Work

AI Demand-Supply Shock Replication

Year: 2026
Type: article
Abstract: Replication package for the paper "Demand shock or supply shock? Identifying the macroeconomic nature of the AI investment boom, with emerging-market spillovers." The AI investment boom raises output through two channels that call for opposite policy responses — a capital-expenditure surge in data centres, semiconductors and electricity, which is inflationary, and the productivity gains AI may eventually deliver, which are disinflationary — and because both move output in the same direction and prices in opposite directions, separating them is a structural identification problem. This archive contains the code, derived data and manuscript source for a four-stage study of that problem: a Monte Carlo audit showing that sign restrictions largely cannot separate two shocks sharing an output sign; a supply/demand decomposition of detailed US PCE categories following Shapiro (2026), which finds that the long-standing disinflation from AI-linked categories vanished and reversed after 2022; a sign-identified SVAR evaluated under Giacomini and Kitagawa (2021) robust Bayesian inference, under which the relative magnitude of the two channels turns out not to be identified; and emerging-market local projections showing that net energy importers face 0.42 to 0.76 percentage points more inflation than exporters from the same shock. The package includes eight documented Python modules, a master script, derived panels, eight figures, five tables and the full LaTeX manuscript, with every stochastic routine seeded; Stage 1 is a pure simulation that runs immediately with no data or API keys. Raw source files are not redistributed, but data/README.md documents every input and its provider (BEA, FRED, the Federal Reserve Bank of San Francisco, and BIS); results use the November 2026 data vintage. Code is released under the MIT License.
Author: Anonymous
Language: English
Open Access status: green
OpenAlex ID: W7228196237
FWCI:
Cites:
Cited by:
Related to: 10