درباره منبع استراتژی های حرکت در بازار سهام ایتالیایی
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|12893||2004||31 صفحه PDF||سفارش دهید||محاسبه نشده|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : International Review of Financial Analysis, Volume 13, Issue 3, Autumn 2004, Pages 301–331
This paper investigates the source of momentum profits, while inferring the validity of the assumptions underlying rational and behavioural theories. Using a unique sample of securities listed in the Italian Stock Exchange from 1950 to 1995, we observe that buying better performing stocks in the previous 3–12 months and selling worse performing stocks over the same period yields significant profits in the short term (less than 1 year). Results also hold when conditioned upon different risk specifications. On the other hand, the continuation effect seems to significantly revert over a longer period. More importantly, in contrast with Conrad and Kaul [Rev. Financ. Stud. 11 (1998) 489], bootstrap and Monte Carlo simulations show that momentum profits are more likely to be generated by stock returns time series properties rather than by their cross-sectional differences. While the overall findings cannot reject the market efficiency hypothesis, we argue that behavioural theory may be a possible “story” to interpret the continuation effect.
Most practitioners firmly believe in the possibility to attain significant profits by identifying time series patterns in securities returns. A mount aggressive attack to this view has been raised from researchers always prompt to demonstrate how beating the market in a systematic way represent more often a “mirage” than a real prospect. At the same time, a puzzling phenomenon remains unsolved to date resisting different academic criticisms: the empirically ascertained profitability of contrarian and momentum strategies. Momentum portfolios, which entail long positions in past best performing stocks (winners) and short positions in past worst performing stocks (losers), have been proven to yield significant positive profits in the medium term (3–12 months). In contrast, a systematic reversal effect is found when a longer holding period (more than 3 years) is considered, and reversing the momentum strategy (i.e., buying past losers and selling past winners) results in production of profitable contrarian profits. These empirical findings, originally reported for the U.S. market in two seminal articles by De Bondt and Thaler (1985) and Jegadeesh and Titman (1993) and subsequently supported by a number of other works, turned out to be particularly provocative in their crucial undermining of the core concept of market efficiency. Therefore, in the following years, a critical understanding of these “anomalies” has become even more urgent, while two main directions have been taken by academics. On one side, continuation and reversal effects have been conceived as a failure of rational models to explain investors' behaviour. New paradigms, able to depict a broader picture, have been proposed based on the assumption of psychological biases in the way individuals respond to new information.1 According to the way investors are hypothesised to deviate from a rational behaviour, at least three broad approaches can be identified. First, Daniel, Hirschleifer, and Subrahmanyam (1998) argue that “overconfidence” and “biased self-attribution” cause individuals to underweight public relative to private information; following that an asymmetric response is expected according to whether public news confirm or disconfirm previous actions. Results would be attributed to “stock-selection skills” in the former case, generating the individual overreaction to news, and to “bad luck” in the latter case, generating the underreaction phenomena. In both cases, a delayed reaction is produced, generating the short-term continuation effect, which eventually reverses as soon as additional public information becomes available and stock prices approach their fair value. Second, while the “conservatism bias”2 leads investors to underreact to firm-specific news, therefore generating the momentum effect, the “representative heuristic bias”3 leads investors to extrapolate prior performance to forecast future expected returns (Barberis, Shleifer, & Vishny, 1998). The latter bias in conjunction with the former is believed to generate the reversal effect. Finally, from a different perspective, in Hong and Stein's (1999) model investors are categorized as informed and noninformed. Informed investors trade using only future cash flow information, while noninformed investors trade on the basis of recent past price information and are ultimately responsible for the observed momentum effect. The information arrival eventually narrows the information gap between the two groups, resulting in the long-term mean-reversion effect. Empirical evidence in favour of this model is reported in Hong, Lim, and Stein (2000) where higher momentum profits are observed for firms smaller in size and with low analyst coverage. Similar results are exhibited in Grinblatt and Moskowitz (1999) for firms characterised by the scarce presence of institutional owners. All these are typically the firms where the speed at which information diffuses into prices is lower. Within a market microstructure framework, an interesting experiment to distinguish among different hypotheses of investor behaviour is proposed by Hvidkjaer (2003). 