Int. Fin. Markets, Inst.

and Money 20 (2010) 5167

Contents lists available at ScienceDirect

Journal of International Financial Markets, Institutions & Money

j o ur na l ho me pa ge : w w w . e l s e v i e r . c o m / l o c a t e / i n t f i n

Pricing assets with higher moments: Evidence from the Australian and us stock markets
Phuong Doan a, Chien-Ting Lin b,, Ralf Zurbruegg b
a b

Business School, University of RMIT, Melbourne, VIC 3000, Australia Business School, University of Adelaide, Adelaide, SA 5005, Australia

a r t i c l e

i n f o

a b s t r a c t
This paper investigates the importance of higher moments of return distributions in capturing the variation of average stock returns for companies listed in the leading S&P US and Australian indices. We nd that Australian stocks are more negatively skewed but less leptokurtic than US stocks. As a result, we nd that co-skewness plays a more important role in explaining Australian returns while co-kurtosis is consistently inuential for US stock returns. We postulate that the differences in results are related to the underlying rm characteristics of the companies in the two indices, where principally the Australian rms are noticeably smaller than their US counterparts and concentrated in a smaller number industry sectors. This implies that for many smaller exchanges around the world higher moment characteristics displayed by the US market may not be applicable. We also show our results are robust to partly explaining average stock returns in the presence of size, value, and momentum effects. 2009 Elsevier B.V. All rights reserved.

Article history: Received 10 January 2009 Accepted 12 October 2009 Available online 22 October 2009 JEL classication: G11 G12 Keywords: Asset pricing Co-skewness Co-kurtosis Fama and French 3 factors Australian stock market

1. Introduction It has long been well documented that stock returns do not follow a normal distribution. For example, Mandelbrot (1963) and Mandelbrot and Taylor (1967) show that stock returns exhibit excess kurtosis, also commonly referred to as fat tail distributions. Fama (1965) nds that large stock returns tend to be followed by stock returns of similar magnitude but in the opposite direction. This can lead to the volatility clustering effect that is related to how information arrives and is received by the market (see Campell and Hentschel (1992)). This clustering in return volatility has raised a fundamental

Corresponding author. Tel.: +61 8 8303 6461; fax: +61 8 8303 7243. E-mail address: Edward.lin@adelaide.edu.au (C.-T. Lin). 1042-4431/$ see front matter 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.intn.2009.10.002

52

P. Doan et al. / Int. Fin. Markets, Inst. and Money 20 (2010) 5167

question on whether a mean and variance asset pricing model using only the rst two moments of the return distribution is adequate in capturing variation in average stock returns. Subsequent voluminous empirical tests on Sharpes CAPM (1964) have largely rejected the validity of the model which assumes that an investors utility function is quadratic and that the co-movement with the market return is the only important factor in pricing stocks (see Campbell et al. (1995) for a comprehensive review). Given that the empirical stock return distribution is observed to be asymmetric and leptokurtic, a natural extension of the elegant but oversimplied two-moment asset pricing model is to incorporate the co-skewness (third moment) and co-kurtosis (fourth moment) factors. An investor whose utility is non-quadratic and is described by non-increasing absolute risk aversion may prefer positive skewness and less kurtosis in the return distribution. Stocks of negative co-skewness and of larger co-kurtosis with the market should therefore be related to higher risk premia. Therefore, movement of higher co-moments unfavourable to the investors risk preferences requires compensation in the form of additional returns. This particular approach of characterizing stock pricing behaviour not only can be intuitively appealing but may also improve the explanatory power of a model on the expected stock returns. In this paper, we examine the importance of co-skewness and co-kurtosis for average stock returns, along with the well documented Fama and French (1993) 3 common risk factors (namely rm size, book-to-market equity (BV/MV), and market returns) and the Jegadeesh and Titman (1993) momentum effect. In particular, we test the presence of higher co-moment effects in the Australian stock market and compare them with those in the US market. Our interest in the behaviour of Australian stocks rests with the glaring absence of any direct studies on the pricing of higher co-moments in the current Australian literature despite some evidence of skewness and kurtosis in the stock return distribution. For instance, Beedles (1986) and Alles and Spowart (1995) nd that Australian stocks exhibit signicant skewness. Furthermore, Bird and Gallagher (2002) and Brands and Gallagher (2004) document that Australian mutual funds are characterized by a leptokurtic distribution. In particular, they noticed that portfolio returns of larger funds had more negative skewness and larger kurtosis relative to smaller mutual funds. Although they suggest that the non-normal distribution may have implications for diversication benets, they did not pursue the analysis to directly measure this through these higher moments. Even for studies on the US, direct examination of higher moments is usually quite limited, and approaches to examining it can be varied. Fang and Lai (1997) examine the importance of co-skewness and co-kurtosis within the four-moment CAPM framework. Dittmar (2002) tests the four moment factors with non-linear pricing kernels to improve the pricing kernels ability to describe the crosssection of returns. His methodology is linked to the nonparametric models of Bansal and Viswanathan (1993) and Chapman (1997) in which the pricing kernel is non-linear in the market return. On the other hand, Kan and Zhou (2003) and Ando and Hodoshima (2006) examine the robustness of the asymptotic covariance matrix of least square errors (LSE) of alphas and betas in a linear asset pricing model when the joint distribution of the factors and error terms may not be normal or conditionally homoskedastic. In contrast, our approach is more consistent with the spirit of Ross APT (1976) or Mertons ICAPM (1973) in which additional factors such as size, BV/MV, and momentum may also capture variation in average stock returns. Our approach can therefore be viewed as a more direct test on the presence of higher co-moments. We draw a comparison of return behaviour between stocks listed as part of the Australian S&P ASX 300 index and the US S&P 500 to highlight the potential different roles that co-skewness and co-kurtosis perform in each market. Since an average Australian rm tends to be smaller and less volatile than in other developed markets, negative skewness could be more dominant than kurtosis in pricing stocks. On the other hand, an average US rm is larger but more volatile (also shown in the descriptive statistics in Tables 1 and 2) such that its variance risk or kurtosis could be a more inuential factor. Despite this casual observation on the different stock market characteristics, no studies to our knowledge have addressed the potential differential pricing effect of co-skewness and co-kurtosis. Most studies, especially those in the US, rather focus simply on the skewness of the return distribution, when kurtosis could be equally or more important. Earlier works including Arditti (1967), Kraus and Litzenberger (1976), Friend and Westereld (1980), Lim (1989), Harvey and Siddique (1999,

Table 1 Summary statistics of the returns of 25 US portfolios formed by size and BV/MV: January 1992July 2007. Size BV/MV Mean SD Unconditional skewness 0.55992 0.03064 0.21915 0.21107 0.21826 0.04318 0.07564 0.01666 0.03459 0.04082 0.14778 0.03205 0.10882 0.10549 0.04084 0.04773 0.21201 0.19771 0.01058 0.02718 0.01686 0.01017 0.05701 0.12154 0.09959 0.124432 Excess unconditional kurtosis 7.05873 2.81503 5.51658 5.52309 4.01917 4.08743 3.28607 4.92388 5.16342 3.03562 3.93310 4.58425 3.77506 3.36970 7.05465 5.97045 6.50191 3.86752 3.95522 4.59737 6.95935 3.38753 3.74465 4.44389 5.63357 4.543455 Normality test JarqueBera 8617.58 1337.54 5166.66 5176.44 2757.41 2819.88 1825.62 4090.45 4498.72 1555.77 2624.53 3546.16 2412.26 1923.17 8397.42 6015.36 7162.45 2549.87 2639.32 3566.29 8171.18 1936.06 2367.89 3341.65 5361.01 3493.092 Probability 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 4049 4049 4049 4049 4049 4049 4049 4049 4049 4049 4049 4049 4049 4049 4049 4049 4049 4049 4049 4049 4049 4049 4049 4049 4049 4049 Observations

Large

Low 2 3 4 High Low 2 3 4 High Low 2 3 4 High Low 2 3 4 High Low 2 3 4 High

Portfolio 11 Portfolio 12 Portfolio 13 Portfolio 14 Portfolio 15 Portfolio 21 Portfolio 22 Portfolio 23 Portfolio 24 Portfolio 25 Portfolio 31 Portfolio 32 Portfolio 33 Portfolio 34 Portfolio 35 Portfolio 41 Portfolio 42 Portfolio 43 Portfolio 44 Portfolio 45 Portfolio 51 Portfolio 52 Portfolio 53 Portfolio 54 Portfolio 55 Index

