Sei sulla pagina 1di 40

A trending walk rather than a random walk?

Time-series momentum in Australia

Faisal Mahboob Supervised by1: Vinod Mishra

An Honours Research Essay submitted in partial fulfilment of the requirements for the Honours degree of Bachelor of Economics, 2013

Department of Economics Faculty of Business and Economics Monash University October, 2013

I would like to thank my supervisor Vinod Mishra for his guidance and support in my research

Table of Contents
Abstract ......................................................................................................................................................... 3 1. Introduction .............................................................................................................................................. 4 2. Literature Review ...................................................................................................................................... 7 2.1 Cross-sectional vs Time-series ............................................................................................................ 8 2.2 Momentum and Autocorrelation........................................................................................................ 9 2.3 A Trending Walk? .............................................................................................................................. 11 2.4 Momentum Explanations .................................................................................................................. 12 2.5 Australian Evidence ........................................................................................................................... 12 3. Data and Preliminaries ............................................................................................................................ 14 3.1 Returns .............................................................................................................................................. 14 3.2 Ex Ante Volatility Estimates .............................................................................................................. 15 4. Time-series Momentum.......................................................................................................................... 17 4.1 Time-series Predictability .................................................................................................................. 17 4.2 Time-series Momentum Trading Strategy ........................................................................................ 20 5. Performance Evaluation of Time-series Momentum.............................................................................. 21 5.1 Sharpe Ratios .................................................................................................................................... 21 5.2 Factor Models ................................................................................................................................... 22 6. Conclusion ............................................................................................................................................... 25 Reference List.............................................................................................................................................. 27 Appendix (figures and tables) ..................................................................................................................... 36

Abstract Following the methodology proposed by Moskowitz, Ooi & Pederson (2012), we report the presence of time-series momentum in ASX 100 stocks. We note that time-series momentum is present only if a large cross-section of the stocks are examined. Even with the presence of timeseries momentum, the random walk hypothesis cannot be dismissed if too few stocks are analysed. Using a sample of 69 stocks for the period January 2000 to January 2013, we find that the most profitable strategy is a look-back period of 3 months and a holding period of 12 months. This strategy was found to be profitable even after adjusting for exposures to the three Fama-French and additional Carhart risk factors. The strategy is not available to retail investors and we caution that future research should be conducted with as many assets as possible, before a market wide conclusion can be drawn.

1. Introduction

One of the puzzling questions in the asset pricing literature is the existence of momentum in asset returns. This study examines the existence of this anomaly in an Australian context through a novel perspective. The financial economics literature usually refers to cross-sectional momentum (CSM) when referring to the phenomenon of momentum. This study looks at timeseries momentum (TSM).

In finance, a phenomenon is considered an anomaly when an assets returns follow a regular pattern, which is reliable, widely known yet it cannot be explained by an asset pricing model. Said differently, the higher average returns observed cannot be attributed to higher risk undertaken and vice versa. Under the framework of the Efficient Market Hypothesis (EMH), investors and traders will eliminate the anomaly by adjusting their trading patterns so as to exploit the widely known anomaly. Moreover, EMH proponents argue that these anomalies cannot be exploited, even if they are present, due to real world aspects related to trading and investing such as transaction costs, credit constraints and institutional/regulatory framework. Due to the complexity involved with incorporating different forms of trading costs, asymmetric information and funding constraints, we will not be concerned with such issues and whether or not the strategy is profitable in the real world. Instead, we will be looking for systematic patterns that persist in the Australian stock market assuming that the investor can buy or short a stock at a moments notice and the investor has only these two choices.

Momentum is defined as the systematic pattern where high (low) asset returns are followed by additional periods of higher (lower) asset returns. Even after accounting for risk measured through exposures to different factors, assets seem to exhibit abnormal returns if a momentum strategy is followed. Empirical evidence suggests that momentum strategies exhibit abnormal performance in almost all markets and asset classes around the world (see Asness, Moskowitz & Pedersen, 2013). It is to be noted, however limited, that there is also some evidence of the reverse (see DeBondt & Thaler, 1985). The pattern of higher (lower) asset returns followed by lower (higher) asset returns is referred to as reversals or contrarian strategies in the literature.

Fama & French (2008) consider momentum, specifically CSM, to be the premier anomaly. They do not find much support for reversal or contrarian strategies or other anomalies that have been documented. Despite the enormous research into momentum, there has not been any consensus on the underlying reasons and causes of the phenomenon. Moskowitz, Ooi & Pederson (2012) find TSM across different asset classes and markets around the world. Rather than observing a securitys returns relative to each other as done in the case of CSM, only the securitys past returns are of concern. There have been many studies looking at the effects of CSM in Australia. However, no detailed work on TSM in Australian markets has been carried out to date. We hope to fill this gap through this study.

This study follows a slightly different route from that taken by Moskowitz, Ooi & Pederson (2012). They show TSM in futures and forwards of different asset classes whereas this research is focused only on individual stocks. There are three reasons for this departure. First, the Australian stock market is much more liquid with more readily available data than the Australian

futures market. Second, we want to check whether this anomaly can persist in one particular asset class. Third, undertaking research into TSM at a more micro level can provide insights into whether this pattern only exists in markets where the majority of market participants are institutional (futures and forward markets) or it exists also in the stock market where the majority of market participants are retail investors. This can shed light on whether the strategy should be considered by retail traders and investors.

Most importantly, finding TSM will pose a significant challenge to the Random Walk Hypothesis (RWH), which is a statistical description of price changes being unforecastable given past information. RWH is usually used as a model to test for market efficiency. However, a rejection of the RWH does not imply a rejection of the sophisticated notions of market efficiency. Under RWH, past prices should not be indicative of whether future prices will go up or down regardless of whether the price represents true underlying value of the asset. Undertaking this research can contribute to the debate on asset pricing and whether the RWH should be dismissed as a model or not.