4 Studying trade imbalances among small and large trades, 5 the author provides evidence in support of the heterogeneity explanation used by Hong and Stein (1999) to model investors' trading behaviour. Accordingly, while the imbalance among smaller investors (momentum traders) is explained by initial underreaction followed by delayed overreaction, simple “rationality” seems to drive the action among larger traders (news watchers). On the other side, several studies have seriously questioned either the real persistence of these controversial effects or their actual nature of “anomalies,” conveying, in both cases, evidences in support to the market efficiency hypothesis. First, empirical patterns in stock returns might be a simple consequence of data-snooping bias. In fact, the ability to process an enormous amount of financial data might reveal just by chance some significant patterns among the large number of possible trading strategies tested. A way to detect this bias due to misleading inference is to verify whether the in-sample results persist in the out-of-sample experiments. To specifically address this issue, Jegadeesh and Titman (2001) replicate their earlier study for the 1990–1998 period (out-of-sample), finding approximately the same results previously obtained for the 1965–1989 period (in-sample). 6 Moreover, the argument that these anomalies are the simple effect of data-snooping bias clashes against the evidence that reversal and continuation results are not confined exclusively to the U.S. market. Rouwenhorst, 1998 and Rouwenhorst, 1999 finds significant momentum effects in 12 European countries (including Italy) and in the emerging markets, while Chui, Titman, and Wei (2000) provide similar evidence for the Asian market. Consistent results come from Griffin, Ji, and Martin (2003), who provide a comprehensive analysis extended to 40 countries around the world. Evidences of the contrarian effect are in the United Kingdom (among others; Campbell & Limmack, 1997 and Nagel, 2001), France (Mai, 1995), and Spain (Alonso & Rubio, 1990). 7 Given the persistence in their results, researchers question whether these effects can be read merely within the market-efficient paradigm and whether they can be interpreted, for instance, as simple market microstructure effects or compensation for risk. First among others, Ball, Kothari, and Shanken (1994) argue that a large part of the contrarian strategies' profitability is driven by microstructure-induced biases.8 An example is provided by contrarian portfolios, which, being likely to be short in higher-priced winner stocks and long in lower-priced loser stocks, eventually give rise to an unhedged position in price-related bias. From a similar point of view, the presence of significant limits to arbitrage (Shleifer & Vishny, 1997) may prevent investors from trading sufficiently to remove systematic profits, ultimately inducing the observed persistent profitability of these strategies. Grundy and Martin (2001) have estimated that only for round trip transaction costs less than 1.5%, momentum strategies become significantly profitable. In addition to these explicit transaction costs, Korajczyk and Sadka (2004) introduce the “invisible” trading cost of the post trade adverse price movement described by Treynor (1994).9 This suggests that once we account for this price impact, only minimising trading costs momentum strategies (built up on liquidity-weighted portfolios) are expected to yield significant profits. Still in a rational world, Sadka (2003) justifies half of the momentum anomaly in terms of liquidity risk10 and views systematic profits as reward for the risk bore by momentum arbitrageurs who are likely to trade in the future against informed traders rather than noise traders. Based on the implications of learning problem models where traders who use past volume statistic information do “better” than traders who do not use such information (see, e.g., Blume, Easley, & O'Hara, 1994), a number of authors have investigated whether volume helps in predicting momentum/reversal effects.11 One rationale behind this is the idea that price changes accompanied by large trading volumes are symptoms of temporary variations in the aggregate demand of liquidity and are hence next expected to reverse (Campbell, Grossman, & Wang, 1993).12 On the other hand, volume can be related to the degree of information asymmetry as formalised in the informed/uninformed traders model developed by Wang (1994). Conrad, Hameed, and Niden (1994) provide results consistent with both these hypotheses, finding price reversals for