Understanding Prototype Theory and How it Can be Useful in Analyzing and Creating SEC XBRL Filings By Charles Hoffman This information is inspired by the book Everything is Miscellaneous: the power of the new digital disorder, by David Weinberger, chapter 9, pages 173 to 198. That chapter has detailed explanations and reasoning which supports prototype theory. Fundamental understanding of prototype theory Fundamentally there are two perspectives to understanding what something is. Aristotle s definition view perspective was that A thing is a member of a category if it satisfies the definition of the thing. The second perspective, prototype theory, is that we can know what something means even if it can t be clearly defined and even if its boundaries cannot be sharply drawn; concepts can be clear without having clear definitions if they re organized around undisputed examples, or prototypes, as Eleanor Rosch the inventor of prototype theory calls them. As an example, one can understand that something is a chair by understanding as many properties as possible about the thing you are looking at, looking at the properties of a chair as defined by a prototype (the undisputed example), and then predicting whether the thing you are looking at is a chair by comparing the properties you are looking at with the properties of a chair. By contrast, the definitional view draws sharp lines whereas the prototype view works because things can be sort of, kind of in a category. Prototype theory relies on our implicit understanding and does not assume that we can even make that understanding explicitly. Problems with SEC XBRL filings SEC XBRL filings provide basically no top level foundation for comparability. Two candidates as a basis for comparison are networks and [Table]s. However, each SEC XBRL filing defines its own networks and no two networks are the same. That rules networks out as a basis of comparison. Within an XBRL taxonomy [Table]s can be used for expressing different sets of information, for example the Statement [Table] is used on the balance sheet, income statement, statement of cash flows, and a number of other statements. Other [Table]s are used multiple times within the US GAAP taxonomy and define different sets of information. As such, there is no mechanism to define a set of information. Looking at this situation from the bottom up, there are approximately 15,000 concepts within the US GAAP taxonomy, too detailed a perspective for any useful comparison at the individual concept level. 1
To exacerbate this situation, SEC filers can extend the US GAAP taxonomy adding additional networks, explicit [Table]s, implicit tables (i.e. everything within a network which is not within an explicit table is within an unnamed implicit table), [Axis], [Line Items] or concepts, and so forth. When an SEC XBRL filer expresses their information, they create new networks which are comparable to no other network, they define [Table]s which could be used to express many different sets of information, tables could be defined implicitly or explicitly, and the [Axis] on each information set have no real pattern. This problem seems unsolvable, but is it really unsolvable? Looking deeper in to SEC XBRL financial filings If you look deeper into financial filings you realize that there are patterns within the information. For example, consider this small slice of the 2011 US GAAP Taxonomy which is used to express nonmonetary transactions: Consider the following: The [Line Items] could be expressed as a text block (i.e. HTML fragment) or detail tagged. The HTML fragment would use the concept Details of Nonmonetary Transaction [Table Text Block] and if the information were detailed tagged it would use some combination of the six concepts within the Nonmonetary Transaction [Hierarchy]. But either way, the information is the same. The concepts within the Nonmonetary Transaction [Line Items] are used nowhere else in the US GAAP Taxonomy. As such, if one sees one or more of these concepts within an SEC XBRL 2
filing, then one can assume with a high level of confidence that the thing which contains one or more of those concepts is highly likely to be a nonmonetary transaction. Financial reporting rules and logic demand that certain concepts be present. In financial reporting rules certain information is always required to be disclosed, certain information is required to be disclosed if a certain event or circumstance occurs during a financial period, certain information is common practice, and certain information is reported at the option of the filer. The base set of information will always exist, it will always be logical based on financial reporting disclosure requirements and logic. For example, an SEC filer would be highly unlikely to report Nonmonetary Transaction, Fair Value Not Determined as the only concept within a nonmonetary transaction. If additional required disclosures which expand the base disclosure is presented, if common practice disclosures are provided, or additional optional information is disclosed; it will always exist with that base, supplementing that base disclosure. Additional information in the form of XBRL calculations enhances the relationships between information within a set of reported information and providing additional clues. Certain base relationships between sets of information further enhance the ability to predict the nature of an information set. For example, there are relationships between the balance sheet, income statement, statement of changes in equity, and cash flow statement which will always exist and can be leveraged. This financial integrity type information can further enhance the ability to predict the nature of a set of information. Prototypes for creation and analysis are the same The prototypes or undisputed examples for creation of SEC XBRL filings are the same as the undisputed examples used for analysis of SEC XBRL filings. These prototypes can be hard to see within the US GAAP Taxonomy because that taxonomy tends to be inconsistent, not uniform. However, this reorganized version of the US GAAP Taxonomy helps one better see the prototypes or undisputed examples within the US GAAP taxonomy: http://www.xbrlsite.com/us-gaap-2011/reorganize/viewer.html It is not the case that there is only one undisputed example, nor does their need to be. For example, there are many different types of balance sheets: classified, unclassified, deposit based operations, insurance based operations, securities based operations, and others for specific industries and financial reporting needs. However, it is not the case that there are an infinite number of balance sheets. Financial information is not random or infinite in nature. 3
Specific undisputed examples can be created and even cross referenced with additional information. Another way of saying this is that there is no need to have only one undisputed example for any piece of a financial report. Further, this idea applies to each piece of a financial report and to the full set of pieces which an SEC XBRL filer might create. This screen shot is a fragment of a report available (rptinformationmodels.pdf) which shows a preliminary set of examples of the prototypes which could be created from the commercial and industrial companies entry point of the 2011 US GAAP Taxonomy. There are a total of 1055 such prototypes in that preliminary list. This is an example to provide an example of the granularity: 4
