8 factors in the bkt model given by the seven fh risk

=8 factors in the BKT model, given by the
seven FH risk factors and the return of our correlation swap (Model 2).
Using GLS estimators, we estimate a negative market price of correlation
risk with respect to models 1 and 2, both with and without Shanken’s asymptotic
correction. The point estimate for the correlation factor risk premium is highly
statistically significant, with t -statistics of − 4.14 and − 4.06, respectively,
in each model.
19
In contrast, the GLS point estimates for other well-known
18
The EIV problem has a number of potential consequences in two-step least squares procedures. First, if standard
errors do not include information that beta coefficients are measured with error, the implied t -statistics might
overstate the precision of the risk premium estimates. Second, different estimators might have substantially
different properties when the linear factor model is misspecified, either because of a missing factor or because of
a latent nonlinearity. Finally, least squares estimators of risk premia in the second step might be biased in finite
samples.
19
The t -statistics after Shanken’s correction are − 4.06 and − 3.92, respectively.
609
Downloaded from
by Queen Mary University of London user
on 02 March 2018

The Review of Financial Studies / v 27 n 2 2014
Table 5
The cross-section of hedge fund excess returns and correlation risk exposures
Panel A: Model 1 (Correlation risk and market risk)
With Shanken’s correction
GLS WLS GLS WLS
Intercept 0 . 10 0 . 16 0 . 10 0 . 16
t
-stat
(3 . 74) (2 . 31) (3 . 47) (2 . 17)
Correl risk
− 4 . 33 − 4 . 54 − 4 . 33 − 4 . 54
t -stat − (4 . 14) − (2 . 7) − (4 . 06) − (2 . 58)
Mkt risk − 0 . 02 0 . 57 − 0 . 02 0 . 57
t -stat − (0 . 07) (1 . 41) − (0 . 07) (1 . 38)
Panel B: Model 2 (Correlation risk factor and FH(2004) model)
Intercept 0 . 10 0 . 18 0 . 10 0 . 18
t -stat (3 . 74) (3 . 4) (3 . 32) (3 . 09)
Correl risk − 4 . 27 − 2 . 61 − 4 . 27 − 2 . 61
t -stat − (4 . 06) − (1 . 9) − (3 . 92) − (1 . 79)
Mkt risk 0 . 01 0 . 39 0 . 01 0 . 39
t
-stat
(0 . 04) (1 . 06) (0 . 04) (1 . 04)
SCMBC
0 . 58 0 . 47 0 . 58 0 . 47
t -stat (2 . 09) (1 . 33) (2 . 05) (1 . 26)
BD10RET − 0 . 04 − 0 . 36 − 0 . 04 − 0 . 36
t -stat − (0 . 19) − (1 . 33) − (0 . 19) − (1 . 24)
BAAmTSY − 0 . 03 − 0 . 08 − 0 . 03 − 0 . 08
t -stat − (0 . 15) − (0 . 36) − (0 . 14) − (0 . 34)
PTFSBD 3 . 18 2 . 06 3 . 18 2 . 06
t
-stat
(2 . 53) (1 . 19) (2 . 44) (1 . 12)
PTFSFX
2 . 89 4 . 79 2 . 89 4 . 79
t -stat (1 . 82) (2 . 11) (1 . 74) (1 . 98)
PTFSCOM 0 . 37 2 . 12 0 . 37 2 . 12
t -stat (0 . 32) (1 . 26) (0 . 31) (1 . 18)
This table reports estimates for the risk premia on the market index, the Fung and Hsieh (2004) factors and
the correlation risk factor (CR). Portfolios are formed based on rolling beta estimates. In Panel A, we report
results for the market and the correlation risk factor (Model 1). In Panel B, we report results for the BKT eight-
factor model (Model 2). The estimation methods are GLS and WLS versions of the (Fama-MacBeth) two-pass
regression methodology.
t
-statistics are in brackets.
t
-statistics in Columns 4 to 5 are calculated using standard
errors based on Shanken (1992) errors-in-variables (EIV) adjustment. The cross-sectional regressions are based
on 75 portfolios (based on a sort using the market betas). Each year from January 1999 to 2011 funds are sorted
into these portfolios based on their betas calculated using the previous 36 monthly returns. The sample period is