Financial markets are complex, interconnected and deterministically chaotic systems in which the price of a stock may be influenced by the economic factors of other stock markets. Thus, a variety of methods that aim to analyze future market behavior have been developed. For example, technical analysis aims to identify, model, extrapolate and combine financial market patterns to improve predictions. In contrast to technical analysis, algorithmic trading has become popular in asset management and trading as it minimizes cost, market impact, and risk. Although these methods have proven useful when analyzing financial markets, it is difficult for investment management and hedge funds to earn consistent annual returns on a systematic basis.
In addition to the strategies mentioned above, machine learning approaches have proven useful to analyze financial data. However, unlike traditional classification and regression problems, predicting these time series requires consideration of the interdependency between stock markets, meaning that all agents involved in a given stock market may exhibit interconnections and correlations, representing important internal forces of the market. That is, the movement of a stock market in a country is likely to be affected by movement of other stocks in both that country and in other regions. These interconnections, also known as interdependencies between stock markets, can be valuable in predicting stock prices.
Previous studies focus exclusively on investigating how the price of one stock is influenced by the economic factors and prices of other stocks; but their models do not consider changes in network structure over time, meaning that the profitable conditions for which the models were optimized may disappear. Thus, we focus on the prediction of stock prices within a stock market interdependence approach using machine learning techniques, avoiding expensive and time-consuming data transportation into an integrated, central data store. Such a distributed learning paradigm is especially critical for big data analysis and real-time learning.
Preliminary results show that the interdependence strategy may enhance prediction accuracy for stocks traded in Germany, representing an advantage over compared methods. In particular, although the US contributes significantly to the global trade volume, our results show that the models learned from stocks traded in Germany are more appropriate to predict other stock markets. This suggest that the US market is more influenced by the European market and not vice versa, which is inline with previous empirical findings. However, to ensure consistency in the reported results, this study is being expanded to additional stock markets and portfolios of financial assets such as commodities and bonds.
More details about this research can be found in the paper “Stock Price Prediction using Kernel Adaptive Filtering within a Stock Market Interdependence Approach” (https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3306250)
Sergio Garcia Vega is based at The University of Manchester 2016-2019, and his research project is Distributed and Real-time Machine Learning for Financial Data Analysis (WP1)