This article was downloaded by: [Chi-Hsiou Hung] On: 30 April 2013, At: 00:22 Publisher: Routledge Informa

Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Quantitative Finance
Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/rquf20

Active momentum trading versus passive naive diversification

ANURAG N. BANERJEE & CHI-HSIOU D. HUNG
a b a b

Durham University, Durham Business School , Durham , UK

School of Business, University of Dundee, Division of Accounting & Finance , Dundee , UK

To cite this article: ANURAG N. BANERJEE & CHI-HSIOU D. HUNG (2013): Active momentum trading versus passive naive diversification, Quantitative Finance, 13:5, 655-663 To link to this article: http://dx.doi.org/10.1080/14697688.2013.766760

PLEASE SCROLL DOWN FOR ARTICLE Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material.

Quantitative Finance, 2013 Vol. 13, No. 5, 655663, http://dx.doi.org/10.1080/14697688.2013.766760

Downloaded by [Chi-Hsiou Hung] at 00:22 30 April 2013

2013 iStockphoto LP

Active momentum trading versus passive 1/ N naive diversication

ANURAG N. BANERJEE and CHI-HSIOU D. HUNG
Durham Business School, Durham University, Durham, UK Division of Accounting & Finance, School of Business, University of Dundee, Dundee, UK
(Received 18 November 2011; in nal form 11 January 2013)

1. Introduction Asset managers in the stock market either opt for active or passive portfolio strategies. One important strategy involving active trading is the return-based momentum strategy that buys recent winner stocks and sells short loser stocks (Jegadeesh and Titman 1993, 2001). The extensive evidence of the profitability of momentum trading has undoubtedly underpinned the interest of investment practitioners. Many studies demonstrate the statistically signicant model alpha of the momentum prot using either the CAPM or multi-factor models (see, for example, Fama and French (1993), Avramov and Chordia (2006), Sagi and Seasholes (2007) and Liu and Zhang (2008)). Korajczyk and Sadka (2004) further show that after controlling for transaction costs the momentum strategies remain profitable.

In contrast, the formation of an equally weighted stock portfolio is widely adopted among passive portfolio managers. Benartzi and Thaler (2001), for example, document this simple 1/ N rule for allocating wealth across assets. DeMiguel et al. (2009) denote the 1/ N allocation rule as a naive diversication strategy and show that the prots of the 1/ N rule are comparable to those of optimal portfolio strategies. Huberman and Jiang (2006) found that 401(k) participants tended to allocate their contributions evenly across their selected funds. Benartzi and Thaler (2007) examine the asset allocation behavior for retirement savings funds and provide evidence that investors are relatively passive . . . and they adopt naive diversication strategies. Banerjee and Hung (2011) provide a new methodology to evaluate the performance of an active momentum asset manager. Their method rewards and penalizes the momentum strategist for using past return information in the formation of the strategies. To this end, they construct naive investors who

*Corresponding author. Email: c.d.hung@dundee.ac.uk

2013 Taylor & Francis

656

Feature On a risk-adjusted basis, the winners outperform the NDS by 14 basis points per month, while the losers under-perform the NDS by 18 basis points per month. The results, taken together, suggest that asset managers might be able to earn higher riskadjusted excess returns than the NDS by either taking a long position in the winner portfolio, or by taking a short position in the loser portfolio. The MS, however, are not better off pursuing the combined longshort trading because a simple strategy that equally weights the feasible set of stocks can achieve the same level of average prot, either raw or risk-adjusted, as the momentum strategies, but with lower idiosyncratic variance. Our ndings are robust with respect to sampling and periodspecic effects as we simulate the distributions of the average returns of the winner and the loser portfolios as well as the momentum prots by re-sampling with replacements. In each of the 100 randomly chosen 10-year periods we use all stocks feasible for trading in the sample during that period, and then construct the empirical distribution of the average prots. At the 5% level, all of the 100 samples accept the null hypothesis that the momentum prots are equivalent to the prots of the NDS. We further consider the momentum strategies based on 12-month formation and 12-month holding periods as in Jegadeesh and Titman (1993). Overall, our results still hold, showing that adopting naive 1/ N diversication is a good investment policy relative to the 12-month momentum strategies. Our ndings are important for practitioners and investors. The results illustrate that the MS requires the analysis of past return information and then takes both sides of the extreme return positions, thereby taking up the risk from the tails of the return distribution. By doing so, they signicantly reduce the exposure to the market risk but face large dispersions in the prots. The NDS, by contrast, simply weights stocks equally and performs as well as the MS, and even has lower idiosyncratic variance. From a practical point of view, pursuing the active momentum strategies would not be benecial once further taking into account the costs of the intensive longshort trading, the risks and regulation restrictions on short sales of stocks. The rest of the paper is structured as follows. Section 2 denes the naive diversication strategy and the momentum strategies. In section 3 we examine the protability of these strategies over the whole sample period, some interesting subsample periods, and 100 randomly selected 10-year periods. Section 4 analyses the theoretical and empirical relations between the momentum and the 1/ N naive diversication strategies, risk factor exposures and idiosyncratic variance. Section 5 concludes. 2. The naive 1/ N diversication strategy and the momentum strategies We dene N stocks that are feasible for trading at a given month as in Jegadeesh and Titman (1993). The set of N stocks we use for forming the NDS strategies is in fact the same
Jones and Lamont (2002), for example, describe the difculties of short sales, including the risks, costs, legal and institutional restrictions, and the need of sufcient stock supply from investors who are willing to lend.

