[Quantlib-dev] Using Inflation Term Structure instead of CPI indices when they overlap

[Francois B. <ig...@gm...>]

Hi,

This is a continuation of the discussion at
http://quantlib.10058.n7.nabble.com/Valuing-CPI-Bond-at-real-yield-curve-td15354.html
.

To recap: I want to value a CPI Bond by discounting at the real yield curve
and input a zero inflation curve consisting only of 0 rates. I've
successfully constructed that curve using InterpolatedZeroCurve. This curve
is used for forecasting CPI indices. I also input historic CPI indices from
far back to the most recent available.

I'm now in a tricky situation where there is a bit of overlap. Suppose my
settlement date is 2014-05-27 and my observation lag is 4 months. What I
want to do is use actual CPI indices up to the lagged date of 2014-01-27
(calculated by linear interpolation between 2014-01-01 and 2014-02-01) and
for anything after that I want to use the zero inflation curve, which
implies any index that is looked up for a date after 2014-01-27 will be
exactly the same (no growth).

My zero inflation curve already has a base date of 2014-01-27, and my
historic CPI indices go up to April 2014. So there is a bit of overlap. At
the moment, the ZeroInflationIndex.fixing() method prefers to use CPI
indices if they are available. I want to rather use the zero inflation
curve. There is a forecastTodaysFixing parameter which is currently ignored.

I propose to implement the forecastTodaysFixing so that when a 'true' is
passed to it, the method will rather using the forecasting part and thus
use the zero inflation curve. In CPICoupon, where this method is called, I
will check whether the zero inflation curve base date is before the
CPICoupon's fixing date and if so, pass 'true'. Or the alternative is to
introduce a new parameter in the CPIBond constructor and pass this through
to CPICoupon to determine which of the two curves get preference.

I'm posting here to ask what I should look out for? I don't want to
introduce a breaking change. The fixing date methods are quite tricky and
it took me a while to get my head around it. Any advice will be appreciated.

As a sidenote, limiting the historic CPI indices up to an earlier date
doesn't solve the problem. The algorithm uses the availability and
observation lags to deduce up to which date the CPI index should be used.
So if I exclude indices after 2014-02-01, then a lookup error occurs.

regards,
Francois Botha