirectional long short term memory (BLSTM)

Show portfolio management financial models, among others: MPT, CAPM, APT, through powerful exercises in Python and R.

Show recent methodologies in measuring Value at Risk and Expected Shortfall as well as the use of adversarial and generative neural networks GAN as an alternative to Monte Carlo simulation.

The validation of the models and the main issues in backtesting are explained in depth.

Present market microstructure models and structural break tests.

WHO SHOULD ATTEND?

This program is aimed at traders, managers, analysts and trading consultants. And to all those people interested in machine learning applied to trading.

Monday, December 01, 2023
Europe: Mon-Fri, CEST 16-19 h
America: Mon-Fri, CDT 18-21 h
Asia: Mon-Fri, IST 18-21 h
Price: 7.900 €
Duration: 30 h
Presentaciones PDF

R , Python, Jupyterlab y Tensorflow
yulia.tsedryk@fermacrisk.com
Banks
Trading AI PDF

Agenda

Module 1: Trading Strategies

Introduction to trading strategies

Trading with securities

Trading with Cryptocurrencies

Technical analysis

Indicators

Oscillators

Moving averages, bands and directional movements

Speed and acceleration

Movement techniques

Divergence in movement

Patterns

Fibonacci numbers

Elliot waves

Trend systems

Directional Trading

MACD

RSI

Bollinger Band

Pairs Trading

Quantitative Trading on Fundamentals

Data collection

Definition of financial ratios

Machine Learning Models

Clusters

Decision trees

Backtesting

Exercise 1: Analysis of MACD, RSI and Bollinger band in R

Exercise 2: Correlation and cointegration of trading pairs in R

Exercise 3: Fundamental Analysis with Clusters and decision trees in R

TRADING STRATEGIES
Module 2: Feature engineering

Feature engineering

Standard data preprocessing

Defining data preprocessing

Definition of feature engineering

Implementation of cross-sectional momentum

Time series features

Residualization of stock returns

Neutralize stock returns

Techniques used to residualize stock returns

Common characteristics of quantitative trading

Cross-sectional characteristics versus time series

Variables based on price

Fundamentals-Based Variables

Variables based on feelings

Text-based variables

Audio-based variables

Image-based variables

Video-based variables

Network-based variables

Common feature normalization techniques

Min-Max

Z Score

Logarithmic normalization

Quantile Normalization

Range Normalization

Other normalizations

Advanced techniques

The fixed time horizon method

Calculation of dynamic thresholds

The triple barrier method

Size learning

Meta labeling

How to use meta labeling

The Quantitative Path

Exercise 4: Cross-sectional momentum and time series features

Exercise 5: Typology of normalizations

Exercise 6: The triple barrier method

Exercise 7: Binning by side and size

Feature Engineering
Module 3: Trading Machine Learning

Definition of Machine Learning

Machine Learning Methodology

Data Storage

Abstraction

Generalization

Assessment

Supervised and Unsupervised Learning

Typology of Machine Learning algorithms

Steps to implement an algorithm

Information collection

Exploratory Analysis

Model training

Model Evaluation

Model improvements

Machine Learning for Trading

Use of RSI, MACD, Bollinge bands

Defining market direction

Probability distribution and prediction

Using Logistic Regression for prediction

Exercise 8: Direction Prediction Using Logistic Regression in R

TRADING MACHINE LEARNING
Module 4: Classification Algorithms

Decision Trees

Modeling

Advantages and disadvantages

Recursion and Partitioning Processes

Recursive partitioning tree

Pruning Decision tree

Conditional inference tree

Tree visualization

Measuring Decision Tree Prediction

CHAID model

Model C5.0

K-Nearest Neighbors

Modeling

Advantages and disadvantages

Euclidean Distance

Distance Manhattan

K value selection

Exercise 9: K-Nearest Neighbors for price prediction in R

Exercise 10: Decision trees for price prediction in R

Module 5: Unsupervised Algorithm

Characteristics of unsupervised algorithms

Clusters

K-Means

Application in trading

Advantages and disadvantages

Exercise 11: K-Means for trading strategy in R

Module 6: Replication of indices using Autoencoders

Replicating an index

Data collection

Implementation of standard Autoencoders

Data exploration and preparation

Model creation and adjustment

Model evaluation

Replicate an index using autoencoders

Explore some variants of autoencoders

The denoising autoencoder

Understand autoencoders in depth

Unsupervised Learning
Module 7: Advanced NN and SVM Algorithms

Support Vector Machine

support vectors

Optimal hyperplane

Add costs

Advantages and disadvantages

SVM visualization

SVM Tuning

Kernel Trick

Neural Networks (NN)

