Digital Financial Reporting

Many people focus on XBRL saying that XBRL about not having to rekey information. While it is true that if XBRL formatted information is expressed correctly that information does not need to be "rekeyed"; but I would suggest that a more enlighten focus might be the following: What does it mean if you don't have to rekey information?

What if you could predictably, reliably, repeatedly grab a piece of information; ascertain the appropriate context of the information which was an OUTPUT of that process; and send that information on as an INPUT to some subsequent process. What if a meaningful exchange of information could be achieved between two computer processes?

What if you could "chain" processes together? That is why people often refer to processes as a supply chain.

One of the things which could occur is the business users can delegate work to software agents.

So what the heck is a "software agent"?  People sometimes use other terms such as "intelligent agent" or "smart agent".  Using terms such as intelligent and smart, in my view, just add hype into the equation which often comes with things related to technology.

I don't want to contribute to this hype by setting unrealistic expectations of the state-of-the-art.  People many times profess exaggerated claims which they can never deliver in order to fool ignorant buyers in order to make a quick buck.  That is not my goal here.

My goal is to explain that some very basic work can be delegated to software agents and if this is done correctly, software agents can perform meaningful work, can reduce costs, and can increase quality.  The first step which is required to make a software agent work appropriately is the meaningful exchange of information to a software agent.

As is said, "garbage in, garbage out".  If a process outputs 98% accurate information; that process can never be completely automated because humans need to be involved in order to fix the 2% of the information which is incorrect. So before processes can be 100% automated, information exchanged via that process needs to be 100% correct, complete, consistent, and accurate.

I found the following two definitions of a software agent which help you understand what a software agent is.  In his paper, Software Agents: an Overview, Hyacinth S. Nwana defines a software agent as:

we define an agent as referring to a component of software and/or hardware which is capable of acting exactingly in order to accomplish tasks on behalf of its user

In their paper Intelligent Agent, Hanh Tran & Thaovy Tran use the term intelligent agent which has the following definition:

An intelligent agent is a software that assists people and act on their behalf. Intelligent agents work by allowing people to delegate work that they could have done, to the agent software. Agents can perform repetitive tasks, remember things you forgot, intelligently summarize complex data, learn from you and even make recommendations to you.

I like both definitions.  Neither has hype in my view.  Two key aspects I would point out are the following.  The first definition states "capable of acting exactingly".  That is critical. What an agent does needs to be predictable and reliable. The second key aspect is "repetitive tasks".  A software agent needs to repeat the same predicable, reliable process over, and over, and over.

I want to take the definition of a software agent just a little bit further and show the categories of software agents pointed out by Hyacinth S. Nwana' s paper.  Here is a diagram which is provided in that paper:

Idea from Hyacinth S. Nwana, Divine T. Ndumu: An Introduction to Agent Technology, Re-Drawn by Mobile Computing, Dept. of IECS, Feng Chua University, R.O.C., 2003.

That paper provides a lot of detail on a lot of rather sophisticated stuff which software agents might be able to do, but I don't want to go there in this very basic explanation of software agents.  Go read the paper if you want that detail.

But I do want to provide some insight on the sorts of things software agents can do.  One good example of a software agent can be seen in the vision of what is called the SEC's RoboCop. What the media calls "RoboCop" is a software agent which enforces an accounting quality model established by the SEC.  What is in the accounting quality model?  Who knows.  Perhaps something as basic as checking to see if SEC filer's have balance sheets that balance or more likely more sophisticated checks of SEC XBRL-based financial filings.

RoboCop might seem "out there" because we cannot see how it works yet.  Here is an example of something which you might be able to better get your head around because you can go check it out.  I pulled reported information for a set of 51 fundamental accounting concepts from SEC XBRL financial filings. I also tested 21 relations between those concepts to try and make sure that the informaton I was grabbing was accurate.  I found the information to be about 98% accurate. Eventually the information will be 100% correct I predict.  That fundamental set of information has the reported facts necessary to compute the sustainable growth rate of every SEC filer.  Here is that formula:

Sustainable Growth Rate = ((Net Income (Loss) / Revenues) * (1+((Assets - Equity) / Equity))) / ((1 / (Revenues / Assets))-(((Net Income (Loss) / Revenues) * (1+(((Assets - Equity) / Equity))))))

Imagine a software agent which monitors the sustainable growth rate of public companies for an analyst looking for investments. The software makes the analyst aware of companies who reach certain thresholds.  This would allow the analyst to then focus additional analysis on the specific entities which have met her criteria.

Another really simple example of an agent is XBRL Cloud's EDGAR Dashboard.  While that dashboard monitors verification/validation of SEC XBRL financial filings which is useful for making sure the SEC XBRL financial filings are properly created; it is not that big of a stretch to imagine other types of these dashboards which could be created.

