AI Portfolio — Strategies Dashboard

I. STRATEGY PIPELINE ideation → testing → shadow → live → sunset

Ideation 2

S-487 — Regime-Aware Value Rebound

Data: fundamentalsDRAFT

Hypothesis: value rebounds after risk-off when credit spreads peak & revisions stabilize.

S-502 — Supply Chain Bottleneck Shock

Data: shipping/satelliteDRAFT

Hypothesis: bottleneck intensity predicts near-term margin revisions in EU industrials.

Research & Feature Build 3

S-266 — Earnings Call “Uncertainty” NLP

Data: transcriptsRESEARCH

Build: embeddings → uncertainty index → sector-neutral long/short selection.

S-318 — Crowding / Flow Reversal

Data: flows/optsRESEARCH

Detect crowding + reversal. Needs cost model stability and borrow constraints.

S-355 — Quality Carry + Low Vol Tilt

Data: factorsRESEARCH

Reduce drawdown; ensure beta remains within portfolio target band.

Statistical Testing 2

S-311 — Event-Vol Overlay (Index Options)

WF: MON · Boot: PASSSHADOW

Testing: calendar triggers + tail scenarios + premium spend constraints (≤ 25bps).

S-404 — Short “Margin Roll-Down” Basket

p=0.04 · t=1.8TEST

Borderline significance. Needs stronger OOS + regime stability before shadow.

Deployed (Live) 3

S-102 — AI Infra Demand (Alt-data)

OOS: 1.92LIVE

Data: supplier checks + logistics + NLP demand. Trades: ASML/IFX/STM + hedged.

S-041 — Quality × Revisions (Fundamental)

OOS: 1.58LIVE

Data: revisions + profitability/ROIC + balance sheet. Low drawdown posture.

S-208 — Services Displacement (AI Agents)

OOS: 1.34LIVE

Data: labor intensity + pricing power + alt/NLP. Borrow + squeeze constraints enforced.

Sunset / Archived 2

S-099 — Mean Reversion (Microstructure)

Decay: HIGHSUNSET

Reason: slippage + turnover breach + failed walk-forward. Removed from execution.

S-144 — “Value Trap” Screen (Legacy)

OOS: 0.31ARCHIVE

Reason: unstable across regimes; correlation with market factor too high.

Researcher View — What “Adaptive Strategies” Means

The AI generates hypotheses inside the L/S Equity mandate, builds features from approved data feeds, and runs a multi-stage test suite (walk-forward, bootstraps, permutation, cost/liquidity models, regime checks, and portfolio risk-gates). Strategies can run in shadow to validate live execution conditions before promotion. If alpha decays or risk drifts, strategies are sunset automatically with reason codes and full lineage.

II. LIVE STRATEGIES performance + drift + portfolio linkage

Live Strategy Monitor

Strategy	Status	MTD P&L	OOS Sharpe	Drift z	Trades (7D)	Portfolio Link
S-102 — AI Infra Demand	LIVE CORE	+$4.86M	1.92	+1.7	21	View in Portfolio
S-041 — Quality × Revisions	LIVE CORE	+$2.94M	1.58	+0.6	14	View in Portfolio
S-208 — Services Displacement	LIVE TACTICAL	+$0.62M	1.34	+2.1	9	View in Portfolio

Trades by Strategy (demo linkage)

Time	Strategy	Action	Ticker	Notional	Reason
14:23	S-102	ADD	IFX	$1.45M	Power-semi inflection + supplier cadence improving
10:30	S-041	ADD	SAP	$2.40M	Revisions up + backlog acceleration; quality screen pass
06:30	S-208	TIGHTEN	KER	—	Squeeze-risk control; stop tightened due to gap volatility
09:45	S-311	ROLL	SX5E Puts	$0.14M	Calendar volatility shift; premium spend within budget

Drift z-score compares live feature distribution vs training window; breaches trigger auto-demotion to shadow + review.

