xref: /petsc/src/tao/complementarity/tutorials/blackscholes.c (revision 6dd63270497ad23dcf16ae500a87ff2b2a0b7474)
1 /*
2     American Put Options Pricing using the Black-Scholes Equation
3 
4    Background (European Options):
5      The standard European option is a contract where the holder has the right
6      to either buy (call option) or sell (put option) an underlying asset at
7      a designated future time and price.
8 
9      The classic Black-Scholes model begins with an assumption that the
10      price of the underlying asset behaves as a lognormal random walk.
11      Using this assumption and a no-arbitrage argument, the following
12      linear parabolic partial differential equation for the value of the
13      option results:
14 
15        dV/dt + 0.5(sigma**2)(S**alpha)(d2V/dS2) + (r - D)S(dV/dS) - rV = 0.
16 
17      Here, sigma is the volatility of the underling asset, alpha is a
18      measure of elasticity (typically two), D measures the dividend payments
19      on the underling asset, and r is the interest rate.
20 
21      To completely specify the problem, we need to impose some boundary
22      conditions.  These are as follows:
23 
24        V(S, T) = max(E - S, 0)
25        V(0, t) = E for all 0 <= t <= T
26        V(s, t) = 0 for all 0 <= t <= T and s->infinity
27 
28      where T is the exercise time and E the strike price (price paid
29      for the contract).
30 
31      An explicit formula for the value of an European option can be
32      found.  See the references for examples.
33 
34    Background (American Options):
35      The American option is similar to its European counterpart.  The
36      difference is that the holder of the American option can exercise
37      their right to buy or sell the asset at any time prior to the
38      expiration.  This additional ability introduce a free boundary into
39      the Black-Scholes equation which can be modeled as a linear
40      complementarity problem.
41 
42        0 <= -(dV/dt + 0.5(sigma**2)(S**alpha)(d2V/dS2) + (r - D)S(dV/dS) - rV)
43          complements
44        V(S,T) >= max(E-S,0)
45 
46      where the variables are the same as before and we have the same boundary
47      conditions.
48 
49      There is not explicit formula for calculating the value of an American
50      option.  Therefore, we discretize the above problem and solve the
51      resulting linear complementarity problem.
52 
53      We will use backward differences for the time variables and central
54      differences for the space variables.  Crank-Nicholson averaging will
55      also be used in the discretization.  The algorithm used by the code
56      solves for V(S,t) for a fixed t and then uses this value in the
57      calculation of V(S,t - dt).  The method stops when V(S,0) has been
58      found.
59 
60    References:
61 + * - Huang and Pang, "Options Pricing and Linear Complementarity,"
62        Journal of Computational Finance, volume 2, number 3, 1998.
63 - * - Wilmott, "Derivatives: The Theory and Practice of Financial Engineering,"
64        John Wiley and Sons, New York, 1998.
65 */
66 
67 /*
68   Include "petsctao.h" so we can use TAO solvers.
69   Include "petscdmda.h" so that we can use distributed meshes (DMs) for managing
70   the parallel mesh.
71 */
72 
73 #include <petscdmda.h>
74 #include <petsctao.h>
75 
76 static char help[] = "This example demonstrates use of the TAO package to\n\
77 solve a linear complementarity problem for pricing American put options.\n\
78 The code uses backward differences in time and central differences in\n\
79 space.  The command line options are:\n\
80   -rate <r>, where <r> = interest rate\n\
81   -sigma <s>, where <s> = volatility of the underlying\n\
82   -alpha <a>, where <a> = elasticity of the underlying\n\
83   -delta <d>, where <d> = dividend rate\n\
84   -strike <e>, where <e> = strike price\n\
85   -expiry <t>, where <t> = the expiration date\n\
86   -mt <tg>, where <tg> = number of grid points in time\n\
87   -ms <sg>, where <sg> = number of grid points in space\n\
88   -es <se>, where <se> = ending point of the space discretization\n\n";
89 
90 /*
91   User-defined application context - contains data needed by the
92   application-provided call-back routines, FormFunction(), and FormJacobian().
