Panel data and instrumental variables
Fit pooled OLS, fixed effects, random effects or first differences. For dynamic panels, use Arellano–Bond difference GMM or system GMM and review instrument counts, serial correlation tests and overidentification diagnostics.
Use two-stage least squares (2SLS) for multiple endogenous regressors, with identification and weak-instrument diagnostics. Use Newey–West HAC inference for ordered regressions.
Causal inference and policy evaluation
Compare two-period or staggered difference-in-differences, including cohort-by-period effects and event-time results. Fit synthetic control with donor weights, pre-treatment fit checks and placebo inference.
Use regression adjustment, IPW or AIPW for binary treatments, with balance and overlap diagnostics. Fit sharp or fuzzy regression discontinuity when assignment changes at a cutoff. Sharp RD includes bandwidth sensitivity and density checks. Fuzzy RD includes first-stage and weak-instrument-robust inference.
Discrete choice, counts and limited outcomes
Use binary logit or probit for two outcomes, multinomial logit for unordered categories, or ordinal logit for ordered categories. Use Poisson or negative binomial regression for counts. In Econometric Models, fit quantile regression with bootstrap intervals, Tobit regression for censored outcomes, or Heckman selection when the outcome is observed only for a selected sample.
Time series, volatility and structural models
Use VAR with Granger predictive-causality tests and impulse responses. Test cointegration with Johansen or Engle–Granger, fit ARDL models, or fit a vector error-correction model (VECM) with diagnostics and level forecasts.
Fit ARCH or GARCH for changing volatility and check standardized residuals. Use Structural VAR (SVAR) with explicit short-run zero restrictions to estimate impulse responses and forecast error variance shares. IID and wild bootstrap options support pointwise intervals and simultaneous impulse-response bands.