III. STRATEGY DETAILS hypothesis + data + test suite + decision — All Live Strategies

Strategy 1: S-102 (AI Infra Demand)

Overview

Hypothesis

Data Sources

Statistical Tests

Performance

Signal Processing

Governance

Strategy Summary

owner: AI · scope: EU L/S equity

Hypothesis: upstream demand signals (supplier, shipments, NLP) lead earnings revisions 30–60D ahead.
Universe: STOXX Europe 600 + EU semicap suppliers (liquidity & ADV filters).
Signals: demand index (alt-data) + revisions momentum + valuation sanity check.
Performance: OOS Sharpe 1.92, MTD P&L +$4.86M, 21 trades in last 7 days.
Status: Live, 8 positions, 28.4% of NAV exposure.

Linked portfolio positions: S-102 bucket

Hypothesis & Signal Construction

Core Investment Thesis

Primary Hypothesis

Upstream demand signals (supplier checks, shipment cadence, NLP sentiment) lead earnings revisions by 30–60 days. This predictive relationship is strongest in semicap suppliers where supply chain visibility is high and demand signals are early indicators of end-market strength.

Signal Construction

Signal Component	Weight	Source	Lookback
Alt-Data: Shipment Cadence	40%	Supply chain scans	30D
NLP: Demand Sentiment	30%	News & transcripts	30D
Fundamentals: Revisions	25%	Earnings revisions	30D
Valuation: Relative	5%	Valuation screens	T-1

Universe & Filters

Universe: STOXX Europe 600 + EU semicap suppliers
Liquidity Filter: ADV ≥ $5M, 30D average
Sector Focus: Technology, Industrials, Materials (semicap supply chain)
Market Cap: Large & mid-cap (≥ €1B)

Portfolio Constraints

Constraint	Limit	Current	Status
Sector Cap	40%	22.4%	✓
Single Name Cap	10%	6.8%	✓
Beta Band	0.30-0.50	0.38	✓
Turnover Budget	22% / day	16 bps	✓

Data Sources & Lineage

audit-friendly

Feed	Use	Freshness	Quality	Version	Update Frequency
Supply chain scans	shipment cadence feature	T-1	● OK	sc_11	Daily
NLP news & transcripts	demand sentiment index	T-0	● OK	nlp_23	Real-time
Fundamentals / revisions	confirmatory filter	T-1	● OK	fs_19	Daily
Liquidity / costs	slippage + capacity model	T-0	● Degraded	cost_08	Intraday

Data Quality & Lineage

Supply Chain Scans (sc_11): Aggregates shipment data from logistics providers, supplier announcements, and port activity. Covers 85% of semicap supply chain. Data validated against company disclosures with 92% accuracy.

NLP News & Transcripts (nlp_23): Real-time sentiment analysis from earnings calls, news articles, and social media. Uses transformer-based models fine-tuned on financial text. Sentiment index correlates 0.68 with 30D forward returns.

Fundamentals/Revisions (fs_19): Aggregates analyst revisions from 12+ data providers. Normalized for coverage differences. Revision momentum calculated as 30D weighted average of estimate changes.

Liquidity/Costs (cost_08): Currently degraded due to data provider issue. Using fallback model with 15% wider confidence intervals. Expected resolution: T+2 days.

Statistical Test Summary

promotion gate outputs

Test	Result	Value	Threshold	Notes
Out-of-sample Sharpe	PASS	1.92	≥ 1.2	rolling 12M OOS
p-value (alpha != 0)	PASS	0.006	≤ 0.05	FDR-adjusted
Walk-forward stability	PASS	8/10	≥ 7/10	fold consistency threshold met
Bootstrap robustness	PASS	94%	≥ 90%	CI excludes 0
Cost / slippage stress	MONITOR	+18%	≤ +20%	risk if spreads widen
Regime sensitivity	MONITOR	2	≤ 3	weakness in sharp risk-off regimes

AI Reasoning — Why This Passed

The signal maintained significance across walk-forward folds and remained robust under bootstrap resampling. Slippage sensitivity was acceptable for the current capital allocation. Portfolio-level risk-gates stayed within beta and concentration constraints, so the strategy was promoted to live with an explicit turnover budget and drift monitoring.

