Koshiyama, Adriano Soares; (2020) Applications of Machine Learning Methods in Financial Markets. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text Adriano Koshiyama - PhD Thesis.pdf - Accepted Version Download (15MB) | Preview

Abstract

This thesis investigates the application of new machine learning algorithms like Generative Adversarial Networks (GANs), Transfer Learning, etc. as building blocks for reliable investment decisions. To improve a machine learning-based trading strategy assessment one needs to consider the problem of backtest overfitting – strategies outperforming on training data but underperform when presented with new data. In this sense, this thesis considers three independent forms to deal with this problem: a) correcting performance metrics, such as the Sharpe ratio, using covariance-penalty methods; b) using GANs to synthesize financial time series data and use them to improve trading strategies calibration and performance assessment; and c) developing novel Transfer Learning-based strategies to deal with data scarcity, spurious correlation and, by consequence, aiding in avoiding backtest overfitting. This research is important for several reasons: (i) backtest overfitting is a pervasive problem that impacts practitioners and researchers across the financial markets; (ii) the increasing adoption of data-driven methodologies, particularly machine learning methods across markets is demanding new forms of model validation and assessment; and (iii) GANs and Transfer Learning are modelling paradigms with proven success in areas such as computer vision, natural language processing, yet with insubstantial amount of research in the finance domain. This research comprises three experiments briefly described below: Avoiding Backtesting Overfitting by Covariance-Penalties – in this experiment we propose a new approach to deal with financial overfitting, a covariance-penalty correction, in which a risk metric is adjusted given the number of parameters and amount of data used to underpin a trading strategy. We outline the theoretical foundation and main results behind the covariance-penalty correction for trading strategies. After that, we pursue an empirical investigation and compare its performance with some other approaches in the realm of covariance-penalties across more than 1,300 assets, using ordinary and total least squares. The findings suggest that covariance-penalties are a suitable procedure to avoid backtesting overfitting, and total least squares outperforms ordinary least squares. Generative Adversarial Networks for Financial Trading Strategies – this experiment proposes the use of Conditional GANs (cGANs) for trading strategies calibration and aggregation. To this purpose, we provide a full methodology on: (i) the training and selection of a cGAN for time series data; (ii) how each sample is used for strategies calibration; and (iii) how all generated samples can be used for ensemble modelling. To provide evidence that our approach is well grounded, we have designed an experiment with multiple trading strategies, encompassing 579 assets. We compared cGAN with an ensemble scheme and model validation methods, both suited for time series. Our results suggest that cGANs are an efficient alternative for strategies calibration and combination, providing outperformance when the traditional techniques fail to generate any alpha. Transferring Learning Across Trading Strategies – in this experiment we introduce QuantNet: an architecture that is capable to transfer knowledge across systematic trading strategies in several financial markets. By having a system that can leverage and share knowledge across them, our aim is two-fold: to circumvent the so-called backtest overfitting problem; and to generate higher risk-adjusted returns and less drawdowns. In order to evaluate QuantNet, we compared its performance to the option of not performing transfer learning, that is, using market-specific old-fashioned machine learning. In summary, our findings suggest that QuantNet performs better than non transfer-based trading strategies, improving Sharpe ratio in 15% and Calmar ratio in 41% across 3103 assets in 58 equity markets across the world. The major contributions of this work are: A covariance-penalty correction formula that can be widely used by quantitative strategists in the financial services; A new approach to synthesise financial time series that can be applied to calibrate and improve model selection and assessment; A new framework to learn systematic trading strategies by cross-pollinating gained knowledge in different financial markets. This work has been done in conjunction with Nomura International, Goldman Sachs International, and The Alan Turing Institute. This research has directly resulted in 7 papers.

Download activity - last month

Export as JSONCSVXML

Download activity - last 12 months

Export as XMLCSVJSON

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item