CSV, Parquet, or JSON: picking the right export for market data

SupaGamma exports every dataset in CSV, Parquet, and JSON. They're not interchangeable — the right pick depends on how much data you're pulling and what reads it next.

Parquet — the default for real work

Parquet is columnar and compressed. For anything beyond a quick look, it's the right choice: files are a fraction of the CSV size, load into pandas/Polars/DuckDB far faster, and preserve types (timestamps stay timestamps, numbers stay numbers). If you're backtesting or doing analysis at scale, download Parquet.

CSV — for eyeballing and spreadsheets

CSV is universal and human-readable. It's the right pick when you want to open a file in a spreadsheet, sanity-check a few rows, or hand data to a tool that only speaks CSV. The cost is size and parse time — a large CSV is many times bigger than the equivalent Parquet and slower to load, and everything is a string until you re-type it.

JSON — for the raw tape and programmatic use

The raw L2 tape is delivered as newline-delimited JSON, because each record is a verbatim venue frame and JSON preserves its nested structure. JSON is also convenient when a small slice feeds directly into code that already speaks it. For flat tabular data at volume, though, prefer Parquet.

A rule of thumb

• Exploring or spreadsheeting a small slice → CSV.
• Backtesting or analysis at any real volume → Parquet.
• Raw L2 frames, or feeding a small slice straight into code → JSON.