use no information and weight risky assets randomly. Overall, they nd that, over the long run, the rewards and penalties cancel out and that the momentum strategist is no better than a simple naive randomizer. One interesting outcome of the construction of this universe of naive investors is that the median naive investor follows the naive diversication strategy or the 1/ N allocation rule. Essentially, such a 1/ N allocation rule is a passive strategy that tracks the equally weighted stock portfolio without using past information. In this paper, we aim to understand the merits and demerits of the active momentum trading and this passive 1/ N strategy. For this purpose we examine their raw and risk-adjusted profits and idiosyncratic variances (in other words, the variance of returns due to stock selection Sharpe (1992)). We further investigate the relation between these strategies in terms of their correlation and the dimensions of risk exposures. We consider a naive diversication strategy (NDS) that is long in the equally weighted (1/ N ) portfolio of stocks and short in the risk-free asset at the beginning of a period. This creates an initial zero-net-worth position as in the momentum strategies. The returns on the 1/ N portfolio in excess of the risk-free rate, using the one-month Treasury bill rate as a proxy, are thus the prots of the zero-net-worth strategies of the NDS. At the end of the period the NDS re-balances the stock portfolio in order to remain equally weighted. The NDS then holds the portfolio for the next period and continues to be nanced at the risk-free rate. This strategy can be implemented practically and tracked over time. The momentum strategies (MS) rst form portfolios of the winner (top 10% past returns) and the loser (bottom 10% past returns) stocks and then hold a long position on the winner portfolio and a short position on the loser portfolio. The momentum strategies further execute intensive trading in order to establish overlapping strategy positions. We use the monthly equity data of the NYSE, AMEX and NASDAQ over the sample period between 1926 and 2005, various sub-sample periods and 100 randomly selected 10year periods. The set of sample stocks that we use to construct the equally weighted portfolio is the same as that for constructing the momentum strategies. We nd that the average prot of the NDS is around 1% per month, close to that of the MS, in our sample period and in the sub-sample periods of Jegadeesh and Titman (1993, 2001). The MS generates the same level of prots as the NDS during the period of bull markets such as the dot-com bubble and makes prots from selling short the losers in the great depression period. Further, we divide the sample into two groups of large and small size stocks for the various sub-sample periods and nd that, in each size group, the difference between the average prots of the NDS and the MS is close to zero and statistically insignicant. Moreover, the variance of the idiosyncratic component of the momentum prots is 20 times higher than the NDS prots. Importantly, the risk-adjusted alpha of the difference in prots between the momentum and the NDS is virtually zero and statistically insignicant. The NDS and the momentum strategies are orthogonal, i.e. their prots have zero correlation. The momentum prots show a positive market beta and a signicant loading of nearly 1 on the momentum factor. In contrast, the NDS has a signicantly positive market beta, but does not have signicant exposure on the momentum factor.

Downloaded by [Chi-Hsiou Hung] at 00:22 30 April 2013

Feature as the set of stocks for forming the momentum strategies. Specically, we use all the monthly equity data of the New York Stock Exchange (NYSE), the American Stock Exchange (AMEX) and NASDAQ les from the Center for Research in Security Price (CRSP) for the period between January 1926 and December 2005. We includes all stocks classied as ordinary common shares (CRSP share codes 10 and 11) with exchange codes of 1, 2 and 3 as of the end of the previous year. We thus exclude all non-common equities such as American deposit receipts, companies incorporated outside of the U.S., shares of benecial interests, certicates, real estate investment trusts, close-end funds, etc. A stock must meet the following criteria in order to be included for analysis: rst, a stock must have observations on returns for the current month and over the past 6 or 12 months (depending on the respective analysis period), stock price, and shares outstanding. Second, a stock must have a price equal to or higher than $5 at portfolio formation. Our tests focus on the representative overlapping momentum strategies that form portfolios by sorting stocks on their past 6-month compounded returns and hold portfolios for 6 months. At the end of each month, the stocks within the top 10% of past returns comprise the winner portfolio FW , and stocks within the bottom 10% of past returns comprise the loser portfolio FL . Portfolios are equally weighted at formation and are held for six months without re-balancing during the holding period. The momentum strategy is then dened as FP = (FW FL ). Denote r = [r1 , r2 , . . . , r N ] as the vector of returns on the feasible stocks in excess of the risk-free rate rf . We compute monthly excess portfolio returns (rW = r FW and rL = r FL ) and the prots, rP = r FP , for the momentum strategies using single-period returns as in Liu and Strong (2008). The momentum strategies have six overlapping strategy positions, each starting one month apart. The monthly portfolio returns from the overlapping strategies are averages of the six strategies as in Jegadeesh and Titman (1993). We construct the 1/ N naive diversication strategy over these N feasible stocks. The excess portfolio returns from the strategy, 1 1 1 1 1= , ,..., , N N N N are then given by r = (1/ N ) iN =1 ri . , where r is effectively the prot of the 1/ N naive diversication strategy. Banerjee and Hung (2011) demonstrate that this strategy does not involve information costs and is the cross-sectional mean and the projection median of their uniformly distributed random strategies.

