1. From Forests to Boosting

Boosting works by sequentially correcting errors of prior trees. Intuition: "a committee of specialists" where each tree focuses on remaining mistakes.

Unlike random forests that average independent trees, boosted trees build an ensemble where each new tree learns from the residuals of the previous ensemble.

2. Why Boosted Trees Dominate Tabular ML

• Strong performance with modest tuning.
• Handle sparse, heterogeneous features.
• Built-in regularization (shrinkage, tree depth, subsampling).

3. XGBoost vs LightGBM

XGBoost:

• Mature, flexible, highly configurable.
• Excellent documentation and community.
• Widely used in production systems.

LightGBM:

• Faster with large feature sets; histogram-based splits.
• Lower memory footprint.

In volarixs, both live under the Tree & Boosted family you pick in the experiment wizard, run against the same datasets, feature sets and target horizon as every other model — so a boosted run is directly comparable to the linear, neural, time-series and volatility candidates beside it.

4. Use Cases

• Cross-sectional 5d/21d returns.
• Realized vol prediction.
• Regime-aware models: feed the macro/regime state in as features, or scope an experiment to a single regime so the booster learns where it actually has to work.

That second idea is where the regime context volarixs records comes in: every run carries the macro state and historical analogues it was made under, so you can read a booster's result against the environment that produced it rather than as a single blended number.