Adaptive Regime Detection via Multi-Layer Binary Modelling
The paper developed a multi-layer binary framework featuring adaptive feature selection to handle high-dimensional, non-linear datasets.
This methodology is directly applicable to Market Regime Detection. By using adaptive feature selection, the model can identify which macro or technical indicators are currently driving market states, allowing for dynamic strategy switching between trending and mean-reverting environments.
Publication
Latent Variable Inference for Asset Correlation & Clustering
The paper built structured clustering and forecasting through latent variable inference to uncover hidden dependencies in complex systems.
The technique is essential for Risk Factor Modelling. This approach allows for the discovery of "hidden" correlations between assets that standard sector classifications miss. It improves the robustness of the Covariance Matrix, leading to more stable Risk Parity and Minimum Variance portfolios.
Publication
Data-Driven Noise Suppression for Volatility Estimation
Novel signal modelling techniques focused on data-driven noise suppression and signal enhancement.
The work translates to Volatility Filtering. In high-frequency data, the "true" volatility signal is often obscured by microstructure noise.
This research provides a mathematical basis for more accurate parameter estimation in GARCH-family models and stochastic volatility frameworks.
Publication
Multi-Modal Integration for Systemic Risk Modelling
A framework for feature-engineered prediction using multi-modal integration, combining disparate data sources into a unified predictive model.
The work is applicable to Systemic Risk and Tail-Risk monitoring.
By integrating heterogeneous data—such as credit spreads, liquidity metrics, and sentiment—this framework can identify cross-asset contagion risks that a single-source model would overlook.
Publication
Spectral Signal Recovery for High-Frequency Microstructure
Analysis of non-stationary signals using Short-Time Fourier Transform (STFT) for robust signal recovery.
The work is highly relevant for High-Frequency Trading (HFT) and Market Making. Spectral analysis allows for the detection of periodicities in the Limit Order Book (LOB) and the identification of execution patterns or "iceberg" orders that operate on specific frequency domains.
Publication