Text-Based Correlation Matrix in Multi-Asset Allocation

TL;DR

This paper proposes a novel method using financial news and central bank texts analyzed through natural language processing to estimate and predict asset correlation structures, addressing limitations of traditional time series methods.

Contribution

It introduces a text-based correlation estimation approach that improves prediction accuracy over conventional historical data methods in financial asset management.

Findings

Text analysis improves correlation prediction accuracy.

Method captures correlation changes during economic phase shifts.

Outperforms traditional time series-based prediction methods.

Abstract

The purpose of this study is to estimate the correlation structure between multiple assets using financial text analysis. In recent years, as the background of elevating inflation in the global economy and monetary policy tightening by central banks, the correlation structure between assets, especially interest rate sensitivity and inflation sensitivity, has changed dramatically, increasing the impact on the performance of investors' portfolios. Therefore, the importance of estimating a robust correlation structure in portfolio management has increased. On the other hand, the correlation coefficient using only the historical price data observed in the financial market is accompanied by a certain degree of time lag, and also has the aspect that prediction errors can occur due to the nonstationarity of financial time series data, and that the interpretability from the viewpoint of fundamentals is a little poor when a phase change occurs. In this study, we performed natural language processing on news text and central bank text to verify the prediction accuracy of future correlation coefficient changes. As a result, it was suggested that this method is useful in comparison with the prediction from ordinary time series data.