Identifying new classes of financial price jumps with wavelets

Area of Science:

• Quantitative Finance
• Time Series Analysis
• Computational Economics

Background:

• Stock price jumps are critical events in financial markets.
• Distinguishing between exogenous (news-driven) and endogenous (market-driven) jumps is essential for risk management.
• Existing methods may not fully capture the complex dynamics of jump behavior.

Purpose of the Study:

• To develop an unsupervised classification framework for stock price jumps.
• To identify key features differentiating jump types.
• To investigate the drivers of stock price cojumps.

Main Methods:

• Utilized a multiscale wavelet representation of time-series data.
• Applied unsupervised classification to stock price data.
• Analyzed features such as volatility asymmetry, local mean-reversion, and trend.

Main Results:

• Confirmed time-asymmetry of volatility as a primary differentiator between exogenous and endogenous jumps.
• Identified local mean-reversion and trend as additional key features for jump classification.
• Discovered that a significant proportion of stock cojumps are driven by endogenous contagion mechanisms.

Conclusions:

• The wavelet-based framework effectively classifies stock price jumps.
• New classes of jumps can be identified using identified features.
• Endogenous contagion plays a surprisingly significant role in stock cojumps.