657

Downloaded by [Chi-Hsiou Hung] at 00:22 30 April 2013

periods from 1929 through 1933 and in the late 1990s, however, there exist large positive and negative deviations. These periods correspond, respectively, to the great depression and the dot-com bubble. We thus carefully examine these subsample periods. Panel A of table 1 presents the average excess returns of the winner and the loser portfolios, the average momentum prots as well as the average differences in excess returns between these portfolios and the NDS. The winner and the loser portfolios have average excess returns of 1.78% and 0.70% per month, respectively. The momentum strategies generate a statistically signicant average prot of 1.08% per month. Strikingly, the average prot of the NDS is as high as 1.06% per month and is highly signicant. The winner portfolio outperforms the NDS by a statistically signicant 0.72% per month, while the loser portfolio under-performs the NDS by a statistically signicant 0.36% per month. Importantly, the average difference between the prots of the momentum and the zero-net-worth NDS strategies is virtually zero and statistically insignicant. We next look into the question of whether size matters in our analysis. For this purpose, we divide the sample, each month, into two groups of large and small size stocks using the median of the market value of all sample stocks at the end of the previous month as the cut-off point. We then form the momentum and the zero-net-worth NDS strategies within each of the groups. Panel A of table 2 shows that, in each size group, the difference between the average prots of the momentum and the NDS strategies is close to zero and statistically insignicant. Within the sample of small size stocks, the average monthly momentum prot is ve basis points lower (but statistically insignicant) than that of the NDS; within the sample of large size stocks, the average momentum prot is 11 basis points higher (but statistically insignicant) than that of the NDS. 3.1. Over sub-sample periods Next, we examine the protability of the momentum and the 1/ N naive diversication strategies over the sample periods of Jegadeesh and Titman (1993, 2001) from 1965 to 1989 and from 1990 to 1998, respectively. Panel B of table 1 shows that, over the former sub-period, the momentum strategies generate a statistically signicant average prot of around 1.1% per month. The NDS has a statistically signicant average monthly prot of 0.9%. The average difference in prots between the momentum and the NDS strategies is 0.18%, but statistically insignicant. Compared with the NDS, the winner portfolio signicantly over-performs by 0.72%, while the loser portfolio signicantly under-performs by 0.36% per month. The results in panel C of table 1 for the latter subperiod are similar. Again, there is virtually no difference in prots between the momentum and the NDS strategies. The middle panels of gure 1 plot the differences in prots between the momentum and the NDS strategies over the sample periods of Jegadeesh and Titman (1993, 2001). The prot differences mainly center around zero and only exhibit sporadic deviations from zero. As discussed above, we observe large prot differences between the momentum and the NDS strategies in the early and later periods of our sample. We therefore further examine the

3. Protability of the strategies The top panel of gure 1 plots the time-series of prot differences between the momentum and the NDS strategies. The prot differences mainly center around zero and exhibit sporadic deviations from zero most of the time. During the
In the case where a stock is delisted during the holding period, the liquidating proceeds are reinvested in the remaining stocks in the portfolio.

658

Feature

Downloaded by [Chi-Hsiou Hung] at 00:22 30 April 2013

Figure 1. Distributions of the differences in prots between the momentum strategy and the NDS for the various sample periods. Table 1. Protability of the six-month momentum and the 1/N naive diversication strategies. This table presents mean excess returns of the momentum strategies and the 1/N naive diversication strategies (NDS) for the whole sample period (panel A), the sub-sample periods of Jagadeesh and Titman (1993, 2001) (panels B and C), the periods during the great depression (panel D) and the dot-com bubble (panel E). We use the one-month T-bill rate as a proxy for the risk-free rate. Winner and Loser are, respectively, the returns on the winner and the loser portfolios in excess of the one-month T-bill rate. WinnerNDS and LoserNDS are, respectively, the return differences between Winner, Loser and the NDS. Momentum prot-NDS is the difference between the prots of the momentum and the NDS. t -Statistics are in parentheses. , and denote statistical signicance at the 10%, 5% and 1% level, respectively. Winner Loser Momentum prots NDS 1.06 (4.96 ) 0.90 (2.64 ) 1.34 (3.02 ) 2.02 (0.98) 1.59 (2.49 ) WinnerNDS 0.72 (6.10 ) 0.72 (4.40 ) 0.88 (3.21 ) 0.33 (0.42) 2.37 (2.81 ) LoserNDS 0.36 (2.99 ) 0.36 (2.87 ) 0.54 (2.17 ) 0.39 (0.27) 0.48 (-1.27) Momentum prot- NDS 0.02 (0.06) 0.18 (0.43) 0.09 (0.18) 2.74 (0.72) 1.27 (1.31)

