This paper introduces a novel multivariate volatility modeling framework, named Long Short-Term Memory enhanced BEKK (LSTM-BEKK), that integrates deep learning into multivariate GARCH processes. By combining the flexibility of recurrent neural networks with the econometric structure of BEKK models, our approach is designed to better capture nonlinear, dynamic, and high-dimensional dependence structures in financial return data. The proposed model addresses key limitations of traditional multivariate GARCH-based methods, particularly in capturing persistent volatility clustering and asymmetric co-movement across assets. Leveraging the data-driven nature of LSTMs, the framework adapts effectively to time-varying market conditions, offering improved robustness and forecasting performance. Empirical results across multiple equity markets confirm that the LSTM-BEKK model achieves superior performance in terms of out-of-sample portfolio risk forecast, while maintaining the interpretability from the BEKK models. These findings highlight the potential of hybrid econometric-deep learning models in advancing financial risk management and multivariate volatility forecasting.

We propose a novel distributionally robust $Q$-learning algorithm for the non-tabular case accounting for continuous state spaces where the state transition of the underlying Markov decision process is subject to model uncertainty. The uncertainty is taken into account by considering the worst-case transition from a ball around a reference probability measure. To determine the optimal policy under the worst-case state transition, we solve the associated non-linear Bellman equation by dualising and regularising the Bellman operator with the Sinkhorn distance, which is then parameterized with deep neural networks. This approach allows us to modify the Deep Q-Network algorithm to optimise for the worst case state transition. We illustrate the tractability and effectiveness of our approach through several applications, including a portfolio optimisation task based on S\&{P}~500 data.

We present a novel approach for predicting the distribution of asset returns using a quantile-based method with Long Short-Term Memory (LSTM) networks. Our model is designed in two stages: the first focuses on predicting the quantiles of normalized asset returns using asset-specific features, while the second stage incorporates market data to adjust these predictions for broader economic conditions. This results in a generalized model that can be applied across various asset classes, including commodities, cryptocurrencies, as well as synthetic datasets. The predicted quantiles are then converted into full probability distributions through kernel density estimation, allowing for more precise return distribution predictions and inferencing. The LSTM model significantly outperforms a linear quantile regression baseline by 98% and a dense neural network model by over 50%, showcasing its ability to capture complex patterns in financial return distributions across both synthetic and real-world data. By using exclusively asset-class-neutral features, our model achieves robust, generalizable results.

Predicting the S&P 500 index volatility is crucial for investors and financial analysts as it helps assess market risk and make informed investment decisions. Volatility represents the level of uncertainty or risk related to the size of changes in a security's value, making it an essential indicator for financial planning. This study explores four methods to improve the accuracy of volatility forecasts for the S&P 500: the established GARCH model, known for capturing historical volatility patterns; an LSTM network that utilizes past volatility and log returns; a hybrid LSTM-GARCH model that combines the strengths of both approaches; and an advanced version of the hybrid model that also factors in the VIX index to gauge market sentiment. This analysis is based on a daily dataset that includes S&P 500 and VIX index data, covering the period from January 3, 2000, to December 21, 2023. Through rigorous testing and comparison, we found that machine learning approaches, particularly the hybrid LSTM models, significantly outperform the traditional GARCH model. Including the VIX index in the hybrid model further enhances its forecasting ability by incorporating real-time market sentiment. The results of this study offer valuable insights for achieving more accurate volatility predictions, enabling better risk management and strategic investment decisions in the volatile environment of the S&P 500.

In modeling multivariate time series, it is important to allow time-varying smoothness in the mean and covariance process. In particular, there may be certain time intervals exhibiting rapid changes and others in which changes are slow. If such locally adaptive smoothness is not accounted for, one can obtain misleading inferences and predictions, with over-smoothing across erratic time intervals and under-smoothing across times exhibiting slow variation. This can lead to miscalibration of predictive intervals, which can be substantially too narrow or wide depending on the time. We propose a continuous multivariate stochastic process for time series having locally varying smoothness in both the mean and covariance matrix. This process is constructed utilizing latent dictionary functions in time, which are given nested Gaussian process priors and linearly related to the observed data through a sparse mapping. Using a differential equation representation, we bypass usual computational bottlenecks in obtaining MCMC and online algorithms for approximate Bayesian inference. The performance is assessed in simulations and illustrated in a financial application.

