Genetically Engineered Quantum Circuits for Financial Market Indicators

TL;DR

The study tackles efficient encoding of financial data onto quantum hardware to enable QSVD-based SVD entropy analysis on NISQ devices. It introduces GASP as a data-loading initializer for VQSVD and benchmarks it against qGAN and AAE using stock-returns-derived correlation matrices, demonstrating that high-fidelity yet shallow encodings yield accurate SVD entropy $S = -\sum_k \lambda_k \ln(\lambda_k)$. The findings indicate diminishing returns beyond roughly $90$–$95\%$ fidelity, highlighting a practical balance between accuracy and circuit depth for quantum-finance algorithms. This work provides a path toward near-term quantum advantages in finance by enabling efficient, hardware-feasible quantum indicators for risk management and market forecasting.

Abstract

Quantum computing holds immense potential for transforming financial analysis and decision-making. Realising this potential necessitates the efficient encoding and processing of financial data on quantum computers. In this study, we propose using the GASP (Genetic Algorithm for State Preparation) framework to optimise the encoding of stock price data into quantum states and show it can enhance both the fidelity and efficiency of the encoding process. We demonstrate the efficacy of our approach by encoding stock price data onto both a simulated and real quantum computer to calculate the Singular Value Decomposition (SVD) entropy. Our results show improvements in fidelity and the potential for more precise financial analysis. This research provides insights into the applicability of GASP for the efficient encoding of real-world data, specifically stock price data, which is crucial for quantum advantage on noisy intermediate-scale quantum (NISQ) era quantum computers.

Figures (8)

• Figure 1: (a) Shows monthly stock opening price data for XOM, WMT, PG, and MSFT, for the period between March 2008 to March 2009, retrieved from Yahoo Finance. (b) Shows the calculated logarithmic rates of return $r_{nt} = \ln(s_{n, t}) - \ln(s_{n, t - 1})$ for each $n$ stock at time $t$.
• Figure 2: Shows the generated correlation matrices for XOM, WMT, PG, and MSFT, for the period between March 2008 to March 2009, retrieved from Yahoo Finance, $T=5$, $C_{nm} = \sum_{t=1}^Ta_{nt}a_{mt}$.
• Figure 3: Shows the generated statevectors from the correlation matrices of XOM, WMT, PG, and MSFT, for the period between March 2008 to March 2009, retrieved from Yahoo Finance, $T=5$, $C_{nm} = \sum_{t=1}^Ta_{nt}a_{mt}$.
• Figure 4: QSVD Experimental circuit structure. The GASP block represents the state encoding circuit produced by GASP. The QSVD section of the circuit is separated into two pieces, the circuit for $U(\vec{\theta})$, and the circuits for $V(\vec{\theta'})$. These circuits are each composed of an $R_z(\theta)$ and $R_y(\theta)$ gate on each qubit, followed by $\rm{CNOT}$ gates to create linear entanglement over the qubits. The $\theta$'s are all initialised as random values. These layers can be repeated to improve the QSVD accuracy.
• Figure 5: Example of an experimental VQSVD circuit with GASP as the state vector initialisation circuit generation technique. The GASP-generated circuit is displayed before the barrier (dashed vertical black line), $U(\vec{\theta})$ and $V(\vec{\theta'})$ composing the QSVD circuit are displayed after the barrier. Circuits generated by GASP will almost always be unique, because of the sporadic nature of genetic algorithms, hence why this is just one example of the generated experimental circuits.