The rest of this research essay is organised as follows. Section 2 presents an overview of the development of momentum as an anomaly and related relevant literature. Section 3 describes the data used in the study and the various data related issues. Section 4 describes the methodology used in the study and documents time-series momentum in the ASX 100. Section 5 tests for whether time-series momentum strategy leads to abnormal performance. Section 6 concludes.

2. Literature Review

Empirically, it has been found that momentum strategies are profitable in almost all markets and asset classes around the world. It has been found as early as in the Victorian Age by Chabot, Ghysels & Jagannathan (2009) using a new hand-collected data set on 1,808 stocks listed on the London Stock Exchange between 1866 and 1907.

Jegadeesh & Titman (1993) were the first to document momentum. They found momentum in individual securities in the stock market through the formation of portfolios of stocks on the NYSE and AMEX from 1965 to 1989. Fama & French (1998), Rouwenhorst (1998), Liew & Vassalou (2000), Griffin, Ji & Martin (2003) and Chui, Wei & Titman (2000) extended the analysis and detected momentum in 40 other international stock markets such as Argentina, Australia and India. Asness, Liew & Stevens (1997) and Bhojraj & Swaminathan (2006)

detected the phenomenon in 38 country equity indices such as Morocco and Venezuela.

Shleifer & Summers (1990), Kho (1996), and LeBaron (1999) identified momentum in currencies such as the Japanese yen and German Mark and Gorton, Hayashi & Rouwenhorst (2008) found momentum in 31 commodity futures between 1969 and 2006. Finally, Asness, Moskowitz & Pedersen (2013) observed the effect within and across asset classes such as bonds, stocks, currencies, equity indices and commodity markets around the world.

Instead of focusing on returns on individual stocks, Moskowitz & Grinblatt (1999) and Grundy & Martin (2001) added other variables such as industry factors to explain abnormal returns from

momentum. They concluded industry momentum and not momentum in the firm-specific component of returns helped explain momentum strategys abnormal performance. It is mostly industry momentum that contributes to the momentum effect in stock returns. Finally, Asness, Porter & Stevens (2000) found both inter- and intra-industry momentum in industries such as Tobacco Products and Candy & Soda when they analysed all the firms listed on the NYSE, AMEX, and Nasdaq stock exchanges from July 1963 (1973 for Nasdaq firms) through December 1998.

2.1 Cross-sectional vs Time-series A distinction has to be made between cross-sectional momentum (CSM) and time-series momentum (TSM). They may be related but are not the same thing. CSM refers to when a security with higher (lower) returns in previous periods relative to its peers tend to have higher (lower) future returns for future periods where the periods are not necessarily equal. In the current finance literature, momentum is usually inferred to be CSM rather than TSM. The literature covered so far has been on CSM.

On the other side of the spectrum is time-series momentum (TSM). TSM focuses purely on a securitys own return rather than its returns relative to other securities. If a security has exhibited high (low) returns in previous periods, it will continue to exhibit high (low) returns in the future but not necessarily for the same number of periods forecasted.

TSM has not been widely studied yet. Only one published paper, that of Moskowitz, Ooi & Pederson (2012), so far has explicitly mentioned TSM. There have been other papers in currency

markets such as Lustig et al. (2011) and Rafferty (2010) following quite close, but not exactly, the TSM methodology. They use an indicator or a trading rule as the signal to buy or sell a security rather than constructing a trading strategy around sign predictability (discussed in section 4.1). However, working papers by Burnside, Eichenbaum & Rebelo (2011) and Baltas & Kosowski (2012) do follow the TSM methodology exactly. Burnside, Eichenbaum & Rebelo (2011) use the methodology to try and explain why the strategy works whereas Baltas & Kosowski (2012) document TSM in futures markets and their relationships with funds employing similar strategies such as commodity trading advisors (CTAs) in the USA.

2.2 Momentum and Autocorrelation Research into return predictability is very broad. It involves conducting various econometric tests such as the Wald test. Economic variables are investigated and economic relationships over various time horizons are examined. Our research agenda is only concerned with past returns since we want to investigate the anomaly of momentum.

It is possible that momentum could just be an alternate way to exploit the pattern of small, statistically significant, high frequency predictability in past returns documented since Fama (1965). There is empirical evidence of positive and negative return autocorrelations (i.e. the correlation of last periods return with this periods) at different time horizons in different asset classes and various countries. Notable literature would be Fama & French (1988), Lo & Mackinlay (1988), Poterba & Summers (1988), Cutler, Poterba & Summers (1991).

Momentum incorporates more than just autocorrelation. Lo & Mackinley (1990) and Lewellen (2002) have argued that CSM might be caused by autocorrelation in returns, lead-lag relations among stocks (cross-serial correlation) or due to the cross-sectional variance in the unconditional means of each stocks returns. Therefore, not only does the stocks own past returns predict higher returns but so does the stocks past returns being negatively correlated with the lagged returns of other stocks and the stocks long run average of returns being higher relative to other stocks.

Moskowitz, Ooi & Pederson (2012) extend this line of reasoning and argue that TSM is composed only of an autocorrelation component and a mean-squared component. A stocks past return and the average squared mean returns of each asset predict higher returns. Notice that CSM and TSM are related through the common component of autocorrelation in returns. They go on to show empirically that the primary driver of TSM, for a given strategy of look-back and holding periods, is the autocorrelation in returns. We refer the reader to their paper for further details and supporting empirical work.

Previous discussion might lead to the belief that TSM is analogous in structure to autocorrelation. However, the superior predictability of TSM over autocorrelation is extracted from looking at a different number of periods of past returns to those forecasted. This flexibility is not possible with the autocorrelation studies which masked a lot of the predictability since the number of periods of past returns and forecasted were exactly the same. Autocorrelation is a necessary precondition for momentum, whether cross-sectional or time-series, to exist but it is not a conclusive proof of momentum.

