With some timeseries data, I have identified certain events that I hypothesise will predict direction changes in the data. If I see event A I expect a rise in the signal, but if I see event B I expect a decline. I assume that those changes should occur within a window of a certain size following the event.

I'm a bit stuck as to how to quantify this as I'm not very familiar with working with timeseries data. I first thought that a peak-to-peak within the window would suffice, but this doesn't actually capture the direction change, only the distance between the most extreme points in the window. Then I thought that I could fit a line to the data in the window and analyse the slope, but as I understand it this is inappropriate for timeseries data as we have autocorrelation and I am also making the assumption of nonlinearity. Is this correct? Or would a polynomial expansion be the best solution?