Perceptron Training

Multilayer Perceptron

Backpropagation algorithm

Training procedures

Tuning NN

NN visualization

Advantages and disadvantages

Exercise 12: Market data extraction from Python and R

Exercise 14: Support Vector Machine for price direction prediction in R

Exercise 15: Neural Networks for price direction prediction in R

Module 8: Ensemble Learning

Set models

Bagging

Random Forest

Boosting

Adaboost

Gradient Tree Boosting

Boosting and Bagging for regression models

Advantages and disadvantages

Stock Price Trend Prediction

Exercise 16: Price trend prediction using Ensemble models in R

Exercise 17: Intraday Price Trend Prediction Using Random Forest in Python and R

Exercise 18: Price Trend Prediction Using Adaboost in R

Exercise 19: Price trend prediction using gradient tree boosting in Python

Module 9: Deep Learning

Activation function

Sigmoidal

Rectified linear unit

Hypertangent

Feedforward network

Multilayer Perceptron

Recurrent Neural Networks

Using Tensorflow

Using Tensorboard

R deep Learning

Python deep learning

Convolutional neural networks

Use of deep learning in image classification

cost function

Gradient Descent Optimization

Using deep learning for trading

Advantages and disadvantages of deep learning

Exercise 20: Trend prediction using Deep Learning with tensorflow and python and Backtesting in Python

Module 10: Trading Rules with Convolved Neural Networks CNN

Trading signals with technical indicators

Get data from public sources

Configure data

Formulation of hypotheses and sample tests

Comparison of alternative models

Simple classification network

Build a convolutional neural network

Model investment logic

Select network architecture

Set the data in the correct format

Train and test the model

Exercise 20: Trading signals: Feed forward Neural Network modeling versus CNN Convolved Neural Network

Supervised Learning
Module 11: Volatility forecasting using LSTM

Volatility measurement

Types of volatility

Historical volatility

Implied volatility

Volatility index

Intraday volatility

Realized Volatility

Implement the LSTM model

Data preparation

Create and adjust the model

Evaluate the model

Improve model performance

Online learning

Stack layers

Adjust hyperparameters

View results

Compare LSTM with other models

RNN GRU model

GARCH model

Cumulative squared error display

Exercise 21: Volatility forecasting using LSTM versus Garch model and backtesting

Volatility Forecasting
Module 12: Forecasting of financial time series

Data treatment

Decomposition of time series

Using Pandas

Moving average

Exponential smoothing

Holt-Winter's exponential

ARIMA models

Data treatment

Normality test

Heavy tail estimation

T-test and F-test

Autocorrelation tests

Non-Stationary Series

Dickey-Fuller test

Cointegration Tests

Durbin-Watson

Stock Price Predictions

Liquidity and illiquidity estimates

Gold price predictions

Bitcoin price predictions

High-Frequency Data Treatment

Exercise 22: Non-stationary Series and Cointegration Tests in R

Exercise 23: Advanced price forecasting using ARIMA in Python

Exercise 24: Treatment of High-frequency data in Python

Module 14: Forecasting Models

Trading strategies with forecasting models

Multivariate Models

VAR Vector Autoregressive Models

ARCH Models

GARCH models

GARCH Models Multivariate Copulas

VEC Error Correction Vector Model

Johansen method

Machine Learning Models

Supported Vector Machine

Neural Network

Market time series forecasting

Data from Yahoo! Finance (r)

Data from Google Finance (r)

FRED's data

Data from Census Bureau, Treasury and BLS

Forecasting market time series returns

NN and SVM algorithms for performance forecasting

Forecasting volatility NN vs. Garch

Development and validation basis

Deep Learning

Recurrent Neural Networks RNN

Elman Neural Network

Jordan Neural Network

Basic structure of RNN

Long short term memory

Temporary windows

Development and validation sample

Regression

Sequence modeling

Time series analysis with Prophet from Facebook

Exercise 25: Stock price modeling with VAR and error correction vectors in R

Exercise 26: Multivariate volatility forecasting GARCH in R

Exercise 27: Forecasting Machine Learning using NN in R

Exercise 28: Forecasting stock prices using Recurrent Neural Networks in Python

Exercise 29: Forecasting stock series with Prophet

Exercise 30: Bitcoin forecasting using neural networks in R

Exercise 31: Garch-Arima trading algorithm

Exercise 32: Cointegration Trading Algorithm

Module 15: Reinforcement Learning

Differences between reinforcement learning and supervised and unsupervised learning

