Return reversal in UK shares Glen C. Arnold 1 Salford Business School, The University of Salford, Salford, Greater Manchester M5 4WT Tel: 0161 295 4305 Fax: 0161 295 5556/5022 [email protected] Rose D. Baker Salford Business School, The University of Salford, Salford, Greater Manchester M5 4WT 0161 295 3861 [email protected] Executive summary A crucial question in business and finance is whether stock markets are efficient at pricing shares. If they are not, then investors will be put off buying shares, society’s scarce resources will be misallocated and managers will receive false signals from the market leading to a number of impacts, from the miscalculation of cost of capital to the demoralisation associated with perverse market reactions the announcement of good performance. This paper demonstrates a serious under pricing of those UK shares that have experienced large falls over the previous five years, and a serious over-pricing of those shares that have risen the most over the previous five years. By buying the ‘losers’ (the 10% worst fallers) an investor has been able to out-perform the market by an average of 8.9% per year over the subsequent five years. On the other hand buying the ‘winner’ portfolio (the best 10% of performers over five years) led to a market lagging performance over the next five years – by 47%. This general pattern is shown for each decade since 1960. One potential explanation for this remarkable result is that the losers are subject to a higher level of risk than the winners and so the differential is due to the market rationally requiring a risk premium for holding losers. We test for risk in seven ways and find that losers are, in fact, lower risk than winners in all the risk tests (propensity to liquidation, reliability, performance in different market/economic states, CAPM-beta, standard deviation, Fama-French three factors). A long recognised phenomenon of financial markets is for small market capitalisation shares to out-perform their larger brethren. Perhaps it is possible that losers out-perform because they tend to be smaller companies (they have generally lost around 95% of market capitalisation of five 1 The author to whom proofs and reprint order forms should be sent: Glen Arnold, Salford Business School, Maxwell Building, The University of Salford, Salford, Greater Manchester, M5 4WT 1 Electronic copy available at: http://ssrn.com/abstract=998418 years). We test for this by separating out the small firm effect from the return reversal effect using two distinct techniques. We find both a strong small firm effect and a separate return reversal effect over the 39 years of the study period. A number of researchers propose that psychological/behavioural and social/institutional factors result in investors neglecting losers and hyping winners leading to ‘an overreaction effect’. Individuals when revising their beliefs, tend to overweight recent information and underweight older (or ‘base-rate frequency’) data. They often fall victim to the representative heuristic under which an individual judges the likelihood of a future event by the similarity of the present (recent) evidence to it. In behaving this way decision takers fail to properly allow for the need to moderate extreme predictions. They overreact to recent unexpected, dramatic and salient news. Winner shares build a reputation over many years for high performance, usually based on corporate performance, such as abnormal rises in earnings per share. The winners are subject to greater publicity than other shares. Their high quality image and acclaim are not readily dissipated. These acknowledged ‘good companies’ are also regarded as ‘good shares’ to hold. They become overvalued. On the other hand loser companies, with a history of disappointing results, are stereotyped as forever on a downward course. The overreaction of investors leads to under-valuation after investors become excessively pessimistic. So why don’t other investors arbitrage shares back to their efficient level? We consider why the arbitrage process may not be strong enough. A preliminary thought is that perhaps investors are unaware of the return reversal phenomenon and the potential for out-performance and so do not act to correct the market “anomaly”. This ignorance explanation is not entirely convincing. It is not necessary for the majority of the investors to be aware of, and respond to, the phenomenon for its elimination over time by arbitrage. We conjecture there must be some obstacles preventing efficient arbitrage. Even when deviations from rationality are encountered potential arbitrageurs are aware that they can lose money by being ‘rational’. The ‘crowd’, in Keynes’ parlance, may push prices to even higher levels of irrationality in the short and medium term, causing arbitrage losses for all but the long-term focused investor (‘noise trader risk’). Thus, rational investors are frequently unable or unwilling to swim against the tide of investor sentiment. It is often assumed that professional investors, working in an institutional setting, would correct the cognitive errors of the “ignorant and emotional” private investors, thus restoring risk-return equilibrium. However, the context of institutional money management means that they are often constrained from correcting market errors. Fund manager decisions need to be justified to sponsors or senior management committees. It is far easier to explain an investment in a well- known company widely acknowledged to be a winning share than one in a share whose performance has declined for the past five years. Loser shares have a higher rate of insolvency than prior winner shares and so a fund manager would be perceived as countenancing the adoption of too much risk in advancing the case for investment in a series of loser shares. Fund managers are often led to believe that it is better to fail conventionally than to succeed unconventionally. This organizational constraint (an example of the principal-agent problem) could reduce the number of active arbitrageurs sufficiently to permit a continuation of the return reversal effect over many decades. In the face of these strong pressures, it is a brave fund manager who stands his/her ground and argues that prior period loser stocks, when taken as a whole, are not fundamentally more risky and that they earn a higher return over an extended period of time – especially when their performance is being judged over short periods. 