0.00027 0.00005 0.00005 0.00003 0.00013 0.00019 0.00034 0.00026 0.00025 0.00044 0.00042 0.00046 0.00037 0.00038 0.00056 0.00027 0.00042 0.00040 0.00045 0.00046 0.00069 0.00044 0.00051 0.00045 0.00048 0.000106

0.01193 0.01076 0.01115 0.01213 0.01593 0.01103 0.00920 0.01060 0.01168 0.01191 0.00941 0.00963 0.00979 0.01137 0.01074 0.01047 0.01001 0.01024 0.01039 0.01109 0.01381 0.01244 0.01097 0.01024 0.01131 0.004235

P. Doan et al. / Int. Fin. Markets, Inst. and Money 20 (2010) 5167

Small

The sample consists of all stocks listed at any point in time during the sample period that were part of the S&P 500. Each portfolio comprises of 4049 observations and is constructed by the intersection of 5 size and 5 BV/MV groups. Portfolio 11 contains large-cap and low BV/MV stocks while portfolio 55 contains small-cap and high BV/MV stocks. Co-higher moments are based on the direct method. The daily returns of each portfolio are the value-weighted returns of stocks in the portfolio. Unconditional skewness and kurtosis are the third and the fourth moment of the daily returns. The JarqueBera normality test is a test of whether the stock returns are normally distributed.

53

54

Table 2 Summary statistics of 25 Australian portfolios formed by size and book-to-market value: January 2001July 2007. Size BV/MV Mean SD Unconditional skewness 0.4141 0.89338 0.14476 1.37125 0.61091 1.1804 0.54744 0.47329 0.7352 0.72905 0.26806 0.53622 0.40085 0.1679 0.55098 0.04241 0.12133 0.57527 2.64937 0.40689 0.507958 1.5602 0.43624 0.41203 0.35203 0.488384 Excess unconditional kurtosis 28.01132 8.17112 5.442801 22.56568 5.383741 11.55367 3.587539 3.539279 4.796208 5.276479 3.737123 6.153249 3.661214 2.891133 3.001116 2.27273 2.1406 4.228345 41.52602 3.543715 6.132786 16.25734 2.173984 2.219143 1.771397 3.058635 Normality test JarqueBera 55659.54 4958.399 2105.548 36623.33 2160.095 9855.93 997.1531 951.3193 1783.619 2123.931 1010.218 2765.017 995.5952 600.4111 724.4126 366.6006 328.9343 1360.986 124207.5 936.9783 2738.833 19422.44 388.9225 397.1606 257.527 725.1418 Probability 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1701 1701 1701 1701 1701 1701 1701 1701 1701 1701 1701 1701 1701 1701 1701 1701 1701 1701 1701 1701 1701 1701 1701 1701 1701 1701 Observations

Large

Low 2 3 4 High Low 2 3 4 High Low 2 3 4 High Low 2 3 4 High Low 2 3 4 High

Portfolio 11 Portfolio 12 Portfolio 13 Portfolio 14 Portfolio 15 Portfolio 21 Portfolio 22 Portfolio 23 Portfolio 24 Portfolio 25 Portfolio 31 Portfolio 32 Portfolio 33 Portfolio 34 Portfolio 35 Portfolio 41 Portfolio 42 Portfolio 43 Portfolio 44 Portfolio 45 Portfolio 51 Portfolio 52 Portfolio 53 Portfolio 54 Portfolio 55 Index

0.001078 0.000641 0.00097 0.000314 0.000682 0.0000503 0.000167 0.000248 0.00003 0.000373 0.000369 0.000331 0.00045 0.00082 0.000275 0.000477 0.000363 0.000313 0.000315 0.000828 0.0005 0.000445 0.000599 0.000637 0.000404 0.000154

0.015451 0.011673 0.013862 0.013093 0.014213 0.009885 0.008852 0.010429 0.011205 0.010618 0.008598 0.009714 0.009091 0.009157 0.012259 0.007567 0.007203 0.008484 0.010089 0.008785 0.009136 0.00817 0.00726 0.007411 0.008728 0.002988

P. Doan et al. / Int. Fin. Markets, Inst. and Money 20 (2010) 5167

Small

The sample consists of all stocks listed at any point in time during the sample period that were part of the S&P ASX 300. Each portfolio comprises of 1701 observations and is constructed by the intersection of 5 size and 5 BV/MV groups. Portfolio 11 contains large-cap and low BV/MV stocks while portfolio 55 contains small-cap and high BV/MV stocks. Co-higher moments are based on the direct method. The daily returns of each portfolio are the value-weighted returns of stocks in the portfolio. Unconditional skewness and kurtosis are the third and the fourth moment of the daily returns. The JarqueBera normality test is a test of whether the stock returns are normally distributed.

P. Doan et al. / Int. Fin. Markets, Inst. and Money 20 (2010) 5167

55

2000) and Smith (2007) have examined the return distribution that only includes skewness. Our study intends to ll the gap in the literature by addressing the relative importance of higher co-moments in markets of different characteristics. Examining these two different markets also provides some robustness checks on the importance of each pricing factor. We nd strong evidence for the higher co-moment factors in the US stocks and co-skewness in Australian stocks. Consistent with the investor preference theory discussed earlier, average stock returns are negatively related to co-skewness but positively related to co-kurtosis. These two factors remain robust when we regress them along with excess market returns, size, BV/MV, and momentum in our data. Our results therefore suggest that both co-skewness and co-kurtosis explain part of the return variation that is not captured by these other well known factors. Our ndings do not support Chung et al. (2006) who argue that Fama and French factors are proxies for the pricing of higher order co-moments, but are more consistent with Smith (2007) who nds that adding co-skewness to the Fama and French 3 factor model improves the explanatory power of the model. Our analysis also shows that although co-skewness is important in both the Australian and US markets, its inuence varies in degree. The co-skewness effect is stronger for the Australian stocks compared to the co-kurtosis effect for US stocks. The importance of the co-skewness effect can perhaps be partly explained by the positive relationship between size and skewness which is found to be more pronounced in Australia than in the US. Since the average size of the sampled Australian rm is smaller than the US rm,1 it follows that Australian stocks may appear to be more sensitive to downside risk given that the return distribution is more negatively skewed. Co-skewness may therefore play a larger role in the Australian market. On the other hand, the return distribution in the US appears to be more leptokurtic as the stock returns tend to be more volatile. Subsequently, the signicance of the co-kurtosis effect is more noticeable in the US data. The source of the larger return volatility in the US market may be traced to the specic characteristics of the US rms. Contrary to Australian rms which are more closely related to primary industries in commodity and mining,2 US rms (at least from our sample of S&P 500 rms) are more represented by technology related rms or high growth rms. Their returns therefore display more extreme values at both tails, leading to co-kurtosis being more inuential for US stocks. The remainder of the paper is as follows. Section 2 describes the data and our methodology while Section 3 presents the empirical results. Section 4 concludes the study. 2. Data and methodology 2.1. Data Our sample consists of all stocks in the Australian S&P ASX 300 and the US S&P 500 indices. The advantage of the data is that they are a good proxy for the Australian and the US market portfolios, plus are derived using similar weighting methodology from the S&P. They include the leading companies, by market capitalization, across different industries for each country. Another advantage is that these relatively large stocks mitigate the nonsynchronous trading problem that tends to be associated with smaller rms. Scholes and Williams (1977) and Dimson (1979) show that small stocks where infrequent trading may be common can cause positive serial correlation in stock returns.3 However, since larger stocks tend to exhibit less skewness and kurtosis, the selection of the rms in our sample may bias against us nding the presence of higher co-moments. All the stock and index daily returns are obtained from Datastream. The sample period for the Australian data starts from inception of the S&P ASX 300 series in January 2001July 2007, yielding approximately 510,000 rm-year observations. Since the US data are available from an earlier date, the

1 At the end of 2007 the average market capitalization of S&P500 and ASX300 stocks were US$ 5.9 billion and US$ 3.3 billion, respectively. 2 Also, the ASX 300 is not well diversied across industries, relative to the S&P 500, with over 51% of stocks in the Australian index (as of the end of 2007) being classied as either a nancial or resource stock. 3 We chose to utilise daily data for our study as Kirchler and Huber (2007) argue that skewness and kurtosis become more prominent when higher frequency data are examined.

