SECFSDSTools

Helper tools to analyze the " Financial Statement Data Sets" from the U.S. securities and exchange commission (sec.gov)

View the Project on GitHub HansjoergW/sec-fincancial-statement-data-set

sec-fincancial-statement-data-set

Helper tools to analyze the Financial Statement Data Sets from the U.S. securities and exchange commission (sec.gov). The SEC releases quarterly zip files, each containing four CSV files with numerical data from all financial reports filed within that quarter. However, accessing data from the past 12 years can be time-consuming due to the large amount of data - over 120 million data points in over 2GB of zip files by 2023.

This library simplifies the process of working with this data and provides a convenient way to extract information from the primary financial statements - the balance sheet (BS), income statement (IS), and statement of cash flows (CF).

Check out my article at Medium Understanding the the SEC Financial Statement Data Sets to get an introduction to the Financial Statement Data Sets.

The main features include:

Principles

The goal is to be able to do bulk processing of the data without the need to do countless API calls to sec.gov. Therefore, the quarterly zip files are downloaded and indexed using a SQLite database table. The index table contains information on all filed reports since about 2010, over 500,000 in total. The first download will take a couple of minutes but after that, all the data is on your local harddisk.

Using the index in the sqlite db allows for direct extraction of data for a specific report from the appropriate zip file, reducing the need to open and search through each zip file. Moreover, the downloaded zip files are converted to the parquet format which provides faster read access to the data compared to reading the csv files inside the zip files.

The library is designed to have a low memory footprint, only parsing and reading the data for a specific report into pandas dataframe tables.

Installation and basic usage

The library has been tested for python version 3.7, 3.8, 3.9, and 3.10. The project is published on pypi.org. Simply use the following command to install the latest version:

pip install secfsdstools

If you want to contribute, just clone the project and use a python 3.7 environment. The dependencies are defined in the requirements.txt file or use the pyproject.toml to install them.

It is possible to write standalone python script but I recommend to first start with interactive jupyter notebooks 01_quickstart.ipynb and 03_explore_with_interactive_notebook.ipynb that are located in notebooks directory.

Upon using the library for the first time, it downloads the data files and creates the index by calling the update() method. You can manually trigger the update using the following code:

from secfsdstools.update import update

if __name__ == '__main__':
    update()

The following tasks will be executed:

  1. All currently available zip-files are downloaded form sec.gov (these are over 50 files that will need over 2 GB of space on your local drive)
  2. All the zipfiles are transformed and stored as parquet files. Per default, the zipfile is deleted afterwards. If you want to keep the zip files, set the parameter ‘KeepZipFiles’ in the config file to True.
  3. An index inside a sqlite db file is created

Moreover, at most once a day, it is checked if there is a new zip file available on sec.gov. If there is, a download will be started automatically. If you don’t want ‘auto-update’, set the ‘AutoUpdate’ in your config file to False.

Configuration (optional)

If you don’t provide a config file, a config file with name secfsdstools.cfg will be created the first time you use the api and placed inside your home directory. The file only requires the following entries:

[DEFAULT]
downloaddirectory = c:/users/me/secfsdstools/data/dld
parquetdirectory = c:/users/me/secfsdstools/data/parquet
dbdirectory = c:/users/me/secfsdstools/data/db
useragentemail = your.email@goeshere.com

The downloaddirectory is the place where quarterly zip files from the sec.gov are downloaded to. The parquetdirectory is the folder where the data is stored in parquet format. The dbdirectory is the directory in which the sqllite db is created. The useragentemail is used in the requests made to the sec.gov website. Since we only make limited calls to the sec.gov, you can leave the example “your.email@goeshere.com”.

Viewing metadata

The recommend way to view and use the metadata is using secfsdstools library functions as described in notebooks/01_quickstart.ipynb

The “index of reports” that was created in the previous step can be viewed using a database viewer that supports the SQLite format, such as DB Browser for SQLite.

