Overreaction: the sensitivity of defining the duration of the formation period

Abstract

Prior research provides evidence suggesting that losers tend to continue to be losers in the short and medium term, whereas there is evidence showing that losers outperform winners in the long term. However, there are some differences in methodology used in the studies, particularly in their definition of the duration of the formation period as well as the definition of returns. This paper aims to investigate the underreaction/overreaction hypothesis and in particular to examine the sensitivity of defining the duration of the formation period. The results confirm that losers tend to continue to be losers when cumulative excess return is calculated over short and medium periods. There is evidence, however, suggesting that losers outperform winners when cumulative excess returns are calculated over a long period, even after six months up to five years of portfolio formation. Furthermore, the results show that neither the size effect nor the January effect has a role in explaining the difference in returns between winners and losers. Moreover, the difference in returns between winners and losers cannot be attributed to change in risk or to change in illiquidity. However, I provide evidence that some part of the winner–loser effect can be attributed to leverage effect.

Notes

¹ Another difference is related to the definition of winners and losers. For example, DeBondt and Thaler (1985, 1987) define winners and losers as being the best and worst 35 performing stocks, respectively. Zarowin (1990) considers the top and bottom quintiles, whereas Kryzanowski and Zhang (1992) look at the extreme deciles.

² The stock market overreaction hypothesis asserts that stock prices take temporary swings away from their fundamental values due to waves of optimism and pessimism (Conrad and Kaul, 1993).

³ These are: the US, the UK, Japan, Canada, Germany, France, Australia, Switzerland, Norway, Denmark, Austria, Hong Kong, Italy, the Netherlands, Spain and Sweden.

⁴ These returns are adjusted for stock dividends, stock split and dividend yields.

⁵ Galariotis (2004) shows that the ‘selection of a value weighted or an equally weighted index does not alter the main findings’.

⁶ This could be used to test for the sensitivity of using such a method compared to quintiles.

⁷ Following Gregory et al. (2004) for each portfolio, first, I construct an image portfolio called the size-adjusted portfolio in which stocks are sorted based on the market value at the end of the previous year. Then the average return for each quintile is computed, referred to as the size-quintiles-return. Second, for the same portfolio I construct another image portfolio called the benchmark portfolio in which stocks are sorted based on the cumulative excess return. Then, for each stock in this portfolio I identify its corresponding size-quintiles-portfolio and its size-quintiles-return. Third, the size-adjusted average return for each quintile is then computed as the difference between its raw return and its size benchmark return.

⁸ Note that the fiscal year-end for all companies listed on Amman Stock Exchange is the end of December. Thus, I allow a four-month gap to ensure that the data are available at the formation date.

⁹ I use the median of the market since the number of the companies over the sample period is just 232.

¹⁰Amihud and Mendelson (1986) investigate the effect of illiquidity on stock returns, measured by the bid–ask spread. They find that expected returns are increasing in the relative bid–ask spread and that there is a clientele effect. This result suggests that investors with longer time horizons will prefer assets with higher transaction costs. Further, they point out that this relation is not an anomaly of market efficiency; rather it is due to rational investor behaviour.

¹¹ Note that for each year and for each stock I calculate the average trading volume over the past five years, then stocks are sorted in an ascending order.

¹² I use the total assets divided by shareholders equity to proxy for leverage.

¹³ Note that I limit the discussion on the LMW portfolios. However, considering the WMB portfolios yield to the same final conclusion.

¹⁴ Also note that I calculate the equally weighted returns. But I limit my discussion to the value-weighted returns. Both methods yield to the same final conclusion. Results are available upon request from the author.

¹⁵ They find that ‘continuation’ is a feature of the Canadian equity market rather than overreaction.

¹⁶ The results are not tabulated here, but available upon request from the author.

¹⁷ Except for the two-year strategy, where the coefficient is statistically significant at 0.10 level.