approaching the question from a foreign or international investor’s point of view.
The regression estimation is by least squares. We employ a within-group fixed effects
estimator with first-order autoregressive disturbances in order to adjust for persistence and
auto-correlation in the error term as well as heteroskedasticity robust standard errors. We
report panel estimations including country fixed effects and both country and time fixed
effects. Time fixed effects have the property of controlling for common global shocks.
Consequently, the domestic GDP growth rate reflects solely the idiosyncratic part of
domestic growth, whereas in the country fixed effects estimation also global factors could
drive the results. The panel regression specification is:
EQKGit = αi + δt + βgit + eit (1)
eit = ρeit-1 + zit
where EQKG is the annual real rate of capital gains on the respective domestic stock
market and g is the real annual rate of domestic GDP growth.
The regression analysis shows the following (Table 1): Both in terms of domestic cur-
rency and in US dollars we find pro-cyclicality of capital gains (significant at the 1% level).
This implies that in our sample a one percentage point increase in the domestic GDP
growth rate co-moves with a 1.3 percentage points increase in the rate of capital gains.
However, the result changes significantly when time fixed effects are included: Insignif-
icant beta coefficients suggests that global factors explained most of the pro-cyclicality
observed before.
In terms of international risk-sharing, this has crucial implications, as we are interested
in isolating the idiosyncratic component of GDP growth. Our results hence imply that
there is only limited evidence for a significant contemporaneous risk sharing mechanism
via domestic stock market capital gains for the period of 1973 to 2006. This means that
in the short-run of one year the specific state of a national economy does not seem to be
reflected in the idiosyncratic part of stock market returns.
In the individual country specification (2), we estimate similarly by general least
squares with a correction for first-order serial correlation in the error term. Moreover,
heteroskedasticity robust standard errors are employed.
EQKGit = αi + βigit + eit (2)
eit = ρeit-1 + zit