Understanding Rankings Of Financial Analysts
Universidade do Porto – Faculdade de Economia (FEP)
Universidade do Porto – Faculdade de Economia (FEP)
Carlos Soares
Universidade do Porto – Faculty of Engineering
November 9, 2015
Abstract:
The prediction of the most accurate analysts is typically modeled in terms of individual analyst characteristics. This approach has the disadvantage that these data are hard to collect and often unreliable. We follow a different approach in which we characterize the general behavior of rankings of analysts based upon state variables rather than individual analyst characteristics or past accuracy. We use a common learning algorithm, naive Bayes, that we adapted to address the problem of ranking the analysts. Our results show that it is possible to model the relation between the selected variables and the rankings. We show that the uncertainty about future stock performance influences the rankings of the analysts while the macroeconomic variables have the most contribution to the changes in rankings.
Understanding Rankings Of Financial Analysts – Introduction
The Efficient Market Hypothesis (EHM) (Fama, 1970) suggests that all public information available to investors is incorporated in prices and new information is immediately reflected in valuations. Yet there are information gathering costs and financial analysts are better than an average investor at processing this information which reflects in issued buy/ sell recommendations. These recommendations, like other news about the general economy as well as about a particular company, influence investors’ perception and beliefs.
Previous studies show that analysts stock recommendations have investment value (Womack, 1996; Barber, Lehavy, McNichols, and Trueman, 2001). The literature also suggests further that foreknowledge of analyst forecast accuracy is valuable (Brown and Mohammad, 2003; Aiguzhinov, Serra, and Soares, 2015). In line with academic research findings, practitioners too pay attention to analyst forecast accuracy rankings. On an annual basis, firms such as The Institutional Investor and StarMine1 publish analysts ratings according to how well they performed, based partly on past earnings forecast accuracy.
The importance of these ratings should not be ignored because the attention that the market gives to the recommendations of different analysts is expected to correlate with them. Typically, the performance of analysts is analyzed in terms of their individual characteristics (e.g., experience, background) (Clement, 1999). The disadvantage of this approach is that the collection of the necessary data is difficult and it is not always reliable. As for practitioners, they rely mostly on past accuracy to predict future accuracy.
In this paper we follow an alternative approach. We model the general behavior of rankings of analysts by using variables that characterize the context (state variables) rather than individual analyst characteristics. The model we propose uses the state variables to distinguish which of them affects the rankings the most; hence, influence the analysts’ forecast accuracy. In summary, our goal is not to understand relative performance of the analysts in terms of their characteristics but rather in terms of the characteristics of the context in which the analysts operate.
To achieve this goal, we, first, build rankings of analyst based on their EPS forecasts accuracy. Then, we select the state variables that are responsible in differences of analysts’ rankings. Finally, we apply a Machine Learning label ranking algorithm to build a model that relates the rankings with the variables and calculates a discriminative power of a variable.
The paper is organized as follows: Section 2 provides the motivation for using rankings of the analysts; Section 3 outlines the state variables that characterize the context; Section 4 outlines the structure of the Machine Learning label ranking model and presents a methodology of building a “variable-ranking” relation; Section 5 describes the datasets used for the experiments; Section 6 summarizes the experiment setup; Section 7 presents and discusses the results; finally, Section 8 concludes this paper.
Rankings as a measure of accuracy
In spite of the Efficiency Market Hypothesis, it is commonly accepted that the recommendations of financial analysts yield an economic value to investors (Womack, 1996); moreover, recommendations of superior analysts have impact on the market (Loh and Stulz, 2011). For this reason, researchers and practitioners have long been interested in understanding how financial analysts affect capital market efficiency (Ramnath, Rock, and Shane, 2008).
Most researchers conclude that financial analysts are better at making EPS forecasts than mathematical models. Specifically, Fried and Givoly (1982); Bouwman, Frishkoff, and Frishkoff (1987); Brown and Kim (1991) show that analysts are better at forecasting EPS values than any time series models (e.g., ARIMA). The analysts superiority is contribute to the fact that they utilize all available information at and after the date of time series model forecasts. Thus, the context in which analysts make decision matters for their accurate forecast.
See full PDF below.
