English
Languages
English
Japanese
German
Korean
Portuguese, Brazilian
Shortcuts

Note

This page was generated from tutorials/finance/10_qgan_option_pricing.ipynb.

Run interactively in the IBM Quantum lab.

Option Pricing with qGANs

Introduction

In this notebook, we discuss how a Quantum Machine Learning Algorithm, namely a quantum Generative Adversarial Network (qGAN), can facilitate the pricing of a European call option. More specifically, a qGAN can be trained such that a quantum circuit models the spot price of an asset underlying a European call option. The resulting model can then be integrated into a Quantum Amplitude Estimation based algorithm to evaluate the expected payoff - see European Call Option Pricing. For further details on learning and loading random distributions by training a qGAN please refer to Quantum Generative Adversarial Networks for Learning and Loading Random Distributions. Zoufal, Lucchi, Woerner. 2019.

[1]:
import matplotlib.pyplot as plt
import numpy as np

from qiskit import Aer, QuantumRegister, QuantumCircuit
from qiskit.circuit import ParameterVector
from qiskit.circuit.library import TwoLocal, NormalDistribution
from qiskit.quantum_info import Statevector

from qiskit.aqua import aqua_globals, QuantumInstance
from qiskit.aqua.algorithms import IterativeAmplitudeEstimation
from qiskit.finance.applications import EuropeanCallExpectedValue

Uncertainty Model

The Black-Scholes model assumes that the spot price at maturity \(S_T\) for a European call option is log-normally distributed. Thus, we can train a qGAN on samples from a log-normal distribution and use the result as an uncertainty model underlying the option. In the following, we construct a quantum circuit that loads the uncertainty model. The circuit output reads

\[\big| g_{\theta}\rangle = \sum_{j=0}^{2^n-1}\sqrt{p_{\theta}^{j}} \big| j \rangle ,\]

where the probabilities \(p_{\theta}^{j}\), for \(j\in \left\{0, \ldots, {2^n-1} \right\}\), represent a model of the target distribution.

[2]:
# Set upper and lower data values
bounds = np.array([0.,7.])
# Set number of qubits used in the uncertainty model
num_qubits = 3

# Load the trained circuit parameters
g_params = [0.29399714, 0.38853322, 0.9557694, 0.07245791, 6.02626428, 0.13537225]

# Set an initial state for the generator circuit
init_dist = NormalDistribution(num_qubits, mu=1., sigma=1., bounds=bounds)

# construct the variational form
var_form = TwoLocal(num_qubits, 'ry', 'cz', entanglement='circular', reps=1)

# keep a list of the parameters so we can associate them to the list of numerical values
# (otherwise we need a dictionary)
theta = var_form.ordered_parameters

# compose the generator circuit, this is the circuit loading the uncertainty model
g_circuit = init_dist.compose(var_form)

Evaluate Expected Payoff

Now, the trained uncertainty model can be used to evaluate the expectation value of the option’s payoff function with Quantum Amplitude Estimation.

[3]:
# set the strike price (should be within the low and the high value of the uncertainty)
strike_price = 2

# set the approximation scaling for the payoff function
c_approx = 0.25

# construct circuit for payoff function
european_call_objective = EuropeanCallExpectedValue(
    num_qubits,
    strike_price=strike_price,
    rescaling_factor=c_approx,
    bounds=bounds
)

Plot the probability distribution

Next, we plot the trained probability distribution and, for reasons of comparison, also the target probability distribution.