56

P. Doan et al. / Int. Fin. Markets, Inst. and Money 20 (2010) 5167

sample period starts from January 1992 to July 2007 for a total of 2.03 million rm-year observations. To directly compare the results between the two markets, we also run the analyses over the same sampled period from January 2001 to July 2007 on both sets of data. Since the outcomes are essentially the same from those over the full sampled period, we did not tabulate them in the paper. For proxies of the risk-free rates, we use the 90-day bank bill and the 30-day Treasury bill rates of Australia and the US, respectively.

2.2. Portfolio formation and measurements To form portfolios based on size and BV/MV, we follow the methodology of Fama and French (1993). Firms of S&P ASX 300 and S&P 500 for each year are ranked according to their market capitalization at the beginning of the year and are divided into ve quintiles with about equal number of stocks in each quintile. We then take the difference between returns of the biggest and smallest portfolios (SMB) to mimic the risk factor in returns relating to rm size. The stocks are further ranked by BV/MV independently and sorted into ve portfolios. The HML factor is then estimated by taking the difference in returns between the highest and lowest BV/MV ratios. The 25 portfolios are subsequently formed by the intersection of ve size and ve BV/MV quintiles as shown in Table 1. We repeat the process each year to rebalance the portfolios and to estimate the size and value factors from the ASX 300 and S&P 500 stocks. For co-skewness (or co-kurtosis) factors, the co-skewness (or co-kurtosis) of each stock is rst calculated according to the following equations; Co-skewness: = E [{ri E (ri )}{rM E (rM )}2 ] E {ri E (ri )} E [{rM E (rM )}]
2 2

iM

Cov(ri , r 2 ) SD(ri )Var(rM )

(1)

Co-kurtosis: iM = E [{ri E (ri )}2 {rM E (rM )}2 ] E {ri E (ri )} E [{rM E (rM )}]
2 2

Cov(r 2 , r 2 ) Var(ri )Var(rM )

(2)

where ri and rM are the returns of stock i and the market returns respectively, and E(ri ) and E(rM ) are the expected returns of stock i and the expected market returns, respectively. The stocks are then ranked based on their co-skewness (or co-kurtosis) and are then sorted into ve quintile portfolios with an approximate equal number of stocks. Therefore, quintile 1 contains the highest co-skewness (or co-kurtosis) and quintile 5 the lowest. The difference of the return of the highest co-skewness (co-kurtosis) portfolio minus the return of lowest co-skewness (co-kurtosis) portfolio captures the return premium that is related to co-skewness (co-kurtosis). Finally, we follow the methodology of Jegadeesh and Titman (1993) to estimate the momentum factor. Stocks are rst sorted into ve quintiles in descending order on the basis of their past daily returns. Based on these rankings, we form an equal-weighted portfolio within each quintile. The rst quintile portfolio containing stocks with the highest returns is the winners portfolio and the bottom portfolio is the losers. The difference in returns between the winner portfolio and the loser portfolio is attributed to the return premium for the momentum strategy.

3. Empirical analysis 3.1. Summary statistics We rst present the summary statistics of the daily returns of the 25 portfolios of the US and Australian stocks in Tables 1 and 2, respectively. In addition to the mean and standard deviation of the

P. Doan et al. / Int. Fin. Markets, Inst. and Money 20 (2010) 5167

57

returns, the reported unconditional skewness and excess kurtosis are calculated as follows: Skewness = 1 T 1
T 3

Rp R
R T 4

(3)

t =1

Excess kurtosis =

1 T 1

RP R
R

(4)

t =1

where Rp is the daily return of a portfolio, R is the standard deviation of the portfolio returns, and T is the number of observations. The excess kurtosis is obtained by subtracting the unconditional kurtosis from (3), the unconditional kurtosis of a normal distribution. For co-skewness and co-kurtosis in our later regression analysis, we estimate them according to Eqs. (3) and (4). Consistent with earlier studies, Table 1 shows that smaller portfolios tend to outperform larger portfolios in the US even after we control for BV/MV. This size effect is most apparent on the returns of the largest portfolio in which 4 out of 5 sub-portfolios sorted by BV/MV earn less average returns than the S&P 500 index. The remaining sub-portfolios within each portfolio however earn higher average returns than the index. Controlling for the size effect, we nd that the value effect is less apparent. The increase in average returns from lowest to highest BV/MV sub-portfolios in each size portfolio is less than monotonic. On the standard deviation of the sub-portfolios, we also fail to detect any systematic patterns across either size or BV/MV. As expected, the average return of each portfolio is asymmetric and leptokurtic. Of the 25 portfolios, 14 and 11 of these portfolios exhibit negative and positive skewness, respectively. Size appears to be positively related to skewness.4 Table 1 shows that 4 out of the 5 smallest portfolios according to size tend to exhibit negative skewness. Overall, the average unconditional skewness is low for the US data, ranging from 0.56 to 0.22. The return distribution however exhibits heavy tails systematically where the unconditional excess kurtosis estimates range from 2.82 to 7.06. In other words, the kurtosis of the sampled portfolios has largely exceeded the kurtosis of 3 for the normal distribution. Based on these observations, if co-kurtosis is priced in the US, our subsequent analysis should nd that the fourthmoment will be more inuential on the average stock returns. The JarqueBera tests of standard normality, which measure the difference of skewness and kurtosis of the series with those of the normal distribution, also show that every sub-portfolio is signicantly non-normal at the 1% level. For the Australian data, Table 2 shows that small portfolios earn on average higher returns than large portfolios. In particular, only the largest sub-portfolios based on size have negative average returns. We also fail to nd a monotonic relationship between average returns and BV/MV after controlling for size. There also appears to be little correlation between size and BV/MV with standard deviation. The stock return behaviour for the rst two moments thus far is similar to that of the US tabulated in Table 1. When we estimate unconditional skewness and co-kurtosis for the Australian data, we nd that the stock returns are more asymmetric but less leptokurtic. The average skewness for the ASX 300 index is 0.4884 compared to 0.1244 for the S&P 500. Furthermore, 24 out of the 25 sub-portfolio return distributions are negatively skewed. The degree of the negative skewness for the sub-portfolios is also larger than that found in the US data. For example, the negative skewness in 9 of the 24 sub-portfolios is larger than the largest negatively skewed portfolio in the US sample. In contrast, Australian stock returns appear to be less leptokurtic than those in the US The excess kurtosis for the Australian and US markets are 3.059 and 4.543, respectively. The excess kurtosis however is more dispersed in the Australian sub-portfolios where it varies from 1.77 to 41.52, compared to the range of 2.82 and 7.06 in the US stocks. Before we run the regression analyses of average portfolio returns on the higher co-moments and other controlled factors, we examine the correlations between these independent variables in both markets. On the US stocks shown in Table 3, the correlations across different variables are generally low. They range from 0.54 to 0.68 where the majority of the correlations fall within 0.2 to 0.2. In particular, the correlations between the two high co-moments and SMB and HML are weak,

We present the correlations in Table 3 which shows that size is positively correlated with skewness in both markets.

58

P. Doan et al. / Int. Fin. Markets, Inst. and Money 20 (2010) 5167

Table 3 The correlations between explanatory variables. RM Rf US RM Rf Co-skewness Co-kurtosis SMB HML Momentum Australia RM Rf Co-skewness Co-kurtosis SMB HML Momentum 1 0.0288 0.5430 0.1631 0.1634 0.0941 1 0.5334 0.3115 0.2144 0.1984 0.1223 Co-skewness Co-kurtosis SMB HML Momentum

1 0.1119 0.0693 0.0475 0.0767

1 0.0528 0.2635 0.4220

1 0.6837 0.1781

1 0.0697

1 0.4416 0.3799 0.2434 0.4740

1 0.0681 0.2937 0.1176

1 0.2082 0.5926

1 0.1690

This table reports the correlations between excess market returns, RM Rf , co-skewness, co-kurtosis, size (SMB), value (HML), and momentum for the US S&P 500 stocks and the S&P ASX 300 Australian stocks.

