Regulatory, institutional and technological change in investment research

Abstract

This thesis comprises four studies, each designed to form an independent contribution. The central theme is change in the provision of investment research. The study focuses on independent investment research firms. The first study documents the mechanisms used to pay for investment research. Through interpretation of documents and event participation, I use qualitative analysis to explain the payment mechanisms used in recent decades and how these mechanisms have changed. A key finding is that since 2017 many investment-management firms have shifted from an opaque, reciprocal arrangement, which resembles a gift exchange economy, towards a neoclassical economic model. Investment management and research firms are adapting to change in payment mechanisms at a time of falling asset management fees and commission rates. One response is to automate. While many firms are exploring machine learning, little evidence on the potential for such approaches exists in the public domain. The second and third studies in this thesis address this gap by evaluating the effectiveness of machine learning in the investment research function. Both studies compare a large, global sample of analyst and machine learning outputs. The second study evaluates the relative effectiveness of analyst and machine learning valuations in predicting future returns. Analysts make unbiased valuations using a standardized discounted cash flow model. Although neither analyst nor machine learning valuations serve as viable predictors of subsequent returns, when used together they offer credible explanatory power which can be increased by adding a price momentum factor. Analyst and machine learning accuracy could be improved by placing greater emphasis on past returns. The third study uses regression analysis to compare analyst and machine learning risk assessments using next quarter volatility as the outcome variable. Both assessments can benefit by incorporating information from the other, and both are more accurate in countries considered to have superior informational environments. Analysts seem to underestimate risk associated with their own “buy” recommendations, but no equivalent miscalibration is apparent in machine learning predictions. The results confirm both the value of analysts’ research and considerable potential for machine learning in financial analysis. The final study summarizes the informational environment in light of regulatory, institutional and technological change. Stock coverage by analysts provides a window on the level of information available to investors. Archival and case study analysis indicates that analysts are providing research on most large and mid-cap listed companies in the US and UK, i.e., stock coverage remains wide. Coverage also remains deep, with most companies receiving coverage from a similar number of analysts and few companies covered by only one analyst. The exception is that fewer companies are now covered by more than 20 analysts, indicating that some surplus research has disappeared. Independent research has expanded over the past decade. Despite this, the tendency for analysts to provide optimistic recommendations persists. Independent analysts rarely contribute to archives and are redefining coverage models. Analyst forecast datasets assembled by established data vendors understate the wider selection of research which is now available, but only to those who can find and afford it. Taken together, there is little evidence to date of a diminished informational environment for equity investors.