Additional Test Details

Test	Value	Interpretation
Information Ratio	0.84	Strong risk-adjusted returns
Max Drawdown (OOS)	-8.2%	Within acceptable range
Win Rate	62%	Above 55% threshold
Average Holding Period	18 days	Short-term tactical
Correlation with Market	0.38	Low correlation, good diversification

Live Performance Metrics

Since Promotion to Live

Metric	Value	Benchmark	Status
MTD P&L	+$4.86M	+$2.1M	✓ Outperform
MTD Return	+8.2%	+3.4%	✓ Outperform
Sharpe Ratio (Live)	2.14	1.92	✓ Outperform
Max Drawdown	-3.8%	-8.2%	✓ Better
Win Rate	68%	62%	✓ Outperform
Trades (7D)	21	15	● Higher

Position-Level Performance

Position	Weight	P&L	Return	Status
ASML	6.8%	+$5,812K	+12.2%	✓
SIE	3.8%	+$1,234K	+6.1%	✓
SU	3.6%	+$892K	+4.2%	✓
IFX	3.2%	+$456K	+2.8%	✓
STM	4.6%	+$1,102K	+5.1%	✓
+ 3 others	6.4%	+$1,234K	+3.8%	✓

Signal Processing Pipeline

How Signals Are Combined

Signal Aggregation Method

Weighted Composite: Signals are weighted by historical predictive power: Alt-data (40%), NLP (30%), Revisions (25%), Valuation (5%). Composite score threshold: ≥ 7.5/10.

Temporal Alignment: All signals normalized to 30-day lookback window. Lag adjustments applied: shipments (T-15), NLP (T-0), revisions (T-5), valuation (T-1).

Cross-Validation: Signal independence confirmed (correlation <0.3 between sources). Bootstrap resampling shows 94% CI excludes zero.

Signal Processing Steps

Data Ingestion: Raw feeds ingested daily (T-1) or real-time (T-0) depending on source.
Feature Engineering: Shipment cadence calculated as 30D rolling average YoY growth. NLP sentiment index uses transformer model fine-tuned on financial text.
Normalization: All signals z-scored relative to 252-day rolling window. Outliers capped at ±3σ.
Weighted Combination: Signals combined using historical IC-weighted scheme. Weights rebalanced quarterly.
Threshold Filter: Composite score must exceed 7.5/10 to generate signal. Additional filters: liquidity, sector constraints.
Position Sizing: Signal strength maps to position size: 7.5-8.0 (3-5%), 8.0-8.5 (5-7%), 8.5+ (7-10%).

Governance & Controls

Risk Management Framework

Control	Limit	Current	Status	Action
Sector Concentration	40%	22.4%	✓	Monitor
Single Name Limit	10%	6.8%	✓	Monitor
Portfolio Beta	0.30-0.50	0.38	✓	Monitor
Turnover Budget	22 bps/day	16 bps	✓	Monitor
Max Drawdown	-15%	-3.8%	✓	Monitor
Signal Decay Threshold	< 6.0	8.1 avg	✓	Monitor

Auto-Rebalancing Rules

Position Trimming: If sector concentration exceeds 40%, positions are trimmed starting with lowest conviction. If single name exceeds 10%, position is automatically reduced to 8%.

Signal Decay: If composite signal strength drops below 6.0/10, position is reduced by 50%. If it drops below 5.0/10, position is closed.

Drift Monitoring: Strategy performance monitored daily. If Sharpe ratio (30D rolling) drops below 1.0, strategy is paused for review. If drawdown exceeds -15%, strategy is automatically paused.