One final example of the utility of a software agent is the analytical review process performed by auditors.  Finding the information, lining it up in an application, computing the variance, finding and adding information about how peers of an entity performed is mindless work which software applications can easily perform for an auditor.  That frees the auditor to focus on the analysis of the actual information and they will have more time for the analysis because software agents performed the mindless part of obtaining and providing the information.

The bottom line here is that the notion of software agents or intelligent agents or smart agents has been around for some time.  The understanding of the sorts of things that these software agents can do for business users is in its early stages. As agents are better understood and as the quality of the information these software agents rely on to function correctly improves, these software agents will deliver increasing utility to business users.  These software agents will never replace highly skilled professionals working with something like financial information.  They will mainly perform mindless, repetitive work; the sort of work computer software does well.  Information gathering will be significantly easier so the humans can focus on the sorts of things that only humans can do. Costs can be reduced and quality can be increased.

A hypercube is simply a table which can have any number of dimensions, or "n" dimensions.  There are two constraints which must be differentiated in your mind: the medium using to display the information and the characteristics one is working with which describe the information.

A scalar is a piece of informtion which has no characteristics, it stands on its own. A scalar does not need to be characterized.  For example, the value of pi is a scalar, it never changes; it always has the same value for everyone. (Pi or p is the ratio of a circle's circumference to its diameter and always has the value of equal to 3.14)

Visually, this is what a scalar looks like:

A list has one dimension.  Dimensions are a model for expressing characteristics of information.  Dimensions effectively conceptualized for unambiguous interpretation.

For example, the following is a list of numbers.  But you don't know any more about the characteristics of the number.  You don't know if the number is for, say, revenue, net income, assets; you don't know if the number is in US Dollars, Euros, or Yen; you don't know if the number is for Microsoft, Apple, IBM, or some other company.  All you have is a list of numbers:

A table has two dimensions physical dimensions: rows and columns which intersect to form cells.  A spreadsheet is a good example of presenting a table.  Other terms used for table are matrix and array.

A table (spreadsheet, matrix, array) can have any number of rows and columns (or, to the limit of the software application; Excel for example does have limits).

So a table can hold pretty much any set of information and the characteristics which describe the information.  But as the number of characteristics, rows, and columns increases; it gets harder and harder to work with the information.  Also, if you are working with DIFFERENT tables which have different dimensions/characteristics things get more challenging.

This is why relational databases are so handy.  Relational databases store information in the form of tables which can basically hold anything.  But as the shapes of the tables change, it gets more and more challenging to put things into relational databases or manage all the things you have in a relational database.

Here is a small table with a value and four characteristics which conceptualize the value in the column "Fact Value":

A cube can be thought of as a three dimensional presentation of a table/matrix/array. "Space" has three dimensions.  Paper only has two dimensions though.  You can present three dimensional information on the two dimensional medium of paper by grouping information and printing one group, then the next, then the next linearly down the piece of paper.  The software application Excel achieves three dimensions by adding the notion of a "sheet".  In the old days of spreadsheets there was only one sheet.  But today, Excel has the notion of a "workbook" which contains "sheets" which then contain rows and columns which intersect and form cells.

A pivot table is an approach to visualizing a complex table of information. There are two things going on with pivot tables.  The first is the number of characteristics a value has and the second is how to show all of the information, including the characteristics, in usable ways.  So you have both the information that you are dealing with and how to present the information.  Again, a computer screen only has two dimensions, it is flat.  It can be a challenge to render information which has more than two dimensions in a two dimensional space. A pivot table is one way of achieving that goal.

A hypercube is an "n" dimensional table/matrix/array. Where "n" represents any number of dimensions or characteristics.  And so every table is basically a hypercube.  Different tables or hypercubes can have different numbers of characteristics.  It is easy to work with tables that have exactly the same characteristics or dimensions together.  The more different the dimensions/characteristics, the more challenging it is to put different tables/hypercubes together.

Saying this another way, different tables/hypercubes have different shapes.

The more dimensions or characteristics that a hypercube has, the more challenging it is to rendering that information in two dimensional space such as paper or a spreadsheet.  Computers can be better for presenting this sort of information to humans in readable forms because computers can be dynamic, like a pivot table.

Imagine an application which is good at handling n-dimensional information (i.e. any number of dimensions or characteristics. Consider this visualization of a four dimensional object or hypercube: 

Or this visualization of a five dimensional object or hypercube:

Because a computer is dynamic, it is easier for a computer to take n-dimensional information and display it (a) in human readable terms and (b) dynamically, the user can change the representation to suit his or her specific preceived needs or demands.

A pivot table is one example of that sort of application.

