D-Series: systematic intelligence across changing regimes.

What changed - and what did not

The July update responds to two changes observed at the same time: market conditions have been rotating more quickly between directional, mean-reverting and defensive phases, while the usable universe of high-quality component strategies has narrowed. That combination matters. A portfolio can be well optimised and still become less effective if its inputs are losing relevance faster than capital can be rotated.

The response is a more selective and more frequently reassessed process. D-Series now applies less tolerance to degrading component strategies, introduces a greater contribution from internally developed quantitative models and reacts earlier when an exposure's original regime is weakening.

The investment mandate has not changed. These enhancements are integrated into the existing D-Series. They are not a separate product version, a retrospective optimisation of a short period or a promise to remove market risk. Systematic decision-making, layered limits and continuous oversight remain the foundation.

Why regime transitions became harder

The difficult part of a changing market is not simply volatility. It is sequence. A trend may dissolve into a short consolidation, appear to mean-revert and then re-emerge. Correlations can change during the same transition, and the strategy that was appropriate at the beginning may become marginal before a full capital rotation is complete.

D-Series therefore reassesses both regime confidence and the suitability of specific return sources more frequently. Volatility, correlations, market breadth, costs and changes in component-strategy quality are considered together. No single indicator is treated as definitive.

Greater selectivity also raises the importance of genuine signal independence. Several strategies can look different in isolation and still carry the same underlying dependency when markets become stressed. Robust portfolio construction requires evidence that a candidate adds a distinct source of return rather than another expression of an existing factor.

Three implemented changes

The update concentrates on the quality of the input universe, the controllable share of the signal stack and the speed at which unsuitable exposures are restricted. All three changes are part of the same decision process.

The practical objective is not to maximise the number of signals. It is to maintain a smaller, more defensible set of return sources and to spend less time in configurations whose original rationale has degraded.

Decision architecture: one controlled sequence

D-Series links market data, feature engineering, regime classification, strategy scoring, portfolio construction and live oversight. The output of one layer becomes the input to the next; no layer makes decisions in isolation.

• Market data covers prices, volatility, correlations and costs, subject to data-quality controls.
• Feature engineering applies robust transformations before information reaches a model.
• Regime classification produces probabilities rather than a single categorical answer.
• Strategy scoring evaluates stability, relevance and independence.
• Portfolio construction allocates subject to concentration, volatility and dependency constraints.
• Live oversight monitors deviations, realised costs and changes in strategy behaviour.

Four control channels run across that sequence: data quality, model robustness, exposure limits and cost discipline. This is important because a statistically valid signal can still be unsuitable if data quality deteriorates, dependencies rise or financing costs overwhelm its expected contribution.

Portfolio and risk controls

Risk is not treated as a number produced at the end of portfolio construction. It is managed through connected filters that begin with input quality and continue through live monitoring and costs.

Models can reduce inconsistency in decision-making; automation does not eliminate risk. Structural breaks, unusual illiquidity, data errors or unexpected interactions between constraints may still produce a response that is slower or different from expectations.

What production monitoring measures

The updated process is assessed through observable operating evidence rather than an illustrative performance chart. Monitoring focuses on whether the system behaves as designed across five areas:

• Selection consistency: score stability, sensitivity to inputs and the share of rejected candidates.
• Rotation efficiency: transition time, transaction costs and unwanted persistence in an outdated configuration.
• Source diversification: risk concentration, overlap and changing dependencies between signals.
• Financing and carry: swap points, position age and the incremental net cost of holding exposure.
• Regime confidence: agreement between classifiers, volatility, correlations and market breadth.

Degradation triggers a defined escalation path: routine review, enhanced assessment, exposure adjustment and, where necessary, full revalidation of the model or strategy logic.

How to read the initial evidence

At the date of the product update, the revised process had been live for approximately ten days. That period is too short to support statistical conclusions and is not presented as evidence of future performance. The initial assessment therefore concentrates on decision stability, rotation quality, cost behaviour and adherence to established limits.

Financing on legacy positions is monitored separately. Where a position carries materially negative swap or holding costs, the next phase allows for gradual closure when appropriate. The aim is to reduce unnecessary financing drag without creating uncontrolled rotation solely to improve a short reporting window.

Further calibration will be made only where the evidence supports it. That discipline matters more than the speed with which a new model can be placed into production.

Read the full product update

The complete July 2026 D-Series Product Update includes the market-regime framework, six-layer decision architecture, implemented changes, control framework, production-monitoring protocol and mandatory risk disclosures.