high-transaction securities and price continuation for low-transactions securities. Alternatively to market microstructure-based explanations, momentum and contrarian strategies may be more simply interpreted as compensation of risk. In a world where capital asset pricing model (CAPM) holds higher beta, stocks are expected to offer higher returns. Accordingly, if winner (loser) portfolios contain higher beta (lower beta) stocks, momentum strategies can be easily justified also in a rational framework. Conversely, a systematically time-reversing risk, following, for example, changes in the debt–equity ratio as effect of extraordinarily better (or worse) performance, might support the reversal effect. However, both Jegadeesh and Titman (1993), for the momentum, and DeBondt and Thaler (1987), for the reversal, present evidences inconsistent with this naı̈ve explanation. In more recent years, Fama and French (1996), hereafter referred to as FF, show how the reversal effect tends to disappear in their more comprehensive three-factor model framework,13 which, however, is still unable to provide any persuasive justification for the continuation effect.14 Given the baffling nature of the short-term momentum effect, financial literature seems to have recently devoted much more attention to this phenomenon relatively to the long-term reversal returns. A new interesting perspective of investigation is introduced by Conrad and Kaul (1998), hereafter CK, who emphasise that momentum profits might not only be uniquely attributed to time series patterns in stock returns, but also to their cross-sectional differences. As higher (lower) expected returns stocks are more likely to experience higher (lower) than average returns in adjacent periods (Lo & MacKinlay, 1990), the cross-sectional variance of stocks' expected returns may be the prime responsible for momentum profits. If the CAPM or the three-factor model adequately captured differences in mean returns, the CK conjecture would inevitably collapse after the FF finding. On the other hand, it is also possible that the asset-pricing models usually employed in the financial literature neglect those risk factors able to account for the continuation effect.15 Simply assuming a time-invariant unconditional mean of stocks' expected returns, CK build a test to overcome this inconvenience. As a result, they ascertain that the cross-sectional variation across stocks plays a major role relative to the time series properties of stock returns in explaining momentums. This might represent a crucial result given that the market efficiency and the behavioural explanation fail only under the assumption that the continuation phenomenon derives from the time series predictability of stock returns (i.e., delay reaction). On the other hand, if this result is accurate, continuation effect can be seen again as pure compensation for risk. In any case, later empirical findings appear to diverge from CK results as documented by Grundy and Martin (2001) and Jegadeesh & Titman, 2001 and Jegadeesh & Titman, 2002. The latter study argues that the long-term reversal effect should represent a fact against the CK conjecture inasmuch as winner portfolios, comprising high expected return stocks, should outperform losers in any post formation period considered even over the longer term (see Fig. 2 p. 712 of Jegadeesh & Titman, 2001). Most importantly, in a paper published a year after, the same authors are able to attribute the discrepancies in empirical evidences to possible biases in the CK experiments. Given the provocative nature of the CK hypothesis, this paper addresses the issue of assessing the validity of the assumptions underlying rational and irrational theories in its ultimate attempt to explain the continuation effect in a market much less developed than the U.S. one. While no study has ever expressly examined contrarian and momentum effects in the Italian equity market, we believe that the key contribution of this work is testing the CK proposition for the first time in a non-U.S. context. Moreover, using a unique data set safe from possible sample biases—that is, small or survivor firm biases that generally affect non-Anglo-Saxon data—we intend to provide results robust to misspecification problems. This is made possible by the exceptionally long time dimension covered by our sample period (1950–1995), the unusually large proportion of yearly listed companies considered for a specific market, and the explicit inclusion of delisted companies. To detect the existence of contrarian and momentum effects in the Italian equity market, we examine the profitability of winner, loser, and zero-cost (winner minus loser) portfolio strategies during the 3 years following portfolio formation. Given that the asset-pricing literature fails to propose a “consensus” model to explain expected returns in the Italian context, we account for different risk factors within both a portfolio and a time series approach (CAPM and the Fama and French, 1996, three-factor model). Moreover, following