Greeners

High-performance econometrics in Rust — statsmodels-grade coverage, compiled-language speed, and Rust's type safety guarantees. Named in honor of William H. Greene.

Why Greeners?

Every serious econometrics tool makes a trade-off:

Greeners fills the gap: the estimator coverage of statsmodels, the performance of compiled code, and Rust's guarantee that type errors and data races are caught at compile time — not at 2 AM in production.

No Python or R runtime. No system BLAS or LAPACK. Built on faer — pure-Rust linear algebra with performance competitive with OpenBLAS, and a single cargo add greeners to install.

Beyond parity, Greeners includes methods statsmodels does not: panel fixed and random effects, Arellano-Bond GMM, IV/2SLS, Difference-in-Differences, and automatic binary variable detection — all in a self-contained library with built-in DataFrame, formula parser, and estimators.

Quick start

use greeners::{OLS, DataFrame, Formula, CovarianceType};

fn main() -> Result<(), Box<dyn std::error::Error>> {
    let df = DataFrame::from_csv("data.csv")?;
    let formula = Formula::parse("wage ~ education + experience + C(region)")?;
    let result = OLS::from_formula(&formula, &df, CovarianceType::HC3)?;
    println!("{}", result);
    Ok(())
}

Installation

[dependencies]
greeners = "1.5"

No system dependencies required.

What's covered

Linear models

Generalized linear models

Discrete choice & count models

Panel data

Time series

State space & decomposition

System estimation

Mixed & multilevel models

Causal & robust methods

Survival analysis

Multivariate & nonparametric

Rolling & recursive

Diagnostics & tests

Regression diagnostics (diagnostics)

Jarque-Bera, Breusch-Pagan, Durbin-Watson, VIF, condition number, leverage, Cook's distance, omnibus, Harvey-Collier, Anderson-Darling.

Specification tests (specification_tests)

White test, RESET, Breusch-Godfrey, Goldfeld-Quandt.

Influence & stability (influence)

Influence measures, CUSUM test.

Time series tests (timeseries)

ADF, KPSS, Phillips-Perron, Zivot-Andrews (structural break), Ljung-Box, ACF/PACF.

Model selection (model_selection)

AIC/BIC comparison, Akaike weights, panel diagnostics (BP LM, F-test, Hausman).

Statistical tests (stats)

One-way and two-way ANOVA, Tukey HSD, Bonferroni post-hoc, regression F-tests.

Multiple testing (multipletests)

Bonferroni, Holm, Hochberg, Hommel, Benjamini-Hochberg, Benjamini-Yekutieli.

Other (proportion, descrstatsw, hausman)

Proportion tests (one/two-sample, equivalence), weighted descriptive statistics, Hausman test.

Inference

All linear models support two inference distributions via .with_inference():

use greeners::InferenceType;

let result = OLS::from_formula(&formula, &df, CovarianceType::HC3)?;
// Default: Student's t (exact, finite-sample)
let result_z = result.with_inference(InferenceType::Normal)?;
// Normal/z (asymptotic, statsmodels-compatible)

Covariance types

CovarianceType::NonRobust               // Classical
CovarianceType::HC1                     // White (1980)
CovarianceType::HC2                     // Leverage-adjusted
CovarianceType::HC3                     // Jackknife (recommended for small samples)
CovarianceType::HC4                     // Cribari-Neto (2004)
CovarianceType::NeweyWest(lags)         // HAC
CovarianceType::Clustered(ids)          // One-way clustering
CovarianceType::ClusteredTwoWay(a, b)   // Two-way clustering (Cameron-Gelbach-Miller)

Formula syntax

// Basic
"y ~ x1 + x2"                    // with intercept
"y ~ x1 + x2 - 1"                // no intercept

// Categoricals
"y ~ x1 + C(region)"             // auto dummy encoding (K-1)

// Transforms
"y ~ x + I(x^2)"                 // polynomial
"y ~ log(x1) + sqrt(x2)"         // functions

// Interactions
"y ~ x1 * x2"                    // full: x1 + x2 + x1:x2
"y ~ x1 : x2"                    // interaction only

// Splines
"y ~ bs(x, df=5)"                // B-splines

DataFrame

// Load from CSV, JSON, URL, or builder
let df = DataFrame::from_csv("data.csv")?;
let df = DataFrame::from_csv_url("https://example.com/data.csv")?;
let df = DataFrame::from_json("data.json")?;
let df = DataFrame::builder()
    .add_column("y", vec![1.0, 2.0, 3.0])
    .add_column("x", vec![4.0, 5.0, 6.0])
    .build()?;

// Missing data
let clean   = df.dropna()?;
let filled  = df.fillna("x", 0.0)?;
let forward = df.fillna_ffill("x")?;
let interp  = df.interpolate("x")?;

// Time series ops
let lagged  = df.lag("price", 1)?;
let diffed  = df.diff("price", 1)?;
let returns = df.pct_change("price", 1)?;

Column types are auto-detected: Float, Int, Bool (including binary detection from any two-value column), DateTime, Categorical, String.

Imputation

use greeners::{MICE, BayesGaussMI};

let imputed = MICE::impute(&df, 10)?;         // Multiple imputation by chained equations
let imputed = BayesGaussMI::impute(&df, 10)?; // Bayesian Gaussian MI

Examples

59 examples in examples/. Run any of them:

cargo run --example formula_example
cargo run --example marginal_effects
cargo run --example specification_tests
cargo run --example panel_model_selection

About the name

William H. Greene is the author of Econometric Analysis — the reference text used in graduate econometrics programs worldwide. Greeners is named in his honor: the goal is to make the methods he systematized available to anyone writing production systems in Rust.

Roadmap

Active development targets full statsmodels parity. See ROADMAP.md for the complete feature matrix with implementation status.

Contributing

Contributions are welcome. See CONTRIBUTING.md for guidelines.

Before opening a PR:

• Add at least one example to examples/ for new estimators
• Verify numerical output against statsmodels or R on a reference dataset
• Run cargo test and cargo clippy -- -D warnings

License

MIT