arXiv subjects

Manuel Noseda

Publications and source records attributed to Manuel Noseda.

3 recordsLinked to original sources

arXiv2026

Predicting Stock Price Direction on Earnings Announcement Days using Multi-modal Deep Learning

Predicting stock price movements during Earnings Announcements (EAs) is a significant challenge due to market noise and high-impact price discontinuities. In this study, we evaluate whether pre-announcement news sentiment, firm fundamentals, and recent market dynamics jointly predict the directional price movement of equities on EA days. We construct a multi-modal feature space combining 15 fundamental metrics, 3 price-based technical indicators and sentiment scores derived from financial news articles processed using FinBERT. We compare a Long Short-Term Memory (LSTM) network and a Transformer-based architecture against a logistic regression baseline, and further assess all models with and without sentiment features to quantify their incremental value. Our results indicate that while the LSTM demonstrates higher precision through a conservative safe-bet strategy, the Transformer model exhibits superior sensitivity in identifying volatile movements, achieving a higher macro F1-score, with ablation experiments showing a consistent benefit from incorporating news sentiment.

cs.LG↗

arXiv2026

Chirality tomography: measuring local helicity from trajectory linking

We present the first three-dimensional helicity maps of fully developed turbulence obtained through chirality tomography, a Lagrangian voxel-based method that reconstructs helicity density from particle trajectories. Our approach builds on an empirically established relation between helicity and trajectory linking, converting local counts of signed crossings $K$ into volumetric maps of dimensionless helicity, $H(\mathbf{x})$. We demonstrate that the entanglement of particle trajectories, quantified by the mean signed crossing number, provides a robust proxy for helicity, not only at the global scale, but also locally in space and time. Our method can reveal local spatial heterogeneities in helicity and relate them to large-scale flow organization, enabling the reconstruction of spatially resolved chiral structures. Applied to von K\'arm\'an experiments and Taylor-Green direct numerical simulations, the method reveals iso-helicity surfaces and coherent chiral features, while time series of $K$ accurately track the evolution of domain-averaged helicity. The proportionality between $K$ and $H$ remains robust across different voxel geometries and different values of particle inertia, but is not held in laminar or time-modulated flows. This study shows that chirality tomography provides a practical helicity diagnostic in turbulent flows, while establishing a direct bridge between trajectory-level topology and a fundamental dynamical invariant of turbulence.

physics.flu-dyn↗

arXiv2025

Federated Learning for Financial Forecasting

This paper studies Federated Learning (FL) for binary classification of volatile financial market trends. Using a shared Long Short-Term Memory (LSTM) classifier, we compare three scenarios: (i) a centralized model trained on the union of all data, (ii) a single-agent model trained on an individual data subset, and (iii) a privacy-preserving FL collaboration in which agents exchange only model updates, never raw data. We then extend the study with additional market features, deliberately introducing not independent and identically distributed data (non-IID) across agents, personalized FL and employing differential privacy. Our numerical experiments show that FL achieves accuracy and generalization on par with the centralized baseline, while significantly outperforming the single-agent model. The results show that collaborative, privacy-preserving learning provides collective tangible value in finance, even under realistic data heterogeneity and personalization requirements.

cs.LG↗