[Quantlib-dev] CDS stubs schedules rule

[Jose Aparicio-N. <ja...@fr...>]

Hi all,
The long/short stub CDS rule (since the "big bang" aka 'old' CDS rule) states
that: (quoting MarkIt docs)

<<
If the trade date falls before 30 days before the first coupon date, the
accrual is due on the first coupon date for the number of days of effective
protection during the period. This is called a “short stub” period. If the
trade date is within 30 days before the first coupon date, there is a “long
stub period”.
No accrual of premium is paid on this first IMM coupon date, rather the long
stub is paid on the following coupon date. That payment will include the
portion of premium owed for protection in the first period plus the full
premium for the second period.
>>

Long stub meaning then a coupon larger thant the CDS tenor.
Ironically, with the big bang, this will no longer be the way the quoted CDS
will have the coupons generated in the US. Still, it is relevant for on the
run positions and non-US ones (though they might change too). But thats another
issue for another post.
This rule makes fwd spreads jump around the cds imm dates, with the effect on
options.

So far I have been using the 20thIMM rule which matches this rule in the short
stub case only.
I have added a new rule to account for the long stub.

For frequencies other than the 'usual' ones I do not think the rule makes sense
(think say, 10days). The stub rule was intended to make the coupon dates
homogeneous. So I have (arbitrarily) chosen to revert to the short stub in
that case. Another alternative is a QL_REQUIRE at the beginning of the Schedule
creation method refusing to take that combination.


I have also forced unadjustment on the first schedule date in this case since
theres no particular convention variable for that date. If we want a convention
on the startDate we pass it already adjusted (as it is done in the helpers).


Another small change I would like to request is for the cds helpers not to
adjust the final protection date accrual, which is the default convention (the
swaps will still adjust the final payment date when building the leg). They
would look like this:

   96     void CdsHelper::initializeDates() {
   97         Date protectionStart_ = evaluationDate_ + settlementDays_;
   98         Date startDate = calendar_.adjust(protectionStart_,
   99                                           paymentConvention_);
  100         Date endDate = evaluationDate_ + tenor_;
  101
  102         schedule_ =
  103             MakeSchedule().from(startDate)
  104                           .to(endDate)
  105                           .withFrequency(frequency_)
  106                           .withCalendar(calendar_)
  107                           .withConvention(paymentConvention_)
  >>>>>>>>>                     .withTerminationDateConvention(Unadjusted) <<<<
  108                           .withRule(rule_);
  109         earliestDate_ = schedule_.dates().front();
  110         latestDate_   = schedule_.dates().back();
  111     }

Alternatively we could have another convention variable in the helper
constructor.
Maybe the rule should be called 20thCDSstub or something different....

Tell me what you think.
Best regards
Pepe

--
Impacted files:
dategenerationrule.Xpp
schedule.cpp