A public research archive for state-native macro relative-value intelligence.
This repository documents an experimental system for studying macro relationships as changing states rather than isolated price forecasts. The research is organized around four interpretable pair relationships spanning sovereign duration, energy, gold, the euro, and the Swiss franc. It combines causal macro framing, statistical diagnostics, architecture comparison, execution-aware simulation, and deliberately skeptical validation.
The work is not presented as a finished trading strategy. The strongest result is methodological: apparently attractive historical architectures become much less certain after temporal separation, combinatorial validation, execution stress, and option-liquidity calibration. That degradation is a central finding, not a footnote.
Can a small, explicit ontology of macro relative-value relationships support a decision system that remains interpretable and useful after regime changes, realistic execution constraints, and model-selection risk are taken seriously?
The current answer is provisional:
- The four-pair ontology produces coherent state descriptions and reusable research structure.
- Several construction families remain directionally stable across chronological splits.
- Architecture selection is unstable under combinatorial out-of-sample testing.
- Futures-led expressions retain more of the modeled relationship than naive ETF translations.
- Multi-leg options results are highly sensitive to quote quality and liquidity assumptions.
- Event windows contain both signal and tail risk; indiscriminate avoidance removes too much information.
- Forecast models are useful as scenario generators, not decision oracles.
- LLMs are more defensible as critics and explainers than as autonomous trade selectors.
- Forward-shadow evaluation has been specified and frozen, but has not accumulated enough cycles for a conclusion.
| Pair | Research interpretation | Primary uncertainty |
|---|---|---|
ZB_GC |
Sovereign duration versus monetary hard asset | Real yields, fiscal credibility, inflation repricing |
BZ_GC |
Growth and energy impulse versus haven demand | Supply shocks, growth expectations, geopolitical stress |
CHF_EUR |
Swiss credibility versus European divergence | Central-bank asymmetry, fragmentation, intervention |
CHF_GC |
Managed fiat refuge versus unmanaged monetary refuge | Risk regime, liquidity preference, monetary confidence |
These are not assumed to be stationary spreads. Each pair can move through continuation, compression, dislocation, and transition states. The research asks which state is active, how confidently it can be identified, and which expression surface - if any - is appropriate.
- State-Native Macro Relative-Value Intelligence - system thesis, ontology, canonical state, geometry, architecture, and conclusions.
- Architecture Selection Under Backtest-Overfitting Risk - temporal validation, CPCV/PBO, stress testing, and model-selection limits.
- Execution Realism and Expression Choice - futures, ETF, and option expression surfaces under liquidity and event constraints.
Publication PDFs are listed in PAPERS.md.
- Pair ontology
- Experiment lineage
- Validation protocol
- Event-conditioned risk
- Scenario generation
- Forward-shadow protocol
- LLM role boundary
- Figure atlas
- Literature and source map
The archive publishes research methods, architecture families, normalized or relative evidence, validation outcomes, negative findings, diagrams, and sanitized figure-level data. Values are indexed or expressed as ranks and relative retention where appropriate.
It does not publish current signals, positions, account values, broker snapshots, trade-level records, exact strategy parameters, raw licensed market data, credentials, or production configuration. This boundary preserves the ability to review the research without presenting a replicable live strategy or exposing private operational state.
Every empirical claim is mapped to a source experiment in EVIDENCE_LEDGER.md. Public charts use the common schema documented in REPRODUCIBILITY.md. The source system is a private working repository; this public archive is a curated research record, not an executable package.
The public evidence emphasizes falsification:
- a training winner fell below the median out-of-sample architecture in most tested folds;
- several option expressions lost most or all of their modeled edge after liquidity adjustment;
- at least one apparently promising regime architecture failed a chronological segment;
- blunt event avoidance materially reduced modeled opportunity;
- forward validation remains incomplete.
research archive / v1.0 / July 2026
No real-capital deployment is approved by this repository. Nothing here is financial advice, an investment recommendation, or an offer to transact.
Text, figures, and sanitized data are licensed under CC BY-NC 4.0. Third-party sources and underlying datasets retain their respective terms.
See CITATION.md.