Conditional forecast  ·  build 2026-07-29    

The Primacy Premium.

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

Markets do not price whether China overtakes the United States. They price how. Three pathways run to one endpoint, and if the outputs converge, the thesis is false.

Thesis variance(pathway) > variance(endpoint) Horizon 2026 to 2035 Draws 10,000 per pathway Channels four of four carried Hypotheses three of three tested
alyssa agard
The question, and the argument
§1

If China attains nuclear parity and Indo-Pacific conventional primacy by 2035, how do global defense expenditure, arms transfer flows, and commercial risk pricing reallocate, and does the pathway to primacy condition those outcomes more strongly than the fact of primacy itself?

The project refuses a single scenario. It builds three, runs each through parameters measured from the historical record, and reports full distributions rather than point estimates. It is deliberately not called prediction. There is no training data for an event that has never occurred, so the model estimates responses from history and imposes each scenario exogenously. Conditional forecasting with uncertainty propagation is the stronger epistemic position, and the project states it plainly rather than burying it.

Four commercial channels carry the transmission: semiconductors, shipping and war risk insurance, energy, and reserve currency composition. All four are carried in this build. Semiconductors as TSMC's own quarterly revenue with the 2027 capacity-tightness note; shipping as fleet, shipbuilding and connectivity structure plus the 2026 Gulf war-risk repricing, a 5.7x premium multiple measured from broker quotes; energy as chokepoint throughput and the Brent path; and reserves end to end, with the renminbi's settlement share measured monthly against its reserve share. The register below reads every artifact off disk with its row count, and the build refuses to publish this page if any is missing.

thesis.  The pathway to Chinese military primacy, gradual accretion, American retrenchment, or violent demonstration, conditions defense and commercial market outcomes more strongly than the endpoint itself. Identical 2035 force balances, reached by different roads, produce different markets.
Three pathways to one endpoint
§2

Each road terminates at the same nominal 2035 balance, which is what lets the model separate the effect of the route from the effect of the destination. If the three produced the same outputs, the thesis would be false.

I

Accretion

China grows, the United States grows more slowly. The balance closes without an event. Hedging is gradual and markets reprice at the margin.

II

Retrenchment

Primacy arrives because Washington withdraws. Allied budgets and supplier choices move first, before any Chinese gain registers.

III

Demonstration

A contingency resolves in Beijing's favor. Repricing is discontinuous: the shock quarter matters as much as the decade around it.

Input register
Every table the model consumes or produces, read off disk with its true row count at build time. The builder raises rather than publish this page if any registered artifact is missing, so a full meter is earned, not asserted.
0 / 0 tables carrying data
Expenditure by actor tier
The six-tier scope from the research design. Tier totals are sums; share of GDP and per capita are tier medians, since summing intensity ratios would be meaningless.
milex_tier_panel
Real change, year on year
Pathway II evidence. Retrenchment is an incumbent choice, not a challenger achievement, so it shows up here before it shows up anywhere else.
Peace dividend, 1985 to 2000
The retrenchment calibration case. Supplies the contraction rate, and the test for whether expansion and contraction are symmetric.
Allocated reserve shares
Channel four, the one this build implements end to end. The renminbi line is the finding: it peaked and reversed, while the dollar's losses went to other currencies instead.
cofer_shares_wide
Conditional forecast to 2035
The model's own output. Ten thousand seeded draws per pathway propagate parameter, innovation and event-timing uncertainty jointly. Read the band, not the line: the median is the least interesting number here.
simulation_fan
Arms transfers and supplier concentration
Deliveries in trend indicator values, a capability measure, deliberately never mixed with the dollar series. Supplier diversification is a falsification tripwire for pathway II.
arms_transfers_global
Hypothesis ledger
What would have to be true, and what this build measures for each claim. Verdicts are written by the pipeline, not by the page.
No.ClaimState
H1 Identical 2035 endpoints, reached by different pathways, produce statistically distinguishable outcomes in defense expenditure and arms transfer flows.
H2 American retrenchment moves allied budgets and supplier market shares faster than Chinese growth alone: the hegemon's exit is the stronger signal.
H3 Reserve composition is the slowest and least pathway-sensitive commercial channel; the dollar's share erodes toward diversification, not toward the renminbi. supported in this build
Hypothesis evidence
The measured series behind each verdict, drawn from hypothesis_panel.csv, one tidy table holding every test, evidence and robustness row the three claims rest on.
hypothesis_panel
What this build finds
§3
    Method, in five stages
    §4

    Each stage consumes only the artifacts of the stage before it, so every figure on this page is reproducible from the raw sources in a single pass.