between 0.26 and 0.07, which implies that co-skewness and co-kurtosis are not proxies for SMB and HML. It is also interesting to note that the highest correlation of 0.68 is found between SMB and HML.5 Similarly, the correlations are also low across the independent variables in the Australian market. Not surprisingly, excess market returns tend to correlate more with other factors but not to the extent that it creates a multicollinearity issue in our regression analyses. The positive correlations between co-skewness and co-kurtosis contradict those in the US market, although they remain very low. It suggests that there may not be a systematic relationship among these independent factors. Momentum however tends to be consistently negatively correlated with both co-skewness and co-kurtosis in both markets. Measured as the returns of winner portfolios minus the returns of loser portfolios, a larger momentum effect is perhaps related to larger negative skewness of the loser portfolio than that of the winner portfolio (see Harvey and Siddique (2000)). 3.2. The co-skewness and co-kurtosis effect In our regression analyses, we rst examine the sensitivity of excess portfolio returns to coskewness and co-kurtosis alone. We therefore regress the excess daily returns of the 25 sub-portfolios formed by size and BV/MV on the two higher co-moments only according to the equation below: Rp,t Rf,t = +
1 CoSt

2 CoKt

+ et

(5)

where Rp,t is the return of the portfolio at time t, Rf,t is the risk-free rate at time t, CoSt is the co-skewness factor at time t, and CoKt is the co-kurtosis at time t. As shown in Table 4, we nd that co-skewness is not consistently related to the US portfolio returns. The signicance of co-skewness is only found in 7 out of 25 sub-portfolios of which they tend to be small in rm size and high in BV/MV. It suggests that co-skewness captures limited variation in returns that is not explained by the size and BV/MV effects. We suspect that the smaller dispersion of skewness in the US sub-portfolios shown in Table 1 may not co-vary well with the variability of returns in the time-series regressions. For co-kurtosis, we nd that it is inuential in 24 out of the 25 US portfolios. Its economic signicance is also quite apparent as a 1% increase in co-kurtosis is related to an increase in average returns of between 0.3% and 0.7% Unlike the US portfolios, we nd that both co-moments are important in the Australian market although the explanatory power of co-skewness seems to be more robust for average portfolio returns.

5 We estimate the variance ination factor (VIF) for each independent factor to examine if the correlations among them could bias the results in our regression analysis. We nd that the highest VIF coefcient obtained is 3.13 relating to co-skewness. Multicollinearity among the variables is not usually considered a problem unless a VIF exceeds 5.

Table 4 Regressions for the 25 US portfolios and Australian portfolios formed by size and BV/MV. BV/MV Size US portfolios Intercept Low Large 2 3 4 Small Large 2 3 4 Small Large 2 3 4 Small Large 2 3 4 Small Large 2 3 4 Small Portfolio 11 Portfolio 21 Portfolio 31 Portfolio 41 Portfolio 51 Portfolio 12 Portfolio 22 Portfolio 32 Portfolio 42 Portfolio 52 Portfolio 13 Portfolio 23 Portfolio 33 Portfolio 43 Portfolio 53 Portfolio 14 Portfolio 24 Portfolio 34 Portfolio 44 Portfolio 54 Portfolio 15 Portfolio 25 Portfolio 35 Portfolio 45 Portfolio 55 0.00023 (1.47) 0.00020 (1.27) 0.00040 (2.83)* 0.00025 (1.61) 0.0071 (3.55)** 0.00006 (0.42) 0.00033 (2.52)* 0.00045 (3.15)* 0.00040 (2.72)* 0.00046 (2.59)* 0.00008 (0.57) 0.00029 (2.08)* 0.00036 (2.56)* 0.00041 (2.93)* 0.00053 (3.52)** 0.00004 (0.25) 0.00029 (1.96)* 0.00042 (3.00)** 0.00047 (3.44)** 0.00070 (3.54)** 0.00017 (0.73) 0.00047 (2.90)* 0.00057 (3.79)** 0.00050 (3.58)** 0.00052 (3.80)** Co-skewness 0.05670 (2.60)* 0.01292 (0.59) 0.00198 (0.10) 0.03870 (1.76) 0.01308 (0.47) 0.01374 (0.65) 0.01052 (0.57) 0.03634 (1.82) 0.09540 (4.59)** 0.02510 (1.01) 0.02377 (1.15) 0.03047 (1.57) 0.01544 (0.77) 0.05354 (2.73)* 0.05317 (3.52)** 0.07114 (2.97)* 0.00095 (0.04) 0.01229 (0.61) 0.04592 (2.39)* 0.01308 (0.46) 0.01251 (0.39) 0.02843 (1.24) 0.04114 (1.93) 0.06743 (3.45)** 0.11730 (6.05)** Co-kurtosis 0.68100 (42.85)** 0.50481 (31.76)** 0.29958 (20.78)** 0.34130 (21.35)** 0.56190 (27.49)** 0.52037 (34.02)** 0.38936 (28.86)** 0.34888 (24.05)** 0.33804 (22.31)** 0.55385 (30.7)** 0.62653 (41.7)** 0.60190 (42.50)** 0.39330 (27.08)** 0.51789 (36.25)** 0.54750 (35.58)** 0.54457 (31.10)** 0.70636 (46.56)** 0.70686 (48.55)** 0.57775 (41.28)** 0.56190 (27.49)** 0.66823 (28.51)** 0.61388 (36.97)** 0.48481 (31.18)** 0.67541 (47.36)** 0.71525 (50.61)** Adj. R 0.318 0.202 0.974 0.105 0.159 0.224 0.173 0.129 0.119 0.192 0.302 0.313 0.154 0.252 0.244 0.200 0.352 0.370 0.302 0.158 0.168 0.253 0.191 0.364 0.400
2

Australian portfolios Intercept 0.00014 (0.43) 0.00054 (2.48)* 0.00074 (3.77)** 0.00066 (3.72)** 0.00067 (3.05)** 0.00012 (0.48) 0.00058 (2.87)* 0.00084 (4.03)** 0.00056 (3.39)** 0.00072 (3.85)** 0.00010 (0.34) 0.00077 (3.41)** 0.00080 (4.05)** 0.00068 (3.53)** 0.00082 (5.22)** 0.00045 (1.53) 0.00062 (2.74)* 0.00119 (5.94)** 0.00066 (2.86)* 0.00094 (5.79)** 0.00018 (0.59) 0.00097 (4.61)** 0.00085 (3.37)** 0.00113 (5.94)** 0.00076 (4.05)** Co-skewness 0.86719 (25.29)** 0.42365 (18.34)** 0.31655 (15.26)** 0.19354 (10.29)** 0.17416 (7.50)** 0.61882 (23.95)** 0.34304 (16.17)** 0.45035 (20.38)** 0.21694 (12.46)** 0.27121 (13.77*)* 0.66438 (20.72)** 0.47066 (19.80)** 0.36176 (17.30)** 0.31988 (15.74*)* 0.26321 (15.74) 0.55726 (17.81)** 0.56930 (23.74)** 0.36517 (17.18)** 0.33819 (13.93)** 0.29751 (17.35)** 0.71778 (22.53)** 0.56772 (25.60)** 0.57780 (21.54*)* 0.33596 (16.66)** 0.36113 (18.10)** Co-kurtosis 0.37255 (8.54)** 0.00146 (0.05) 0.00389 (0.14) 0.06981 (2.92)* 0.05748 (1.95)* 0.07447 (2.26)* 0.01964 (0.72) 0.02293 (0.81) 0.10047 (4.53)** 0.07575 (3.02)** 0.16839 (4.12)** 0.04948 (1.64) 0.12297 (4.62)** 0.01096 (0.42) 0.14281 (6.71)** 0.21372 (5.37)** 0.12183 (3.98)** 0.09379 (3.47)** 0.08929 (2.89)* 0.08085 (3.70)** 0.05033 (1.24) 0.11614 (4.11)** 0.16197 (4.75)** 0.15084 (5.87)** 0.10795 (4.25)** Adj. R2 0.276 0.198 0.145 0.094 0.05 0.279 0.155 0.239 0.142 0.147 0.213 0.236 0.223 0.156 0.220 0.160 0.327 0.208 0.148 0.213 0.262 0.359 0.297 0.226 0.232 P. Doan et al. / Int. Fin. Markets, Inst. and Money 20 (2010) 5167

This table reports the regression results of the excess portfolios returns on co-skewness and co-kurtosis factors over the 25 portfolios formed by size and BV/MV. * and ** denote t-statistics at the 5% and 1% level, respectively.