(The location of the SQLite database file is specified in the dbdirectory field of the config file, which is set to <home>/secfsdstools/data/db in the default configuration. The database file is named secfsdstools.db.)

There are only two relevant tables in the database: index_parquet_reports and index_parquet_processing_state.

The index_parquet_reports table provides an overview of all available reports in the downloaded data and includes the following relevant columns:

For instance, if you want to have an overview of all reports that Apple has filed since 2010, just search for “%APPLE INC%” in the name column.

Searching for “%APPLE INC%” will also reveal its cik: 320193

If you accidentally delete data in the database file, don’t worry. Just delete the database file and run update() again (see previous chapter).

A first simple example

Goal: present the information in the balance sheet of Apple’s 2022 10-K report in the same way as it appears in the original report on page 31 (“CONSOLIDATED BALANCE SHEETS”): https://www.sec.gov/ix?doc=/Archives/edgar/data/320193/000032019322000108/aapl-20220924.htm

from secfsdstools.e_collector.reportcollecting import SingleReportCollector
from secfsdstools.e_filter.rawfiltering import ReportPeriodAndPreviousPeriodRawFilter
from secfsdstools.e_presenter.presenting import StandardStatementPresenter

if __name__ == '__main__':
    # the unique identifier for apple's 10-K report of 2022
    apple_10k_2022_adsh = "0000320193-22-000108"
  
    # us a Collector to grab the data of the 10-K report. an filter for balancesheet information
    collector: SingleReportCollector = SingleReportCollector.get_report_by_adsh(
          adsh=apple_10k_2022_adsh,
          stmt_filter=["BS"]
    )  
    rawdatabag = collector.collect() # load the data from the disk
    
   
    bs_df = (rawdatabag
                       # ensure only data from the period (2022) and the previous period (2021) is in the data
                       .filter(ReportPeriodAndPreviousPeriodRawFilter())
                       # join the the content of the pre_txt and num_txt together
                       .join()  
                       # format the data in the same way as it appears in the report
                       .present(StandardStatementPresenter())) 
    print(bs_df)

Overview

The following diagram gives an overview on SECFSDSTools library.

Overview

It mainly exists out of two main processes. The first one ist the “Date Update Process” which is responsible for the download of the Financial Statement Data Sets zip files from the sec.gov website, transforming the content into parquet format, and indexing the content of these files in a simple SQLite database. Again, this whole process can be started “manually” by calling the update method, or it is done automatically, as it described above.

The second main process is the “Data Processing Process”, which is working with the data that is stored inside the sub.txt, pre.txt, and num.txt files from the zip files. The “Data Processing Process” mainly exists out of four steps:

The diagramm also shows the main classes with which a user interacts. The use of them is described in the following chapters.

General

Most of the classes you can interact with have a factory method which name starts with get_. All this factory method take at least one optional parameter called configuration which is of type Configuration.

If you do not provide this parameter, the class will read the configuration info from you configuration file in your home directory. If, for whatever reason, you do want to provide an alternative configuration, you can overwrite it.

However, normally you do not have to provide the configuration parameter.

Index: working with the index

The first class that interacts with the index is the IndexSearch class. It provides a single method find_company_by_name which executes a SQL Like search on the name of the available companies and returns a pandas dataframe with the columns name and cik (the central index key, or the unique id of a company in the financial statements data sets). The main purpose of this class is to find the cik for a company (of course, you can also directly search the cik on https://www.sec.gov/edgar/searchedgar/companysearch).

from secfsdstools.c_index.searching import IndexSearch

index_search = IndexSearch.get_index_search()
results = index_search.find_company_by_name("apple")
print(results)

Output:

