# V103 — Transfer-response estimands and the thesis calibration gate

Status: targeted published-paper extraction, October 29, 2026. This supersedes only S5's abstract-only review status in v102. It does not supersede the other four source reviews or modify simulation parameters.

## Source extraction
Ganong et al. (2024), AER 114(9), DOI 10.1257/aer.20220973. Published author copy: https://bpb-us-w2.wpmucdn.com/voices.uchicago.edu/dist/b/1275/files/2024/08/Ganong-Greig-Noel-Sullivan-Vavra_2024_AER_Spending-and-Job-Finding-Impacts-of-Expanded-Unemployment-Benefits.pdf

Table 1 (p.2911): one-month additional spending per benefit dollar; estimates and household-clustered standard errors are in the accompanying JSON. Waiting compares early/delayed recipients (total UI); supplement designs compare matched employed households, except the interstate comparison. Identification requires counterfactual parallel spending changes. Spending broadly approximates nondurables and excludes debt payments. Table 4 (p.2926) contains structural-model outputs, not independent causal estimates; longer horizons are not reliably identified directly.

Official appendix: https://www.aeaweb.org/articles/materials/21546
A.1–A.4 document simultaneous eligibility changes, selected short spells, account screens and imperfect payroll visibility. D.5 separates transfer size, persistence and targeting in model comparisons; calibration targets are not validation observations. E.1 discusses calendar aggregation.

## Interpretation for MacroVision
The numeric rows are archived as source estimates, not proposed parameter values. The published Table 1 is the authoritative extraction: rounded numbers appearing elsewhere in model comparisons must not silently replace it.

Our thesis asks whether consumption can persist under automation-related job loss, given prices, ownership, debt and feasible transfers. That question requires a path of resources and prices. A monthly marginal response alone neither supplies the resources nor establishes long-run consumption.

The inherited .90 coefficient multiplies a modeled spendable-income object. Replacing it with any extracted MPC would change both the economic object and the time unit. An annual rate cannot be obtained by multiplying a monthly MPC by twelve: later responses and transfers overlap, and saving remains a stock. Likewise, a difference between liquidity groups does not identify the effect of assigning a household extra liquid wealth.

Sampling uncertainty, identification risk and transportability must remain separate. Reported clustered standard errors address the first under the source design. They do not quantify the error in applying a pandemic estimate to permanent unemployment, another country or a changing price system. Model-output rows have no extracted confidence interval; null does not mean zero uncertainty. No pooled coefficient or confidence interval is manufactured.

## Concrete next design
Before another simulation, specify a monthly incremental-consumption module:
- Define baseline spending and the funded transfer separately, with receipt dates.
- Track incremental spending, saving and debt service in distinct cash-flow accounts.
- Specify a finite response horizon and aggregation rule; reconcile each year's monthly flows to the existing annual accounts.
- Hold price and fiscal mechanisms fixed when checking the response module, then vary them transparently.
- Treat source estimates as contextual comparison targets only after matching shock, population, outcome and horizon. A model fitted to a target cannot validate itself using that target.

This is a proposed measurement contract, not an implemented or estimated response model. The immediate next task is to formalize that contract and audit whether the current annual simulator can support it without inventing monthly observations.

## Access and checks
The published author-hosted text and selected official appendix sections were inspected. This is not an exhaustive appendix review. The AEA full-text endpoint required access; the replication endpoint returned 403. No replication code was executed, restricted records acquired or paid access attempted. Public source bytes are not checksum archived.

The JSON distinguishes empirical rows and structural outputs and records closed calibration gates. Publication checks cover JSON validity, public links, existing CI and live deployment; they do not certify causal validity.

See [v102 evidence map](THESIS_EVIDENCE_MAP_V102.md) and [v101 thesis synthesis](THESIS_SYNTHESIS_V101.md).