Menu
▾
▴
Re: [Quantlib-users] Bootstraping difference between python QuantLib and Bloomberg.
|
From: Luigi B. <lui...@gm...> - 2018-03-29 07:34:19
|
Glad to hear it. Thanks!
Luigi
On Thu, Mar 29, 2018 at 2:04 AM Jose Pedro Melo <jm...@li...> wrote:
> Finally, after many iterations, i found were was the problem. I changed
> this part of the code:
>
> handle = ql.YieldTermStructureHandle()
> helpers +=
> [ql.OISRateHelper(settlement_days_icp,ql.Period(months,ql.Months),
> ql.QuoteHandle(ql.SimpleQuote(rate/100)),ICP,handle,False,0,ql.Following,ql.Monthly)
> for rate, months in zip(swap_month,months)]
>
> helpers +=
> [ql.OISRateHelper(settlement_days_icp,ql.Period(years,ql.Years),
> ql.QuoteHandle(ql.SimpleQuote(rate/100)),ICP,handle,False,0,ql.Following,ql.Semiannual)
> for rate, years in zip(swap_years,years)]
>
> Seems like the OISRateHelper payment frequency was different than our ICP
> swaps. This modification yields the close results to BBG bootsrapping:
>
> 0 0.000000 0.000000
> 1 0.008333 2.500000
> 2 0.261111 2.486395
> 3 0.513889 2.483810
> 4 0.769444 2.501550
> 5 1.025000 2.560555
> 6 1.527778 2.691903
> 7 2.038889 2.884934
> 8 3.061111 3.315383
> 9 4.072222 3.674813
> 10 5.083333 3.976807
> 11 6.097222 4.306079
> 12 7.113889 4.607043
> 13 8.127778 4.903621
> 14 9.147222 5.175711
> 15 10.158333 5.471312
> 16 15.233333 6.497894
> 17 20.308333 7.977794
>
> Thanks to all for your time. I'll leave the complete code here for the
> people interested:
>
> import QuantLib as ql
> import pandas as pd
> def create_calendar_chile(start_year,n_years):
> Chile = ql.WeekendsOnly()
> days = [1,14,15,1,21,26,2,16,15,18,19,9,27,1,19,8,17,25,31]
> months = [1,4,4,5,5,6,8,9,9,10,10,11,12,12,12,12]
> name = ['Año Nuevo','Viernes Santo','Sabado Santo','Dia del
> Trabajo','Dia de las Glorias Navales','San Pedro y San Pablo','Elecciones
> Primarias','Dia de la Virgen del Carmen','Asuncion de la
> Virgen','Independencia Nacional','Glorias del Ejercito','Encuentro de dos
> mundos','Día de las Iglesias Evangélicas y Protestantes','Día de todos los
> Santos','Elecciones Presidenciales y Parlamentarias','Inmaculada
> Concepción','Segunda vuelta Presidenciales','Navidad','Feriado Bancario']
> for i in range(n_years+1):
> for x,y in zip(days,months):
> date = ql.Date(x,y,start_year+i)
> Chile.addHoliday(date)
> return Chile
> def get_curve(date,swap,currency = 'CLP'):
> calendar = create_calendar_chile(2001,50)
> dayCounter_Act360 = ql.Actual360()
> settlement_days_icp = 2
> # OIS quotes up to 20 years
> ICP = ql.OvernightIndex("ICP", settlement_days_icp, ql.CLPCurrency(),
> calendar, dayCounter_Act360)
> fixingDays = 0
> if currency == 'CLP':
> # setup DepositRateHelper for 0-1 days
> TPM = swap[0]
> months = [3,6,9,12,18]
> swap_month = swap[1:6]
> years =[2,3,4,5,6,7,8,9,10,15,20]
> swap_years = swap[6:]
> helpers
> =[ql.DepositRateHelper(ql.QuoteHandle(ql.SimpleQuote(TPM/100)),ql.Period(1,ql.Days),
> fixingDays,calendar, ql.Following, False,ql.Actual360())]
> else:
> months = [3,6,9,12]
> swap_month = swap[0:3]
> years =[2,3,4,5,6,7,8,9,10,15,20]
> swap_years = swap[4:]
