Combining Core News and Financial Times for Equity Selection

News-based sentiment strategies have become established tools for systematic equity selection, and most sophisticated approaches already rely on thousands of diverse news sources. Our own data feed, for example, processes over 45,000 news sources. Following our strategic partnership announcement with the Financial Times, our latest research explores a more nuanced question: Does adding a single, highly authoritative source move the needle when one already has an extensive news feed?

We tested an ensemble approach that combines RavenPack Core News (with its thousands of sources) with Financial Times content for selecting U.S. equities. The results demonstrate that, yes, thoughtfully integrating a complementary, authoritative source provides a strong amplification signal, producing signals that are stronger than those from the broad feed alone.

Strategy design

The strategy targets the top 1,000 U.S. companies by market capitalization, implementing a long-short, dollar-neutral approach with daily rebalancing.

Signal construction

For each trading day, we calculate company-level sentiment scores from two sources:

RavenPack Core News: We focus on Earnings Growth events, which include Earnings, Revenues, Dividends, and Assets announcements. This captures sentiment specifically related to financial performance.

Financial Times: Financial Times contributes institutional-grade insight and high-impact company coverage shaped by rigorous editorial standards. As an authoritative news provider, FT serves as a strong amplification and validation signal, with content that is less mined and more selectively curated.

Each signal is cross-sectionally normalized across the universe of stocks daily, ensuring comparability. The composite signal sums the normalized signals from both sources with equal weighting, creating a consensus score.

Portfolio implementation

We construct long-short, dollar-neutral portfolios rebalanced daily. Capital allocation is proportional to both the sign and magnitude of the composite signals.

A critical implementation detail: we use a 30-minute cutoff before U.S. market close, ensuring only information available up to 24 hours prior influences that day's allocation. This prevents look-ahead bias, but still includes the most recent available information.

To manage turnover, we apply exponential smoothing to the signals. This technique allows performance evaluation across different effective holding periods: slower decay extends holding periods but reduces alpha capture, while faster decay does the opposite.

Performance results

Since 2010, the ensemble strategy has consistently outperformed traditional news-based strategies.

At a 2-day effective holding period, adding Financial Times content to Core News improved performance by approximately 120 basis points. The Information Ratio increased from 0.59 (Core News alone) to 0.73 (ensemble).

The combined value is synergistic, not just additive. While RavenPack Core News excels at comprehensive, bottom-up event detection using proprietary NLP across two decades of data, the Financial Times provides curated, institutional-grade insights. As an authoritative source, the FT acts as a powerful amplification signal, validating key events and enhancing the relevance of our core news feed.

Why Ensemble methods work

The performance improvement comes from several factors:

Complementary Coverage: Our extensive Core News feed captures a wide spectrum of specific corporate events from over 45,000 sources. The Financial Times, however, specializes in high-impact company coverage. Because investors place greater emphasis on FT news, it acts as a high-conviction amplification signal. Furthermore, FT content has been less mined as an alpha dataset, enhancing the predictive power of our combined signal.

Signal Breadth: Combining sources increases the number of independent observations, improving signal stability and reducing idiosyncratic noise.

Consensus Filtering: Our ensemble approach utilizes a consensus voting mechanism, not merely to identify agreement, but primarily to mitigate the inherent model risk associated with signals from each source. The resulting signal is not limited by the weakest input; rather, the mechanism strategically combines the high-quality features of both sources, naturally weighting companies where both provide clear, reinforcing signals, leading to a more robust final output.

Reduced Source Risk: Relying on a single news source creates dependency risk. Ensemble approaches are more robust to changes in editorial focus or coverage patterns from any single provider.

Implementation considerations

Several technical details matter for successful implementation:

Normalization: Cross-sectional normalization each day is essential. Without it, differences in scoring scales between sources would dominate the composite signal.

Equal Weighting: We found that equal weighting of normalized signals from both sources produced the best results.

Turnover Management: Exponential smoothing provides an effective way to balance alpha capture against transaction costs. The optimal effective holding period depends on trading costs and market conditions.

Signal Cutoff: To ensure signals are genuinely tradable and free from look-ahead bias, a 30-minute cutoff is applied. This means portfolio formation uses only information that was demonstrably available 30 minutes prior to the trading decision, allowing us to accurately harvest returns from T0 (the date of the cutoff) to T1.

Looking forward

This research demonstrates that ensemble approaches can meaningfully improve sentiment-based equity selection strategies. The key is combining sources that provide genuinely different information rather than redundant coverage of the same events.

Future research directions include exploring optimal weighting schemes that adapt to market regimes, extending the approach to other regions and market caps, and incorporating additional news sources where they provide complementary information.

The full whitepaper includes detailed performance metrics across different holding periods, turnover analysis, and complete methodology specifications.