SEC XBRL financial filings are nothing more than many, many hypercubes.  Those hypercubes; referred to as explicit [Table]s or implied tables which have only reporting entity (or CIK), a period, and a concept which contextualize reported facts; must be managed by software application which work with the information.

Some of those software applications show static, unchangeable views of the information.  The SEC Interactive Data Viewer is an example of a static application.  Other applications let you change the view of the information, such as the XBRL Cloud Viewer or CoreFiling's Matnify.

An ontology seems to be a means to an end. Probably the most important notion to understand about digital financial reporting is that of secondary use ontologies.  Secondary use ontologies are a currency of the information age. Understanding this currency is a key to thriving in the information age.

I don't have all of these pieces synthesized in my mind yet, like many of my blog posts this is more of a stream of consciousness dump on the journey toward understanding.

Something that I though was true may not be true at all.  Or rather, it may only be partially true. The premise of "the semantic web" which was held by some is "anyone can say anything about anything."  I thought that that statement was hogwash.  Why?  Well, think about the accounting equation: Assets = Liabilities and Equity. Who would dispute that fundamental building block of accounting?

Well, perhaps no one would dispute that specific fact.  However, there are many other facts which exist which might be disputed by others.  Said another way, there are likely some facts that few if anyone disputes and there are likely some facts that people do rightly dispute.

So what is the point?  The point is this, maybe there is not just one financial report ontology.  Maybe there are many.  Perhaps there is some base ontology that everyone agrees on such as "Assets = Liabilities and Equity" and other fundamental accounting concepts and relations. Everyone agrees at this level. It is kind of like how two attorneys who are arguing a case establish the facts of a case, the things that neither side disputes. This allows them to focus on the things where they don't agree to be the focus of the case they are arguing.

This is actually not my notion; there are actually names for these different camps.  Remember that an ontology is a description of the "things" and the "relations between things" which exist within a some determined field of study or "domain".

So, an ontology is a representation of something that exists in reality. An ontology is a "window on reality". There are two views of "reality" which are common in philosophy:

• Ontological materialism: The belief that reality exists regardless of human observers. (i.e. there is only ONE reality)
• Ontological idealism: The belief that reality is constructed in the mind of the observer. (i.e. reality is determined by the observer)

Does this mean that everyone can create their own reality? No.  That is where epistemology comes in. Epistemology is the theory of knowledge.

So, what is knowledge? Some say that knowledge is "justified true belief".  Basically, it seems that you need to have solid grounds for holding a belief and that you must be aware of such grounds.  You can't just make stuff up.  You must be able to justify your claims or beliefs.

How do you justify beliefs or claims? Evidence. Evidence must be of good quality and evidence should be logical and reasonable.  Hard to argue with that.  But how do you gather this evidence?  Well, again from philosophy, there seems to be two approaches to that:

• Empiricism: Evidence of true knowledge is primarily found on input from our senses; by experience and by observations.
• Rationalism: Evidence of true knowledge is primarily found based on reasoning; by research results which are verified by reasoning.

How does knowledge differ from opinion or belief? 

Epistemology has other aspects which explain how knowledge is obtained such as "propositional knowledge" which is the knowledge of facts and "practical knowledge" which is know how.

Why is knowledge important? Here is a great quote by Nicholas Rescher, a philosopher:

...Knowledge brings great benefits. The release of ignorance is foremost among them. We have evolved within nature into the ecological niche of an intelligent being. In consequence, the need for understanding, for "knowing one's way about," is one of the most fundamental demands of the human condition.

"Knowing one's way about." Before digital financial reporting, the only way one had to "know one's way about" was the information one had in their head.  What one person had in their head may have been different than what another person had in their head.

The rigor and discipline of trying to articulate in the form of an ontology what is in your head yields two things as I see it.  First, it yields clarity. Articulating this information in a form that both parties can better observe yields clarity.  Second, it yields something that a computer can understand. Digitizing financial reports is a lot of work, but it makes it so computers can interact with those digital financial reports.

At some level, there needs to be agreement on "reality".  For example, the fundamental accounting concepts and relations between those concepts of a financial report is such a level. At other levels, secondary use ontologies which build upon the fundamental level, there may be different views of "reality".  Distinguishing between these two levels seems critical.  It is highly likely that more variability will exist in secondary use ontologies.  However, those creating those secondary use ontologies cannot just make things up.  They must justify their beliefs and claims with quality evidence, logic, and reasoning.

I put together a play list of a handful of short videos which digs a little deeper into semantics, ontology, and epistemology.  These are terms business people are not that comfortable enough with.  Given that digital financial reporting is inevitable, perhaps business people need to get more comfortable.

Business people need to understand the currency of the information age. Computer readable knowledge is part of that currency.