> helpers = []
>
> # setup OISRateHelper from 3 months to 20 years
> #helpers
> +=[ql.DepositRateHelper(ql.QuoteHandle(ql.SimpleQuote(rate/100)),ql.Period(months,ql.Months),settlement_days_icp,calendar,ql.Following,False,ql.Actual360())
> for rate, months in zip(swap_month,months)]
> handle = ql.YieldTermStructureHandle()
> helpers +=
> [ql.OISRateHelper(settlement_days_icp,ql.Period(months,ql.Months),
> ql.QuoteHandle(ql.SimpleQuote(rate/100)),ICP,handle,False,0,ql.Following,ql.Monthly)
> for rate, months in zip(swap_month,months)]
> helpers +=
> [ql.OISRateHelper(settlement_days_icp,ql.Period(years,ql.Years),
> ql.QuoteHandle(ql.SimpleQuote(rate/100)),ICP,handle,False,0,ql.Following,ql.Semiannual)
> for rate, years in zip(swap_years,years)]
> icp_curve = ql.PiecewiseLogCubicDiscount(date, helpers, ql.Actual360())
> icp_curve.enableExtrapolation()
> return icp_curve
> def print_zero(date,yieldcurve):
> day_count = ql.Actual360()
> spots = []
> tenors = []
> for d in yieldcurve.dates():
> yrs = day_count.yearFraction(date, d)
> print(yrs,d)
> compounding = ql.Simple
> freq = ql.Annual
> zero_rate = yieldcurve.zeroRate(yrs, compounding, freq)
> print(zero_rate)
>
> tenors.append(yrs)
> eq_rate =
> zero_rate.equivalentRate(day_count,compounding,freq,date,d).rate()
> zero_rate.equivalentRate(day_count,compounding,freq,date,d).rate()
> spots.append(100*eq_rate)
> datatable = {'Years': tenors,'Zero': spots}
> datatable = pd.DataFrame.from_dict(datatable)
> print(datatable)
> #Eval. Date
> date_ql = ql.Date(12,1,2018)
> ql.Settings.instance().evaluationDate = date_ql
> swap_clp = [2.5, 2.48, 2.47, 2.48, 2.53, 2.64, 2.82, 3.17, 3.43, 3.62,
> 3.81,3.96, 4.09, 4.19, 4.29, 4.45, 4.62]
> yieldcurve_clp = get_curve(date_ql,swap_clp)
> print_zero(date_ql,yieldcurve_clp)
>
>
> ------------------------------
> *De:* Raul Muñoz <rm...@gm...>
> *Enviado:* martes, 20 de marzo de 2018 03:22 a. m.
> *Para:* Luigi Ballabio
> *CC:* Jose Pedro Melo; qua...@li...
> *Asunto:* Re: [Quantlib-users] Bootstraping difference between python
> QuantLib and Bloomberg.
>
> Hi:
>
> You need to try: SWDF DFLT, with this comand you can see the defaults that
> bbg is using (attached). Another thing that I didn't see in your code is
> the spot lag. In chile the spot lag is t+2 (as usual) so that the
> bootstrapp should match to all par values to t+2. And finally I suggest you
> to match discount factors instead of rates.
>
> Best Regards,
>
> RAMR
>
>
>
> 2018-03-19 14:09 GMT-03:00 Luigi Ballabio <lui...@gm...>:
>
> I think there's still something we don't know about the way the Bloomberg
> curve is built. For instance, in the first couple of rows of the Bloomberg
> data you reported:
>
> 10-01-2018 11-01-2018 1 1,000069 ZERO 2,500
> 12-01-2018 12-04-2018 90 0,993838 ZERO 2,480
>
> The first one has a positive rate, yet the discount factor is larger than
> 1. I think this suggests that the curve has its reference date on the
> 12th; but in your code you have to set it to the 10th in order to include
> the first overnight rate, because our curve doesn't go before the
> reference.
