Black-Scholes for 17BII+ with 12 Digit Accuracy

The Museum of HP Calculators

HP Forum Archive 15

Black-Scholes for 17BII+ with 12 Digit Accuracy
Message #1 Posted by Bob Wang on 3 June 2005, 5:42 p.m.

Black-Scholes for 17BII+ with 12 Digit Accuracy
Tony Hutchins has an article in DataFile V24N2pp13-21 Wronski's Lex Suprema on the HP-12C that presents a Black-Scholes program accurate to 12 digits. Tony has graciously given me permission to post the HP 17BII+ version here. The INIT.VARS equation is to initialize all one-letter variables to allow L() and G() to work properly. I also included run-times for this version; as strike prices become less probable, times increase to around 9 minutes.
PS=52, PE=47, RF%=0.5, T=6, S=0.2054

INIT.VARS:A+B+C+D+E+F+G+H+I+J+K+L+M+N+O+P+Q+R+S+T+U+V+W+X+Y+Z

BLK.SCHLS,11.9: 0xPSxPExRF%xTxS+IF(S(CALLV): 0xL(X:(LN(PS÷PE)+(RF%÷100+S^2÷2)xT)÷S÷SQRT(T)) xL(Y:G(X)-SxSQRT(T))+PSxIF(G(X)^2>49: IF(G(X)>0:1:0):.5+G(X)xEXP(-(G(X)^2)÷2)÷SQRT(2xPI)x(1+ SIGMA(N:1:10xABS(G(X))+IF(ABS(G(X))<1.4:5:0):1:(2xG(X)^2)^NxFACT(N)÷FACT(2xN+1)))) -PExEXP(-RF%xT÷100)xIF(G(Y)^2>49: IF(G(Y)>0:1:0):.5+G(Y)xEXP(-(G(Y)^2)÷2)÷SQRT(2xPI)x(1+ SIGMA(N:1:10xABS(G(Y))+IF(ABS(G(Y))<1.4:5:0):1:(2xG(Y)^2)^NxFACT(N)÷FACT(2xN+1)))) -CALLV: -PS+PExEXP(-RF%xT÷100)+CALLV-PUTV)

Exact Call Calculated Call Error Execution Time PE=47, d1=.6957 13.208118745061 13.208118745100 -3.92344E-11 39 sec PE=37, d1=.9876 18.961966848997 18.961966849000 -3.1477E-12 58 sec PE=22, d1=2.0209 30.902477228531 30.902477228600 -6.94556E-11 1 min 34 sec PE=13, d1=3.0666 39.393284752365 39.393284752200 1.64839E-10 2 min 50 sec PE=02, d1=6.7869 50.059108932926 50.059108932900 2.57856E-11 8 min 55 sec PE=01, d1=8.1645 51.029554466452 51.029554466500 -4.85088E-11 7 sec
An alternative is to use a 5th order polynomial approximation that always executes in 13 seconds, but only has 7.4 digits of accuracy:

BLK.SCHLS,7.4: 0xPSxPExRF%xTxS+IF(S(CALLV): 0xL(B:1÷(1+.2316419xABS(L(D:L(C:(LN(PS÷PE) +(RF%÷100+S^2÷2)xT)÷S÷SQRT(T))-SxSQRT(T))))) xL(A:1÷(1+.2316419xABS(G(C))))+PSxABS(IF(G(C)<0:0:-1) +(.254829592xG(A)-.284496736xG(A)^2+1.421413741xG(A)^3 -1.453152027xG(A)^4+1.061405429xG(A)^5)xEXP(-G(C)^2÷2)÷2) -PExEXP(-RF%xT÷100)xABS(IF(G(D)<0:0:-1) +(.254829592xG(B)-.284496736xG(B)^2+1.421413741xG(B)^3 -1.453152027xG(B)^4+1.061405429xG(B)^5)xEXP(-G(D)^2÷2)÷2) -CALLV: -PS+PExEXP(-RF%xT÷100)+CALLV-PUTV)

