AI/ML & Algorithmic Trading Packages in R

I recently came across Guevara et al.'s (2026) book on algorithmic trading with R, sharing a summary of one of the introductory sections, which brings together the essential R packages that power AI/ML workflows and strengthen algorithmic trading infrastructure.

What is the "R quant stack"?

Not one package, 10 categories covering the whole workflow: data → modeling → risk → execution.

The image maps Category to Package(s). Here's what each does:

1. Market Data — quantmod is the entry point; tidyquant wraps it for tidyverse. IBrokers links to live execution, alphavantager covers free data, Rblpapi reaches Bloomberg — prototyping to production.
2. General ML Frameworks — caret unified training/resampling first; mlr3 modernized it with parallelization and tuning; tidymodels adds tidyverse-style composability — together, reproducible pipelines.
3. Regression — glmnet regularizes with LASSO/Ridge/Elastic Net for high-dimensional, collinear data; e1071 adds SVMs; nnet gives lightweight nets for regime detection.
4. Ensembles — randomForest aggregates trees to cut variance; ranger reimplements it faster for large-scale data; xgboost, lightgbm, catboost push gradient boosting further.
5. Deep Learning — keras (high-level) and tensorflow (low-level graph) build CNNs/RNNs/transformers for sequential data and financial NLP; torch adds PyTorch-style GPU tensors; h2o brings AutoML.
6. Bayesian ML — rstan + brms handle Bayesian hierarchical models and risk/time-series forecasting; MCMCpack offers a simpler MCMC route to predictive return distributions.
7. Explainability — lime visualizes local decisions; DALEX covers global behavior and breakdown profiles; iml unifies Shapley values and partial dependence — audit before deploying.
8. Strategy & Backtesting — PortfolioAnalytics evaluates strategies, handles constraints, and assesses portfolio risk — absorbing the accounting role once handled by blotter.
9. Risk & Performance — PerformanceAnalytics benchmarks Sharpe, Sortino, drawdowns, and CVaR; FRAPO covers risk modeling/optimization, RiskMetrics covers volatility/covariance estimation, VaR models Value-at-Risk, and fPortfolio optimizes portfolios; rugarch/rmgarch handle GARCH volatility forecasting.
10. Execution — RQuantLib brings QuantLib pricing engines for valuation/hedging; plumber turns scripts into REST APIs; shiny builds live monitoring dashboards.

Bottom line

It's one reproducible ecosystem spanning the whole pipeline — but watch for deprecated legacy packages and uneven data-feed quality.

References

Guevara, J., Bulavs, R., & Linares, O. (2026). Algorithmic trading via AI/machine learning with R. Routledge.