Systematic Intelligence for
Digital Asset Markets.
Syrax Quant is the quantitative trading division of Syrax Global. We combine quantitative research, AI-driven market intelligence and automated execution to develop systematic, risk-controlled strategies across digital asset markets.
Quantitative trading is engineering, not prediction.
Markets are competitive systems. An edge is not found by forecasting harder than everyone else; it is found by identifying a structural reason that a price is where it is, and building a process that acts on that reason under control. Syrax Quant treats strategy development as an engineering discipline with the evidentiary standards that implies.
Mechanism before model
The economic rationale is stated first and tested second. This ordering is what separates a strategy from a pattern that happened to hold.
Evidence gates progress
Each stage of the research lifecycle has an explicit standard. Work that fails a gate stops there rather than advancing on judgement.
Costs are assumptions
Fees, slippage, funding, latency and market impact are modelled as first-class inputs, because a result that ignores them is not a result.
A backtest is a hypothesis, not a result.
The research programme runs a single lifecycle, and a strategy advances only by passing each stage in order. Most candidates do not reach the end, which is the point of having the process at all.
What the process guards against
- Overfitting — parameter sets tuned until history cooperates, with no reason to expect the fit to persist.
- Look-ahead bias — information reaching a decision earlier in the test than it could have in reality.
- Survivorship bias — universes that quietly exclude the assets and venues that failed.
- Unrealistic execution — fills assumed at prices no participant of that size would have received.
- Flattering cost models — omitted fees, funding or impact that turn a marginal idea into an attractive one.
- Regime dependence — behaviour that holds in one market condition and is never tested outside it.
The strategy universe under evaluation.
The programme assesses several families of systematic strategy. Each is described here as an area of research and assessment. None is described as deployed, and inclusion below is not a statement that a strategy has been selected.
Market-neutral
Relative-value and statistical approaches that seek to isolate a specific spread or dislocation while holding directional market exposure close to neutral.
Market-making
Continuous two-sided quoting with inventory and adverse-selection controls, where the mechanism is compensation for providing liquidity and bearing inventory risk.
Systematic directional
Rules-based participation in persistent market behaviours, sized by volatility and governed by the same risk limits as every other family.
Assessment is uniform across families: stated mechanism, capacity and liquidity constraints, behaviour under stress, correlation with the rest of the universe, and the operational cost of running it. A strategy that cannot be monitored and controlled is not a candidate regardless of its research profile.
Built as separable layers, each independently testable.
The platform is organised so that research cannot reach production by accident and execution cannot bypass risk. Venue specifics are abstracted, so no component above the execution layer is written against a particular exchange.
Data layer
Market and reference data captured with provenance and timestamps, normalised across venues, and versioned so a study can be reproduced exactly.
Research environment
Isolated from production by design. Reproducible studies, tracked datasets and recorded parameters, so a result can be re-derived rather than remembered.
Strategy engine
Strategies expressed against a common interface, which is what makes them comparable, testable in isolation and replaceable without touching the layers around them.
Execution engine
Order handling, venue abstraction and execution logic. Exchange-agnostic through an adapter layer, so venue selection stays an operational decision rather than an architectural one.
Risk engine
An independent layer, not a strategy feature. Limits are evaluated outside the strategy that they constrain, and cannot be relaxed by it.
Monitoring
Continuous instrumentation of positions, exposures, connectivity and control state, with alerting on the conditions that precede a problem rather than only on the problem.
Where AI is used — and where it is deliberately excluded.
AI supports systematic decision-making; it does not replace it. Research, market intelligence and performance analysis are AI-augmented. Risk control and execution are deterministic by design.
- Research support — hypothesis generation, feature exploration and surfacing relationships worth testing under the same evidence gates as any other idea.
- Market intelligence — synthesis of unstructured market context into a form a researcher can act on.
- Performance analysis — attribution and diagnostics that explain why behaviour changed.
- Anomaly surfacing — drawing attention to conditions that merit a human look.
- No model places an order.
- No model widens, raises or removes a risk limit.
- No model overrides a control or a kill switch.
- Every automated input carries a deterministic fallback, and the system is tested with every model disabled.
Risk is a layer, not a setting.
Risk control is independent of the strategies it governs, and authority is explicit at every level. No automated system has unrestricted authority to deploy capital.
Independent risk layer
Limits are evaluated outside the strategy. A strategy cannot grant itself more room, and a change to a limit is a governed action rather than a parameter edit.
Limit hierarchy
Constraints at strategy, portfolio and venue level, each binding independently, so a breach at any level acts without waiting for agreement from another.
Kill switches
Deterministic halt paths that stop new orders and reduce exposure. They are tested as a normal part of operation, not documented and left alone.
Drawdown governance
Defined responses to loss thresholds, decided in advance and applied mechanically, which is what stops a decision being made under pressure.
Counterparty & venue risk
Venues assessed on operational reliability, settlement behaviour and concentration, with exposure limits applied per venue.
Human oversight
Named accountability for changes to limits, strategy state and deployment. Automation executes a decision; it does not make the decision to commit capital.
Controls stated; specifics withheld.
The controls below describe how the environment is operated. Hosts, credentials and topology are deliberately not published.
API key isolation
Keys scoped to the narrowest function required, separated per environment, and rotated as a routine operation rather than an incident response.
No withdrawal permissions
Trading credentials are provisioned without withdrawal rights, so a compromised trading key cannot move assets off a venue.
IP whitelisting
Venue access restricted to known egress addresses, so a credential alone is not sufficient to reach an exchange.
Environment separation
Research, paper and production are separate environments with separate credentials. Promotion between them is deliberate and recorded.
Audit logging
Configuration changes, limit changes and order activity recorded so the sequence of events can be reconstructed after the fact.
Disaster recovery
Documented recovery procedures with defined objectives, exercised rather than assumed, including the case where a venue is unreachable.
Speak to the division.
Syrax Quant works with sophisticated institutional counterparties. If you would like to understand the research programme, the risk framework or the technology in more detail, send a short note and we will respond directly.
This page is provided for information only. It is not an offer, a solicitation, or an invitation to invest, and it does not constitute financial or investment advice. Syrax Quant trades proprietary capital only.
