Can Lsqnonlin iterative output display current point for the algorithm? - matlab

Question

My optimization using "Lsqnonlin" is running into an error in the 18th iteration. I was wondering if I could see what is the current input point that the algorithm is using for each iteration. It may help me diagnose what's going wrong. Thanks
EDIT: First Pass at Solution
I created myoutput.m
function stop = myoutput(x,optimValues,state)
stop = false;
indicator = x;
disp(indicator)
Then added OutPut Fcn to me options
options = optimset('disp','iter-detailed','MaxFunEvals',1000,'TolFun',1e-5,'OutputFcn',#myoutput);
HW1Fparams= lsqnonlin(HW1Fobjfun4,x0,lb,ub,options)
But I am getting hideous looking results like these:
I'd appreciate it if someone can help me make it look nicer. Below the break is the rest of the original question.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Full code below. I am using some Financial Toolbox functions. The idea is to calibrate the Hull White One Factor Model to market data. It's a straightforward exercise and I must be specifying things incorrectly because it's totally tripping me up.
ValuationDate = '10-01-2014';
Settle = datenum(ValuationDate);
CurveDates = [735874;
735882;
735906;
735936;
735950;
736040;
736133;
736224;
736314;
736424;
736606;
736788;
736971;
737153;
737336;
737518;
737701;
737884;
738069;
738251;
738433;
738615;
738797;
738979;
739162;
739345;
739528;
739710;
739893;
740075;
740260;
740442;
740624;
740806;
740989;
741171;
741354;
741536;
741719;
741901;
742084;
742269;
742451;
742633;
742815;
742997;
743180;
743362;
743545;
743728;
743911;
744093;
744278;
744460;
744642;
744824;
745006;
745189;
745372;
745554;
745737;
745919;
746102;
746284;
746469;
746651;
746833;
747015;
747198;
747380;
747563;
747745;
747928;
748111;
748296;
748478;
748660;
748842;
749024;
749206;
749389;
749572;
749755;
749937;
750120;
750302;
750487];
ZeroRates = 1.0e-03*[0.0172;
0.0188;
0.0191;
0.0221;
0.0249;
0.0244;
0.0269;
0.0333;
0.0423;
0.0571;
0.0789;
0.1021;
0.1253;
0.1435;
0.1617;
0.1749;
0.1881;
0.1973;
0.2064;
0.2158;
0.2253;
0.2311;
0.2370;
0.2429;
0.2488;
0.2547;
0.2607;
0.2640;
0.2672;
0.2706;
0.2738;
0.2772;
0.2807;
0.2842;
0.2877;
0.2913;
0.2948;
0.2964;
0.2979;
0.2995;
0.3011;
0.3026;
0.3043;
0.3060;
0.3077;
0.3095;
0.3112;
0.3118;
0.3125;
0.3132;
0.3138;
0.3146;
0.3152;
0.3160;
0.3167;
0.3175;
0.3183;
0.3186;
0.3189;
0.3192;
0.3196;
0.3199;
0.3202;
0.3206;
0.3209;
0.3213;
0.3217;
0.3217;
0.3216;
0.3216;
0.3216;
0.3216;
0.3216;
0.3216;
0.3216;
0.3216;
0.3216;
0.3217;
0.3217;
0.3218;
0.3218;
0.3219;
0.3219;
0.3220;
0.3220;
0.3221;
0.3221];
Compounding = 2;
RateSpec = intenvset('Compounding', 2,'ValuationDate', ValuationDate,'StartDates', ValuationDate,'EndDates', CurveDates,'Rates', ZeroRates);
InstrumentMaturity = datenum('12-Sep-2044');
SwaptionBlackVol = [ 0.5940 0.5550 0.4450 0.3710 0.3400 0.3110 0.2910 0.2750 0.2630 0.2520 0.2250 0.2140 0.2080 0.2050;
0.5630 0.5470 0.4420 0.3690 0.3360 0.3090 0.2900 0.2740 0.2630 0.2520 0.2260 0.2150 0.2090 0.2060;
0.5760 0.5330 0.4400 0.3730 0.3410 0.3150 0.2970 0.2820 0.2700 0.2590 0.2330 0.2220 0.2170 0.2140;