Review Schedule

Daily: Performance monitoring, risk limit checks
Weekly: Signal quality review, position-level analysis
Monthly: Full strategy review, parameter optimization
Quarterly: Signal weight rebalancing, universe review

Strategy 2: S-041 (Quality × Revisions)

Overview

Hypothesis

Data Sources

Statistical Tests

Performance

Signal Processing

Governance

Strategy Summary

owner: AI · scope: EU L/S equity

Hypothesis: High-quality companies (ROIC, FCF conversion, balance sheet) with positive earnings revisions generate alpha with lower drawdown risk.
Universe: STOXX Europe 600 large & mid-cap (liquidity filters).
Signals: Quality z-score (ROIC, FCF, balance sheet) + revisions momentum + valuation sanity check.
Performance: OOS Sharpe 1.58, MTD P&L +$2.94M, 14 trades in last 7 days.
Status: Live, 6 positions, 18.2% of NAV exposure.

Linked portfolio positions: S-041 bucket

Hypothesis & Signal Construction

Core Investment Thesis

Primary Hypothesis

High-quality companies (measured by ROIC, FCF conversion, balance sheet strength) with positive earnings revisions generate alpha with lower drawdown risk. Quality provides downside protection while revisions provide upside momentum. This combination historically outperforms in both bull and bear markets.

Signal Construction

Signal Component	Weight	Source	Lookback
Quality z-score (ROIC, FCF, BS)	40%	Fundamentals	252D
Revisions Momentum	35%	Earnings revisions	30D
NLP: Sentiment	20%	Earnings calls	30D
Valuation: Relative	5%	Valuation screens	T-1

Universe & Filters

Universe: STOXX Europe 600 large & mid-cap
Liquidity Filter: ADV ≥ $5M, 30D average
Quality Filter: Quality z-score ≥ +1.5 (top decile)
Revisions Filter: Revisions momentum ≥ +2.0% (30D)

Data Sources & Lineage

audit-friendly

Feed	Use	Freshness	Quality	Version	Update Frequency
Fundamentals / revisions	quality score + revisions	T-1	● OK	fs_19	Daily
NLP earnings calls	sentiment confirmation	T-0	● OK	nlp_23	Real-time
Valuation screens	relative valuation filter	T-1	● OK	val_12	Daily

Data Quality & Lineage

Fundamentals/Revisions (fs_19): Aggregates financial data from 12+ providers. Quality z-score calculated from ROIC (5Y avg), FCF conversion (3Y), and balance sheet strength (net cash/debt ratio). Revision momentum uses 30D weighted average of estimate changes.

NLP Earnings Calls (nlp_23): Real-time sentiment analysis from earnings transcripts. Uses transformer models fine-tuned on financial text. Sentiment index correlates 0.72 with 30D forward returns for quality stocks.

Valuation Screens (val_12): Relative valuation metrics vs 5Y median and peer group. Used as sanity check to avoid overpaying for quality.

Statistical Test Summary

promotion gate outputs

Test	Result	Value	Threshold	Notes
Out-of-sample Sharpe	PASS	1.58	≥ 1.2	rolling 12M OOS
p-value (alpha != 0)	PASS	0.012	≤ 0.05	FDR-adjusted
Walk-forward stability	PASS	9/10	≥ 7/10	fold consistency threshold met
Bootstrap robustness	PASS	96%	≥ 90%	CI excludes 0
Regime sensitivity	PASS	0	≤ 3	stable across regimes

AI Reasoning — Why This Passed

Quality + revisions combination showed consistent alpha across regimes with lower drawdown. Walk-forward stability was strong (9/10 folds). Bootstrap robustness confirmed (96% CI excludes 0). Portfolio risk-gates remained within beta and concentration constraints. Strategy promoted to live as core position.

Live Performance Metrics

Since Promotion to Live

Metric	Value	Benchmark	Status
MTD P&L	+$2.94M	+$1.2M	✓ Outperform
MTD Return	+5.8%	+2.4%	✓ Outperform
Sharpe Ratio (Live)	1.72	1.58	✓ Outperform
Max Drawdown	-2.1%	-5.2%	✓ Better
Win Rate	71%	65%	✓ Outperform

Signal Processing Pipeline

How Signals Are Combined

Signal Aggregation Method

Weighted Composite: Signals are weighted by historical predictive power: Quality (40%), Revisions (35%), NLP (20%), Valuation (5%). Composite score threshold: ≥ 7.5/10.

