This is a financial data cleaning question. I have raw price and yield data for US cash treasury across the curve. In the time-series there are jumps on the day after the treasury auction results come out. Prior to using the data, is it good practice to manually remove the jumps? Thanks.
No, do nothing. It is important to have "as-of" data for backtesting.
Answer below is when I misread the question and thought you were trying to smooth out jumps in the rolling on-the-run series. I thought I'd keep it for reference purpose:
It depends on what you're doing. If this is purely to show where the rolling on-the-run bond yields have been trading, then no, do nothing. Practitioners are well aware of these "rolls." (Make sure your computational convention is correct – WIs should be computed using the US Treasury method for forward settlement on the issue date.)
If the time series is used for relative value trading, then these data need to be processed. The most common technique is to build a spline that generates constant maturity par yields.
You need to have a process to fix the auction-day prices. Often the coupon is changed and you will have a previous price for the old coupon, which causes the jump. It happens fairly frequently; at my place we finally automated it.