The eXtensible Business Reporting Language (XBRL) has standardised nancial reporting and provide a machine-interpretable format that makes nancial and business reports easier to access and consume. Leveraging XBRL with Open Linked Data for purposes such as multi-dimensional regulatory querying and investigation requires XBRL formalisation as RDF. This paper investigates the use of o -the-shelf Semantic Web technologies to formalise accounting regulations speci ed in XBRL jurisdictional taxonomies. Speci cally the use of the SPARQL Inferencing Notation (SPIN) with RDF to represent these accounting regulations as rule constraints, not catered for in the RDF abstract model is investigated. We move beyond previous RDF to XBRL transformations and investigate how SPIN enhanced formalisation enables inferencing of nancial statement facts associated with nancial reporting concepts and sophisticated consistency checks, which evaluate the correctness of reported nancial data with respect to the calculation requirements imposed by accounting regulation. The approach illustrated through two use cases demonstrates the use of SPIN to meet central requirements for nancial data and regulatory modelling.
Despite the proliferation of nancial information available from sources such as company websites, institutions and regulating bodies there remains a lack of transparency with regard to nancial information. Two areas that can contribute to enhancing transparency are the adoption of a standard data representation formalism and greater levels of interoperability with and between di erent nancial sources.
Multiple heterogeneous formats (e.g. HTML, PDF, CSV), ensure that data
usage and interpretation typically has a dependency on manual intervention
with a knock on e ect for accurate and timely analysis of for example, nancial
reports [
XBRL, an XML-based format de nes nancial concepts and their relations based on jurisdictional Generally Accepted Accounting Practices (GAAP'S). Relations include calculation rules for monetary concepts { for example, the value for the nancial element Assets is calculated from the sum of Current assets and Non-current assets { in addition to other more complex business rules expressed through XBRL formula.
XBRL o ers automated processing of nancial reports and increased
interoperability between reporting instances. Even though XBRL provides a common
interoperable format its document-centric nature has been identi ed as an
inhibitor to integration of nancial information from multiple sources [
In this paper, we present an approach to implementing XBRL semantics using o -the-shelf Semantic Web technologies and speci cally SPIN3, ) to semantically model regulatory requirements mentioned. The resulting representation can be used to infer values for monetary concepts, and consistency check reported values without the need for customised XBRL software. The SPIN vocabulary, developed for business rule representation, can capture the intended semantics of accounting regulations in a transparent and intuitive way. SPIN adoption is supported by tools such as TopBraid Composer4), in progress standardisation e orts and an open-source Java API 5. 1 XBRL V2.1 Taxonomy Speci cation Recommendations http://www.xbrl.org/
SpecRecommendations/. 2 See http://xbrl.sec.gov/. 3 Speci cation at http://www.spinrdf.org/ 4 Seehttp://topquadrant.com/products/TB_Suite.html 5 See http://www.spinrdf.org/faq.html for more details.
After a brief introduction to XBRL, nancial reporting using XBRL
taxonomies and SPIN, we position our work with respect to recent e orts from
Garc a and Gil [
This section introduce the main aspects of XBRL and nancial reporting as relevant to this work (Section 2.1), as well as the basic features of SPIN (Section 2.3). Section 2.2 discusses related work, rstly with respect to transforming XBRL data to its RDF equivalent, and secondly regarding the use of SPARQL for business-related modelling issues. 2.1
XBRL is an XML-based formalism which aims to replace dependency on
proprietary formats usage in nancial reports preparation and compilation [
As shown in (1), values in an instance document refer to concepts de ned in XBRL taxonomies, which specify, for example, that Cash and cash equivalents is a monetary concept with balance debit, which is measured at a point in time { as opposed to over a duration of time. In addition to adopting internationally standardised taxonomies like the International Financial Reporting Standards (IFRS), companies can also provide their own taxonomy extensions, in instances where they need to report a value for a concept which is not covered by the standard. For example, the concept Software revenue is not found in the 2009 IFRS taxonomy, but is provided as a custom taxonomy extension by SAP, with the facts in the instance document. Each concept in a taxonomy is linked to a set of XLink linkbases8 (called resource and relation networks in the gure). These specify labels for the concepts, as well as other information such as, how the values of the concepts should be displayed in di erent statement types . For example, the International Financal Accounting Standard 9 \Statement of nancial position, current/non-current" speci es that the concept \Assets" should be displayed above \Non-current assets" and \Current assets" in a consolidated ling, whereas the nancial instrument \Statement of nancial position, order of liquidity" places \Assets" above \Property, plant and equipment", \Investment property". To express such relationships, XBRL uses XML Linking Language (XLink:10) arcs and extended links, which can be used to group any number of 6 Note: XBRL has also been used by the Global Reporting Initiative https://www.
