Date post: | 30-Dec-2015 |
Category: |
Documents |
Upload: | rhona-perkins |
View: | 44 times |
Download: | 1 times |
Daniel Chi-Hsiou Hung
Systematic Risks and Nonlinear Market Models in International Size and Momentum Strategies
Research questions asked:
Do higher order Capital Asset Pricing Models better describe asset returns than the standard CAPM?
Can higher co-moment risks (coskewness and cokurtosis) capture the stylized effects of momentum and size strategies in international stock markets?
Why and how are they relevant to finance?
The beta of the Capital Asset Pricing Model (CAPM) may not be sufficient to describe systematic risks
Risk management, where the estimation and
control of risk profiles of hedged positions of a company or an investment are critical elements of effective hedging
The estimation of the cost of capital of a company, which is directly related to corporate valuation and capital budgeting
Why and how are they relevant to finance? (continued)
Insight from higher co-moments can be beneficial to the formation of portfolio strategies, especially for hedge funds that typically have highly skewed return distributions
Performance of managed funds could be
evaluated by comparing average return of a managed portfolio with that of benchmark portfolios that have similar beta, coskewness and cokurtosis
Research questions are answered in many aspects:
Higher co-moments are priced in 20 international equity markets
When higher co-moments are included into the two-moment CAPM, model intercepts become insignificant in all cases for examining the two-way sorted, momentum-size portfolios
Research questions answered (continued):
Develop and test a cubic-market model, which shows better performance than the linear CAPM in explaining the stylised effects
Market models are found to predict payoffs from momentum strategies that buy the past return winners and sell the past return losers
When up and down markets are tested separately, the CAPM beta is highly significant in explaining the cross-section of international stock returns
The smallest size decile has positive average returns in both up- and down-markets
Research questions answered (continued):
Higher order systematic risks and evidence
The intuition for the preference of skewness
If a risky asset has a return distribution with a long tail in the negative direction, it is more likely to have more extreme negative returns.
Other things being equal, a rational investor will require a higher mean rate of return on assets that contribute negative skewness to the market
Kraus and Litzenberger (1976) Non-increasing absolute risk aversion leads
to a preference for positive skewness. (U.S. stock data)
Harvey and Siddique (2000) Assets that make the portfolio returns more
