LIVE
About The firm · the method · the people writing the briefings est. 2024 · London
About Arcane

An intelligence
service, not a platform.

Arcane exists because the people running books on professional desks were, as a group, underserved by the tools they had available — Bloomberg for data, Excel for thinking, and wire services that wrote for readers rather than for capital.

We are a research firm first. The software is the distribution channel for the research; the research is what actually matters. The model is named, versioned, and accountable. Every briefing is signed.

Founded
2024
London & Zürich
People
28
17 research · 11 engineering
Assets modelled
4,200
across 18 venues
Seats active
6,419
across 412 firms
Operating principles

Six commitments that run the firm.

Principle 01

Every model is named and dated.

VIDI v8.4 is the current regime engine. Each version ships with a changelog. We do not silently retrain and push.

Principle 02

The briefing is signed.

A person's byline goes on every briefing. If the call is wrong, it's wrong with a name attached. Accountability lives above automation.

Principle 03

Uncertainty is quantified, not hidden.

Every signal publishes its confidence, its stability window, and the conditions that would flip it. We report spreads, not point estimates.

Principle 04

We don't take your trade data.

The model is trained on public market microstructure. Your positions, orders, and P&L stay with your broker. Arcane will never ask for them.

Principle 05

No advisor relationship.

Arcane is not a registered investment adviser or broker-dealer. Nothing published is advice. This is research you apply to your own judgment.

Principle 06

The product is the research.

The software is a distribution channel. If the work is wrong, no amount of UX fixes it. If it's right, the UI can be plain and still worth reading.

Methodology

How a regime
is read.

A regime is a period during which an asset's return distribution, correlation structure, and sentiment proxies are internally consistent. Regimes don't end cleanly — they soften at the edges, transition through a noisier middle, and resolve into the next state. The point of Arcane is to read those edges earlier than headline price action does.

Every regime reading passes through five stages. Signals from any stage can be consumed independently via Build, but the published reading is the full pipeline.

Stage 01
Ingest
Market microstructure — trades, quotes, options surfaces, futures basis, ETF flows — plus a macro layer of policy rates, spreads, and cross-asset positioning. No alternative-data pretension, no satellite imagery, no scraped social sentiment.
p50 · 4 ms
Stage 02
Factor decomposition
Each asset is projected into a factor space built from 47 structural readings — breadth, dispersion, concentration, dealer gamma posture, carry, quality, and several proprietary ones. Factors are observable; nothing here is a black box.
47 factors · 30-day window
Stage 03
Regime classification
VIDI assigns probabilistic membership to a discrete regime set: Institutional, Growth, Defensive, and four sub-regimes used only inside the model. Classification is probabilistic, not binary — the number we publish is the one we would bet on.
VIDI v8.4 · HMM ensemble
Stage 04
Transition detection
A separate detector watches for the statistical signature of regime change — not the change itself. Early transitions are intentionally noisy; we surface them as Elevated rather than High Risk until confirmation. False positives are a tax we pay for earliness.
p50 alert · 2.2 s
Stage 05
Editorial pass
A human reads the automated output and decides what a briefing should say. The model writes data; the human writes judgment. If the model is confident and the human disagrees, the briefing reflects both — with the disagreement explicit.
Every briefing · every time
The people

Who's actually writing this.

Arcane is small on purpose. Twenty-eight people, no chief-of-anything, nobody's title is invented. The research bench reports to the CIO; engineering reports to the CTO; everyone reads briefings before publication.

MH
Mira Halvorsen
Chief Investment Officer · co-founder
15y at Brevan Howard & Millennium. Built the first regime engine at MOCA in 2019.
SO
Saoirse O'Connor
Chief Technology Officer · co-founder
Ex-Palantir Foundry, ex-Citadel Tech. Authors the Build API.
TR
Takeshi Ramirez
Head of Quant Research
Wrote his PhD on regime-switching HMMs before most firms had heard of them.
EW
Elena Wójcik
Head of Editorial
Former senior markets editor, Financial Times. Signs the midday briefing most days.
AJ
Aarav Joshi
Staff Quant · factor modelling
Previously Two Sigma. Maintains the factor decomposition engine.
LC
Liesel Chen
Staff Quant · transition detection
Caltech physics; wrote the change-point detector VIDI uses at Stage 04.
NK
Noor Khatun
Principal Engineer · real-time infra
Built the sub-50ms delivery bus that powers alerts and webhooks.
BO
Benedikt Ohlsen
Head of Institutional
20y relationship coverage at JP Morgan & UBS. Runs institutional onboarding.
Capitalization

Privately held. Aligned backers.

We've taken a single Series A from investors whose portfolio and time horizons match ours. No growth-at-all-costs pressure; no pivot to consumer; no pressure to become a broker.

Index Ventures
LEAD · SERIES A · 2025
Point72 Ventures
STRATEGIC · 2025
LocalGlobe
SEED · 2024
12 individuals
ANGELS · OPERATORS · 2024–25

If what we've described here matches how you actually think about markets, Arcane is probably the right read.