-27-
(last available data point)8. Moreover, there may have been regime switches in central
bank behavior. Apart from structural time variations (propagation), the variance of
exogenous shocks could have changed as well. All in all, it is rather unlikely that the data
generating process in 1984 is the same as in 2007. This in turn could undermine the
stability of our SFAVAR. We therefore examine three types of structural breaks: in the
factor dynamics (Φ), in the global shocks (Gηt ) and in the factor loadings (Λ).9
(9)
Ft =φ(L)Ft-1 + Gηt where φ(L)Ft-1
Φ1(L), t ≤ K
φ 2(L), t > K
and var(ηt ) =
< σt, t ≤ τ
σ 22, t > τ
Equation (9) allows the global SFAVAR coefficients and the global structural shocks to
break at potentially different dates (K and τ respectively).
(10)
∣Λ1 Ft,t ≤ υ
Xt = ΛF+ + D(L)Xt 1 + vt where Xt = <
t t ’ t-1 t t [Λ 2 Ft, t > υ
In equation (10), it is allowed for a break in the factor loadings at date υ .
We start by testing jointly for the stability of all the coefficients on the lags of a
given global variable using the Andrews-Quandt structural breakpoint test. In contrast to
the traditional Chow test, this structural change test does not assume any prior knowledge
about potential break dates. Instead, the Andrews-Quandt sup-F statistic is the maximum
of a sequence of traditional Chow tests for structural change each based on a different
potential breakpoint. As is common in the literature, we applied a heteroskedasticity-
robust version of the Andrews-Quandt test (Stock, Watson 2002). The range of the
sample is trimmed by 15% from each side. Of the seven tests performed, none rejects the
null hypothesis of stability at conventional confidence levels. Moreover, additional
Andrews-Quandt tests are implemented separately for three variables in each of the seven
SFAVAR equations: global monetary liquidity, global GDP and global house prices.
Given the previously obtained empirical evidence, we regard these three variables as
short-run driving forces of the global economy. Only in the global liquidity equation do
8 The world trade statistic and the readings for financial cross-border holdings stem from the IMF and Lane
and Milesi-Ferretti (2007), respectively. Both figures are our own calculations.
9 The dynamics D(L) on a national level and the idiosyncratic errors vt may undergo profound changes as
well. These types of time variation are neglected in our work.