=Paper=
{{Paper
|id=Vol-1517/JOWO-15_WoMO_paper_3
|storemode=property
|title=OntoBacen: A Modular Ontology for Risk Management in the Brazilian Financial System
|pdfUrl=https://ceur-ws.org/Vol-1517/JOWO-15_WoMO_paper_3.pdf
|volume=Vol-1517
|dblpUrl=https://dblp.org/rec/conf/ijcai/PolizelCS15
}}
==OntoBacen: A Modular Ontology for Risk Management in the Brazilian Financial System==
https://ceur-ws.org/Vol-1517/JOWO-15_WoMO_paper_3.pdf
OntoBacen: A Modular Ontology for Risk Management
in the Brazilian Financial System
Filipe Polizel Sara Casare Jaime Sichman
Instituto de Matemática e Estatı́stica IBM do Brasil Escola Politécnica
Universidade de São Paulo sjcasare@uol.com.br Universidade de São Paulo
fpolizel@ime.usp.br jaime.sichman@poli.usp.br
Abstract Financial Industry Risk Management
This paper presents a first semantic formalization of The main good practices for the governance of the global fi-
the Brazilian financial system risk management poli- nancial system were established and formalized by the Basel
cies, called OntoBacen, that is based on a modularized Committee on Banking Supervision (BCBS), forming part
approach. We show some partial results generated by of the Bank for International Settlements (BIS), that in post-
a knowledge-based system that uses an ontology con- crisis periods identified the need for a robust data manage-
structed to address some domain questions. ment framework. This must ensure that banks have the ca-
pacity to aggregate risk exposure data in an integrated man-
Episodes like global crisis undermine people’s confidence ner, reaching all the corporation levels, in addition to stan-
in the financial system, but also provide lessons for the fu- dardized risk reporting practices (BIS 2013a), confering the
ture. The 2007-2008 meltdown resulted in a significative degrees of assertiveness and timeliness required by institu-
advancement of governance policies followed by financial tional leaders for decision-making in times of stress.
institutions worldwide, with some of them treating classic Once established the core principles for banking supervi-
data management problems, as integrity and completeness. sion (BIS 1997; BIS 2012), central banks around the globe
To enable the adoption of an integrated and robust global fi- have taken them as basis for the establishment of their own
nancial system, IT companies and financial institutions are regulatory norms. Moreover, they also considered the sin-
joining efforts for the creation and adoption of a technolog- gularities of their domestic financial systems, what partially
ical framework to better meet the industry needs. mitigated the level of heterogeneity of the global financial
The main goal of this work is to explore alternative ap- system risk management domain. However, this was not suf-
proaches for the conceptualization and definition of busi- ficient to achieve the goal of representing these domain con-
ness rules present in governance policies of the Brazilian cepts and their relationships in an integrated manner world-
financial system, more specifically those related to risk man- wide, which demands the use of highly integrable and log-
agement. To this end, it proposes an ontology, called Onto- ically grounded tools, such as the use of formal ontologies
Bacen, that expresses the concepts (and their relationships) (Guarino 1995) in the Semantic Web (Berners-Lee, Handler,
of this domain, and by using inference algorithms, can ver- and Lassila 2001).
ify the compliance of hypothetical financial institutions with
The policies and guidelines to be followed by the Brazil-
those policies.
ian financial institutions are created and maintained by the
For such a wide and complex domain, modularity must
local monetary authorities, but mainly by the executive au-
play a central role in the design of the proposed solution,
thority of the national financial system, the Brazil Central
to ensure that it results in an coherent, understandable and
Bank, also known as BACEN. In order to align their gover-
scalable knowledge-based system.
nance policies with the Basel principles, the Brazilian mon-
In the following section, the risk management setting of
etary authorities created a series of norms, known as pru-
the Brazilian financial system is briefly introduced. We then
dential regulation (BACEN 2014), to be followed by local
present the main initiatives involving ontologies for the fi-
banks and financial institutions. The main goal of the pru-
nancial industry, followed by a description of OntoBacen
dential regulation is to consolidate a national system for risk
by means of its properties and requirements. In the sequence,
management and capital adequacy.
we show the modularization approach adopted by the pro-
These regulations are arranged to take into account the
posed solution, followed by a section that details how On-
main types of risk, described as follows:
toBacen has been developed and how it’s meant to be used.
Some test cases illustrate the use of the proposed ontol- Credit Risk Associated with the risk of default, the failure
ogy, followed by current conclusions and future develop- to comply with obligations and responsibilities.
ment steps.
Market Risk Related to the volatility of rates or prices over
the time, such as currency exchange and interest rates, or
prices of securities and commodities.