0.5840 0.5020 0.4240 0.3730 0.3480 0.3240 0.3060 0.2920 0.2810 0.2710 0.2430 0.2300 0.2230 0.2190;
0.5630 0.4750 0.4100 0.3700 0.3450 0.3230 0.3070 0.2940 0.2830 0.2740 0.2470 0.2330 0.2260 0.2210;
0.5510 0.4520 0.3980 0.3660 0.3410 0.3220 0.3070 0.2950 0.2850 0.2760 0.2500 0.2360 0.2290 0.2240;
0.4630 0.4010 0.3660 0.3440 0.3250 0.3100 0.2990 0.2890 0.2790 0.2720 0.2470 0.2320 0.2260 0.2210;
0.4230 0.3750 0.3480 0.3290 0.3140 0.3030 0.2930 0.2840 0.2760 0.2690 0.2420 0.2300 0.2240 0.2190;
0.3700 0.3470 0.3280 0.3110 0.2960 0.2880 0.2800 0.2730 0.2680 0.2620 0.2360 0.2240 0.2190 0.2150;
0.3420 0.3250 0.3100 0.2970 0.2850 0.2770 0.2700 0.2640 0.2590 0.2540 0.2280 0.2180 0.2140 0.2110;
0.3230 0.3010 0.2900 0.2810 0.2720 0.2650 0.2590 0.2540 0.2500 0.2470 0.2230 0.2130 0.2090 0.2060;
0.3010 0.2860 0.2760 0.2670 0.2580 0.2530 0.2480 0.2450 0.2420 0.2390 0.2160 0.2060 0.2030 0.2000;
0.2850 0.2750 0.2650 0.2560 0.2480 0.2440 0.2400 0.2370 0.2350 0.2320 0.2100 0.2000 0.1970 0.1940;
0.2710 0.2600 0.2510 0.2440 0.2380 0.2340 0.2310 0.2290 0.2260 0.2240 0.2040 0.1940 0.1910 0.1890;
0.2580 0.2470 0.2400 0.2350 0.2300 0.2270 0.2240 0.2210 0.2190 0.2170 0.1980 0.1890 0.1860 0.1840;
0.2460 0.2370 0.2320 0.2270 0.2240 0.2210 0.2180 0.2150 0.2130 0.2110 0.1980 0.1840 0.1820 0.1800;
0.2040 0.1980 0.1950 0.1920 0.1900 0.1890 0.1890 0.1880 0.1880 0.1870 0.1720 0.1660 0.1640 0.1620;
0.1790 0.1750 0.1740 0.1730 0.1730 0.1710 0.1710 0.1700 0.1690 0.1690 0.1530 0.1510 0.1500 0.1480;
0.1650 0.1650 0.1660 0.1670 0.1680 0.1670 0.1670 0.1680 0.1680 0.1680 0.1550 0.1580 0.1560 0.1530;
0.1530 0.1570 0.1590 0.1620 0.1640 0.1650 0.1660 0.1670 0.1680 0.1690 0.1560 0.1650 0.1620 0.1590];
SwaptionExerciseDates = cellstr(['1M ';'2M ';'3M '; '6M ';'9M ';'1Y ';'18M';'2Y ';'3Y ';'4Y ';'5Y ';'6Y ';'7Y ';'8Y ';'9Y ';'10Y';'15Y';'20Y';'25Y';'30Y']);
SwaptionTenors = cellstr(['1Y ';
'2Y ';
'3Y ';
'4Y ';
'5Y ';
'6Y ';
'7Y ';
'8Y ';
'9Y ';
'10Y';
'15Y';
'20Y';
'25Y';
'30Y']);
testmat = zeros(length(SwaptionExerciseDates),1);
for i = 1:length(SwaptionExerciseDates)
if SwaptionExerciseDates{i}(end)=='Y'
testmat(i) = addtodate(Settle,str2double(SwaptionExerciseDates{i}(1:end-1)),'year');
elseif SwaptionExerciseDates{i}(end)=='M'
testmat(i)=addtodate(Settle,str2double(SwaptionExerciseDates{i}(1:end-1)),'month');
end
end
EurExDates= testmat;
EurExDatesFull = repmat(testmat,1,length(SwaptionTenors));
testmat2 = zeros(length(SwaptionExerciseDates),length(SwaptionTenors));
for i = 1:size(EurExDatesFull,1)
for j = 1:size(EurExDatesFull,2)
if SwaptionTenors{j}(end)=='Y'
testmat2(i,j) = addtodate(EurExDatesFull(i,j),str2double(SwaptionTenors{j}(1:end-1)),'year');
elseif SwaptionTenors{j}(end)=='M'
testmat2(i,j)= addtodate(EurExDatesFull(i,j),str2double(SwaptionTenors{j}(1:end-1)),'month');
end
end
end
EurMatFull = testmat2;
relidx = find(EurMatFull <= InstrumentMaturity);
SwaptionBlackPrices = zeros(size(SwaptionBlackVol));
SwaptionStrike = zeros(size(SwaptionBlackVol));
for iSwaption=1:length(SwaptionExerciseDates)
for iTenor=1:length(SwaptionTenors)
[~,SwaptionStrike(iSwaption,iTenor)] = swapbyzero(RateSpec,[NaN 0],Settle, EurMatFull(iSwaption,iTenor),...
