Research toolkit for Python quant workflows

The labeled-array model, pandas extensions, persistence, automation, and introspection behind every VBT workflow. These pages cover how many experiments fit into one array, the compiled operations that run on them, and the tools that save, schedule, and inspect your research. This section is for readers who already run backtests and want to understand why VBT shapes data the way it does, work faster with it, or build on it.

Find the tool for your next step

What do you want to do?	Where to start
Compare assets, parameter sets, or strategies without losing their labels	Multidimensional research explains broadcasting, selection, grouping, and stacking.
Calculate rolling statistics or apply date and session rules	Time-series operations adds compiled calculations to pandas objects.
Reopen a backtest or reuse a project's settings	Configuration and persistence saves objects, inputs, and configuration.
Run updates or research jobs on a schedule	Workflow automation covers tasks, scheduling, progress, and notifications.
Inspect an unfamiliar object or debug custom code	Developer tools covers help, source, prepared inputs, and extension tools.

Multidimensional research

Run assets, parameters, and strategies as labeled columns, then index and stack them

Time-series operations

Run compiled rolling windows, reductions, and date logic on labeled pandas data

Configuration and persistence

Save any research object with compression, reload it fast, and keep settings in files

Workflow automation

Schedule data updates, send Telegram alerts, run tasks in parallel, and track progress

Developer tools

Search the API, read source, draw reference graphs, and extend classes with factories

Keep your Python workflow

Start with the pandas Series and DataFrames you already use. The .vbt accessor adds calculations and array operations while keeping labeled results, so you can move between pandas analysis and VBT's compiled functions. As the experiment grows, named column levels keep symbols, parameter values, and strategies attached to the results they produced.

A typical workflow calculates signals, compares backtests, saves the chosen inputs and portfolios, and reopens them later for more analysis. The same calculation can run across a list of symbols or on a schedule. You can inspect its arguments and source when you need to change it, then turn a reusable calculation into a custom indicator.

For larger workloads, Parallel execution and caching covers dividing work into batches and reusing calculations. For preparing market data before the experiment, see Financial data pipelines.

Start here

Browse the Cookbook for practical recipes on arrays, indexing, configuration, and persistence.