A multivariate extension of the univariate nonlinearity test of Tsay (1986)
is presented. Simulation results show that the multivariate test is more p
owerful than its univariate counterpart, especially for series having nonli
near structure involving several components of the vector process and weakl
y or moderately cross-correlated process error terms. For illustration, the
test is applied to a set of seasonally adjusted quarterly capital expendit
ures and appropriations in U.S. manufacturing and a vector nonlinear model
for the data is constructed.