# Stage 111 — Tax-response evidence and calibration gate

## Decision

Keep all simulation parameters and the stage-110 sensitivity grid unchanged. The evidence below does not identify the model’s cohort-specific, recurring-tax response q over twenty years. This is a selected primary-source review, not a systematic review, empirical replication, or access to restricted microdata.

The model distinguishes its baseline 0.90 income-spending rule from q, the incremental desired-spending response to an accrued surcharge. Actual spending can also be constrained by cash. Neither coefficient becomes empirical merely because a published number looks similar.

## Source ledger

Exact numeric extracts, units, sample qualifiers, locators and source URLs are in [the structured ledger](data/metadata/tax_response_evidence_v111.json).

| Source and pinned version | Evidence type | Implication for this model |
| --- | --- | --- |
| Bracha & Cooper, Boston Fed brief 13-4 (2013) | Survey reports and plans around a payroll-tax increase and refunds | An intention is not an observed, recurring spending response. |
| Baugh et al., AER (2021), Table 3 | Account transactions around annual tax settlements | Separate liability news from cash settlement before interpreting smoothing. |
| Sahm, Shapiro & Slemrod, FEDS 2015-037 | Main spending, saving or debt adjustment reported in surveys | Household response shares cannot be substituted for dollars of consumption. |
| Wong, CESifo 11185 (June 2024) | Administrative homeowner records; leveraged vehicle-spending proxy | The outcome basket and housing-wealth channel need an explicit bridge. |

These studies answer different questions. The two payroll-tax surveys also concern the same historical expiration, so they are not independent policy experiments. No coefficients are averaged or pooled.

## Interpretation safeguards

- The Boston estimate numerically resembling 0.90 does not validate either the baseline rule or the common-q scenario.
- In the AER extract, values are percentages of settlement amounts and standard errors are percentage points. A small or insignificant payment response does not identify a zero response to a permanent tax liability.
- A respondent choosing “mostly spending” is not a measurement of the fraction of each dollar spent.
- Vehicle borrowing, saving transfers, debt repayment and broad consumption must remain separate accounting objects.
- Wealth-owning simulation cohorts are not interchangeable with mortgaged homeowners. A tax change coupled with home appreciation differs from the model’s fixed-price surcharge experiment.

These are model-mapping judgments, not additional results attributed to the papers.

## Calibration gate

Before changing q, document a comparable population, tax-increase direction, expectations, liability-news timing, payment timing, horizon, consumption definition, identification and uncertainty. Then map observed spending to the model’s desired consumption and cash constraint separately. The current extraction does not pass this gate.

The research question remains whether lower prices and funded transfers can preserve purchasing power under automation, including distributional effects through debt and assets. This review strengthens the treatment of uncertainty in financing those transfers; it does not test the automation hypothesis itself.

## Coverage and next step

Primary-source searches covered payroll-tax increases, annual tax settlements and homeowner property-tax responses. Selected methods, results and instrument wording were inspected; full appendices and replication code were not audited. The Kueng NBER candidate remained excluded from coefficient extraction after access failures. The 2024 Wong version is pinned rather than silently substituted with a newer author-listed draft.

Next: audit Wong’s denominator, event-time design, reported uncertainty and heterogeneity, then decide whether any defensible bridge to broad spending is possible. Until that work passes the gate, retain stage 110’s declared scenarios and funding exclusions.

## Validation

Documentation-only publication: structured numeric units, source identifiers, unchanged-parameter flag and the public asset allowlist are checked before publication. Repository CI remains the merge gate.