Panel A: The whole sample: 01-1927 to 07-2005 Mean return (%) 1.78 0.70 1.08 (2.4 ) (5.86 ) (6.97 ) Panel B: JagadeeshTitman (1993): 01-1965 to 12-1989 Mean return (%) 1.62 0.54 1.13 (3.90 ) (1.33) (4.68 ) Panel C: JagadeeshTitman (2001): 01-1990 to 12-1998 Mean return (%) 2.22 0.79 1.60 (1.32) (4.36 ) (3.41 ) Panel D: Great depression: 08-1929 to 03-1933 Mean return (%) 1.69 2.41 0.72 (1.02) (0.76) (0.34) Panel E: Dot-com bubble: 01-1995 to 02-2000 Mean return (%) 3.96 1.11 2.86 (3.13 ) (1.21) (3.37 )

sub-sample periods between August 1929 and March 1933 during the great depression as designated by the NBER and between January 1995 and March 2000 for the dot-com bubble. Interestingly, we nd that during the great depression period, as shown in panel D of table 1, the winner portfolio loses 1.69% per month, but the loser portfolio incurs an even larger average monthly loss of 2.41%. The average momentum prot of 0.72%, although statistically insignicant, mainly comes from the short position on the loser portfolio. The NDS also suffers an average monthly loss of 2.02%. The average of the excess returns on the winner portfolio is only 33 basis points higher than the average prot of the NDS, but statistically insignicant. The average difference in prots between the

momentum and the NDS strategies, although as high as 2.74%, is statistically insignicant because of the greater volatility in this period. The right-hand panel of the lower part of gure 1 shows that the prot differences over this period display a few very large negative outliers. Panel E of table 1 shows that, during the dot-com bubble period, the winner portfolio has an average monthly excess return of 3.96%, which is 2.37% higher than that of the NDS. The average excess return on the loser portfolio is 48 basis points lower than the NDS, albeit statistically insignicant. The average difference between the prots of the momentum and the NDS strategies is 1.27%, but statistically insignicant. The left-hand panel of the lower part of gure 1 shows that the

Feature

659

Table 2. Protability of six-month momentum and the 1/N naive diversication strategies for large and small size stocks. This table presents mean excess returns of the momentum strategies and the 1/N naive diversication strategies for the sample periods as described in table 1. We use the one-month T-bill rate as a proxy for the risk-free rate. NDS is the prot for the NDS . Momentum prot NDS is the prot difference between the momentum strategies and the NDS. t -Statistics are in parentheses. , and denote statistical signicance at the 10%, 5% and 1% level, respectively. Large size (greater than the NDS of the market value) Momentum prots NDS Momentum protsNDS 0.11 (0.37) 0.21 (0.58) 0.11 (0.20) 2.84 (0.70) 1.75 (1.58) Small size (smaller than the NDS of the market value) Momentum prots 1.27 (6.98 ) 1.47 (6.41 ) 2.00 (5.58 ) 0.44 (0.23) 2.72 (3.47 ) NDS 1.32 (5.52 ) 1.24 (3.32 ) 1.37 (2.63 ) 2.43 (-1.10) 1.72 (2.43 ) Momentum protsNDS -0.05 (0.14) 0.23 (0.52) 0.63 (1.08) 2.88 (0.75) 1.00 (1.03)

Downloaded by [Chi-Hsiou Hung] at 00:22 30 April 2013

Panel A: The whole sample: 011927 to 072005 Mean return (%) 0.94 0.83 (4.23 ) (4.76 ) Panel B: JagadeeshTitman (1993): 01-1965 to 12-1989 Mean return (%) 0.80 0.58 (1.95 ) (3.58 ) Panel C: JagadeeshTitman (2001): 011990 to 12-1998 Mean return (%) 1.11 1.00 (2.28 ) (2.95 ) Panel D: Great depression: 08-1929 to 03-1933 Mean return (%) 0.75 2.08 (0.31) (1.05) Panel E: Dot-com bubble: 01-1995 to 02-2000 Mean return (%) 3.22 1.48 (2.41 ) (3.02 )

prot differences over this period mostly center around zero. The results in panels B, C, D and E of table 2 show that, in these sub-sample periods, the differences between the average profits of the momentum and the NDS strategies are statistically insignicant for both the large and small size groups. Overall, the momentum strategies do not outperform the NDS in all periods considered. The evidence suggests that, on average, an investor would be better off if he holds, each month, a portfolio of all feasible stocks with equal weights that is nanced by borrowing at the risk-free rate. This is because this passive investment strategy generates an equivalent average prot and does not require sophisticated trading as do the momentum strategies.

the NDS strategies using the following t -statistic: rPs r s , s = 1, . . . , 100, ts = T std s (rP r ) where rPs , r s and std s (rP r ) are the average momentum prot, the average prot of the NDS strategies and the standard deviation of the difference in prots between the momentum and NDS strategies, respectively, for the randomly chosen sample period s . We reject the null at the % signicance level when ts > t , and count the number of non-rejections to indicate the number of periods when momentum strategies signicantly outperform NDS strategies. We nd that, at the 5% level, none of the 100 samples rejects the null and, at the 10% level, only four of the 100 samples reject the null. Figure 2 presents the distributions of these sample averages. The modes of average returns of the winner portfolio, the loser portfolios and the momentum prots are 1.83%, 0.69% and 1.19%, respectively. The average prots of the NDS have a mode of 1.18%. We also observe that, 95% of the time, the NDS gives statistically positive prots, whereas the corresponding gures for the winner portfolio, the loser portfolio and the momentum prots are 100%, 87% and 95%, respectively. The differences in prots between the momentum and the NDS strategies have a mode of 0.06 with a 53% chance of being positive. Hence, the 1/ N diversier is almost as good as the momentum strategist.

