PPML with separated zeros

The task

Task statement

Estimate Poisson pseudo maximum likelihood for trade. The index contains agreement, pair fixed effects, exporter-year fixed effects, and importer-year fixed effects. Use the supplied identifiers. All country pairs, including equal exporter and importer identifiers, are synthetic trading units. Detect every separated zero observation. A separation direction h satisfies Xh=0 on positive outcomes and Xh<=0 on zeros. A zero is separated if some such direction gives its index a strict negative value. Remove these observations. Retain ordinary zeros. Resolve collinear dummy columns without changing the policy span.

Return the agreement coefficient and exp(coefficient)-1. Compute its sandwich standard error clustered by pair. Use bread inverse(X' diag(mu) X) and cluster scores sum_i X_i*(y_i-mu_i). Multiply the covariance by G/(G-1)*(N-1)/(N-K). N is the retained sample size, G the retained pair count, and K the retained design rank including all fixed effects. Return retained observations, separated observations, and retained zero observations. For every result, n_obs is N, the retained estimation sample size.

Data

input.csv: exporter = Synthetic exporter identifier; importer = Synthetic importer identifier; year = Calendar year identifier; pair = Exporter-importer pair identifier and covariance cluster; exporter_year = Exporter-year fixed-effect identifier; importer_year = Importer-year fixed-effect identifier; agreement = Trade-agreement indicator; trade = Nonnegative trade outcome

Files

input.csv

Required output

Scored estimates: agreement_coefficient, proportional_mean_change.

Reference output

A run succeeds when every scored estimate lies within 1 percent or 0.005 of its reference value, whichever is larger. Sample sizes are not scored.