>
> Does anybody have any insights on this?
>
> Luigi
>
>
>
> On Sun, Feb 25, 2018 at 5:35 AM Jose Pedro Melo <jm...@li...> wrote:
>
> Unfortunately i haven't advance much. Why do i get different results when
> i bootstrap the curve with the swap and OIS rate helpers?
>
> 2018-02-13 7:29 GMT-03:00 Luigi Ballabio <lui...@gm...>:
>
> Apologies for dropping the ball. Did you make any progress?
>
> Luigi
>
>
> On Thu, Jan 25, 2018 at 6:41 PM Jose Pedro Melo <jm...@li...> wrote:
>
> Thanks for the fast response,
>
> I'm not sure which interpolation method Bbg uses, but i tried changing it
> from cubic to linear in quantlib -PiecewiseLinearZero and
> PiecewiseCubicZero- to check if that would help, without much success. In
> the following table i changed both the interpolation method and the start
> date:
>
> Start Date End Date DF Days Year Fraction Interpolation Output
> 12/01/2018 12/01/2038 0.386246 7305 20.2916667 Cubic 7.830933
> 12/01/2018 12/01/2038 0.386246 7305 20.2916667 Linear 7.831006
>
> I changed the settlement days to 0 in order to get the same starting
> dates. For reference, i tried a basic bootstrapping spreadsheet i had and
> i'm getting closer result to bloomberg's output.
>
>
> El 24 ene. 2018 12:54 p. m., Luigi Ballabio <lui...@gm...>
> escribió:
>
> To check that, you can try passing the spot date instead of today's date
> to the curve. That should cause the start date to go to January 12.
>
> Apart from that, do you know what kind of interpolation Bloomberg uses?
> That might have an effect too.
>
>
> On Wed, Jan 24, 2018 at 4:48 PM Jose Pedro Melo <jm...@li...> wrote:
>
> I did the following comparison:
>
> Source Start Date End Date Days Disc. Factor Year Fraction
> (1/df-1)*360/days Original Output
> Quantlib 10/01/2018 12/01/2038 7307 0.386192 20.2972222 7.831% 7.831
> Bloomberg 12/01/2018 12/01/2038 7305 0.382045 20.2916667 7.971% 7.971
>
> There is a small difference in the amount of days used to calculate the
> rates. I don't know if these difference on start dates could have a
> relevant impact in the bootsrapping algorithm.
>
>
> 2018-01-24 12:13 GMT-03:00 Luigi Ballabio <lui...@gm...>:
>
> Did you try using the same formula, or comparing the discount factors
> directly? (I'd try to understand if the difference is in the underlying
> curve or in the equivalentRate calculation.)
>
>
>
> On Wed, Jan 24, 2018 at 4:06 PM Jose Pedro Melo <jm...@li...> wrote:
>
> Thanks for the response Luigi,
>
> The 'semiannual' tag appears only to show the swap original frequency. I
> got the rates using the excel formula =BCurveInt(A1;"zc.mid";B1;"Frequency=Zero"),
> where A1 is the curve bbg ID and B1 is the date for which you are getting
> the rate. The formula behind the rates is (1/DF-1)*360/Days.
>
>
> El 24 ene. 2018 11:45 a. m., "Luigi Ballabio" <lui...@gm...>
> escribió:
>
> What is Bloomberg quoting, precisely? The "Semiannual" they're displaying
> seems to go against the simple-compounding rate you're extracting...