2 Electronic copy available at: http://ssrn.com/abstract=998418 In conclusion, we find strong evidence suggesting that those shares with a bad history produce abnormally high returns, whereas thus with very high return histories prove disappointing. Those that have fallen are accordingly undervalued and those that have risen tend to rise too much relative to their potential. This leads, not only to opportunities for investors but to inappropriate market signals being sent to managers and, potentially, to a misallocation of society’s scarce resources. Short author biography Rose D. Baker Rose Baker is an applied statistician and Professor in Statistics with considerable research experience and interests in modelling real problems and developing data fitting techniques. Her main research and teaching interests are applied statistics, permutation and non-parametric tests, mathematical modelling of maintenance of machinery, epidemiology and clinical trials analysis. Her first degree from Cambridge was a joint first in Theoretical Physics and she holds a Ph.D. in High Energy Physics. She has had overseas experience in India. Glen Arnold Professor of Finance, Salford University and the director of the Centre for Economic and Finance Research He is author of the best-selling UK based corporate finance textbook, Corporate Financial Management, now in its third edition, as well as many other publications, e.g. The Financial Times Guide to Investing. Ph.D. Economics (Loughborough), B.Sc.Econ.(Hons.) (Cardiff) His main research interest is stock market behaviour, investment theory and the systematic mispricing of shares; drawing on the developments in behavioural finance, practitioner insight and traditional financial economics. ABSTRACT We investigate whether shares that have experienced extreme stock market performance over a five year period become mis-priced leading to opportunities for exploiting that mis-pricing. The results cast serious doubt on the efficiency of the UK stock market to price these shares. This means that managers of these companies may be receiving distorted signals from the market, which will have a knock-on effect on a range of managerial decisions, from the estimation of the equity cost of capital to the timing of share issues. More specifically we find that the 10 per cent of shares with the worst total share return record over five years go on to produce the highest 3 returns in the subsequent five years with an average out-performance of 8.9% per year – defying those who would write these companies off as the ‘dogs’ of the market. Those that are currently flying high on the market (the 10% best performers over five years) go on to under-perform by 47% over the next five years – suggesting that the ‘darlings’ of the market tend to become over- priced leading to the possibility of excessive resource allocation in their favour. This systematic long-term reversal of share returns is observed over the entire period 1960-2002. Loser shares (the worst performing shares in the prior five years) out-perform winner shares (the best performing shares over the prior five years) by about 14% per year. By separating the firm size (market capitalization) effect from the return reversal effect we show the presence of both. This evidence is in direct contradiction to Clare and Thomas (1995), who found no return reversal in UK shares following an adjustment for size. Return reversal is a feature of large as well as small companies and a seven-part consideration of risk does not substantiate the argument that loser out-performance is compensation for risk. We also consider some the arguments concerning the behavioural and institutional constraints that may permit the continuance of the return reversal effect over a long period with insufficient arbitrage pressure to eliminate it. JEL classification: G14, information and market efficiency Key words: Return reversal, Overreaction, Market inefficiency, Market efficiency 1. Introduction A strong return reversal effect has been shown in US studies (e.g. De Bondt and Thaler, 1985, 1987; Chopra, et. al., 1992). That is, portfolios of shares showing the worst total return performance over a three or five year period subsequently significantly out-perform portfolios of shares made up of prior period ‘winners’, and the market portfolio, over the long horizon. This 4 out-performance is remarkably high: extreme prior losers out-perform extreme prior winners by 5-10 per cent per year. Studies from around the world have drawn similar conclusions1. Contradicting the consensus developed in the US and elsewhere, in 1995 Clare and Thomas published the most comprehensive analysis to that date of UK shares, and concluded that the return reversal effect disappeared if the influence of market capitalization (size effect) is adjusted for. Thus, we have doubt thrown on belief in the generality of the return reversal effect. If it is not present in the second most important stock market in the world where does that leave the argument that share returns tend to go into reverse across markets and at various points in stock market histories? Are the reported results a product of mining data in particular markets in specific time periods? Clare and Thomas’ work, however, suffered from some serious methodological flaws. This paper provides a more robust analysis. It also extends the study period to 48 years, ending in 2002, rather than 1990, permitting us to observe the phenomenon at the height of the bull market. Our results indicate the presence of an economically significant return reversal effect in the UK share market. Moreover, it is unlikely that this effect can be attributed to the influence of size or risk. We find that extreme losers outperform extreme winners by about 14% per year in the five years following portfolio formation. The effect is present in both the full sample and in the largest 20% of firms, but is stronger for smaller companies. 