That was a funny challenge: cramming the Black-Scholes equation in an HP-12C.

- Fill "i" register with the interest, as usual (e.g. 12 for 12% per year).
- Fill "PV" with the spot value of the underlying asset.
- Fill "PMT" with the volatility, as a percentage (e.g. 25 for 25% per year).
- Fill "n" with the time to expiration. Use the same time unit as interest and volatility. If you enter interest and volatility as % per year, you must supply the time to expiration expressed in years (e.g. 2 months = 0.1666 years).
- Fill "FV" with the strike of the option.
- Press R/S to get the call value. It takes several seconds.
- The value on display is the call value. Press x⇄y to see the put value, and again to go back to the call value.

The option parameters are not changed during calculation, so you can fiddle with them and press R/S as many times as you want.

Opcode # | Operation | Remarks |
---|---|---|

0.196854 | ||

STO 3 | N(x) 1st coefficient | |

0.115194 | ||

STO 4 | N(x) 2nd coefficient | |

0.000344 | ||

STO 5 | N(x) 3rd coefficient | |

0.019527 | ||

STO 6 | N(x) 4th coefficient | |

f P/R | Goes into programming mode | |

f CLEAR PRGM | Clears programming memory | |

06 | RCL n | t |

07 | RCL i | 100.r |

08 | % | rt |

08 | CHS | -rt |

08 | e^{x} | e^{-rt} |

16 | RCL FV | K |

13 | × | K.e^{-rt} |

02 | STO EPX | √t |

15 | RCL PV | S |

15 | ÷ | K.e^{-rt}/S |

15 | 1/x | S.e^{rt}/K |

18 | g LN | ln(S/K) + rt |

02 | STO 0 | √t |

01 | RCL n | t |

02 | g √x | √t |

03 | RCL PMT | 100.σ |

04 | % | √t.σ |

17 | STO ÷ 0 | S/K |

05 | STO 2 | Let mem2 = √t.σ |

10 | 2 | |

17 | ÷ | S/K |

05 | STO - 0 | Let mem2 = √t.σ |

05 | RCL 0 | Let mem2 = √t.σ |

32 | 4 | calculate cummulative normal distribution |

33 | Y^{x} | d^{4} |

34 | RCL 6 | c4 |

35 | x⇄y | |

36 | × | c4.d^{4} |

37 | g LSTx | d^{4} |

38 | g √x | d^{2} |

39 | RCL 4 | c2 |

40 | x⇄y | |

41 | × | c2.d^{2} |

42 | g LSTx | d^{2} |

43 | g √x | d (guaranteed to be positive) |

44 | RCL 3 | c1 |

45 | × | c1.d |

46 | + | c1.d+c2.d^{2} |

47 | + | c1.d+c2.d^{2}+c4.d^{4} |

48 | RCL 0 | d |

49 | 6 | |

50 | Y^{x} | d^{6} |

51 | g √x | d^{3} (guaranteed to be positive) |

52 | RCL 5 | c3 |

53 | × | c3.d^{3} |

54 | + | c1.d+c2.d^{2} + c3.d^{3} + c4.d^{4} |

55 | 1 | |

56 | + | 1+c1.d+c2.d^{2} + c3.d^{3} + c4.d^{4} |

57 | 4 | |

58 | CHS | |

59 | Y^{x} | (1+c1.d+c2.d^{2} + c3.d^{3} + c4.d^{4})^{ - 4} |

60 | 2 | |

61 | ÷ | (1+c1.d+c2.d^{2} + c3.d^{3} + c4.d^{4})^{ - 4}/2 |

62 | 0 | Current result is 1-N(abs(d)) |

63 | RCL 0 | |

64 | g x≤y | d < 0? |

65 | GTO 72 | d<0, N(d)=1-N(-d), we are done, jump ahead |

66 | R↓ | d>0, need to do 1-(1-(N(d)) |

67 | R↓ | Restore 1-N(d) into register X |

68 | CHS | |

69 | 1 | |

70 | + | N(d) corrected, we are done |

71 | GTO 74 | Jump ahead |

72 | R↓ | Restore N(d) into register X |

73 | R↓ | |

75 | RCL 2 | mem2 is √t.σ at 1st round, zero otherwise |

76 | g x=0 | mem2 is zero? |

77 | GTO 81 | Jump to calculate call and put |

78 | RCL + 0 | Let mem0 = d2 + √t.σ = d1 |

79 | STO - 2 | Let mem2 = √t.σ - √t.σ = zero |

73 | R↓ | Recalls N(d2) |

74 | STO 1 | Let mem1 = N(d2) |

80 | GTO 26 | Go back to calculate N(d1) |

81 | R↓ | Recalls N(d1) |

81 | RCL PV | S |

36 | × | S.N(d1) |

82 | STO 0 | Let mem0 = S.N(d1) |

89 | RCL EPX | K.e^{-rt} |

90 | STO × 1 | Let mem1 = K.e^{-rt.}N(d2) |

91 | RCL PV | S |

92 | - | K.e^{-rt} - S |

93 | RCL 0 | S.N(d1) |

94 | RCL 1 | K.e^{-rt} . N(d2) |

95 | - | S.N(d1) - K.e^{-rt} . N(d2) = Call |

96 | + | Call + K.e^{-rt} - S = Put |

97 | g LSTx | Call at X, put at Y |

f P/R | Back to normal mode |

It was difficult to fit the equation in 99 programming steps, and using all steps reduces the available memory, which in turn makes the program more complicated.

The most critical function is the cummulative normal distribution, that we implement using a rational approximation with precision of 4 decimal digits. The coefficients are entered as part of the program but they are stored on memory positions (the STO 3, STO 4... at the beginning of the table). There are better approximations but they could not be fitted in such a small "computer".