Moral Sentiment: Investigating the Roles of Ethics and Affect in Determining Asset Returns

Abstract

Research in behavioural finance claims that markets are inefficient, in the sense that market prices can deviate from their fundamental value. Proponents of this theory suggest that investors succumb to emotional and psychological biases that can sometimes cause them to make irrational decisions. This irrational behaviour can lead them to overvalue or undervalue an asset beyond its fundamental value. Work in sentiment analysis suggests that an insight into investors' thought processes can be gleaned by analysing the texts they typically read. It is believed that affective language functions as a proxy for investor sentiment, and that affective news contains information that has not been incorporated into market prices. Typically the categories of positive or negative affect are used, describing the bullishness or bearishness of a market at a particular point in time. However, a great deal of this negative and positive language tends to be rooted in ethical judgements. Words like "good" and "bad", for instance, have underlying ethical connotations as well as evaluative ones. In this thesis I postulate that negative evaluation is rooted in ethical judgements, and therefore I explore whether an investor sentiment proxy can be constructed using ethical terms.

This research therefore explores the relationship between ethical and affective language, and their respective roles in determining asset prices. The aim is to see whether ethical language can describe variability in market returns, that cannot be described using other market variables. To explore this relation I develop a text analysis system, which constructs an ethical and affect time series from unstructured texts. These time series are subsequently used in an econometric framework, to explore their impact on financial returns. Overall, I observe that ethical and affect terms can explain variability in financial returns that cannot be accounted for by any other market variables. The extent of this relationship depends on the data being used, as well as the time period under investigation. The relationship is also observed to be episodic in nature, suggesting that the impact of news on financial returns is confined to particular periods.