25 papers · October 2026
Sovereign Debt & Fixed Income
Working Paper 22 · October 2026 · PDF
Five ways out of debt: Britain's century-long grind, America's postwar financial repression, Sweden's legislated rules, Canada's cuts, Belgium's euro-anchored surpluses. The repression did more work than the surpluses. Japan and Italy are the red flags.
Working Paper 21 · October 2026 · PDF
Five disinflations, one anatomy: deeply positive real yields, a named credible policymaker, and a forcing crisis preceded every bond rally. Israel killed 445% inflation with GDP growing; Turkey's real rates sat above 30% for two years into reform.
Working Paper 20 · October 2026 · PDF
In dollar terms the frontier carry trade does not survive: Ghana earned -78% and Kyrgyzstan -13% from 2010 to 2024. Starting yield still predicts forward dollar returns at long horizons, but the currency eats half the local carry; rising debt-to-GDP, not IMF programs, is the ruin signal.
Working Paper 19 · October 2026 · PDF
Bond markets exceed equity markets in 9 of 10 economies; the US is the sole exception with its ratio falling from 1.22 to 0.89. China's bond market grew 24-fold to $28.7 trillion; equities run 2 to 7 times hotter than bonds everywhere, so the ratio moves on the equity leg.
Working Paper 17 · October 2026 · PDF
Euro-area spreads never returned to pre-2008 levels: France trades about 80bp over Germany against near zero in 2006, the UK spread more than doubled; Korea and Australia compressed about 200bp as Japanese yields normalized; the largest spikes were episodic and country-specific.
Working Paper 04 · September 2026 · PDF
US federal interest rose from 10.9% of revenue in 2020 to 20.3% in 2024; about 93% of 2024 US gross issuance refinanced maturing debt.
Working Paper 03 · September 2026 · PDF
Major-market 10-year yields rose about 3 points from their 2019–2020 troughs; roughly 80% of 2025 OECD sovereign borrowing refinanced maturing debt.
Banking & Credit
Working Paper 27 · October 2026 · PDF
From 1,495 primary documents across 29 campuses: of $1,165.00B in uncommenced lease obligations only 3.3% names a counterparty in public records, and the gap is structural, since no register records leases and no issuer files them; hyperscalers buy land for cash while operators carry the recorded mortgage debt, and registered funds report 36.5% of the largest project bond issue but 3.33% of the second.
Working Paper 24 · October 2026 · PDF
Economic fixed obligations run 9 to 17 times reported debt at the largest AI borrowers; the funding architecture resembles pre-crisis shadow banking in structure, no fraud was found, and a solvent core funds a fragile periphery: Microsoft, Amazon, and Alphabet never breach coverage in any scenario while CoreWeave breaches from 2027.
Working Paper 12 · September 2026 · PDF
No observable system-level banking stress: 56 reconstructed banks hold capital and liquidity buffers intact, but long-end yields, bank equity drawdowns, and rate volatility are deteriorating; the binding stress sequence runs through funding before capital, and the 2023 backstops are gone.
Working Paper 10 · September 2026 · PDF
No aggregate crowding out: US corporate bond issuance averaged $1.90T per year in 2023–2025 versus $1.68T in 2015–2022 while AI-side capex reached $389B; spreads narrowed and the euro-area credit premium held flat, but manufacturing loan stocks contracted and AI financing shifted toward private credit, project vehicles, and foreign-currency bonds.
AI & Technology Economics
Working Paper 11 · September 2026 · PDF
Each GW of AI facility capacity must generate about $3.55B per year to clear hurdle rates; 2030 capacity forecasts span 108 to 490.7 GW once definitions are separated; the Jevons price channel does not support demand backfire; long-dated contracts written against short-lived accelerators are the central stranding mechanism.
Working Paper 06 · September 2026 · PDF
Hyperscaler capex grew 31.6% per year, from $23B in 2015 to $357.5B in 2025; staged 2030 US supply of 87–152 GW faces demand scenarios of 56–132 GW.
Working Paper 05 · September 2026 · PDF
Estimated learning rates of 38.6% for DRAM, 32.5% for computers, and 27.7% for solar PV; median real cost trend of −4.05% per year for manufactures versus +0.51% for commodities.
Energy & Commodities
Working Paper 23 · October 2026 · PDF
Turkey built 57 percent more capacity but generated only 34 percent more power; coal never left, gas became the swing fuel against hydro's swings, and the 2022 price spike hit $211/MWh before the lira absorbed the rest. Households were shielded; industry paid.
Working Paper 18 · October 2026 · PDF
World sesame comes from rainfed smallholders with no strategic stocks; weather explains none of country-level yield moves and world production none of US import prices because buyers switch origins. A 50% price spike costs the average McDonald's $98 a year, but allergen law and rigid supplier approvals turn a short shortage into an operational crisis.
Working Paper 15 · October 2026 · PDF
A composite burden index across 54 economies tests whether low nominal tariffs mean low effective burden: Nigeria's $0.049/kWh commercial tariff coexists with an effective burden 5 to 10 times higher, while Ireland's $0.29/kWh industrial price maps to the second-lightest burden; aluminium energy cost per tonne runs $280 in Qatar against $3,745 in Germany.
Working Paper 14 · September 2026 · PDF
Gold and silver curves sit in clear contango, copper\'s squeeze is regional and tariff-driven, and macro factors explain 18 to 41 percent of return variance; paper markets price macro risk and plumbing, not depletion.
Working Paper 13 · September 2026 · PDF
Copper\'s 2025 refined market ran a 380 kt surplus even as concentrate treatment charges hit zero; silver posted six straight deficits but the 2025 print was revised from 117.6 to 40.3 Moz; no metal shows acute physical scarcity.
Working Paper 08 · September 2026 · PDF
A 12-month energy-driven delay on a stylized 500 MW, $6.0B data center costs $857.6M in present value; waiting beats alternative power for delays under about two years.
Science & Engineering
Working Paper 28 · October 2026 · PDF
How sensitively does light propagation depend on initial conditions, and how much source information survives partial observation? The ratio of Lyapunov exponent to orbital frequency organizes predictability across black holes, a wormhole throat and plasma lenses; the one genuinely chaotic system computed, the Majumdar–Papapetrou di-hole, confines its chaos to a thin fractal set whose outcomes remain 99.3% recoverable under coarse-graining.
Working Paper 26 · October 2026 · PDF
Which geometries direct a broad family of light rays from one point to another with synchronized arrival? A survey of point-to-point null-geodesic convergence across general relativity, optical geometry, and laboratory analogues, with a proved obstruction lemma for wormhole refocusing and a new invertibility hierarchy for light-transport reconstruction.
Working Paper 16 · October 2026 · PDF
Canephora carries twice the caffeine of arabica, but arabica's sucrose and lipid advantage explains the sensory gap better than any single compound; dark roast "strength" is manufactured bitterness, not caffeine; caffeine per serving spans tenfold, from a 30 mg finjan of qahwa to a 300 mg Vietnamese phin.
Working Paper 01 · September 2026 · PDF
Lean load-adaptive switching between hydrogen and a hydrogen–methane blend gains 0.95 efficiency points over the best fixed fuel; blend scheduling on fixed hardware gains 0.015.
Decision Science
Working Paper 02 · September 2026 · PDF
Across seven processes, wrong-side outcome rates follow a common curve when sample size is normalized by the variance horizon; published odds describe the process, not the participant.