I have run into a small road block and would appreciate some direction from more seasoned subject experts. Here is how the "road block" comes into play:1. Several swaptions and a yield curve are used to calibrate Hull-White 1 factor affine model, i.e. produce mean reversion and short rate vol (both are time independent)2. The fit is usually good, as the level of mispricing of the swaptions is generally low enough relative to Black prices/implied vols3. Mean reversion and short rate vol obtained via model calibration in step 1. are then used to reprice individual calibrating swaptions via analytical (Jamshidian) solution4. At this point my expectation is that: level of mispricing observed in 2. should be comparable to that produced in 3. This is not the case, as the mispricing between analytical and Black's pricing of swaptions is significantly higher.I am not exactly sure whether it's a simple error on my part (although results are benchmarked across a few different implementations) or there is a theoretical issue that I am overlooking in my comparison.

Jamaicaman, I think that normally you would use the Jamshidian method to analytically price swaptions given 1-factor Hull-White dynamics, and that you use this prices to calibrate your Hull-White model to the market data. So there is no need then to reprice in step 3.Note that you can also use the method of Schrager and Pelsser to calibrate swaptions for affine models, this is much simpler and faster.

Church, thank you for suggesting Pelsser & Schrager paper. I am currently in the process of setting up calibration of the model via Jamshidian, and rely on the calibrated lattice as a convenient benchmark. My working assumption is that calibrating Hull-White via a trinomial tree and via Jamshidian would result in fairly similar mean reversion and short rate vol parameters (provided that calibrating instruments are the same). Put differently, given mean reversion and short rate vol, I'd expect that level of mispricing of calibrating instruments via either of the methods would be similar. This is not what I observe so far. Does my logic make sense to you though, or I am way off on assuming that calibrated parameters will be quite similar?