59

60

P. Doan et al. / Int. Fin. Markets, Inst. and Money 20 (2010) 5167

Co-skewness in 24 of the 25 sub-portfolios is signicant compared to 17 sub-portfolios for the cokurtosis. Furthermore, 4 out of the 17 sub-portfolios have negative signs (rather than positive signs) for the stock returns. Therefore, while the returns of the majority of the sub-portfolios can be explained by co-kurtosis, these are not as consistent as with co-skewness. One possible explanation for the differences in the importance of the high co-moments in the two markets may be related to their rm characteristics. An average rm in the US sample tends to be larger than its counterpart in Australia. If smaller rms tend to associate with negative skewness, then co-skewness should play a larger role in asset pricing. Comparing unconditional skewness across portfolios in Tables 1 and 2 does show that the Australian portfolios, on average, exhibit more negative skewness than the US portfolios. The larger daily return volatility of 0.004% among the US stocks compared to 0.0001% of the Australian stocks however indicates that excess kurtosis is on average higher for the US stocks (4.54 in the US vs. 3.06 in Australia). With a larger number of high growth stocks, such as those in the computer related industries represented in the S&P 500 data, it may explain why our US sampled stocks exhibit higher excess kurtosis. It follows that co-kurtosis may be more inuential on the average returns for the US stocks. On the contrary, the sampled Australian rms tend to be concentrated in the mining and resource sectors. These rms tend to be more mature and their corresponding volatilities are also lower. As a result, co-kurtosis plays a lesser role than those found for the average US stock. 3.3. Multivariate regression analysis and robustness checks If higher co-moments do explain average returns, then they must also remain important in the presence of other well known factors. We therefore add the excess market returns to our regressions below: Rp,t Rf,t = + (Rm,t Rf,t ) +
1 CoSt

2 CoKt

+ et

(6)

where Rm,t is the return of the market index at time t, and the other variables are dened earlier in Eq. (5). Eq. (6) allows us to test whether the covariance of the market return volatility with the portfolio returns (co-skewness) and that with the portfolio volatility (co-kurtosis) captures variation in average stock returns in addition to the covariance of the market return and portfolio return. Hence, in this test, we examine if the second moment of the market index is just as important as its rst moment in explaining portfolio returns and volatility. Table 5 shows that at least one of the higher co-moments remains signicant in both markets when we add the excess market returns. In fact, co-skewness and co-kurtosis in the US returns explain 15 and 17 of the 25 sub-portfolio returns, respectively. Overall, 24 out of the 25 sub-portfolios have at least one higher co-moment statistically signicant at the 5% level. Co-skewness in the Australian portfolios remains an important factor where 18 out of 25 subportfolios are statistically signicant in explaining average portfolio returns. However, the results for co-kurtosis are mixed and of those sub-portfolios that can be explained by a signicant fourth moment, it is unclear how it affects the returns as the direction associated with the moment is inconsistent and switches between being positive and negative. Next we regress the average portfolio returns in each sample by size, BV/MV, and momentum effect. That is: Rp,t Rf,t = + (Rm,t Rf,t ) + s SMB + h HML +
3 CoSt

4 CoKt

+ M + et

(7)

where SMB is small minus big, HML is high minus low, and M is the momentum by taking the difference in the returns of the winner and loser portfolios. The remaining variables are dened in the earlier equations. Table 6 shows that the negative relationship between co-skewness and the average returns, and the positive relationships between co-kurtosis and the average returns in the US portfolios have changed little from earlier analyses, despite us adding size, value, and momentum effects. More specically, coskewness and co-kurtosis are signicant in 19 and 21 of the 25 sub-portfolios, respectively. It suggests that not only the higher co-moments are important in pricing assets but they also capture different elements of risk that other established factors fail to explain. Results in Table 7 for the Australian portfolios are again consistent with our earlier ndings. The signicance of co-skewness remains robust

Table 5 Regressions for the 25 US portfolios and 25 Australian portfolios formed by size and BV/MV. BV/MV Size US portfolios Intercept Low Large 2 3 4 Small 2 Large 2 3 4 Small 3 Large 2 3 4 Small 4 Large 2 3 Portfolio 11 Portfolio 21 Portfolio 31 Portfolio 41 Portfolio 51 Portfolio 12 Portfolio 22 Portfolio 32 Portfolio 42 Portfolio 52 Portfolio 13 Portfolio 23 Portfolio 33 Portfolio 43 Portfolio 53 Portfolio 14 Portfolio 24 Portfolio 34 0.00037 (2.87)* 0.00004 (0.29) 0.00025 (2.19)* 0.00010 (0.80) 0.00059 (3.15)** 0.00007 (0.62) 0.00018 (1.79) 0.00280 (2.57)* 0.00023 (1.98)* 0.00025 (1.78) 0.00005 (0.36) 0.00014 (1.26) 0.00019 (1.77) 0.00024 (2.26)* 0.00032 (3.03)** 0.00009 (0.63)* 0.00015 (1.19) 0.00028 (2.43)* RM Rf 1.48370 (40.29)** 1.56396 (43.44)** 1.52175 (48.33)** 1.48339 (39.95)** 1.30854 (25.01)** 1.44240 (40.88)** 1.49602 (52.37)** 1.62114 (53.06)** 1.68938 (53.85)** 1.89818 (48.08)** 1.35979 (38.59)** 1.49858 (48.52)** 1.66189 (55.28)** 1.66557 (57.13*)* 1.90148 (63.81)** 1.41157 (33.15)** 1.45905 (42.04)** 1.46689 (45.04)** Coskewness 0.08519 (4.63)** 0.04294 (2.39)* 0.02723 (1.73) 0.06717 (3.61)** 0.01205 (0.46) 0.01395 (0.79) 0.03924 (2.74)* 0.06746 (4.42)** 0.12783 (8.00)** 0.06151 (3.11)** 0.00233 (0.13) 0.05923 (3.84)** 0.01645 (1.09) 0.08550 (5.86)** 0.08964 (6.01)** 0.09824 (4.61)** 0.02706 (1.56) 0.01587 (0.97) Co-kurtosis 0.33080 (20.68)** 0.13565 (8.67)** 0.05961 (4.35)** 0.00884 (0.55) 0.25302 (11.17)** 0.17990 (11.73)** 0.03623 (2.92)* 0.03377 (2.54)* 0.06072 (4.37)** 0.10581 (6.17)** 0.30556 (19.97)** 0.24817 (18.49)** 0.00102 (0.08) 0.12474 (9.85)** 0.09866 (7.62)** 0.21136 (11.43)** 0.36196 (24.01)** 0.36061 (25.47)** Adj. R2 0.51 0.46 0.43 0.36 0.27 0.45 0.51 0.49 0.48 0.49 0.49 0.56 0.52 0.59 0.62 0.37 0.55 0.58 Australian portfolios Intercept 0.00002 (0.07) 0.00030 (1.41) 0.00034 (1.91) 0.00025 (1.57) 0.00017 (0.57) 0.00005 (0.2) 0.00027 (1.40) 0.00043 (2.24)* 0.00011 (0.81) 0.00015 (1.05) 0.00024 (0.78) 0.00048 (2.21)* 0.00032 (1.87) 0.00016 (1.01) 0.00027 (2.48)* 0.00028 (0.95) 0.00037 (1.65) 0.00075 (4.18)** RM Rf 0.53927 (4.27)** 0.80207 (9.63)** 1.32247 (18.96)** 1.39429 (22.95)** 1.88696 (25.95)** 0.23337 (2.44)* 1.03058 (13.84)** 1.37761 (18.46)** 1.49359 (28.05)** 1.88959 (33.39)** 0.45220 (3.80)** 0.94695 (11.15)** 1.60368 (23.98)** 1.72902 (27.68)** 1.85196 (43.55)** 0.57805 (5.03)** 0.84969 (9.84)** 1.46622 (20.91)** Coskewness 0.79076 (20.53)** 0.30996 (12.21)** 0.12912 (6.07)** 0.00407 (0.22) 0.09326 (4.21)** 0.58574 (20.10)** 0.19699 (8.67)** 0.25511 (11.20)** 0.00527 (0.32) 0.00341 (0.19) 0.60029 (16.64)** 0.33646 (12.98)** 0.13448 (6.59)** 0.07498 (3.93)** 0.00075 (0.05) 0.47534 (13.55)** 0.4488 (17.04)** 0.15738 (7.35)** Co-kurtosis 0.39120 (8.96)** 0.02629 (0.91) 0.04964 (2.06)* 0.02157 (1.02) 0.00780 (0.31) 0.08254 (2.49)* 0.05529 (2.15)* 0.02473 (0.96) 0.04879 (2.65)* 0.01037 (0.53) 0.18404 (4.50)** 0.01673 (0.56) 0.06749 (2.92)* 0.04883 (2.26)* 0.07874 (5.35)** 0.23372 (5.87)** 0.09243 (3.10)** 0.04306 (1.77) Adj. R2 0.28 0.24 P. Doan et al. / Int. Fin. Markets, Inst. and Money 20 (2010) 5167 0.29 0.31 0.32 0.28 0.24 0.37 0.41 0.48 0.22 0.29 0.42 0.42 0.63 0.17 0.36 0.37 61

