RE: [Quantlib-dev] decreasing discount factors

[Ferdinando A. <fer...@am...>]

>Using the discount factor formula, if the change in rates is big enough
>relative to the time period, it is quite possible for the discount factors
>to increase!
>
>E.g:
>
>                         Rate    Days            Df
>                 3M      2.30    90/365  0.99436
>                 6M      1.10    180/365 0.99460
>
>This is unlikely but of course not impossible
this is simply impossible in the real world, as it would imply
a forward 3x6 rate of -0.10%, that is a negative interest rate, way out of 
order with rates at a level of 1%-2%
It is not impossible to have negative interest rate, but this happened in 
Japan on very short spot rates (overnight, tom-next). Negative forward 
rates would be an arbitrage opportunity.

Which leads me back to my original argument. Marco, what kind of benefit 
are you getting out of the "relaxed" constraints? What are you using the 
discount curve for?
I'm not asking this out of curiosity, but since it would put a considerable 
burden on my applications I would prefer to revert your commit unless I do 
understand the real benefit of the change. Not that I'm in a hurry, since I 
will not need a new executable at work in the next few months.

Just in case you wonder what kind of burden I'm talking about: I receive a 
discount curve as input from another application that uses FinCAD 
libraries. When this application has problems it usually output a curve 
with non-decreasing discounts, and that QuantLib requirement help me 
immediately detect such problems. I've also investigated how it is possible 
for that application to output non-decreasing discounts and, you know what, 
the problem is with a flawed Reuters/Bloomberg feed that provides wrong 
market rates, of the kind Andre has given as example, and that are not 
possible for EUR/USD/YEN these days.

This said I'm not against checking the discounts in an intermediate layer, 
but I would like to understand which benefit we get in return.

cooperatively yours -- Nando