A computing platform for pairs-trading online implementation via a blended Kalman-HMM filtering approach

Area of Science:

• Quantitative Finance
• Financial Engineering
• Algorithmic Trading

Background:

• Pairs-trading strategies require efficient real-time platforms to capture stock return spread dynamics.
• Modeling heavy-tailed, mean-reverting spread processes and time-varying parameters is crucial for accurate trading.
• Existing methods often lack integration, hindering effective execution in dynamic market environments.

Purpose of the Study:

• To design an efficient, automated platform for real-time pairs-trading implementation.
• To effectively model and capture the stylized features of stock return spread processes.
• To develop a method for optimal recovery of time-varying parameters in return-spread models.

Main Methods:

• Fusion of Kalman filtering and hidden Markov model (HMM) multi-regime dynamic filtering approaches.
• Integration of HMM's expectation-maximization algorithm with Kalman filtering for automated parameter estimation.
• Development of a hybrid signal-processing algorithm incorporating data fusion techniques.

Main Results:

• The proposed method provides self-updating optimal parameter estimates for pairs-trading.
• Back-testing on Coca-Cola and Pepsi stock data demonstrated potential for non-zero profits (without transaction costs).
• Trading simulations confirmed the method's success and highlighted potential gains with low transaction fees.

Conclusions:

• The hybrid filtering method offers a novel and effective approach for pairs-trading actualization.
• Data fusion plays a critical role, advancing predictive analytics for big financial datasets.
• The strategy shows significant potential for profit in markets with low transaction costs.