-14-
monetary aggregates, like M2 or M3; short-term interest rates which include 3M interest
rates; and share prices which include the MSCI share price index in domestic currency.
Estimates for the global factors are obtained as the first principal component for each
sub-set (category) of series. Each segment of X t is therefore explained by exactly one
factor. For example, global monetary liquidity is estimated as the first principal
component from the set of monetary aggregates in the G-7 countries plus the euro zone.
|
^ Xt 1 ^ |
^Λι |
0. |
. 0 ■ |
~ F 1 |
Г et11 | |||
|
(4) |
Xt 2 |
— ^ |
0 ... |
Л 2 . ... . |
. 0 .. ... |
Ft2 |
+ |
2 |
|
. ×', . |
_ 0 |
0. |
. Л I |
... L Ft \ |
... L et J |
In the second step, a VAR with the estimated (Ft1,Ft2,...,FtI) is implemented. The
innovations in the VAR model can be identified by applying standard procedures, like a
Cholesky decomposition. However, a simple recursive ordering may not be appropriate in
an international context. Given its low level of flexibility, one has inevitably to make
extreme assumptions about the interaction between global variables. For example, if one
puts global money before the global short-term interest rate, the interest elasticity of
global money demand is constrained to be zero. In contrast, if the global interest rate is
predetermined for global money, money supply’s interest elasticity is assumed to be zero
(Leeper and Roush, 2003). As a result, the derived monetary policy shocks might be
contaminated. Non-recursive schemes allow for more general contemporaneous
interactions among variables than recursive orderings. A global structural FAVAR
(SFAVAR) postulates more reasonable economic structures and reflects better the
complexities of international policy-making.
We see two major advantages by implementing a FAVAR in our context. First, it
is possible to derive structural shocks on a worldwide level and their impact on other
global variables. For example, global money demand shocks can be disentangled from
global money supply shocks. Second, it is possible to derive the effects of each global
shock on specific national variables (cf. equation 3). Hence, impulse response functions
can be constructed for any variable included in the informational data set (Xt).