All companies are required to file registration statements, periodic reports, and other forms to the SEC and these filings are popular sources of data for researchers at the GSB. This topic guide covers some of the resources available to facilitate research using these filings and a few sample workflows.
The SEC maintains a comprehensive collection of filings on their EDGAR system and you can search for filings by company name, ticker, and CIK here.
While viewing individual filings on EDGAR is useful, for those who need bulk access to filings, there are a couple of primary resources.
The DARC team maintains a mirror of the raw text filings on the Yens by checking for new filings on the EDGAR system on a weekly basis and then downloading those filings. An example path for an individual filing located on the Yens is:
For clarity, the middle portion of this path can be used to find the same filing on EDGAR:
/zfs/data/NODR/EDGAR_HTTPS/ + edgar/data/1050122/0001047469-03-017249 + .txt
https://www.sec.gov/Archives/ + edgar/data/1050122/0001047469-03-017249 + -index.htm
The above provides a clear way to identify specific documents, but not a way to “query” those documents. To address this, the DARC team also maintains a SQL dataset on Redivis that can be queried for specific filings.
This dataset contains a table,
edgar-filings, with the following fields:
cik company_name form_type date_filed filename year quarter filepath
For example, to produce a list of a few 10-K filings from 1994, you can run the following query via Redivis:
SELECT * FROM `edgar_filings` WHERE form_type='10-K' and year=1994 LIMIT 10
You may then use the
filepath field to locate these filings on the Yens or
filename to construct the appropriate URLs to those filings on EDGAR.
For your convenience, we have also crafted a Python-based example project to illustrate how one might parse EDGAR filings on the Yens. In this example (an actual use case we encountered), we attempt to seek out sentences that contain a specific word, that also have adjacent sentences containin key target phrases. Please refer to the
edgar_filings_example_notebook.ipynb notebook in the following directory for more information on how to explore this project:
If you have a WRDS account (you can request one here), you can also access EDGAR filings from their server. Have a look at this link for additional information about the WRDS collection of filings.
To access these filings, you can connect to WRDS Cloud and either navigate to
/wrds/sec/warchives/ for raw filings or
/wrds/sec/wrds_clean_filings/ for cleaned filings.
In an effort to standardize data submitted by companies to the SEC, the SEC commission adopted amendments in 2018 requiring the use of the Inline XBRL (eXtensible Business Reporting Language) format in submissions of financial statement information. As a result, many recent forms on EDGAR (see here for a list of form types with XBRL) can be viewed with XBRL elements highlighted (example Form 10-K from AT&T in 2020).
Besides viewing this XBRL data in an interactive format on EDGAR, users have a few ways of accessing bulk XBRL data.
Both the SEC and XBRL offer Application Programming Interfaces (APIs) that allow researchers to query specific XBRL data in a programmatic fashion. These APIs are great options if you desire to track specific XBRL tags on a regular basis or would like to query a moderate amount of data at once (i.e. not all tags from all companies).
If your project requires a large amount of data, the SEC also offers regular dumps of XBRL data on a quarterly (2009 through 2020 Q3) and monthly (October 2020 through current) basis. A detailed description of the tables and columns in the data dump can be found in this file.
Because many projects require data over the course of several years and it may be inconvenient to work with the zipped data dumps, the DARC team also maintains updated copies of the tables in these dumps in this Redivis dataset. This dataset is updated on a monthly basis, in concordance with the release of the dumps by the SEC.
We have also created an example project on Redivis that illustrates a use case for this XBRL data and how to pull data from the tables. You can check the description field in each transform to understand the queries made.
WRDS maintains a comprehensive collection of tools called the WRDS SEC Analytics Suite. It is a “one-stop” research platform that provides standardized service tools to enable users to overcome the challenges in systematically parsing regulatory reports on the SEC website, and quantifying and analyzing information which is qualitative in nature. Here’s a link to the manual.
Here are a few tools and their descriptions that the DARC team has found useful in the WRDS SEC Analytics Suite.
