I am working with sales time-series data, I have a history of 9 years of monthly data. I am trying to forecast sales for the next 12 months. I am using XGboost regression to build multivariate time series, forecasting model. The problem I am facing is, I can see a strong correlation, Pearson correlation, between my stationery sales and stationary temperature for some month and some years e.g. the strong correlation between my stationary sales which is my target and temperature can be observed between 2004 and 2007 in May, June, and July. I have decomposed the data, removed the trend and seasonality and kept the level only. However XGboost model is not showing this kind of strong correlation between the two. I am feeding original time series of 9 years, non-stationary but providing some trend information but still can't see the good forecast for the 12 months in general and for those months in specific. Could anyone tell me why strong individual strong correlations are not caught by the model?