Basic concepts:

Agents

Policies

Around

Actions

Rewards

Markov decision process

Bellman equations

Dynamic programming

Monte Carlo methods

Application to finance

Trading Application

Exercise 33: Reinforcement Learning applied to trading strategies in Python and Tensorflow

Forecasting Time Series
Module 16: Sentiment Analysis on Twitter to predict stock prices

Sentiment-Based Stock Price Prediction Models

Twitter

Information sources

Twitter API connection

Token and Keys

Information processing

Definition Text Mining

Unstructured Data

Exploratory Analysis

Treemaps

Predictive modeling in Text Mining

K-Nearest Neighbors

Text Mining on Social Networks

Keyword Search

Classification algorithms

Clustering Algorithms

Feeling in linguistics and psychology

Subjectivity

Facticity

Sentiment Analysis on Twitter

Polarity Analysis and Score

Support Vector Machine

Neural Networks

Exercise 34: Text Mining of a document in R

Exercise 35: Analysis of words and tweet associations

Exercise 36: Twitter Sentiment Analysis and predictive model using Support Vector Machine in R

Module 17: Using NLP with BLSTM for news

Natural Language processing NLP

Sentiment analysis for finance

Represent text data: words into vectors

Frequency-Based Word Vectors

Counting vectorization

TF-IDF vectorization

Word embeddings

WordVec

CBOW

Skip-gram

FastText

GloVe

Loading and splitting data

Implementation of the BLSTM model

Data preparation

Model creation and adjustment

Model evaluation

Performance improvement

Handling unbalanced classes

Applying pre-trained word embeddings

Considering separate decisions

Exercise 37: NLP Modeling with BLSTM Using Stock Price News

Sentiment Analysis
Module 18: Portfolio Management

Asset Management

Modern MPT Portfolio Theory

Objectives: minimize risk, VaR and ES.

Maximize sharpe ratio

utility function

Efficient Frontier

rebalancing of positions

Capital Asset Pricing Model CAPM

Beta estimation and adjustments

Asset Pricing Theory APT

Multifactor Model

Exercise 38: Efficient frontier estimation, minimizing ES using Python

Exercise 39: Optimization and rebalancing of securities positions in R

Exercise 40: Estimating betas and Capital Asset Pricing Model (CAPM) in Python

Exercise 41: Asset Pricing Theory (APT) model

Module 19: Asset Allocation LSTM-CNN

Tactical Asset Allocation Modeling

Joint forecast for an asset class

Forecast and individual bets

Configure data

Build the model

Understand the deep learning model

Implement a CNN-LSTM model

Alternative Bayesian VAR model

Test and validate the model

Exercise 42: Hybrid model Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) for asset allocation

Asset Management
Module 20: Algorithm validation

Verification of p-values in regressions

R squared, MSE, MAD

Waste diagnosis

Goodness of Fit Test

Deviance

Bayesian Information Criterion (BIC)

Akaike Information Criterion

Cross validation

Bug Bootstrapping

confusion matrix

Kappa

Main discriminant power tests:

K.S.

