Conditional forecast  ·  build 2026-11-21

The Primacy Premium.

Conditional forecasts of defense and commercial markets under Chinese military primacy, 2026 to 2035

Three pathways to Chinese military primacy by 2035 are simulated from parameters estimated on historical data. The project tests whether allied defense expenditure and reserve currency composition depend on the pathway.

Horizon to Draws per pathway Seed Hypotheses three tested
Research question and design
§1

If China reaches nuclear parity with the United States and conventional military primacy in the Indo-Pacific by 2035, how do defense expenditure, arms transfers, and reserve currency composition change, and do the changes depend on the pathway by which primacy is reached?

The project specifies three pathways and simulates each from parameters estimated on historical data. Results are distributions conditional on the pathway. The pathways are assumptions, not forecasts of what will occur. The historical record contains no instance of the outcome, so no model can be fitted to it directly. Response parameters are therefore estimated from observed episodes, and each pathway is imposed.

Two outcomes are simulated: military expenditure by block, and the US dollar and renminbi shares of official foreign exchange reserves. Three further channels (semiconductors, shipping and war risk insurance, and energy) are documented with measured exposure series and are not simulated.

thesis.  Allied defense expenditure and reserve composition in 2035 depend on how Chinese primacy is reached (gradual accretion, US retrenchment, or a military demonstration), not only on whether it is reached.
Three pathways
§2

Each pathway is defined by growth rates or an event taken from a measured historical episode. The pathways share the premise of Chinese primacy by 2035. They do not impose an identical 2035 force balance: US expenditure in 2035 is lower under retrenchment than under the other two pathways.

I

Accretion

Chinese and US real expenditure continue at their compound growth rates of the last ten years of data.

II

Retrenchment

Chinese expenditure as in accretion. US real expenditure contracts at the rate observed from 1988 to 1998.

III

Demonstration

Accretion, plus one event in a year drawn from 2028 to 2031. Reserve shares and allied expenditure growth repeat the changes observed in 2022.

Input register
All tables read or written by the pipeline, with row counts read from disk at build time. The build stops if a registered table is missing.
0 / 0 tables built
Expenditure by actor tier
Six analytic tiers. Dollar measures are tier totals, chain-linked over members reported in consecutive years. Share of GDP, per capita, and share of government spending are tier medians.
milex_tier_panel
Real change, year on year
Annual real growth by tier over the last ten years of data.
US expenditure, 1985 to 2000
The 1988 to 1998 contraction sets the US growth rate in the retrenchment pathway.
Reserve shares by currency
Shares of official foreign exchange reserves, IMF COFER. The renminbi is reported separately from 2016-Q4.
cofer_shares_wide
Simulated reserve shares to 2035
Percentiles of the simulated paths by pathway. Parameter, innovation, and event-timing uncertainty are included.
simulation_fan
Arms transfers and supplier concentration
World deliveries in SIPRI trend-indicator values, a volume measure and not a financial value. Concentration is the Herfindahl index of delivered volume by five-year order window.
arms_transfers_global
Hypothesis ledger
Hypotheses, test criteria, and verdicts. Verdicts are computed by the pipeline and read from hypothesis_tests.json.
No.Hypothesis and criterionVerdict
H1 Allied defense expenditure in 2035 differs by pathway. Criterion: the median pathway separation of the four scored outcomes exceeds twice the separation of the dollar reserve share. The statistic describes the simulated distributions and is not a significance test.
H2 Allied expenditure responds more strongly to US expenditure than to Chinese expenditure. Criterion: one-sided test that the US coefficient exceeds the Chinese coefficient in magnitude in the pooled country panel, at the 5 percent level.
H3 Reserve composition is the least pathway-sensitive outcome, and the dollar's decline accrues mainly to currencies other than the renminbi. Criteria: the expenditure coefficient is not distinguishable from zero; dollar share separation is below the median expenditure separation; the renminbi's payments share exceeds its reserve share.
Hypothesis evidence
Series and statistics behind each verdict, from hypothesis_panel.csv.
hypothesis_panel
Results
§3
    Method
    §4

    The pipeline has five stages. Each stage reads only the outputs of earlier stages.