Temporal Alignment: All signals normalized to 30-day lookback window. Lag adjustments applied: quality (T-1), revisions (T-5), NLP (T-0), valuation (T-1).

Cross-Validation: Signal independence confirmed (correlation <0.2 between sources). Bootstrap resampling shows 96% CI excludes zero.

Governance & Controls

Risk Management Framework

Control	Limit	Current	Status	Action
Sector Concentration	40%	18.2%	✓	Monitor
Single Name Limit	10%	5.4%	✓	Monitor
Quality Threshold	≥ +1.5	+1.8 avg	✓	Monitor
Revisions Threshold	≥ +2.0%	+3.2% avg	✓	Monitor

Strategy 3: S-208 (Services Displacement)

Overview

Hypothesis

Data Sources

Statistical Tests

Performance

Signal Processing

Governance

Strategy Summary

owner: AI · scope: EU L/S equity

Hypothesis: AI automation compresses labor-heavy services margins, creating short opportunities in BPO/IT services with high labor intensity and low pricing power.
Universe: STOXX Europe 600 services sector (BPO, IT services, call centers).
Signals: Labor intensity + AI adoption mentions (NLP) + margin compression + options flow.
Performance: OOS Sharpe 1.34, MTD P&L +$0.62M, 9 trades in last 7 days.
Status: Live, 5 short positions, -12.8% of NAV exposure.

Linked portfolio positions: S-208 bucket

Hypothesis & Signal Construction

Core Investment Thesis

Primary Hypothesis

AI automation compresses labor-heavy services margins, creating short opportunities in BPO/IT services with high labor intensity and low pricing power. Companies with 80%+ labor costs and declining pricing power are structurally vulnerable to AI agent displacement. This trend is accelerating, not stabilizing.

Signal Construction

Signal Component	Weight	Source	Lookback
NLP: AI Adoption Mentions	35%	Transcripts & news	30D
Alt-Data: Labor Metrics	30%	Job postings, churn	30D
Fundamentals: Margin Compression	25%	Financials	30D
Options: Put/Call Ratio	10%	Options flow	T-0

Universe & Filters

Universe: STOXX Europe 600 services sector (BPO, IT services, call centers)
Labor Intensity Filter: Labor costs ≥ 75% of total costs
Pricing Power Filter: Contract renewals showing -3% or worse pricing
Liquidity Filter: ADV ≥ $3M, 30D average

Data Sources & Lineage

audit-friendly

Feed	Use	Freshness	Quality	Version	Update Frequency
NLP transcripts/news	AI adoption mentions	T-0	● OK	nlp_23	Real-time
Alt-data: labor metrics	job postings, churn data	T-1	● OK	lab_07	Daily
Fundamentals	margin compression, pricing	T-1	● OK	fs_19	Daily
Options flow	put/call ratio, vol skew	T-0	● OK	opt_15	Intraday

Data Quality & Lineage

NLP Transcripts/News (nlp_23): Real-time sentiment analysis from earnings calls, news articles. Tracks "AI automation" mentions and customer switching patterns. Sentiment index correlates 0.68 with 30D forward returns for services stocks.

Alt-Data: Labor Metrics (lab_07): Aggregates job posting data from LinkedIn, Indeed, and company career pages. Tracks churn data from employee review sites. Job posting trends correlate -0.52 with 60D forward returns.

Fundamentals (fs_19): Margin compression tracked via quarterly financials. Pricing power measured via contract renewal data and customer switching patterns.

Options Flow (opt_15): Real-time put/call ratio and implied volatility skew. Elevated put/call ratios (>1.5) indicate negative sentiment. Vol skew shows puts trading rich vs calls.

