Synthetic-Data-For-Finance

This repository complements the CFA Institute's Research and Policy Center Synthetic Data in Investment Management report. It aims to serve as a centralized hub for generative AI (genAI) approaches to synthetic data generation and their applications within finance. The repository provides a curated list of libraries, papers and case studies that can be used for synthetic data generation to aid practitioners and is regularly updated.

📘 Contents

• 🧠 Overview
• 🛠️ Libraries
• 📁 Case Studies
• 📚 Papers

---

🧠 Overview

Synthetic data is artificially generated data designed to resemble real data. It can be used to address data-related challenges such as:

• Lack of historical data
• Privacy and compliance concerns around data-sharing
• Overfitting in backtesting and model training
• Imbalanced datasets

This repository focuses on genAI approaches to synthetic data generation, focusing on the following:

• Variational Autoencoders (VAEs)
• Generative Adversarial Networks (GANs)
• Diffusion models
• Large Language Models (LLMs)

These methods are more flexible than traditional statistical methodologies, allowing for each data type to be modelled - from textual datasets to time-series and tabular data. As a result, synthetic data has a wide range of use cases within the industry, from enhanced risk modelling and portfolio optimization approaches to forecasting and sentiment analysis.

---

🛠️ Libraries

• Synthetic Data Vault: General-purpose synthetic data generation with statistical and genAI approaches.
• Synthetic Data SDK, MOSTLY.AI: Python Library for high-quality synthetic data generation
• HuggingFace Synthetic Data Generator: No-code natural language synthetic dataset builder.
• nbsynthetic: GAN-based synthetic tabular dataset creation.
• synthcity: GenAI based synthetic data library covering various data types.
• DoppelGANger: GAN-based time-series generation.
• CTGAN: GAN-based model for synthetic tabular datasets.

---

📁 Case Studies

See /LLM for an example using synthetic data to improve the performance of a fine-tuned small LLM (Qwen3-0.6B) for financial sentiment classification.

📚 Papers

Variational Autoencoders

Paper	Release Date	Type of Data Modeled	Codebase
An Overview of Variational Autoencoders for Source Separation, Finance, and Bio-Signal Applications	2021	N/A	No official repo
TimeVAE: A Variational Auto-Encoder for Multivariate Time Series Generation	2021	Time Series	GitHub
Variational Autoencoders: A Hands-Off Approach to Volatility	2021	N/A	Implied Volatility

Generative Adversarial Networks

Paper	Release Date	Type of Data Modeled	Codebase
SeriesGAN: Time Series Generation via Adversarial and Autoregressive Learning	2024	Time Series	GitHub
Time-series Generative Adversarial Networks	2019	Time Series	GitHub
Simulating Asset Prices using Conditional Time-Series GAN	2024	Time Series	GitHub
CorrGAN: Sampling Realistic Financial Correlation Matrices Using Generative Adversarial Networks	2019	Financial Correlation Matrices	No official repo
cCorrGAN: Conditional Correlation GAN for Learning Empirical Conditional Distributions in the Elliptope	2021	Financial Correlation Matrices	No official repo
Conditional Sig-Wasserstein GANs for Time Series Generation	2020	Time Series	GitHub
Deep Hedging: Learning to Simulate Equity Option Markets	2019	Equity Options	No official repo
GANs and synthetic financial data: calculating VaR	2024	Time-Series	No official repo
A Modified CTGAN-Plus-Features Based Method for Optimal Asset Allocation	2023	Time-Series	No official repo
Autoencoding Conditional GAN for Portfolio Allocation Diversification	2022	Time-Series	No official repo
Data Synthesis based on Generative Adversarial Networks	2018	Tabular	GitHub
Financial Thought Experiment: A GAN-based Approach to Vast Robust Portfolio Selection	2021	Time Series	No official repo
Improved Data Generation for Enhanced Asset Allocation: A Synthetic Dataset Approach for the Fixed Income Universe	2023	Financial Correlation Matrices	No official repo
MTSS-GAN: Multivariate Time Series Simulation Generative Adversarial Networks	2020	Time Series	GitHub
PAGAN: Portfolio Analysis with Generative Adversarial Networks	2019	Time Series	No official repo
Quant GANs: Deep Generation of Financial Time Series	2019	Time Series	No official repo
Tail-GAN: Learning to Simulate Tail Risk Scenarios	2022	Time Series	GitHub
Time Series Simulation by Conditional Generative Adversarial Net	2019	Time Series	No official repo

Diffusion models

Paper	Release Date	Type of Data Modeled	Codebase
Denoising Diffusion Probabilistic Model for Realistic Financial Correlation Matrices	2024	Financial Correlation Matrices	GitHub
FinDiff: Diffusion Models for Financial Tabular Data Generation	2023	Tabular	GitHub
High-Resolution Image Synthesis with Latent Diffusion Models	2021	Image	GitHub

Large Language Models

Paper	Release Date	Type of Data Modeled	Codebase
AugGPT: Leveraging ChatGPT for Text Data Augmentation	2023	Text	GitHub
Data Augmentation using LLMs: Data Perspectives, Learning Paradigms and Challenges	2024	N/A	No official repo
FinLLMs: A Framework for Financial Reasoning Dataset Generation with Large Language Models	2024	Text	No official repo
Simulating Financial Market via Large Language Model based Agents	2024	Time Series	No official repo

📣 Contribute

Feel free to contribute if you’d like to add a new paper, case study or tool.