Multivariate and Online Prediction of Closing Price Using Kernel Adaptive Filtering

Area of Science:

• * Computational Finance
• * Machine Learning
• * Time Series Analysis

Background:

• * Traditional stock price prediction models rely on batch-oriented, independent data processing.
• * Existing methods often fail to capture the dynamic, real-time nature of financial markets.
• * The need for adaptive, online learning algorithms in financial forecasting is significant.

Purpose of the Study:

• * To propose and evaluate a multivariate, online stock price prediction framework using kernel adaptive filtering (KAF).
• * To challenge the conventional reliance on batch learning for stock market analysis.
• * To investigate the efficacy of various KAF algorithms on granular, high-frequency trading data.

Main Methods:

• * Implementation of ten distinct kernel adaptive filtering (KAF) algorithms for online stock price prediction.
• * Utilizing granular data across multiple time windows (1 minute to 1 day).
• * Testing the proposed framework on 50 stocks from the Indian Nifty-50 index.

Main Results:

• * Demonstrated the effectiveness of online learning and KAF for stock price prediction.
• * Validated the applicability of the proposed KAF approach on diverse time-granularity data.
• * Showcased the potential for deploying KAF in high-frequency trading environments.

Conclusions:

• * Online kernel adaptive filtering (KAF) offers a viable alternative to traditional batch methods for stock price prediction.
• * The proposed KAF framework is suitable for analyzing granular market data across various trading time windows.
• * KAF methods show practical deployment potential for high-frequency trading strategies.