qufin 1.1.2

The open-source framework for quantum-enhanced quantitative finance.
Research-grade algorithms. Production-grade engineering. Honest benchmarks.

_{159 modules · 14 subpackages · 11 backends · 8 error-mitigation strategies · 6 QAE algorithms}

Installation · Quickstart · Capabilities · Backends · Benchmarks · Docs

Why qufin

Most quantum finance libraries are toy demos or locked to one framework. qufin is different.

Every quantum algorithm ships alongside the best classical solver for the same problem. Results are compared head-to-head on identical inputs, with identical metrics, on standardized benchmark sets.

• Backend-agnostic — Write once, run on 11 backends (Aer, IBM, PennyLane, Cirq, Braket, CUDA-Q, D-Wave, IonQ, Quantinuum, noisy sim, mock)
• Mathematically correct — Grover global phase, IQAE multi-branch theta, canonical QPE. Details matter for derivative pricing.
• Production patterns — Typed configs, reproducibility manifests, noise-aware simulation, 8 error-mitigation strategies, finance-optimized transpilation.

Project status: 1.x, actively developed. The classical quant core (Black-Scholes/Greeks, Monte Carlo, VaR/CVaR, mean-variance, HRP, GARCH, backtesting) is well-tested and matches textbook values. Several quantum paths are research-stage and evolving, and may change between minor versions. The Quickstart below is exercised end-to-end by examples/quickstart.py.

Installation

pip install qufin

Requires Python 3.10+

Quickstart

Option pricing: classical vs. quantum

import numpy as np
from qiskit.circuit import QuantumCircuit

from qufin.options.classical.black_scholes import call_price
from qufin.options.amplitude_estimation.estimation_problem import EstimationProblem
from qufin.options.amplitude_estimation.iqae import (
    IterativeAmplitudeEstimation, IQAEConfig,
)
from qufin.backends.qiskit_backend import QiskitAerBackend

# Classical: Black-Scholes closed form
classical = call_price(s=100, k=105, r=0.05, sigma=0.2, T=1.0)

# Quantum: Iterative Quantum Amplitude Estimation of a = sin^2(theta)
theta = np.pi / 5
oracle = QuantumCircuit(1)
oracle.ry(2 * theta, 0)  # A|0> = cos(theta)|0> + sin(theta)|1>
problem = EstimationProblem(state_preparation=oracle, objective_qubits=[0], n_qubits=1)

backend = QiskitAerBackend(method="automatic", seed=42)
result = IterativeAmplitudeEstimation(
    problem, IQAEConfig(epsilon_target=0.01, shots_per_round=2048), backend,
).estimate()

print(f"Black-Scholes call: {classical:.4f}")
print(f"IQAE estimate:      {result.estimate:.4f}  (true {np.sin(theta) ** 2:.4f})")

Portfolio optimization with QAOA

import numpy as np
from qufin.portfolio.qubo import PortfolioQUBO
from qufin.portfolio.optimizers.qaoa import QAOAPortfolio, QAOAConfig
from qufin.backends.qiskit_backend import QiskitAerBackend

rng = np.random.default_rng(42)
n_assets = 6
mu = rng.uniform(0.05, 0.15, n_assets)
factor = rng.standard_normal((n_assets, n_assets))
cov = (factor @ factor.T) / n_assets

qubo = PortfolioQUBO(mu=mu, cov=cov, gamma=0.5, cardinality=3, encoding="one_hot")
config = QAOAConfig(p=2, mixer="xy_ring", cardinality=3, shots=2048, seed=42)

result = QAOAPortfolio(qubo, config, QiskitAerBackend(seed=42)).run()
print(f"Selected (bitstring): {result.best_bitstring}")
print(f"Objective:            {result.best_objective:.6f}")

from qufin.data.synthetic import gbm_paths, heston_paths

# GBM: shape (n_paths, n_steps + 1)
gbm = gbm_paths(s0=100, mu=0.08, sigma=0.2, T=1.0,
                n_steps=252, n_paths=10_000, seed=42)

# Heston returns (prices, variances), each (n_paths, n_steps + 1)
prices, variances = heston_paths(
    s0=100, v0=0.04, kappa=2.0, theta=0.04, xi=0.3, rho=-0.7,
    mu=0.08, T=1.0, n_steps=252, n_paths=10_000, seed=42,
)

import numpy as np
from qufin.backtesting.engine import BacktestEngine

returns = np.random.default_rng(0).normal(0.0004, 0.01, size=(800, 5))

def equal_weight(mu, cov):           # strategy: (mu, cov) -> weights
    return np.ones(len(mu)) / len(mu)

engine = BacktestEngine(returns, train_window=252, test_window=21)
result = engine.run(equal_weight, strategy_name="equal_weight")
print(f"Sharpe: {result.summary.sharpe_ratio:.2f}")

from qufin.backends.auto_select import auto_select_backend

backend = auto_select_backend(circuit)  # GPU -> Aer -> Mock

Capabilities

Portfolio Optimization

• Classical: Mean-Variance, Black-Litterman, Risk Parity, HRP, Multi-Period, ADMM, Factor Models
• Quantum: QAOA (4 mixers), VQE, Warm-Start, Szegedy Walk, Robust CVaR QUBO, Grover Search, Quantum IPM, Simulated Quantum Annealing
• Annealing: D-Wave QUBO solver (Pegasus/Zephyr/Chimera)
• Constraints: Cardinality, sector, turnover, transaction cost, budget