CK, we adjust returns for the unconditional mean of expected returns to be able to discriminate between different sources of the continuation effect. Further investigation is carried out implementing both bootstrap and Monte Carlo experiments. The idea behind these tests is to remove any predictable time series patterns in stocks returns while maintaining their cross-sectional property; following that if momentum profits are found to persist in the simulation samples, their source will necessarily have a cross-sectional nature. In doing these analyses, we take into account the possible “small bias” described in Jegadeesh and Titman (2002), which, however, does not seem to significantly affect results in our sample. Unlike the CK evidence, the overall findings of this study appear to converge towards a main time series explanation of momentum profits, suggesting that behavioural models, being in fact satisfied with their underlying assumption, might tell us a possible “story” behind both contrarian and momentum profits. The remainder of this article is organised as follows. Section 2 provides a detailed description of our data and methodology. Section 3 presents the results on the unconditional profitability of momentum strategies over short and long horizons. The persistence of these results is investigated within a risk–reward framework in Section 4. Section 5 evaluates the contribution of the cross-sectional versus the time series source in generating momentum profitability. Section 6 contains the main conclusions of this work and its implications.
نتیجه گیری انگلیسی
This paper presents evidence on contrarian and momentum profits in the Italian equity market. Using a unique sample of monthly returns for all securities listed from 1950 to 1995, we aim to assess both the profitability of these strategies and their underlying generating sources. While the former problem is addressed using different methodologies and accounting for different risk specifications, the latter issue, which is somehow the essence of the analysis, is tackled by carrying out a number of simulation experiments. We first find that momentum strategies are profitable in the medium term (from 3 to 12 months). Past winners outperform past losers by about 0.90% per month. On the other hand, and similarly to the U.S. evidence, longer term analysis reveals that momentum profits dramatically shrink after the first year. This result seems to be more pronounced in the recent period due to slimmer profits in the short and medium term. Independently of the time horizon, we do not find any seasonal effect influencing the magnitude of the continuation effect. Although this issue is not specifically addressed in this study, we believe that unraveling the reasons behind it may be a source of fruitful future research. Second, we analyse the persistence of the above results, adjusting momentum profits by a number of possible risk factors. Both conventional single- and multifactor models are proved to fail in explaining momentum-driven returns. Once documented that the strength of these strategies is not built on differences in risk, we move on, in the attempt to explicitly identify the alternative possible sources of these profits. Bootstrap and Monte Carlo simulations show that momentum strategies carried out on stock returns samples that match all the characteristics of our data while eliminating any time series pattern are nil, implying that continuation effect should come from this latter source. Results are robust to the use of the bootstrap “in block” methodology and hold even when the possible small bias described in JT is taken into account. So, unlike in CK, we provide evidence in favour of the time series component as the main explanation of momentum profits; therefore, we are able to infer that the assumption underlying the behavioural models cannot be rejected. At the same time, these findings do not go in the direction of rejecting the market efficiency hypothesis in explaining momentum profits as often and incorrectly argued in many previous papers. In fact, a question recently and broadly debated in the literature is the possibility that these effects might be captured by more complex time-varying risk models, generally ignored by traditional studies, and that actively contribute to continuously renewing the interest on this issue. Evidences of the presence of momentum profits in equity markets around the world, so different in terms of size, microstructure, and social–cultural background (DeBondt, Schiereck, & Weber, 1999, point to the similarity between German and U.S. results as evidence in support of behavioural models), are in fact inconclusive when considering that these “anomalies” can be easily expressed in terms of common risk factors. On the other hand, our results support the idea that behavioural models coming from the stock returns' time series pattern should not be discarded a priori, as they might tell us a possible “story” to interpret both contrarian and momentum profits.