In this paper, we examine semiparametric nonlinear autoregressive models wi
th exogenous variables (NLARX) via three classes of artificial neural netwo
rks: the first one uses smooth sigmoid activation functions; the second one
uses radial basis activation functions; and the third one uses ridgelet ac
tivation functions, We provide root mean squared error convergence rates fo
r these ANN estimators of the conditional mean and median functions with st
ationary beta -mixing data. As an empirical application, we compare the for
ecasting performance of linear and semiparametric NLARX models of U.S. infl
ation. We find that all of our semiparametric models outperform a benchmark
linear model based on various forecast performance measures, In addition,
a semiparametric ridgelet NLARX model which includes various lags of histor
ical inflation and the GDP gap is best in terms of both forecast mean squar
ed error and forecast mean absolute deviation error.