93 */
94 
95 typedef struct {
96   PetscReal *Vt1; /* Value of the option at time T + dt */
97   PetscReal *c;   /* Constant -- (r - D)S */
98   PetscReal *d;   /* Constant -- -0.5(sigma**2)(S**alpha) */
99 
100   PetscReal rate;                /* Interest rate */
101   PetscReal sigma, alpha, delta; /* Underlying asset properties */
102   PetscReal strike, expiry;      /* Option contract properties */
103 
104   PetscReal es;     /* Finite value used for maximum asset value */
105   PetscReal ds, dt; /* Discretization properties */
106   PetscInt  ms, mt; /* Number of elements */
107 
108   DM dm;
109 } AppCtx;
110 
111 /* -------- User-defined Routines --------- */
112 
113 PetscErrorCode FormConstraints(Tao, Vec, Vec, void *);
114 PetscErrorCode FormJacobian(Tao, Vec, Mat, Mat, void *);
115 PetscErrorCode ComputeVariableBounds(Tao, Vec, Vec, void *);
116 
117 int main(int argc, char **argv)
118 {
119   Vec        x;    /* solution vector */
120   Vec        c;    /* Constraints function vector */
121   Mat        J;    /* jacobian matrix */
122   PetscBool  flg;  /* A return variable when checking for user options */
123   Tao        tao;  /* Tao solver context */
124   AppCtx     user; /* user-defined work context */
125   PetscInt   i, j;
126   PetscInt   xs, xm, gxs, gxm;
127   PetscReal  sval = 0;
128   PetscReal *x_array;
129   Vec        localX;
130 
131   /* Initialize PETSc, TAO */
132   PetscFunctionBeginUser;
133   PetscCall(PetscInitialize(&argc, &argv, NULL, help));
134 
135   /*
136      Initialize the user-defined application context with reasonable
137      values for the American option to price
138   */
139   user.rate   = 0.04;
140   user.sigma  = 0.40;
141   user.alpha  = 2.00;
142   user.delta  = 0.01;
143   user.strike = 10.0;
144   user.expiry = 1.0;
145   user.mt     = 10;
146   user.ms     = 150;
147   user.es     = 100.0;
148 
149   /* Read in alternative values for the American option to price */
150   PetscCall(PetscOptionsGetReal(NULL, NULL, "-alpha", &user.alpha, &flg));
151   PetscCall(PetscOptionsGetReal(NULL, NULL, "-delta", &user.delta, &flg));
152   PetscCall(PetscOptionsGetReal(NULL, NULL, "-es", &user.es, &flg));
153   PetscCall(PetscOptionsGetReal(NULL, NULL, "-expiry", &user.expiry, &flg));
154   PetscCall(PetscOptionsGetInt(NULL, NULL, "-ms", &user.ms, &flg));
155   PetscCall(PetscOptionsGetInt(NULL, NULL, "-mt", &user.mt, &flg));
156   PetscCall(PetscOptionsGetReal(NULL, NULL, "-rate", &user.rate, &flg));
157   PetscCall(PetscOptionsGetReal(NULL, NULL, "-sigma", &user.sigma, &flg));
158   PetscCall(PetscOptionsGetReal(NULL, NULL, "-strike", &user.strike, &flg));
159 
160   /* Check that the options set are allowable (needs to be done) */
161 
162   user.ms++;
163   PetscCall(DMDACreate1d(PETSC_COMM_WORLD, DM_BOUNDARY_NONE, user.ms, 1, 1, NULL, &user.dm));
164   PetscCall(DMSetFromOptions(user.dm));
165   PetscCall(DMSetUp(user.dm));
166   /* Create appropriate vectors and matrices */
167 
168   PetscCall(DMDAGetCorners(user.dm, &xs, NULL, NULL, &xm, NULL, NULL));
169   PetscCall(DMDAGetGhostCorners(user.dm, &gxs, NULL, NULL, &gxm, NULL, NULL));
170 
171   PetscCall(DMCreateGlobalVector(user.dm, &x));
172   /*
173      Finish filling in the user-defined context with the values for
174      dS, dt, and allocating space for the constants
175   */
176   user.ds = user.es / (user.ms - 1);
177   user.dt = user.expiry / user.mt;
178 
179   PetscCall(PetscMalloc1(gxm, &user.Vt1));
180   PetscCall(PetscMalloc1(gxm, &user.c));
181   PetscCall(PetscMalloc1(gxm, &user.d));
182 
183   /*
184      Calculate the values for the constant.  Vt1 begins with the ending
185      boundary condition.
