Anchor Reversion: The Case of the 52-Week High and Asset Prices

AbstractThis study attempts to jointly identify the effects of anchoring and mean reversion on asset prices. In particular, we define “anchor reversion” as the tendency for stock prices to revert back toward an anchor, for example, the 52-week high. Our results show that the further a stock moves away from its 52-week high, the more likely it is to revert back toward that 52-week high in the following month. Conversely, if a stock moves toward its 52-week high in a given month, it is less likely to experience a large movement toward that 52-week high in the following month. Portfolios constructed according to a long-short anchor reversion strategy result in future returns that range from 1.40% to 1.62% per month. These results are robust to controls for various risk factors as well as several cross-sectional stock characteristics, such as the monthly reversal phenomena.Keywords: Anchoringasset pricingbehavioral financemean reversionJEL Codes: G10G19G40G41 Disclosure statementNo potential conflict of interest was reported by the authors.Notes1 Baker, Pan, and Wurgler (Citation2012) find that prior stock-price peaks act as reference points and affect several aspects of mergers and acquisitions. In particular, the authors find that M&A offer prices are biased toward recent price peaks and that the probability of a targets’ acceptance of an acquirers’ offer is higher if the offer price is above the price peak.2 Much of the volatility forecasting literature is based on the notion of mean reversion (see, e.g., Engle, Citation1982; Bollerslev, Engle, and Wooldridge Citation1988; Bollerslev Citation1990; Bollerslev, Chou, and Kroner Citation1992; Bollerslev and Engle Citation1993; Engle Citation2002b).3 In column [5], without including controls, both the coefficients on Away and Closer are negative and significant. The difference between the two coefficients is 0.0079, but the t statistic, testing the significance of that difference, is only 0.79, suggesting that the two estimates are statistically similar.4 These results are available from the authors upon request.