Synthetic Data Generation – Part 2: Visual Distribution Modelling with Synner

In Part 1, we covered synthetic data creation using Plaitpy, a lightweight Python library. Here in Part 2, we explore Synner—an open-source tool with a rich visual UI for designing statistical data distributions.

Refresher: The FX Futures Volume Problem

Imagine an FX futures broker launching a new desk with projected 1-year and 5-year transaction volumes across different maturities (6M, 1Y, 5Y) and notional trade brackets (200k, 1M, 10M). Because it is a brand-new desk, there is no historical production data to sample from—we have to generate realistic Day-Zero test data from scratch.

Modelling Conditional Distributions in Synner

Once Synner is built and running, you can define the distribution of trade amounts across the brackets using a helper column (_above) and then attach a Gaussian distribution conditional on that bracket.

Better still, for the 5-year volume scenario—where the split of trade sizes changes depending on whether the maturity is 6M, 1Y, or 5Y—Synner allows you to make _above directly dependent on the Maturity column.

You can either generate the dataset directly from the UI or export the model as a JSON file and run it headlessly from the command line (see our example model files on GitHub).

Conclusion on Synner

Pros:

• Clear visual representation of statistical distributions as you design them
• Rich support for conditional dependencies and custom distribution profiles
• JSON model files can be version-controlled in Git
• Easily reproducible and scriptable in CI/CD pipelines

Cons:

• Requires some technical effort to compile and set up initially