186   */
187   for (i = 0; i < gxm; i++) {
188     sval        = (gxs + i) * user.ds;
189     user.Vt1[i] = PetscMax(user.strike - sval, 0);
190     user.c[i]   = (user.delta - user.rate) * sval;
191     user.d[i]   = -0.5 * user.sigma * user.sigma * PetscPowReal(sval, user.alpha);
192   }
193   if (gxs + gxm == user.ms) user.Vt1[gxm - 1] = 0;
194   PetscCall(VecDuplicate(x, &c));
195 
196   /*
197      Allocate the matrix used by TAO for the Jacobian.  Each row of
198      the Jacobian matrix will have at most three elements.
199   */
200   PetscCall(DMCreateMatrix(user.dm, &J));
201 
202   /* The TAO code begins here */
203 
204   /* Create TAO solver and set desired solution method  */
205   PetscCall(TaoCreate(PETSC_COMM_WORLD, &tao));
206   PetscCall(TaoSetType(tao, TAOSSILS));
207 
208   /* Set routines for constraints function and Jacobian evaluation */
209   PetscCall(TaoSetConstraintsRoutine(tao, c, FormConstraints, (void *)&user));
210   PetscCall(TaoSetJacobianRoutine(tao, J, J, FormJacobian, (void *)&user));
211 
212   /* Set the variable bounds */
213   PetscCall(TaoSetVariableBoundsRoutine(tao, ComputeVariableBounds, (void *)&user));
214 
215   /* Set initial solution guess */
216   PetscCall(VecGetArray(x, &x_array));
217   for (i = 0; i < xm; i++) x_array[i] = user.Vt1[i - gxs + xs];
218   PetscCall(VecRestoreArray(x, &x_array));
219   /* Set data structure */
220   PetscCall(TaoSetSolution(tao, x));
221 
222   /* Set routines for function and Jacobian evaluation */
223   PetscCall(TaoSetFromOptions(tao));
224 
225   /* Iteratively solve the linear complementarity problems  */
226   for (i = 1; i < user.mt; i++) {
227     /* Solve the current version */
228     PetscCall(TaoSolve(tao));
229 
230     /* Update Vt1 with the solution */
231     PetscCall(DMGetLocalVector(user.dm, &localX));
232     PetscCall(DMGlobalToLocalBegin(user.dm, x, INSERT_VALUES, localX));
233     PetscCall(DMGlobalToLocalEnd(user.dm, x, INSERT_VALUES, localX));
234     PetscCall(VecGetArray(localX, &x_array));
235     for (j = 0; j < gxm; j++) user.Vt1[j] = x_array[j];
236     PetscCall(VecRestoreArray(x, &x_array));
237     PetscCall(DMRestoreLocalVector(user.dm, &localX));
238   }
239 
240   /* Free TAO data structures */
241   PetscCall(TaoDestroy(&tao));
242 
243   /* Free PETSc data structures */
244   PetscCall(VecDestroy(&x));
245   PetscCall(VecDestroy(&c));
246   PetscCall(MatDestroy(&J));
247   PetscCall(DMDestroy(&user.dm));
248   /* Free user-defined workspace */
249   PetscCall(PetscFree(user.Vt1));
250   PetscCall(PetscFree(user.c));
251   PetscCall(PetscFree(user.d));
252 
253   PetscCall(PetscFinalize());
254   return 0;
255 }
256 
257 /* -------------------------------------------------------------------- */
258 PetscErrorCode ComputeVariableBounds(Tao tao, Vec xl, Vec xu, void *ctx)
259 {
260   AppCtx   *user = (AppCtx *)ctx;
261   PetscInt  i;
262   PetscInt  xs, xm;
263   PetscInt  ms   = user->ms;
264   PetscReal sval = 0.0, *xl_array, ub = PETSC_INFINITY;
265 
266   PetscFunctionBeginUser;
267   /* Set the variable bounds */
268   PetscCall(VecSet(xu, ub));
269   PetscCall(DMDAGetCorners(user->dm, &xs, NULL, NULL, &xm, NULL, NULL));
270 
271   PetscCall(VecGetArray(xl, &xl_array));
272   for (i = 0; i < xm; i++) {
273     sval        = (xs + i) * user->ds;
274     xl_array[i] = PetscMax(user->strike - sval, 0);
275   }
276   PetscCall(VecRestoreArray(xl, &xl_array));
277 
278   if (xs == 0) {
279     PetscCall(VecGetArray(xu, &xl_array));
280     xl_array[0] = PetscMax(user->strike, 0);
281     PetscCall(VecRestoreArray(xu, &xl_array));
282   }
283   if (xs + xm == ms) {
284     PetscCall(VecGetArray(xu, &xl_array));
285     xl_array[xm - 1] = 0;
