Extracting market events from limit order book data : a data-driven approach to financial feature engineering
Εξαγωγή συμβάντων αγοράς από δεδομένα "limit order book" : μια δεδομενοκεντρική προσέγγιση για τη μηχανική χαρακτηριστικών στη χρηματοοικονομική ανάλυση
View/
Keywords
Limit order book ; High frequency trading ; Market microstructure ; Market dynamics ; Feature engineering ; Financial forecastingAbstract
Financial markets have evolved into highly complex systems, where the intricate dynamics of
trading are captured in the limit order book (LOB), a critical component of market microstructure.
This thesis explores the challenge of analyzing LOB data to infer meaningful patterns and predictive
signals, leveraging machine learning techniques to address the inherent complexity and non-linearity
of the data. Our work focuses on bridging the gap between raw LOB snapshots and the more detailed
market by order (MBO) data, which offers richer insights but is often unavailable or underutilized in
academic research.
To this end, we developed a greedy algorithm to reconstruct synthetic MBO data from high
frequency LOB snapshots. This reconstructed dataset serves as the foundation for constructing a
feature vector set inspired by prior studies, incorporating inferred market event data to compute
time-sensitive features aimed at capturing temporal dynamics. The resulting features aim to provide
a deeper understanding of market activity and support applications such as alpha signal prediction,
a crucial element in financial modeling.
By simplifying access to granular market information and demonstrating the feasibility of ex
tracting temporal dynamics from raw LOB data, this thesis lays the groundwork for future work in
predictive modeling and feature validation. Experimental results demonstrate the potential of this
method to contribute to trading strategies and market analysis while highlighting avenues for further
exploration and optimization.