Today on The Systematic Desk: the infrastructure layer beneath systematic trading is compressing fast. Following recent breakthroughs in autonomous coding pipelines, ING is now building e-trading systems in hours with AI, while a new MCP-based portfolio analytics stack ships for fund operators. Plus: a novel backtesting diagnostic tackles overfitting, Bermuda broadens its on-chain economy push, Hong Kong finalizes its digital asset management licensing, and we look at the MiCA jurisdiction map as it stands in May 2026.
QuantJourney released two Model Context Protocol servers that connect Claude and other AI agents directly into IBOR/PMS, risk-engine, and market-data layers, enabling end-to-end investment workflows — concentration analysis, mark-to-market, VaR/CVaR, factor exposure, stress testing, and Brinson attribution. The article demonstrates ten live PM/CIO workflows with actual prompts and outputs, emphasizing full portfolio context rather than isolated price lookups.
Why it matters
This is the clearest working example to date of what 'agent-native' fund infrastructure actually looks like: the AI agent has controlled, authenticated access to portfolio state, real-time pricing, risk models, and audit trails through a standardized protocol layer. The MCP pattern — curated data + API access + human-approved execution gates — is directly replicable for anyone building systematic fund operations. The difference between this and a financial chatbot is the difference between a Bloomberg terminal and a Google search: context, not just data.
ING's electronic trading team is using Anthropic's Claude to generate trading infrastructure code — including entire credit e-trading systems and real-time analytics dashboards — in hours rather than weeks. The bank credits an in-house AI currency pricing model with a 50% increase in large-ticket FX trades. Separately, ING is cutting 1,250 operations roles globally, signaling a structural workforce shift as AI compresses development cycles.
Why it matters
This is a tier-1 bank deploying AI code generation in production trading infrastructure, not a proof-of-concept. The 50% improvement in large-ticket execution suggests the AI-generated tools are directly influencing market outcomes, not just developer velocity. For small systematic fund operators, the implication is that the cost structure of building and iterating on trading tools is dropping fast — but the governance challenge of deploying AI-generated code into systems routing billions in daily volume remains unresolved. The workforce reduction confirms this isn't additive; it's substitutive.
Martyn Tinsley introduces Walk Forward Correlation (WFC), a method that measures whether an entire in-sample parameter optimization surface retains predictive consistency out-of-sample — rather than testing whether a single best parameter survives. WFC computes correlation across the full parameter grid, separating predictability from profitability and identifying overfitting by evaluating optimization surface geometry rather than point estimates.
Why it matters
Most backtesting validation asks: does the best parameter work out-of-sample? WFC asks a better question: does the shape of the optimization surface transfer? If the entire surface correlates between in-sample and out-of-sample periods, the strategy captures structural market behavior rather than a lucky parameter. For systematic traders running parameter sweeps across FX, gold, or crypto signals, this is a complementary diagnostic to Sharpe deflation and walk-forward analysis — and it's computationally cheap to implement on existing backtesting infrastructure.
WuBlockchain details how perpetual futures liquidation cascades, rather than spot buying, now drive crypto price discovery. With perpetuals trading 10–15× spot volume, thin spot order books are exploited to trigger short liquidations, forcing market makers to hedge by buying spot — creating a positive-feedback spiral. The mechanism explains price moves that appear disconnected from fundamentals.
Why it matters
For anyone building crypto algorithmic strategies, this is a market microstructure primer on the dominant price formation mechanism. The liquidation-driven feedback loop means that funding rates, open interest concentration, and liquidation heat maps are more informative signals than spot order flow. Understanding how market-maker hedging amplifies liquidation cascades — and where the cascade breaks — directly informs position sizing, stop placement, and the design of mean-reversion strategies around liquidation events.
Hong Kong's FSTB and SFC published consultation conclusions on May 26 finalizing a licensing regime for virtual asset advisory and management services. Two license tracks mirror traditional SFO Type 4 and Type 9 licenses, with HK$5M paid-up capital, liquid capital thresholds of HK$100K–HK$3M depending on custodial activity, and no transitional deeming arrangements — meaning firms must be licensed before operating. Legislation is expected to be introduced to LegCo later in 2026.
Why it matters
The no-transition-period requirement means firms planning to manage digital assets in Hong Kong need to begin pre-application work now. The capital requirements (HK$5M base, up to HK$3M liquid capital for custodial firms) are materially lower than Singapore's but higher than most offshore alternatives. For operators evaluating Asian domiciles for tokenized fund management, Hong Kong is now a defined option with clear compliance costs, alongside MAS (SGD 250K base, strict physical presence) and VARA (AED 600K–1.5M). The alignment with existing SFO license types means existing regulated firms can extend rather than rebuild.
