🇬🇧 English · 🇪🇸 Español

Observed household policy studies

These development examples connect observed household spending, explicitly financed compensation and robust policy comparisons. They require the 6.1 development source. Their price changes and preference parameters are declared scenarios, not estimates of a tariff's causal effect.

Reproduce the ENIGH study offline

python -m puremacro.examples.household_trade_policy --output enigh-study
python -m puremacro.examples.trade_adjustment_households --output adjustment-study

Each output directory must be new. CSV tables and JSON manifests retain the assumptions, inputs, financing rules and numerical comparisons. The first study also copies its authenticated source payload into the output directory.

Source: INEGI ENIGH 2022 nueva serie, concentradohogar. The retained columns contain all 90,102 original rows and the expansion factor, stratum and PSU. Of these, 69 rows have zero monetary spending and are outside the positive-budget preference domain. They represent 21,945 households and remain in the source and exclusion record. The analysis uses 90,033 sampled households representing 37,538,178 households.

Money is quarterly MXN at August 2022 prices. alimentos covers food, beverages and tobacco; the second category is gasto_mon - alimentos. Income quintiles use household expansion weights and current income per member; ties stay in the lower quintile. These are household-weighted groups, not population quintiles weighted by persons. Source archive, CSV and payload hashes accompany the data. The original all-household weighted mean current income is 63,695.46 pesos, consistent with INEGI's rounded published figure.

The INEGI free-use terms apply to these data. Attribute INEGI and retain the source metadata. The transformations and analysis are PureMacro's; INEGI does not endorse them.

Household policies and financing

The static experiment compares 5%, 10% and 20% increases in the first category's purchaser price. Each household has its own calibrated CES basket; substitution elasticities 0, 1 and 2 are assumptions. Nonlinear welfare is evaluated for each household before expansion-weighted aggregation.

Compensation rules allocate the same gross budget, 1% of aggregate baseline monetary spending: equal benefits per household, benefits for the bottom two income quintiles, or benefits proportional to baseline food expenditure. Lump-sum taxes proportional to baseline monetary spending finance the benefits. The zero-compensation case is a separate zero-budget control. There is no endogenous labor, saving, tax-base or eligibility response; net transfers are fully spent.

from puremacro.trade import balanced_compensation
from puremacro.trade.distributional import compute_distributional_welfare

# groups, scores, tax_base and prices are explicitly labeled inputs.
plan = balanced_compensation(groups, gross_total,
                            benefit_scores=scores, tax_base=tax_base)
result = compute_distributional_welfare(
    groups, baseline_prices, counterfactual_prices,
    household_transfers=plan.net_transfers, transfer_total=0., rule="ces",
    ces_elasticity=1.,
)

The weighted sum of signed transfers must equal the declared aggregate funding. Positive EV means a welfare gain measured in baseline purchasing power. The studies separately compare mean EV for the whole analysis population and for the bottom two income quintiles. Those objectives can rank policies differently; neither is silently substituted for the other.

Survey strata and PSUs are preserved but these experiments report no survey confidence intervals. The uncertainty set contains economic assumptions, not sampling draws. A causal identification strategy and survey-aware inference remain additional research work.

Recorded results

The installed-wheel run retained in tests/fixtures/household_trade_policy/ illustrates the distinction between objectives. With a 10% first-category price increase and CES elasticity one, mean EV in quarterly baseline MXN is:

Policy	All households	Bottom two income quintiles
No compensation	-1,406.05	-1,140.51
Equal benefits	-1,408.01	-1,018.04
Targeted benefits	-1,409.81	-444.96
Food-spending benefits	-1,408.62	-1,090.04

Targeting reduces the lower-income group's loss in every enumerated static state. It does not eliminate that loss in this example. Redistributing nominal resources across households with different price indexes also changes the population mean of money-metric EV. The policy objectives therefore have separate payoff and regret tables. These rankings condition on the stated model and sample; they are not an empirical optimal-policy result.

The recorded independent two-good CES formula agrees with household EV to 1.1e-15 scaled by baseline budgets. To regenerate both studies, source hashes, CSV tables and checks into a new directory from a checkout:

PYTHONPATH=. python tools/reference_validation/check_household_trade.py --output study-evidence

Robust comparisons

puremacro.study.compare_policy_payoffs takes a policy-by-state DataFrame. It reports best/worst outcomes, dominance and maximum regret, where regret is the gap from the best available policy in each state. All payoffs must share an explicit population, horizon, numeraire and welfare definition. Missing, nonfinite or failed cells block the comparison. Expected payoffs appear only when the caller supplies labeled probabilities summing to one. Grid density never implies probabilities. Tied policies remain tied.

Trade adjustment and lifetime incidence

The second study uses dynamic sourcing. A temporary 10% or 25% tariff raises the assumed imported input's delivered price at dates 2–5. A two-origin composite has assumed baseline shares 0.7/0.3 and substitution elasticity 3. Switching costs are either zero or two. Full unit procurement and adjustment costs pass through to the ENIGH first-category price by assumption. Tariff revenue is not treated as extra household income or used to finance compensation. This is not a national resource-accounting closure.

household_incidence_path computes a constant consumption equivalent over the same finite horizon for every policy. It uses separable CRRA utility of each household's CES bundle, quarterly discount 0.99 and curvature 1 or 2. Household budgets and baskets repeat at baseline, with no saving. The same gross compensation budget and its financing apply each quarter, including quarters before and after the tariff. A twelve-quarter sourcing path is compared with the first twelve dates of a 24-quarter sourcing solve; the acceptance tolerance is 1e-4 in baseline unit cost. This comparison concerns the sourcing path, not unmodeled lifetime utility after date twelve.

Tests include analytical CES welfare, conservation of financing, a discrete dynamic-programming sourcing oracle, and the constant-path equivalence of static EV and consumption-equivalent welfare. They do not establish empirical elasticities or general-equilibrium validity.