Cookbook

Short, single-task pages. One question, one runnable answer, under thirty lines of code, with the real output printed underneath. If you know what you want to do and just need the call that does it, start here.

Every block on every page was executed against the installed library and the output pasted verbatim. Each recipe stands alone — the data are simulated in the snippet with a fixed seed, so you can paste any page into a fresh interpreter with nothing but pip install tsecon and reproduce the numbers exactly.

Not sure which model your problem calls for? Start at Which model when?. Want the ideas rather than the call? That is the Guide. Want the assumptions, defaults, and failure modes of a specific estimator? That is a model card.

---

Diagnostics — before you model anything

Recipe	What it does
Screen a new series before you model it	Run check_series, read the report, and act on its ordered recommendations
Test whether a series has a unit root	The ADF + KPSS confirmatory quadrant via check_stationarity, and when to reach for phillips_perron
Find structural breaks when you don't know the dates	sup_f_test for one break, bai_perron for several, with break-date confidence intervals

Inference — standard errors that survive review

Recipe	What it does
Get HAC (Newey-West) standard errors on a regression	The se_type ladder on ols, and why hc1 does nothing for time series
Driscoll-Kraay and clustered SEs for a panel local projection	Why clustering by entity shrinks your SEs under a common shock, and what fixes it

Systems — VARs, forecasts, and identification

Recipe	What it does
Put confidence bands on a VAR impulse response	var_irf_bands with bootstrap or asymptotic bands, sliced like var_irf
Forecast a VAR with prediction intervals	var_forecast with a stability check and a coverage knob
Identify a monetary policy shock with sign restrictions	A set-identified SVAR in a dozen lines, and how to read the acceptance rate

Causal effects

Recipe	What it does
Estimate a fiscal multiplier from an instrumented shock	lp_multiplier (Ramey-Zubairy), the first-stage F, and the ratio-of-peaks trap it avoids

Risk and volatility

Recipe	What it does
Model fat-tailed volatility with GARCH and forecast it	garch_fit(dist="t"), robust SEs, a variance path, and the post-fit residual check
Compute growth-at-risk	Quantiles of future GDP growth conditional on financial conditions

Reporting

Recipe	What it does
Export a results table to LaTeX or Markdown	tsecon.summarize for the console, pandas for the paper

---

Conventions in every recipe

• Arrays in, dicts out. Every estimator takes plain arrays (or anything array-like — pandas Series and DataFrames are coerced) and returns a dict of documented keys. There is no fit/predict object to construct.
• You supply the constant. Functions taking a design matrix (ols, sup_f_test, bai_perron, recession_probit) use x exactly as given.
• alpha is the non-coverage. alpha=0.10 is a 90% band, alpha=0.05 a 95% one.
• Seeds are arguments, not globals. Anything stochastic — bootstraps, rotation draws, posterior sampling — takes a seed. Report it.
• No data loaders, no network calls. import tsecon depends only on NumPy; you bring the data.

Missing a recipe you expected to find? Open an issue — the list is demand-driven.