Perceptron Intelligence – Research Portal

Sector Forecasting Model — Out-of-Sample Validation

• Conducted rigorous walk-forward validation of a monthly relative strength forecasting model across all 25 GICS Industry Groups vs the US_REBALANCE benchmark.
• Methodology: model trained exclusively on historical data, with each forecast made strictly before the outcome period — no hindsight bias in any prediction.
• Validated accuracy of 54.7% on a 4-week forward horizon; rising to 55.4% on high-conviction signals where the model expressed a clear directional view.
• Strongest forecasting edge in Consumer Durables/Apparel, Capital Goods, Technology Hardware/Equipment, and Real Estate Management/Development — all above 80% accuracy over the validation window.
• Model exhibits a coherent bearish stress detection factor: performs best when medium-term relative weakness is persistent and identifiable from price structure alone.
• Sector asymmetry confirmed: Energy, Utilities, and Semiconductors remain harder to forecast from price data alone — these sectors respond primarily to commodity cycles, interest rate expectations, and earnings revision cycles not captured in relative price series.
• 100 live forecasts outstanding for May–June 2026, providing a real-time prospective validation window.
• Probability calibration confirms the model is expressing genuine uncertainty rather than overconfident directional calls — well-suited for ranking and relative allocation, not binary timing.

Momentum Rotation — US Sectors (SPDR)

• Weekly relative momentum analysis across 11 US SPDR sector ETFs vs S&P 500.
• Identifies sectors in improving vs deteriorating relative momentum quadrants for tactical allocation input.
• April 16 update reflects post-tariff announcement positioning shifts across cyclical and defensive sectors.

Platform Index — AI Supply Chain Theme

• Updated constituent analysis and index performance for the AI Supply Chain thematic basket.
• Tracks relative performance of infrastructure, hardware, and supply chain enablers within the AI capex cycle.

Research Priorities — Next Period

• Extend forecasting model to incorporate interest rate and commodity momentum as additional inputs for Energy, Utilities, and Financials sectors.
• Assess whether the bearish stress factor can be decomposed into regime-specific sub-signals for more precise sector allocation timing.

Factor Research — PCA Walk-Forward Analysis

• Rolling 5-year PCA windows evaluated against expanding-window approach; rolling windows show better regime sensitivity.
• Three return targets examined: total return, benchmark-adjusted (x-beta), and industry-adjusted (x-beta + industry).
• PC4 identified as a high-signal component — loads on long-term reversal, momentum, and profitability simultaneously, suggesting a composite quality-momentum factor.
• Factor loadings time series proposed as a novel approach to detecting market structure regime changes over time.

Sector Forecasting — Signal Review

• Banks sector showing elevated probability of near-term outperformance based on current technical positioning.
• Sector rotation analysis identifies four-quadrant momentum structure across all GICS groups vs benchmark.
• Energy and rates sector signals flagged for independent verification — directional calls inconsistent with macro backdrop; investigation ongoing.
• Biweekly publication cadence established for sector-level signal summaries.

Alpha Research — Macro & Alternative Data

• PCA framework expansion proposed to cover all asset classes: equities, fixed income, commodities, and macro time series.
• Text-based alpha research initiated: Fed communications, earnings call transcripts, and macro commentary as quantifiable data sources.
• Objective: construct custom sentiment indices from unstructured data to complement price-based signals.
• Macro factor integration identified as critical given elevated cross-asset volatility in current environment.

Next Steps

• Verify energy and rates data inputs with Jacob; publish sector rotation research once validated.
• Bond market PCA research to be circulated ahead of next meeting.
• Biweekly research cadence confirmed; next meeting scheduled week of March 9.

Return Decomposition Approach

• Research consensus: apply PCA to ex-industry returns (residuals after removing industry effects) rather than raw active returns, to isolate cross-sectional factor structure more cleanly.
• Rationale: active returns conflate industry allocation and stock selection effects; ex-industry returns isolate the pure cross-sectional signal.

Principal Component Analysis — Rolling Methodology

• Rolling PCA windows adopted over a single full-period estimation — reduces stale factor loadings and improves responsiveness to changing market regimes.
• PC1 ticker rankings to be updated and published as a live relative value signal across the Russell universe.

Theme Detection Research

• Initiated investigation into thematic clustering using Early Warning System (EWS) signals as inputs — seeks to identify emerging sector and style themes before they appear in consensus factor models.

Coppock Signal Research — Multi-Timeframe Architecture

• Evaluated a 3×3 signal matrix across weekly, monthly, and daily Coppock timeframes — each providing distinct cyclical information at different horizons.
• Monthly Coppock identified as the primary strategic signal; weekly Coppock useful for tactical entry refinement.
• MA50/MA200 crossover signals integrated as confirmation layer: Golden Cross (bullish regime confirmation), Dark Cross (bearish regime confirmation).
• Individual (non-aggregated) Coppock scores preferred over composite scoring — preserves sector-level dispersion and avoids signal dilution.

Regime Classification Framework

• Tree-structured state model proposed: securities bucketed into regime nodes based on the directional alignment of weekly, daily, and monthly Coppock scores.
• Framework enables systematic identification of sectors in early recovery, confirmed uptrend, deteriorating, or confirmed downtrend regimes.
• Overlay of aggregate vs sector-level Coppock divergence identified as a leading indicator of rotational opportunities.

Research Output

• Latest Coppock signal by GICS industry group to be published on a biweekly basis as a structured research deliverable.

Hit Ratio Analysis — 24 GICS Sector Time Series

• Systematic hit ratio analysis completed across all 24 GICS sector series to assess directional forecasting accuracy.
• Data integrity review completed: sector and index series mappings verified and corrected.

Signal Construction & Portfolio Logic

• Buy/Hold/Sell framework defined: long exposure to high-conviction positive signals, short or underweight on low-conviction negative signals.
• Each of the 24 GICS sectors treated as an independent tradable unit — consistent with industry group-level portfolio construction.
• Both aggregate (composite) and individual sector-level signal views maintained for different use cases.

Forecasting Horizon Research

• Optimal forecasting setup validated: 5-year rolling feature estimation window; 3-week forward prediction target.
• Out-of-sample validation confirmed over 4+ week forward windows — consistent with monthly rebalancing cadence.