    1. Data engineering. Seven parsers turn published sources into tidy, validated tables: IMF COFER (parse_cofer.py, with an independent share reconciliation that must pass to 0.05 pp or the build fails); the SIPRI expenditure workbook (parse_milex.py, locating headers by content so the parser survives SIPRI's annual re-layout); both SIPRI arms transfer files (parse_arms_transfers.py, a thirty-thousand-row deal register aggregated into supplier concentration); twenty-seven monthly Swift trackers (parse_swift.py, reading the data month from each document, never the filename); four UNCTAD maritime tables (parse_maritime.py); the EIA chokepoint report, the STEO workbook and nine TSMC quarterly statements (parse_energy_semis.py); and the NATO burden table, the FAS nuclear notebooks, recipient-level imports and the curated event, war-risk and alliance tables (parse_security.py). Every extraction passes range gates so a source redesign fails loudly instead of shipping a wrong number.
    2. Historical calibration. Every parameter the simulation consumes is estimated, none assumed. The reserve channel response is a Newey-West regression of quarterly dollar-share changes on the year-on-year change in the logged China to US expenditure ratio, with indicator terms for the 2022 sanctions quarters (calibrate.py). The alliance feedback system regresses each allied tier's real growth on both principals' lagged growth, HAC(3) over 1951 to 2025, and measures the 2022 mobilization differential and the named retrenchment windows (calibrate_layer2.py). All parameters ship with confidence intervals in two JSON artifacts.
    3. Conditional simulation. Two engines propagate parameter, innovation and event timing uncertainty jointly through ten thousand seeded draws per pathway: the reserve channel quarterly to 2035-Q4 (simulate.py), and a five-block system with endogenous allied budgets annually to 2035 (simulate_layer2.py), which scores the H1 separation test, projects arms demand through the measured delivery elasticity, and attaches the 2026 Gulf war-risk multiple to demonstration events. Innovations are bootstrapped from measured residuals, so simulated periods keep the true fat tails instead of an imposed Gaussian.
    4. Visualization. Four static plates (make_figures.py) and this page, whose every number is injected from the pipeline's own artifacts by build_dashboard_data.py and build_index.py.
    5. Analytic handoff. Tidy exports load directly into Power BI or Tableau, led by hypothesis_panel.csv, a single table holding every measurement behind the three hypotheses (build_hypothesis_panel.py), alongside the fans, the processed tables and payload.json; the repository remains the model of record.
    Falsification tripwires
    §5

    A model someone can use is a model that names, in advance, the observables that would break it.

    Status of the work
    §6

    All five layers built

    • Layer 1, data: seven parsers over fourteen source families, validation gates throughout
    • Layer 2, system dynamics: alliance feedback estimated 1951 to 2025, allied budgets endogenous
    • Layer 3, simulation: two seeded engines, 10,000 draws per pathway, three uncertainty sources
    • Layer 4, events: the exercise register, the 2026 Gulf repricing multiple, arms demand elasticity
    • Layer 5, delivery: this page, four plates, and the BI-ready hypothesis dataset

    Tested, with the caveats carried

    • H1 supported on the median criterion, 2.3x the reserve channel; Indo-Pacific allies the mapped exception
    • H2 directionally supported in all three tiers, unresolved at annual frequency, p 0.22 to 0.35
    • H3 supported: near-null defense coefficient, 0.36 sd separation, settlement-reserve gap live
    • Both replayed shocks rest on n = 1 precedents, stated wherever they are used
    • Thirty artifacts in the register, every one read off disk at build time
    Data, reproduction, and license
    §7

    Sources: International Monetary Fund, Currency Composition of Official Foreign Exchange Reserves (COFER), quarterly through 2026-Q1, retrieved 24 July 2026. Stockholm International Peace Research Institute, Military Expenditure Database 1949 to 2025 (v1.2, April 2026) and Arms Transfers Database, retrieved 24 and 29 July 2026. Chinese expenditure figures are SIPRI estimates. Swift, RMB Tracker and Global Currency Tracker, twenty-seven monthly issues, April 2024 data through June 2026 data. NATO, Defence Expenditure of NATO Countries, 2026 edition. UNCTADstat maritime transport tables, retrieved 29 July 2026. US Energy Information Administration, World Oil Transit Chokepoints and Short-Term Energy Outlook, July 2026. Federation of American Scientists, Nuclear Notebook, 2025 and 2026 issues, with DOD China Military Power Report projections. TSMC, consolidated condensed financial statements, nine quarterly issues. TrendForce, 1Q26 foundry ranking, public highlight only. Joint War Committee circular JWLA-033 and Argus Media war-risk reporting, 2026. Kiel Institute, Ukraine Support Tracker, as context for the 2022 mobilization case.

    Raw files are not redistributed in the repository, in keeping with the providers' terms; the README explains where to obtain each of the fourteen source families and where to place them. The simulation is seeded, so a re-run on the same data vintages reproduces every number on this page exactly, in the order parse_cofer, parse_milex, parse_arms_transfers, parse_swift, parse_maritime, parse_energy_semis, parse_security, calibrate, calibrate_layer2, simulate, simulate_layer2, build_hypothesis_panel, make_figures, build_dashboard_data, build_index.

    cc by-nc 4.0
    This page, its text, its figures and the code that produced them are © 2026 Alyssa Agard and licensed under the Creative Commons Attribution-NonCommercial 4.0 International License. You may share and adapt the material with attribution for noncommercial purposes. The underlying IMF and SIPRI data are excluded from this license: they remain the property of their publishers and are used here with citation under their own terms.