I'm currently studying autocorrelation in Python and exploring the ArmaProcess module within the statsmodels.tsa.arima_process package. This module allows us to set the parameters ar and ma. When dealing with the ar parameter, if we want to consider the negative impact of adding a lag to the previous data, we assign a negative coefficient to it.
However, sometimes I've come across situations where people take a different approach regardless of the sign of the coefficient. In other words, if the coefficient is truly negative, they consider it positive or vice versa for certain lags.
Could someone please explain this phenomenon to me?
here is an example:
# Simulate AR(1) with phi=+0.6
ma = np.array([1])
ar = np.array([1, -0.6])
AR_object = ArmaProcess(ar, ma)
simulated_data_1 = AR_object.generate_sample(nsample=5000)