TGM AL MAMUN
Open-source econometric workspace

Open Econometric Systems

Every model in the Time-Series, Panel Data, Causal Inference, and Machine Learning guides is ready to run. Bayesian VAR and Markov-Switching models are also included. Choose a model and load its example dataset or upload your own CSV file. Get estimation results, diagnostic tests, statistical outputs, and charts. No installation or external software is required. Python runs directly in your browser.

31 models6 model groupsRuns in your browser: your data stays on your deviceFree and open source
01 · Model

Choose a model

Each model links to its guide, which explains when to use it and how to write up the results. The Bayesian and Markov-switching groups have their own guides, the Bayesian Guide and the Markov-Switching Guide.

02 · Data

Load data

Use the dataset from the model's guide, or your own CSV file with one column per variable (and unit and time columns for panel data).

No data loaded.
or drop a CSV file here
03 · Variables (optional)

Create variables

Lags, first differences, logs, growth rates, cross-sectional demeaning and the estimation sample. Lags and differences are taken within each unit when a unit column is given. The new columns appear in the model fields below.

Operation
Variables
04 · Specification

Specify the model

Choose a model and load data first.

06 · Test suite

Test every model in your browser

Runs each model on its guide example here, in your browser, and compares the first reported estimate with the reference value computed with the same package versions. The engine is also tested against statsmodels, linearmodels, scikit-learn and XGBoost and against published results (for example, Arellano and Bond's employment equation) in the repository's test suite.

07 · About

How it works

Python in the browserThe estimators are Python (NumPy, pandas, SciPy, statsmodels, scikit-learn, XGBoost) running in Pyodide. The first run downloads Python once (about 30 MB); later runs start at once. Your file is read on your device and never uploaded.
Bayesian and regime-switching modelsThe Bayesian VARs follow the ECB's BEAR toolbox (Minnesota, Normal-Wishart, Independent Normal-Wishart and Normal-diffuse priors, FAVAR, steady-state, panel, TVP, stochastic-volatility, mixed-frequency and threshold VARs). Markov-switching models are estimated by maximum likelihood; the regression and AR versions reproduce Stata's and Hamilton's published estimates.
The same code as the guidesDifference GMM, the panel ARDL estimators, the causal designs and double machine learning use the helper files published with the guides. Static panel models reproduce linearmodels, including its small-sample scaling.
What is not hereDSGE models are solved and estimated in Dynare; see the DSGE Guide. The causal forest is a doubly robust forest, not EconML's implementation, which does not run in a browser.
Large filesEverything runs on your computer, so very large datasets and the forest-based methods take longer. Fewer trees or folds speed them up.