# GenXEcon

> An AI-driven platform that turns a plain-language description of an economy
> into a fully solved, verified DSGE (dynamic stochastic general equilibrium)
> model — then lets you estimate it against real data and test whether it
> actually resembles the world.

GenXEcon's guiding principle is that **the LLM proposes and verified
mathematics disposes**. Seven role-specialised agents design and write a
model's equations; a symbolic compiler then has to prove those equations
solve before any result is shown. Nothing about a model's dynamics is
templated or faked — every number traces back to an equation the model
actually contains.

## What makes it different

- **Generated models are verified, not trusted.** The compiler solves the
  nonlinear steady state, log-linearizes, and solves the rational-expectations
  system via Klein's (2000) QZ method, checking Blanchard–Kahn determinacy. It
  reproduces known analytical RBC and New Keynesian solutions to 1e-6.
- **Failures are repaired, not papered over.** Compiler and reviewer failures
  are fed back to the modeler with escalating detail. If a model cannot be made
  to work, the app says so and suggests a rephrasing — it never falls back to a
  canned template.
- **Every result states its own assumptions.** A Cholesky ordering must be
  given explicitly rather than defaulted. Sign restrictions are labelled as set
  identification. Estimates arrive with identification diagnostics, because an
  unidentified parameter still produces a confident-looking number that is
  really just the prior.
- **Limitations are documented, not buried.** First-order solutions only, no
  risk-adjusted steady state, and recursive SVARs being structurally
  incompatible with dense DSGE impact matrices are all stated plainly in the
  documentation.

## Capabilities

### Model generation and solution
- Multi-agent generation from plain-language descriptions (7 specialist agents)
- Symbolic equation compiler: nonlinear steady state, log-linearization
- Klein (2000) QZ solution with Blanchard–Kahn determinacy checking
- 5 verified reference models: RBC, New Keynesian, New Keynesian with capital,
  small open economy, oil-importing hybrid
- Second-order perturbation with accuracy checking

### Analysis of a solved model
- Impulse responses for any declared shock, with parameter overrides
- Exact population business-cycle moments from the Lyapunov equation
- Taylor-rule policy analysis; unstable and indeterminate rules are flagged
- Lucas (1987) / Otrok (2001) consumption-equivalent welfare cost
- Publication-ready PDF reports with rendered equations
- Standalone Python export that solves and plots without the app

### Empirical econometrics (16 methods, no model required)
- Trend/cycle filters: Hodrick–Prescott, Baxter–King, Hamilton (2018)
- OLS with Newey–West HAC and White heteroskedasticity-robust errors
- Unit-root and stationarity tests: ADF, KPSS, Phillips–Perron
- VAR with Granger causality, variance decomposition, forecasting
- Structural VAR: recursive (Cholesky), long-run (Blanchard–Quah), sign
  restrictions — all with bootstrap confidence bands
- Cointegration: Engle–Granger, Johansen trace and maximum-eigenvalue, VECM

### Structural estimation
- Joint maximum likelihood over structural parameters and shock volatilities,
  using a Kalman-filter likelihood on the model's own state space
- Bayesian posteriors by adaptive random-walk Metropolis–Hastings, with beta,
  gamma, inverse-gamma, normal and uniform priors
- Convergence diagnostics: split-R-hat, per-chain Geweke, effective sample size
- Local identification diagnostics on every run

### Model-vs-data validation
- Business-cycle moment comparison (model simulated at the data's own sample
  length, so both sides carry identical finite-sample bias)
- Impulse-response matching against an identified SVAR, normalised and scored
- Out-of-sample forecast accuracy against a VAR and a random walk, with
  Diebold–Mariano tests of equal predictive accuracy

### Data
- FRED (Federal Reserve Economic Data) — free API key required
- World Bank indicators — no key required
- Search by name or fetch by series code

## Facts

- **License**: MIT — self-host the entire platform at no cost
- **Stack**: Python / FastAPI backend, React + TypeScript + Vite frontend
- **Tests**: 279, none requiring network access
- **API**: 64 endpoints under `/api/v1/`
- **Storage**: models are plain JSON files; no database required
- **Source**: https://github.com/irsal2012/GenXEcon

## Pricing

- **Researcher** — free. Unlimited local models, full solver, econometrics and
  estimation, bring your own LLM API key.
- **Professional** — $49/month. Hosted workspace, managed model generation,
  PDF reports, priority compute for Bayesian estimation.
- **Institution** — custom. Shared team library, SSO, on-premise deployment,
  dedicated support.

Note: pricing is illustrative in the current build; the repository contains no
billing or subscription logic.

## Who it is for

Academic macroeconomists, central-bank and policy-institution research staff,
graduate students in economics, and quantitative analysts who need a DSGE model
built, solved, estimated and stress-tested against data without writing the
solver themselves.

## Links

- Full text for ingestion: /llms-full.txt
- Source code: https://github.com/irsal2012/GenXEcon
- License: https://opensource.org/licenses/MIT