3.2. One hundred randomly selected 10-year periods ptWe designed a simulation experiment to determine whether the momentum strategies outperform the NDS in any given period and for any given set of assets. For each run, we selected an experimental period of 120 months with the starting month randomly chosen between July 1926 and December 1995. We then used all sample stocks in that period and the portfolio formation methods described above to form the momentum strategies and the NDS. The set of sample stocks that are used to construct the equally weighted portfolios is the same as that for constructing the momentum portfolios. We draw 100 samples with replacements and obtain 100 sample averages for monthly excess returns of each of the winner and the loser portfolios as well as the prots of both the momentum and the NDS strategies. We compute average excess returns of the momentum portfolios and an average monthly prot of the zero-net-worth NDS over the 120 months. We test, for each sample, the null hypothesis H0 : E [rP ] = E [r ] against the alternative H1 : E [rP ] > E [r ] using the T = 120 monthly prots of the momentum and

4. The relation between the momentum and the 1/ N diversication strategies We analyse the theoretical relation between the momentum and the 1/ N naive diversication strategies by assuming that, for every period t , the cross-section of excess returns is characterized by a factor model: rt = t + Bt xt + t , t N (0, t2 I Nt ), t = 1, . . . , T , (1)

660

Feature

where t = [1t , . . . , it , . . . , Nt t ], Bt = [1t it , . . . , N t ] is the vector of factor loadings, xt are common risk factors, t2 is the idiosyncratic variance at time t and Nt is the total number of assets at time t . The idiosyncratic variance of NDS is simply Var(r t | xt ) = t2 (1/ Nt ). Also note that Cov(r jt , r 1(1/ Nt ) | xt ) = t2 F jt 1 (1/ Nt ) = t2 (1/ Nt ), j = W,L. Therefore, it is easy to see that the conditional mean of the excess returns on the winner and loser portfolios can be given by E (r jt | xt , r t ) = (t + Bt xt ) F jt + r t (t + Bt xt ) 1 1 Nt .

Since F jit = 10/ Nt , if the i th asset is in the winner portfolio, we obtain the idiosyncratic variance as Var(r jt | x, r ) = 2 F jt F jt 10 = t2 . Nt

Therefore, it immediately follows that the conditional mean and idiosyncratic variance of the momentum prots are

Figure 2. Distributions of the average prots of the NDS and the excess returns of the winner and the loser portfolios and the momentum prots over 100 random sample periods. The difference in prots between the momentum strategy and the NDS is denoted by protNDS. For each run of the simulation, we select an experimental period of 120 months with the starting month randomly chosen between July 1926 and December 1995.

Downloaded by [Chi-Hsiou Hung] at 00:22 30 April 2013

E [rPt (t + Bt x) FP | xt , r t ] = 0, Var[rPt | xt ] = t2 20 . Nt

(2) (3)

The above results suggest the theoretical relations that: (1) the momentum strategies and NDS are orthogonal to each other, and (2) the idiosyncratic variance of prots is 20 times higher than the idiosyncratic variance of the NDS. We test these predictions below.

4.1. Orthogonality Under the assumptions made for the factor model (1) and from (2), the prots of the momentum strategies are orthogonal to the prots of the 1/ N naive diversication strategies. We examine this prediction by plotting the excess returns on the winner portfolio, the loser portfolio and the momentum prots against the 1/ N naive diversication strategies (r ) in gure 3. The scatter plot shows that the slope coefcients for both rW and rL seem to be close to one and that the momentum prots are randomly scattered against r . Formally, we run a weighted (by t2 / Nt ) linear regression (WLS) of the momentum prots on the prots of the NDS. Specically, we examine the following time-series model to test the null H0 : P = 0: r jt = j + j r t + u jt , t = 1, . . . , T , j {W,L,P}, Var(u jt ) where t2 t2 , Nt

Figure 3. Scatter plots of the excess returns on the winner (blue line) and the loser (green line) portfolios and the momentum prots (red line) versus the prots of the NDS.

4.2. Risk-adjusted prots and risk exposure We examine the risk-adjusted prots and risk factor exposures of the momentum and the 1/ N naive diversication strategies. Specically, we run the time-series regression r jt = j + j MKT t + s j SMBt + h j HMLt + m j MOM t + v jt , j {W,L,P,NDS}, (4) where r jt is the excess portfolio return or prot for strategy j at time t , and MKT is the return on the CRSP value-weighted market index at time t in excess of the one-month T-bill rate. The Fama and French (1993) size factor, SMB, is dened as the monthly return difference between two portfolios that consist of large and small stocks. The FamaFrench value factor, HML, is dened as the monthly return difference between two portfolios with high and low book-to-market equity ratio. The momentum factor, MOM , is the monthly return difference between two portfolios with high and low returns over the past

is the variance of t in (1), and Nt is the total number of assets at time t . The WLS results show a P of 0.04 (t = 0.43), which is in line with the predictions of (2) and suggests that the momentum prots are orthogonal to the prots of the NDS. We also test whether the sensitivities of the winner and the loser portfolios on the NDS are statistically different from one. We run WLS regressions of excess portfolio returns on the prots of the NDS to test the null of H0 : W = L = 1. The results show that W = 0.85 and L = 0.81 are both smaller than one with t = 1.91 and t = 2.52, respectively.