With MiCA enforcement now live, 204 CASPs are registered across the EU — 51 authorized in 2026 alone. Germany leads on count (55), Malta dominates exchanges (13 CASPs including OKX, Crypto.com, Gemini), Lithuania remains the startup default (fast 3–5 month process), and Estonia has sharply reversed to just 1 CASP. Authorization typically costs €200K–€475K in year one with 6–9 month timelines. The article provides a jurisdiction-by-jurisdiction breakdown with live register data.
Why it matters
This is the first data-driven map of where digital asset service providers are actually obtaining MiCA licenses — not marketing claims but registry counts. For operators evaluating where to domicile a tokenized fund or trading infrastructure, the data directly informs jurisdiction selection: Malta for exchange-grade credibility, Lithuania for lean English-language setup, or non-EU alternatives (Dubai/Singapore) if non-EU customers dominate. The emphasis on real local management substance — not mailbox incorporation — confirms that regulatory arbitrage within the EU is diminishing.
Hamilton Lane ($1T AUM) launched a tokenized share class for its Global Private Assets Fund with Allfunds Blockchain providing the technology layer, Apex Group ($3.5T assets serviced) as transfer agent, and BBVA Asset Management as the first investor and exclusive initial distributor across EMEA and Latin America. The structure enables blockchain-based subscription, administration, and servicing with reduced friction in cross-border distribution.
Why it matters
This is private markets tokenization moving from proof-of-concept to institutional deployment at scale. The partnership model — tech provider (Allfunds Blockchain), regulated transfer agent (Apex), and institutional bank distributor (BBVA) — shows how tokenized fund infrastructure avoids requiring standalone crypto licensing by layering blockchain distribution atop existing regulated fund and banking rails. For fund formation, this is the clearest template yet for how a tokenized share class can be distributed through traditional institutional channels without rebuilding compliance from scratch.
OKX announced Exchange OS, a protocol upgrade on its X Layer L2 enabling developers and institutions to deploy customized financial markets — spot, perpetuals, and prediction markets — through shared institutional-grade infrastructure: millisecond matching, 300,000 TPS, unified margin and settlement. Venue deployment requires staking OKB, with configurable compliance, oracle, and revenue models. OKX's own World Cup prediction market launches in June as the first venue on the infrastructure.
Why it matters
Exchange OS moves exchange mechanics from siloed venue control to a composable protocol layer. For fund operators evaluating where to deploy tokenized trading products, the unified margin system across spot, perpetuals, and outcome markets — combined with configurable compliance frameworks and isolated risk environments — directly addresses the operational fragmentation problem of running multi-strategy capital across multiple venues. The architecture is the on-chain analogue of what prime brokers provide in TradFi: shared infrastructure with configurable access controls.
Caladan, an Asia-headquartered market maker, launched an API liquidity product providing institutional streaming prices and RFQ for spot and perpetuals across 100+ tokens by aggregating pricing from 20+ on-chain protocols and 47+ centralized exchanges. Connectivity via FIX (4.2, 4.4, 5.0), platform partners (Talos, Finery Markets, CrossX), and settlement through Hidden Road, BitGo, Customers Bank CUBIX, and stablecoins.
Why it matters
For a systematic fund building digital asset execution infrastructure, this is a concrete build-vs-buy data point. Caladan's aggregated model addresses the fragmentation problem — no single venue provides reliable fills across long-tail tokens — through a single API with FIX connectivity that plugs into existing OMS/EMS stacks. The breadth of settlement options (bilateral, platform, fiat, stablecoin) is particularly relevant for fund operators who need flexibility across custodial and prime brokerage arrangements.
Senior data engineer Chris Riccomini discusses real-world patterns for using LLMs in financial data workflows: 41% hallucination rates on finance queries, the case for Substrait as a deterministic query intermediate representation replacing SQL for AI-generated pipelines, and the Ralph Loop iterative technique for managing LLM context and output quality. The interview distinguishes between decisioning layers (requiring explainability) and data engineering pipelines (enforcing invariants through quality gates).
Why it matters
The 41% hallucination rate on finance-specific queries establishes the baseline risk for anyone integrating AI into quantitative research or fund operations. Riccomini's distinction between using AI for data pipeline automation (where invariant checks catch errors) versus decision-making (where explainability is mandatory) is directly applicable to systematic fund infrastructure: LLMs can accelerate ETL and data wrangling while deterministic engines handle pricing, NAV calculation, and risk. The Ralph Loop — a structured iteration pattern for managing LLM context drift — is a practical technique for anyone building agentic data workflows.
