QuantLib › quantlib-users

convergence error for flat yield curve using cubic spline

Classic

List

Threaded

4 messages Options

jlee

Sep 12, 2014; 6:22pm

convergence error for flat yield curve using cubic spline

9 posts

Hi all,

Can someone help point out what I am doing wrong? I am getting convergence errors when trying to bootstrap a yield curve using cubic spline. I have narrowed it down to a test case which is just a flat yield curve with 3 instruments (1w and 1m cash and 2y swap) all at 6%.

The main code is:

depoFRASwapInstruments.push_back(d1w);

depoFRASwapInstruments.push_back(d1m);

depoFRASwapInstruments.push_back(s2y);

boost::shared_ptr<YieldTermStructure> depoFRASwapTermStructure(new PiecewiseYieldCurve<ZeroYield, Cubic>(settlementDate, depoFRASwapInstruments, termStructureDayCounter, tolerance, Cubic(CubicInterpolation::Spline, false, CubicInterpolation::SecondDerivative, 0.0, CubicInterpolation::SecondDerivative, 0.0)));

// get 1 month discfact

std::cout << "1-month discount factor: " << std::setprecision(8) << depoFRASwapTermStructure->discount(1 / 12.0, false);

the error it returns is "convergence not reached after 99 iterations".

I think the code is fine as sometimes it does work (ie. if I use flat 5% it works). But I often get convergence errors making things unstable. Am i missing something? Does it just not converge?

Have attached the entire code.

Thanks,

Jon

------------------------------------------------------------------------------
Want excitement?
Manually upgrade your production database.
When you want reliability, choose Perforce
Perforce version control. Predictably reliable.
http://pubads.g.doubleclick.net/gampad/clk?id=157508191&iu=/4140/ostg.clktrk
_______________________________________________
QuantLib-users mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/quantlib-users

swap.cpp (16K) Download Attachment

Peter Caspers-4

Sep 13, 2014; 10:39am

Re: convergence error for flat yield curve using cubic spline

452 posts

Hi Jon,
your tolerance (1E-15) is very tight, try to relax that (in your
example 1E-14 already works, but the standard value 1E-12 should be
sufficient, too ?).
Peter

On 12 September 2014 20:22, Jon Lee <[hidden email]> wrote:

> Hi all,
> Can someone help point out what I am doing wrong? I am getting convergence
> errors when trying to bootstrap a yield curve using cubic spline. I have
> narrowed it down to a test case which is just a flat yield curve with 3
> instruments (1w and 1m cash and 2y swap) all at 6%.
>
> The main code is:
>
> depoFRASwapInstruments.push_back(d1w);
> depoFRASwapInstruments.push_back(d1m);
> depoFRASwapInstruments.push_back(s2y);
>
> boost::shared_ptr<YieldTermStructure> depoFRASwapTermStructure(new
> PiecewiseYieldCurve<ZeroYield, Cubic>(settlementDate,
> depoFRASwapInstruments, termStructureDayCounter, tolerance,
> Cubic(CubicInterpolation::Spline, false,
> CubicInterpolation::SecondDerivative, 0.0,
> CubicInterpolation::SecondDerivative, 0.0)));
>
> // get 1 month discfact
> std::cout << "1-month discount factor: " << std::setprecision(8) <<
> depoFRASwapTermStructure->discount(1 / 12.0, false);
>
> the error it returns is "convergence not reached after 99 iterations".
>
> I think the code is fine as sometimes it does work (ie. if I use flat 5% it
> works). But I often get convergence errors making things unstable. Am i
> missing something? Does it just not converge?
>
> Have attached the entire code.
>
> Thanks,
> Jon
>
> ------------------------------------------------------------------------------
> Want excitement?
> Manually upgrade your production database.
> When you want reliability, choose Perforce
> Perforce version control. Predictably reliable.
> http://pubads.g.doubleclick.net/gampad/clk?id=157508191&iu=/4140/ostg.clktrk
> _______________________________________________
> QuantLib-users mailing list
> [hidden email]
> https://lists.sourceforge.net/lists/listinfo/quantlib-users
>

... [show rest of quote]

jlee

Sep 15, 2014; 9:00pm

Re: convergence error for flat yield curve using cubic spline

9 posts

Hi Peter,
Thank you for spotting this. Most of the code was copied for the demo swap valuation. Out of the office this week but will give this a try when I get back.