62

Table 5 (Continued ) BV/MV Size US portfolios Intercept 4 Small 5 Large 2 3 4 Small Portfolio 44 Portfolio 54 Portfolio 15 Portfolio 25 Portfolio 35 Portfolio 45 Portfolio 55 0.00030 (2.96)* 0.00270 (2.87)* 0.00002 (0.05) 0.00031 (2.26)* 0.00042 (3.35)** 0.00035 (3.14)** 0.00035 (3.44)** RM Rf 1.65434 (58.60)** 1.83673 (70.55)** 1.53906 (25.85)** 1.58506 (41.58)** 1.51777 (42.94)** 1.51027 (48.60)** 1.65781 (57.77)** Coskewness 0.07770 (5.50)** 0.06077 (4.67)** 0.01703 (0.57) 0.00200 (0.10) 0.01200 (0.67) 0.09642 (6.21)** 0.14911 (10.39)** Co-kurtosis 0.18726 (15.27)** 0.10072 (0.11) 0.30495 (11.79)** 0.23974 (14.49)** 0.12655 (8.24)** 0.31892 (23.62)** 0.32394 (25.99)** Adj. R 0.62 0.67 0.29 0.48 0.45 0.60 0.67
2

P. Doan et al. / Int. Fin. Markets, Inst. and Money 20 (2010) 5167

Australian portfolios Intercept 0.00006 (0.31) 0.00031 (3.19*)* 0.00015 (0.48) 0.00069 (3.38)** 0.00041 (1.72) 0.00063 (3.93)** 0.000174 (1.20) RM Rf 2.0020 (26.47)** 2.08963 (54.87)** 0.11370 (0.96) 0.94336 (11.97)** 1.48593 (16.07)** 1.67597 (26.79)** 1.97128 (35.07)** Coskewness 0.05450 (2.36) 0.00136 (0.12) 0.70166 (19.50)** 0.43402 (18.04)** 0.36721 (13.02)** 0.09843 (5.15)** 0.08176 (4.76)** Co-kurtosis 0.02003 (0.76) 0.00856 (0.64) 0.05426 (1.33) 0.08350 (3.06)** 0.110557 (3.45)** 0.09286 (4.29)** 0.03975 (2.05)* Adj. R2 0.40 0.71 0.26 0.41 0.39 0.45 0.55

This table reports the regression results of the excess portfolios returns on market excess returns, co-skewness and co-kurtosis over the 25 portfolios formed by size and BV/MV. * and ** denote t-statistics at the 5% and 1% level, respectively.

Table 6 Regressions for 25 U.S portfolios formed by size and BV/MV.

Co-skewness Size Big 2 3 4 Small Low BV/MV 0.1426 (1.28) 0.0559 (3.61)** 0.0117 (0.81) 0.0697 (3.98)** 0.0111 (0.45) SMB Size Big 2 3 4 Small Low BV/MV 0.5189 (26.72)** 0.0308 (1.29) 0.0712 (3.19)** 0.1852 (6.91)** 0.3699 (9.64)** RM Rf Size Big 2 3 4 Small Low BV/MV 1.7003 (65.79)** 1.5935 (50.29)** 1.4735 (49.63)** 1.4652 (40.97)** 1.2571 (24.64)** 2 1.6696 (53.38)** 1.5497 (55.26)** 1.6252 (54.72)** 1.6430 (51.64)** 1.7881 (44.94)** 3 1.5712 (48.84)** 1.5726 (50.42)** 1.6775 (56.11)** 1.6600 (56.01)** 1.8092 (60.60)** 4 1.5789 (37.61)** 1.5605 (44.21)** 1.5254 (45.69)** 1.6346 (55.99)** 1.7683 (66.74)** High BV/MV 1.8675 (33.96)** 1.6688 (43.37)** 1.5666 (43.96)** 1.5414 (50.89)** 1.5989 (67.28)** 2 0.7464 (31.75)** 0.2088 (9.91)** 01330 (5.95)** 0.1287 (5.37)** 0.4047 (13.54)** 3 0.6784 (28.06)** 0.2304 (9.82)** 0.1409 (6.26)** 0.0245 (1.10) 0.2885 (12.86)** 4 0.6139 (19.46)** 0.1949 (7.35)** 0.0556 (2.21)* 0.0031 (0.14) 0.2185 (10.97)** High BV/MV 1.3051 (31.59)** 0.4585 (15.96)** 0.3594 (13.42)** 0.0934 (4.11)** 0.2833 (15.86)** 2 0.0638 (4.17)** 0.0491 (3.58)** 0.0618 (4.25)** 0.1162 (7.46)** 0.0395 (2.03)* 3 0.0495 (.3.14)** 0.0761 (4.98)** 0.0150 (1.02) 0.0819 (5.64)** 0.0685 (4.69)** 4 0.1315 (6.41)** 0.0567 (3.28)** 0.0361 (2.21)* 0.0700 (4.90)** 0.0454 (3.50)** High BV/MV 0.0771 (2.86)* 0.0106 (0.57) 0.0119 (0.68) 0.1039 (7.01)** 0.1412 (12.15)** Co-kurtosis Low BV/MV 0.3437 (11.28)** 0.2431 (15.31)** 0.0892 (5.99)** 0.0848 (4.73)** 0.3736 (14.61)** HML Low BV/MV 0.4810 (24.75)** 0.6851 (28.75)** 0.5118 (22.92)** 0.5874 (21.84)** 0.7910 (20.62)** Momentum Low BV/MV 0.0946 (8.14)** 0.0163 (1.14) 0.1234 (9.24)** 0.0239 (1.48) 0.0533 (2.32)* 2 0.0300 (2.14)* 0.0378 (2.99)* 0.1170 (8.75)** 0.0638 (4.45)** 0.0230 (1.28) 3 0.0392 (2.71)* 0.0107 (0.76) 0.0852 (6.33)** 0.0521 (3.91)** 0.0566 (4.21)** 4 0.0078 (0.41) 0.1084 (6.83)* 0.1197 (7.96)** 0.0503 (3.83)** 0.0325 (2.72)* High BV/MV 0.0711 (2.87)* 0.0987 (5.74)** 0.1086 (6.77)** 0.0574 (4.21)** 0.1063 (9.94)** 2 0.1738 (7.39)** 0.1466 (6.95)** 0.2704 (12.10)** 0.3933 (16.43)** 0.4323 (14.44)** 3 0.1594 (6.58)** 0.0167 (0.71) 0.1861 (8.27)** 0.1846 (8.28)** 0.3682 (16.40)** 4 0.2756 (8.73)** 0.0327 (1.23) 0.1577 (6.28)** 0.0421 (1.92) 0.2186 (10.97)** High BV/MV 0.6854 (16.57)** 0.2665 (9.27)* 0.2940 (10.97)** 0.4379 (19.23)** 0.3892 (21.77)** 2 1.1062 (6.77)** 0.0708 (5.04)** 0.0634 (4.26)** 0.0413 (2.59)* 0.2054 (10.30)** 3 0.2325 (14.42)** 0.2364 (15.12)** 0.0682 (4.55)** 0.1801 (12.12)** 0.1994 (13.33)** 4 0.1446 (6.87)** 0.2954 (16.70)** 0.2734 (16.34)** 0.2191 (14.97)** 0.1610 (12.19)** High BV/MV 0.1731 (6.28)** 0.2253 (11.77)** 0.1164 (6.51)** 0.2140 (14.09)** 0.2182 (18.32)**

P. Doan et al. / Int. Fin. Markets, Inst. and Money 20 (2010) 5167 63

This table reports the regression results of the excess portfolios returns on market excess returns, co-skewness, co-kurtosis, SMB (small minus big), HML (high minus low), and momentum (winner minus loser) over the 25 portfolios formed by size and BV/MV. * and ** denote t-statistics at the 5% and the 1% level, respectively.