>
>
> On Mon, Jan 15, 2018 at 9:28 AM josepedro123 <jm...@li...> wrote:
>
> I'm trying to get the Zero rates implied by the swap CLP curve (ICP). The
> following code works but it doesn't produce similar levels compared to
> Bloomberg's Zero curve. Here is a snippet of the code:
>
> import QuantLib as ql
> import pandas as pd
>
> def create_calendar_chile(start_year,n_years):
> Chile = ql.WeekendsOnly()
> days = [1,14,15,1,21,26,2,16,15,18,19,9,27,1,19,8,17,25,31]
> months = [1,4,4,5,5,6,8,9,9,10,10,11,12,12,12,12]
> name = ['Año Nuevo','Viernes Santo','Sabado Santo','Dia del
> Trabajo','Dia de las Glorias Navales','San Pedro y San Pablo','Elecciones
> Primarias','Dia de la Virgen del Carmen','Asuncion de la
> Virgen','Independencia Nacional','Glorias del Ejercito','Encuentro de dos
> mundos','Día de las Iglesias Evangélicas y Protestantes','Día de todos los
> Santos','Elecciones Presidenciales y Parlamentarias','Inmaculada
> Concepción','Segunda vuelta Presidenciales','Navidad','Feriado Bancario']
> for i in range(n_years+1):
> for x,y in zip(days,months):
> date = ql.Date(x,y,start_year+i)
> Chile.addHoliday(date)
> return Chile
> def get_curve(date,swap,currency = 'CLP'):
> calendar = create_calendar_chile(2001,50)
> dayCounter_Act360 = ql.Actual360()
> settlement_days_icp = 2
> # OIS quotes up to 20 years
> ICP = ql.OvernightIndex("ICP", settlement_days_icp, ql.CLPCurrency(),
> calendar, dayCounter_Act360)
> fixingDays = 0
>
> if currency == 'CLP':
> # setup DepositRateHelper for 0-1 days
> TPM = swap[0]
> months = [3,6,9,12,18]
> swap_month = swap[1:6]
> years =[2,3,4,5,6,7,8,9,10,15,20]
> swap_years = swap[6:]
> helpers =
> [ql.DepositRateHelper(ql.QuoteHandle(ql.SimpleQuote(TPM/100)),
> ql.Period(1,ql.Days), fixingDays,
> calendar, ql.Following, False,
> ql.Actual360())]
> else:
> months = [3,6,9,12]
> swap_month = swap[0:3]
> years =[2,3,4,5,6,7,8,9,10,15,20]
> swap_years = swap[4:]
> helpers = []
>
>
> # setup OISRateHelper from 3 months to 20 years
> helpers +=
> [ql.DepositRateHelper(ql.QuoteHandle(ql.SimpleQuote(rate/100)),
> ql.Period(months,ql.Months),
> settlement_days_icp,
> calendar,
> ql.Following,
> False,
> ql.Actual360())
> for rate, months in zip(swap_month,months)]
>
> #helpers += [ql.OISRateHelper(settlement_days_icp,
> ql.Period(months,ql.Months),
> #
> ql.QuoteHandle(ql.SimpleQuote(rate/100)),ICP)
> # for rate, months in
> zip(swap_month,months)]
>
> helpers += [ql.OISRateHelper(settlement_days_icp,
> ql.Period(years,ql.Years),
>
> ql.QuoteHandle(ql.SimpleQuote(rate/100)),ICP)
> for rate, years in
> zip(swap_years,years)]
>
> icp_curve = ql.PiecewiseCubicZero(date, helpers, ql.Actual360())
> icp_curve.enableExtrapolation()
> return icp_curve
> def print_zero(date,yieldcurve):
> day_count = ql.Actual360()
> spots = []
> tenors = []
> for d in yieldcurve.dates():
> yrs = day_count.yearFraction(date, d)
> print(yrs,d)
> compounding = ql.Simple
> freq = ql.Annual
> zero_rate = yieldcurve.zeroRate(yrs, compounding, freq)
> tenors.append(yrs)
> eq_rate =
> zero_rate.equivalentRate(day_count,compounding,freq,date,d).rate()
