Context Analytics Blog

Exploring CompuText's Predictive Insights in 8-K Filings

Written by CA Research Team | Nov 6, 2024 6:49:24 PM

CompuText leverages Context Analytics' Machine Readable Filings (MRF) and Global Machine Readable Filings (GMRF) datasets, containing over 1.5 million domestic and 1.25 million international documents, along with S&P Earnings Call Transcripts, to analyze company filings. CompuText identifies and tags sentences relevant to specific financial topics, extracting insights across Income Statements, Balance Sheets, and Cash Flow Statements. Covering over 38 financial items such as Sales, Inventory, and Operating Cash Flow, CompuText is continuously expanding. Each sentence is tagged with specific financial items, contextual relevance, sentiment indicators, and unique identifiers for the document and company.

 

CompuText's analysis includes various filing types, including annual reports, quarterly reports, 10-Ks, and 8-Ks. Here, we explore how CompuText can reveal insights from 8-K filings. Companies are required to file an 8-K for significant material events, so identifying key language in these filings with CompuText can reveal indicators of subsequent market movement.

 

For this analysis, we focus on the Russell 3000 universe, using CompuText data on 8-K filings from 2014 onward. We calculate a sentiment score for each company's 8-K filing day using the term polarity field, defined as (Total Positive Count - Total Negative Count) / Total Hit Count, with a score range of -1 to 1, indicating the sentiment surrounding each report.

 

To ensure meaningful data, we filter for documents with at least three relevant phrases. Although 8-K filings are shorter than other government filings, this minimum threshold ensures score reliability. We then categorize each company's filing into tertiles (top 33%, middle 33%, and bottom 33%) based on sentiment scores. We chose tertiles over quintiles given the phrase count threshold, allowing for a balanced distribution across categories.

 

Securities enter tertiles based on their sentiment score and are held for the next 10 market days, re-bucketing daily and using updated sentiment scores if a new 8-K filing is released within that period. Returns are calculated on an equally weighted, close-to-close daily basis until the holding period ends.

 

Over time, this analysis reveals a clear return spread across tertiles: the top tertile shows positive returns, the middle tertile remains neutral, and the bottom tertile shows negative returns. This trend highlights a connection between positive CompuText language in 8-K filings and favorable market performance. Conversely, predominantly negative CompuText language can signal a likelihood of price declines.

 

For more information on CompuText, visit www.contextanalytics-ai.com .