Operational Risk Associated with the probability of loss These initiatives are being conducted with the support of
resulting from internal processes failures or deficiencies, the Governance, Risk and Compliance Technology Centre3
including legal risks, such as damages to third parties aris- (GRCTC), also responsible for the development of the Fi-
ing from its activities, or violation of rules established in nancial Industry Regulatory Ontology (FIRO) and the Finan-
their jurisdiction. cial Governance, Risk and Compliance Ontology (FIGO).
This business division of the domain suggests that mod-
ularity should be taken into account (and exploited to the OntoBacen Proposal
fullest) when dealing with risk management. The final and main goal of this work is to create an ontol-
ogy to represent the Brazilian financial system, by means
Ontologies for Financial Industry of its risk management concepts, and to be implemented in
In recent years, with the greater control over the financial the Web Ontology Language (OWL). In this sense, it is re-
systems by regulatory agencies, the need for information lated to some of the ontologies mentioned in the previous
systems interoperability and data integration has increased, section, while considering BACEN’s governance policies, as
which strengthened initiatives related to finance on the Se- published in its norms.
mantic Web; these initiatives, allied with the ontologies’ se- A Brazilian financial jargon says that BACEN’s regula-
mantic formalism, have gained their place and importance in tions are the “tropicalization of Basel”. To exemplify this
this specific industry. statement, when comparing BACEN and Basel standardized
As an example, the Suggested Upper Merged Ontology approaches to evaluate market risks, one can conclude that
(Niles and Pease 2001), also known as SUMO, has included the BACEN approach has an additional component of risk
its own finance domain ontology years ago, dealing with for fixed interest rates denominated in the local currency
concepts related primarily to financial services, typical of (real), making it a conservative adaptation of the Basel ap-
commercial banks, such as bank accounts, payments, loans, proach (BIS 2013b).
etc. A more recent work (in progress at the time of writ- The specification of OntoBacen, by means of its require-
ing), is the Financial Report Ontology1 (FRO), which pro- ments, is given by a set of competence questions (Grüninger
vides formal and structured meta-information about finan- and Fox 1995), as shown in the following:
cial reports, such as balance-sheets; it is primarily based
CQ1: What is the capital structure, by means of its compo-
on a well-known XML schema for this application domain,
nents, of a financial institution that belongs to the Brazil-
XBRL (Engel et al. 2013), that stands for eXtensible Busi-
ian financial system?
ness Reporting Language.
Other relevant initiative is the Financial Industry Busi- CQ2: What are the maximum and minimum constraints for
ness Ontology2 (FIBO), a series of standards being devel- the capital components of a Brazilian financial institution?
oped by the Enterprise Data Management Council (EDMC) CQ3: What are the cash amounts that represent each capital
and published following the technical governance process of component of a Brazilian financial institution?
the Object Management Group (OMG). FIBO currently pro-
vides a framework of conceptual definitions concerning the CQ4: Do the capital components of a Brazilian financial
wide spectrum of financial applications, currently available institution respect its constraints?
for use in two modules. The definition of capital component used in this paper is
The first module, FIBO foundations, defines high-level that of a grouping of assets or liabilities, possibly weighted
financial concepts such as currency or contracts, and even by some factor and represented by an amount of money.
non-financial concepts such as autonomous agent or coun- Aditionally, for the development of the proposed ontol-
try, that are need for the definition of more specific finan- ogy, a methodology partially based on the GRCTC method-
cial concepts. The second module, FIBO Business Entities, ology (adopted by FIRO and FIGO) will be used, as shown
defines concepts such as legal persons and corporations, en- in the Construction and Usage section.
tities that could incur legal obligations such as establishing
business contracts with other entities. OntoBacen Modularization
There are also ontological initiatives concerning financial
regulations, where lies the scope of this work. Abi-Lahoud The BACEN’s prudential regulation is the starting point for
et al. (2013) developed an ontology concerning the com- the definition of OntoBacen. It was analyzed in a top-down
pliance with American anti-money-laundering regulations; approach, beginning with the most high-level view, repre-
later, Abi-Lahoud, OBrien, and Butler (2013) presented an sented by the notions related to the methodology for cal-
experimental discussion about the adopted approach, an it- culation of the Reference Capital, later reaching the lower-
erative process based on subject-matter expertise and on the level related concepts. In this sense, by the interpretation
use of structured natural language, more precisely based of a series of BACEN documents, the lower level concepts
on SBVR (OMG 2008), that stands for Semantics of Busi- were identified, introducing definitions related to several ap-
ness Vocabulary and Business Rules, a structured vocabu- proaches for the measurement of different types of risk.
lary founded in formal logic. After this domain analysis, some core modules were iden-
tified for the development process, as shown in Figure 1.
1
See: http://xbrl.squarespace.com/financial-report-ontology/.
2 3
See: http://www.omgwiki.org/OMG-FDTF/doku.php. See: http://www.grctc.com/platform-research/.
