Systematic, risk-managed research &
market intelligence

Tail risk, regime transitions, and market-structure shifts — measured rigorously, validated forward, and claimed only when proven.

250+registered hypotheses
~100recorded null results
5papers on SSRN
7+
Independent Signal Engines
167
Academic References
55000+
Prospective Observations
24/7
Live Computation
Research & Perspectives
The Big Ideas
Autonomous allocation changes everything: the structural advantages of continuous, automated portfolio management — and why they compound over time.
75 days
Traditional Latency
<1 sec
Autonomous Latency
$1.9M
30yr Fee Differential
Explore Ideas
Signals we can measure. Claims we can defend.
Our research measures each signal's overlap with conventional quant factors and records the result in the register. Each capability below is generated using the statistics of stochastic processes.
 MeasurementReading
01Structural Tail Risk Estimation
Maintains a continuous reading of structural stress in market factor geometry. The reading is pre-registered in our research log and evaluated forward-prospectively; it is a monitored risk state under forward test.
Forward test · risk-state reading
03Adaptive Rebalancing Triggers
Schedules portfolio maintenance from measured change in factor structure rather than from the calendar. The scheduling mechanism is live; whether it improves on calendar rebalancing remains an open question in our research log.
LIVE · event-driven trigger cadence
04Market Regime Classification
Classifies the current market environment into one of four regimes on a continuous coordinate system. The classifier is under forward evaluation; regime labels inform risk posture rather than prescribe an allocation.
4 · quadrant classifier · updated every cycle
06Multi-Scale Regime Velocity
Analyzes how market dynamics shift across timescales — from intraday through weekly. Cross-scale divergence surfaces when short- and long-horizon dynamics decouple and feeds the regime read as one input among several — treated as a hypothesis under forward validation, not a settled early-warning signal.
intraday–weekly · cross-scale regime input

Lineage, stated plainly: the eigenspace-rotation measurement builds on established results on the dynamics of correlation-matrix eigenvectors (Allez & Bouchaud, 2012) and on principal-angle geometry between subspaces. Our contributions are the topological stability measure published in the SSRN series and the use of the rotation rate as a live rebalancing trigger.

What the output looks like
These are representative instruments from the research feed, shown with the portfolio decision each one informs.
Structural Fragility
P( drawdown > 5% · 30d forward )
illustrative ·
Elevated
Trims concentrated equity, adds tail hedges and raises cash the moment fragility enters the elevated band — a rules-based response, not a forecast.
Factor Eigenspace
Principal-subspace rotation rate
per day · illustrative
Rotating
Tracks the orientation of the hidden factor structure over time and reports when it is rotating.
Multi-Horizon Risk
Stress intensity across forecast horizons · schematic
horizons elevated · schematic
Distinguishes the horizon at which stress concentrates — short-fuse shocks and slow regime drift call for different responses. Hedge selection from this reading is a research direction, recorded as such in the register.
Pre-registered, validated forward
Every signal is pre-registered with a frozen threshold before observations are recorded, then validated forward under statistical rigor. A signal is not promoted to live sizing until it clears the Harvey (2016) t>3 bar on a frozen, out-of-sample window.
Pre-Registered & Tracking
Live

Tail Risk Monitoring

Primary crash signal spanning SPY, QQQ, IWM, DIA, and EEM, with a frozen threshold and lead-time window pre-registered in our research log. It continues live tracking for prospective Harvey-significance validation — a forward, out-of-sample evaluation rather than a retrospective replay of historical stress events.

Live Deployment
24/7

Forward-Validation Harness

A multi-asset research book runs live and continuously — equities, digital assets, and derivatives — as the prospective, out-of-sample test bed for the signal registry. Allocation and risk management operate automatically so that every frozen-threshold hypothesis accrues genuine forward observations, not in-sample fits.

The live book is a validation instrument. We do not headline live returns: a signal earns a capital allocation only after it clears statistical validation, not before.
Pre-Registered
11

Hypotheses Under Test

Every signal has a pre-registered hypothesis with defined null conditions, required sample sizes, and multiple-testing correction. Over 55,000 prospective observations recorded across the registry. Validation architecture enforces statistical discipline automatically — signals are not promoted to live sizing until they clear the Harvey 2016 t>3 threshold on a frozen sample.

Live Tracking
27

Numerai Signals · Rounds Resolved

Our research model submits weekly to the Numerai Signals blind tournament for independent third-party validation. Resolved rounds counted here; full per-round scores published on the performance page. We do not claim a score until the tournament’s own robust-evaluation quorum is reached — Numerai’s Alpha metric requires a multi-month track record before it stabilises.

External validation is a multi-month process. We report resolved scores as they come in — not as they were expected to.
What we tested — and what we killed
Most “AI market prediction” sells a confident guess about the one thing that is provably hard to estimate: which way price moves next. We do the opposite. We measure what markets actually reveal, keep only the edges that survive out-of-sample testing, and openly record the ones that don’t. You can trust what we do claim precisely because we show you what we rejected.
CONFIRMED
Volatility-risk-premium income. Selling insurance the market systematically overpays for. Survives multiple-testing correction across stress periods in the historical characterisation; forward accrual continues in the live book.
CONFIRMED
Trend-following (sign, not magnitude). You can’t forecast how much an asset returns, but its direction carries and self-corrects. A confirmed, risk-managed carry of that persistence.
FORWARD
Positioning / crowding density. A measurement of where capital is concentrated — not a price forecast. Pre-registered and accruing out-of-sample observations before any claim is made.
REJECTED
“Predict the next move.” An asset’s expected return is provably hard to estimate from its own price history — the reason most AI-prediction products quietly fail. We tested it. It does not survive. We don’t sell it.
REJECTED
Cross-asset “gap” trades. Tested across hundreds of asset pairs over two decades. Efficient markets have already arbitraged the tradeable ones flat. No durable edge — we moved on.
Most hypotheses we test are rejected — that is the discipline, not a shortcoming. The full methodology and the register of results are in our published research.
Rigorous foundations, peer-quality standards
Our signal architecture applies the statistics of stochastic processes, grounded in 167+ academic references. Six papers from the research program are published on SSRN, with further work in the publication pipeline. The full-length working papers behind them (the whitepaper, educational series, literature survey, and system specification) are in preparation for formal publication and are available on request; proprietary implementation details remain internal.
Derivative design as a research tool
Beyond pricing existing contracts, the same mathematical architecture can be used to explore new payoff structures — for example, instruments referenced to signal quality, regime transitions, or factor stability. This is an active research direction, not a product offering: candidate structures are studied and stress-tested internally, not sold.
Research, market intelligence, and related publications
Selected academic work, market observations, and internal research directly connected to the mathematical methods underpinning our signal engines.

Twelve articles — five results, three methods, four essays. Full index

Subscriptions are not currently offered
The published research, papers, and insights remain freely available — read the research.

Download a sample report, or reach out to discuss how Quark fits your workflow.

SAMPLE REPORT  ·  [email protected]  ·  X/@QUARKRESEARCH