to limit interest rate paths to a 'reasonable' range (if we could define reasonable). Now we calibrate log-normal skew and mean reversion monthly to robust basket of atm swaptions and in and out caps. Sometimes resulting in at least five percent of the paths exceeding 40% (1 month forward rate) and reaching over 250%. Most / many months - the highest paths never exceed 50%. Also - any recommendations on how to analyze how this affects resulting metrics such as OAS or effective duration for mortgage-related assets?
If you start removing some of the paths, then the expectation of the bond prices that you calculate will not fit the market prices. Since these paths are on the extremes, they should not be affecting your OAS or effective duration much (the probability on these paths are low, and all these statistics depend on the expectation).