globalreporting.org/ to report on sustainability issues. 7 http://xbrl.squarespace.com 8 e.g. the U.S. 2009 GAAP taxonomy contains 450 linkbases. 9 http://www.ifrs.org/XBRL/IFRS+Taxonomy/IFRS+Taxonomy+2011/IFRS+
Taxonomy+2011.htm. 10 http://www.w3.org/TR/xlink/. arcs and link them to other resources For the case just described, the XBRL presentation linkbase speci es how the concepts are linked using parent-child arcs, and that a set of presentation arcs are associated with a particular accounting standard using an extended link role.
For the purpose of our investigations we focus on the XBRL calculation linkbases, which de nes how concept values are calculated as de ned by speci c accounting standards and general business rules (see sections 3.2 and 6. Example (2) shows the XBRL representation of a calculation arc taken from the calculation linkbase of the IFRS taxonomy. The XBRL formula linkbase is outside the research scope and left for future work considerations. (2) <loc xlink:type="locator" xlink:label="ifrs_CashAndCashEquivalents" xlink:href="http://xbrl.iasb.org/taxonomy/2009-04-01/
ifrs-cor_2009-04-01.xsd#ifrs_CashAndCashEquivalents" /> <calculationArc xlink:type="arc" xlink:arcrole="http://www.xbrl.org/2003/arcrole/summation-item" xlink:from="ifrs_CurrentAssets" xlink:to="ifrs_CashAndCashEquivalents" order="1" weight="1" /> The arc states that the concept \CurrentAssets" is linked to \CashAndCashEquivalents" through a \summation-item" relation. The arc weight of 1 indicates that the concept values are added and a weight of -1 that the values are subtracted. Currently XBRL calculation links only allow for the summation of items. Clearly Semantic Web formalisms o er a far more compact and intuitive representation for expressing such statements. To that end we next discuss related approaches that convert XBRL data to its RDF equivalent. 2.2
Transforming XBRL to Semantic Web formalisms. Bao et al. [
In addition to this, their strategy for representing linkbase arcs is based on
the assumption that the intended interpretation of arcs holds between instances
of the respective concepts linked by a particular arc. The XBRL Speci cation
does not note this as being the intended interpretation, but instead states that
arcs relate \one XBRL concept to one other XBRL concept"11. While the
assumption seem reasonable for concrete-numeric-concepts, abstract concepts by
de nition do not have instances. Irrespective they are still related by means of
parent-child arcs, and it is not apparent how [
Bao et al. are not \mechanical" in their preservation of the structural
properties of XBRL while other approaches by Declerck and Krieger [
Using SPARQL in the context of business information. Furber and Hepp [
According to its developers, the SPARQL Inferencing Notation (SPIN) \is the de-facto industry standard to represent SPARQL rules and constraints on Semantic Web models"12. SPIN has been developed out of the necessity to perform calculations on property values, a task which is largely unsupported in current Semantic Web formalisms, as well as the need for constraints checking with closed-world semantics (see Section 5).