Statistical Test Summary

promotion gate outputs

Test	Result	Value	Threshold	Notes
Out-of-sample Sharpe	PASS	1.34	≥ 1.2	rolling 12M OOS
p-value (alpha != 0)	PASS	0.021	≤ 0.05	FDR-adjusted
Walk-forward stability	PASS	7/10	≥ 7/10	fold consistency threshold met
Bootstrap robustness	MONITOR	91%	≥ 90%	CI excludes 0 but lower than core
Borrow availability	PASS	OK	≤ 80 bps	costs within budget
Squeeze risk	MONITOR	—	≤ 15%	short interest rising

AI Reasoning — Why This Passed

Signal showed strong negative momentum with multiple independent sources converging. OOS Sharpe of 1.34 met threshold. Walk-forward stability was acceptable (7/10 folds). Bootstrap showed robustness (91% CI). Borrow availability and squeeze risk were monitored. Strategy promoted to live as tactical short with enhanced monitoring.

Live Performance Metrics

Since Promotion to Live

Metric	Value	Benchmark	Status
MTD P&L	+$0.62M	+$0.3M	✓ Outperform
MTD Return	+4.8%	+2.1%	✓ Outperform
Sharpe Ratio (Live)	1.48	1.34	✓ Outperform
Max Drawdown	-4.2%	-6.8%	✓ Better
Win Rate	64%	58%	✓ Outperform

Signal Processing Pipeline

How Signals Are Combined

Signal Aggregation Method

Weighted Composite: Signals are weighted by historical predictive power: NLP (35%), Labor metrics (30%), Fundamentals (25%), Options (10%). Composite score threshold: ≥ 7.0/10.

Temporal Alignment: All signals normalized to 30-day lookback window. Lag adjustments applied: NLP (T-0), labor (T-5), fundamentals (T-10), options (T-0).

Cross-Validation: Signal independence confirmed (correlation <0.25 between sources). Bootstrap resampling shows 91% CI excludes zero.

Governance & Controls

Risk Management Framework

Control	Limit	Current	Status	Action
Borrow Cost	≤ 80 bps	42 bps avg	✓	Monitor
Short Interest	≤ 15%	8.2% avg	✓	Monitor
Squeeze Risk	MONITOR	Flagged	●	Daily review
Max Drawdown	-20%	-4.2%	✓	Monitor
Signal Decay Threshold	< 6.0	7.1 avg	✓	Monitor

Auto-Rebalancing Rules

Borrow Cost Monitoring: If borrow cost exceeds 80bps, position is reduced by 50%. If it exceeds 120bps, position is closed.

Squeeze Risk: If short interest exceeds 15%, position is reduced by 50%. If it exceeds 20%, position is closed. Daily monitoring enabled.

Signal Decay: If composite signal strength drops below 6.0/10, position is reduced by 50%. If it drops below 5.0/10, position is closed.

IV. GOVERNANCE & HUMAN CONTROLS promotion gates + kill-switch + approvals

Autonomy Settings

researcher / PM controls

Allow auto-promotion from Shadow → Live (if all gates pass)

Require human sign-off for new LIVE strategies

Auto-demote Live → Shadow on drift breach

Block strategies with data-feed degradation

Allow new strategy sleeves outside L/S Equity mandate

Max risk budget per strategy (bps/day)

22

Minimum OOS Sharpe to go Live

1.2

Max turnover budget (portfolio % / day)

22%

Emergency Controls

hard stops

Control	Description	Status
Global Kill Switch	Disable all strategy execution immediately	Armed
Strategy Quarantine	Freeze a strategy; keep monitoring in shadow	Available
Data Feed Gate	Auto-block if critical feed degrades	Enabled
Risk Gate Override	Human override of promotion/position gates	Enabled

Governance Note

Strategy autonomy is bounded by: mandate constraints, risk budgets, statistical gates, and a human approval framework. Every promotion/demotion is logged with: dataset versions, feature hashes, test outputs, and decision rationale.

V. STRATEGY EVENT LOG transparent audit trail