Reference¶
Two complementary references for the tsecon API.
API reference¶
The complete callable surface — every function's signature and one-line contract, generated directly from the type stub so it never drifts from the shipped module. Start here when you know the function and need its arguments.
Results objects¶
tsecon.results — an opt-in layer of dict subclasses that render themselves:
.summary() output an economist can read, plus the standard plots. Because they
subclass dict, the plain-dict contract every estimator already returns is
preserved exactly; adopting them is additive.
Testing & validation¶
The single map of how the library is tested: ten tiers from Rust goldens to a published-result replication, what each can and cannot prove, the commands to run them, and an honest list of what is not covered. Start here if you are deciding whether to trust a number this library produced.
Validation matrix¶
What every method family is validated against — the reference implementation
(statsmodels, SciPy, arch, linearmodels, scikit-learn, ArviZ) or the
documented closed form, with the fixture, the test, and the tolerance for each.
Correctness, made auditable. Its runtime companion is the
Monte Carlo suite, which verifies the statistical
properties (size, coverage, consistency) that a fixture match cannot.
Speed dashboard¶
The benchmark harness's output as a page: the cross-library parity matrix
first (every operation tsecon and statsmodels/arch/SciPy/scikit-learn both
compute, with the measured disagreement and the asserted tolerance), then the
timings with the machine and build that produced them, then the commands that
regenerate it. Rendered from a committed JSON, never typed by hand.
Model cards¶
One card per method family, each with the same anatomy: what it estimates · assumptions · when to use (and when not) · key arguments and defaults (and why) · how to read the output · failure modes · what it's validated against · references · a runnable example. Start here when you want to know whether a method fits your problem and how to trust its output.
| Family | Functions |
|---|---|
| Diagnostics | acf, pacf, ljung_box, jarque_bera, arch_lm, adf, kpss, check_stationarity |
| Phillips-Perron & Ouliaris tests | phillips_perron, phillips_ouliaris |
| Check series (one-call battery) | check_series (plus the tsecon.results.check_series facade with .summary() / .plot_diagnostics()) |
| ARIMA | arima_fit, ar_loglik |
| Spectral analysis | periodogram, welch, coherence |
| Volatility | garch_fit, gas_volatility, dcs_local_level, ccc_garch, dcc_garch |
| VAR / SVAR | var_fit, var_irf, var_irf_bands, var_fevd, var_granger, var_forecast, sign_restricted_svar, zero_sign_svar, favar, connectedness, var_girf |
| Structural identification (advanced) | long_run_svar, max_share_svar, proxy_svar, hetero_svar |
| Local projections | lp, lp_iv, lp_multiplier, lp_state, smooth_lp |
| Bayesian | bvar_fit, bvar_hierarchical, bvar_ssvs, bvar_irf_draws, mcmc_diagnostics |
| GMM | iv_gmm, gmm_nonlinear |
| Cointegration & regimes | johansen, vecm, ou_fit, spread_zscore, markov_switching_ar, setar, setar_test, setar_threshold_ci, star, star_eval, star_test, threshold_vecm, hansen_seo_test, threshold_var, threshold_var_test, threshold_var_girf |
| Forecasting | backtest, dm_test, cw_test, gw_test, theta_forecast, accuracy |
| Machine learning | ridge, lasso, elastic_net, adaptive_lasso, lasso_path, cv_splits |
| Structured penalties & post-selection | group_lasso, post_lasso, pds_lasso |
| Kernel methods | kernel_ridge, kernel_regression |
| Regression trees & random forests | regression_tree, random_forest |
| L1 trend filtering & boosting | l1_trend_filter, boosting |
| Neural regressors | mlp_regression, echo_state_network |
| Panel | panel_fe, panel_distributed_lag, panel_lp, mean_group_var, panel_mean_group, panel_pmg |
| Panel unit-root tests | panel_unit_root (LLC, IPS, Fisher/Maddala-Wu-Choi) |
| Nowcasting & MIDAS | dfm_nowcast, dfm_news, midas_weights, umidas, weighted_midas |
| Term structure | nelson_siegel, svensson, dynamic_ns, acm_term_premium, jsz_fit, jsz_loadings |
| Realized volatility | realized_measures, har_rv, realized_quarticity, tripower_quarticity, bns_jump_test, realized_range |
| Predictive regressions & IVX | predictive_regression, ivx_test |
| Recession probability | recession_probit |
| Survey expectations | cg_regression, forecast_efficiency, forecast_disagreement |
| Long memory | frac_diff, frac_integrate, long_memory_d |
| Specification & diagnostic tests | heteroskedasticity_test, reset_test, chow_test, cusum_test |
| Arbitrage-free Nelson-Siegel | afns_adjustment |
| DSGE (linear RE solver) | dsge_solve |
| Quantile regression & growth-at-risk | quantile_regression, quantile_lp, growth_at_risk |
| Functional shocks (FVAR/FLP) | functional_pca, flp, flp_scenario, fvar_scenario |
| Structural breaks | bai_perron, sup_f_test |
| Static copulas | pseudo_obs, copula_fit, copula_select |
| Cointegrating regressions (FM-OLS / DOLS / CCR) | fmols, dols, ccr |
| VAR conditional forecasts & residual diagnostics | var_conditional_forecast, var_diagnostics, var_select_order |
| Multiple forecast comparisons | spa_test, stepm_test, model_confidence_set |
| Exponential smoothing (ETS) | ets_fit, auto_ets |
| Structural time-series models (unobserved components) & TVP regression | unobserved_components, tvp_regression |