left-skewed are less desirable and should command higher expected returns. (U.S. stock data)
Higher order systematic risks and evidence
The intuition for the preference of kurtosis
Since kurtosis measures the probability of extreme outcomes, a rational investor will prefer short-tailed distributions to long-tailed distributions
Other things being equal, the effect of a risky asset contributing to market leptokurtosis will be to increase the required mean rate of return on the asset
Higher order systematic risks and evidence
Fang and Lai (1997) Propose an extended model that incorporates a
cokurtosis term which is significant in explaining the cross-section of U.S. stock returns, 1969 to 1988
Dittmar (2002) Decreasing absolute prudence leads to an
aversion for kurtosis
Higher order systematic risks and evidence
Christie-David and Chaudhry (2001) for U.S. commodity contracts
Hung, Shackleton and Xu (2004) for U.K. stock data
Higher order systematic risks and evidence
Methodology
Sorts and portfolio formation
Size/Momentum/Country sorts
Equally-weighted deciles are formed for examining the characteristics of portfolios
36 two-way sorted, size-momentum portfolios
Returns of the two extreme deciles of size and momentum sorts are examined in the time series
Data and descriptive Statistics
Monthly U.S. dollar returns, from August 1988 to November 2003 (Datastream)
44,290 stocks from 20 markets: Canada, U.S., Belgium, Denmark, Finland, France, Germany, Italy, Netherlands, Norway, Spain, Sweden, Switzerland, U.K., Australia, Hong Kong, Japan, Korea, Singapore and Taiwan
Market value of equity and the London Financial Times Euro dollar one-month rate (Datastream)
Panel A Value weighted countries (markets) returns, from September 1988 to November 2003
Country N. stocks Mean MV Mean Median Stdev, Max Min Skewness Kurtosis J. B. Test Total 44,250 1,964 0.0072 0.0083 0.0335 0.1005 -0.1027 -0.20 3.49 3.05* Canada 5,350 263 0.0054 0.0074 0.0446 0.2172 -0.1633 0.15 7.56 159*** U.S. 14,040 1,388 0.0096 0.0157 0.0415 0.1089 -0.1300 -0.54 3.63 12.1*** Belgium 699 6,848 0.0092 0.0121 0.0346 0.1158 -0.1168 -0.62 4.51 29.1*** Denmark 382 674 0.0088 0.0120 0.0446 0.1360 -0.1152 -0.37 3.40 5.38* Finland 284 1,137 0.0093 0.0115 0.0727 0.2242 -0.2342 -0.10 3.60 3.03* France 2,121 12,506 0.0110 0.0079 0.0384 0.2202 -0.0971 1.35 8.67 301*** Germany 4,433 2,336 0.0099 0.0066 0.0513 0.1599 -0.1758 0.04 3.79 4.78** Italy 493 1,903 0.0084 0.0096 0.0646 0.1994 -0.1860 0.15 3.45 2.22 Netherlands 521 3,193 0.0078 0.0128 0.0414 0.0866 -0.2802 -2.23 15.25 1296*** Norway 471 425 0.0080 0.0117 0.0530 0.1614 -0.2113 -0.29 3.93 9.26*** Spain 247 4,794 0.0079 0.0079 0.0538 0.1384 -0.1696 -0.38 3.86 10.02*** Sweden 995 1,155 0.0070 0.0134 0.0511 0.1438 -0.1809 -0.45 3.65 9.24*** Switzerland 737 4,395 0.0087 0.0101 0.0335 0.1484 -0.0757 0.43 4.72 28.4*** U.K. 5,159 1,756 0.0039 0.0033 0.0298 0.0822 -0.0782 -0.08 3.42 1.56 Australia 1,955 490 0.0078 0.0093 0.0461 0.1215 -0.1615 -0.37 3.62 7.1** Hong Kong 892 870 0.0121 0.0131 0.0696 0.2094 -0.3348 -0.47 6.10 79.8*** Japan 3,077 1,276 0.0004 -0.0084 0.0613 0.2631 -0.1534 0.70 4.41 30.3*** Korea 601 406 0.0107 0.0007 0.1090 0.4028 -0.3010 0.57 4.39 24.6*** Singapore 767 446 0.0067 0.0072 0.0604 0.2099 -0.1936 0.09 5.06 32.6*** Taiwan 1,066 383 0.0047 -0.0009 0.1121 0.3384 -0.3886 -0.10 3.88 6.25** Euo-$ 1M - - 0.0041 0.0041 0.0017 0.0081 0.0009 -0.001 2.63 1.04
Panel B. Equally weighted size portfolios, from October 1988 to November 2003
N. stocks MV ($ M) Past
return Mean Stdev Skewness Kurtosis J. B. Test Beta Gamma Delta
Small 2551.4 1 -0.0106 0.0310 0.0425 0.65 4.74 35.9*** 0.42 2.24 0.59 2 2551.3 5 0.0300 0.0183 0.0400 0.59 5.81 71*** 0.53 2.01 0.65 3 2551.3 14 0.0469 0.0117 0.0345 0.07 4.19 10.9*** 0.55 1.53 0.67
4 2551.1 28 0.0377 0.0091 0.0325 -0.16 4.13 10.4*** 0.58 1.50 0.69 5 2550.9 51 0.0460 0.0078 0.0332 -0.23 4.14 11.5*** 0.65 1.37 0.75
6 2550.9 93 0.0505 0.0071 0.0341 -0.24 4.37 16.1*** 0.72 1.35 0.83 7 2550.9 172 0.0569 0.0068 0.0357 -0.24 4.32 15*** 0.80 1.29 0.90 8 2550.9 353 0.0635 0.0054 0.0372 -0.26 4.67 23.2*** 0.86 1.28 0.96 9 2550.8 920 0.0682 0.0058 0.0378 -0.36 4.58 23.1*** 0.90 1.30 0.98
Big 2550.8 14,614 0.0640 0.0067 0.0352 -0.44 4.32 19.2*** 0.90 1.25 0.93 S - B - -14,613 -0.0746 0.0242 0.0073 1.09 0.43 - -0.48 0.99 -0.34
Panel C. Equally weighted momentum portfolios, from February 1989 to November 2003
N. stocks MV ($ M) Past
return Mean Stdev Skewness Kurtosis J. B. Test Beta Gamma Delta
Loser 2469.8 365 -0.4511 0.0218 0.0795 0.61 5.90 75.3*** 1.03 2.58 1.24 2 2469.7 935 -0.2154 0.0074 0.0512 -0.06 5.29 40.2*** 0.82 1.90 1.01
3 2469.6 1,760 -0.1150 0.0055 0.0384 -0.36 6.09 76.7*** 0.69 1.58 0.86 4 2469.5 1,870 -0.0503 0.0040 0.0302 -0.34 5.24 41.7*** 0.56 1.19 0.71 5 2469.4 1,824 -0.0109 0.0059 0.0251 -0.21 5.68 56.2*** 0.47 0.87 0.53 6 2469.3 1,798 0.0238 0.0078 0.0214 -0.004 3.98 7.29** 0.39 0.73 0.41 7 2469.2 1,790 0.0697 0.0091 0.0238 0.03 3.18 0.27 0.48 0.46 0.46 8 2469.2 2,590 0.1381 0.0091 0.0317 -0.65 4.24 24.7*** 0.66 1.38 0.71
9 2469.1 2,311 0.2641 0.0127 0.0373 -0.75 5.01 47.9*** 0.80 1.86 0.90 Winner 2468.9 1,548 0.9621 0.0203 0.0545 -0.90 7.24 162*** 1.01 2.62 1.14
W - L - 1,183 1.4131 -0.0015 -0.0249 -1.51 1.35 - -0.02 0.04 -0.10
Panel D. Equally weighted momentum-size, two-way sorts, February 1989 to November 2003
Momentum Size
Number
Stocks
MV
($ M)
RP
(P. 6M)
Rp
(Mean) Stdev Skew Kurt J. B. Test Beta Gamma Delta
1 1016.1 2 -0.4356 0.0509 0.081 1.12 6.79 147*** 0.64 3.98 0.98
2 871.4 14 -0.3758 0.0101 0.067 0.65 5.88 76.1*** 0.83 2.49 1.04
3 711.9 44 -0.3487 0.0024 0.072 0.78 8.22 226*** 0.96 2.04 1.14