SPIN provides the syntax to attach SPARQL queries to resources in an RDFcompliant way using RDF properties spin:rule, spin:constraint, and superproperty spin:query. The spin:rule property accepts SPARQL CONSTRUCT queries as value and can be used to infer new triples on the basis of the statements in the query's WHERE clause. A basic example is provided in (3), and the corresponding SPIN representation in Turtle syntax in (4). (3) CONSTRUCT { ?this a ?c2 . }
WHERE { ?c1 rdfs:subClassOf ?c2 .
?this a ?c1 . } (4) [ a sp:Construct ; sp:templates ([ sp:subject spin:_this ; sp:predicate rdf:type ; sp:object _:b1 ]) sp:where ([ sp:subject _:b3 ; sp:predicate rdfs:subClassOf ; sp:object _:b1 ] [ sp:subject spin:_this ;
sp:predicate rdf:type ; sp:object _:b3 ]) ] 12 See http://www.spinrdf.org.
The example, based on that available from TopBraid's SPIN website13, formalises the semantics of rdfs:subClassOf and illustrated variable use. While variables in a SPARQL query are generally mapped to blank nodes, in SPIN RDF notation, the variable ?this refers to the resource spin: this. This variable, like the corresponding keyword in object-oriented programming languages, refers to an instance of the class to which the respective rule has been attached. As a result, if the rule in Example (3) was attached to owl:Thing, it would be applied to every instance of owl:Thing satisfying the statements in the WHERE clause.
The spin:constraint property can be used to model consistency constraints, using the SPARQL ASK queries. Where an ASK query evaluates to true, the respective instance is indicated as violating the constraint. Finally, the general property spin:query can be used to generally attach SPARQL queries to RDF resources, i.e. also SELECT queries. As will be shown in Section 3, we make use primarily of CONSTRUCT and ASK queries in order to capture the intended semantics of accounting regulations.
In addition to standard SPARQL operators like UNION, OPTIONAL and FILTER, SPIN supports SPARQL extensions such as the ARQ keyword LET14, which allows for value assignment to variables, as well as the possibility to de ne custom functions.
With SPIN having recently started standardisation activities as a W3C member submission and the fact that SPARQL is already the query language of choice in numerous Semantic Web applications { there is a solid basis for its wide-spread adoption by the Semantic Web community. 3
This section discusses the conversion of XBRL to RDF, with focus on the SPINbased representation of accounting regulations. After a brief introduction to the general underlying ideas in Section 3.1, we discuss the representation of calculation rules (3.2) and consistency constraints used (3.3) to transform the accounting regulations from XBRL to RDF. 3.1
Our approach adheres to what Bao et al. [
Each of the graph patterns in the query represents one calculation arc. References to URIs for the context and units, ensure that the values of relevant instances are only taken into account. This excludes cases where a particular value refers to di erent entities or di erent segments of the same entity, as well as cases in which values are reported for di erent time periods. This is a normal occurrence as nancial statements generally contain gures for both the current and preceding reporting periods. Finally, the LET clause speci es how the values of the individual concept instances should be combined. For accounting rules, this is limited to summation and subtraction. XBRL provides a single arc role summation-item for this purpose, and uses the value of the weight attribute { either 1 or -1 (example (2) above) { to indicate whether the value of a particular concept should be added or subtracted. For further more complex calculations possible using SPIN we refer the reader to the SPIN vocabulary speci cation 15.
The example illustrates how rules can make use of previously calculated values to calculate further values. This allows value calculation for composite monetary concepts (i.e. those whose values are calculated on the basis of the values of other concepts) by specifying values for atomic concepts and then applying the rules iteratively (see Section 4.1 for more details).
Moreover, it should be noted that in our RDF representation, rules are not modelled as blank nodes, but instead carry a URI. This has the bene t of enabling other instances to dereference the rules, allowing a particular calculation rule be reused across di erent nancial reports of the same accounting standard. It also further allows attachment of additional properties to a rule and a reference to the type of nancial statement to which the rule applies. Therefore, in 15 http://www.spinrdf.org. addition to the actual rule, the instance representing the rule in (5) is the subject of a triple relating it to the URI representing SAP's Consolidated Statement of Financial Position by means of xbrlrdf:roleRef.
