Applied Quantum Computing in Finance: Case Study

High‑Frequency Trading - Quantum Boltzmann Machines for Volatility Spike Prediction

1 | Problem
HFT desk predicts < 30 s volatility spikes on NASDAQ futures. GPU LSTM pipeline good at macro regime shifts but struggles with sudden micro‑flashes caused by cross‑venue order imbalances.

2 | Hybrid Tactic
Replace final LSTM attention head with Quantum Boltzmann Machine (QBM) sampler executed on Rigetti Aspen‑M 128Q to approximate partition function of limit‑order‑book (LOB) state.

3 | Flatcar & K8s
Edge‑4 H200 nodes for LSTM; QPU gateway as CPU Deployment with nodeSelector role=lowlatency.
PodDisruptionBudget guarantees two gateway replicas.

Code repository for doing hands on exercise Indicative Project Source code

4 | Python Quantum Compute library Qiskit Code

from qiskit_machine_learning.neural_networks import SamplerQNN
from qiskit_machine_learning.connectors import TorchConnector
from qiskit import QuantumCircuit
import torch, torch.nn as nn

qc = QuantumCircuit(8, name='qbm')
for q in range(8):
    qc.h(q)
qc.barrier()
# pairwise ZZ entanglement
for i in range(7):
    qc.cx(i, i+1)
    qc.rz(0.5, i+1)
    qc.cx(i, i+1)

qnn = SamplerQNN(circuit=qc, sampling shots=1024)
model = TorchConnector(qnn)

class HybridHead(nn.Module):
    def __init__(self, in_dim):
        super().__init__()
        self.linear = nn.Linear(in_dim, 8)

    def forward(self, x):
        logits = self.linear(x)
        return model(logits)

5 | Gateway Enhancements

• gRPC method RunQbmSampler() multiplexes over same port.

• Redis result key = SHA‑256(input_vector + weights).

6 | Latency Budget
Total inference P99 = 6.3 ms.
QPU round‑trip 4.1 ms measured NY4 Datacenter (Secaucus) ↔ AWS us‑east‑1 ↔ Rigetti QCS us‑west‑1.
Still inside 10 ms arbitrage window.

7 | Profit Impact
Back‑test on 30 days tick data → Sharpe + 0.42, daily pnl + 3.7 %.
Live two‑week pilot confirmed + 2.9 % pnl vs. baseline, no major slippage increase.

8 | Risk & Compliance
Added “quantum on/off” feature flag; fallback to classical attention head if QPU errors > 1 %.
COSIGN signed Qiskit container; SBOM stored for audit.