4 614.7 120 -0.3334 0.0009 0.077 0.61 6.46 102*** 1.16 1.76 1.35
5 527.5 382 -0.3144 0.0014 0.079 0.42 5.37 48.1*** 1.28 1.45 1.42
Lowest
Past
Return
6 374.5 5,097 -0.2903 0.0021 0.076 0.19 4.71 23.3*** 1.27 1.36 1.28
1 580.6 2 -0.1191 0.0193 0.043 0.32 4.78 27.2*** 0.43 1.99 0.68
2 695.0 15 -0.1190 0.0068 0.036 -0.12 5.06 32.7*** 0.52 1.69 0.71
3 711.0 45 -0.1173 0.0042 0.037 -0.39 5.79 63.9*** 0.61 1.69 0.79
4 707.9 121 -0.1179 0.0026 0.041 -0.20 6.60 100*** 0.74 1.54 0.92
5 705.4 393 -0.1169 0.0022 0.044 -0.22 5.74 58.8*** 0.86 1.29 0.99
2
6 716.0 8,671 -0.1147 0.0029 0.044 -0.43 4.68 27.1*** 0.92 1.35 1.00
1 786.1 2 -0.0199 0.0088 0.026 0.03 3.72 3.93* 0.34 1.17 0.42
2 684.2 15 -0.0232 0.0051 0.025 -0.14 3.95 7.48** 0.38 1.07 0.49
3 668.1 44 -0.0238 0.0044 0.025 0.19 6.26 82.3*** 0.42 0.90 0.54
4 649.1 121 -0.0239 0.0037 0.026 0.13 6.42 89.5*** 0.52 0.96 0.62
5 657.5 391 -0.0241 0.0035 0.030 0.04 6.35 85.4*** 0.64 0.88 0.75
3
6 670.8 10,733 -0.0248 0.0042 0.031 -0.02 4.88 26.9*** 0.71 0.88 0.77
1 795.1 2 0.0347 0.0149 0.025 0.80 3.96 26.5*** 0.24 0.87 0.25
2 690.6 15 0.0373 0.0088 0.021 0.10 3.87 6.11** 0.28 0.57 0.29
3 685.0 44 0.0387 0.0078 0.020 0.08 3.43 1.63 0.34 0.48 0.33
4 669.5 121 0.0394 0.0068 0.021 0.06 4.05 8.47** 0.42 0.53 0.41
5 630.8 390 0.0396 0.0066 0.025 0.14 3.82 5.66** 0.54 0.56 0.51
4
6 644.5 10,752 0.0405 0.0065 0.027 0.08 3.87 6.02** 0.64 0.47 0.56
1 441.3 2 0.1364 0.0214 0.047 1.61 9.39 391*** 0.42 1.45 0.50
2 555.1 15 0.1405 0.0145 0.035 0.12 6.16 76.7*** 0.53 1.54 0.60
3 656.2 45 0.1409 0.0095 0.030 -0.54 4.57 27.7*** 0.58 1.35 0.65
4 730.3 122 0.1416 0.0080 0.030 -0.56 4.08 18.4*** 0.64 1.25 0.71
5 802.2 397 0.1426 0.0058 0.032 -0.55 3.92 15.8*** 0.73 1.32 0.79
5
6 930.3 10,201 0.1449 0.0061 0.035 -0.51 3.74 12.0*** 0.86 1.30 0.90
1 497.2 2 0.9796 0.0363 0.070 0.58 4.59 29.4*** 0.66 3.57 0.86
2 619.3 15 0.8460 0.0216 0.055 -0.09 7.55 158*** 0.73 2.69 0.92
3 683.5 45 0.6964 0.0168 0.048 -0.92 8.15 228*** 0.85 2.18 0.98
4 744.1 122 0.6354 0.0150 0.049 -1.02 7.70 200*** 1.00 2.17 1.12
5 792.1 391 0.6064 0.0113 0.051 -0.64 6.54 108*** 1.11 2.01 1.21
Highest
Past
Return
6 779.2 7,784 0.5083 0.0097 0.049 -0.44 6.18 82.9*** 1.07 1.84 1.14
Cross-sectional tests of co-moment pricing
100 size portfolios
100 momentum portfolios
36 two-way sorted, size-momentum portfolios
Portfolio beta, gamma (Coskewness) and delta (Cokurtosis) estimation
1
60
21
60
/t
t
mtm
t
t
mtmptppt rrrrrr
1
60
31
60
2/
t
t
mtm
t
t
mtmptppt rrrrrr
1
60
41
60
3/
t
t
mtm
t
t
mtmptppt rrrrrr
(2)
Cross-sectional tests of co-moment pricing
The estimates of βpt, γpt and δpt are used in cross-sectional regressions to estimate premia ηβ, η γ and ηδ associated with covariance, coskewness and cokurtosis
ptpttpttptttptr 0 (4)
Cross-sectional tests of co-moment pricing