ROC curve

Lift Curve

Gini Index

Cumulative Accuracy Profile

Confidence intervals

Jackknifing with discriminant power test

Bootstrapping with discriminant power test

K-Fold Cross Validation

Exercise 43: Goodness of Fit Test Logistic Regression

Exercise 44: Gini, CAP, ROC Estimation in R

Exercise 45: ROC Bootstrapping of R Parameters

Exercise 46: K-Fold Cross Validation in R

Module 21: Cross-validation in finance

The goal of cross validation

Why K-Fold CV fails in finance

Solution A: Bleeding CV K-Fold

Training set purge

Embargo

The purged K-Fold class

Module 22: Errors in cross validation

Feature Importance

The importance of the importance of features

Importance of characteristics with substitution effects

Average impurity decrease

Average Decrease Accuracy

Importance of characteristics without substitution effects

Importance of a single feature

Orthogonal features

Parallelized vs Importance of Stacked Features

Experiments with synthetic data

Module 23: Hyperparameter Tuning with Cross Validation

Grid Search Cross Validation,

Random Search Cross Validation

Uniform logarithmic distribution

Hyperparameter scoring and tuning

Model Validation
Module 24: Advanced Backtesting

Backtesting Analysis

Bet size

Strategy-independent bet sizing approaches

Bet Size from Predicted Odds

Average active bets

Size discretization

Dynamic bet sizes and limit prices

Risks of backtesting

Clean and dirty backtesting

Backtesting problems

Some general recommendations

Strategy selection

Backtesting using cross validation

The Forward-Looking Method

Walk-Forward Method Errors

The cross validation method

The combinatorial method of purged cross-validation

Combinatorial divisions

Debugged combinatorial cross-validation

Backtesting algorithm

How Clean Combinatorial Cross-Validation Addresses Backtesting Overfitting

Synthetic data backtesting

Trading rules

The problem

Our framework,

Optimal trading rules

Algorithms

Implementation

Experimental results

Cases with zero long-run equilibrium

Cases with positive long-term balance

Cases with long-term negative equilibrium

Backtesting Statistics

Types of backtesting statistics

Evaluation metrics

Information Coefficient and R2

General characteristics

Performance

Time-weighted rate of return

Execution statistics for performance evaluation

Implementation deficit

Sharpe's relationship

The probabilistic Sharpe relation

The deflated Sharpe ratio

Efficiency statistics

Classification and attribution scores

Exercise 47: Backtesting metrics and estimation of the Sharpe ratio of a trading strategy

Backtesting
Module 25: Value at Risk (VaR) and Expected Shortfall (ES)

Linear and non-linear portfolios

Volatility Estimation

Parametric Models

Normal VaR

t-student distribution

Lognormal Distribution

Linear Model for Stocks and Bonds

Quadratic Model for options

Expected Shortfall

Exercise 48: Expected Shortfall and VaR in Python

Module 26: Historical Simulation and Monte Carlo

VaR Historical Simulation

Volatility Adjustment

Bootstrapping

VaR Monte Carlo Simulation

Simulation with a risk factor

Simulation with multiple risk factors

Variance Reduction Methods

VaR Monte Carlo based on Gaussian copula

VaR Monte Carlo based on t-student copula

Exercise 49: VaR Estimation: Using Monte Carlo Simulation in Python

Exercise 50: Historical Simulation in Excel

Exercise 51: Historical Simulation Backtesting in Excel

Exercise 52: VaR using Gaussian copula and tStudent in R

Module 27: Risk measurement using GAN

Estimation of value at risk

Computing methods and disadvantages

Introduction of generative adversarial networks

Generative models

Discriminative models

GAN inner workings

Implementation of a risk model using GAN

Definition of our model

GAN model implementation

Benchmarking results

Exercise 53: Estimation of VaR using GAN neural networks and Monte Carlo simulation

Risks: VaR, ES and Machine Learning
Module 28: Structural breaks

Types of structural failure tests

Entropy functions

Shannon entropy

The complementary (or maximum likelihood) estimator

Lempel-Ziv estimators

Coding schemes

Binary encoding

Quantile coding

Sigma Coding

Entropy of a Gaussian process

Entropy and generalized mean

Some financial applications of entropy

Market efficiency

Maximum entropy generation

Portfolio concentration

Market microstructure

Module 29: Microstructural characteristics

First generation: price sequences

The tick rule

The roll model

CUSUM tests

Brown-Durbin-Evans CUSUM test in recursive

Waste

Chu-Stinchcombe-White CUSUM Test in Tiers

Explosivity tests

Dickey-Fuller Chow type test,

Supreme Augmented Dickey-Fuller

Submartingale and supermartingale tests

High-Low Volatility Estimator

Second Generation: Strategic Business Models

Kyle's Lambda

Amihud Lambda

Hasbrouck Lambda

Third generation: sequential business models

Information Based Trading Probability

Probability of information synchronized with volume

Trade

Additional features of microstructural data sets

Order Size Distribution

Cancellation rates, limit orders, market orders

Time Weighted Average Price Execution Algorithms

Options Markets

Serial correlation of signed order flow

What is market microstructural information?

Exercise 54: Structural break test: The Chow Test

Exercise 55: Structural break test: The CUSUM Test

Market Microstructure