'StartDate',EurExDatesFull(iSwaption,iTenor),'LegReset',[1 2],'Basis',2);
SwaptionBlackPrices(iSwaption,iTenor) = swaptionbyblk(RateSpec,'call', SwaptionStrike(iSwaption,iTenor),Settle, ...
EurExDatesFull(iSwaption,iTenor), EurMatFull(iSwaption,iTenor),SwaptionBlackVol(iSwaption,iTenor));
end
end
TimeSpec = hwtimespec(Settle,daysadd(Settle,30*(1:370),6), 12);
% B = (214:224) produces error free solutions.
B = (150:224);
HW1Fobjfun4 = #(x) SwaptionBlackPrices(relidx(B)) - ...
swaptionbyhw(hwtree(hwvolspec(ValuationDate,testmat,x(2),testmat,x(1),'spline'), RateSpec, TimeSpec), 'call',SwaptionStrike(relidx(B)),EurExDatesFull(relidx(B)), 0,EurExDatesFull(relidx(B)), EurMatFull(relidx(B)),'Basis',2, 'SwapReset',12);
options = optimset('disp','iter','MaxFunEvals',1000,'TolFun',1e-5);
x0 = [.1 .01];
lb = [0 0];
ub = [1 1];
HW1Fparams = lsqnonlin(HW1Fobjfun4,x0,lb,ub,options)

Answer 1

Your best bet may be to modify the source lsqnonlin.m file. This can be a somewhat in-depth process, but it gives you the maximum control over what's going on.
Open the file by typing lsqnonlin at the command prompt, highlighting it, then right-clicking and clicking on Open Selection. Before you do anything else, save a copy of the file to your default Matlab working directory (e.g. C:\Users\username\Documents\MATLAB\ for Windows 7. Matlab puts your default working directory at the top of the search path, so if you have a program that's the same name as a Matlab built-in one, then Matlab will find yours first and use it instead. I don't have that particular function myself, so I can't give you the exact code to put in there, but the solution should be simple enough for you to implement.
With your locally-saved version of the code open, note that on the first line of the program, there's a function declaration that looks something like
function [output1,output2,...]=lsqnonlin(input1,input2,...)
From the MATLAB help page, it looks like x is the first output. Presumably, it's called x in the code itself or something similar, but if not, just use the first output parameter. Now that we know the name of the variable that is being output, we can go through the code and find where it is being calculated. MATLAB will probably have this routine be a wrapper around a more fundamental numerical code. For lsqnonneg, it calls lsqncommon, which then calls either snls or levenbergMarquardt, depending on the details of the problem. Any code that is iteratively solving something will eventually end up in a while loop, since it has to perform the same calculation an unknown number of times to converge on a solution. Once you find the while loop, it's simply a matter of adding a little code to output whatever parameter(s) you'd like to look at.
Just remember that as long as you have a file of the same name in your workspace, you'll be calling that one, not the original code, so you may want to delete (or at lease move) your modified code after you've finished debugging.