Digital Financial Reporting

Many people try and use information from a financial report at the level of the individual facts in the report.  That is one way.  But there is another way.  That other way is working with sets of facts.  For example, the "balance sheet" is a set of facts, the "income statement" is a set of facts, the "maturities of long-term debt" disclosure is a set of facts.

But how do you identify those sets of facts?  That is what I will explain.

So, I will be using this Microsoft 10-K filing to explain. You can go there and get the XBRL instance, use the HTML version of the 10-K.  Or, you can look at that same report using the XBRL Cloud Viewer to view the Microsoft filing here.

To start, have a look at this Excel spreadsheet (within a ZIP archive). That spreadsheet contains information on the Network, the Components, and what I call Blocks contained within that Microsoft 10-K.

There are 127 Networks.  Now, Networks is generally not to right level to be looking at sets of information within a report.  Why?  Because companies represent information differently.  For example, companies could put two Tables (or Hypercubes) within one Network; so just looking at the Network would give you different results.  So, what about Tables (or Hypercubes)?

There are 127 Tables in the Microsoft 10-K.  Why the same as Networks? Because Microsoft used a Table (Hypercube) to represent the information in each Network.  So for Microsoft there are the same number of Networks and Tables, there is a one-to-one correlation.  But, this is not true for every XBRL-based reports.  Sometimes companies put TWO Tables within a Network.  I call the Network + Table combination a Component.  Those are on the second tab in that Excel spreadsheet.

There is another layer which I call the Block.  There are 191 Blocks in that Microsoft 10-K.  A Block goes one layer deeper.  It breaks Tables down into individual "roll ups" and "roll forwards" and other sets of information.  Now, these sets are 100% consistent across all XBRL-based reports created by any company.  Why?  Because the set is determined by logic, not by the preferences of how information is presented in a report.

What I mean is this.  The "Inventory, Net, Components Roll Up" of a report will ALWAYS be one set of information.  That set will be a roll up. This is by definition.  That roll up always has a total concept, every roll up does, again by definition of what a roll up is.  That total concept is "us-gaap:InventoryNet".  That Level 4 Disclosure roll up detail is always represented one of two Level 3 Disclosure Text Blocks.  There are other rules.  Here is a human readable version of those rules.  Here is that same information in machine-readable form (it was actually used to generate the human readable version of those rules).

That information can be used to identify what information is being represented by a Block of information in an XBRL-based report.  That information can also be used to identify where that information is in the report no matter what Network or what Table.

Here are the 191 Blocks represented in the Microsoft 10-K in machine readable form:  Model Structure | Fact Tables.

Here is information for about 70 of those 191 blocks in human readable form. Why do I only have human readable information for 70 of the 191 Blocks?  Because I only have machine-readable rules for 70 disclosures. Here is the evidence package for Microsoft.

Pretty nifty!  (If you don't understand why this is nifty, read Computer Empathy.) If you want to understand more, read Putting the Expertise into an XBRL-based Knowlege Based System for Creating Financial Reports.

There are a number of people using this US GAAP test data for the 2017 10-Ks.  Here is a summary of that information:

• Human readable list of the 5,734 filings
• Machine readable RSS feed with 5,734 filings
• ZIP archive containing Excel file with the list of 5,734 filings
• ZIP archive containing XML infosets of model structure of each of the 5,734 filings taxonomy
• Second ZIP archive with Excel spreadsheet with list of 5,743 filings with additional metadata: total assets, number of facts reported
• Reporting Styles Used by Companies US GAAP
• Overview of reporting styles, US GAAP
• Inconsistencies with reporting styles, US GAAP
• Reporting checklist rules, US GAAP
• Disclosure mechanics rules, US GAAP
• Disclosures
• Topics
• Fundamental Accounting Concept Relations Rules
• Conceptual model
• US GAAP Ontology Prototype
• US GAAP Concept Roll Ups

The Forbes article, Will Artificial Intelligence And Cloud Accounting Replace The Accountants Of Tomorrow?, explains that robots and artificial intelligence will replace 7% of all jobs in the U.S. by 2025.  So, that 7% is actually a net of a 16% loss of existing jobs and a 9% increase in new jobs, per Forrester.

The article says:

In this not-too-distant future, businesses will be able to afford powerful CFO advice without the cost of a full-time employee. All of those menial, time-consuming tasks that used to eat up hours of a day will be gone – a victim (or virtue, depending on your perspective) of the automation process. This doesn't eliminate the need for accountants, but rather empowers them. It frees up time in their day to work with clients in a more effective way.

The article goes on to explain that disruption is what you make of it:

When people hear the word "disruption," they tend to assume it has a negative connotation. However, this doesn't have to be the case --technological disruption is more about creating an opportunity than creating a threat. For accountants working with small business owners, disruption is less about, "What I do can now be automated, putting me out of a job," and more about, "Now that I can automate X, Y and Z, I can change my job for the better and for the future."

Artificial intelligence has an issue.  You need to be aware of this issue because it means that artificial intelligence should not be employed for solving certain classes of problems.  The issue is straight forward.

There are two types of reasoning: deduction and induction. These two videos help you understand the difference between deduction and induction: Video 1; Video 2.

The bottom line is that deduction is certain; you can be logically certain of the conclusions reached using deduction. That is the goal of deduction, certainty. Induction is not certain, induction is based on probability.

