Internal Relevance between Analysts’ Forecasts and Target Prices - Informativeness and Investment Value
Abstract
ABSTRACT Analysts’ decision-making process, through which they use earnings and non-earnings forecasts to provide their target price revisions, remains opaque. Based on the decision tree analysis, we develop a new multivariate information metric, Internal Relevance (IR), to measure the extent to which the revisions of analysts’ consensus earnings and sales forecasts are incorporated into the revisions of consensus target prices. We show that stocks with higher IR have stronger market reactions to target price revisions in terms of abnormal return, abnormal trading volume, and abnormal return volatility. These results remain consistent across a series of robustness checks. Finally, we perform portfolio analysis to show that IR can be used to screen stocks to improve trading profitability. **KEYWORDS:** Decision tree analysis, earnings and sales forecasts, market reactions, target prices, portfolio analysis **JEL CLASSIFICATION:** G10, G24, G14, G11 **Disclosure statement** No potential conflict of interest was reported by the author(s). **Notes** 1. In this paper, we focus on consensus outputs rather than individual analysts’ outputs for two reasons. First, IR is conceptualized as a stock-level, rather than analyst-level, characteristic. Second, the limited number of observations at the analyst level may result in unreliable estimation for IR. It should be noted that earnings forecasts, sales forecasts, target prices, and their revisions in the remainder of the paper are consensus, unless otherwise noted. 2. Brown and Huang (Citation2013) define a pair of recommendation and earnings forecast to be consistent if both are above (or below) the consensus level. Unlike Brown and Huang (Citation2013), Iselin, Park, and Van Buskirk (Citation2021) define analyst consistency based on three outputs: earnings forecasts, stock recommendations, and target prices. An analyst is identified consistent if the analyst revises all three outputs upward (or downward) compared with the prior estimates. 3. Indeed, several studies consider more than two analysts’ outputs (Asquith, Mikhail, and Au Citation2005; Bradshaw, Brown, and Huang Citation2013; Brav and Lehavy Citation2003). However, the major foci have been on the investigation of the incremental informativeness of one output relative to others, i.e., the market reaction to target price revisions controlling for stock recommendation and earnings forecast revisions. In contrast, we explore the contemporaneous interconnections between two intermediate outputs and one terminal output. 4. We treat target price revisions as a categorical variable since the CART algorithm, in modelling analysts’ revisions, performs better in classification tree than in regression tree. Decision tree analysis is prone to overfitting. To address that issue, we specify a relatively large cost-of-complexity value to generate an initial tree and then perform the pruning process by checking whether the tree is associated with the minimum of cross-validation errors. Our results are robust regardless of whether the pruning process is performed or not. 5. We thank the anonymous reviewer for pointing this out. 6. The literature documents the impacts of regulatory changes on analyst behaviour such as Reg FD, NASD Rule 2711, NYSE Rule 472, e.g. see Chan, Lo, and Su (Citation2014); Mohanram and Sunder (Citation2006); Palmon and Yezegel (Citation2011). Reg FD was implemented in October 2000 and NASD Rule 2711 (NYSE Rule 472) was initially adopted in 2002–2003. 7. The consensus target price revision event refers to the date when consensus target price changes. It can be triggered either when one or more analysts revise their own target prices, or when one or more analysts stop issuing the target prices, or when one or more analysts start to follow the stocks and issue new target prices, or a combination thereof. 8. We thank the anonymous reviewer for suggesting this approach to address the issue of sample selection bias. 9. We obtain the weekly returns and factors extracted from Kenneth R. French data library at https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html. 10. Similar to our decision tree analysis, we rescale the change of weekly consensus by stock price to make the calculation comparable. To obtain reliable regression fits, we further require that each stock-year sub-sample contain at least 20 non-missing observations. 11. The definition of (in)consistency is different from those in the literature in several ways: first, existing literature examines the consistency of a pair of revisions issued by the same analyst, while ours is based on the joint movement of the weekly consensus; second, existing literature treats (in)consistency as a binary variable to capture the directions of the link, while ours focuses on the proportion of (in)consistency to capture the strength of the link.
Faculty Members
- Jahangir Sultan - Department of Finance, Bentley University, Waltham, MA, USA
- Wenxiu (Vince) Nan - School of Business, Salisbury University, Salisbury, MD, USA
- Tao Li - School of Business, SUNY New Paltz, New Paltz, NY, USA
Themes
- Portfolio Management and Trading Strategies
- Impact of Consensus Earnings and Sales Forecasts
- Analyst Decision-Making Processes
- Market Reactions to Forecast Revisions
- Development of Financial Metrics