ptpttpttptttptr 0
Panel A. 100 Size sorted portfolios
Model η0 ηβ ηγ ηδ Adj. R2
1 0.0312 -0.0343 0.2325 5.76 -5.91 2 0.0112 -0.0297 0.0100 0.3028 3.18 -5.25 5.77 3 0.0167 -0.0739 0.0500 0.2911 5.04 -6.55 5.63 4 0.0104 -0.0499 0.0080 0.0228 0.3377 3.40 -4.85 4.45 2.54
ptpttpttptttptr 0
Panel B. 100 Momentum sorted portfolios
Model η0 ηβ ηγ ηδ Adj. R2
1 -0.0072 0.0183 0.2214 -2.88 2.72 2 -0.0085 0.0300 -0.0040 0.2818 -2.96 2.92 -1.67 3 -0.0067 0.0406 -0.0206 0.2708 -2.71 3.10 -1.96 4 -0.0071 0.0343 -0.0048 -0.0071 0.3430 -2.56 2.85 -1.26 -0.51
ptpttpttptttptr 0
Panel C. 36 Momentum and size two-way sorted portfolios
Model η0 ηβ ηγ ηδ Adj. R2
1 0.0085 -0.0054 0.2028 3.26 -1.02 2 0.0015 -0.0185 0.0105 0.3386 0.61 -3.09 4.88 3 0.0031 -0.0515 0.0446 0.3269 1.12 -4.10 3.54 4 0.0007 -0.0363 0.0083 0.0212 0.4134 0.26 -2.75 3.01 1.41
Pettengill et al.(1995): significance tests of beta premia should be separately conducted according to up- or down-market status
According to the CAPM, the ex ante market premium should be positive and that higher beta portfolios should have higher expected returns than lower beta portfolios
Tests of ex-post beta and return relationships
Ex post, the market premium can be negative in some periods. Thus, averaging all cross-sectional periods might result in an insignificant market premium
Even though the market premium is significant according to Sharp ratio of the market (Eq. 5)
Tests of ex-post beta and return relationships
m
fmppmfp
RRRR
(5)
And also those portfolios with higher betas can have more negative (lower) realised returns than that of lower beta portfolios. Thus reduce the average realised returns of higher beta portfolios.
Consequently, the coefficient for beta might appear insignificantly
Tests of ex-post beta and return relationships
Mean Monthly Returns of Size Sorted Portfolios
-0.03
-0.02
-0.01
0.00
0.01
0.02
0.03
0.04
0.05
1 2 3 4 5 6 7 8 9 10
Portfolios
UpDownMean
Mean Monthly Returns of Momentum Sorted Portfolios
-0.03-0.02
-0.010.00
0.010.02
0.030.04
0.050.06
1 2 3 4 5 6 7 8 9 10
Portfolios
UpDownMean
ptpttpttptttptr 0
36 Momentum and size two-way sorted portfolios Model Status η0 ηβ ηγ ηδ Adj. R2
Up 0.0068 0.0205 0.1935 1 2.73 4.34 Down 0.0109 -0.0427 0.2163 3.95 -10.31 Up -0.0002 0.0113 0.0092 0.3150 2 -0.09 2.24 6.29 Down 0.0040 -0.0614 0.0123 0.3725 1.73 -12.31 4.28 Up 0.0008 -0.0175 0.0393 0.3181 3 0.28 -1.40 2.96 Down 0.0063 -0.1003 0.0523 0.3395 2.52 -8.98 4.47 Up -0.0030 -0.0056 0.0072 0.0223 0.4026 4 -1.02 -0.38 3.48 1.32 Down 0.0061 -0.0805 0.0099 0.0195 0.4290 2.49 -8.70 2.78 1.67
Time-series tests of nonlinear market models
A cubic model (Eq. 6), which is consistent with the four-moment extension of the CAPM is applied to explain the time-series returns of size, momentum and country sorted portfolios
