QARIMA: A Quantum Approach To Classical Time Series Analysis
A quantum-assisted twist on classical time-series forecasting shows modest promise on environmental datasets, but real quantum advantage remains unproven.
The Thesis
ARIMA — short for AutoRegressive Integrated Moving Average — is the workhorse model economists and engineers use to forecast everything from electricity demand to pollution levels. This paper proposes QARIMA, which replaces several ARIMA setup steps with quantum-computing subroutines: swap-test circuits (a technique that measures similarity between quantum states) identify which historical lags matter, and variational quantum circuits (small, trainable quantum programs) estimate the model's coefficients. The appeal is reduced manual tuning overhead during model selection, since quantum autocorrelation functions do some of the screening work automatically. The catch is substantial: the paper tests on a handful of environmental and industrial datasets using today's noisy, small-scale quantum hardware simulators, and the accuracy gains over classical automated ARIMA are incremental at best. Until fault-tolerant quantum hardware exists at scale, this approach is primarily a research prototype, not a production tool.
Catalyst
Variational quantum circuits became tractable research objects in the last three to four years as simulators improved and frameworks like PennyLane and Qiskit matured enough to let researchers prototype hybrid quantum-classical pipelines quickly. Simultaneously, the automated ARIMA (auto-ARIMA) baseline has become sufficiently polished that beating it is a credible benchmark. The combination creates a moment where researchers can cleanly ask whether quantum subroutines add value at all to classical forecasting.
What's New
Classical ARIMA requires practitioners to manually inspect autocorrelation plots and run information-criterion searches (like AIC or BIC — statistical penalties that balance model fit against complexity) to choose model orders. Prior quantum-classical hybrid forecasting work, such as quantum LSTM variants, largely swapped out the prediction network wholesale rather than surgically replacing specific diagnostic steps. QARIMA instead targets the model-selection pipeline itself — the lag-screening and differencing steps — with quantum circuits, leaving the overall ARIMA structure intact and making it easier to isolate exactly where (if anywhere) quantum processing helps.
The Counter
The Diebold-Mariano test results reported in the paper compare QARIMA against automated classical ARIMA on a small collection of environmental and industrial datasets — a narrow empirical base that makes generalization risky. Quantum circuits in the NISQ era (noisy intermediate-scale quantum — today's hardware with hundreds of qubits but significant error rates) introduce shot noise and approximation errors that could easily swamp any signal the swap-test autocorrelation is supposed to extract cleanly. Classical methods for lag selection, including LASSO-penalized regression and information-criterion grid search, are already fast, cheap, and well-understood; it is not obvious that replacing them with quantum circuits provides anything beyond novelty. The paper also runs circuits on simulators, not real quantum hardware, so latency and noise characteristics on actual QPUs are unknown and could be far worse. Finally, the fixed VQC ansatz — the paper's way of preventing researchers from cherry-picking quantum circuit designs to inflate results — also constrains expressivity, potentially handicapping the quantum estimator relative to a fully optimized classical alternative.
Longs
- IONQ (IONQ) — trapped-ion hardware most relevant to near-term hybrid quantum-classical workloads
- RGTI (Rigetti Computing) — superconducting QPU maker whose VQC tooling underpins this class of research
- QTUM (ETF) — broad quantum-computing equity index exposure
- IBM (IBM) — Qiskit framework and cloud quantum access are direct infrastructure for this research
Shorts
- Classical auto-ARIMA software vendors — if quantum lag screening genuinely reduces tuning overhead, niche forecasting products built around automated classical model selection face substitution pressure, though this threat is distant
- Consulting firms selling bespoke time-series model tuning — automation of the order-selection step, quantum or otherwise, compresses billable hours for that specific service
Enablers (Picks & Shovels)
- PennyLane (Xanadu open-source) — the dominant framework for building and simulating variational quantum circuits
- Qiskit (IBM open-source) — alternative VQC toolchain with cloud QPU backend access
- pmdarima (Python library) — the auto-ARIMA baseline the paper competes against; its quality makes the benchmark meaningful
- NISQ-era cloud QPUs (IBM Quantum, IonQ cloud) — hardware on which these circuits eventually need to run without simulation
Private Watchlist
- Xanadu (private) — PennyLane framework creator; directly enables VQC research like QARIMA
- QuEra Computing (private) — neutral-atom hardware targeting near-term algorithm research
- Multiverse Computing (private) — focuses on quantum finance and time-series applications explicitly
Resources
The Paper
We present a quantum-inspired ARIMA methodology that integrates quantum-assisted lag discovery with fixed-configuration variational quantum circuits (VQCs) for parameter estimation and weak-lag refinement. Differencing and candidate lags are identified via swap-test-driven quantum autocorrelation (QACF) and quantum partial autocorrelation (QPACF), with a delayed-matrix construction that aligns quantum projections to time-domain regressors, followed by standard information-criterion parsimony. Given the screened orders (p,d,q), we retain a fixed VQC ansatz, optimizer, and training budget, preventing hyperparameter leakage, and deploy the circuit in two estimation roles: VQC-AR for autoregressive coefficients and VQC-MA for moving-average coefficients. Between screening and estimation, a lightweight VQC weak-lag refinement re-weights or prunes screened AR lags without altering (p,d,q). Across environmental and industrial datasets, we perform rolling-origin evaluations against automated classical ARIMA, reporting out-of-sample mean squared error (MSE), mean absolute percentage error (MAPE), and Diebold-Mariano tests on MSE and MAE. Empirically, the seven quantum contributions (1) differencing selection, (2) QACF, (3) QPACF, (4) swap-test primitives with delayed-matrix construction, (5) VQC-AR, (6) VQC weak-lag refinement, and (7) VQC-MA collectively reduce meta-optimization overhead and make explicit where quantum effects enter order discovery, lag refinement, and AR/MA parameter estimation.