10

2.3 A Trending Walk? The Random Walk Hypothesis (RWH) asserts that stock prices follow a random walk, that prices are unpredictable. Although used as a test for a particular form of market efficiency, a random walk is a precise mathematical formulation of a stochastic process formed by successive summation of independent, identically distributed random variables. Market efficiency, on the other hand, is more arbitrary and harder to test as there are various models and interpretations2. A layman definition of market efficiency asserts that investors/traders cannot earn returns above the average stock market return on a consistent basis. An implication of the rejection of RWH is to also discredit a form of market efficiency. Even so, this study will only scrutinize the RWH and not consider any implications for market efficiency.

The anomaly of TSM is a direct challenge to RWH since the anomaly suggests past returns can be used to forecast future returns while not being captured by standard risk factors such as FamaFrench factors SMB and HML and Carhart factor UMD. Furthermore, there are no constructions of portfolios of securities. Lo & MacKinlay (1990) and Conrad & Kaul (1998) argue that the cross-sectional variation in returns rather than the time-series component lead to momentum profits. They initially assumed that individual stocks followed a random walk, yet returns from a momentum strategy were positive. Thus, they concluded that time-series predictability is negligible. However in TSM, it is only the securitys own past returns that determine its abnormal returns. It is through examining correlations from different time horizons that gives rise to the time-series predictability. Moskowitz, Ooi & Pederson (2012) document significant TSM in various asset classes (except individual stocks) and across 58 liquid instruments with
2

Guerrien & Gun (2011) provide a neat discussion of the issues involved in their article "Efficient Market Hypothesis: What are we talking about?" real-world economics review 56 (2011): 19-30.

11

over 25 years of data. To date, this paper poses the most significant challenge to the RWH.

2.4 Momentum Explanations Despite the enormous research into momentum, there has not been any consensus on the underlying reasons and causes of the phenomenon. There are behavioural explanations (See Tversky and Kahneman (1974), Barberis, Shleifer & Vishny (1998), Daniel, Hirshleifer & Subrahmanyam (1998), Hong & Stein (1999), Frazzini (2006)). An interesting paper by Chui, Titman, & Wei (2010) suggested that momentum returns could be due to cross-country cultural differences which were measured using an individualism index developed by Hofstede (2001).

Chordia & Shivakumar (2002) attempt to link momentum returns to macroeconomic factors which are the dividend yield, default spread, bond yields and term structure spread. On the other hand, Cooper, Gutierrez & Hameed (2004) do not find any such relationship. Momentum returns have been shown to be linked to firm-specific factors such as size (Hong, Lim & Stein (2000)), credit rating(Avramov, Chordia, Jostova & Philipov (2007)), revenue growth volatility (Sagi & Seascholes (2007)) and likelihood of bankruptcy (Eisdorfer, (2008)).

There is also evidence of a link between momentum returns with trading volume (Lee & Swaminathan (2000)), transaction costs (Lesmond, Schill & Zhou (2004) and Korajczyk & Sadka (2004)) and information such as analyst coverage (Hong, Lim & Stein (2000)).

2.5 Australian Evidence In Australia, there has been considerable research into CSM. Drew, Veeraraghavan & Ye (2007)

12

find CSM by examining its relationship with trading volume while observing periods between 1988 to 2002 in stocks. Brailsford and OBrien (2008) find CSM in the top 500 stocks by market capitalisation and relate it to firm size observing periods from 1979 to 2005. Hurn and Pavlov (2003), Demir et al. (2004) and Bettman et al. (2009) all reported the profitability of CSM strategies in the ASX stocks. Galariotis (2010) provides comprehensive and robust tests of CSM on the ASX 100 stocks and all market securities for different time periods and market states. Indeed, the author claims that the momentum effect is greater in Australia than most of the other developed markets such as European countries. O'Brien, Brailsford and Gaunt (2010) suggest that the momentum effect may just be due to the small size effect which is a stock return characteristic where firms with low market capitalization have higher returns due to inherent higher risk. Nonetheless, they find CSM for portfolios of firms with large market capitalization. CSM, in general, has been found and has been suggested to be strong enough for earning above average risk-adjusted returns in Australia.

As mentioned before, examining TSM is closely related to the calculation of autocorrelations of stock returns. Autocorrelation studies in Australia have been documented since 1969. Praetz (1969), Officer (1975), Brown, Keim, Kleidon & Marsh (1983) and Groenewold & Kang (1993) have found autocorrelations with differing lags. More recently, Gaunt & Gray (2003) examined autocorrelations structures of returns on the top and bottom 200 Australian stocks by market capitalisation. They report statistically significant autocorrelations only for the bottom 200 Australian stocks. Any autocorrelation found was due to either illiquidity or the small firm effect as Australian companies are generall y smaller than other developed nations stock markets. There is a general consensus among the authors of these studies of not being able to trade

13

profitably based on returns exhibiting autocorrelation if transaction costs were taken into account.

3. Data and Preliminaries

We will discuss here the data source, the benchmarks used and the construction of the ex ante volatility estimates in our analysis.

3.1 Returns Monthly returns index on the one hundred stocks part of the ASX 100 index from the period July 1988 to January 2013 was retrieved from Datastream. The ASX 100 index was chosen due to its liquidity and companies with high market capitalisation. It covers firms with large and medium market capitalisation. Only stocks from the ASX 100 were analysed to avoid market microstructure effects such as stale prices and illiquidity from contaminating the results. Monthly returns were used rather than returns of a higher or lower frequency due to the issue of sign predictability. This will be covered in more detail in section 4.2.

The stocks were split into two samples, Sample 1 and Sample 2, and a subsample of Sample 1 called Sample 3. This allowed us to have samples where the time-dimensions of the returns matched up with each other for each stock. For the analysis there was a natural trade-off between the number of stocks and stocks with the longest history of data. Sample 1 has a higher number of past returns ranging from July 1988 to January 2013 with 18 stocks whereas Sample 2 has a higher number of stocks with 69 stocks with historical data ranging from January 2000 to

14

January 2013. Sample 3 consisted of all the stocks in Sample 1 except with returns from January 2000 to January 2013. This allowed us to carry out a comparison between Sample 2 and Sample 3. Fundamentally, the rationale between such sample selections was data availability as many stocks did not exist from 1988. Note that there is an overlap of the same stock between the samples.