                           name      cik
      APPLE GREEN HOLDING, INC.  1510976
   APPLE HOSPITALITY REIT, INC.  1418121
                      APPLE INC   320193
         APPLE REIT EIGHT, INC.  1387361
          APPLE REIT NINE, INC.  1418121
         APPLE REIT SEVEN, INC.  1329011
             APPLE REIT SIX INC  1277151
           APPLE REIT TEN, INC.  1498864
         APPLETON PAPERS INC/WI  1144326
  DR PEPPER SNAPPLE GROUP, INC.  1418135
   MAUI LAND & PINEAPPLE CO INC    63330
          PINEAPPLE ENERGY INC.    22701
  PINEAPPLE EXPRESS CANNABIS CO  1710495
        PINEAPPLE EXPRESS, INC.  1654672
       PINEAPPLE HOLDINGS, INC.    22701
                PINEAPPLE, INC.  1654672

Once you have the cik of a company, you can use the CompanyIndexReader to get information on available reports of a company. To get an instance of the class, you use the get get_company_index_reader method and provide the cik parameter.

from secfsdstools.c_index.companyindexreading import CompanyIndexReader

apple_cik = 320193
apple_index_reader = CompanyIndexReader.get_company_index_reader(cik=apple_cik)

First, you could use the method get_latest_company_filing which returns a dictionary with the latest filing of the company:

print(apple_index_reader.get_latest_company_filing())

Output:

{'adsh': '0001140361-23-023909', 'cik': 320193, 'name': 'APPLE INC', 'sic': 3571.0, 'countryba': 'US', 'stprba': 'CA', 'cityba': 'CUPERTINO', 
'zipba': '95014', 'bas1': 'ONE APPLE PARK WAY', 'bas2': None, 'baph': '(408) 996-1010', 'countryma': 'US', 'stprma': 'CA', 
'cityma': 'CUPERTINO', 'zipma': '95014', 'mas1': 'ONE APPLE PARK WAY', 'mas2': None, 'countryinc': 'US', 'stprinc': 'CA', 
'ein': 942404110, 'former': 'APPLE INC', 'changed': 20070109.0, 'afs': '1-LAF', 'wksi': 0, 'fye': '0930', 'form': '8-K', 
'period': 20230430, 'fy': nan, 'fp': None, 'filed': 20230510, 'accepted': '2023-05-10 16:31:00.0', 'prevrpt': 0, 'detail': 0, 
'instance': 'ny20007635x4_8k_htm.xml', 'nciks': 1, 'aciks': None}

Next there are two methods which return the metadata of the reports that a company has filed. The result is either returned as a list of IndexReport instances, if you use the method get_all_company_reports or as pandas dataframe if you use the method get_all_company_reports_df. Both method can take an optional parameter forms, which defines the type of the report that shall be returned. For instance, if you are only interested in the annual and quarterly report, set forms to ["10-K", "10-Q"].

# only show the annual reports of apple
print(apple_index_reader.get_all_company_reports_df(forms=["10-K"]))

Output:

