Artificial Intelligence for Financial Markets: The Polymodel Approach (Financial Mathematics and Fintech)

✍ Scribed by Thomas Barrau, Raphael Douady

• Publisher: Springer
• Year: 2022
• Tongue: English
• Leaves: 182
• Edition: 1st ed. 2022
• Category: Library

✦ Synopsis

This book introduces the novel artificial intelligence technique of polymodels and applies it to the prediction of stock returns. The idea of polymodels is to describe a system by its sensitivities to an environment, and to monitor it, imitating what a natural brain does spontaneously. In practice this involves running a collection of non-linear univariate models. This very powerful standalone technique has several advantages over traditional multivariate regressions. With its easy to interpret results, this method provides an ideal preliminary step towards the traditional neural network approach. 
The first two chapters compare the technique with other regression alternatives and introduces an estimation method which regularizes a polynomial regression using cross-validation. The rest of the book applies these ideas to financial markets. Certain equity return components are predicted using polymodels in very different ways, and a genetic algorithm is described which combines these different predictions into a single portfolio, aiming to optimize the portfolio returns net of transaction costs. Addressed to investors at all levels of experience this book will also be of interest to both seasoned and non-seasoned statisticians.

✦ Table of Contents

Foreword
Preface
Contents
Chapter 1: Introduction
1.1 Financial Market Predictions: A Concise Literature Review
1.2 A Simple Model of Portfolio Returns
1.3 Plan of the Book
References
Chapter 2: Polymodel Theory: An Overview
2.1 Introduction
2.2 Mathematical Formulation
2.3 Epistemological Foundations
2.3.1 A Statistical Perspectivism
2.3.2 A Phenomenological Approach
2.4 Comparison of Polymodels to Multivariate Models
2.4.1 Reducing Overfitting
2.4.2 Increasing Precision
2.4.3 Increasing Robustness
2.5 Considerations Raised by Polymodels
2.5.1 Aggregation of Predictions
2.5.2 Number of Variables Per Elementary Model
2.6 Conclusions
References
Chapter 3: Estimation Method: The Linear Non-Linear Mixed Model
3.1 Introduction
3.2 Presentation of the LNLM Model
3.2.1 Definition
3.2.2 Fitting Procedure
3.3 Evaluation Methodology
3.4 Results
3.4.1 For 126 Observations (Tables 3.1 and 3.2)
3.4.2 For 252 Observations (Tables 3.3 and 3.4)
3.4.3 For 756 Observations (Tables 3.5 and 3.6)
3.4.4 For 1,260 Observations (Tables 3.7 and 3.8)
3.4.5 Computation Time (Table 3.9)
3.4.6 Interpretations
3.5 Conclusions
References
Chapter 4: Predictions of Market Returns
4.1 Introduction
4.2 Data
4.3 Systemic Risk Indicator
4.3.1 Model Estimation
4.3.2 Systemic Risk Indicator Definition
4.3.3 Primary Analysis
4.3.4 Roots of the Predictive Power
4.4 Trading Strategy
4.4.1 Methodology
4.4.2 Results
4.5 Robustness Tests
4.5.1 Sensitivity to the Noise Filter
4.5.2 Sensitivity to Future Returns Windows
4.5.3 Sensitivity to Half-life
4.5.4 Sensitivity to Rolling Window Length
4.5.5 Asset-Class Specific Performances
4.5.6 Sensitivity to Trading Strategy
4.6 Conclusions
References
Chapter 5: Predictions of Industry Returns
5.1 Introduction
5.2 Data
5.2.1 Summary Statistics
5.3 Methodology
5.3.1 Measuring Antifragility
5.3.2 Predicting Industry Returns
5.4 Results
5.4.1 Long/Short Trading Strategy
5.4.2 Regressions on Future Returns
5.4.3 Comparisons to Classical Factors
5.4.4 Separating Concavity and Convexity
5.4.5 Separating Positive and Negative Market Returns
5.5 Robustness Tests
5.5.1 Stability of the Signal
5.5.2 Sensitivity to Hyper-Parameters
5.6 Conclusions
References
Chapter 6: Predictions of Specific Returns
6.1 Introduction
6.2 Methodological Foundations
6.2.1 Data
6.2.2 Time-series Predictions, Cross-sectional Portfolio Construction
6.2.3 Subtracting the Average Return
6.3 Predictions Selection
6.3.1 Root Mean Squared Error Filter Evaluation
6.3.2 p-value Filter Evaluation
6.3.3 Bayesian Information Criterion Filter Evaluation
6.3.4 Selection Using Dynamic Optimal Filtering
6.4 Aggregation of Predictions
6.4.1 Prediction Aggregation Using Bayesian Model Averaging
6.4.2 Prediction Aggregation Using Added Value Averaging
6.4.3 Uncertainty of Aggregate Predictions
6.5 Trading Strategy
6.5.1 Methodology
6.5.2 Performance
6.5.3 Significance of the Methods
6.6 Robustness Tests
6.6.1 Correlations with Standard Factors
6.6.2 Sensitivity to the Filter Regression Window
6.6.3 Stability of the Performance
6.6.4 Sensitivity to the Believability Measure Window
6.6.5 Sensitivity to the Originality Measure Window
6.6.6 Unaccounted Parameter Sensitivities
6.7 Conclusions
References
Chapter 7: Genetic Algorithm-Based Combination of Predictions
7.1 Introduction
7.2 Analysis of Strategies
7.2.1 Predictions of Market Returns
7.2.2 Predictions of Industry Returns
7.2.3 Predictions of Specific Returns
7.2.4 Correlation Analysis
7.3 Risk Parity Combination
7.3.1 Introducing Risk Parity
7.3.2 Transaction Costs Matter
7.3.3 Ex-ante Optimal Reduction of the Transaction Costs
7.4 Genetic Algorithm-Based Combinations
7.4.1 Methodology
7.4.2 Results
7.5 Robustness Tests
7.5.1 Mutation Probability
7.5.2 Number of Chromosomes
7.5.3 Number of Epochs
7.5.4 Seed
7.5.5 Optimal Trading Rate Window
7.6 Conclusions
References
Chapter 8: Conclusions
Appendix
Representative RMSE Distributions per Stock Index
Market Timed Portfolio and Systemic Risk Indicator per Stock Index
Industry Buckets Summary Statistics
AFI Scores per Industry over Time