Sample 1 started out with twenty stocks whereas Sample 2 had 71 stocks. Some stocks returns were zero for a significant period of time due to non-trading months where a period is one month. A data cleaning rule was put in place to deal with this issue. Any stock with more than ten periods of returns of 0% was deleted. Two stocks needed to be deleted from both the samples according to this rule. This was followed with linear interpolation to fill the returns of those periods for those stocks which had less than ten periods of returns of 0%.

3.2 Ex Ante Volatility Estimates To allow comparison, each stocks return will need to be scaled by its volatility since different stocks exhibit different levels of volatility. Every stocks return will be divided by its volatility

. The use of the volatility last period, called ex ante volatility, is used to avoid any lookahead bias.

Look-ahead bias occurs when historical data is used in testing a strategy that would not have been known or available during the period being analysed. Similarly, more sophisticated volatility models such as GARCH were avoided since they require parameter estimates over the

15

whole sample to generate the volatility estimates. These parameter estimates would have incorporated the look-ahead bias.

The ex ante volatility

was estimated using a close-to-close estimator as a proxy for realized

volatility (Shu, Jinghong, and Jin E. Zhang, 2003). These volatility estimates will be used in section 4.1 to scale the coefficients in our regressions. Realized volatility is a specific measure of historical volatility. Data is sampled at a very high-frequency to compute ex-post volatility at a lower frequency. In our analysis, daily returns have been used to calculate monthly volatility. It is calculated as follows:

(1)

where

is 21 since an average month is assumed to have 21 trading days,

is the daily return

on date .

3.3 Asset Pricing Benchmarks To check the performance of the strategies, we evaluate the returns of a given strategy relative to excess market index return, Fama-French factors SMB (Small Minus Big) and HML (High Minus Low) and Carhart factor UMD (Up Minus Down)3. The SMB factor is constructed by sorting the difference in returns of stocks with low market capitalisation (called Small) to those with high market capitalisation (called Big). Similarly, HML factor is constructed by sorting

We are extremely grateful to Stefano Marmi for making this data freely available at: http://homepage.sns.it/marmi/Data_Library.html#Australia

16

the difference in returns of stocks with high book value to market value ratio (called High) to those with low book value to market value ratio (called Low).

The specifications of the factors were made by Stefano Marmi and Flavia Poma as including all stocks for July of year to June of which have market equity data for the last fiscal year

end before March and June of time , and positive book equity data for the last fiscal year end before March . These factors capture particular characteristics of stock returns which have been considered as risk in empirical finance research.

4. Time-series Momentum

We first inspect time-series predictability of a stocks return across different time horizons at the monthly level. Then, we construct a trading strategy to exploit the time-series predictability to test for TSM similar to Moskowitz, Ooi & Pederson (2012). To determine whether there is timeseries predictability, we search for a pattern in the t-statistics of our regressions. The pattern should exhibit positive and significant t-statistics for the first couple of months followed by insignificant t-statistics. This pattern suggests a return continuation for the first couple of months with no trend after.

4.1 Time-series Predictability There are two possible ways to discern time-series predictability. One way is through a regression on lags and the other is to look at direction-of-change. Please note robustness checks have not been made in this study which is a major limitation. 17

Current literature suggests that there is a direct connection between asset return volatility dependence and asset return sign dependence. If that is the case, then the sign of next periods return can be predicted. There is considerable literature showing asset return sign forecasting can done successfully: Breen et al. (1989), Leitch & Tanner (1991), Wagner et al. (1992), Pesaran & Timmermann (1995), Kuan & Liu (1995), Larsen & Wozniak (1995), Womack (1996), Gencay (1998), Leung et al. (2000), Elliott & Ito (1999), White (2000), Pesaran & Timmermann (2004), and Cheung et al. (2003). These studies usually analysed stocks in the USA. An important point to note is that direction-of-change can only be forecasted at the monthly level (Christoffersen & Diebold, 2006). Hence, our analysis is based only on monthly returns data.

We regress the stocks excess return

at month

on its return lagged by

months. The

regression is a pooled time-series regression. Both the returns and their lagged values are scaled by their ex ante volatility as defined in section 3.2: (2) We stack all the scaled returns of all the dates and compute the t-statistics using lags from

Figure 2 and figure 3, shown in the appendix, plots t-statistics from the pooled time-series regressions of Sample 1 and Sample 2 by month lag h. The positive t-statistics indicate significant upward trend of returns and negative t-statistics indicate trend reversals. The first sample seems to continue exhibiting TSM at higher lags such as 43 and 47 with statistically significant correlation which is an unlikely event. The second sample is a bit more reasonable as

18

it displays the pattern we are trying to detect. Nevertheless, it still displays statistically significant negative correlation between the return at lagged month 60 with the return at month .

The second way to look at time-series predictability is by observing only the sign of the past excess return. It is this concept that underlies the trading strategy as it exploits the fact that, at least in sample, there is information in the sign of Eichenbaum and Rebelo, 2011): (3) The left-hand side of the regression is scaled again and the right-hand side doesnt require so since it only take the value of +1 or -1. The use of dummy variables was avoided as they would not be able to capture the ability to short the stock which is captured by -1. The results are similar to the first regression and we obtain almost the same pattern. Sample 1 does not seem to show any sort of pattern for time-series predictability whereas Sample 2 does. This is visible from the pattern of return continuation exhibited by Sample 2 only through positive, significant t-statistics for the first couple of months followed by insignificant and smaller t-statistics. about the sign of (Burnside,

There is considerable possibility of data snooping bias due to our small samples which occurs when a model is fit to random historical patterns that makes performance of a strategy look superior. The continuing significance of t-statistics and lack of negative signs for the t-statistics at higher lags is worrisome. However, the point of the regression is to only look for the pattern of return continuation mentioned before. We see that it has been the case where the first few lags are more significant than the rest. This is especially true and clear for Sample 2 as shown in the appendix.

