Three essays on investor confidence: a new measure, empirical application and experimental evidence

This PhD research is committed to contributing to the literature on investor overconfidence, one of the most robust findings in the field of behavioural finance. Overconfidence, a cognitive bias where decision makers tend to be overly optimistic not only about their aptitudes and skills, but also about the precision of their forecasts and information, is associated with poor decision making. Individuals suffering from overconfidence tend to be excessive stock traders, Chief Executive Officers (CEOs) who rush into mergers and acquisitions, risky drivers, naïve entrepreneurs and sloppy retirement planners. The literature yields the many attempts to link stock market phenomena to overconfidence. However, existing measures that have been used to test these hypotheses are typically only loosely related to the overconfidence of investors in their own abilities, or use proxies that lack a formal model of cognitive psychology. In the first of three research projects, I propose a measure of aggregate investor confidence that is based on a cross-disciplinary model containing determinants of confidence. The measure captures major economic events intuitively, and is statistically distinct from exiting proxies. Using a 1926-2011United States (US) sample, I find that the new measure is a better predictor of aggregate trading activity than past stock returns, which have been used in prior studies.The second research project explores the role of aggregate investor confidence in asset pricing factors. Empirical tests reveal interesting patterns. Firstly, and in line with a behavioural model by Daniel, Hirshleifer, and Subrahmanyam (1998), aggregate investor confidence partially explains variations in the profitability of momentum strategies. Additionally, aggregate investor confidence appears to play a key role in the size factor, complementing an early hypothesis by Roll (1981). Indeed, investors seem to systematically change their risk perceptions which ultimately impacts on market outcome. The third research project takes a qualitative stance. Using a new methodology proposed by Glaser, Langer, and Weber (2013), we utilise the ability to assess time series variations of individual overconfidence levels in an experimental asset market. We find that arriving signals that strongly support prior decisions cause overconfidence to prevail, while strongly opposing signals cause the effect to vanish 'overconfidence crashes'. However, previously lost overconfidence can re-emerge when these opposing signals reverse .Additionally, we find strong evidence in favour of the hypothesis by Hongaund Stein (2007) which states that investors interpret arriving information differently with opposing feedback having particularly strong effects. We also find measurement bias in the methodology proposed by Glaser et al. (2013). This is consistent with methodological concerns documented by Langnickeland Zeisberger (2016) and Biais, Hilton, Mazurier, and Pouget (2005) who report that assessment tasks using confidence intervals typically yield inflated overconfidence scores, as individuals tend to be insensitive to confidence levels in their estimations.