Feature

661

Table 3. Risk exposures of the six-month momentum and the 1/N naive diversication strategies. Panel A reports the factor loadings of the excess returns on the winner (Winner) and the loser (Loser) portfolios, the momentum prots and the NDS for the period from January 1927 to December 2005. Momentum prot is for the strategies that are long in the winner and short in the loser portfolios. Panel B reports the factor loadings of WinnerNDS, LoserNDS and Momentum prot-NDS, which are the differences between the excess returns of the winner and loser portfolios, the momentum prots and the prots of the NDS, respectively. MKT is the return on the CRSP value-weighted market index in excess of the one-month T-bill rate. SMB is the monthly return difference between two portfolios that consist of large and small size stocks. HML is the monthly return difference between two portfolios with high and low book-to-market equity ratios. MOM is the monthly return difference between two portfolios with high and low returns over the past 2 to 12 months. The t -statistics in parentheses are calculated by applying the Newey-West heteroskedasticity-and-autocorrelation-consistent standard errors. , and denote statistical signicance at the 10level, respectively. Constant MKT SMB 0.8001 (10.70) 0.7992 (16.85) 0.0010 (0.01) 0.4251 (14.50) 0.3751 (4.67) 0.3741 (6.59) 0.4241 (3.93) HML 0.0484 (1.18) 0.0635 (1.29) 0.1119 (1.64) 0.0888 (4.48) 0.1372 (2.93) 0.0253 (0.45) 0.2007 (2.89) MOM 0.5495 (15.17) 0.4471 (12.45) 0.9966 (21.51) 0.0186 (0.88) 0.5680 (13.46) 0.4286 (11.51) 1.0151 (18.63)

Downloaded by [Chi-Hsiou Hung] at 00:22 30 April 2013

Panel A: Excess returns of the momentum strategy and the NDS Winner 0.0039 1.2230 (44.14) (4.85) Loser 0.0007 1.1580 (0.74) (35.89) Momentum prot 0.0032 0.0649 (1.72) (2.49) NDS 0.0025 1.0502 (96.80) (7.10) Panel B: Return differences between the momentum strategy and the NDS WinnerNDS 0.0014 0.1728 (6.66) (1.72) LoserNDS 0.0018 0.1079 (3.20) (1.93) Momentum protNDS 0.0007 0.9852 (0.50) (26.85)

2 to 12 months. The coefcients s j , h j and m j are the loadings on the size, value and the momentum factors, respectively. Table 3 shows the regression results for the entire sample period. The NeweyWest standard errors are applied to correct for heteroskedasticity and autocorrelation of residuals. Panel A shows that both the winner and loser portfolios are signicantly exposed to the market, SMB and momentum factors. The momentum strategies have an average risk-adjusted prot of 0.3% per month and are marginally exposed to the market factor and with a highly signicant loading of nearly 1 on the momentum factor, but with insignicant loadings on the SMB and HML factors. The NDS has a highly signicant alpha of 0.25% per month with statistically signicant loadings of 1.05, 0.42 and 0.09, respectively, on the market, SMB and HML factors. The NDS does not have signicant exposure on the MOM factor. Thus the momentum and the NDS bear different sources of risk. After accounting for the factor exposures, the magnitude of the risk-adjusted momentum prot is close to that of the NDS. Panel B of table 3 shows that the differences between the excess returns of the winner portfolio and the prots of the NDS have a positive and marginally signicant alpha with positive and highly signicant exposures to the market, SMB and MOM , but with a negative exposure to the HML factor. The differences between the excess returns of the loser portfolio and the prots of the NDS have a negative and marginally signicant alpha with positive and highly signicant exposures to the market and the SMB factor, but with a negative and highly signicant exposure to the MOM factor. Importantly, the difference in prots between the momentum and the NDS has a very small and statistically insignicant alpha of seven basis points per month.
The returns on these factors are obtained from Ken Frenchs data library.

4.3. Idiosyncratic variance ratio We test the null hypothesis that the ratio of idiosyncratic variance (CVR = Var[rPt |xt ]/Var(r t |xt )) between the momentum prots and the NDS prots are equal to 20 as is the prediction of (3). Specically, we test the null hypothesis: H0 : Var[rPt | xt ] = 20. Var(r t | xt )

Since the idiosyncratic variance (t2 ) and the number of assets ( Nt ) vary over time, we cannot use the traditional variance ratio test. We test this hypothesis by constructing a moment-based estimator for the CVR, CVR = 1 T
T

CVRt ,
t =1

where CVRt =

2 vP t 2 vNDSt

where vPt and vNDSt are the residuals from the regressions in (4). Note that E (CVRt ) = E
2 vP t 2 vNDS t

2 E vP t 2 E vNDS t

Var[rPt | xt ] . Var(r t | xt )

The second equality follows from Heijmans (1999). Finally, using (3) we obtain the expected variance ratio. Therefore, under the null hypothesis by the central limit theorem, the standardized statistic is ZCVR = (CVR E (CVR)) D T N (0, 1). stdev (CVR) (5)

Under the null hypothesis E (CVR) = 20. We obtain a calculated ZCVR of 1.005 over the whole sample period, and, hence, accept the hypothesis that the idiosyncratic variance of the momentum portfolio is 20 times higher than the idiosyncratic variance of the NDS.

