Be careful with what you're seeing there. The weighted-fitting approach is actually forecasting the past. What you need is the correlation between a clearly defined span of out-of-sample future compared to the last weight adjustment using a certain amount of latest historical data.