286     PetscCall(VecRestoreArray(xu, &xl_array));
287   }
288   PetscFunctionReturn(PETSC_SUCCESS);
289 }
290 /* -------------------------------------------------------------------- */
291 
292 /*
293     FormFunction - Evaluates gradient of f.
294 
295     Input Parameters:
296 .   tao  - the Tao context
297 .   X    - input vector
298 .   ptr  - optional user-defined context, as set by TaoAppSetConstraintRoutine()
299 
300     Output Parameters:
301 .   F - vector containing the newly evaluated gradient
302 */
303 PetscErrorCode FormConstraints(Tao tao, Vec X, Vec F, void *ptr)
304 {
305   AppCtx    *user = (AppCtx *)ptr;
306   PetscReal *x, *f;
307   PetscReal *Vt1 = user->Vt1, *c = user->c, *d = user->d;
308   PetscReal  rate = user->rate;
309   PetscReal  dt = user->dt, ds = user->ds;
310   PetscInt   ms = user->ms;
311   PetscInt   i, xs, xm, gxs, gxm;
312   Vec        localX, localF;
313   PetscReal  zero = 0.0;
314 
315   PetscFunctionBeginUser;
316   PetscCall(DMGetLocalVector(user->dm, &localX));
317   PetscCall(DMGetLocalVector(user->dm, &localF));
318   PetscCall(DMGlobalToLocalBegin(user->dm, X, INSERT_VALUES, localX));
319   PetscCall(DMGlobalToLocalEnd(user->dm, X, INSERT_VALUES, localX));
320   PetscCall(DMDAGetCorners(user->dm, &xs, NULL, NULL, &xm, NULL, NULL));
321   PetscCall(DMDAGetGhostCorners(user->dm, &gxs, NULL, NULL, &gxm, NULL, NULL));
322   PetscCall(VecSet(F, zero));
323   /*
324      The problem size is smaller than the discretization because of the
325      two fixed elements (V(0,T) = E and V(Send,T) = 0.
326   */
327 
328   /* Get pointers to the vector data */
329   PetscCall(VecGetArray(localX, &x));
330   PetscCall(VecGetArray(localF, &f));
331 
332   /* Left Boundary */
333   if (gxs == 0) {
334     f[0] = x[0] - user->strike;
335   } else {
336     f[0] = 0;
337   }
338 
339   /* All points in the interior */
340   /*  for (i=gxs+1;i<gxm-1;i++) { */
341   for (i = 1; i < gxm - 1; i++) {
342     f[i] = (1.0 / dt + rate) * x[i] - Vt1[i] / dt + (c[i] / (4 * ds)) * (x[i + 1] - x[i - 1] + Vt1[i + 1] - Vt1[i - 1]) + (d[i] / (2 * ds * ds)) * (x[i + 1] - 2 * x[i] + x[i - 1] + Vt1[i + 1] - 2 * Vt1[i] + Vt1[i - 1]);
343   }
344 
345   /* Right boundary */
346   if (gxs + gxm == ms) {
347     f[xm - 1] = x[gxm - 1];
348   } else {
349     f[xm - 1] = 0;
350   }
351 
352   /* Restore vectors */
353   PetscCall(VecRestoreArray(localX, &x));
354   PetscCall(VecRestoreArray(localF, &f));
355   PetscCall(DMLocalToGlobalBegin(user->dm, localF, INSERT_VALUES, F));
356   PetscCall(DMLocalToGlobalEnd(user->dm, localF, INSERT_VALUES, F));
357   PetscCall(DMRestoreLocalVector(user->dm, &localX));
358   PetscCall(DMRestoreLocalVector(user->dm, &localF));
359   PetscCall(PetscLogFlops(24.0 * (gxm - 2)));
360   /*
361   info=VecView(F,PETSC_VIEWER_STDOUT_WORLD);