Applying prototype theory to SEC XBRL filings So, how can these ideas be leveraged with SEC XBRL filings? These appear to be the steps: 1. Determine the level at which the US GAAP Taxonomy will be prototyped. The report above is only a preliminary possible starting point. The precise point needs to be determined. 2. Create a set of agreed upon, rock solid undisputed examples from that list. These will be created by reorganizing the 2011 US GAAP Taxonomy, adding XBRL Formulas as additional meta data to enforce numeric relations, adding metadata to enforce consistency or utilize the XBRL US GAAP consistency suite. This will be in the form of either an XBRL taxonomy or some internal proprietary format and will serve as the master prototype set. This will be maintained in the future at the individual prototype level. We would suspect that there would be between 2000 and 5000 such prototypes initially if the entire US GAAP taxonomy were modeled for all entry points. 3. All SEC XBRL filings created would follow the prototypes created. Policies and procedures would enforce this. This would ensure financial integrity within each prototype and where one prototype relates to other prototypes. 4. Work with the FASB, the SEC, and XBRL US to help others see where financial integrity does not exist within the US GAAP taxonomy and within SEC XBRL filings so that others can move to a more sound base information model. There are two areas where base consistency and uniformity must exist to be successful: a. Consistent use of [Axis] with and across [Table]s b. Base financial integrity 5. Apply these same practices, but with different prototypes or undisputed examples, to other implementations of XBRL, leveraging what was learned from SEC XBRL filings within all products, professional services delivered, software constructed, etc. More information: http://en.wikipedia.org/wiki/concept_learning Concept Learning: Concept learning, also known as category learning and concept attainment, is largely based on the works of the cognitive psychologist Jerome Bruner. Bruner, Goodnow, & Austin (1967) defined concept attainment (or concept learning) as "the search for and listing of attributes that can be used to distinguish exemplars from non exemplars of various categories." More simply put, concepts are the mental 5
categories that help us classify objects, events, or ideas and each object, event, or idea has a set of common relevant features. Thus, concept learning is a strategy which requires a learner to compare and contrast groups or categories that contain concept-relevant features with groups or categories that do not contain concept-relevant features. Concept learning also refers to a learning task in which a human or machine learner is trained to classify objects by being shown a set of example objects along with their class labels. The learner will simplify what has been observed in an example. This simplified version of what has been learned will then be applied to future examples. Concept learning ranges in simplicity and complexity because learning takes place over many areas. When a concept is more difficult, it will be less likely that the learner will be able to simplify, and therefore they will be less likely to learn. Colloquially, task is known as learning from examples. Most theories of concept learning are based on the storage of exemplars and avoid summarization or overt abstraction of any kind. Prototype Theory of Concept Learning The prototype view on concept learning holds that people abstract out the central tendency (or prototype) of the experienced examples, and use this as a basis for their categorization decisions. The prototype view on concept learning holds that people categorize based on one or more central examples of a given category followed by a penumbra of decreasingly typical examples. This implies that people do not categorize based on a list of things that all correspond to a definition; rather, a hierarchical inventory based on semantic similarity to the central example(s). Exemplar Theories of Concept Learning Exemplar theory is the storage of specific instances (exemplars), with new objects evaluated only with respect to how closely they resemble specific known members (and nonmembers) of the category. This theory hypothesizes that learners store examples verbatim. This theory views concept learning as highly simplistic. Only individual properties are represented. These individual properties are not abstract and they do not create rules. An example of what Exemplar theory would look at is, water is wet; it simply knows that some (or one, or all) stored examples of water have the property wet. Exemplar based theories have become more empirically popular over the years with some evidence suggesting that human learners use exemplar based strategies only in early learning, forming prototypes and generalizations later in life. An important result of exemplar models in psychological literature has been a de-emphasis of complexity in concept learning. Some of the best known exemplar theory of concept learning is the Generalized Context Model (GCM). Multiple-Prototype Theories of Concept Learning 6
More recently, cognitive psychologists have begun to explore the idea that the prototype and exemplar models form two extremes. It has been suggested that people are able to form a multiple prototype representation, besides the two extreme representations. For example, consider the category spoon. There are two distinct subgroups or conceptual clusters: spoons tend to be either large and wooden or small and made of steel. The prototypical spoon would then be a medium-size object made of a mixture of steel and wood, which is clearly an unrealistic proposal. A more natural representation of the category spoon would instead consist of multiple (at least two) prototypes, one for each cluster. A number of different proposals have been made in this regard (Anderson, 1991; Griffiths, Canini, Sanborn & Navarro, 2007; Love, Medin & Gureckis, 2004; Vanpaemel & Storms, 2008). These models can be regarded as providing a compromise between exemplar and prototype models. http://courses.umass.edu/psy315/prototype.html Exemplars Rather than relying on a single prototype, we can also represent a category by storing many or all known exemplars of the category. When a new item is encountered, it is compared against all the members. Of course, this increases the memory requirement. Exemplar theories of concepts can also explain graded membership. The more exemplars that a stimulus matches, the better it fits into a category. The experimental results that support prototypes can generally be explained by exemplar theories. One Advantage of Exemplar Models over Prototype Models Exemplar models can preserve information about the correlation of different features within a category. Medin et. al. (1982) showed that subjects do use information about correlated features. Their subjects first studied a set of examples to learn a category. Then they had to decide whether new instances belonged in the category or not. After learning the category of patients with burlosis, subjects receive pairs of new patients and must decide which is more likely to have burlosis. 7
Both patients have the same number of symptoms, but in the first patient, the last two symptoms are correlated, while in the second patient they are not. None of the earlier patients had a pattern like the second patient. Subjects were more likely to pick the first patient. See description of Medin et. al. in textbook for more details. Some evidence suggests that prototypes are more likely to be used than exemplars after long experience with a concept. 8