Additional variable makes AR term insignificant

[HTTrader]

Hi,

I'm building a model for forecasting that is looking good. I was going to introduce one extra term, a term that accounts for customers receiving deposits throughout the year which can contributes to outflows, the dependent variable in my model. It is zero for most observations, but can be significant, being particularly large for the last few observations in my sample.

Now I thought this would be a perfectly reasonable variable to add. However, adding it has the effect of completely wiping out the AR(1) term in my model. It now has a p-value of 0.97. Looking at the correlogram, however, it is clear that autocorrelation has returned, so it is not clear why it becomes insignificant.

Can anyone tell me why this is happening and how I can fix it?