Multi-strategy platforms including Millennium, Point72, and Citadel are expanding into private credit and asset-backed lending. Millennium is raising a $5B evergreen opportunities fund targeting complex private-market credit; Point72 hired Todd Hirsch from Blackstone to lead its private capital division. The move brings relative-value discipline and cross-asset risk infrastructure into a market historically dominated by direct lenders and specialist credit managers.
Why it matters
This signals a structural convergence: the operational infrastructure and risk systems built for liquid markets are now extending into private credit, bringing pod-level risk management, cross-asset correlation analysis, and institutional-scale data infrastructure to an asset class that has operated with less quantitative rigor. For emerging managers and fund builders, the competitive implication is clear — mega-platform entry compresses spreads and raises the operational bar. The evergreen fund structure (permanent capital, no vintage-year lockups) is worth watching as a template for tokenized private credit vehicles.
Bermuda's Monetary Authority announced plans to become the world's first economy to go fully on-chain, partnering with Circle (treasury management via Circle Mint), Coinbase (consumer/institutional onboarding), and Stellar (Bermuda Digital Dollar sovereign stablecoin). The BMA distributed $100 in USDC to residents for a physical pop-up marketplace pilot, with MoneyGram providing fiat conversion. Government digital asset payment acceptance starts with the Department of Motor Vehicles. Legislation updates to property, contract, and securities law are underway to recognize smart contracts as valid legal instruments.
Why it matters
This is the most ambitious sovereign blockchain deployment to date — not a CBDC study or a sandbox, but a phased plan to move government services, property law, and financial infrastructure onto distributed ledger rails. The operational details matter: MoneyGram as the fiat liquidity bridge, DMV as the first high-transaction agency, and Circle Mint for government treasury operations. For anyone evaluating the Bahamas or Bermuda as a base for building financial infrastructure, this signals a jurisdiction that is actively legislating smart-contract validity and building the rails, not just issuing licenses.
AI code generation enters production trading infrastructure ING building credit e-trading systems in hours with Claude, QuantJourney shipping MCP servers for portfolio analytics, and Cortex running 769 AI-reviewed PRs per month all point to a structural compression of the build cycle for financial software. The pattern is consistent: human oversight shifts from reviewing individual outputs to tuning policy and guardrails.
Tokenized fund infrastructure reaches multi-jurisdictional production scale Hamilton Lane launching a tokenized private assets share class with BBVA as distributor, Plume's Bermuda Class M vault license, and Bermuda's sovereign on-chain economy initiative demonstrate that tokenized fund structures are graduating from pilot to production across multiple regulatory frameworks simultaneously.
Regulatory maps are crystallizing — jurisdiction selection is now a data problem The MiCA CASP register now shows 204 authorized providers with jurisdiction-level granularity, Hong Kong has finalized capital requirements for digital asset managers, and UAE's VARA has formalized token design classification. Operators can now make evidence-based domicile decisions rather than relying on regulatory speculation.
Execution infrastructure consolidates across asset classes on single protocol layers OKX's Exchange OS, Caladan's 100+ token API liquidity product, and Grvt integrating RWA yield into perpetual futures all reflect a convergence where execution, margining, and settlement for multiple asset classes collapse into unified infrastructure stacks rather than siloed venue architectures.
Backtesting rigor intensifies as survivorship and overfitting problems get named Walk Forward Correlation offers a new diagnostic for parameter surface stability, while academic research documents how crypto momentum returns are driven entirely by coins that later collapse. Both point to a maturing quantitative culture where surface-level backtesting is no longer credible.
What to Expect
2026-06-01—Japan FSA stablecoin and intermediary regime takes effect — bond-backed reserves and new intermediary category go live.
2026-06-04—Caribbean Development Bank convenes Nassau seminar on innovative debt instruments and the Regional Multi-Guarantee Debt-for-Resilience (MGDR) initiative.
2026-06-30—Malta MFSA consultation on RWA tokenization closes — guidance expected on settlement finality, smart-contract legal standing, and custody.
2026-H2—Binance-BlockShoals Philippines StratBox sandbox testing projected to begin; Hong Kong crypto advisory/management legislation expected for LegCo introduction.
How We Built This Briefing
Every story, researched.
Every story verified across multiple sources before publication.
🔍
Scanned
Across multiple search engines and news databases
734
📖
Read in full
Every article opened, read, and evaluated
182
⭐
Published today
Ranked by importance and verified across sources
12
— The Systematic Desk
🎙 Listen as a podcast
Subscribe in your favorite podcast app to get each new briefing delivered automatically as audio.
Apple Podcasts
Library tab → ••• menu → Follow a Show by URL → paste