Option Pricing

• Classical: Black-Scholes (full Greeks), Monte Carlo, CRR Binomial, LSM American, Implied Vol (SABR/SVI)
• Quantum (6 QAE algorithms): Canonical QAE, IQAE, MLAE, FQAE, MRQAE, QMC (Montanaro)
• Option wrappers: European, Asian, American, Path-Dependent, Multi-Asset, QSP pricing
• Exotics: Bermudan, lookback, cliquet, autocallable, basket

Risk Management

• Classical: VaR (historical/parametric/MC), CVaR, stress testing, CVA/DVA, tail risk (EVaR, spectral)
• Quantum: Quantum VaR, Egger credit-risk, Quantum Stress Testing, HHL Linear Systems, Quantum Entropy
• Credit: Gaussian copula, NIG copula

Hedging

• Classical: Delta hedging, deep hedging (PyTorch)
• Quantum: Quantum deep hedging, RL-quantum hedging, PPO with VQC policy

Machine Learning

• Classical: Standard classifiers, PCA anomaly detection
• Quantum: Kernel methods, VQC, qGAN, HQGAN, reservoir computing, Boltzmann machine, credit scoring, transfer learning, quantum autoencoder

Error Mitigation (8 strategies)

• Level 1: Readout calibration, TREX
• Level 2: ZNE (Richardson), Dynamical Decoupling (XY4/CPMG/Uhrig)
• Level 3: PEC, CDR, M3 (matrix-free)
• Adaptive: Noise-aware variational optimization

Data & Infrastructure

• Market Data: Yahoo Finance, FRED, Bloomberg, Refinitiv, CoinGecko crypto
• Streaming: Alpaca, Polygon, IEX WebSocket
• Synthetic: GBM, Heston, Merton jump-diffusion
• Warehouse: Parquet storage, PyArrow, auto-compaction
• Backtesting: Walk-forward engine, permutation test, CSCV overfitting detection

Enterprise

• REST API: FastAPI (optimize, price, risk)
• Job Queue: Celery + Redis with priority routing
• Caching: SQLite/Redis with TTL
• Deployment: Docker, Kubernetes Helm chart
• Compliance: Audit trail, SR 11-7/SS1/23, Shapley explainability

Backends

All quantum algorithms accept any Backend implementation. Swap without changing algorithm code.

from qufin.backends.auto_select import auto_select_backend
backend = auto_select_backend(circuit)

Benchmarks

Standardized suites for honest quantum-vs-classical comparison.

from qufin.benchmarks.runner import BenchmarkRunner, SolverEntry
from qufin.benchmarks.problems import portfolio_small

runner = BenchmarkRunner()
runner.register(SolverEntry(name="mean_variance", family="classical",
                            solve_fn=my_mean_variance_fn))
runner.register(SolverEntry(name="qaoa-p2", family="quantum",
                            solve_fn=my_qaoa_fn))

rows = runner.run_problem(portfolio_small())   # 15-asset benchmark
for row in rows:
    print(row.solver_name, row.family, row.objective, row.wall_seconds)

Each registered solver is a callable problem -> {"objective": ..., ...}; the runner returns a list of BenchmarkRow records.

• Problem sets: 15, 25, 50 asset portfolios (portfolio_small/medium/large)
• Metrics: objective, relative error vs. reference, wall time, circuit depth
• Reproducibility: Hardware, versions, seeds manifest
• Transpiler: QUBO-aware ZZ optimization via Qiskit optimization_level=3 (gate cancellation, commutation, template matching). CNOT reduction depends on circuit structure; dense QAOA cost layers see little reduction (see the measured benchmark in docs/).

Architecture

src/qufin/
  backends/           11 backends + error mitigation + transpiler
  options/
    classical/        Black-Scholes, binomial, Monte Carlo
    amplitude_estimation/  QAE, IQAE, MLAE, FQAE, Asian, QMC, QSP
  portfolio/
    classical/        MVO, Black-Litterman, HRP, Risk Parity
    optimizers/       QAOA, VQE, warm-start, annealing, Grover, IPM
  risk/               VaR, CVaR, stress, entropy, tail risk, HHL
    credit/           Egger, Gaussian copula, NIG copula
  hedging/            Delta, deep, quantum RL (PPO + VQC)
  ml/                 Kernels, VQC, qGAN, Boltzmann, autoencoder
  derivatives/        Bermudan, lookback, cliquet, autocallable
  data/               Yahoo, FRED, Bloomberg, Refinitiv, crypto
  backtesting/        Walk-forward, permutation test, CSCV
  benchmarks/         Problem sets, runner, resource estimation
  viz/                Plotly widgets, Dash dashboard
  api/                FastAPI + Celery + Redis cache
  compliance/         Audit trail, validation, explainability
  utils/              Circuit cache, parallel exec, sparse Pauli
  cli.py              CLI: optimize, price, risk, benchmark
  plugins.py          Entry-point plugin discovery

Testing

pytest                             # Full suite (2,507 tests collected)
pytest tests/unit/                 # Unit tests (fast)
pytest -m "not slow"               # Skip slow tests
pytest -m "not hardware"           # Skip hardware tests

Contributing

Contributions welcome. See CONTRIBUTING.md.

License

Apache 2.0. See LICENSE.

Citation

@software{qufin,
  author  = {Adarsh Keshri},
  title   = {qufin: Quantum Algorithms for Quant Finance},
  year    = {2026},
  version = {0.1.dev},
  url     = {https://github.com/anonymousAAK/qufin},
  license = {Apache-2.0}
}