💰 quantitative finance

Classification of Financial Data Using Quantum Support Vector Machine

This paper presents the first systematic study applying Quantum Support Vector Machines to the Dhaka Stock Exchange Broad Index dataset, demonstrating that specific quantum kernels can outperform classical RBF-kernel baselines within the empirical quantum advantage framework while correlating performance with the Phase Space Terrain Ruggedness Index.

Original authors: Seemanta Bhattacharjee, MD. Muhtasim Fuad, A. K. M. Fakhrul Hossain

Published 2026-07-08

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Seemanta Bhattacharjee, MD. Muhtasim Fuad, A. K. M. Fakhrul Hossain

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to predict the weather in a very chaotic city. You have a lot of data: temperature, humidity, wind speed, and even the price of gold. A classical computer (like the laptop you use today) tries to find patterns in this mess using a standard map. It's good, but sometimes the weather is so weird and the patterns so tangled that the map gets confused.

This paper is about testing a Quantum Computer to see if it can draw a better map for this specific chaotic city: the Dhaka Stock Exchange (DSEx) in Bangladesh.

Here is the breakdown of their experiment in simple terms:

1. The Goal: Finding the "Right" Map

The researchers wanted to see if a Quantum Computer could predict whether the stock market index would go up or down tomorrow better than a classical computer.

The Data: They didn't use a massive global dataset. Instead, they created a specific, smaller dataset (460 data points) combining the top stocks in Dhaka and the price of gold over ten years. They chose a smaller size on purpose because real-world financial data is often messy and hard to get, making it a tougher challenge to solve.
The Contest: They pitted a standard "Classical SVM" (a popular, reliable algorithm) against several "Quantum SVMs" (algorithms that use the strange rules of quantum physics).

2. The Quantum Secret Sauce: "Feature Maps"

To make the quantum computer work, the researchers had to translate the stock data into a language the quantum machine understands. They call this a Feature Map.

The Analogy: Imagine you have a flat piece of paper (the stock data). A classical computer tries to draw a line to separate "Up" from "Down" on that flat paper. Sometimes the lines get too messy.
The Quantum Trick: A quantum computer can "fold" that paper into a complex 3D shape (like origami). Suddenly, the "Up" and "Down" groups might be on different layers of the paper, making it easy to slice them apart with a single straight cut.
The Specific Tool: They tested different ways to fold the paper. They found that one specific folding technique, called the Pauli Y YY Feature Map, was the best at untangling the stock market's mess.

3. The Results: The Quantum Advantage

When they ran the tests:

The Winner: The Quantum Computer using the Pauli Y YY map consistently beat the Classical Computer.
The "Zero-Advantage" Line: They drew a line representing where the classical computer performs its worst. The quantum computer stayed above this line in almost every scenario, meaning it was more accurate even when the data was tricky.
The Terrain Check: To understand why this worked, they used a metric called PTRI (Phase Space Terrain Ruggedness Index). Think of this as checking how "rocky" the ground is. They found that the quantum computer is like a mountain goat—it handles "rocky" (noisy, complex) terrain much better than the classical computer, which is like a car that struggles on bumpy roads.

4. The Reality Check: Resources and Limits

The paper is very honest about what it takes to run this:

The Cost: They calculated exactly how many "gates" (the basic switches of a quantum circuit) and how much "depth" (how many steps the circuit takes) are needed. They provided a formula so other scientists can estimate the cost before trying it themselves.
The Hardware: They tested this on a real quantum chip (IBM's 7-qubit processor) and found the results matched their simulations.
The Caveat: They admit this was a "small-scale" study. They used a small dataset and didn't spend hours tweaking the settings (hyperparameters) to get the absolute best score. It's like a "clinical trial" to prove the concept works, not a full-blown commercial product yet.

Summary

In short, this paper is a proof-of-concept. It says: "We took a specific, messy set of stock market data from Bangladesh. We tried a new quantum method called the Pauli Y YY map, and it predicted market trends more accurately than the standard classical method. We also figured out exactly how much 'fuel' (quantum resources) is needed to run this engine."

They aren't saying quantum computers are ready to replace Wall Street traders tomorrow, but they have shown that for this specific type of financial puzzle, the quantum approach has a clear edge.

