They use them as part of their modelling technique. The closer to the present day the outlier is the greater impact it will have (this can be adjusted for in the hyperparameters, and will impact the variance/confidence interval you can find out more on the data science learning path