19

4.2 Time-series Momentum Trading Strategy We explore different trading strategies based on TSM. There are three steps involved: a) we lag the returns for different number of months which is called the look-back period b) we consider whether the past excess return over the look-back period is positive or negative which will determine whether the investor goes long or short c) we then calculate the returns based on the number of months the stock is held for called the holding-period. We have avoided setting the position size to be inversely proportional to the asset classs ex ante volatility like Moskowitz, Ooi & Pederson (2012) since we are only analysing one asset class which is stocks.

Each trading strategy of a particular look-back period and holding period gives a single timeseries of monthly returns. This single series is derived following the methodology used by Jegadeesh and Titman (1993) such that the return at time represents the average return of every stock that is being held from the past. As an example, suppose an investor has decided to hold for two periods and bought a stock on January and another on February. There is an overlap of returns as the return on March is dependent on the stock bought on Januarys second holding period return and the stock bought on Februarys first holding period. Said differently, the stock bought on January is still active in March. We take this into account by taking the arithmetic mean of both the returns in March after having incorporated sign predictability.

To determine whether the investor/trader goes long or short we multiply the returns found earlier by either +1 (long) or -1 (short). This is how we integrate sign predictability into our trading strategy. For a given stock, the time- return is based on the sign of the past return from to . We then compute the time- return based on the sign of the past return from

20

to

, and we continue until we get to a point where the sign of the past return is the final to . For each trading strategy of a particular

return that is still being used from

(k,h), we get a monthly time-series of positive and negative returns by taking the average of all of the currently active stocks.

Finally, the returns are converted to excess returns, defined as returns minus the risk free rate. This will remove any predictability there may be in returns from the inclusion of the risk free rate and allow us to run factor models. We use the risk free rate used by Stefano Marmi and Flavia Poma of yields on 90 days Australian bank-accepted bills This gives us the TSM excess returns, .

5. Performance Evaluation of Time-series Momentum

To determine whether the strategy can lead to abnormal returns after adjusting for risk, we look at Sharpe ratios and alphas from the Fama-French three-factor model (Fama & French, 1993) and the Carhart four-factor model (Carhart, 1997). Sharpe ratio (Sharpe, 1966) is the ratio of expected returns less the risk-free rate over the standard deviation of the return. The factor models are regressions of the returns against patterns of stock market returns related to specific characteristics which are considered risk.

5.1 Sharpe Ratios Assuming that the returns are independently and identically distributed, we compute the annualized Sharpe ratio from monthly TSM returns using: 21

(4)

Figure 1 shows the Sharpe ratios for Sample 1, Sample 2 and Sample 3 for only one strategy. Only the strategy of a look-back and holding periods of 3 months and 12 months respectively is regarded as it is the most profitable in our analysis using factor models discussed in section 5.2.

The Sharpe ratios were calculated by first creating equally-weighted portfolios from a (3,12) strategy and taking arithmetic means of the returns, risk free rates and standard deviations over the whole sample. As expected, the ratios are above 1 suggesting that average differential return is greater than per unit of historic variability and the strategy is reasonably profitable. Sample 2 has higher Sharpe ratio than either of the samples suggesting a possibility of improvement in the returns of the strategy from expanding investments into further assets.

5.2 Factor Models We compute the alphas from the following time-series regression, the Fama-French three-factor model: (5) is the overall risk of the stock market. This controls for risk the investor could have taken by just buying an index fund rather than investing in a stock. controls for risk attributed to

investing in small firms. These are usually stocks which have low market capitalization. controls for risk taken on by the investor through investments in value companies. These are

22

usually stocks with high ratios of book value to market equity value. The exact details of the construction of the factors are available on Stefano Marmis website4.

Table 1 and table 2 in the appendix show the t-statistics of the estimated alphas for different combinations of the strategy for Sample 1 and Sample 2 respectively. The strategy backtested on Sample 1 exhibit extraordinary levels of abnormal performance as some of the t-statistics are significant and extremely high. We conjecture that errors are arising from this sample as a result of too few stocks. The overall average returns from TSM are not being captured. Note that the strategy seems to try and extend the look-back and holding periods as far as possible. We infer from this that the strategy is merely trying to capture the largest historic trends which is unreasonable as a trading strategy. On the other hand, Sample 2, shows results consistent with Moskowitz, Ooi & Pederson (2012) and also show above average risk-adjusted returns. There is abnormal performance for many grids of look-back and holding period combinations. Most notably is the strategy with a look-back period of 3 months and a holding period of 12 months which yields a positive, significant t-statistic of 6.48. We find that Sample 2 exhibits abnormal performance due to the presence of many assets. The strategy is most profitable when applied across a spectrum of many assets, especially across different asset classes as done by Moskowitz, Ooi & Pederson (2012).

We also compute the alphas from the following Carhart four-factor model: (6)

http://homepage.sns.it/marmi/Data_Library.html#Australia

23

This regression is similar to the Fama-French three-factor model with the addition of

This factor controls for risk taken on by the investor by investing in stocks that exhibit crosssectional momentum.