    1. Data. Seven parsers convert published sources to long-format tables: IMF COFER (parse_cofer.py); the SIPRI expenditure workbook, including tier and country panels (parse_milex.py); the SIPRI arms transfer tables and trade register (parse_arms_transfers.py); monthly Swift tracker issues (parse_swift.py); four UNCTAD maritime tables (parse_maritime.py); the EIA chokepoint report, the Short-Term Energy Outlook, and quarters of TSMC results (parse_energy_semis.py); and NATO, nuclear, arms import, event, and alliance tables (parse_security.py). Extracted values are range-checked and the build stops on a failed check. The COFER dollar share is recomputed from claims and must match the published share within 0.05 percentage points.
    2. Estimation. The reserve channel is a regression of the quarterly change in the dollar share on the year-on-year change in the log China/US expenditure ratio and an indicator for 2022-Q1 and 2022-Q2, with Newey-West standard errors (four lags); the renminbi share is modeled the same way in log-odds (calibrate.py). For each allied tier, real growth is regressed on the previous year's US and Chinese growth, to , with Newey-West standard errors (three lags). H2 is tested on a country panel with fixed effects and Driscoll-Kraay standard errors (calibrate_layer2.py).
    3. Simulation. draws per pathway. Reserve shares are simulated quarterly to 2035-Q4 (simulate.py). Expenditure is simulated annually to 2035 for six blocks: the United States, China, three allied tiers, and the rest of the world (simulate_layer2.py). Coefficients are drawn jointly from their estimated sampling distributions. Innovations are resampled from estimated residuals; expenditure residuals are resampled in runs of consecutive years, jointly across blocks. The event date in the demonstration pathway is random.
    4. Hypothesis assembly. Test statistics, evidence series, and context are written to hypothesis_panel.csv, and verdicts to hypothesis_tests.json (build_hypothesis_panel.py).
    5. Output. Four static figures (make_figures.py), this page (build_dashboard_data.py, build_index.py), a static summary, a report, and a data dictionary. The processed tables load directly into Power BI or Tableau.
    Observable checks
    §5

    Observations that would contradict the model's assumptions or the interpretation of its results.

    Limitations
    §6

    Estimation

    • The regressors in H2 vary only over time, so the effective sample is years. The result depends on expenditure weighting.
    • The tier equations are static. Their residuals are serially correlated, most strongly for NATO Europe, and most coefficient intervals include zero.
    • The renminbi model is estimated on quarters. Its simulated intervals are wide.
    • COFER shares are valued at current exchange rates and mix transactions with valuation changes.
    • Chinese expenditure figures are SIPRI estimates. SIPRI does not report the United Arab Emirates after 2014.

    Simulation

    • The reserve and expenditure responses to a demonstration event each repeat a single episode (2022).
    • The H1 statistic is descriptive. US expenditure differs across pathways by assumption, and the threshold of twice the reserve separation is a convention.
    • Rest-of-world growth is set to its mean over the last ten years of data, a period that includes wartime expenditure growth in Russia and Ukraine.
    • Semiconductor, shipping, and energy exposure series are reported but not simulated. Foundry concentration is not computed because market share data are not available to the project.
    Data, reproduction, and license
    §7

    Sources. International Monetary Fund, Currency Composition of Official Foreign Exchange Reserves (COFER), to . Stockholm International Peace Research Institute, Military Expenditure Database to (v1.2, April 2026) and Arms Transfers Database; Chinese expenditure figures are SIPRI estimates. Swift, RMB Tracker and Global Currency Tracker, data months to . NATO, Defence Expenditure of NATO Countries, 2026 edition. UNCTADstat maritime transport tables. US Energy Information Administration, World Oil Transit Chokepoints and Short-Term Energy Outlook (). Federation of American Scientists, Nuclear Notebook, 2025 and 2026, with US Department of Defense projections. TSMC consolidated financial statements, to . Joint War Committee circular JWLA-033 and Argus Media reporting on war risk premiums, 2026. Full citations are in REFERENCES.md.

    Raw source files are not redistributed. data/raw/README.md lists each file and where to obtain it. The simulations are seeded; rerunning the pipeline on the same input files reproduces the published figures. The run order is given in README.md.

    cc by-nc 4.0
    Text, figures, and code are © 2026 Alyssa I. Agard and licensed under the Creative Commons Attribution-NonCommercial 4.0 International License. The underlying source data are excluded from this license, remain the property of their publishers, and are used with citation under their terms.