As mentioned, rules can be executed iteratively, making use of previously inferred values. In order to make sure that a particular calculation rule with atomic concepts (i.e. those concepts which lack regulatory calculation rules attached) can also be applied, we add the default calculation shown in (6) to atomic monetary concepts. This rule then assigns the reported value of the respective instance to xbrlrdf:calculatedValue. (6) CONSTRUCT { ?this xbrlrdf:calculatedValue ?value . }
WHERE { ?this xbrlrdf:value ?value } . 3.3
On the basis of the SPIN rules presented above, each monetary concept which participates in some calculation and has reported values in the respective context is assigned a calculated value. The next step in modelling the accounting regulation is to specify that calculated values need to match reported values. As SPIN rules and constraints are applied to all instances of the class to which they have been attached, as well as to the instances of its subclasses, this can be achieved by attaching a single SPIN constraint to the top monetary concept: (7) ASK WHERE { ?this xbrlrdf:value ?value ;
xbrlrdf:calculatedValue ?cvalue .
FILTER (?value != ?cvalue) }
Additionally, SPIN constraints can be used to formalise more general constraints imposed by the XBRL speci cation. Below, we illustrate this using a constraint (restriction) which states that if two concepts have the same balance type (i.e. credit or debit), they can only be added to one another, not subtracted. In other words, the value of the XBRL weight attribute, which is preserved in our structural representation of XBRL, has to be positive: (8) ASK WHERE { ?this xlink:from ?from ; xlink:to ?to ;
xbrlrdf:weight ?weight . ?from a ?balance . ?to a ?balance . ?balance rdfs:subClassOf xbrlrdf:BalanceType .
FILTER (?weight < 0) } 4
The SPIN rules and constraints given in Section 3 above can be used to infer values for instances of the respective reporting concepts, as well as check their consistency. We next illustrate how these representations integrate with TopBraid Composer, taking as example the SAP 2009 annual report reported against its custom extension of the IFRS 2009 taxonomy. 4.1
In order to test whether the rules explained in Section 3 actually behave as desired, we have rst generated a modi ed version of the report such that it contained reported values for atomic monetary concepts only. All composite monetary concepts were assigned their values through iterative application of the SPIN calculation rules. This allowed evaluation as to whether the available information triggered the application of all rules necessary to calculate the missing information and whether the calculated gures corresponded to those reported in the original ling. Table 1 summarises the results, with the gures from the original report recorded in parentheses.
Concepts in IFRS 2009 taxonomy and SAP extension Regulatory calculation rules Reported monetary values Monetary concepts with reported values Inferred monetary values Monetary concepts with inferred values Monetary values inferred by default rule Monetary values inferred by regulatory rules Regulatory rules applied Total number of monetary values Total number of correct monetary values absolute relative
Tuples 3 and 4 of the table detail that the modi ed report contains 351 reported values for 129 monetary concepts, compared to 482 and 171, respectively, from the original report. After applying the calculation rules, values are inferred for 97.66% of these 171 concepts, indicating that the modi ed report contains 458 values, as opposed to 482 original report values. 25.11% of the 458 inferred values are due to regulatory rules, outlining that the remaining 343 values have been inferred for atomic concepts by means of the default rule. Over all 8.54% of all the regulatory rules available in the IFRS 2009 taxonomy and the SAP 2009 extension have been applied.
The gures illustrate that for 4 of the monetary concepts no value could be
inferred. Analysis revealed that this is due to values for ifrs:BasicEarningsLossPerShare
and ifrs:DilutedEarningsLossPerShare, being missing from the XBRL
report instance, despite being part of a composite concept. As a result, the
corresponding rules could not be applied (see Section 5 for a discussion regarding the
optionality of calculation arcs).