> zero_rate.equivalentRate(day_count,compounding,freq,date,d).rate()
> spots.append(100*eq_rate)
>
> datatable = {'Years': tenors,'Zero': spots}
> datatable = pd.DataFrame.from_dict(datatable)
> print(datatable)
>
> #Eval. Date
> date_ql = ql.Date(10,1,2018)
> ql.Settings.instance().evaluationDate = date_ql
> swap_clp = [2.5, 2.48, 2.47, 2.48, 2.53, 2.64, 2.82, 3.17, 3.43, 3.62,
> 3.81,
> 3.96, 4.09, 4.19, 4.29, 4.45, 4.62]
> yieldcurve_clp = get_curve(date_ql,swap_clp)
> print_zero(date_ql,yieldcurve_clp)
> Results:
> Years Zero
> 0 0.000000 0.000000
> 1 0.002778 2.500000
> 2 0.255556 2.480773
> 3 0.508333 2.470668
> 4 0.763889 2.480488
> 5 1.025000 2.530187
> 6 1.522222 2.639855
> 7 2.036111 2.864088
> 8 3.050000 3.287675
> 9 4.063889 3.640960
> 10 5.077778 3.937202
> 11 6.091667 4.259577
> 12 7.111111 4.553990
> 13 8.122222 4.843389
> 14 9.136111 5.108135
> 15 10.150000 5.396163
> 16 15.225000 6.400718
> 17 20.297222 7.830561
> BBG:
> Start End Days DF Frequency Zero
> 10-01-2018 11-01-2018 1 1,000069 ZERO 2,500
> 12-01-2018 12-04-2018 90 0,993838 ZERO 2,480
> 12-01-2018 12-07-2018 181 0,987734 ZERO 2,470
> 12-01-2018 12-10-2018 273 0,981540 ZERO 2,480
> 12-01-2018 14-01-2019 367 0,974857 ZERO 2,530
> 12-01-2018 12-07-2019 546 0,961501 ZERO 2,640
> 12-01-2018 13-01-2020 731 0,944621 SEMIANNUAL 2,887
> 12-01-2018 12-01-2021 1096 0,908255 SEMIANNUAL 3,318
> 12-01-2018 12-01-2022 1461 0,870133 SEMIANNUAL 3,678
> 12-01-2018 12-01-2023 1826 0,832049 SEMIANNUAL 3,980
> 12-01-2018 12-01-2024 2191 0,792244 SEMIANNUAL 4,309
> 12-01-2018 13-01-2025 2558 0,753263 SEMIANNUAL 4,610
> 12-01-2018 12-01-2026 2922 0,715195 SEMIANNUAL 4,906
> 12-01-2018 12-01-2027 3287 0,679005 SEMIANNUAL 5,178
> 12-01-2018 12-01-2028 3652 0,642988 SEMIANNUAL 5,473
> 12-01-2018 12-01-2033 5479 0,502562 SEMIANNUAL 6,504
> 12-01-2018 12-01-2038 7305 0,382045 SEMIANNUAL 7,971
>
> There is a +10 bp difference in the long term part. I tried using the swap
> rate helpers and tweacking the days to get the same year fractions but it
> doens't seem to be the problem. I wonder how is the oisratehelper
> configurated (payment conventions, rates). Any ideas of where could be my
> mistake?
> Thanks,
>
>
>
> --
> Sent from: http://quantlib.10058.n7.nabble.com/quantlib-users-f3.html
>
>
> ------------------------------------------------------------------------------
> Check out the vibrant tech community on one of the world's most
> engaging tech sites, Slashdot.org! http://sdm.link/slashdot
> _______________________________________________
> QuantLib-users mailing list
> Qua...@li...
> https://lists.sourceforge.net/lists/listinfo/quantlib-users
>
>
>
>
>
>
> ------------------------------------------------------------------------------
> Check out the vibrant tech community on one of the world's most
> engaging tech sites, Slashdot.org! http://sdm.link/slashdot
> _______________________________________________
> QuantLib-users mailing list
> Qua...@li...
> https://lists.sourceforge.net/lists/listinfo/quantlib-users
>
>
>
|
