Note that this process assumes the price return is characterized as a submartingale
process (Ross, 1989) and X denotes the vector of macroeconomic explanatory variables
at time t. Since we are interested in volatility linkages, the conditional standard deviation
of returns can be written as:
Et (σt∖It-ι) = f ( Et (σtX )) (2)
Macroeconomic factors are observed monthly, therefore, we rely on the methodology
developed by Davidian and Carroll (1987) when estimating the conditional standard
deviations. It is noteworthy that this approach is used in prior work by Sadorsky (2003),
Kearney (2000) and Kautolas and Kryzanowski (1996) among others. Allow σtx to
denote the unconditional standard deviation of the vector of macroeconomic variables
and htx to denote the conditional standard deviations of these variables. The conditional
standard deviations are then estimated as htx = σtx - u2x,t utilizing the equation below,
where
12
σtx=β1(L)σtx+∑βs,jDUMj,t+utx,t (3)
j
and β1(L) is a 12th order polynomial in the lag operator, while DUM is a monthly
seasonal dummy variable. The series σx =∖ u1x,t ∖ is calculated as the residuals from the
following regression
12
u1x,t =∆Xt-Et(∆Xt ∖It-1)=∆Xt-λ1(L)∆Xt-∑λs,jDUMj,t (4)
j=1
This approach is based on the notion that standard deviations based on the absolute value
of the prediction errors are more robust than measures based on the squared residuals
alone (Davidian and Carroll, 1987). Moreover, as argued by Sadorsky (2003), the