Conference paper and talk
Variational Quantum Linear Solver for Time Series Forecasting: A Practical Implementation on Financial Data
S. Siyanwal, B. Kape, C. Pushkarna, K. Islur, T. Kumar, N. Khan, K. Dave
Overview
A variational quantum linear solver (VQLS) applied to time series forecasting on real financial market data. Hankel-like trajectory matrices are built from historical Reliance Industries stock prices and the resulting linear system is solved with a quantum-classical hybrid approach: Hadamard test circuits evaluate the cost function, COBYLA optimizes the variational parameters, and a SWAP test estimates the overlap of the solution state with recent market data to give the next-day forecast. Quantum advantage is not realized at this problem scale (2 × 2 system), and classical baselines reach equivalent accuracy; the workflow runs end to end on the Aer simulator with shallow circuits (depth ~6) compatible with NISQ hardware, and the paper identifies its scalability bottlenecks.
Problem / motivation
Existing VQLS demonstrations focus on synthetic test cases; practical deployment on domain-specific data remains sparse. The question was whether the quantum-classical hybrid architecture can be validated on an authentic forecasting workflow before hardware scales.
Methods
A Hankel-like trajectory matrix is built from daily closing prices of Reliance Industries (January to April 2023) and decomposed into a Pauli basis. Hadamard test circuits evaluate the VQLS cost function, COBYLA optimizes the variational parameter with 10,000 shots per circuit, and a SWAP test estimates the overlap of the solution state with recent market data to give the next-day forecast. All circuits run in Qiskit on the Aer simulator.
Results / outcomes
The workflow runs end to end on the Aer simulator with shallow circuits (depth ~6) compatible with NISQ hardware, and the solution is verified against AR(1) and linear regression baselines. Quantum advantage is not realized at this 2 × 2 scale, and the classical baselines reach equivalent accuracy in negligible time. The paper identifies the scalability bottlenecks: exponential growth of the Pauli decomposition and COBYLA shot overhead.
My role
First author.