Topics in Bayesian machine learning for finance
<p>This thesis presents novel Bayesian machine learning-based approaches to selected problems in finance. We investigate and model perceived market inefficiencies with respect to curated financial data sets, then assess market trading and investment strategies, yielding the following three co...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
2024
|
| Subjects: |
| _version_ | 1824459072357793792 |
|---|---|
| author | Spears, T |
| author2 | Roberts, S |
| author_facet | Roberts, S Spears, T |
| author_sort | Spears, T |
| collection | OXFORD |
| description | <p>This thesis presents novel Bayesian machine learning-based approaches to selected problems in finance. We investigate and model perceived market inefficiencies with respect to curated financial data sets, then assess market trading and investment strategies, yielding the following three contributions.</p>
<p>Firstly, we consider a modern deep learning prediction model given intraday limit order book data constituting of a subset of Eurodollar futures contracts, the popular and liquidly traded set of interest rate derivatives. The data is spatio-temporal in nature, such that the model utilises convolutional neural networks for automated feature extraction and recurrent neural networks for time series prediction. We show how to train the network to yield short-term forecasts paired with aleatoric uncertainty estimates by specifying a suitable loss function. Further, we estimate an approximation to epistemic uncertainty via a pseudo-Bayesian deep learning method. This work demonstrates the utility of the model output for deciding the relative allocation of risk capital across trades. That is, investment sizes are scaled between trade opportunities in a principled and data-driven way. We calculate the trading benefit and demonstrate model outperformance relative to strategies that either do not take uncertainty into account, or that utilize an alternative-yet-common market-based statistic as a proxy for uncertainty.</p>
<p>The Black-Litterman model extends the framework of the Markowitz Modern Portfolio Theory to incorporate investor views. In our second work, we recognise this extension as a Bayesian formulation, and relate view uncertainty to its aleatoric or epistemic counterpart. We consider the novel case where multiple view estimates, including uncertainties, are given for the same underlying subset of assets at a point in time. We find that data fusion techniques -- for combining similar information from multiple sources -- can help to synthesise a collection of views into a single investment strategy. In particular, we present consistency-based methods that yield fused view and uncertainty pairs; such methods are not common to the quantitative finance literature. We show a relevant, modern case of incorporating machine learning model-derived view and uncertainty estimates, and the impact on portfolio allocation, with an example subsuming Arbitrage Pricing Theory. Hence we show the value of the Black-Litterman model in combination with information fusion and artificial intelligence-grounded prediction methods.</p>
<p>Our third contribution further leverages preceding work. We devise a novel Bayes-based conditional factor model for a multivariate portfolio of United States equities in the context of analysing a statistical arbitrage trading strategy. A state space framework underlies the factor model whereby asset returns are assumed to be a noisy observation of a linear combination of factor values and latent factor risk premia. Filter and state prediction estimates for the risk premia are retrieved in an online way. Such estimates induce filtered asset returns that can be compared to measurement observations, with large deviations representing candidate mean reversion trades. Further, in that the risk premia are modelled as time-varying quantities, non-stationarity in returns is de facto captured. We study an empirical trading strategy respectful of transaction costs and demonstrate performance over a long history of 29 years, for both a linear and a non-linear state space model. Our results show that the model is competitive relative to other methods, including simple benchmarks and other comparable innovative approaches published in the literature. Also of note, while strategy performance degradation is noticed through time – especially for the most recent years – the strategy continues to offer compelling economics, and has scope for further advancement.</p> |
| first_indexed | 2025-02-19T04:35:58Z |
| format | Thesis |
| id | oxford-uuid:35b9e5a9-a872-4257-b7ee-2bff683cc039 |
| institution | University of Oxford |
| language | English |
| last_indexed | 2025-02-19T04:35:58Z |
| publishDate | 2024 |
| record_format | dspace |