Using Betfair's time series data, an analysis of the United Kingdom (UK) horse racing market reveals an interesting paradox: a market with short tails, rapidly decaying autocorrelations, and no long-term memory. There seems to be a remarkably high level of informational efficiency in betting exchange returns, in contrast to financial assets that are characterized by heavy tails and volatility clustering. The generalized Gaussian unconditional distribution with a light tail point to a market where knowledge is quickly assimilated and reflected in prices. This is further supported by the extremely quick fading of autocorrelations and the absence of gain-loss asymmetry. Therefore, in addition to measuring long-range memory, the Hurst exponent also shows mean reversion, a sign that markets respond quickly to fresh information.

This study aims to compare multiple deep learning-based forecasters for the task of predicting volatility using multivariate data. The paper evaluates a range of models, starting from simpler and shallower ones and progressing to deeper and more complex architectures. Additionally, the performance of these models is compared against naive predictions and variations of classical GARCH models. The prediction of volatility for five assets, namely S&P500, NASDAQ100, gold, silver, and oil, is specifically addressed using GARCH models, Multi-Layer Perceptrons, Recurrent Neural Networks, Temporal Convolutional Networks, and the Temporal Fusion Transformer. In the majority of cases, the Temporal Fusion Transformer, followed by variants of the Temporal Convolutional Network, outperformed classical approaches and shallow networks. These experiments were repeated, and the differences observed between the competing models were found to be statistically significant, thus providing strong encouragement for their practical application.

Abstract: We consider the problem of simultaneously approximating the conditional distribution of market prices and their log returns with a single machine learning model. We show that an instance of the GDN model of Kratsios and Papon (2022) solves this problem without having prior assumptions on the market's "clipped" log returns, other than that they follow a generalized Ornstein-Uhlenbeck process with a priori unknown dynamics. Abstract: Graph Neural Networks (GNNs) are often used for tasks involving the geometry of a given graph, such as molecular dynamics simulation. While the distance matrix of a graph contains the complete geometric structure information, whether GNNs can learn this geometry solely from the distance matrix has yet to be studied. In this work, we first demonstrate that Message Passing Neural Networks (MPNNs) are insufficient for learning the geometry of a graph from its distance matrix by constructing families of geometric graphs which cannot be distinguished by MPNNs.

We consider the problem of simultaneously approximating the conditional distribution of market prices and their log returns with a single machine learning model. We show that an instance of the GDN model of Kratsios and Papon (2022) solves this problem without having prior assumptions on the market's "clipped" log returns, other than that they follow a generalized Ornstein-Uhlenbeck process with a priori unknown dynamics. We provide universal approximation guarantees for these conditional distributions and contingent claims with a Lipschitz payoff function.

This study examines the weak form of the efficient market hypothesis for Bitcoin using a feedforward neural network. Due to the increasing popularity of cryptocurrencies in recent years, the question has arisen, as to whether market inefficiencies could be exploited in Bitcoin. Several studies we refer to here discuss this topic in the context of Bitcoin using either statistical tests or machine learning methods, mostly relying exclusively on data from Bitcoin itself. Results regarding market efficiency vary from study to study. In this study, however, the focus is on applying various asset-related input features in a neural network. The aim is to investigate whether the prediction accuracy improves when adding equity stock indices (S&P 500, Russell 2000), currencies (EURUSD), 10 Year US Treasury Note Yield as well as Gold&Silver producers index (XAU), in addition to using Bitcoin returns as input feature. As expected, the results show that more features lead to higher training performance from 54.6% prediction accuracy with one feature to 61% with six features. On the test set, we observe that with our neural network methodology, adding additional asset classes, no increase in prediction accuracy is achieved. One feature set is able to partially outperform a buy-and-hold strategy, but the performance drops again as soon as another feature is added. This leads us to the partial conclusion that weak market inefficiencies for Bitcoin cannot be detected using neural networks and the given asset classes as input. Therefore, based on this study, we find evidence that the Bitcoin market is efficient in the sense of the efficient market hypothesis during the sample period. We encourage further research in this area, as much depends on the sample period chosen, the input features, the model architecture, and the hyperparameters.