Reproducing kernel Hilbert spaces with odd kernels in price prediction.

Area of Science:

• Quantitative Finance
• Machine Learning
• Econometrics

Background:

• Modeling futures contract prices is crucial for financial markets.
• Existing models often struggle with predictive accuracy and overfitting.
• Conditional price changes in time series require sophisticated modeling techniques.

Purpose of the Study:

• To develop a novel regression model for futures contract price time series.
• To incorporate an odd symmetry constraint within a reproducing kernel Hilbert space framework.
• To enhance predictive accuracy and mitigate overfitting in financial forecasting.

Main Methods:

• Utilized regression in a reproducing kernel Hilbert space (RKHS).
• Introduced and implemented an odd function constraint on the regression.
• Modified the kernel to transform the constrained optimization into an unconstrained problem.
• Derived odd and even kernels from symmetry constraints.

Main Results:

• Demonstrated that the odd kernel constraint can be reduced to an unconstrained optimization problem.
• Showcased the natural emergence of odd and even kernels from symmetry constraints.
• Empirically validated the model on large real-world datasets for four futures contracts.
• Achieved higher predictive accuracy compared to standard methods.
• Observed a reduced tendency for the model to overfit the data.

Conclusions:

• The oddness assumption in the proposed RKHS regression model is valid and practically useful.
• Employing odd kernels significantly enhances predictive performance for futures contract prices.
• The method offers a robust approach to financial time series forecasting, reducing overfitting.