Context Analytics Blog

Index created with CA sentiment data vs. S&P 500 RavenPack AI Sentiment Index

Written by CA Research Team | Jan 17, 2025 6:15:50 PM

Introduction

Last year Context Analytics (CA) released a news sentiment data feed. This paper will explore an index created with CA sentiment data compared to the S&P 500 RavenPack AI Sentiment Index (Bloomberg ticker: CSRPAISE). Both use news-based sentiment data to construct a sentiment-driven, sector-based index.

Following the foundational structure of CSRPAISE, this study involves creating quarterly sentiment scores for S&P 500 constituents, selecting four sectors from GIC 11 based on aggregated sentiment scores, and applying market cap weighting within those sectors.

Notable methodological differences influence the outcomes of our index.

Underlying News Data Differences:

  • CSRPAISE uses RavenPack Average ESS Sentiment Score, which is market cap-weighted within sectors before aggregation.
  • Context Analytics employs a proprietary score from its quantitative news feed, using equal weighting for sentiment scores within sector summations. This approach reduces the weight of large securities like Nvidia, Apple, and Tesla, placing greater emphasis on actual sentiment versus mkt. cap.

Sector Weighting Differences:

  • CSRPAISE applies a sector-wise market cap weighting and further incorporates a risk-adjusted weighting, where sectors are weighted inversely proportional to their estimated volatility.
  • Context Analytics equally weights the four selected sectors for simplicity and relies primarily on sentiment for selections, without incorporating any risk adjustments.

The benchmark for this study is CSRPAISE. The full methodology can be found in references at the end of the paper. Characteristics of the RavenPack methodology include:

  1. Stock Weight Calculation: Stocks are weighted by their closing index market cap relative to the total sector index market cap in the S&P 500.
  2. Sector Weight Calculation: Sector weights are determined by the sum of stock index market caps within each sector relative to the total market cap of all S&P 500 constituents.
  3. Sector Sentiment Scores: Sentiment scores for each sector are calculated by multiplying a stock's average Event Sentiment Score (ESS, by RavenPack) by its weight and summing the results for all stocks within the sector.
  4. Sector Selection: The top four sectors with the highest sentiment scores are selected for the index.
  5. Tie-Breaking: Ties are resolved by prioritizing sectors with the largest weight in the S&P 500.
  6. Sector Weights: Assigned by Market Cap Summation within sectors & estimated volatility.
  7. Rebalancing: Sector weights and portfolios are rebalanced quarterly.

The objectives of this research:

To isolate and evaluate the performance of Context Analytics' news sentiment as a long-term predictor, applying a methodology comparable to the benchmark index.

 

Data and Methodology

Data Sources

  1. News Sentiment Data: Context Analytics’ proprietary financial news sentiment data, filtered to include only non-reposted articles with 100% relevancy, each uniquely discussing a specific S&P 500 constituent.
  2. Market Data: Market capitalizations for S&P 500 constituents and GIC sector indices.
  3. S&P 500 RavenPack AI Sentiment Index Returns: Daily total returns provided by S&P Global.

Sentiment Score Calculation: To replicate and refine the quarterly sentiment score methodology, the following steps were applied:

  1. Daily Aggregation: For each S&P 500 ticker, the following metrics were calculated daily:
    • Positive Hits: Count of positive sentiment words in articles.
    • Total Hits: Total count of sentiment words in articles.
  2. Monthly Aggregation: Daily positive hits and total hits were aggregated at the end of each month.
  3. Quarterly Weighted Sum: To emphasize recent sentiment trends, a 3x exponential weighting scheme was applied over the previous quarter, using multipliers of 1, 3, and 9. This ensures greater influence for sentiment closer to the rebalancing date.
  4. Positivity Ratio: For each stock, the positivity ratio was calculated as:
     

Sector Sentiment Calculation

Sector sentiment scores were derived by summing the positivity ratios across all stocks within each sector and dividing by the number of stocks within that sector. The top four sectors with the highest sentiment scores at each rebalancing date were selected.

 

Portfolio Construction

For the selected sectors:

  1. Stocks within each sector were weighted by their market caps.
  2. Returns from the top four sectors were equally weighted to form the final index return.
  3. Rebalancing occurred quarterly.

Benchmark Comparison

The CSRPAISE methodology incorporates additional steps for volatility adjustment during rebalancing. While the CSRPAISE methodology accounts for volatility, the Context Analytics approach focuses solely on sentiment data and market capitalization.

 

Results

Performance Comparison- Returns & metrics for both are calculated by Context Analytics. The link to raw data for CSRPAISE is found under Data Sources.

 

The Context Analytics Sentiment Index was back tested from Q4 2021 to the present, yielding the following observations:

  1. Consistent Improvement: The Context Analytics strategy demonstrated consistent improvement in returns compared to the benchmark.
  2. Enhanced Sector Selection: By emphasizing recent sentiment trends through exponential weighting, the strategy effectively identified sectors with strong near-term sentiment momentum.
  3. Simplified Construction: The positivity ratio and market cap weighting streamlined the process while maintaining robust performance.
  4. Risk Reduction: Sharpe and Sortino ratios for the Context Analytics strategy showed significant improvement.

Volatility and Sector Weighting: Although the S&P 500 RavenPack AI Sentiment Index methodology adjusts sector weights based on volatility, the Context Analytics index achieved improved results without incorporating this additional complexity.

 

 

Conclusion

This research successfully created an index comparable to the S&P 500 RavenPack AI Sentiment Index using Context Analytics' news sentiment data. By introducing a quarterly exponential weighting scheme and focusing on the positivity ratio, the Context Analytics Sentiment Index demonstrated consistent performance improvements over the benchmark.

The findings underscore the effectiveness of Context Analytics’ news sentiment data in constructing sector-based indices and portfolio strategies. Future research could explore incorporating volatility adjustments or experimenting with alternative weighting methodologies to further enhance returns.

 


References

  1. S&P Global and RavenPack, "S&P 500 RavenPack AI Sentiment Index Methodology," https://www.spglobal.com/spdji/en/documents/methodologies/methodology-sp-500-rvnpck-ai-sentiment-indices.pdf.
  2. S&P Global and RavenPack, “S&P 500 RavenPack AI Index,” https://www.spglobal.com/spdji/en/documents/brochure/brochure-sp-500-ravenpack-ai-index.pdf.