MATS: An AI‑First Proprietary Trading System — The Oracle’s First Three Trials

It’s been a while for me to refine my thought on this idea, since 2016, till LLM. My first thought, MoDeA (mixture of domain expert and agent), reflected my interpretation of how AI could be embedded into established organisations. But a more bold approach kept tugging at me, not AI inside an organisation, but an organisation built around AI from the ground up. An AI‑first system.  

Somewhere along the way, that idea crystallised into something sharper:a multi‑agent proprietary trading system (MATS).  

This post is the beginning of what I expect to be a year‑long series - a way to document the evolution of this architecture, and a way to reason about markets without drowning in noise, narratives, and the endless churn of financial content. What started as a desire for clarity has evolved into something far more ambitious: a multi‑agent, AI‑first proprietary trading architecture.

Paradigm Selection: Why Agentic, Not Skills

The AI industry moves at a speed that makes even seasoned engineers feel like they’re sprinting on a treadmill. Anthropic’s recent skills pattern is elegant - modular, scalable, engineered for large teams.

But for a "one‑person, lean startup", the agentic approach is simply faster.There are more frameworks. More low‑code options. More ways to test feasibility without building an entire engineering pipeline.

Skills are great for scale. Agents are great for velocity. And right now, velocity matters more.

The Architecture: Five Agents in the Arena

The system is simple in structure, but deep in behaviour. There are 5 agents, each mirroring a real role inside a trading firm:The Market Sentinel - The Oracle, The Quant Researcher, The Quant Developer, The Adversarial Auditor, and The Trading Desk, minimal but essential.

The workflow looks like this:User provides information or a scenario. The Oracle transforms it into a trading concept or macro thesis. The Quant Researcher turns that concept into a mathematical model. The Quant Developer converts the model into production‑ready code. The Adversarial Auditor attacks the logic, finds weaknesses, and forces iteration. Once approved, The Trading Desk executes with discipline.

Each agent is guard‑railed by proprietary system prompts, specialised through proprietary playbooks via RAG, empowered by proprietary domain‑specific tools via MCP

And the first team member to come online is The Oracle

Testing the Oracle: 3 baseline trials

Before I let the Oracle inform any downstream agent, I needed to understand how it thinks. So I ran three baseline tests, each probing a different dimension of reasoning.

Use Case 1 — The Naive Retail Investor

A simple prompt:
If I were a retail investor buying a global ETF monthly, what should I know about 2026?

The response:

Thesis: Tactical monthly accumulation of the MSCI ACWI in 2026 should focus on the H1 Federal Reserve leadership transition and the historically significant Q3 midterm election "slump" to optimize entry ahead of a high-probability 16.3% average post-election recovery rally.

Macro Context:
Jerome Powell's term as Fed Chair expires May 15, 2026, with the January 31 expiration of Governor Kugler’s term serving as the catalyst for naming a "shadow chair" successor and triggering a mid-year volatility window.

U.S. midterm election years (Year 2 of the presidential cycle) are historically the most volatile, with the S&P 500 typically bottoming in September (averaging -0.72% returns) before resolving into a powerful post-election recovery.

Monetary policy is forecasted to reach a terminal neutral rate of 3.25% by mid-2026, shifting market focus from broad liquidity-driven "narratives" to forensic evaluations of corporate cash flow and pricing power.

Tactical entry is optimized by targeting the Mondays preceding quarterly "Triple Witching" (March 20, June 18, September 18, December 18) or MSCI index rebalancing effective dates (March 2, June 1, September 1, December 1) to capture seasonal "Monday effect" dips before systematic institutional flows stabilize.

Risk Factors:
Persistence of high inflation (sticky at 3%) or an unexpected "hard landing" in the labor market (unemployment rising to 4.6%) could prevent the FOMC from reaching its projected neutral rate, leading to sustained equity weakness.

Geopolitical flashpoints, such as the February 2026 expiration of the New START nuclear treaty or renewed US-China trade tensions following 2025 sunset clauses, could trigger global "risk-off" sentiment regardless of the political cycle.

A forensic shift in market sentiment from AI "promise" to margin-based results could cause a sharp derating of the mega-cap tech constituents that dominate the MSCI ACWI if monetization fails to materialize by 2026.7

Use case test observation:

This test was about clarity, discipline, and non‑hallucination. It didn’t predict, it contextualised, It behaved like a calm macro analyst explaining the world to someone who is a newbie.

