An ARMAX model adds input variables or covariate on the right-hand side of the ARMA time series model (assuming the series is stationary so we can skip differencing):
This resembles a linear regression model but is quite difficult to interpret because the effect of β on yt is not the effect of an increase in xt by one unit as in linear regression. Instead, the presence of lagged values of yt on the right-hand side of the equation implies that the coefficient can only be interpreted given the lagged values of the response variable, which is hardly intuitive.