Table 3 and 4 of the appendix show the t-statistics of alphas from the Carhart four-factor model on Sample 1 and Sample 2 respectively. Sample 1s t-statistics have decreased but they are still not valid results. Sample 2 has more interesting results since the t-statistics of the alphas are more reasonable. The t-statistics have decreased but the significant ones have not become insignificant with the addition of the factor. The strategy with a look-back period of 3

months and a holding period of 12 months again yields the highest positive and significant tstatistic of 5.82. We can deduce that TSM excess returns, though related, are quite different from CSMs excess returns as the TSM. factor has not fully captured the abnormal performance of

For Sample 3, only the sign predictability regression and Fama-French three-factor model were done for the analysis. From the sign predictability regression results in figure 6 in the appendix we can see that most of the t-statistics are not statistically significant. Only lags 7, 9 and 19 are significant which does not allow for much inference. Similar to Sample 1, the results from the Fama-French three-factor model shown in table 3 also suggest that the strategy is trying to capture the largest trends as the t-statistics of alphas increase with holding period. However, there is a disparity. The t-statistics of alphas do not seem to increase with look-back periods. We conjecture that the strategy cannot trace back too far in history for the largest trends due to the reduced sample size.

24

The studys results show that a sample with a small number of stocks such as Sample 1 and Sample 3 do not exhibit TSM whereas a sample with many stocks such as Sample 2 does exhibit TSM. Our conjecture was made more convincing when the strategy was applied to one stock. There were no correlations of past returns to this periods returns. TSM was not present and it exhibited a random walk. Although the strategy is implementable in the Australian stock market where most retail investors invest, it is still not available to them. The strategy is only profitable when the investor is extremely diversified with investments in many stocks.

The inconsistency seems to arise from the role of the mean squared term. Moskowitz, Ooi & Pederson (2012) find the term to be insignificant to have any effect. We believe that the term possibly has a negative impact on the strategy if invested in too few assets and becomes insignificant when implemented across many assets. The latter case is where the autocorrelation in returns can be exploited fully to implement a TSM strategy.

6. Conclusion

Significant time-series momentum was found in a larger sample of stocks of the ASX 100 than in a smaller sample. The larger sample consists of 69 stocks from the ASX 100 with returns from the year 2000 to 2013 whereas the small sample consists of only 18 stocks from the ASX 100 with returns from the year 1988 to 2013.

Similar to Moskowitz, Ooi & Pederson (2012), the larger sample exhibits time-series momentum for the first 12 months. However, the look-back period is different. The most profitable 25

strategy is a look-back of 3 months and a holding period of 12 months as it results in an alpha that is the most significant and has the highest t-statistic. Time-series momentum is different from cross-sectional momentum as the cross-sectional momentum factor doesnt fully capture time-series momentums excess returns.

We note that time-series momentum can persist in one asset class such as stocks and not only in the futures and forward markets. Although the majority of market participants are retail investors, the strategy is not optimal for retail investors as it requires access to many liquid assets.

Time-series momentum is a direct challenge to the random walk hypothesis. Even in the presence of time-series momentum, the random walk hypothesis cannot be completely dismissed. When a smaller sample size was used, no time-series momentum was found which supported the random walk hypothesis. Caution needs to be taken to ensure a wide range of assets are considered when implementing the strategy or undertaking future research.

26

References
Asness, Clifford S., John M. Liew, and Ross L. Stevens. "Parallels between the cross-sectional predictability of stock and country returns." The Journal of Portfolio Management 23, no. 3 (1997): 79-87.

Asness, Clifford S., R. Burt Porter, and Ross Stevens. "Predicting stock returns using industry-relative firm characteristics." In available at SSRN: http://ssrn. com/abstract, vol. 213872. 2000.

Asness, Clifford S., Tobias J. Moskowitz, and Lasse Heje Pedersen. "Value and momentum everywhere." The Journal of Finance 68, no. 3 (2013): 929-985.

Asness, Clifford S., R. Burt Porter, and Ross Stevens. "Predicting stock returns using industry-relative firm characteristics." In available at SSRN: http://ssrn. com/abstract, vol. 213872. 2000.

Avramov, Doron, Tarun Chordia, Gergana Jostova, and Alexander Philipov. "Momentum and credit rating." The Journal of Finance 62, no. 5 (2007): 2503-2520.

Baltas, Akindynos-Nikolaos, and Robert Kosowski. "Momentum strategies in futures markets and trend-following funds." In Paris December 2012 Finance Meeting EUROFIDAI-AFFI Paper. 2012.

Barberis, Nicholas, Andrei Shleifer, and Robert Vishny. "A model of investor sentiment." Journal of financial economics 49, no. 3 (1998): 307-343.

Bettman, Jenni L., Thomas RB Maher, and Stephen J. Sault. "Momentum profits in the Australian equity market: A matched firm approach." Pacific-Basin Finance Journal 17, no. 5 (2009): 565-579.

Bhojraj, Sanjeev, and Bhaskaran Swaminathan. "Macromomentum: Returns Predictability in International Equity Indices." The Journal of Business 79, no. 1 (2006): 429-451.

27

Brailsford, Tim, and Michael A. O'Brien. "Disentangling size from momentum in Australian stock returns." Australian Journal of Management 32, no. 3 (2008): 463-484.

Breen, William, Lawrence R. Glosten, and Ravi Jagannathan. "Economic significance of predictable variations in stock index returns." The Journal of Finance 44, no. 5 (1989): 1177-1189.

Brown, Philip, Donald B. Keim, Allan W. Kleidon, and Terry A. Marsh. "Stock return seasonalities and the tax-loss selling hypothesis: Analysis of the arguments and Australian evidence." Journal of Financial Economics 12, no. 1 (1983): 105-127.

Burnside, Craig, Martin S. Eichenbaum, and Sergio Rebelo. Carry trade and momentum in currency markets. No. w16942. National Bureau of Economic Research, 2011.

Chabot, Benjamin, Eric Ghysels, and Ravi Jagannathan. Momentum Cycles and Limits to Arbitrage Evidence from Victorian England and Post-Depression US Stock Markets. No. w15591. National Bureau of Economic Research, 2009.

Cheung, Yin-Wong, Menzie Chinn, and Antonio Garcia Pascual. "What do we Know About Recent Exchange Rate Models? In-Sample Fit and Out-Sample Performance Evaluated." (2003): 03-13.

Christoffersen, Peter F., and Francis X. Diebold. "Financial asset returns, direction-of-change forecasting, and volatility dynamics." Management Science52, no. 8 (2006): 1273-1287.

