从 EDGAR XBRL 文件构建财务表

Question

When companies submit their reports to SEC, a number of files are made available. Eg the latest 10-K of AAPL . From this even the SEC website (and many others) make the tables as structured data .

What I would like to do is to reproduce it myself, but got stuck. Can somebody point me to a detailed, step-by-step description to how to do it?

Usually SO users ask for more specific questions and the list of things the OP tried, so, here is what I tried, what I understand:

• The six files at the bottom (8-12, 15 in the example) have all data used. Basically 8-12 submitted by the company and 15 is an extract from the in-line Xbrl of the filing itself (1 in the example).
• The extract file (15) has all the Xbrl instances listed and all the contexts. It is very clear.
• The XSD file (8) has a list of all the forms and all company specific elements. The first is given in link:roleType blocks, giving a Definition and a list of linkbases where it appears (although sometimes they do not appear). The second is given as <xs:element>-s.
• The presentation file (12) has the same list of tables.
• The definition (10) and label (11) linkbases should give some more details, eg the company specific label of a certain data.
• The calculation linkbase is not really needed (I guess) it is more a validation that the totals are indeed calculated how it is indicated.

What I do not understand though:

• What is the right approach to build up the tables from these files? Is it by going through the XSD/PRE forms and find the data for them in the Extract file or the other way round?
• Regardless how hard I tried I could not find the link (with all the locators and arcs) between a datapoint in the Extract file and a label in the LAB file. As a human, it is "easy", but as a machine, the names are always slightly different (eg (a) loc_XYZ changing to lab_XYZ; (b) a name "XYZ" has its own version and an "XYZAbstract" version, (c) names like XYZ have numbers attached to them XY_123) and so the link between the "two ends" I cannot establish.

This is why I would like a step-by-step explanation, like:

• Take file... first. There iterate through the <...> tags. For every tag find a <...> tag in file..., where attribute... is equal to the... attribute of the iterated tag. Etc.

Thanks,

PS (I am not interested in available software and services that already do this, neither in some specific libraries to call. I simply would like to extract the information, using the plain text files.)

Answer 1

If you're looking to process XBRL without re-using existing XBRL software, then the best place to start would be the XBRL Specifications . In particular, the section on XLink in XBRL will explain how XBRL Linkbases work, including the labels used in the xlink:from and xlink:to attributes. The short answer is that those attributes just contain arbitrary identifiers that reference the xlink:label attribute of an element elsewhere in the file.

The specific question of how you construct financial tables from the XBRL data is tricky. An XBRL Report does not contain any explicit information that associates facts in the report with tables. You can build a list of the concepts in a section of a financial report from the presentation linkbase, but you'll often find that you have more facts than expected using those concepts. For example, if you build a list of concepts from the Balance Sheet section, you'll often find that you have facts that use those concepts but with additional dimensions because they were tagged from a note providing a breakdown of that concept.

The SEC use a heuristic-based approach to organising facts into tables. This process is documented in section 6.24 of the Edgar Filer Manual .