| spelling | oxford-uuid:35b9e5a9-a872-4257-b7ee-2bff683cc0392025-02-03T09:51:08ZTopics in Bayesian machine learning for financeThesishttp://purl.org/coar/resource_type/c_db06uuid:35b9e5a9-a872-4257-b7ee-2bff683cc039FinanceMachine learningEnglishHyrax Deposit2024Spears, TRoberts, SZohren, S<p>This thesis presents novel Bayesian machine learning-based approaches to selected problems in finance. We investigate and model perceived market inefficiencies with respect to curated financial data sets, then assess market trading and investment strategies, yielding the following three contributions.</p> <p>Firstly, we consider a modern deep learning prediction model given intraday limit order book data constituting of a subset of Eurodollar futures contracts, the popular and liquidly traded set of interest rate derivatives. The data is spatio-temporal in nature, such that the model utilises convolutional neural networks for automated feature extraction and recurrent neural networks for time series prediction. We show how to train the network to yield short-term forecasts paired with aleatoric uncertainty estimates by specifying a suitable loss function. Further, we estimate an approximation to epistemic uncertainty via a pseudo-Bayesian deep learning method. This work demonstrates the utility of the model output for deciding the relative allocation of risk capital across trades. That is, investment sizes are scaled between trade opportunities in a principled and data-driven way. We calculate the trading benefit and demonstrate model outperformance relative to strategies that either do not take uncertainty into account, or that utilize an alternative-yet-common market-based statistic as a proxy for uncertainty.</p> <p>The Black-Litterman model extends the framework of the Markowitz Modern Portfolio Theory to incorporate investor views. In our second work, we recognise this extension as a Bayesian formulation, and relate view uncertainty to its aleatoric or epistemic counterpart. We consider the novel case where multiple view estimates, including uncertainties, are given for the same underlying subset of assets at a point in time. We find that data fusion techniques -- for combining similar information from multiple sources -- can help to synthesise a collection of views into a single investment strategy. In particular, we present consistency-based methods that yield fused view and uncertainty pairs; such methods are not common to the quantitative finance literature. We show a relevant, modern case of incorporating machine learning model-derived view and uncertainty estimates, and the impact on portfolio allocation, with an example subsuming Arbitrage Pricing Theory. Hence we show the value of the Black-Litterman model in combination with information fusion and artificial intelligence-grounded prediction methods.</p> <p>Our third contribution further leverages preceding work. We devise a novel Bayes-based conditional factor model for a multivariate portfolio of United States equities in the context of analysing a statistical arbitrage trading strategy. A state space framework underlies the factor model whereby asset returns are assumed to be a noisy observation of a linear combination of factor values and latent factor risk premia. Filter and state prediction estimates for the risk premia are retrieved in an online way. Such estimates induce filtered asset returns that can be compared to measurement observations, with large deviations representing candidate mean reversion trades. Further, in that the risk premia are modelled as time-varying quantities, non-stationarity in returns is de facto captured. We study an empirical trading strategy respectful of transaction costs and demonstrate performance over a long history of 29 years, for both a linear and a non-linear state space model. Our results show that the model is competitive relative to other methods, including simple benchmarks and other comparable innovative approaches published in the literature. Also of note, while strategy performance degradation is noticed through time – especially for the most recent years – the strategy continues to offer compelling economics, and has scope for further advancement.</p> |
| spellingShingle | Finance Machine learning Spears, T Topics in Bayesian machine learning for finance |
| title | Topics in Bayesian machine learning for finance |
| title_full | Topics in Bayesian machine learning for finance |
| title_fullStr | Topics in Bayesian machine learning for finance |
| title_full_unstemmed | Topics in Bayesian machine learning for finance |
| title_short | Topics in Bayesian machine learning for finance |
| title_sort | topics in bayesian machine learning for finance |
| topic | Finance Machine learning |
| work_keys_str_mv | AT spearst topicsinbayesianmachinelearningforfinance |