64

Table 7 Regressions for 25 Australian portfolios formed by size and BV/MV.

Co-skewness Size Big 2 3 4 Small Low BV/MV 0.2396 (6.31)** 0.2881 (8.48)** 0.2043 (7.08)** 0.0346 (1.36) 0.0060 (0.19) SMB Size Big 2 3 4 Small Low BV/MV 0.8752 (21.64)** 0.0160 (0.44) 0.0265 (0.86) 0.0512 (1.89) 0.1647 (4.99)** RM Rf Size Big 2 3 4 Small Low BV/MV 2.0765 (19.97)** 0.9259 (9.94)** 1.3194 (16.69)** 1.3678 (19.59)** 1.6868 (19.88)** 2 1.4117 (14.17)** 1.1509 (13.01)** 1.4408 (16.27)** 1.6026 (25.41)** 1.6519 (25.13)** 3 1.4742 (11.08)** 0.9156 (9.05)** 1.5966 (20.66)** 1.5298 (20.71)** 1.7060 (34.03)** 4 1.7669 (13.98)** 0.9692 (9.48)** 1.3421 (16.22)** 1.9036 (21.33)** 1.9194 (43.00)** High BV/MV 1.6242 (14.58)** 1.0881 (12.25)** 1.3896 (13.33)** 1.6437 (22.80)** 1.5934 (25.90)** 2 0.7630 (19.68)** 0.0644 (1.87) 0.0179 (0.52) 0.0570 (2.32)* 0.1942 (7.59)** 3 0.6747 (13.04)** 0.0065 (0.16) 0.0026 (0.08) 0.1331 (4.63)** 0.0951 (4.87)** 4 0.8320 (16.92)** 0.0434 (1.09) 0.0452 (1.40) 0.0513 (1.50) 0.1047 (6.03)** High BV/MV 1.0898 (25.14)** 0.1636 (4.73)** 0.0267 (0.65) 0.0218 (0.77) 0.2467 (10.30)** 2 0.0979 (2.69)* 0.1222 (3.79)** 0.2145 (6.63)** 0.0503 (2.19)* 0.0841 (3.50)** 3 0.1641 (3.38)** 0.3733 (10.11)** 0.1637 (5.63)** 0.1667 (6.18)** 0.0792 (4.33)** 4 0.0332 (0.72) 0.3629 (9.73)** 0.2581 (8.55)** 0.0653 (2.00)* 0.0834 (5.12)** High V/MV 0.0249 (0.61) 0.3615 (11.15)** 0.3120 (8.20)** 0.1072 (4.07)** 0.1569 (6.99)** Co-kurtosis Low BV/MV 0.0398 (1.20) 0.0878 (2.95)* 0.0104 (0.41) 0.0676 (3.03)** 0.0104 (0.38) HML Low BV/MV 0.9630 (24.95)** 0.4604 (13.32)** 0.3898 (13.28)** 0.2857 (11.02)** 0.2535 (8.05)** Momentum Low BV/MV 0.1159 (3.27)** 0.0686 (2.16)* 0.0543 (2.02)* 0.0199 (0.80) 0.0318 (1.09) 2 0.0233 (0.68) 0.0558 (1.86) 0.0567 (1.86) 0.0374 (1.74) 0.0827 (3.69)** 3 0.0160 (0.35) 0.0531 (1.54) 0.0367 (1.35) 0.0086 (0.34) 0.0249 (1.45) 4 0.1402 (3.25)** 0.1035 (2.97)** 0.1218 (4.32)** 0.0161 (0.53) 0.0269 (1.76) High BV/MV 0.0216 (0.57) 0.0937 (3.09)** 0.0612 (1.72) 0.0636 (2.59)** 0.0989 (4.71)** 2 0.0022 (0.06) 0.0097 (0.30) 0.0799 (2.42)* 0.0407 (1.74) 0.1659 (6.79)** 3 0.1191 (2.41)* 0.0036 (0.10) 0.0790 (2.70)* 0.0594 (2.17)* 0.0648 (3.47)** 4 0.1803 (3.84)** 0.0661 (1.74) 0.1284 (4.32)** 0.1716 (5.18)** 0.0091 (0.55) High BV/MV 0.9749 (23.57)** 0.4359 (13.21)** 0.5350 (13.57)** 0.2558 (9.55)** 0.2652 (11.61)** 2 0.1119 (3.51)** 0.0177 (0.63) 0.0177 (0.62) 0.0856 (4.25)** 0.0276 (1.32) 3 0.0347 (0.82) 0.0067 (0.21) 0.0652 (2.56)* 0.0743 (3.14)** 0.0565 (3.53)** 4 0.1233 (3.05)** 0.1585 (4.85)** 0.0247 (0.93) 0.0158 (0.55) 0.0266 (1.86) High BV/MV 0.0289 (0.81) 0.0066 (0.23) 0.0375 (1.12) 0.0261 (1.13) 0.0248 (1.26)

P. Doan et al. / Int. Fin. Markets, Inst. and Money 20 (2010) 5167

This table reports the regression results of the excess portfolios returns on market excess returns, co-skewness, co-kurtosis, SMB (small minus big), HML (high minus low), and momentum (winner minus loser) over the 25 portfolios formed by size and BV/MV. * and ** denote t-statistics at the 5% and the 1% level, respectively.

P. Doan et al. / Int. Fin. Markets, Inst. and Money 20 (2010) 5167 Table 8 Adjusted R2 for regressions of 25 US portfolios and 25 Australian portfolios formed by size and BV/MV. Low BV/MV US stocks Size Big 2 3 4 Small Australian stocks Size Big 2 3 4 Small 2 3 4

65

High BV/MV

0.78 0.61 0.53 0.44 0.35

0.58 0.56 0.54 0.52 0.51

0.60 0.59 0.56 0.60 0.64

0.43 0.56 0.59 0.62 0.68

0.43 0.51 0.47 0.65 0.79

0.66 0.32 0.36 0.36 0.35

0.45 0.24 0.37 0.42 0.51

0.30 0.29 0.42 0.43 0.64

0.30 0.37 0.38 0.41 0.72

0.53 0.74 0.45 0.49 0.63

This table reports the adjusted R2 of the regression results reported in Tables 6 and 7.

as it captures variations of 21 out of the 25 sub-portfolio returns. The co-kurtosis effect, however, continues to be weak where it explains only 7 out of 25 sub-portfolios with the correct positive sign. In summary, the picture remains similar throughout our analyses where we nd co-kurtosis is especially important for the US stock returns, while co-skewness plays a more important role for Australian stocks. Although not tabulated in the paper, we also conducted further checks on our results by incorporating a GARCH (1,1) effect in the time-series regressions to adjust for conditional heteroscedasticity. However, the results remain consistent to what we previously illustrated. Our overall ndings are consistent with Arditti (1967), Scott and Horvath (1980), Fang and Lai (1997), and Galagedera et al. (2002) who argue that investors have a negative preference for even moments (i.e. variance and kurtosis) and a positive for odd moments (i.e. return and skewness). Furthermore, Smith (2007) nds that by adding co-skewness to the Fama and French 3 factor model, the model performs better than either the 3 factor or three-moment models alone. Table 8 shows the adjusted R2 for the regressions in Tables 6 and 7. Similar to Smith (2007), adding Fama and French 3 factors and momentum to higher co-moments increases the explanatory power of the model. The adjusted R2 for the 25 sub-portfolios range from 0.24 to 0.79. Our ndings, however, do not nd support Chung et al. (2006) who suggest that the Fama and French 3 factors are proxies for the higher co-moments. 3.4. Fama and Macbeth regressions To investigate if co-skewness and co-kurtosis are priced and command signicant risk premiums in the Australian and US stock markets, we run cross-sectional regressions following the Fama and MacBeth (1973) methodology. First, we estimate the sensitivity of a rms excess return to the risk premium related to each risk factor for each day: Rp,t Rf,t = p,k,t + p,k,t RPk,t + ep,t t = 59, , 0 (8)

where Rp,t Rf,t is portfolio ps excess return for day t, Rk,t is the risk premium related to the kth factor which includes co-skewness, co-kurtosis, size, BV/ME, market, and momentum. The regression is run across each of these six factors so that at the end of each day portfolio p possesses a vector of sensitivity estimates to these factors. This estimation process is repeated daily for the sample period. Next, we run the following cross-sectional regression at the end of each day to examine if coskewness and co-kurtosis are priced or have predictive power over portfolio returns:
K