The previous sections detail how the SPARQL CONSTRUCT rules and ASK
queries capture the semantics of the XBRL data in an intuitive and transparent
way. However the XBRL does allow for calculation arcs and if so to have an
assumed value of zero. This usage of default values, cannot be naturally handled
with monotonic logics such as OWL and would normally require the use of a
formalism such as Reiter's default logic [
WHERE { ?x0 a ifrs:CurrentTaxAssets ;
xbrlrdf:calculatedValue ?cv0 .
OPTIONAL { ?x1 a ifrs:CashAndCashEquivalents ;
xbrlrdf:calculatedValue ?cv1 . }
LET(?cvalue := 1.0 * ?cv0 + 1.0 * ?cv1) }
If it is assumed that that rules can be applied in any order, this rule would be applied before the rule that calculates ?cv1, and a di erent value produced. We therefore take the position that every value should be de ned or explicitly stated as not unde ned. While OWL has no speci c vocabulary to state that a value has not been de ned we can achieve the same result using an OWL class axiom.
For example, in order to state that SAP did not report a value for ifrs:BasicEarningsLossPerShareFromDiscontinuedOperations in units iso4217:EUR year ending December 200916, the following would need to be asserted. (10) ifrs:BasicEarningsLossPerShareFromDiscontinuedOperations v
:((9xbrli:contextRef:fFYp0YTDg) u (9xbrli:unitRef:fiso4217:EURg))
With the non-existence of such an instance explicitly asserted it can be combined with the SPIN rules by providing a default rule that speci es the value of the property if it is known not to exist as follows: (11) OPTIONAL { ?x4 a ifrs:CashAndCashEquivalents; xbrli:contextRef ?context; xbrli:unitRef ?unit; xbrlrdf:calculatedValue ?cv4 } .
OPTIONAL { NOT EXISTS { ?x4 a ifrs:CashAndCashEquivalents; xbrli:contextRef ?context; xbrli:unitRef ?unit; } .
LET (?cv4 := 0) . }
Here the assumption is made that NOT EXISTS predicate evaluates to true where it is provable that the dataset does not contain the predicate. Interesting future work could look to determine whether a default logic rule language such as RIF-SILK17 could be used. 16 In SAP's XBRL instance document, the period context represented by FYp0YTD. 17 See http://silk.semwebcentral.org/RIF-SILK.html.
The paper outlines SPARQL's capability to meet complex query requirements, central to modelling nancial data, and speci cally their accounting regulations on the Semantic Web. The approach transforms nancial reports represented in XBRL to RDF, and uses the RDF-compatible SPARQL Inferencing Notation to capture the regulatory rules expressed by the XBRL calculation linkbase. The resulting representation was evaluated against XBRL nancial data, both with respect to inferring values for instances of monetary concepts and checking their consistency. Additionally the use of the representation to formalise additional constraints to address the well-formedness and high quality of the data was discussed.
The approach taken can be extended to cater for the more complex mathematical operations of the XBRL formula speci cation18. For example the formula speci cation de nes that value calculations apply to instances that refer to identical contexts, and more generally to concept instances which are p(arent)-equal, c(ontext)-equal and u(nit)-equal. p-equality and u-equality have been previously shown through rule attachment to the composite class and including the reference to the unit in the rule. Alternatively c-equality could be inferred beforehand, by specifying that two contexts which share the same entity and period are linked by owl:sameAs. When applying the rules iteratively to a repository that is OWL-aware, the rules shown above can be applied as is.
Arguments for nancial information integration include the ability to
conduct nancial metrics comparison [
We have demonstrated SPIN's viability but others rule languages such as
the Semantic Web Rule Language 19 suggested by [
The work presented in this paper has been funded by the Multilingual Ontologies for Networked Knowledge (MONNET) project under EU Grant No. 248458, and Science Foundation Ireland under Grant No. SFI/08/CE/I1380 (Lion-2).