부산대학교 도서관

Investors group assets into different classes based on some similarity among them. For example, stocks can be categorized into broad classes such as small versus large stocks, value versus growth stocks, prior winners versus losers, or categorized by different industry sectors. The asset classes are called “styles” and the process allocating money among styles is called “style investing” and investors must consider styles because portfolio allocation among different styles is required by law. Much of academic literature has shown that certain styles outperform other styles in the long run. In particular, small-cap (value) stocks outperformed large-cap (growth) stocks historically. However, the relative performance between these styles is not stable over time. Chan et al., (2000), for example, show that large-cap (growth) stocks outperform small-cap (value) stocks in 13 years (8 years) out of their 29 year sample period from 1970 to 1998. Style based strategy can produce long periods of poor performance. Thus, style rotation strategy, switching from one style to another, could generate additional returns when we can forecast the relative performance between styles. In this study, we examine seasonal patterns in the cross-section of expected returns on twelve style portfolios that we composed from daily return data for stocks listed in the Korean Stock Exchange (KRX). We find that style returns exhibit substantial variations across calendar months. For example, over the sample period of January 1982 to December 2013, in January the mean return of the Small/Neutral portfolio is 6.2 percent and that of the Big/Up portfolio is only 2.3 percent. In March, however, the mean return of the Small/Neutral portfolio is -0.6 percent and that of the Big/Up portfolio is 2.5 percent. Our finding is consistent with previous literature on seasonality in stock returns which suggests the outperformance of some style against another in a specific calendar month. For example, Keim (1983), Reinganum (1983), and Roll (1983) find that small-cap stocks outperform large-cap stocks in January. Branch (1977) and Dyl (1977) suggest that tax-loss selling creates a downward price pressure on loser stocks in December and a price rebound in January. Lakonishok, Shleifer, Thaler, and Vishny (1991) find that pension funds dump prior loser stocks at the end of every quarter. However, these studies explored only the turn-of-the-year period or the end of each quarter. Our finding shows that the seasonal pattern of the style returns is not limited to January or the end of each quarter. Small stocks perform poorly in August and the Big/Down portfolios beat the market in July. We employed the style rotation strategy suggested by Choi(2014) using the seasonal patterns among style returns. We take the long positions of styles with good performance in a specific calendar month and the short positions of styles that have done poorly in the same calendar month. For example, we rank the twelve style portfolios according to their average returns during the previous five Januaries to construct a zero investment portfolio for the next January. We repeat this for each of twelve calendar months. The strategy yields profits across all calendar months except August. Specifically, the mean profit in January alone is 5.8 percent. Overall, our seasonal strategy yields economically and statistically significant profits of 14.2 percent per year. The possible source of the profit from our strategy is seasonal autocorrelation in style returns (predictability component) or cross-sectional variation (dispersion component) in mean style returns as suggested by Lo and MacKinlay (1990) and Conrad and Kaul (1998). The decomposition of the profit shows that the main source of the profit is the predictability component. The predictability component explains more than 90 percent of the profit in every calendar month. In order to check the robustness of seasonal regularities in style returns and the profitability of our zero-investment strategy, we extended our approach to the sector portfolios. The strategy based on the one-year historical performance in each month yields profits of 21.7 percent in January and 28 percent annually but they are not statistically significant due to the large volatility. However, when we construct the strategy with 5 or 10 year historical performance in each month, the strategy yields annul profit of higher than 20 percent, which is statistically significant. The decomposition of the profit shows that the main source of the profit is the predictability component as well. Therefore, the seasonal patterns among style returns have significant power to forecast future relative style performance, which would be inconsistent with the efficient market hypothesis.