Is my response here relevant? The common approach in time series literature is to just include everything up to the highest justifiable order, then strongly bias the coefficients towards zero.
I don’t know the frequency of your data, but at 17 lags you’re probably looking at seasonal effects.