Chordia, Tarun, and Lakshmanan Shivakumar. "Momentum, business cycle, and time varying expected returns." The Journal of Finance 57, no. 2 (2002): 985-1019.

Chui, Andy CW, Sheridan Titman, and KC John Wei. "Momentum, legal systems and ownership structure: An analysis of Asian stock markets." University of Texas at Austin working paper (2000).

28

Chui, Andy CW, Sheridan Titman, and KC John Wei. "Individualism and momentum around the world." The Journal of Finance 65, no. 1 (2010): 361-392.

Conrad, Jennifer, and Gautam Kaul. "An anatomy of trading strategies." Review of Financial Studies 11, no. 3 (1998): 489-519.

Cooper, Michael J., Roberto C. Gutierrez, and Allaudeen Hameed. "Market states and momentum." The Journal of Finance 59, no. 3 (2004): 1345-1365.

Cutler, David M., James M. Poterba, and Lawrence H. Summers. "Speculative dynamics." The Review of Economic Studies 58, no. 3 (1991): 529-546.

Daniel, Kent, David Hirshleifer, and Avanidhar Subrahmanyam. "Investor psychology and security market underand overreactions." the Journal of Finance 53, no. 6 (1998): 1839-1885.

Demir, Isabelle, Jay Muthuswamy, and Terry Walter. "Momentum returns in Australian equities: The influences of size, risk, liquidity and return computation." Pacific-Basin Finance Journal 12, no. 2 (2004): 143-158.

Drew, Michael E., Madhu Veeraraghavan, and Min Ye. "Do momentum strategies work? Australian evidence." Managerial Finance 33, no. 10 (2007): 772-787.

Eisdorfer, Assaf. "Empirical evidence of risk shifting in financially distressed firms." The Journal of Finance 63, no. 2 (2008): 609-637.

Elliott, Graham, and Takatoshi Ito. "Heterogeneous expectations and tests of efficiency in the yen/dollar forward exchange rate market." Journal of Monetary Economics 43, no. 2 (1999): 435-456.

29

Fama, Eugene F. "The behavior of stock-market prices." The journal of Business 38, no. 1 (1965): 34-105.

Fama, Eugene F., and Kenneth R. French. "Dividend yields and expected stock returns." Journal of Financial Economics 22, no. 1 (1988): 3-25.

Fama, Eugene F., and Kenneth R. French. "Common risk factors in the returns on stocks and bonds." Journal of financial economics 33, no. 1 (1993): 3-56.

Fama, Eugene F., and Kenneth R. French. "Value versus growth: The international evidence." The Journal of Finance 53, no. 6 (1998): 1975-1999.

Fama, Eugene F., and Kenneth R. French. "Dissecting anomalies." The Journal of Finance 63, no. 4 (2008): 16531678.

Frazzini, Andrea. "The disposition effect and underreaction to news." The Journal of Finance 61, no. 4 (2006): 2017-2046.

Galariotis, Emilios C. "What should we know about momentum investing? The case of the Australian Security Exchange." Pacific-Basin Finance Journal 18, no. 4 (2010): 369-389.

Gaunt, Clive, and Philip Gray. "Short-term autocorrelation in Australian equities." Australian Journal of Management 28, no. 1 (2003): 97-117.

Gencay, Ramazan. "The predictability of security returns with simple technical trading rules." Journal of Empirical Finance 5, no. 4 (1998): 347-359.

Gorton, Gary B., Fumio Hayashi, and K. Geert Rouwenhorst. "The fundamentals of commodity futures returns." Review of Finance 17, no. 1 (2013): 35-105.

30

Griffin, John M., Xiuqing Ji, and J. Spencer Martin. "Momentum investing and business cycle risk: Evidence from pole to pole." The Journal of Finance 58, no. 6 (2003): 2515-2547.

Grinblatt, Mark, and Tobias J. Moskowitz. "Predicting stock price movements from past returns: The role of consistency and tax-loss selling." Journal of Financial Economics 71, no. 3 (2004): 541-579.

Groenewold Fraser, Nicolaas. "Share prices and macroeconomic factors." Journal of Business Finance & Accounting 24, no. 910 (1997): 1367-1383.

Groenewold, Nicolaas, and Kuay Chin Kang. "The SemiStrong Efficiency of the Australian Share Market." Economic Record 69, no. 4 (1993): 405-410.

Grundy, Bruce D., and J. Spencer Martin. "Understanding the nature of the risks and the source of the rewards to momentum investing." Review of Financial Studies 14, no. 1 (2001): 29-78.

Guerrien, Bernard, and Ozgur Gun. "Efficient Market Hypothesis: What are we talking about?." real-world economics review 56 (2011): 19-30.

Hofstede, Geert H. Culture's consequences: Comparing values, behaviors, institutions and organizations across nations. Sage, 2001.

Hong, Harrison, and Jeremy C. Stein. "A unified theory of underreaction, momentum trading, and overreaction in asset markets." The Journal of Finance54, no. 6 (1999): 2143-2184.

Hong, Harrison, Terence Lim, and Jeremy C. Stein. "Bad news travels slowly: Size, analyst coverage, and the profitability of momentum strategies." The Journal of Finance 55, no. 1 (2000): 265-295.

31

Hurn, Stan, and Vlad Pavlov. "Momentum in Australian stock returns." Australian Journal of Management 28, no. 2 (2003): 141-155.

Jegadeesh, Narasimhan, and Sheridan Titman. "Returns to buying winners and selling losers: Implications for stock market efficiency." The Journal of Finance48, no. 1 (1993): 65-91.

Kho, Bong-Chan. "Time-varying risk premia, volatility, and technical trading rule profits: Evidence from foreign currency futures markets." Journal of Financial Economics 41, no. 2 (1996): 249-290.