Use Case 2 — The Experienced Investor’s Thinking Framework

This time, I fed the Oracle a multi‑layered mental model, the prompts contains: structural incentives, geopolitical alignment, liquidity regimes, reflexivity loops and institutional flows:

The test prompt: 
I recently encountered a compelling historical framework for analysing U.S. government debt through the lens of British fiscal history. At several points in the 18th and 19th centuries, Britain’s public‑debt‑to‑national‑income ratio exceeded 200%. Yet this did not trigger national collapse; instead, it coincided with the emergence of long‑run prosperity. 

A key reason was that British government debt was overwhelmingly held by domestic elites. This created strong alignment between the state and its wealth‑holding class — effectively making Britain function like a shareholder‑owned enterprise. In this view, the debt ratio was less a solvency problem and more a question of internal ownership structure.

By contrast, today’s U.S. national debt is held largely by global institutions, and American elites themselves operate globally diversified corporations and portfolios. This creates a structural misalignment: the U.S. behaves less like a shareholder‑aligned entity and more like a “debt company” whose obligations are externally owned. Under this framework, the sustainability of U.S. debt becomes a question of geopolitical and ownership structure rather than pure fiscal arithmetic.

From this perspective, de‑globalisation, reshoring, and tighter alliances with countries sharing strategic interests could be interpreted as mechanisms to re‑internalise debt ownership and restore incentive alignment.

Can you explore this framework in depth and extract potential global‑macro trading concepts or regime‑shift themes that might emerge from it?

This wasn’t a question. It was a worldview. 

The response:

Establishing your vision through the lens of a Global Macro Strategist, the transition from a "globalized debt company" to a "sovereign shareholder entity" is more than a historical analogy—it is a live structural shift. The 19th-century British model worked because the debt was a transfer between the state and its own "shareholders" (the domestic elite), creating a self-reinforcing loop of growth and stability. In contrast, the US has spent decades as a "debt company" with globalized liabilities, where foreign holders (currently ~30% of public debt) seek exit liquidity rather than long-term national prosperity.

The "Alpha" in this concept lies in the forced reshoring of the Treasury market. As the US pivots toward "friend-shoring" and domestic alignment, we are moving toward a regime of "Financial Repression 2.0"—where domestic capital is incentivized or mandated to absorb the debt, effectively converting "foreign liabilities" into "domestic equity".

Thesis: Long US "Real Assets" and "Domestic Compounders" funded by a short-duration/short-USD position, betting on a structural shift where the US government "re-aligns" incentives by forcing domestic capital to prioritize US debt and growth over global diversification.

Macro Context:
Declining Foreign "Official" Demand: Foreign central bank ownership of US Treasuries has dropped from a peak of 49% in 2011 to roughly 29-32% in 2025. This creates a "funding gap" that must be filled by domestic institutions and "friends" (e.g., UK, Japan), effectively shifting the US toward a "shareholder" model with aligned allies.

Deglobalization Stagflation: De-globalization and stronger "home bias" are historically stagflationary, raising domestic inflation while potentially lowering long-term growth as global trade efficiencies are sacrificed for national security.

Fiscal Multipliers & Ownership: Research indicates that fiscal multipliers are smaller when debt is held domestically (like 1950s US or modern Japan) because domestic debt acquisition "crowds out" private consumption, whereas foreign-held debt allows for larger immediate stimulative effects. A shift to domestic ownership implies a more "disciplined" but slower-growth "shareholder" economy.

Regime Shift to Safety & Convexity: As global institutions rotate out of "risk-on" global assets, the US domestic market becomes the primary "Safe Haven," but with higher volatility due to the "de-risking" of international companies' global portfolios.

Primary Assets Affected:
Short: Long-end US Treasuries (10y-30y): Increased domestic supply and inflationary pressures from deglobalization should drive yields higher and prices lower.

Long: US Real Assets (TIPS, Real Estate, Infrastructure): These act as hedges against the "Financial Repression" needed to inflate away the domestic debt burden.

Long: "Friend-Shoring" Leaders (Mexico, UK, Japan): These nations are becoming the primary external "shareholders" of the US macro-entity.

Short: Global Growth Equities: Companies with heavy reliance on unaligned global supply chains will see margins compressed by reshoring costs.

Risk Factors:
Deflationary Shock: A sudden global recession would cause a "flight to quality," temporarily surging demand for all US Treasuries (including foreign), delaying the reshoring thesis.