The Black Swan Problem of Induction clearly explains the problem of induction.

“No amount of observations of white swans,” he said, “can allow the inference that all swans are white, but the observation of a single black swan is sufficient to refute that conclusion.”

Many employing artificial intelligence use inductive reasoning systems.  They believe that these systems can work because the problems caused by induction can be managed effectively.  That may be true.  But it likewise may not be true for certain knowledge domains.  The call here should be made by domain experts, not computer scientists that do not have a deep understanding of the domain.

#############################

One problem to explain why AI works

Reporting styles break financial reports into patterns of reporting styles.  I have reporting styles for US GAAP and IFRS. Reporting styles organize fundamental accounting concepts into different allowed relations that appear in financial reports.  I don't make the reporting styles up, I simply identify the patterns of reporting styles within US GAAP and IFRS reports that have been submitted to the SEC.

I verify the fundamental accounting concept relations per each reporting style each quarter for both US GAAP and IFRS as of last quarter.

I am now going one layer deeper within reports, what concepts are used within roll ups to the fundamental accounting concepts.

So, consider this reporting style: COMID-BSC-CF1-ISM-IEMIB-OILY-SPEC6.  This same style is used in US GAAP and in IFRS. The model structure is essentially identical. There are some minor differences but for all practical purposes, they are the same.

Here are the mappings of the fundamental accounting concepts to the US GAAP and IFRS XBRL Taxonomy concepts:  US GAAP | IFRS.

If you notice those mappings there is a link for all the XBRL concepts, US GAAP or IFRS.  For example, us-gaap:Assets and ifrs-full:Assets have links.  Click on the link and you will be taken to a web page that shows you how SEC filers are using those concepts relative to other concepts.

So, for example, if you click on us-gaap:Assetsyou see the 532 different concepts that public compainies used as line items that contribute to the total "us-gaap:Assets" in SEC filings.  Note that this list includes 100% of the companies that use the COMID-BSC-CF1-ISM-IEMIB-OILY-SPEC6 reporting style.  Here is the same information for ifrs-full:Assets. There are 270 different concepts that are used within roll ups of "ifrs-full:Assets".  Now, for IFRS I am using all 339 filings.  Currently I am experimenting to see what gives me the based information.  It seems like doing the breakdowns by reporting style is best.

I also created a web service that provides this information for US GAAP and IFRS. That can be useful to those creating reports.  I am not going to go into detail on this.

So, I am finding some interesting things.  Consider the roll up of the components of concept Gross Profit. Here that is for US GAAP and for IFRS.  If you look at either of those breakdowns of concepts you see the same thing for US GAAP and for IFRS. You see that most reports contain "Revenues" and "Cost of Revenues", which you would expect, and then you see pretty much odds-and-ends being reported.

Now, I have to do some additional analysis to figure out exactly what I am seeing.  I think I am seeing some reporting inconsistencies, some missing taxonomy concepts, and errors filers are making creating extension concept which they really should be be creating.

Here are some clear errors.  The US GAAP XBRL Taxonomy has a concept us-gaap:NoncurrentAssets.  It also has the concept us-gaap:AssetsNoncurrent.  Those concepts mean two different things.  I think what is going on is that a few filers are providing detailed breakdowns of us-gaap:NoncurrentAssets in their geographic area disclosure.

Here are some clear errors:

• Cost of revenue included in operating expenses: If you look at the total us-gaap:OperatingExpensesyou will see the concepts "us-gaap:CostOfRevenue" and "us-gaap:CostOfServicesDirectLabor" and "us-gaap:CostOfServicesDepreciationAndAmortization" among other DIRECT operating expenses included in the total "us-gaap:OperatingExpenses" which represents INDIRECT operating expenses.  That is clearly an error.  If a filer is reporting DIRECT and INDIRECT expenses together, the concept "us-gaap:CostsAndExpenses" should be used which includes both direct and indirect expenses.
• Inappropriate extension concepts: If you search on us-gaap:AssetsCurrentand you look at the list you see "none:Inventory", "mpaa:NetInventory", "smp:FIFOInventoryNet", "emkr:InventoryCurrent", "dpm:OtherCurrentAssets"; all of these are clearly inappropriate extension concepts.
• Inappropriate use of commitments and contingencies: As I understand it, the line item "Comitments and contingencies" is a US GAAP line item and never used in IFRS.  Yet, if you search the total ifrs-full:EquityAndLiabilitiesyou see filers using concepts "bce:CommitmentsAndContingencies1", "emitf:CommitmentsContingenciesLiensAndCollaterals", "forty:CommitmentAndContingencies", "iag:CommitmentsAndContingencies1", "abx:CommitmentandContingencies1", "nept:CommitmentsAndContingentLiabilities", "oncyf:CommitmentsAndContingencies1", "qbmi:IfrsCommitmentsAndContingencies", "umc:CommitmentsAndContingencies1"; either (a) these are all errors or (b) the IFRS XBRL Taxonomy is wrong and needs to include this concept and some accountants that think that this is NOT allowed are mistaken and this line item is allowed on the balance sheet.

If you find some interesting errors when you fiddle with this tool please let me or the SEC or ESMA or FASB or IASB know.

This PDF walks you through a presentation of this information.