This Forbes article, XBRL Would Be Wonderful If It Always Worked, is right. For the information to be useful, it has to be 100% correct.  Or close.  My fundamental accounting concept analysis shows an accuracy rate of 97.6%.  The analysis also shows every SEC XBRL financial filing which contributes to an anomaly being detected.  The filers could fix their filings to get this information correct, then the 100% would be realized.

Or, the SEC could institute a set of less than probably 50 automated tests which are executed with the other testing which is done to determine whether the filing will be accepted by the SEC.  That would push the accuracy rate for that specific area to 100%.

Sure, more automated tests would be necessary, but these fundamental accounting concepts would be a good start!

As part of my efforts to understand SEC XBRL financial filings, I analyzed 7,160 such SEC XBRL financial filings, all 10-K filings, looking at a set of 51 fundamental accounting concepts and the relations between those concepts I would expect to see in those filings.

What do I mean by "fundamental accounting concept" and "relations between those concepts"? Well, you can get all that information here on this Fundamental Accounting Concepts Wiki. Examples of fundamental accounting concepts include: Assets, Current Assets, and Noncurrent Assets.  An example of a relationship is: Assets = Current Assets + Noncurrent Assets.

And this is not about requiring a filer to report noncurrent assets or thinking that if a filer does not report noncurrent assets that you cannot figure out what noncurrent assets is.  If assets is reported and current assets is reported, a child with a calculator can figure out that noncurrent assets = assets - current assets.

And this considers that some filers report a classified balance sheet and others provide an unclassified balance sheet.  Also, some filers have a single-step income statement and others report a multi-step income statement.  Some filers combined their income statement and statement of comprehensive income into one statement, some report that information in two statements.  That does not matter. Some filers don't even report comprehensive income.  Why?  Because it is zero, it is not because they don't have to; it is implied to be zero.

Now, don't confuse these fundamental accounting concepts with the US GAAP XBRL Taxonomy concept used to express that fundamental accounting concept.  A simple example will show what I am talking about. So for example, the 2012 US GAAP XBRL taxonomy or earlier allowed Liabilities and Equity to be expressed using one of two different concepts: us-gaap:LiabilitiesAndStockholdersEquity or us-gaap:LiabilitiesAndPartnersCapital.

In the 2013 US GAAP XBRL Taxonomy deprecated (retired) the concept us-gaap:LiabilitiesAndPartnerCapital. The fact that the concept is deprecated shows two things.  First, that the US GAAP XBRL Taxonomy expresses many fundamental accounting concepts using multiple taxonomy concepts.  Second, that these issues will likely be corrected over time as this issue was corrected.

However, until such time one needs to map concepts used by SEC filers to a set of fundamental accounting concepts in order compare information appropriately.  That is exactly what I have done.  You can see that mapping here in this computer readable XML file. And here is all this metadata you can read in Excel.

If you look through the mapping file you start to understand why something like the fundamental accounting concepts are necessary. Analysis software does not understand these relations unless you tell the software about the relations.

So are these fundamental accounting concepts correct?  Are the relations correct?  Well, 97.6% of SEC XBRL financial filings say that the concepts and relations are correct.  See this summary:

(Click image for larger version)

This information helps you see that these fundamental accounting concepts and the relations between those concepts are correct:

Here is the same information above shown graphically:

(Click image to view larger image)

The data and the chart of the data show that 91% of all SEC XBRL financial filings have 3 or less anomalies (i.e. they follow the 51 fundamental accounting concepts and the relations between the concepts in all but 3 cases within their filing).

What does this mean? Well, it means a number of things.  First, as I mentioned in another blog post; automated testing of things like this can help improve information quality of SEC XBRL financial filings. Second, the fundamental accounting concepts and relations seem to be correct.  If they are not correct, then what?  Change them until you get a set that everyone agrees are correct.

Are these the only relations which exist in SEC XBRL financial filings? Certainly not.  These 51 concepts and 21 relations are a beachhead.  If this information is correct then EVERY SEC XBRL financial filing can be compared at this level. If they can be compared at this high level, then one can drill down into lower levels and both see the concepts and relations, write the business rules, and improve the quality at the next level down.

The many different disclosures have business rules which are followed.  For example, a disclosure of the components of inventory would always have total inventories, should always foot, and would always exist if the line items inventories existed on the balance sheet.  And so it is with the 1000 to 2000 (I don't know what the number is, but it is not infinite) other disclosures.

All these business rules make up a knowledgebase of information which will be useful for many things. One thing that this knowedgebase will be used for is making sure SEC filers create their digital financial reports correctly. This is not about changing US GAAP or telling filers how to report.  This is about common sense, rational, sensible use of XBRL to express the information you want to express.

Balance sheets balance.  These concepts and relations are fundamental.