Dollar Debasement: If the "shareholder alignment" fails and the US resorts to aggressive money printing, the USD could lose its status faster than domestic capital can absorb the debt, leading to a "sell everything" event.

Policy Reversal: A return to aggressive multilateralism and "hyper-globalization" would re-validate the "global debt company" model and punish domestic-centric trades.

Use case test observation:
The Oracle didn’t just respond, it expanded the framework. It connected incentives, flows, policy, and history into a coherent macro thesis. This was the moment it stopped feeling like a prototype and started feeling like a collaborator. 

However, it missed the recent GENIUS act and stable coin factor. So the following prompts from human in the loop is needed, along with deep thinking/research mode

Use Case 3 — Alternative Testing (Sports → Macro)

I gave the Oracle a BBC Sports article about Premier League penalty records - something completely unrelated to finance — and asked it to derive a macro trading thesis.

The response:

Thesis:
Exploiting a "bull-steepening" regime as the Federal Reserve transitions from aggressive easing to a cautious, data-dependent pause, favoring high-quality cyclicals and front-end rate sensitivity over defensive duration.

Macro Context:
Monetary Policy Pivot: The Federal Reserve cut rates by 25bps in December 2025 to a range of 3.50%–3.75%, but three dissents and hawkish language regarding "the extent and timing" of future moves suggest an imminent pause.

Yield Curve Normalization: The 2y–10y spread has moved into positive territory (~32 bps), signaling a transition from an inverted "recession-scare" regime to a "bull-steepener" where short-end rates fall faster than the long-end due to easing.

Labor Market Cooling: Nonfarm payrolls grew by a soft 64k in November with unemployment rising to 4.6%, providing the "cover" for the Fed to ease even as core inflation remains sticky at 3.0%.

Equity Regime Shift: Markets are moving from "Growth at Any Price" to a more volatile "Value and Cyclical" leadership as higher real rates pressure tech multiples while fiscal stimulus (One Big Beautiful Bill Act) supports broader GDP growth.

Alpha Driver:
The core opportunity lies in the information shock of the Fed’s "Hawkish Cut". While headline rates fell, the shift in guidance creates a "higher-for-longer" floor (estimated at 3.0%) that prevents long-end yields from collapsing, rewarding trades that benefit from a steeper curve and resilient domestic growth rather than a simple flight to safety.

Primary Assets Affected:
US Treasuries: Long position in 2-year notes (front-end sensitivity) while staying underweight/short long-duration (30y) bonds to capture curve steepening.

Equities (S&P 500/Small Caps): Overweight Small Caps (Russell 2000) and rate-sensitive cyclicals that benefit from lower borrowing costs and a "soft landing" scenario.

Currencies (USD): Short USD bias as rate differentials narrow against the Euro, with the DXY approaching structural weakness near the 98-point level.

Commodities/Gold: Long Gold as a hedge against "debasement trades" and fiscal sustainability concerns, despite USD volatility.

Risk Factors:
Inflation Resurgence: If core PCE stalls above 3% due to tariff passthrough, the Fed may be forced to hike, causing a "bear-flattening" that crushes both equities and bonds.

Fiscal Shock: Massive government deficits could drive a "term premium" spike in long-end yields, causing a disorderly steepening that triggers a "sell-everything" liquidity event.

Recessionary Gap: If the labor market loosening (4.6% unemployment) accelerates into a hard landing, the 60/40 correlation could turn positive, invalidating cyclical equity bets.

Use case test observation:
This was the real stress test, can the Oracle abstract across domains, I had a follow up prompt asking it to explain how it dd it.

Below is the mapping table

Sports Concept	Macro Equivalent	Trading Action
100% Conversion	Fed Policy Credibility	Long Quality/Alpha assets
Waiting for the Keeper	Data Dependency	Front-end Treasury sensitivity
Record Breaking	Yield Curve Normalization	Curve Steepener (2s10s)
The Miss (Risk)	Policy Error / Inflation	Credit Protection / Volatility

From that, it built a coherent macro thesis involving: bull‑steepening, hawkish cuts, front‑end sensitivity, and volatility windows.

Summary: What the Three Tests Reveal
Each test probes a different dimension of intelligence: Retail → clarity, Experienced investor → depth  
Alternative domain → abstraction, The Oracle showed some interesting robustness.