I have a dataset of accident cases starting from January 2001 to September 2024 on a monthly basis. I plan to do a SARIMAX model with X as a binary variable (1: festive month, 0: non-festive month). Since my dataset spans the years 2001 through 2024, it includes months with lockdowns, which indicates a sharp drop in the number of accident cases. Then, I decide to do a single structural break during the month that experiences lockdown. But after adding a structural break, the model exhibits autocorrelation, which means my residual independent assumption is not met. Is there a reason why this occurs, and how can I ensure that my assumption is met?