Hunting the quicksilver: Using textual news and causality analysis to predict market volatility

Abstract

This paper proposes that the dynamics of bond volatility may be understood by studying textual news sentiments. In this new approach, a modified framework is used to understand the atypical characteristics of bond market news. The paper proceeds in two steps. First, a word list of sentiment terms is generated using three sentiment word lists to determine negative and positive news sentiment scores. Second, four measures of volatility are estimated and combined with a nonlinear technique adapted from information theory to understand the correlation and direction of causality between sentiment scores and measures of volatility. This paper shows that sentiments extracted from textual news published in the newspapers can explain bond returns volatility or the quicksilver. The empirical results support that news sentiment is highly correlated with the measures of volatility and that information flows unidirectionally from news to volatility. This study, perhaps the earliest work in text mining to examine the run of causality between news signals and bond return volatility, adapts a nonlinear technique from information theory to describe the nonlinear behavior of Indian debt markets and understand the volatility dynamics of the benchmark bond.