Paper: arXiv 2412.00036
Authors: Andrew Lesniewski, Giulio Trigila
Abstract
We propose a highly efficient and accurate methodology for generating synthetic financial market data using a diffusion model approach. The synthetic data produced by our methodology align closely with observed market data in several key aspects: (i) they pass the two-sample Cramer - von Mises test for portfolios of assets, and (ii) Q - Q plots demonstrate consistency across quantiles, including in the tails, between observed and generated market data. Moreover, the covariance matrices derived from a large set of synthetic market data exhibit significantly lower condition numbers compared to the estimated covariance matrices of the observed data. This property makes them suitable for use as regularized versions of the latter. For model training, we develop an efficient and fast algorithm based on numerical integration rather than Monte Carlo simulations. The methodology is tested on a large set of equity data.
Complexity vs Empirical Score
- Math Complexity: 8.0/10
- Empirical Rigor: 7.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced mathematics including diffusion models, stochastic differential equations (SDEs), and score matching with detailed derivations, indicating high math complexity. It demonstrates strong empirical rigor through validation on equity data, statistical tests (Cramer-von Mises), and practical regularization benefits (lower condition numbers), making it backtest-ready.
Research Flowchart
flowchart TD A["Research Goal: Efficient Generation of<br>Financial Market Synthetic Data"] --> B["Data Input: Large Equity Dataset"] B --> C["Core Methodology: Diffusion Model<br>with Numerical Integration Algorithm"] C --> D["Computational Process: Generate<br>Large Set of Synthetic Market Data"] D --> E["Key Finding 1: Statistical Validation<br>Passes C-vM & Q-Q Tests"] D --> F["Key Finding 2: Covariance Matrices<br>Lower Condition Number vs. Observed"] D --> G["Key Finding 3: Methodology<br>Highly Efficient & Accurate"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes a highly efficient and accurate methodology for generating synthetic financial market data using a diffusion model approach.
Source sentence
We propose a highly efficient and accurate methodology for generating synthetic financial market data using a diffusion model approach.
Assumes 3
- Daily observations of asset prices are used. data
Source sentence
for clarity, we assume daily observations.
- No specific assumptions on the distribution of asset returns. statistical
Source sentence
Notably, we impose no specific assumptions on the distribution of these returns.
- Asset returns are modeled with a linear SDE having deterministic drift and diffusion coefficients. model
Source sentence
Specifically, a d-dimensional linear SDE driven by a d-dimensional Brownian motion
Data
- large set of equity data market-prices unknown
Source sentence
AAPL, AXP, BA, CAT, COST, CSCO, CVX, DD, DIS, GE, GS, HD, IBM, INTC, JNJ, JPM, KO, LLY, MCD, MMM, MRK, MSFT, NIKE, NVDA, PFE, PG, TRV, TSLA, UNH, V, VZ, WMT, XOM.
Costs
Unclear - sources disagree
Source sentence (the authors' words)
The Bloomberg closing prices are adjusted for stock splits but not for dividends. We believe this does not impact our simulation methodology
Finds 1
- Synthetic data passes the two-sample Cramer‑von Mises test with p‑value 1.00 and yields covariance matrices with substantially lower condition numbers than historical data (e.g., κhist=109.64 vs κsynth=53.46). statistical
Source sentence
pCvM = 1.00, κhist = 109.64, κsynth = 53.46.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 4 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (29)
- A Simple Method for Predicting Covariance Matrices of Financial Returns (2023) — link
- Diffusion Models: A Comprehensive Survey of Methods and Applications (2022) — link
- Gotta Go Fast When Generating Data with Score-Based Models (2021) — link
- Score-Based Generative Modeling through Stochastic Differential Equations (2020) — link
- Statistical Foundations of Data Science (2020) — link
- Denoising Diffusion Probabilistic Models (2020) — link
- Improved Techniques for Training Score-Based Generative Models (2020) — link
- Quant GANs: deep generation of financial time series (2019) — link
and 21 more
- Generative Modeling by Estimating Gradients of the Data Distribution (2019) — link
- The Market Generator (2019) — link
- Robust multivariate nonparametric tests via projection averaging (2018) — link
- Deep Unsupervised Learning using Nonequilibrium Thermodynamics (2015) — link
- A Family of Nonparametric Density Estimation Algorithms (2013) — link
- Simulating (2012) — link
- A Connection Between Score Matching and Denoising Autoencoders (2011) — link
- The Fokker-Planck Equation (2010) — link
- Estimation of Non-Normalized Statistical Models by Score Matching (2005)
- A well-conditioned estimator for large-dimensional covariance matrices (2004) — link
- Stochastic differential equations : an introduction with applications (1987) — link
- TIME REVERSAL OF DIFFUSIONS (1986) — link
- Reverse-time diffusion equation models (1982) — link
- On the Distribution of the Two-Sample Cramer-von Mises Criterion (1962) — link
- Numerical Solution of Stochastic Differential Equations (2015)
- Stochastic Processes in Physics and Chemistry (1995) — link
- Stochastic differential equations and diffusion processes (1981) — link
- ii) Encoding and decoding of the training (historically observed) data
- Once we determine the score function of the DSDE
- the terminal probability is white noise
- Projected Cramer - von Mises test, in preparation
0 of 29 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.