This thesis consists of three applications of machine learning techniques to empirical asset pricing.
In the first part, which is co-authored work with Alexis Marchal, we develop a new method that detects jumps nonparametrically in financial time series and significantly outperforms the current benchmark on simulated data. We use a long short-term memory (LSTM) neural network that is trained on labelled data generated by a process that experiences both jumps and volatility bursts. As a result, the network learns how to disentangle the two. Then it is applied to out-of-sample simulated data and delivers results that considerably differ from the benchmark: we obtain fewer spurious detection and identify a larger number of true jumps. When applied to real data, our approach for jump screening allows to extract a more precise signal about future volatility.
In the second part, which is co-authored work with Alexis Marchal, we develop a methodology for detecting asset bubbles using a neural network. We rely on the theory of local martingales in continuous-time and use a deep network to estimate the diffusion coefficient of the price process more accurately than the current estimator, obtaining an improved detection of bubbles. We show the outperformance of our algorithm over the existing statistical method in a laboratory created with simulated data. We then apply the network classification to real data and build a zero net exposure trading strategy that exploits the risky arbitrage emanating from the presence of bubbles in the US equity market from 2006 to 2008. The profitability of the strategy provides an estimation of the economical magnitude of bubbles as well as support for the theoretical assumptions relied on.
In the third part, I propose a new methodology to construct interpretable, fundamental-based pricing factors from news to explain Bitcoin returns. Each news article from a specialized cryptocurrency website is classified in a semi-supervised manner into one of the few prede- fined topics. Topic sentiments become factors contributing to the price variation. I use a cutting-edge NLP algorithm (SBERT network) to embed linguistic data into a vector space, which allows the application of an intuitive classification rule. This approach permits the exclusion of news pieces that describe the price movements per se from the analysis, thus mitigating endogeneity concerns. I show that non-endogenous news contains fundamental information about Bitcoin. Thus I reject the concept of Bitcoin price being based on pure spec- ulation and show that Bitcoin returns are partially explained by fundamental topics. Among those, the adoption of cryptocurrencies and blockchain technology is the most important aspect. On top of that, I study the media expressed attitude toward Bitcoin from the functions of money perspective. I show that investors consider Bitcoin as the store of value rather than the medium of exchange.