Kuan, ChungMing, and Tung Liu. "Forecasting exchange rates using feedforward and recurrent neural networks." Journal of Applied Econometrics10, no. 4 (1995): 347-364.

LeBaron, Blake. "Technical trading rule profitability and foreign exchange intervention." Journal of International Economics 49, no. 1 (1999): 125-143.

Lee, Charles, and Bhaskaran Swaminathan. "Price momentum and trading volume." The Journal of Finance 55, no. 5 (2000): 2017-2069.

Leitch, Gordon, and J. Ernest Tanner. "Economic forecast evaluation: profits versus the conventional error measures." The American Economic Review(1991): 580-590.

Lesmond, David A., Michael J. Schill, and Chunsheng Zhou. "The illusory nature of momentum profits." Journal of Financial Economics 71, no. 2 (2004): 349-380.

Leung, Mark T., Hazem Daouk, and An-Sing Chen. "Forecasting stock indices: a comparison of classification and level estimation models." International Journal of Forecasting 16, no. 2 (2000): 173-190.

32

Lewellen, Jonathan. "Momentum and autocorrelation in stock returns." Review of Financial Studies 15, no. 2 (2002): 533-564.

Liew, Jimmy, and Maria Vassalou. "Can book-to-market, size and momentum be risk factors that predict economic growth?" Journal of Financial Economics 57, no. 2 (2000): 221-245.

Lo, Andrew W., and Archie Craig MacKinlay. "Stock market prices do not follow random walks: Evidence from a simple specification test." Review of financial studies 1, no. 1 (1988): 41-66.

Lo, Andrew W., and Archie Craig MacKinlay. "When are contrarian profits due to stock market overreaction?." Review of Financial studies 3, no. 2 (1990): 175-205.

Long, J. Bradford, Andrei Shleifer, Lawrence H. Summers, and Robert J. Waldmann. "Positive feedback investment strategies and destabilizing rational speculation." The Journal of Finance 45, no. 2 (1990): 379-395.

Lustig, Hanno, Nikolai Roussanov, and Adrien Verdelhan. "Common risk factors in currency markets." Review of Financial Studies 24, no. 11 (2011): 3731-3777.Moskowitz, Tobias J., and Mark Grinblatt. "Do industries explain momentum?"The Journal of Finance 54, no. 4 (1999): 1249-1290.

Moskowitz, Tobias J., Yao Hua Ooi, and Lasse Heje Pedersen. "Time series momentum." Journal of Financial Economics 104, no. 2 (2012): 228-250.

OBrien, Michael A., Tim Brailsford, and Clive Gaunt. "Interaction of size, booktomarket and momentum effects in Australia." Accounting & Finance 50, no. 1 (2010): 197-219.

Officer, Robert R. "Seasonality in Australian capital markets: Market efficiency and empirical issues." Journal of Financial Economics 2, no. 1 (1975): 29-51.

33

Pesaran, M. Hashem, and Allan Timmermann. "Predictability of stock returns: Robustness and economic significance." The Journal of Finance 50, no. 4 (1995): 1201-1228.

Pesaran, M. Hashem, Davide Pettenuzzo, and Allan Timmermann. "Forecasting time series subject to multiple structural breaks." The Review of Economic Studies 73, no. 4 (2006): 1057-1084.

Poterba, James M., and Lawrence H. Summers. "Mean reversion in stock prices: Evidence and implications." Journal of Financial Economics 22, no. 1 (1988): 27-59.

Praetz, Peter D. "Australian share prices and the random walk hypothesis." Australian Journal of Statistics 11, no. 3 (1969): 123-139.

Rafferty, Barry. "The returns to currency speculation and global currency realignment risk." manuscript, Duke University (2010).

Rouwenhorst, K. Geert. "International momentum strategies." The Journal of Finance 53, no. 1 (1998): 267-284.

Sagi, Jacob S., and Mark S. Seasholes. "Firm-specific attributes and the cross-section of momentum." Journal of Financial Economics 84, no. 2 (2007): 389-434

Sharpe, William F. "The sharpe ratio." Streetwise-The best of the Journal of Portfolio Management, Princeton University Press Princeton, New Jersey (1998): 169-185.

Shleifer, Andrei, and Lawrence H. Summers. "The noise trader approach to finance." The Journal of Economic Perspectives 4, no. 2 (1990): 19-33.

Shu, Jinghong, and Jin E. Zhang. "The relationship between implied and realized volatility of S&P 500 index." Wilmott magazine (2003): 83-91.

34

Tversky, Amos, and Daniel Kahneman. "Judgment under uncertainty: Heuristics and biases." science 185, no. 4157 (1974): 1124-1131.

Wagner, Jerry C., Steve Shellans, and Richard Paul. "Market timing works where it matters most in the real world." The Journal of Portfolio Management18, no. 4 (1992): 86-90.

Womack, Kent L. "Do brokerage analysts' recommendations have investment value?" The Journal of Finance 51, no. 1 (1996): 137-167.

35

Appendix
2 1.8 Annualized Sharpe Ratio 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 Sample - 1 Sample - 2 Figure 1 Sample - 3

Sharpe ratios of (k,h) = (3,12) strategy

Regression 1 Sample 1
5 4 3 2 1 t-statistic 0 -1 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 -2 -3 -4 -5 -6 Month lag

Figure 2

36

Regression 2 Sample 1
4 3 2 1 t-statistic 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 -1 -2 -3 -4 Month lag

Figure 3

Regression 1 Sample 2
8 6 4 t-statistic 2 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 -2 -4 -6

Month lag

Figure 4

37

Regression 2 Sample 2
5 4 3 2 1 t-statistic 0 -1 -2 -3 -4 -5 -6 Month lag 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59

Figure 5

Regression 2 Sample 3

4 3 2 t-statistic 1 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 -1 -2 -3

Month lag

Figure 6

38

Fama-French three-factor model

Table 1

Table 2

Table 3

39

Carhart four-factor model

Table 4

Table 5

40

Potrebbero piacerti anche