Enhancing Japanese Equity Performance with Hierarchical Portfolio Construction & Earnings Sentiment

The academic evidence consistently shows that signals, when properly aggregated with robust risk controls, can generate meaningful alpha.

Hierarchical portfolio construction through multi-nested tilts offers a rigorous framework for translating alternative data signals into persistent alpha. Our latest research demonstrates how combining Earnings News Sentiment from RavenPack Edge Analytics with sequential sector and stock-level tilting can significantly enhance risk-adjusted performance in the Japanese Top-500 equity universe.

Since 2018, this multi-level approach has improved Information Ratios from 0.53 to 0.83 while maintaining a moderate 7% tracking error, delivering nearly 700 basis points of annual excess return versus the benchmark, outperforming standalone sector or stock-level tilting by approximately 300 basis points.

The challenge: aggregating signals

Alternative data signals often exhibit low information coefficients , making them difficult to incorporate into active portfolio management without triggering excessive active risk or concentration. Traditional single-level optimization approaches struggle to extract value from these signals while maintaining acceptable tracking error.

The solution lies in hierarchical portfolio construction: a theoretically grounded approach that decomposes active risk across multiple portfolio layers, effectively separating systematic and idiosyncratic sources of alpha.

Our approach: multi-level benchmark tilting

Our framework implements sequential optimization at two distinct levels:

1. Sector-level tilting: First, we adjust sector weights away from the benchmark using market-cap weighted aggregations of constituent earnings sentiment scores. This captures broad sectoral momentum while controlling sector concentration through absolute (40%) and relative (4x) weight caps.

2. Stock-level tilting: Within each reweighted sector, we apply stock-specific tilts based on individual earnings news sentiment. Position-level constraints (15% absolute, 10x relative) ensure diversification while allowing meaningful exposure to high-conviction names.

Signal construction

Our earnings sentiment scores are powered by RavenPack Edge Analytics, capturing earnings announcements and guidance, revenue reports, dividend news, and both factual events and forward-looking statements. The scores are exponentially weighted over a six-month lookback with monthly decay, balancing timely alpha capture with disciplined turnover control.

Performance results

Testing across 1,681 portfolio configurations from 2018-2025 on the 500-stock Japanese universe with daily rebalancing reveals:

• Information Ratio: 0.53 → 0.83
• Annual Excess Return: ~700 basis points
• Tracking Error: 7%
• Annualized Return: 19.1% vs. 12.5% benchmark
• Outperformance vs. Single-Level: +300 bps

The hierarchical approach consistently delivers superior risk-adjusted returns across the parameter space. Even at moderate tilting levels, the strategy maintains attractive Information Ratios above 0.80 while keeping tracking error below 7%.

Hierarchical construction works - here’s why

The effectiveness of multi-level tilting stems from several key principles:

• Risk decomposition: by separating sector and stock decisions, the framework prevents unintended exposures and concentrations that often plague single-layer approaches.
• Signal aggregation: signals gain statistical power through proper aggregation. Sector-level tilts capture systematic patterns, while stock-level tilts exploit idiosyncratic information.
• Robust implementation: nested constraints at multiple levels enhance out-of-sample stability, particularly valuable when working with noisy alternative data.
• Scalable architecture: the layered approach aligns with practical index construction frameworks, making it suitable for institutional implementation and client-aligned investment solutions.

Download the full research

Hierarchical portfolio construction using earnings news sentiment shows that well-structured, multi-level optimization can unlock meaningful alpha from alternative data while maintaining institutional-grade risk controls. The 700 basis points of annual outperformance at 7% tracking error highlights a strong risk–return profile for active equity managers and materially outperforms single-level approaches.

For full methodology, parameter analysis, and comprehensive backtesting results, download the complete whitepaper.