Rp,t +1 = k,t +1 +
k=1

k,t +1 p,k,t

+ ep,t +1

p = 1, , N

(9)

66

P. Doan et al. / Int. Fin. Markets, Inst. and Money 20 (2010) 5167 Table 9 FamaMacbeth regression estimates. Variables Co-skewness Co-kurtosis SMB HML Market Momentum US 0.0001 (0.38) 0.0019* (2.10) 0.0005 (1.86) 0.0003 (1.08) 0.0003 (1.30) 0.0024** (3.40) Australia 0.0122** (2.80) 0.0007 (0.12) 0.0172** (2.72) 0.0025 (1.24) 0.0008 (0.52) 0.0181 (1.78)

This table presents the results of FamaMacbeth regressions for co-skewness, co-kurtosis, SMB, HML, market, and momentum. At the end of each day, the following cross-sectional regresK

sion is run, Rp,t +1 = k,t +1 +

k =1

k,t +1 p,k,t

+ ep,t +1 ,

p = 1, , N where Rp,t+1 is the portfolio

return at t, p,k,t is the portfolio beta for the kth factor at time t, and ep,t+1 is the error term at
T

time t. The test statistic for

k,t+1

is calculated as tk = ( k 0/

k ) where k = r

1 T j=1

k,j .

* and

** denote t-statistics at the 5% and the 1% level, respectively.

Under the null hypothesis, test statistics for tk = k 0

k r T k,j j =1

Ak ,t +1

can be obtained as follows: (10)

where k =

1 T

Table 9 reports the results of cross-sectional regressions according to the Fama and MacBeth (1973) methodology. Consistent with the results of time-series regressions reported earlier, we nd that cokurtosis and co-skewness are signicantly related to expected portfolio returns in the US and Australia, respectively, in the presence of size, book-to-market, market, and momentum factors. US stocks which are characterized by larger variance and less skewness can be better explained by the variance risk or kurtosis of the return distribution. On the other hand, Australian stocks which tend to be smaller and more negatively skewed are more correlated with the co-skewness factor. As a result, at least one of the higher co-moment factors is priced for the stocks. In sum, risk related to higher co-moment does not appear to be proxied by the well documented four common risk factors. 4. Conclusions This paper suggests that co-skewness and co-kurtosis are important in pricing stocks. The degree of the importance, however, depends on the rm characteristics of the stocks and the risk preference of investors. As Australian stocks tend to be more skewed and less leptokurtic, we nd that co-skewness plays a more signicant role in explaining average stock returns. For US stocks, co-kurtosis is more inuential as returns exhibit higher excess kurtosis. We believe the differences in results between the US and Australian markets come from the fact that the Australian stocks are relatively small to begin with, when compared to their counterparts in the US. The implication being that for many medium to small sized exchanges, co-skewness may be a more relevant factor than co-kurtosis. Our results are also robust to a number of model specications. Although size, BV/MV, and momentum are

P. Doan et al. / Int. Fin. Markets, Inst. and Money 20 (2010) 5167

67

correlated with co-skewness and co-kurtosis, the importance of co-skewness and co-kurtosis remain largely unchanged in their presence. It implies that the higher co-moments capture parts of variation in average stock returns that are not explained by their effects. Adding co-skewness and co-kurtosis also improves the explanatory power of the Carhart (1997) four-factor model that includes market, size, BV/MV, and the momentum factors. This is despite the fact that we believe our sample is bias against us nding higher co-moment effects, as we focus on only analyzing the larger capitalized rms in the US and Australia. If our study included smaller cap stocks we would expect even further evidence of the presence of higher moments. We therefore believe that our ndings do support the need to incorporate higher co-moments into asset pricing models and developing a theoretical asset pricing model that addresses the non-normality of the return distribution should remain important for future research. References
Alles, L., Spowart, A., 1995. Higher moments of Australian equity returns: characteristics and determinants. Accounting Research Journal 8, 6676. Ando, M., Hodoshima, J., 2006. The robustness of asset pricing models: coskewness and cokurtosis. Finance Research Letters 3, 133146. Arditti, F.D., 1967. Risk and required return on equity. Journal of Finance 22, 1936. Bansal, R., Viswanathan, S., 1993. No-arbitrage and arbitrage pricing: a new approach. Journal of Finance 48, 12311262. Beedles, W., 1986. Asymmetry in Australian equity returns. Australian Journal of Management 11, 111. Bird, R., Gallagher, D., 2002. The evaluation of active manager returns in a non-symmetrical environment. Journal of Asset Management 2, 303324. Brands, S., Gallagher, D., 2004. Portfolio selection, diversication, and funds-of-funds: a note. Accounting and Finance 45, 185197. Campell, J.Y., Hentschel, L., 1992. No new is good news: an asymmetric model of changing volatility in stock returns. Journal of Financial Economics 31, 281318. Campbell, J., Lo, A., MacKinlay, C., 1995. The Econometric of Financial Markets. Princeton University Press, Princeont, NJ. Carhart, M., 1997. On persistence in mutual fund performance. Journal of Finance 52, 5782. Chapman, D., 1997. Approximating the asset pricing kernel. Journal of Finance 52, 13831410. Chung, Y.P., Johnson, H., Schill, M., 2006. Asset pricing when returns are non-normal: Fama-French factors vs higher-order systematic co-moments. Journal of Business 79, 923940. Dimson, E., 1979. Risk measurement when shares are subject to infrequent trading. Journal of Financial Economics 7, 197226. Dittmar, R.F., 2002. Nonlinear pricing kernels, kurtosis preference, and evidence from the cross section of equity returns. Journal of Finance 57, 369403. Fama, E., 1965. The behaviour of stock market prices. Journal of Business 38, 34105. Fama, E., French, K., 1993. Common risk factors in the returns on stocks and bonds. Journal of Financial Economics 33, 356. Fama, E., MacBeth, J.D., 1973. Risk return and equilibrium: empirical test. Journal of Political Economy 81, 607636. Fang, H, Lai, T.Y., 1997. Co-kurtosis and capital asset pricing. The Finance Review 32, 293307. Friend, I., Westereld, R., 1980. Co-skewness and capital asset pricing. Journal of Finance 35, 10851100. Galagedera, D.U.A., Henry, D., Silvapulle, P., 2002. Conditional relation between higher-order co-moments and stock returns: evidence from Australian data. In: Econometric Society Australasian Meeting, Brisbane, Australia. Harvey, C., Siddique, A., 1999. Autoregressive conditional skewness. Journal of Financial and Quantitative Analysis 34, 465488. Harvey, C., Siddique, A., 2000. Conditional skewness in asset pricing tests. The Journal of Finance 55, 12631295. Jegadeesh, N., Titman, S., 1993. Returns to buying winners and selling losers: implications of stock market inefciency. The Journal of Finance 58, 6591. Kan, R., Zhou, G., 2003. Modelling non-normality using multivariate t: implications for asset pricing. Working paper, Washington University, St. Louis. Kirchler, M., Huber, J., 2007. Fat tails and volatility clustering in experimental asset markets. Journal of Economic Dynamics & Control 31, 18441874. Kraus, A., Litzenberger, R.H., 1976. Skewness preference and the valuation of risk assets. The Journal of Finance 38 (4), 10851100. Lim, K.G., 1989. A new test of the three-moment CAPM. Journal of Financial and Quantitative Analysis 24, 205216. Mandelbrot, B.B., 1963. The variation of certain speculative prices. Journal of Business 40, 393413. Mandelbrot, B.B., Taylor, H., 1967. On the distribution of stock price differences. Operations Research 15, 10571062. Merton, Robert C., 1973. An intertemporal capital asset pricing model. Econometrica 41, 867887. Ross, S.A., 1976. The arbitrage theory of capital asset pricing. Journal of Economic Theory 13, 341360. Scholes, M., Williams, J., 1977. Estimating betas from nonsynchronous data. Journal of Financial Economics 5, 309328. Scott, R.C., Horvath, A., 1980. On the direction of preference for moments of higher order than the variance. The Journal of Finance (4), 915919. Sharpe, W.F., 1964. Capital asset prices: a theory of market equilibrium under condition of risk. The Journal of Finance 19 (3), 425442. Smith, D., 2007. Conditional coskewness and capital asset pricing. The Journal of Empirical Finance 91, 91119.