Extracting the multi-timescale activity patterns of online financial markets

Area of Science:

• Quantitative Finance
• Complex Systems Analysis
• Data Mining

Background:

• Online financial markets exhibit complex dynamics across various time scales.
• Understanding these dynamics is crucial for market stability and risk management.
• Existing methods may not fully capture multi-timescale trading behaviors.

Purpose of the Study:

• To develop a novel methodology for extracting multi-timescale trading dynamics from financial market data.
• To apply Non-Negative Tensor Factorization (NTF) for uncovering hidden activity patterns.
• To characterize trading behaviors and crisis impacts in real-world financial markets.

Main Methods:

• Development of a Non-Negative Tensor Factorization (NTF) based methodology.
• Application to synthetic data for validation.
• Analysis of a large dataset of Italian interbank transactions (e-MID, 2001-2015).

Main Results:

• NTF successfully extracts multi-timescale trading dynamics.
• Identified distinct trading strategies (e.g., normal, early, flash trading).
• Revealed "crisis modalities" in trading during the 2008 financial crisis.

Conclusions:

• NTF is an effective tool for analyzing complex financial market data.
• The methodology uncovers hidden patterns and characterizes market behavior during crises.
• Provides insights into bank trading strategies and systemic risk.