OISRateHelper Payment and Fixing different calendars

[LZ1153]

Referring to URL below, I can replicate the issue today for SOFR Swap where 3Y maturity date is 14th April 2028, which 14th April 2028 is Government Bond but not Fed holiday. I gave a try by using Sofr() as Overnight Index, and passing Fed Holiday as in both MakeOIS & OISRateHelper. I couldn't get 3Y fairRate() the same as inputs, other tenors are perfectly price back to curve build inputs. Please could someone help look into fixing this issue?

https://quant.stackexchange.com/questions/81096/quantlib-oisratehelper-accrual-and-fixing-different-calendars

[lballabio]

May you post some sample code where you create the curve and reprice (or misprice) the input rate? Thanks!

[LZ1153]

Sorry for late response @lballabio , here is the sample code please.

import QuantLib as ql

reference_date = ql.Date(10, 4, 2025)
ql.Settings.instance().evaluationDate = reference_date

day_count = ql.Actual360()
settlement_days = 2
calendar = ql.UnitedStates(ql.UnitedStates.FederalReserve)

tenors = ["1Y", "2Y", "3Y"]
maturities = [3.7755, 3.4115, 3.3417]

sofr = ql.Sofr()

ois_helpers = [
    ql.OISRateHelper(
        settlement_days,
        ql.Period(tenor),
        ql.QuoteHandle(ql.SimpleQuote(rate / 100)),
        sofr,
        paymentFrequency=ql.Annual,
        paymentCalendar=calendar,
    )
    for tenor, rate in zip(tenors, maturities)
]

sofr_curve = ql.PiecewiseLogCubicDiscount(settlement_days, calendar, ois_helpers, day_count)
sofr_handler = ql.YieldTermStructureHandle(sofr_curve)

sofr = ql.Sofr(sofr_handler)

sofr_swap = ql.MakeOIS(ql.Period("3Y"), sofr, 0.0, ql.Period("0D"), calendar=calendar)
sofr_swap.setPricingEngine(ql.DiscountingSwapEngine(sofr_handler))

print(sofr_swap.startDate())
print(sofr_swap.maturityDate())
print(sofr_swap.fairRate())

[lballabio]

In the call to MakeOIS, replace calendar=calendar with paymentCalendar=calendar, like you did in the helper. The second causes the Fed calendar to be used to adjust the payment dates, while the SOFR calendar (which is slightly different) is used to create the swap schedule. The first causes the Fed calendar to be used for both.

[LZ1153]

Thank you @lballabio , I did try to use paymentCalendar=calendar instead and able to price back the input. Somehow this will result the maturity date to be 17th April 2028, where the correct maturity date should be 14th April 2028 instead.

[lballabio]

Ok, then instead of making the test swap equal to the one in the OIS helper, you'd have to make the swap in the OIS helper equal to the test swap.

In this particular case, this turns out to be tricky.

The maturity of your swap is April 14th 2028, which happens to be Good Friday. Usually, SOFR doesn't fix on that day (for instance, this year it didn't fix on Good Friday, April 18th). So the 14th is a holiday for the SOFR fixing calendar but not for the Federal Reserve calendar that you're specifying.

If not specified otherwise, the OIS helper uses the index calendar to create the schedule, which means that the 14th will be a holiday and the maturity will roll to the 17th (the same effect you saw when you passed paymentCalendar=calendar to the test swap and didn't pass calendar). First problem: as of now, the OIS helper is missing the parameters to specify a different calendar. The missing one has just been added in #2226 so in version 1.39 you'll be able to say:

    ql.OISRateHelper(
        settlement_days,
        ql.Period(tenor),
        ql.QuoteHandle(ql.SimpleQuote(rate / 100)),
        sofr,
        paymentFrequency=ql.Annual,
        fixedCalendar=calendar,
        overnightCalendar=calendar,
    )

and have the desired schedule generated.

Unfortunately I tried using the code that will be released in version 1.39, and there's a second problem. The schedule was correct, but the bootstrap failed because the instrument was trying to access the curve outside its defined range. The helper reported the maturity date as April 14th, and the curve used it as the date of the next node being bootstrapped, and therefore the maximum date in the range of the curve; but the instrument then tried to forecast the SOFR index for the 13th (the last fixing being compounded in that coupon) and since the 14th is a holiday, it asked the curve for the forward rate between the 13th and the 17th. I'll have to write some code in the helper to detect this case and set the node accordingly.

So, long story short: please wait for version 1.39. Sorry for the inconvenience.