There has been a change of plan. I no longer want to scrape the form 4 filings. I now want to scrape forms with the code DEF14A. These come out on an annual basis. I would like to scrape data for as many companies as possible. This would include the constituents of the S&P 500 and maybe some of the larger indices. I also need the historical data. You will have get data from the DEF 14A file for every available year. I want to extract just two fields: the name of the company and the percentage of the shares owned by company insiders.
Extracting this information will be fairly challenging because the DEF 14A filings do not follow a set pattern. You will need to create a program capable of scanning through the document and locating the relevant text. You will have to employ some kind of proximity system and make use of "wildcard" searches.
Here is an example, its the most recent DEF 14A filing submitted by Microsoft: [url removed, login to view] The relevant table appears on page 11. The table lists the names of the managers and the percentage of the company that they personally own. Collectively the managers own [url removed, login to view] percent of the company. In this example I would be looking to extract the word "Microsoft" and the number 9.46.
I would like to do this for every form DEF 14A filing in Microsoft's history. The big problem is that not all the companies use the same language: Here is a list of different phrases that various companies use:
All Directors and Executive Officers as a group (including Named Executives) (32 persons) beneficially owned [url removed, login to view]% of Ford common stock or securities convertible into Ford common stock as of February 1, 2012 6 1,226,353 9,476,285 13.3%
All directors and current executive officers as a group (12 persons) 3,991,056 6,348,957 10,340,013 2.6%
Directors, nominees and Named Executive Officers as a group (11 persons) 6,185,034 (12) [url removed, login to view] Executive officers and directors as a group (13 persons)(19) 1,490,847 6.7%
All executive officers and directors as a group (14 persons) 4,782,931 6.4 All directors and executive officers as a group (18 persons) 661,671 1,440,299 269,802,371,776 4.3%
All directors and executive officers as a group (12 persons)(6) 870,542 [url removed, login to view] All Company directors and executive officers as a group (19 persons) 433,960 596,312 1,030,272 1.5%
All nominees, continuing directors and executive officers as a group (20 persons) 5,944,103 16,824,264 139,82 8,03,926,234 (4) 23%
All directors, director nominees and executive officers as a group (12 persons) 13,412,40 17.0% All current executive officers and directors as a group (10 persons) (7)...................... 19,059,809 1,275,405 52.1%
All directors and executive officers as of November 13, 2012 as a group (13 persons) 17,011,477 624,969 17,636,446 54.8%
There does appear to be certain words that recur most of the time. If you created a set of rules that said something like:
LOCATE TEXT WITH THE WORDS: "as a group (?? persons)"
EXTRACT THE NEXT AVAILABLE NUMBER BETWEEN 1 AND 100 THAT IS FOLLOWED BY A PERCENT SIGN
If you managed to devise rules like that then you would extract the correct percentage most of the time. The failure rate would be fairly high. I do not mind if a fairly high percentage of the data is either missing or incorrect.
5 freelancers are bidding on average £582 for this job
Hello, sounds like interesting and challenging project. If you are interested in my bid, please contact me so i can continue with the research. Samples of previous work will be provided upon request. Best Regards.
Got a bevy of bots scraping full text breaking news 24/7 using rules-based text processing systems for similarly difficult-to-scrape data (for similar reasons, too).