The Python quant stack

Python is the default language of quantitative research because of the libraries built around it. You rarely write low-level code yourself — you compose proven tools.

The core data libraries

• NumPy — fast numerical arrays and vectorized math. Almost everything else is built on it.
• pandas — labelled tables (the DataFrame) for time series, the workhorse for loading and manipulating OHLCV data.
• SciPy — scientific computing: optimization, statistics, signal processing.
• statsmodels — classical statistics and econometrics (regressions, ARIMA, stationarity tests).

The machine-learning libraries

• scikit-learn — the standard for classical ML: regression, classification, cross-validation, pipelines.
• XGBoost and LightGBM — gradient-boosted decision trees, often the strongest performers on tabular financial features.
• PyTorch — deep learning, used for neural networks and sequence models when you have enough data.

How they fit together

A typical research loop is: load data with pandas, engineer features with pandas/NumPy, model with scikit-learn or LightGBM, evaluate the statistics with statsmodels/SciPy. You add PyTorch only when a simpler model is genuinely insufficient — complexity is a cost, not a goal.

A note before we go further

None of these tools make a strategy profitable. They make you fast at testing ideas — most of which will not work. That speed is the real edge: it lets you reject bad ideas cheaply.