362   */
363   PetscFunctionReturn(PETSC_SUCCESS);
364 }
365 
366 /* ------------------------------------------------------------------- */
367 /*
368    FormJacobian - Evaluates Jacobian matrix.
369 
370    Input Parameters:
371 .  tao  - the Tao context
372 .  X    - input vector
373 .  ptr  - optional user-defined context, as set by TaoSetJacobian()
374 
375    Output Parameters:
376 .  J    - Jacobian matrix
377 */
378 PetscErrorCode FormJacobian(Tao tao, Vec X, Mat J, Mat tJPre, void *ptr)
379 {
380   AppCtx    *user = (AppCtx *)ptr;
381   PetscReal *c = user->c, *d = user->d;
382   PetscReal  rate = user->rate;
383   PetscReal  dt = user->dt, ds = user->ds;
384   PetscInt   ms = user->ms;
385   PetscReal  val[3];
386   PetscInt   col[3];
387   PetscInt   i;
388   PetscInt   gxs, gxm;
389   PetscBool  assembled;
390 
391   PetscFunctionBeginUser;
392   /* Set various matrix options */
393   PetscCall(MatSetOption(J, MAT_IGNORE_OFF_PROC_ENTRIES, PETSC_TRUE));
394   PetscCall(MatAssembled(J, &assembled));
395   if (assembled) PetscCall(MatZeroEntries(J));
396 
397   PetscCall(DMDAGetGhostCorners(user->dm, &gxs, NULL, NULL, &gxm, NULL, NULL));
398 
399   if (gxs == 0) {
400     i      = 0;
401     col[0] = 0;
402     val[0] = 1.0;
403     PetscCall(MatSetValues(J, 1, &i, 1, col, val, INSERT_VALUES));
404   }
405   for (i = 1; i < gxm - 1; i++) {
406     col[0] = gxs + i - 1;
407     col[1] = gxs + i;
408     col[2] = gxs + i + 1;
409     val[0] = -c[i] / (4 * ds) + d[i] / (2 * ds * ds);
410     val[1] = 1.0 / dt + rate - d[i] / (ds * ds);
411     val[2] = c[i] / (4 * ds) + d[i] / (2 * ds * ds);
412     PetscCall(MatSetValues(J, 1, &col[1], 3, col, val, INSERT_VALUES));
413   }
414   if (gxs + gxm == ms) {
415     i      = ms - 1;
416     col[0] = i;
417     val[0] = 1.0;
418     PetscCall(MatSetValues(J, 1, &i, 1, col, val, INSERT_VALUES));
419   }
420 
421   /* Assemble the Jacobian matrix */
422   PetscCall(MatAssemblyBegin(J, MAT_FINAL_ASSEMBLY));
423   PetscCall(MatAssemblyEnd(J, MAT_FINAL_ASSEMBLY));
424   PetscCall(PetscLogFlops(18.0 * (gxm) + 5));
425   PetscFunctionReturn(PETSC_SUCCESS);
426 }
427 
428 /*TEST
429 
430    build:
431       requires: !complex
432 
433    test:
434       args: -tao_monitor -tao_type ssils -tao_gttol 1.e-5
435       requires: !single
436 
437    test:
438       suffix: 2
439       args: -tao_monitor -tao_type ssfls -tao_max_it 10 -tao_gttol 1.e-5
440       requires: !single
441 
442    test:
443       suffix: 3
444       args: -tao_monitor -tao_type asils -tao_subset_type subvec -tao_gttol 1.e-5
445       requires: !single
446 
447    test:
448       suffix: 4
449       args: -tao_monitor -tao_type asils -tao_subset_type mask -tao_gttol 1.e-5
450       requires: !single
451 
452    test:
453       suffix: 5
454       args: -tao_monitor -tao_type asils -tao_subset_type matrixfree -pc_type jacobi -tao_max_it 6 -tao_gttol 1.e-5
455       requires: !single
456 
457    test:
458       suffix: 6
459       args: -tao_monitor -tao_type asfls -tao_subset_type subvec -tao_max_it 10 -tao_gttol 1.e-5
460       requires: !single
461 
462    test:
463       suffix: 7
464       args: -tao_monitor -tao_type asfls -tao_subset_type mask -tao_max_it 10 -tao_gttol 1.e-5
465       requires: !single
466 
467 TEST*/
468