Re: [Quantlib-dev] boostraping a YC in the past

[Bianchetti M. <mar...@ca...>]

Concerning the following version for the calcualtion of historical
correlations on forward rates to be used in the Market Model:

> loop currentDate backward from the latest date to the earliest one
> get fixings from Indexes and set them to your relevant Quotes
> set the EvaluationDate equal to currentDate
> (re-)build RateHelpers and YieldCurve
> calculate forward rates at constant interval (e.g. 6M), not constant
dates

This last point is controversial.
For LMM we need a correlation matrix among forward rates of kind
L(t,S,T), observed at t<S, between dates S<T.

1) So for each past date t'<t-D one would calculate the forward rate
between the *same* dates L(t',S,T), the only difference being the
observation time.=20

Instead:

2) Calculating the forward rates in the past at constant interval
L(t',S',T') with S'=3DS-D and T'=3DT-D amounts to calculate *different*
forward rates, so the financial meaning is poorer in my opinion.

This is also consistent with the discussion given in Brigo & Mercurio,
2nd edition, par. 6.19.1.

Any other opinion ?
Well, probably the best thing to do at the end is to implement both
methods (not much extra work) and go for testing on the market...

Ciao
Marco

> calculate fwd rate differences, and add them to a SequenceStatistics
class
> get correlation from SequenceStatistics