Replication package: The Specification Ladder - A Seasonal Interaction That Reports an Effect Where There Is None

Replication package for the paper "The Specification Ladder: A Seasonal Interaction That Reports an Effect Where There Is None" (MPRA Paper No. 131703). The paper shows that a seasonal interaction between a time-invariant regional exposure and a calendar quarter, estimated under two-way fixed effects, converges to Cov(K,s)/Var(K) - the cross-sectional slope of seasonal amplitude on exposure. That quantity belongs entirely to the pre-period and has no relation to any shock, so the design reports a coefficient of the predicted sign and a large t-statistic when the true effect is zero. The package contains everything needed to reproduce that demonstration and the substantive null it implies for Portuguese minimum wages and housing transactions. Contents: four self-contained Python scripts (the specification ladder and its channel regressions; the 96-specification multiverse; permutation inference; the size-and-power simulation behind the minimum detectable effect), the three constructed panels they read, the reference output of the ladder for diffing, the four figures of the paper, a matplotlibrc that fixes font embedding, and a README documenting which script produces which reported number. All inputs are public: Statistics Portugal (INE) quarterly housing statistics by NUTS III, the Quadros de Pessoal administrative employer-employee series, and published Euribor fixings. No restricted, firm-level or proprietary data are used. Each script resolves its inputs relative to its own directory, so the package runs from wherever it is unpacked; the four scripts were re-run from this package and reproduce the values tabulated in the README. Requires Python 3.11+ with pandas, numpy, statsmodels and matplotlib.