662

Feature

Table 4. Protability of the 12-month momentum and the 1/N naive diversication strategies. This table presents mean excess returns of the momentum strategies (12 month by 12 month) and the 1/N naive diversication strategies (NDS) for the whole sample period (panel A), the sub-sample periods of JagadeeshTitman (1993, 2001) (panels B and C), the periods during the great depression (panel D) and the dot-com bubble (panel E). We use the one-month T-bill rate as a proxy for the risk-free rate. Winner and Loser are, respectively, the returns on the winner and the loser portfolios in excess of the one-month T-bill rate. WinnerNDS and LoserNDS are, respectively, the return differences between Winner, Loser and the NDS. Momentum protNDS is the difference between the prots of the momentum and the NDS. t -Statistics are in parentheses. , and denote statistical signicance at the 10%, 5% and 1% level, respectively. Winner Loser Momentum prots NDS 1.04 (4.79 ) 0.83 (2.33 ) 1.71 (4.02 ) 1.12 (1.34) 5.11 (0.90) WinnerNDS 0.35 (2.70 ) 0.45 (2.22 ) 0.66 (1.70 ) 0.28 (0.53) 1.19 (0.60) LoserNDS 0.07 (0.56) 0.19 (1.34) 0.26 (1.07) 0.17 (0.33) 0.04 (0.02) Momentum protsNDS 0.76 (2.42 ) 0.20 (0.46) 1.31 (2.85 ) -1.00 (1.04) 6.26 (0.70)

Downloaded by [Chi-Hsiou Hung] at 00:22 30 April 2013

Panel A: The whole sample: 01-1928 to 01-2005 Mean return (%) 1.38 1.11 0.28 (1.55) (5.29 ) (3.76 ) Panel B: JagadeeshTitman (1993): 01-1965 to 12-1989 Mean return (%) 1.28 0.64 0.63 (1.50) (2.45 ) (2.75 ) Panel C: JagadeeshTitman (2001): 01-1990 to 12-1998 Mean return (%) 2.36 1.96 0.40 (0.96) (3.14 ) (3.42 ) Panel D: Great depression: 08-1929 to 03-1933 Mean return (%) 1.40 1.28 0.12 (1.15) (1.09) (0.21) Panel E: Dot-com bubble: 01-1995 to 02-2000 Mean return (%) 3.93 5.08 1.15 (1.03) (0.72) (0.33)

Table 5. Risk exposures of the 12-month momentum and the 1/N naive diversication strategies. Panel A reports the factor loadings of the excess returns on the (12 month by 12 month) winner (Winner) and the (12 month by 12 month) loser (Loser) portfolios, the momentum prots and the NDS for the period from January 1927 to December 2005. Momentum prot is for the (12 month by 12 month) strategies that are long in the winner and short in the loser portfolios. Panel B reports the factor loadings of Winner-NDS, Loser-NDS and Momentum prot-NDS, which are the differences between the excess returns of the winner and loser portfolios, the momentum prots and the prots of the NDS, respectively. M K T is the return on the CRSP value-weighted market index in excess of the one-month T-bill rate. S M B is the monthly return difference between two portfolios that consist of large and small size stocks. H M L is the monthly return difference between two portfolios with high and low book-to-market equity ratios. M O M is the monthly return difference between two portfolios with high and low returns over the past 2 to 12 months. The t -statistics in parentheses are calculated by applying the NeweyWest heteroskedasticity-andautocorrelation-consistent standard errors. , and denote statistical signicance at the 10level, respectively. Constant MKT SM B 0.7377 (9.04) 0.9209 (14.19) 0.1832 (1.56) 0.4251 (14.37) 0.3126 (4.08) 0.4958 (6.02) 0.6084 (5.71) HML 0.1945 (3.43) 0.2894 (4.69) 0.4839 (5.59) 0.0894 (4.46) 0.2839 (4.41) 0.2000 (3.33) 0.5734 (6.04) MOM 0.5280 (12.28) 0.1302 (2.43) 0.6582 (11.13) 0.0183 (0.86) 0.5463 (11.47) 0.1119 (2.32) 0.6765 (9.64)

Panel A: Excess returns of the momentum strategy and the NDS Winner 0.0010 1.2493 (1.07) (42.23) Loser 0.0013 1.1573 (1.01) (27.14) Momentum prot 0.0002 0.0920 (0.14) (2.26) NDS 0.0025 1.0498 (95.76) (7.03) Panel B: Return differences between the momentum strategy and the NDS WinnerNDS 0.0015 0.1994 (1.55) (6.42) LoserNDS 0.0013 0.1075 (1.03) (2.68) Momentum protNDS 0.0028 0.9579 (21.03) (1.68)