SEC Filings Index
Comprehensive Daily and Quarterly SEC Filings Index can be found at the WRDS SEC Filings Index Search site. From here you can get an idea of what kind of forms were available from which companies dating back to the 1970s. You can search for a single or a list of CIK, TICKER, or GVKEY company code. Some of the output variables are:
SEC Central Index Key (CIK) Filing Date (FDATE) First SECDate with Index Record Information (FINDEXDATE) Last SECDate with Index Record Information (LINDEXDATE) SEC Form (Form) Company Name (CoName) Reference Name of Complete Report Filing (FName) Reference Name of Report HTML Index (IName) Index Source: 1-Daily, 2-Full, 3-Both (SOURCE)
Readability and Sentiment Analysis
The Readability and Sentiment Analysis tool is a relatively new addition to the suite. Please note that the data for this web query is only reliable up to 2019; data after 2019 maybe be incomplete. This work is expected to complete by March of 2023.
The WRDS SEC Readability and Sentiment data extends the WRDS SEC Analytics Suite by providing academic researchers a clean set of text files for every SEC filing since 1994, along with basic sentiment and readability scores. Researchers can use the pre-computed scores to further academic research, and can also compute their own features based on the raw text. You can find more information in the manual about this tool here.
Some of the many counts and scores it outputs include:
Complete Report File Size (FSIZE) Paragraph Count (PARAGRAPHCOUNT) Raw Character Count (CHARCOUNT_RAW) Average Number of Words per Sentence (AVERAGEWORDSPERSENTENCE) Word Count (WORDCOUNT) Sentence Count (SENTENCECOUNT) Flesch Reading Ease Index (FLESCH_READING_EASE) Smog Readability Index (SMOG_INDEX) Loughran-McDonald Negative Word Proportion (FINTERMS_NEGATIVE) Loughran-McDonald Uncertainty Word Count (FINTERMS_UNCERTAINTY_COUNT) Harvard General Inquirer Negative Word Proporation (HARVARDIV_NEGATIVE) etc...
SEC Filings Search is where you can search the contents of over 15 million SEC filings. The dataset contains 720 form types, including 10-Ks, 10-Qs, 8-Ks, Proxy and Registration Statements, 40-F Annual Reports, Uploads, and Correspondence. You can do a syntax search with terms and logic like the following:
Example: "fried eggs" +(eggplant | potato) -frittata
Syntax supports the following:
+ signifies AND operation
| signifies OR operation
- excludes a specific term from your search
"" signifies an exact phrase search
* at the end of a prefix acts as a wildcard – e.g., auto*, thermo *
use () to establish order of precedence for complex queries
~N after a word signifies edit distance, or fuzziness. The edit distance, N, is the number of character changes required to transform one word into another – e.g., Anderson~1 will return Anderson, Andersen and Andersan.
~N after a phrase indicates how many words can separate the terms in a successful match – e.g., “CEO died”~4 will return “… CEO of this great company, died peacefully.”
SEC Linking Tables
WRDS provides tables you can use to join datasets with different IDs across time. It is a good idea to ask a GSB librarian for help on how exactly to use these tables so that you can get the right company matches especially if you are looking at a large span of time. Linking tables include:
- Historical CIK-CUSIP Link Table
- Historical Company Names of SEC Filers
- Historical GVKEY-CIK Link Table
Bag of Words
WRDS Bag of Word (BoW) interface provides WRDS SEC Analytics Suite subscribers an alternative way to perform content analysis against company SEC filings. A frequency count of all words in SEC filings and amendments from 1993 through 2017 was prepared. Users can query word counts across all filings, or against specific company filings. Total word occurrence per filing, cosine similarity, Jaccard similarity, minimum edit distance between current and previous filings, word stemming and lemmatization, English word, positive or negative word, stop word, geographic, company name, patent related (requires KtMINE subscription) word indicators etc. are also available. You can read more about how words are contructed and what summary measures were applied in the manual.
Here are some of the modules available:
Filing Summary Similarity Measures Word Distribution Word Distribution (all filings) Word Summary
Connect with us