def F_plus_capital():
indicator = np.where(x_space <= 0, 1, 0)
trimmed_x_space = indicator * x_space # чтобы при "сильно положительных" x не росла экспонента
integral = ffts.make_rad_ifft(integrand_plus(u_space, gamma, sigma),
dx)
exponent = exp(-trimmed_x_space * omega_minus)
return indicator * exponent * integral
def F_minus_capital():
indicator = np.where(x_space >= 0, 1, 0)
trimmed_x_space = indicator * x_space # чтобы при "сильно отрицательных" x не росла экспонента
integral = ffts.make_rad_ifft(
integrand_minus(u_space, gamma, sigma, pois_lambda, pois_delta,
pois_mu), dx)
exponent = exp(-trimmed_x_space * omega_plus)
return indicator * exponent * integral
def evaluate_option_by_wh(T, H_original, K_original, r_premia, V0, kappa,
theta, sigma, rho, N, M):
if 2 * kappa * theta < sigma**2:
print(
"Warning, Novikov condition is not satisfied, the volatility values could be negative"
)
r = log(r_premia / 100 + 1)
omega = sigma
# time-space domain construction
x_space = ffts.make_fft_spaces(M, dx)[0] # prices array
first_step_of_return = [elem + V0 * rho / sigma for elem in x_space]
original_prices_array = H_original * exp(first_step_of_return)
# array_to_csv_file(original_prices_array, "../output/routine/price_line_original.csv")
# time discretization
delta_t = T / N
# making volatilily tree
markov_chain = build_volatility_tree(T, V0, kappa, theta, omega, N)
V = markov_chain[0]
pu_f = markov_chain[1]
pd_f = markov_chain[2]
f_up = markov_chain[3]
f_down = markov_chain[4]
rho_hat = sqrt(1 - rho**2)
q = 1.0 / delta_t + r
factor = (q * delta_t)**(-1)
F_n_plus_1 = zeros((len(x_space), len(V[N])), dtype=complex)
F_n = zeros((len(x_space), len(V[N])), dtype=complex)
for j in range(len(x_space)):
for k in range(len(V[N])):
F_n_plus_1[j,
k] = array(G(H_original * exp(x_space[j]), K_original))
# the global cycle starts here. It iterates over the volatility tree we just constructed, and goes backwards in time
# starting from n-1 position
# print("Main cycle entered")
# when the variance is less than that, is is reasonable to assume it to be zero, which leads to simpler calculations
treshold = 1e-6
discount_factor = exp(r * delta_t)
for n in range(len(V[N]) - 2, -1, -1):
print(str(n) + " of " + str(len(V[N]) - 2))
with profiler():
for k in range(n + 1):
# to calculate the binomial expectation one should use Antonino's matrices f_up and f_down
# the meaning of the containing integers are as follows - after (n,k) you will be in
# either (n+1, k + f_up) or (n+1, k - f_down). We use k_u and k_d shorthands, respectively
k_u = k + int(f_up[n][k])
k_d = k + int(f_down[n][k])
# initial condition of a step
f_n_plus_1_k_u = array(
[F_n_plus_1[j][k_u] for j in range(len(x_space))])
f_n_plus_1_k_d = array(
[F_n_plus_1[j][k_d] for j in range(len(x_space))])
H_N_k = -(rho / sigma) * V[n, k] # modified barrier
local_domain = array(
[x_space[j] + H_N_k for j in range(len(x_space))])
if V[n, k] >= treshold:
# set up variance-dependent parameters for a given step
sigma_local = rho_hat * sqrt(V[n, k])
gamma = r - 0.5 * V[n, k] - rho / sigma * kappa * (
theta - V[n, k]) # also local
# beta_plus and beta_minus
# beta_minus = - (gamma + sqrt(gamma**2 + 2*sigma_local**2 * q))/sigma_local**2
# beta_plus = - (gamma - sqrt(gamma**2 + 2*sigma_local**2 * q))/sigma_local**2
# factor functions
# phi_plus_array = array([beta_plus/(beta_plus - i*2*pi*xi) for xi in xi_space])
# phi_minus_array = array([-beta_minus/(-beta_minus + i*2*pi*xi) for xi in xi_space])
phi_plus_array = mex.make_phi_minus(
M, dx, omega_plus, gamma, sigma_local, q)
phi_minus_array = mex.make_phi_plus(
M, dx, omega_minus, gamma, sigma_local, q)
# factorization calculation
f_n_k_u = factor * \
ffts.make_rad_ifft(phi_minus_array *
ffts.make_rad_fft(
indicator(
ffts.make_rad_ifft(phi_plus_array * ffts.make_rad_fft(f_n_plus_1_k_u, dx), dx),
local_domain, H_N_k), dx), dx)
f_n_k_d = factor * \
ffts.make_rad_ifft(phi_minus_array *
ffts.make_rad_fft(
indicator(
ffts.make_rad_ifft(phi_plus_array * ffts.make_rad_fft(f_n_plus_1_k_d, dx), dx),
local_domain, H_N_k), dx), dx)
elif V[n, k] < treshold:
f_n_plus_1_k_u = [
F_n_plus_1[j][k_u] for j in range(len(x_space))
]
f_n_k_u = discount_factor * f_n_plus_1_k_u
f_n_plus_1_k_d = [
F_n_plus_1[j][k_d] for j in range(len(x_space))
]
f_n_k_d = discount_factor * f_n_plus_1_k_d
f_n_k = f_n_k_u * pu_f[n, k] + f_n_k_d * pd_f[n, k]
for j in range(len(f_n_k)):
# here we try some cutdown magic. The procedure without it returns great bubbles to the right
# from the strike. And the more L the greater this bubble grows.
# what we are going to do there is to try to cut off all the values on prices greater than, say,
# 4 times bigger then the strike
# we use S>4K and, therefore, y > ln(4K/H) + (pho/sigma)*V inequality to do this
if local_domain[j] < log(3.5 * K_original / H_original +
(rho / sigma) * V[n][k]):
F_n[j][k] = f_n_k[j]
else:
F_n[j][k] = complex(0)
# plt.plot(original_prices_array, f_n_plus_1_k_u)
# plt.show()
# for j in range(len(y)):
# tree_to_csv_file(y[j], "../output/routine/price_slices/Y" + str(original_prices_array[j]) + ".csv")
# for j in range(len(F)):
# tree_to_csv_file(F[j], "../output/routine/answers/F" + str(original_prices_array[j]) + ".csv")
answer_total = open("../output/routine/answer_cumul.csv", "w")
answers_list = array([F_n[j][0] for j in range(len(x_space))])
for elem in list(zip(original_prices_array, answers_list)):
answer_total.write(str(elem[0]) + ',')
answer_total.write(str(elem[1].real) + ',')
# answer_total.write(str(elem[1].imag) + ',')
answer_total.write('\n')
# for j in range(len(F)):
# tree_to_csv_file(F[j], "../output/routine/answers/F" + str(original_prices_array[j]) + ".csv")
plt.plot(
original_prices_array[(original_prices_array > 75)
& (original_prices_array < 200)],
answers_list[(original_prices_array > 75)
& (original_prices_array < 200)])
plt.savefig("../output/figure.png")
plt.show()
plt.close()