SynthFlow | Nippofin

SynthFlow on Snowflake Marketplace →

SynthFlow generates synthetic daily SPY price paths — calibrated regime dynamics, volatility clustering, and fat tails — for strategy robustness testing beyond the single realized history.

The Backtesting Problem

Historical data gives one realized path of SPY and every backtest runs on that path. A strategy can look robust because it happened to fit the sequence of events that occurred. Was it robust, or just fit to one possibility? What if the price paths were different?

SynthFlow addresses this by generating many statistically plausible alternate histories for the same asset, so a strategy can be tested against a range of market conditions rather than the one that happened to occur.

How SynthFlow Works

You specify a historical period and a seed; SynthFlow returns a reproducible two-year synthetic price path with a full statistical passport attached.

SynthFlow’s engine runs three layers in sequence. A first-order Markov chain simulates a regime sequence: quiet, normal, or stress. An EGARCH volatility recursion with a leverage term sets each day’s conditional volatility. Filtered Historical Simulation (FHS) draws each day’s shock from a pool of real SPY standardized residuals. The three layers combine into a daily log-return, compounded into a price path. No neural networks, no GPU.

Using synthetic market data to evaluate strategies against conditions beyond the single realized record is an established application area in quantitative finance (Potluru et al., 2023). Deep generative approaches such as diffusion models are an active alternative direction in the literature (Takahashi & Mizuno, 2025). SynthFlow’s approach also differs from generative deep-learning methods for synthetic OHLCV data, such as (Homayounfar & Luoni, 2026), in favoring classical, auditable statistical methods over black-box generation.

What Makes It Unique

Auditable by design. Every path carries a passport: the seed, the realized regime sequence, full return/tail/volatility/drawdown statistics, and similarity scores against real SPY. A path is not a number series; it’s a claim you can inspect and challenge.
Transparent about its limits. The validated horizon is two years. The generator refuses longer requests rather than producing paths outside validated scope. Paths run modestly hot on volatility and drawdown depth relative to real SPY — disclosed, and directly measurable against a companion real-SPY reference table.
Grounded in real market structure. No GAN, no diffusion model. EGARCH and regime-switching are straightforward, auditable statistical methods with a decades-long track record in quantitative finance.

Use Cases

Test whether a strategy’s backtest performance is robust or a lucky fit to one historical sequence.
Stress a strategy across a taxonomy of market moods — calm grind-ups, sharp crises, V-shaped recoveries — including moods real history offers only once.
Benchmark synthetic-data generation methods against a transparent, disclosed baseline.

Sample Dataset

A free sample dataset — 240 synthetic two-year SPY paths spanning twelve historical periods from 2001 to 2024, each with full passport metadata and a real-SPY reference table — is available on Snowflake Marketplace.

Snowflake Marketplace →

Further Details

For research collaborations or licensing inquiries, contact us.

References

2025

T&F

Generation of synthetic financial time series by diffusion models

Tomonori Takahashi, and Takayuki Mizuno

Quantitative Finance, 2025

Abs HTML

Despite its practical significance, generating realistic synthetic financial time series is challenging due to statistical properties known as stylized facts, such as fat tails, volatility clustering, and seasonality patterns. Various generative models, including generative adversarial networks (GANs) and variational autoencoders (VAEs), have been employed to address this challenge, although no model yet satisfies all the stylized facts. We alternatively propose utilizing diffusion models, specifically denoising diffusion probabilistic models (DDPMs), to generate synthetic financial time series. This approach employs wavelet transformation to convert multiple time series (into images), such as stock prices, trading volumes, and spreads. Given these converted images, the model gains the ability to generate images that can be transformed back into realistic time series by inverse wavelet transformation. We demonstrate that our proposed approach satisfies stylized facts.

2023

arXiv

Synthetic data applications in finance

Vamsi K Potluru, Daniel Borrajo, Andrea Coletta, and 8 more authors

arXiv preprint arXiv:2401.00081, 2023

Abs HTML

Synthetic data has made tremendous strides in various commercial settings including finance, healthcare, and virtual reality. We present a broad overview of prototypical applications of synthetic data in the financial sector and in particular provide richer details for a few select ones. These cover a wide variety of data modalities including tabular, time-series, event-series, and unstructured arising from both markets and retail financial applications. Since finance is a highly regulated industry, synthetic data is a potential approach for dealing with issues related to privacy, fairness, and explainability. Various metrics are utilized in evaluating the quality and effectiveness of our approaches in these applications. We conclude with open directions in synthetic data in the context of the financial domain.

IEEE

Constraint-Guaranteed Synthesis of OHLCV Data

Kambiz Homayounfar, and Mario Luoni

In Proceedings of the 2026 IEEE Conference on Computational Intelligence for Financial Engineering and Economics, Tokyo, Japan, 2023

Abs

Existing generative adversarial network (GAN) based approaches to financial time series synthesis produce univariate return series only. They provide no mechanism to enforce the geometric constraints that every valid bar must satisfy: the High must weakly exceed both the Open and the Close, and the Low must weakly fall below both. We solve this through a constraint-by-construction reparameterization of each OHLCV bar into five components (close-to-close return, overnight gap, upper shadow, lower shadow, and volume) under which the bar constraints are algebraically guaranteed for every generated bar regardless of generator output. Zero violations were recorded across 1,255,500 generated bars; this result is independent of training quality and holds for any output of the trained generator. Building on QuantGAN, we extend the stochastic volatility neural network to a five-channel joint generator trained with the Wasserstein GAN with gradient penalty (WGAN-GP) and auxiliary losses targeting return drift, volume collapse, and volume-volatility co-movement. Evaluated on S&P 500 daily data from 2016 to 2025, the model reproduces the intrabar range distribution with an 8% mean error and produces a contemporaneous volume-volatility correlation of 0.23 against a real value of 0.41. Volatility clustering remains an open problem, which we discuss alongside a concrete proposed resolution.