tmtmtpmtmtpmtpppt rrCrrCrCCr 3
3
2
210(6)
tmtmtpmtmtpmtpppt rrCrrCrCCr 3
3
2
210
Panel A. Explaining returns of Small and Big portfolios
Model C0 C1 C2
C3 Adj. R2
1 Small 0.0253 0.4605 0.1251 8.48 5.18 Big -0.0004 0.9643 0.8444 -0.42 31.36 2 Small 0.0315 0.4210 -5.4257 0.1706 9.10 4.82 -3.30 Big 0.0013 0.9531 -1.5423 0.8496 1.10 31.22 -2.68 3 Small 0.0260 0.3116 37.8746 0.1301 8.62 2.28 1.43 Big -0.0001 0.8940 17.8854 0.8468 -0.08 18.96 1.96 4 Small 0.0315 0.3605 -5.1523 15.8891 0.1676 9.08 2.67 -3.01 0.59 Big 0.0013 0.9066 -1.3325 12.1996 0.8501 1.11 19.30 -2.23 1.30
Panel B. Explaining returns of Winner and Loser portfolios
Model C0 C1 C2
C3 Adj. R2
1 Winner 0.0135 1.0164 0.3888 4.18 10.63 Loser 0.0147 1.1193 0.2198 2.78 7.11 2 Winner 0.0199 0.9710 -5.6381 0.4192 5.33 10.29 -3.19 Loser 0.0211 1.0748 -5.5192 0.2306 3.36 6.80 -1.86 3 Winner 0.0140 0.9116 26.4639 0.3883 4.28 6.17 0.93 Loser 0.0159 0.8847 59.2217 0.2225 2.96 3.65 1.27 4 Winner 0.0199 0.9593 -5.5862 3.0399 0.4159 5.32 6.61 -3.04 0.11 Loser 0.0210 0.9262 -4.8553 38.8624 0.2291 3.36 3.81 -1.58 0.81
tmtmtpmtmtpmtpppt rrCrrCrCCr 3
3
2
210
Model C0 C1 C2
C3 Adj. R2
Canada 1 -0.0011 0.7767 0.3376 -0.42 9.68 2 0.0025 0.7533 -3.2071 0.3502 0.80 9.39 -2.12 3 -0.0012 0.7892 -3.1790 0.3339 -0.44 6.35 -0.13 4 0.0025 0.8226 -3.5200 -18.2135 0.3485 0.80 6.64 -2.24 -0.74 US 1 0.0027 0.8954 0.5248 1.26 14.21 2 0.0057 0.8760 -2.6635 0.5352 2.29 13.93 -2.25 3 0.0025 0.9211 -6.5517 0.5225 1.17 9.44 -0.35 4 0.0057 0.9496 -2.9959 -19.3477 0.5352 2.29 9.79 -2.44 -1.00
tmtmtpmtmtpmtpppt rrCrrCrCCr 3
3
2
210
Belgium 1 0.0022 0.8916 0.7496 1.74 23.36 2 0.0043 0.8786 -1.7884 0.7567 2.84 23.13 -2.50 3 0.0026 0.8260 16.6999 0.7512 1.97 14.05 1.47 4 0.0043 0.8414 -1.6204 9.7792 0.7562 2.84 14.36 -2.18 0.83 Denmark 1 0.0020 0.8681 0.4266 0.79 11.68 2 0.0051 0.8480 -2.7677 0.4357 1.74 11.39 -1.98 3 0.0022 0.8163 13.1929 0.4245 0.88 7.10 0.59 4 0.0051 0.8423 -2.7423 1.4803 0.4325 1.74 7.32 -1.88 0.06 Netherlands 1 0.0010 0.8498 0.4738 0.45 12.84 2 0.0029 0.8374 -1.7005 0.4762 1.11 12.56 -1.36 3 0.0002 1.0064 -39.8340 0.4827 0.09 9.93 -2.03 4 0.0029 1.0308 -2.5738 -50.8269 0.4912 1.13 10.18 -2.00 -2.51
tmtmtpmtmtpmtpppt rrCrrCrCCr 3
3
2
210
Australia 1 0.0011 0.8179 0.3470 0.38 9.89 2 0.0054 0.7904 -3.7671 0.3642 1.65 9.59 -2.43 3 0.0001 1.0102 -48.9123 0.3575 0.03 7.97 -1.99
4 0.0054 1.0574 -4.9723 -70.1497 0.3878 1.67 8.48 -3.14 -2.81
Taiwan 1 -0.0024 0.9462 0.0752 -0.30 3.97
2 0.0134 0.8455 -13.8137 0.1179 1.44 3.60 -3.13 3 -0.0001 0.4808 118.3849 0.0842 -0.01 1.31 1.67 4 0.0134 0.6016 -12.7127 64.0875 0.1169 1.44 1.66 -2.77 0.89
tmtmtpmtmtpmtpppt rrCrrCrCCr 3
3
2
210
Model predicted returns of momentum and size deciles
