Research Area
Allocation Intelligence
Qualification answers whether a decision has earned the right to act. Allocation intelligence answers a fundamentally different question: among all qualified decisions, which deserve priority? These are distinct problems with distinct information requirements. Conflating them — treating qualification as sufficient for allocation — leads to naive equal-weighting among strategies with vastly different attractiveness profiles.
Research Thesis
Qualification answers whether a decision has earned the right to act. Allocation intelligence answers a fundamentally different question: among all qualified decisions, which deserve priority? These are distinct problems with distinct information requirements. Conflating them — treating qualification as sufficient for allocation — leads to naive equal-weighting among strategies with vastly different attractiveness profiles.
Allocation intelligence operates in the space between binary qualification and actual capital deployment. It produces a continuous score (0-100) across six dimensions: regime alignment, performance quality, lifecycle health, behavioral consistency, stability, and diversity contribution. The key insight is that returns alone are a poor allocation signal. A strategy with strong returns but active degradation flags, concentrated regime exposure, and behavioral drift may be far less attractive than a lower-returning strategy that is stable, diversifying, and regime-aligned.
The system is explicitly advisory — it ranks attractiveness but never determines capital amounts, position sizes, or deployment permission. This separation is deliberate: allocation intelligence provides decision support, while execution and risk management remain in systems designed for enforcement. The research challenge is calibrating the scoring weights so that the ranking genuinely reflects forward-looking attractiveness, not backward-looking performance.
Evolution of the Discipline
Understanding how this problem has been approached across eras reveals both recurring patterns and persistent gaps.
Equal Allocation
Pre-2000Limited computational capacity for portfolio optimization. Few systematic strategies to allocate across.
Equal-weight or discretionary allocation. Capital split evenly or based on manager conviction.
Ignored regime sensitivity, behavioral differences, and concentration risk. Equal allocation to correlated strategies amplified drawdowns.
Mean-Variance Optimization
2000-2010Required stable covariance estimates that rarely existed in practice.
Markowitz optimization, risk parity, minimum variance portfolios.
Extreme sensitivity to input estimates. Small changes in expected returns or correlations produced wildly different allocations. Solutions were mathematically elegant but practically fragile.
Factor-Based Allocation
2010-2020Factor definitions were debatable and factor crowding introduced systemic risk.
Allocate across risk factors (momentum, value, carry) rather than individual strategies.
Factor-based allocation assumed stable factor definitions and independent factor behavior. Factor correlations spiked during crises — precisely when diversification was most needed.
Multi-Signal Scoring
2020-PresentCombining heterogeneous signals (performance, health, regime, diversity) into a single actionable score without overfitting.
Weighted composite scoring with deterministic rules, explicit component breakdown, and advisory-only output.
Open question: whether fixed weights can capture the time-varying importance of different scoring components, or whether adaptive weighting is necessary.
Landscape Review
How different domains approach this problem today — their assumptions, strengths, weaknesses, and open questions.
Institutional Asset Management
Allocation is driven by risk budgets, mandate constraints, and expected return estimates from quantitative models.
Sophisticated risk modeling, scenario analysis, and regulatory compliance frameworks.
Models are backward-looking. Risk budgets are typically rebalanced quarterly, missing intra-period regime shifts. Allocation decisions are made by committees with long feedback loops.
Open question: Can continuous, automated allocation scoring provide better regime responsiveness than quarterly committee-based allocation?
Robo-Advisory
Risk tolerance questionnaires can determine appropriate asset allocation. Rebalancing is mechanical (threshold or calendar based).
Low-cost, disciplined, removes emotional bias from rebalancing decisions.
No strategy-level intelligence. Treats all assets within an allocation bucket as interchangeable. No awareness of behavioral consistency, degradation signals, or regime dynamics.
Open question: Can strategy-level allocation intelligence be applied to asset-level allocation without introducing excessive complexity?
Algorithmic Fund Management
Strategy allocation is based on backtested performance metrics (Sharpe, drawdown, correlation) with periodic review.
Quantitative, systematic, reproducible.
Backtest-derived metrics may not reflect live performance characteristics. No integration of lifecycle health signals, behavioral drift, or regime-conditional performance.
Open question: How should allocation intelligence weight live behavioral data versus backtested performance when the two diverge?
Core Mental Models
Reusable frameworks for thinking about this research area.
Qualification vs Allocation Separation
Qualification determines eligibility (binary). Allocation determines priority (continuous). These must be separate systems with separate ownership. A strategy can be qualified but unattractive, or qualified and highly attractive. Conflating the two leads to either over-restrictive qualification or naive allocation.
Returns Are Not Priority
High returns are a necessary condition for attractiveness but not a sufficient one. A high-return strategy with degradation flags, regime misalignment, and behavioral instability may be less attractive than a moderate-return strategy that is stable, diversifying, and healthy. Allocation intelligence captures this nuance.
