I have a monthly time series that stretches about 18 years. I examine the over-the-month differences in a time series and observe that 3 or 4 OTM values are very extreme. I can identify these values by flagging values that are over 2.5 standard deviations from the mean. I then replace these values with the average of the OTM values that are remaining. Lastly, I apply the adjusted OTM values backwards from my last observed value, giving me an adjusted time series without the outlier values.
Are there any good reasons why this method should not be used?
I find that since I am removing these extreme values, the variance of my models are much lower and my confidence intervals are smaller. So I believe I am better identifying the overall trend of the time series.