                 adsh     cik       name  form     filed    period                                           fullPath  originFile originFileType                                                url
 0000320193-22-000108  320193  APPLE INC  10-K  20221028  20220930  C:\Users\hansj\secfsdstools\data\parquet\quart...  2022q4.zip        quarter  https://www.sec.gov/Archives/edgar/data/320193...
 0000320193-21-000105  320193  APPLE INC  10-K  20211029  20210930  C:\Users\hansj\secfsdstools\data\parquet\quart...  2021q4.zip        quarter  https://www.sec.gov/Archives/edgar/data/320193...
 0000320193-20-000096  320193  APPLE INC  10-K  20201030  20200930  C:\Users\hansj\secfsdstools\data\parquet\quart...  2020q4.zip        quarter  https://www.sec.gov/Archives/edgar/data/320193...
 0000320193-19-000119  320193  APPLE INC  10-K  20191031  20190930  C:\Users\hansj\secfsdstools\data\parquet\quart...  2019q4.zip        quarter  https://www.sec.gov/Archives/edgar/data/320193...
 0000320193-18-000145  320193  APPLE INC  10-K  20181105  20180930  C:\Users\hansj\secfsdstools\data\parquet\quart...  2018q4.zip        quarter  https://www.sec.gov/Archives/edgar/data/320193...
 0000320193-17-000070  320193  APPLE INC  10-K  20171103  20170930  C:\Users\hansj\secfsdstools\data\parquet\quart...  2017q4.zip        quarter  https://www.sec.gov/Archives/edgar/data/320193...
 0001628280-16-020309  320193  APPLE INC  10-K  20161026  20160930  C:\Users\hansj\secfsdstools\data\parquet\quart...  2016q4.zip        quarter  https://www.sec.gov/Archives/edgar/data/320193...
 0001193125-15-356351  320193  APPLE INC  10-K  20151028  20150930  C:\Users\hansj\secfsdstools\data\parquet\quart...  2015q4.zip        quarter  https://www.sec.gov/Archives/edgar/data/320193...
 0001193125-14-383437  320193  APPLE INC  10-K  20141027  20140930  C:\Users\hansj\secfsdstools\data\parquet\quart...  2014q4.zip        quarter  https://www.sec.gov/Archives/edgar/data/320193...
 0001193125-13-416534  320193  APPLE INC  10-K  20131030  20130930  C:\Users\hansj\secfsdstools\data\parquet\quart...  2013q4.zip        quarter  https://www.sec.gov/Archives/edgar/data/320193...
 0001193125-12-444068  320193  APPLE INC  10-K  20121031  20120930  C:\Users\hansj\secfsdstools\data\parquet\quart...  2012q4.zip        quarter  https://www.sec.gov/Archives/edgar/data/320193...
 0001193125-11-282113  320193  APPLE INC  10-K  20111026  20110930  C:\Users\hansj\secfsdstools\data\parquet\quart...  2011q4.zip        quarter  https://www.sec.gov/Archives/edgar/data/320193...
 0001193125-10-238044  320193  APPLE INC  10-K  20101027  20100930  C:\Users\hansj\secfsdstools\data\parquet\quart...  2010q4.zip        quarter  https://www.sec.gov/Archives/edgar/data/320193...
 0001193125-09-214859  320193  APPLE INC  10-K  20091027  20090930  C:\Users\hansj\secfsdstools\data\parquet\quart...  2009q4.zip        quarter  https://www.sec.gov/Archives/edgar/data/320193...

Collect: collecting the data for reports

The previously introduced IndexSearch and CompanyIndexReader let you know what data is available, but they do not return the real data of the financial statements. This is what the Collector classes are used for.

All the Collector classes have their own factory method(s) which instantiates the class. Most of these factory methods also provide parameters to filter the data directly when being loaded from the parquet files. These are

It is also possible to apply filter for these attributes after the data is loaded, but since the Collector classes apply this filters directly during the load process from the parquet files (which means that fewer data is loaded from the disk) this is generally more efficient.

All Collector classes have a collect method which then loads the data from the parquet files and returns an instance of RawDataBag. The RawDataBag instance contains then a pandas dataframe for the sub (subscription) data, pre (presentation) data, and num (the numeric values) data.

The framework provides the following collectors:

Have a look at the collector_deep_dive notebook.

Raw Processing: working with the raw data

When the collect method of a Collector class is called, the data for the sub, pre, and num dataframes are loaded and being stored in the sub_df, pre_df, and num_df attributes inside an instance of RawDataBag.

The RawDataBag provides the following methods:

It is simple to write your own filters, just get some inspiration from the once that are already present in the Framework (module secfsdstools.e_filter.rawfiltering:

Have a look at the filter_deep_dive notebook.

Joined Processing: working with joined data

When the join method of a RawDataBag instance is called an instance of JoinedDataBag is returned. The returned instance contains an attribute sub_df, which is a reference to the same sub_df that is in the RawDataBag. In addition to that, the JoinedDataBag contains an attribut pre_num_df, which is an inner join of the pre_df and the num_df based on the columns adsh, tag, and version. Note that an entry in the pre_df can be joined with more than one entry in the num_df.

The JoinedDataBag provides the following methods:

Present

It is simple to write your own presenter classes. So far, the framework provides the following Presenter implementations (module secfsdstools.e_presenter.presenting):

Links