Technical Summary: Classification of Financial Data Using Quantum Support Vector Machine

Problem Statement
Classical machine learning techniques, particularly Support Vector Machines (SVMs), face limitations when processing the high-dimensional, non-linear, and noisy data characteristic of financial markets. Stock market data often exhibits non-stationary behavior, where statistical properties change over time, making the prediction of daily index changes a challenging task. While classical SVMs have proven effective for certain tasks, they struggle with intricate patterns found in real-world financial datasets. This study addresses the potential for Quantum Support Vector Machines (QSVMs) to outperform classical baselines in this domain, specifically focusing on the Dhaka Stock Exchange (DSEx) Broad Index.

Methodology
The research employs an Empirical Quantum Advantage (EQA) framework to benchmark quantum kernels against a classical Radial Basis Function (RBF) kernel SVM. The study utilizes a self-curated dataset comprising 460 data points, derived by merging historical DSEx Broad Index data (sourced from Investing.com) with a decade of gold price data (from usagold.com). The dataset is designed to predict daily changes in the index, utilizing subsets ranging from 200 to 400 data points with 5 to 7 features.

Quantum Approach: The authors construct quantum feature maps to encode classical input data into quantum states. The primary focus is on the Pauli Y YY feature map, which generates a unitary operation $U_\Phi(x)$ involving alternating layers of Hadamard gates, single-qubit rotations parameterized by Pauli-Y operators, and entangling two-qubit interactions. This map is conjectured to be classically hard to simulate and is implementable as a short-depth circuit. The authors also tested Pauli Z, Pauli ZZ, Pauli Y ZZ, and Pauli Z ZZ feature maps for comparison.
Experimental Setup: Experiments were conducted using Qiskit and IBM Quantum simulators (specifically the qasm simulator without noise models) as well as actual hardware (IBM's 7-qubit ibm_nairobi QPU). To ensure consistency, all comparisons utilized a single random seed.
Metrics: Performance was evaluated using Balanced Accuracy and F1 Score. The study also employed the Phase Space Terrain Ruggedness Index (PTRI) as a global metric to identify configuration spaces where quantum kernels might offer an advantage.
Resource Estimation: The paper provides mathematical formulations to estimate the quantum resources (gate counts and circuit depth) required for the Y YY feature map based on the number of features ( $F$ ) and circuit repetitions ( $R$ ).

Key Results

Performance Superiority: In most configurations, the quantum kernels outperformed the classical RBF-kernel SVM baseline in predicting daily changes in the DSEx Broad Index.
Optimal Kernel: The quantum kernel built using the Pauli Y YY feature map consistently demonstrated the best performance, outperforming all other quantum kernels and the classical SVM across every point in the feature-dataset configuration space tested.
EQA Verification: The study observed Empirical Quantum Advantage (EQA) for all problem instances within the tested configuration space. The PTRI analysis indicated that while classical SVM performance remained relatively stable regardless of training set size, the quantum kernels performed notably better on "smoother terrain" within the phase space.
Hardware Consistency: Results obtained from the ibm_nairobi hardware consistently aligned with noiseless simulation results, despite the experiments running with a maximum of 1024 shots.

Key Contributions

First Systematic Study: This work presents the first systematic application of quantum kernel methods to the DSEx Broad Index dataset.
Benchmarking Framework: It establishes a robust comparison framework for various quantum kernels against classical baselines on financial data, utilizing small-sized datasets to mimic the challenging classification problems faced by financial institutions.
Resource Estimation: The paper provides explicit equations for calculating the total number of gates, specific gate counts (H, Rx, P, Cx), and circuit depth required for the Y YY feature map, offering a practical guide for future practitioners estimating resource needs.
PTRI Application: The study validates the use of the PTRI metric to delineate regions of potential Empirical Quantum Advantage within financial datasets.

Significance and Claims
The authors position this study as a foundational framework for future practitioners, analogous to clinical trials in other fields. They claim that the results demonstrate the potential for gaining an empirical quantum advantage over classical baselines on complex, real-world financial datasets. The paper emphasizes that while the current study uses small datasets and noiseless simulations (with a single random seed), it successfully documents the intrinsic capabilities of quantum kernels. The authors modestly conclude that as quantum hardware improves, larger and more rigorous studies will be necessary to fully clarify the extent to which these techniques can benefit the finance industry. They suggest future directions include exploring classically hard-to-simulate kernels and developing custom feature maps specifically tailored to the characteristics of financial data.