[4]:
# Evaluate trained probability distribution
values = [bounds[0] + (bounds[1] - bounds[0]) * x / (2 ** num_qubits - 1) for x in range(2**num_qubits)]
uncertainty_model = g_circuit.assign_parameters(dict(zip(theta, g_params)))
amplitudes = Statevector.from_instruction(uncertainty_model).data

x = np.array(values)
y = np.abs(amplitudes) ** 2

# Sample from target probability distribution
N = 100000
log_normal = np.random.lognormal(mean=1, sigma=1, size=N)
log_normal = np.round(log_normal)
log_normal = log_normal[log_normal <= 7]
log_normal_samples = []
for i in range(8):
    log_normal_samples += [np.sum(log_normal==i)]
log_normal_samples = np.array(log_normal_samples / sum(log_normal_samples))

# Plot distributions
plt.bar(x, y, width=0.2, label='trained distribution', color='royalblue')
plt.xticks(x, size=15, rotation=90)
plt.yticks(size=15)
plt.grid()
plt.xlabel('Spot Price at Maturity $S_T$ (\$)', size=15)
plt.ylabel('Probability ($\%$)', size=15)
plt.plot(log_normal_samples,'-o', color ='deepskyblue', label='target distribution', linewidth=4, markersize=12)
plt.legend(loc='best')
plt.show()
../../_images/tutorials_finance_10_qgan_option_pricing_7_0.png

Evaluate Expected Payoff

Now, the trained uncertainty model can be used to evaluate the expectation value of the option’s payoff function analytically and with Quantum Amplitude Estimation.

[5]:
# Evaluate payoff for different distributions
payoff = np.array([0,0,0,1,2,3,4,5])
ep = np.dot(log_normal_samples, payoff)
print("Analytically calculated expected payoff w.r.t. the target distribution:  %.4f" % ep)
ep_trained = np.dot(y, payoff)
print("Analytically calculated expected payoff w.r.t. the trained distribution: %.4f" % ep_trained)

# Plot exact payoff function (evaluated on the grid of the trained uncertainty model)
x = np.array(values)
y_strike = np.maximum(0, x - strike_price)
plt.plot(x, y_strike, 'ro-')
plt.grid()
plt.title('Payoff Function', size=15)
plt.xlabel('Spot Price', size=15)
plt.ylabel('Payoff', size=15)
plt.xticks(x, size=15, rotation=90)
plt.yticks(size=15)
plt.show()
Analytically calculated expected payoff w.r.t. the target distribution:  1.0666
Analytically calculated expected payoff w.r.t. the trained distribution: 0.9805
../../_images/tutorials_finance_10_qgan_option_pricing_9_1.png
[6]:
# construct A operator for QAE
european_call = european_call_objective.compose(uncertainty_model, front=True)
[7]:
# set target precision and confidence level
epsilon = 0.01
alpha = 0.05

# construct amplitude estimation
ae = IterativeAmplitudeEstimation(epsilon=epsilon, alpha=alpha,
                                  state_preparation=european_call,
                                  objective_qubits=[num_qubits],
                                  post_processing=european_call_objective.post_processing)
[8]:
result = ae.run(quantum_instance=Aer.get_backend('qasm_simulator'), shots=100)
[9]:
conf_int = np.array(result['confidence_interval'])
print('Exact value:        \t%.4f' % ep_trained)
print('Estimated value:    \t%.4f' % (result['estimation']))
print('Confidence interval:\t[%.4f, %.4f]' % tuple(conf_int))
Exact value:            0.9805
Estimated value:        1.0353
Confidence interval:    [0.9209, 1.1497]
[10]:
import qiskit.tools.jupyter
%qiskit_version_table
%qiskit_copyright

Version Information

Qiskit SoftwareVersion
Qiskit0.24.0
Terra0.16.4
Aer0.7.6
Ignis0.5.2
Aqua0.8.2
IBM Q Provider0.12.1
System information
Python3.8.8 (default, Feb 19 2021, 19:42:00) [GCC 9.3.0]
OSLinux
CPUs2
Memory (Gb)6.7913818359375
Thu Mar 04 21:21:43 2021 UTC

This code is a part of Qiskit

© Copyright IBM 2017, 2021.

This code is licensed under the Apache License, Version 2.0. You may
obtain a copy of this license in the LICENSE.txt file in the root directory
of this source tree or at http://www.apache.org/licenses/LICENSE-2.0.

Any modifications or derivative works of this code must retain this
copyright notice, and modified files need to carry a notice indicating
that they have been altered from the originals.