Exact Call Calculated Call Error Execution Time PE=47, d1=.6957 13.208118745061 13.208118448700 2.96361E-07 13 sec PE=37, d1=.9876 18.961966848997 18.961966877200 -2.82031E-08 13 sec PE=22, d1=2.0209 30.902477228531 30.902481308000 -4.07947E-06 13 sec PE=13, d1=3.0666 39.393284752365 39.393281519000 3.23336E-06 13 sec PE=02, d1=6.7869 50.059108932926 50.059108932900 2.57856E-11 11 sec PE=01, d1=8.1645 51.029554466452 51.029554466500 -4.85088E-11 11 sec
Bob

Re: Black-Scholes for 17BII+ with 12 Digit Accuracy
Message #2 Posted by tony on 4 June 2005, 4:11 a.m.,
in response to message #1 by Bob Wang

Gidday Bob! This really seems to highlight yet another issue with the 17BII+. I don't dispute your results, but on the 19BII execution time is not really an issue. Your excellently chosen worst case for the algorithm with d1 & d2 up there just under 7 executes in 30 seconds on the 19BII - about 18 times faster than on the 17BII+. The case for d1 near 3 executes in 13 seconds and the rest in 10 seconds or less. The program takes about 15 seconds to "verify" on the 19BII so is fairly cumbersome to start, but exceutes well. In practice d1 and d2 will be under 3 so the execution time will be under about 10 seconds. Anything taking longer is probably a theoretical case :-) Cheers, and thanks for your contributions and ability to make the 17BII+ work! Tony

Re: Black-Scholes for 17BII+ with 12 Digit Accuracy
Message #3 Posted by tony on 6 June 2005, 7:53 a.m.,
in response to message #1 by Bob Wang

Bob, me again, only a day or two late :-)
This may work on the 17BII+, as it uses L() only, not G() and does not use IF(S(..., so we limit the things that can go awry :-).It works on a 19BII.
It uses the same 5th degree polynomial approximation as your BLK.SCHLS,7.4, just evaluated differently.

BLK.SCHLS:0*L(D1:(LN(PS/PE)+(RF%/100+S^2/2)*T)/S/SQRT(T)) *L(D2:D1-S*SQRT(T))*D2*PUTV+PS*ABS(IF(D1<0;0;-1) +SIGMA(I:1:5:1:ITEM(NORM:I)*SPPV(23.16419*ABS(D1):I)) /EXP(D1^2/2))-PE/EXP(RF%*T/100)*ABS(IF(D2<0:0:-1) +SIGMA(I:1:5:1:ITEM(NORM:I)*SPPV(23.16419*ABS(D2):I)) /EXP(D2^2/2))-CALLV+0*L(PUTV:CALLV-PS+PE/EXP(RF%*T/100))
The first page of Solver shows, here on the 19BII:
PS PE RF% S T MORE
These are the five *inputs*.
The second page shows:
D1 D1 PUTV CALLV MORE
These are the 4 *outputs*. They are all calculated simultaneously, by pressing CALLV. Then RCL PUTV shows PUTV and likewise for D1 and D2. It is often useful to have D1 and D2 available.**Only CALLV is "solved for"as such**. This is the main difference - we don't first solve for CALLV and then for PUTV, we just do the one solve.
NORM is a 5 element SUM List:

1 0.127414796 2 -0.142248368 3 0.710706871 4 -0.726576013 5 0.530702714 ----------- Total 0.500000000
SUM Lists are quite handy - here it reduces the typing required for the equation quite a bit (by about a third) - maybe the nearest thing we have to subroutines :-)
For the inputs: PS=52, RF%=.5, S=.2054 and T=6, and the various PE you tabulate, this produces basically the same results you list. RCL PUTV shows zero for the last 2 examples (PE=2 and PE=1). Even for PE=13, RCL PUTV shows only 1 cent. For PE=47 D1 is .5121. PE=45 specifies the same example used on the HP web site for the 17BII Black Scholes equation. The above equation is similar to one given in my "Wronski" article in DataFile. It does slow down Solver slightly as it is calculating more than one output at once, and I don't know if this is possible on the 17BII+ :-)
Cheers, Tony

[ Return to Index | Top of Index ]

Go back to the main exhibit hall