מסמך טכני מיומן המחקר · באנגלית, כפי שנכתב בשלב העבודה
MacroVision v102: evidence-to-parameter map for the central thesis
29 October 2026. Targeted primary-source screen, not a systematic review, full replication, current tax-law guidance or numerical calibration.
Decision
External research supports investigating the mechanisms, but none of the model's core numerical assumptions is newly approved for calibration. Keep the v101 conditional theoretical conclusion separate from empirical evidence.
The immediate priority is full-text extraction of estimands, methods, uncertainty and replication availability for the published transfer/spending paper. It supplies a concrete starting point for measuring consumption response; the automation-to-price bridge remains a particularly important gap for the full thesis.
Map of the three central links
| Link | Model object | Evidence currently supports | What it does not identify |
|---|---|---|---|
| Productivity -> prices | productivity_gain, pass_through | Context-dependent price reductions and a role for market structure | A global consumer-price response to 20% automation unemployment or the grid's pass-through levels |
| Economic profits -> collectible tax revenue | profit_tax_rate, capture_share | The importance of tax base, international allocation, implementation and behavioral responses | The model's accessible-profit share or net funding available for permanent transfers |
| Transfers -> displaced consumption | consumption rule, buffers, benefit funding | Material spending response to targeted temporary UI and household heterogeneity | The .90 consumption share, permanent benefit adequacy, or a causal effect of having more liquidity |
No new source coefficient is inserted into the simulation.
Primary sources and reviewed scope
S1 — Aghion, Antonin, Bunel and Jaravel (2020), CEPR DP14443. The official abstract describes French manufacturing data, event studies and shift-share IV; it reports lower prices, higher profits/sales and positive employment effects from automation. This is useful directional evidence and also cautions against assuming every automation episode raises unemployment. Full identification assumptions, tables and price measures have not been audited here. Sectoral historical estimates are not a global CPI coefficient. https://cepr.org/publications/dp14443
S2 — Firooz, Liu and Wang, revised October 18, 2024, SF Fed WP2022-05. The official abstract combines IV evidence about automation/concentration with a calibrated heterogeneous-firm model. In that model productivity gains coexist with markup distortions. This is a countermechanism to assuming all cost savings reach consumers, not a measured pass-through parameter for MacroVision. https://www.frbsf.org/research-and-insights/publications/working-papers/2023/04/automation-and-the-rise-of-superstar-firms/
S3 — OECD, January 2024 minimum-tax impact summary. The four-page summary describes projections using multinational data and assumptions about implementation, top-up taxes and profit shifting. Use it to map fiscal channels, not to infer realized cash revenue or the fraction of automation profits a government can collect. A projected percentage increase in tax revenues is not a tax rate or a capture-share estimate. https://www.oecd.org/tax/beps/summary-economic-impact-assessment-global-minimum-tax-january-2024.pdf
S4 — IMF staff, International Corporate Tax Reform, February 2023. The executive summary projects modest revenue gains, differing country effects and possible investment costs. These are policy-scenario results, not observed reform outcomes. They have different coverage/assumptions from S3 and must not be pooled as independent estimates of the same quantity. This historical document is not used to assert today's implementation status. https://www.imf.org/-/media/files/publications/pp/2023/english/ppea2023001.pdf
S5 — Ganong, Greig, Noel, Sullivan and Vavra (2024), American Economic Review 114(9), DOI 10.1257/aer.20220973. The published abstract reports large spending and small job-finding effects from expanded pandemic UI. It also reports high spending propensities among normally low-liquidity households even after balances increase, pointing to persistent heterogeneity. This supports targeted-transfer research but does not isolate the causal protection of liquidity or establish an indefinitely sustainable automation benefit. Full identification and numerical estimates await extraction. https://www.aeaweb.org/articles?id=10.1257/aer.20220973
These reviewed scopes are deliberate: reading an official abstract does not certify a causal estimate or replication. One additional NBER page returned 403 and is recorded as uncompleted, not included as a fully reviewed core source.
Why a source number cannot simply replace a model parameter
The inherited price formula is:
``` price level = 1 - pass_through * .60 * (1 - wage_index / efficiency)
``` A source elasticity of prices with respect to an automation measure is not pass_through in this formula. Mapping requires the source technology measure, productivity change, wage response, markups, price basket and horizon. A relative producer/export price change also differs from an aggregate consumption deflator. The .60 baseline wage share is a normalized model assumption.
Similarly:
``` modeled cash profit tax = modeled profit rate * accessible share * positive modeled profits
``` Accounting tax expense, taxes actually paid, taxable income and economic profit have different definitions and timing. Country-level receipts/aggregate profit ratios alone do not identify the causal capture share. Wage-tax losses, existing spending and interest must be deducted before interpreting funds as available for transfers.
Finally, the model consumes .90 of a specified spendable-income object. A study's marginal spending response to a temporary transfer is a derivative with a particular horizon; it is not that average consumption share. Temporary income replacement, permanent basic income and consumption of only food are different estimands.
Evidence gaps that now govern progress
- Prices: choose sector/country and comparable consumer-price/productivity measures; inspect direct estimates and market structure before proposing a pass-through anchor.
- Fiscal capacity: define a reconciled cash-tax base with timing and coverage; estimate net transferable resources instead of applying a statutory rate to all modeled profits.
- Consumption: extract full published transfer-response definitions, horizons, units and uncertainty; assess transportability separately from statistical identification.
- Common macro scope: 20% global unemployment, long-run production, permanent transfers and simultaneous price/ownership changes have not been jointly identified by these sources.
The sources can support mechanism discussion and candidate empirical designs. They cannot currently validate the grid's chosen values or turn 75 preservation cases into a probability.
Ranked next research action
First inspect S5's full published methods, estimates, appendix and replication terms. Create a source table for each admissible estimate containing population, shock, outcome, units, horizon, identification assumptions, uncertainty and access limits. Do not acquire restricted records, run paid access or alter parameters at this step.
Then work on a direct price/productivity measurement bridge. Retain contradictory or context-limited evidence; do not select studies for coefficients that make the simulated thesis succeed. A systematic review or full replication, if later undertaken, needs its own declared search/extraction rules.
Traceability and validation
The machine-readable ledger at data/metadata/thesis_evidence_map_v102.json records five primary sources, review scope, parameter targets, inference limits and blocked gates. Review date is October 29, 2026. URLs and source metadata are public; full source bytes are not archived or checksum-pinned in this screen. No participant data or external numerical estimates were imported.
Validation is documentary: parameter names/formula checked against inherited v71/v77 rules; source types and review depth are explicit. Existing site CI and post-live asset checks verify publication, not causal correctness.
References: v101 central thesis synthesis, v90 empirical uncertainty. No original model or calibration gate was changed.