Thanks,
Jon

> On 13 Sep 2014, at 12:39, Peter Caspers <[hidden email]> wrote:
>
> Hi Jon,
> your tolerance (1E-15) is very tight, try to relax that (in your
> example 1E-14 already works, but the standard value 1E-12 should be
> sufficient, too ?).
> Peter
>
>
>> On 12 September 2014 20:22, Jon Lee <[hidden email]> wrote:
>> Hi all,
>> Can someone help point out what I am doing wrong? I am getting convergence
>> errors when trying to bootstrap a yield curve using cubic spline. I have
>> narrowed it down to a test case which is just a flat yield curve with 3
>> instruments (1w and 1m cash and 2y swap) all at 6%.
>> The main code is:
>>
>> depoFRASwapInstruments.push_back(d1w);
>> depoFRASwapInstruments.push_back(d1m);
>> depoFRASwapInstruments.push_back(s2y);
>>
>> boost::shared_ptr<YieldTermStructure> depoFRASwapTermStructure(new
>> PiecewiseYieldCurve<ZeroYield, Cubic>(settlementDate,
>> depoFRASwapInstruments, termStructureDayCounter, tolerance,
>> Cubic(CubicInterpolation::Spline, false,
>> CubicInterpolation::SecondDerivative, 0.0,
>> CubicInterpolation::SecondDerivative, 0.0)));
>>
>> // get 1 month discfact
>> std::cout << "1-month discount factor: " << std::setprecision(8) <<
>> depoFRASwapTermStructure->discount(1 / 12.0, false);
>>
>> the error it returns is "convergence not reached after 99 iterations".
>>
>> I think the code is fine as sometimes it does work (ie. if I use flat 5% it
>> works). But I often get convergence errors making things unstable. Am i
>> missing something? Does it just not converge?
>>
>> Have attached the entire code.
>>
>> Thanks,
>> Jon
>>
>> ------------------------------------------------------------------------------
>> Want excitement?
>> Manually upgrade your production database.
>> When you want reliability, choose Perforce
>> Perforce version control. Predictably reliable.
>> http://pubads.g.doubleclick.net/gampad/clk?id=157508191&iu=/4140/ostg.clktrk
>> _______________________________________________
>> QuantLib-users mailing list
>> [hidden email]
>> https://lists.sourceforge.net/lists/listinfo/quantlib-users
>>

... [show rest of quote]

------------------------------------------------------------------------------
Want excitement?
Manually upgrade your production database.
When you want reliability, choose Perforce.
Perforce version control. Predictably reliable.
http://pubads.g.doubleclick.net/gampad/clk?id=157508191&iu=/4140/ostg.clktrk
_______________________________________________
QuantLib-users mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/quantlib-users

jlee

Sep 15, 2014; 9:29pm

Re: convergence error for flat yield curve using cubic spline

9 posts

In reply to this post by jlee

Hi,

Thanks for the reply and highlighting about the compounding issue. Yes I agree the 6 month depo is not 6% if the the 3m and 3x3 FRA are 6% because of the compounding. In the code I might have created the quote but in the curve construction I only pass in 3 instruments (1w, 1m and 2y). This was just a test scenario so not real data but was getting errors. I think Peter's suggestion about the tolerance levels helps so will try that out. Thanks for looking at it.

Jon
On 13 Sep 2014, at 19:49, BL BL <[hidden email]> wrote:

Hi,

i just had a quick look at your swap.cpp. Seems the deposit and the fras have some redundant quotes. Did you consider the compounding? I mean doing a 3M deposit and then a 3x3 FRA gives you a payment after 3 month. The 6 M deposit does "compound" those two 3M Periods and pays only once at the end.

So flat 6% is not ok for the par quotes. Not all of them can have 6%. This holds only for the 6% cc yield cuve for example. But be aware of the discrete rates because of compounding.

2014-09-12 20:22 GMT+02:00 Jon Lee <[hidden email]>:
Hi all,
Can someone help point out what I am doing wrong? I am getting convergence errors when trying to bootstrap a yield curve using cubic spline. I have narrowed it down to a test case which is just a flat yield curve with 3 instruments (1w and 1m cash and 2y swap) all at 6%.

The main code is:

depoFRASwapInstruments.push_back(d1w);
depoFRASwapInstruments.push_back(d1m);
depoFRASwapInstruments.push_back(s2y);

boost::shared_ptr<YieldTermStructure> depoFRASwapTermStructure(new PiecewiseYieldCurve<ZeroYield, Cubic>(settlementDate, depoFRASwapInstruments, termStructureDayCounter, tolerance, Cubic(CubicInterpolation::Spline, false, CubicInterpolation::SecondDerivative, 0.0, CubicInterpolation::SecondDerivative, 0.0)));

// get 1 month discfact
std::cout << "1-month discount factor: " << std::setprecision(8) << depoFRASwapTermStructure->discount(1 / 12.0, false);

the error it returns is "convergence not reached after 99 iterations".

I think the code is fine as sometimes it does work (ie. if I use flat 5% it works). But I often get convergence errors making things unstable. Am i missing something? Does it just not converge?

Have attached the entire code.

Thanks,
Jon

------------------------------------------------------------------------------
Want excitement?
Manually upgrade your production database.
When you want reliability, choose Perforce
Perforce version control. Predictably reliable.
http://pubads.g.doubleclick.net/gampad/clk?id=157508191&iu=/4140/ostg.clktrk
_______________________________________________
QuantLib-users mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/quantlib-users

... [show rest of quote]

... [show rest of quote]

------------------------------------------------------------------------------
Want excitement?
Manually upgrade your production database.
When you want reliability, choose Perforce.
Perforce version control. Predictably reliable.
http://pubads.g.doubleclick.net/gampad/clk?id=157508191&iu=/4140/ostg.clktrk
_______________________________________________
QuantLib-users mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/quantlib-users