4.4. Twelve-month momentum For a robustness check, we further consider the momentum strategies based on 12-month formation and 12-month holding periods as in Jegadeesh and Titman (1993). Using exactly the same criteria for the selection of the stock set, the stocks that are feasible for trading exclude those whose prices are below $5 at portfolio formation. Consistent with Jegadeesh and Titman (1993), panel B of table 4 reports the average (12-by-12) momentum prot as 0.63% per month and statistically signicant. Panel A for the entire sample period shows that the winner

and the loser portfolios have average excess returns of 1.38% and 1.11% per month, respectively. The average momentum prot is 0.28% per month, but statistically insignicant. The average prot of the NDS remains statistically signicant at 1.04% per month. The momentum strategies under-perform the zero-net-worth NDS strategies by 0.76% per month. The overall pattern of the results presented in table 4 shows that momentum strategies do not outperform the NDS strategies. Table 5 shows, for the entire sample period, the regression results and the t -statistics using the NeweyWest standard errors. In contrast to the results in table 3, panel A of table 5

Feature shows that the HML factor becomes statistically signicant for the winner portfolio, the loser portfolio and the momentum prot. Further, the average risk-adjusted return on the winner portfolio is essentially zero and statistically insignicant, while the alpha of the NDS remains statistically signicant at 0.25% per month. Panel B of table 5 shows that the model alpha of the difference between the excess return on the winner portfolio and the NDS prot is now negative, reecting the insignicant model alpha of the winner portfolio as shown in panel A, while the pattern of risk exposures is similar to that of the 6-month-by-6month momentum. The model alpha of the difference between the loser portfolio and the NDS prot is close to zero and statistically insignicant. The model alpha of the difference in prots between the momentum and the NDS is a negative 28 basis points per month and statistically signicant. Overall, the 12-month momentum strategies under-perform the NDS strategies. Overall, our results still hold, and, in fact, the 12-month momentum strategies under-perform the NDS strategies, showing that the use of adopting the naive 1/ N diversication is a good investment policy relative to the 12-month momentum strategies. 5. Conclusions In this paper we examine the merits and demerits of the active momentum strategies that use stocks of the top and bottom 10% returns and the passive 1/ N naive diversication strategy which is long in the equally weighted portfolio of stocks and short in the risk-free asset. The momentum and the NDS strategies are orthogonal. The NDS generates an average prot of 1% per month, close to that of the momentum strategies over the sample period between 1926 and 2005, and various sub-sample periods, including those examined by Jegadeesh and Titman (1993,2001). The evidence shows that the differences in the average prots between the momentum and the NDS strategies are economically and statistically insignicant. The ndings are robust with respect to market capitalization effects as well as sampling and period-specic effects in our simulations,

663

where we randomly select 10 years for 100 times. From the viewpoint of investment, our ndings suggest that pursuing the active momentum trading would not be benecial relative to the passive 1/ N naive diversication strategy.

References
Avramov, D. and Chordia, T., Asset pricing models and nancial market anomalies. Rev. Financ. Stud., 2006, 19, 10011040. Banerjee, A. and Hung, C-H., Informed momentum trading versus uninformed naive investors strategies. J. Bank. Finance, 2011, 35, 30773089. Benartzi, S. and Thaler, R.H., Naive diversication strategies in dened contribution saving plans. Am. Econ. Rev., 2001, 91, 7998. Benartzi, S. and Thaler, R.H., Heuristics and biases in retirement savings behavior. J. Econ. Perspect., 2007, 21, 81104. DeMiguel, V., Garlappi, L. and Uppal, R., Optimal versus naive diversication: how inefcient is the 1/ N portfolio strategy? Rev. Financ. Stud., 2009, 22, 19151953. Fama, E.F. and French, K.R., Common risk factors in the returns on stocks and bonds. J. Financ. Econ., 1993, 33, 356. Fama, E.F. and French, K.R., Multifactor explanations of asset pricing anomalies. J. Finance, 1996, 51, 5584. Heijmans, R., When does the expectation of a ratio equal the ratio of expectations? Statist. Pap., 1999, 40(1), 107115. Huberman, G. and Jiang, W., Offering vs. choice in 401(k) plans: equity exposure and number of funds. J. Finance, 2006, 61, 763 801. Jegadeesh, N. and Titman, S., Returns to buying winners and selling losers: implications for stock market efciency. J. Finance, 1993, 48, 6591. Jegadeesh, N. and Titman, S., Protability of momentum strategies: an evaluation of alternative explanations. J. Finance, 2001, 56, 699720. Jones, C. and Lamont, O., Short-sale constraints and stock returns. J. Financ. Econ., 2002, 66, 207239. Korajczyk, R.A. and Sadka, R., Are momentum prots robust to trading costs? J. Finance, 2004, 59, 10391082. Liu, L.X. and Zhang, L., Momentum prots, factor pricing, and macroeconomic risk. Rev. Financ. Stud., 2008, 21, 24172448. Liu, W. and Strong, N., Biases in decomposing holding period portfolio returns. Rev. Financ. Stud., 2008, 21, 22432274. Sagi, J.S. and Seasholes, M.S., Firm-specic attributes and the cross section of momentum. J. Financ. Econ., 2007, 84, 389434. Sharpe, W., Asset allocation: management style and performance measurement. J. Portfol. Mgmt, 1992, 719.

Downloaded by [Chi-Hsiou Hung] at 00:22 30 April 2013