Re: [Quantlib-users] Calibration short rate models

[philippe h. <pha...@ma...>]

Even without the stochastic element, the behavior of linear forward curve is well known to be extremely unstable to small moves in key rates used to bootstrap it. For instance, if you calculate the Key Rate dv01s of à bond using linear forwards, you will see a substantial amount of propagation into adjacent instruments. The total dc01 is still correct but the allocation across input instruments is a bit insane. This property is totally avoided by flat continuous forwards when bootstrapping the curve.
By extension; I assume that the bad propagation behavior of a linear forward curve strongly destabilizes a stochastic rate calibration model.

Regards

Philippe Hatstadt
+1-203-252-0408
https://www.linkedin.com/in/philippe-hatstadt


> On Sep 23, 2020, at 5:23 AM, oyvfos--- via QuantLib-users <qua...@li...> wrote:
> 
> 
> Hi, 
> It seems to be common practice to use a flat forward rate when calibrating the short rate models using QuantLib (example). What is the explanation for that? Is there any rationale for choosing a specific forward term? Using the full term structure yields quite different results. I am a bit confused. Thanks 
> 
> _______________________________________________
> QuantLib-users mailing list
> Qua...@li...
> https://lists.sourceforge.net/lists/listinfo/quantlib-users