Diversity as a First-Class Signal
A strategy's value to a portfolio depends not just on its individual quality but on what it adds to the portfolio that nothing else provides. A mediocre strategy that covers an unserved regime may be more valuable than an excellent strategy that duplicates existing coverage.
Advisory, Not Enforcement
Allocation intelligence produces recommendations, not constraints. The execution path never reads allocation scores. This separation ensures that allocation intelligence can be iteratively improved without risk to the safety-critical execution path.
Conservative Under Uncertainty
When data is missing or insufficient, the system defaults to low confidence rather than optimistic estimates. Unknown regime alignment scores 40, not 50. Missing behavioral data scores 50, not 75. The system should never recommend increasing allocation based on absence of negative signals.
Canonical Questions
The research questions that define this area. These are not rhetorical — they represent genuine uncertainties that guide investigation.
What is the optimal weighting across the six scoring components, and should weights be fixed or regime-adaptive?
Can allocation intelligence incorporate predicted regime transitions (from ML models) without introducing excessive false-signal risk?
How should asset-level correlation be integrated alongside regime-level diversity scoring to prevent hidden concentration in correlated assets?
At what confidence score threshold does the 'increase' recommendation become reliably predictive of forward returns?
Should allocation intelligence produce confidence intervals rather than point estimates, and how would interval-based recommendations change user behavior?
How should the system handle the tension between allocating to proven winners (exploitation) and maintaining allocation to strategies still accumulating evidence (exploration)?
Working Hypotheses
Not conclusions — working hypotheses. Each includes our current confidence level and the evidence or counterarguments we are aware of.
Regime alignment should be the highest-weighted component because regime mismatch is the strongest predictor of near-term strategy failure.
Regime-mismatched strategies show significantly worse performance. However, some behavioral families (directional momentum, scalp) appear regime-insensitive, suggesting uniform regime weighting may overpenalize certain strategy types.
Counterargument: Behavioral intelligence research suggests that behavioral family, not regime alignment, is the more predictive axis for some strategy types. A family-conditional weighting scheme may be more accurate.
Diversity contribution at 10% weight is underweighted relative to its portfolio-level importance.
Portfolio theory suggests diversification is one of the few 'free lunches' in investing. Current weighting treats it as a minor tiebreaker rather than a primary signal.
Counterargument: At the current portfolio scale (fewer than 30 strategies), diversity scoring is noisy. Increasing its weight could produce unstable recommendations as individual strategy additions and removals cause large swings.
Fixed scoring weights are sufficient for the current system scale and will not require dynamic adjustment for at least 6 months of operation.
The system has been in production since May 2026 with fixed weights and the recommendations appear directionally correct based on operator review.
Counterargument: No formal backtesting of allocation recommendations against realized forward returns has been conducted. Directional correctness assessed by operators may reflect confirmation bias.
Open Problems
Unsolved questions that define the frontier of this research area.
Position sizing integration: allocation intelligence currently ranks strategies but does not suggest capital amounts. Bridging from 'this is attractive' to 'allocate X dollars' requires a sizing engine that respects strategy-level risk limits, account-level safety constraints, and portfolio-level diversification targets simultaneously.
Temporal dynamics: the system evaluates attractiveness at a point in time (every 6 hours). It does not model how attractiveness is likely to change, even though regime transition probabilities and degradation trajectories are available. Incorporating temporal predictions could improve allocation timing but risks introducing prediction error.
Cross-strategy correlation: diversity scoring currently measures regime coverage. It does not measure equity-curve correlation between strategies, asset concentration, or directional overlap. Multiple BTC strategies appear 'diverse' by regime but carry concentrated asset risk.
Exploration budget: the system has no mechanism for allocating capital to strategies still accumulating evidence. All allocation flows to proven, high-confidence strategies — which is safe but may underinvest in strategies that need live data to complete their qualification journey.
Institutional multi-user context: allocation intelligence currently scores globally. In a multi-user deployment, different users may have different risk profiles, regime beliefs, and portfolio compositions. Per-user allocation scoring is an unsolved problem.
Implications
Capital Allocators
Allocation intelligence provides a systematic, multi-dimensional alternative to return-chasing. It surfaces attractiveness dimensions (regime alignment, behavioral health, portfolio diversity) that pure performance analysis misses.
Decision System Architects
The strict separation between qualification (eligibility) and allocation (priority) is a transferable design pattern. Any system that must choose among qualified options benefits from separating the 'is it allowed?' question from the 'how attractive is it?' question.
Risk Managers
Advisory allocation scoring provides an early warning layer. A strategy whose allocation score is declining, even while performance remains acceptable, may be approaching a regime or behavioral transition that performance metrics have not yet captured.
Related Research
Methodologies
Foundational Concepts
See these ideas implemented in Orqis