For each portfolio, the intercept and slope coefficients (C0, C1, C2 and C3) of time-series regression model (Eq. 6) are estimated in each month on a rolling basis from the month of portfolio formation to the 5th month following formation for momentum sorts (the 11th month for size sorts)
tmtmtpmtmtpmtpppt rrCrrCrCCr 3
3
2
210 (6)
In the second stage, the estimates of C0, C1, C2 and C3 for each asset are used to predict excess return of the asset in the next period by utilizing realized excess market return in the next period
31,1,3
21,1,21,101,
^
tmtmptmtmptmpptp rrCrrCrCCr (10)
Model predicted returns of momentum and size deciles
Linear Market Model Predicted Momentum Portfolio Return
-0.01
0.00
0.01
0.02
0.03
0.04
1 2 3 4 5 6
Month following portfolio formation
Pre
dict
ed
Re
turn
Winner
Loser
Linear model predicted returns of momentum deciles
Linear Market Model Predicted Size Portfolio Return
-0.01
0
0.01
0.02
1 2 3 4 5 6 7 8 9 10 11 12
Month following Portfolio Formation
Pre
dict
ed R
etur
n
Small
Big
Deflated Excess Returns of Momentum Deciles
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
0.0 0.5 1.0 1.5 2.0 2.5Linear Model P redicted Mean Return, % per month
Realis
ed M
ean R
etu
rn, %
Panel A. Predicted versus Realised Returns of the Equally Weighted Size Deciles
Deciles Small D2 D3 D4 D5 D6 D7 D8 D9 Big Mean
Square Error
Realised 0.0211 0.0129 0.0078 0.0049 0.0030 0.0024 0.0009 0.0000 0.0002 0.0016
Linear Model Predicted
0.0109 0.0075 0.0051 0.0042 0.0048 0.0052 0.0058 0.0066 0.0077 0.0083 3.22×10-5
Quadratic Model Predicted
0.0104 0.0074 0.0054 0.0047 0.0052 0.0060 0.0064 0.0070 0.0084 0.0086 3.65×10-5
Cubic Model Predicted
0.0102 0.0073 0.0056 0.0049 0.0054 0.0064 0.0069 0.0074 0.0091 0.0094 4.1×10-5
Panel B. Predicted versus Realised Returns of the Equally Weighted Momentum Deciles
Deciles Small D2 D3 D4 D5 D6 D7 D8 D9 Big Mean
Square Error
Realised 0.0166 0.0036 0.0001 -0.0002 -0.0001 0.0026 0.0040 0.0058 0.0086 0.0147
Linear Model Predicted
0.0064 0.0048 0.0033 0.0016 0.0001 0.0009 0.0033 0.0059 0.0104 0.0235 2.03×10-5
Quadratic Model Predicted
0.0077 0.0057 0.0038 0.0021 0.0002 0.0011 0.0036 0.0063 0.0112 0.0255 2.29×10-5
Cubic Model Predicted
0.0075 0.0065 0.0047 0.0026 0.0001 0.0014 0.0034 0.0064 0.0113 0.0263 2.64×10-5
Summary and conclusion
By using a large international stock data, this paper shows evidence for the pricing of higher order systematic risks in returns of size and momentum portfolios
The inclusion of coskewness and cokurtosis to the standard CAPM can provide incremental explanatory power on stock returns of size and momentum sorts
Summary and conclusion (continued)
This paper also develops and tests a cubic market model that is consistent with the four-moment CAPM
In time-series tests, the benefit of adopting non-linear market models is evidenced for both size and momentum sorts and also for eight international markets