2. UK Empirical Studies In comparison with the US there have been few studies of return reversal in UK shares. Those that are available differ in the period studied, length of study and methodology used. Power, Lonie and Lonie (1991) take the list of the ‘top 200’ UK companies in the June 1982 edition of Management Today. They calculate the share price changes plus dividends for each company over the 10 years to 1982. The 30 best performing companies are assigned to a winner 5 portfolio. The bottom 30 companies combine to form a loser portfolio. They report positive cumulative abnormal returns for prior-period losers of 86% in the five-year test period, while the prior-period winners showed negative abnormal returns of -47%. The study by MacDonald and Power (1991) cumulates market-adjusted excess return over three year formation periods. The winners are defined as the top 5% of firms and the losers as the bottom 5%. They establish eight sets of portfolio formations at three-year intervals between 1959 and 1985, and examine cumulative excess returns of the portfolios over three year holding periods. They find a reversal in the performance of the winner and loser portfolios in the test period. The average cumulative abnormal return from the arbitrage strategy of selling short the shares in the winner portfolios and buying shares in the loser portfolios is a statistically significant 29.15%. However, MacDonald and Power (1993) examine the predictability of 40 individual UK shares over the period 1982 to 1990 and conclude that rather than return reversal, ‘for the vast majority of the firms in the sample, share prices appear to follow a random-walk process….the findings of this paper, therefore, would appear to lend support to the proponents of a traditional view of stock market efficiency’. (p.38) Clare and Thomas (1995) declared, after examining a random sample of up to 1000 shares between 1955 and 1990, that losers outperform winners by a statistically significant 1.7% per annum, “However, after controlling for firm size we find that this return difference can be explained by the small firm effect. Our findings thus provide no evidence of overreaction in the UK stockmarket” (p.963). There are a number of problems with their approach, which, when combined, may help to explain why they were unable to detect return reversal. First, they tested only six series of portfolio returns (six portfolio formations) for periods of three years. We argue that many more portfolios need to be formed and tested to gain insight into the persistence of return reversal. This study examines 39 portfolio formations over 12, 24, 36, 48 and 60 month holding periods. Second, their method of adjusting for size lacked sensitivity – we undertake a 6 more comprehensive and sensitive analysis separating the return reversal effect from the size effect. Third, Clare and Thomas required firms to be “continuously quoted” to be included in their sample. As they freely admit, this introduced survivorship bias. Fourth, they used quintile portfolios, rather than deciles, leading to less sensitive tests. Fifth, they only examined portfolios formed by equally weighting the constituent shares, rather than considering the impact of value weighting within the portfolios. Sixth, Clare and Thomas used cumulative abnormal returns (CAR) rather than buy-and-hold returns to measure both formation and test period returns. Given the extensive literature describing the distortions caused by the use of CARs in long-run studies (e.g. Dissanaike, 1994, Barber and Lyon, 1997) it is important that the data be analysed using the buy-and-hold method Dissanaike (1997, 2002) conducts an analysis of return reversal, extending the test periods to 48 months. He shows return reversal for eight out of the ten portfolio formations examined even after adjustment for firm size. However he left open the question of whether the return reversal effect is evident in the entire cohort of London Stock Exchange (LSE) listed companies, because only those companies in the FT 500 share index (the largest 500 by market capitalization) are included in these studies. Following the imposition of the condition that only those companies without missing returns over the prior 48 months are included he was left with an average of 450 companies in each formation year. Dissanaike focused on returns on portfolios formed over the short period between 1979 and 1988. It is possible for sceptical readers to draw the conclusion that this evidence of a tendency for loser firms to outperform winner firms can easily be explained away as an event occurring in the 1980s, and so does not pose an overwhelming challenge to the efficient markets hypothesis. Campbell and Limmack (1997) in examining portfolios of loser companies and portfolios of winner companies found, over a five year period following portfolio formation, that a “reversal in the abnormal returns of winner and loser portfolios was experienced over each of the 7 years 2-5, thus lending support to the winner-loser effect” (p.537). In their study the companies are defined as winners and losers on the basis of abnormal returns calculated over only 12 months. This is unusual in the ‘long-term’ return reversal literature, given that prior period returns are generally calculated over 3 or 5 year periods. In sum, the studies reported in the literature have produced somewhat conflicting results. It seems clear that there is a return reversal effect, but its size is unclear, and the question of whether it is subsumed by the firm size effect remains. It is therefore important to ascertain the strength of the return reversal phenomenon when a larger data set and up-to-date analytical techniques are employed. This paper extends the study period to 48 years, ending in 2002, permitting us to observe the phenomenon at the height of the recent bull market as well as observing the strength of return reversal in each of the decades from the 1960s to the 1990s, thereby gaining a much broader picture. In all, this study examines 39 portfolio formations, far more than any previous paper. Furthermore, test period market-adjusted returns are reported for 12, 24, 36, 48 and 60 months. In addition to using the buy-and-hold method to analyse the data we extend the analysis further by considering market capitalization weighting within portfolios and examine returns against both an equally weighted and a value weighted market portfolio. In addition, these portfolios are formed from companies ranked by prior period returns and formed into deciles, leading to a more sensitive test than a quintile based analysis. The survivorship bias problem is dealt with by including all firms in the sample whether or not they are liquidated during the test period. We perform two tests to separate the size effect from the return reversal effect and conduct a seven-part consideration of risk, which includes making use of the Fama and French three-factor model (1993, 1996). This paper also contributes to the literature by observing the return reversal effect in both large and small firms. 8 3. Hypotheses We test the following hypotheses: o On average shares that exhibit extreme negative movements in returns over a period of five years will, during the subsequent five years, experience high returns relative to both the market index and to those shares that produced very high returns in the prior five-year period. o The more extreme the return in the prior five years, the greater will be the subsequent adjustment. o The reversal of return performance is explained by additional risk carried by the extreme loser shares. o The return reversal effect is subsumed by the size effect. o The return reversal effect is present in the investible universe relevant for most institutional investors, i.e. in the largest 20% of companies, as well as in the small company cohort. 4. Return reversal and overreaction The term overreaction has become so linked with the phenomenon of long-term share return reversal that this area of study has become known as the overreaction literature. However, it is important to maintain the distinction between the observation of share return reversals and the concept of overreaction. The two may be consistent with each other, and may provide mutually supportive evidence, but they have different roots and meanings. The study of return reversal is the examination of time series share (strictly, share portfolio) relationships, and the untangling of this effect from other potential explanations for patterns observable in the data, such as the size effect, or the impact of beta risk. 9 The overreaction discussion owes its origin to departures from Bayesian rationality documented in the field of experimental psychology. The Bayesian hypothesis for learning is for the consistent use of conditional probabilities for changing beliefs on the basis of new information. It would seem that such high levels of rationality are not an accurate characterization of how individuals behave when faced with new data (Kahneman et al., 1982, Kahneman and Tversky, 2000). Individuals, when revising their beliefs, tend to overweight recent information and underweight prior (or ‘base-rate frequency’) data. They often fall victim to the representative heuristic (Tversky and Kahneman, 1971, 1974, Kahneman and Tversky, 1972, Arrow 1982) under which an individual judges the likelihood of a future event by the similarity of the present (recent) evidence to it. In behaving this way decision takers fail to properly allow for the necessity of moderating extreme predictions. They overreact to recent unexpected, dramatic and salient news. Grether (1980), for example, found this in his laboratory experiments with students. Gilovich et al. (1985) found it in the sports market, where players and fans of basketball believe that a player is more likely to hit a shot if his previous shot was a hit, despite the evidence of a lack of correlation between successive shots. Clapp and Tirtiroglu (1994) found positive feedback in the housing market where recent rates of change become over weighted information used by decision-makers. There are many more examples in a wide variety of fields. Coming back to finance, it has been observed that the actions of professional security analysts, private investors and institutional fund managers display behaviour consistent with this overreaction view; for example, Dreman and Berry (1995), Dreman (1998), Bauman et al. (1999) and Dreman and Lufkin (2000) observing returns on value and growth shares, and Stein (1989) in the options market. When applied to share return patterns the overreaction interpretation is as follows. Winner shares build a reputation over many years for high performance, usually based on corporate 10