FIBO Foundations FIBO Business Entities was added to treat the risk related to interest rates of bank-
ing book assets, i.e. those acquired and meant to be held until
their maturity, because the risk exposure calculation method
Foundations
for these cases is different from the standard market risk for
interest rates approach.
Capital RWA An additional reason to adopt such modularization is that
Tier 1 Core Market
RWA Risk these subdomains are commonly subject of distinct business
areas in financial institutions (specially the big ones). In this
Tier 1 Complementary Credit Risk
sense, the compliance area is more willing to treat issues
Tier 2 Operational Risk addressed by the Capital ontologies, the market risk man-
Permanent Limit Banking Book Risk agement area to those of the Market Risk ontology, and so
forth, in such a way that those areas could consult specific
in development (blank)
prototype ready (filled) ontologies as semantics repositories or even use its inference
dependency relation Capital Requirements capabilities to address their own issues.
Finally, an additional ontology, Capital Requirements,
Figure 1: OntoBacen modules deals with the capital adequacy questions mentioned in the
proposal section. It is related to concepts defined in both
Primarily, the FIBO Foundations and FIBO Business En- Capital and RWA ontologies, and contains the definitions
tities ontologies were chosen to provide the highest level of of business rules necessary for addressing CQ2.
semantics, such as the has part property, a whole-part rela- Currently, the Foundations, Market Risk and Capital
tionship frequently used by OntoBacen, or even concepts Requirements ontologies are ready to be used, as detailed
common in finance, such as currency. in the sequence. Consequently, the capital adequacy compe-
Some additional concepts not present in FIBO were de- tence questions address only market risk exposure.
fined to provide mid-level semantics, constituting the mod-
ule Foundations, that is supposed to be used by other mod- Foundations Module
ules and ontologies of OntoBacen; it is composed by a set This module is composed of four foundational ontologies,
of ontologies described later in this section. described as follows:
The lower level ontologies are then separated in two major
disjoint groups: Capital and RWA. The first group addresses Common Relations Ontology that defines a set of data and
the definition of capital, that is represented in an account- object properties required by other ontologies, such as:
ing perspective by shareholder’s equity, savings and other 1. mathematical properties, to define product, minimum,
kinds of liabilities. The second group name is an acronym maximum, and other derivation relationships;
for Risk Weighted Assets, that also in an accounting per- 2. time constraints properties, such as initial and ending
spective consists of the values of assets weighted by distinct date (xsd:dateTime), minimum and maximum duration
types of risks according to the degree of risk exposure; these (xsd:duration), needed for time intervals definition;
two groups encompass the concepts that must be defined for
addressing CQ1. 3. the has reference date property, needed for the defini-
tion of financial system governance contexts;
To exemplify how RWA captures the notion of risk, one
could think, for instance, in a market risk perspective, if you 4. the has decimal value property, used to define factors;
have a given amount of resources applied in the stock mar- BACEN Factors Ontology for the definition of constant or
ket and the same amount of resources applied in treasury time dependant factors present in BACEN regulations.
bonds, it is expected that the weighting factor of the first Time dependancy is defined by fixed xsd:dateTime inter-
group is higher than the second one’s weighting factor, and vals or fixed xsd:duration intervals relative to a specific
consequently its RWA, because of the probability that the date;
company which some stocks was acquired enter bankruptcy
is higher than that of the government defaulting. Financial Institutions Ontology encompassing concepts
Each one of these groups were then divided in minor (and that define the Brazilian financial system agents, such as:
also disjoint) modules. Capital group takes into account 1. The financial institutions classification, specializing the
BACEN major definitions to do these separation, being de- concept for legal person from FIBO Business Entities.
composed in Tier 1 Core, Tier 1 Complementary and Tier Examples are commercial banks, investiment banks,
2 ontologies. The same group also encompasses a Perma- securities brokerages, etc.;
nent Limit ontology, that is used to reduce the amount of 2. The concept of monetary authority, that also is a legal
capital used for requirements checking, in cases that perma- person, and used to define BACEN;
nent assets exceed a certain limit.
3. The concepts for classification of institutions into com-
The RWA group considered the division for risk man-
pliant or not-compliant to specific regulatory norms;
agement mentioned in the first section, resulting in differ-
ent ontologies for Market Risk, Credit Risk and Opera- Contextualized Monetary Amount Ontology for the defi-
tional Risk. An additional ontology Banking Book Risk nition of a generic context concept, and also:
1. The contextualized monetary amount concept is a spe- RWAMINT RWAMPAD is derived from
Sm
cialization of FIBO Foundations monetary amount con- (int. model
approach)
considering
Sm factor
product of
factor
cept, something that has a currency and an amount, and RWAMINT
has part
is derived from is derived from
that additionally has a context (uses FIBO Foundations components
(omitted)
is derived from maximum of product of
has currency, has amount and has context properties); maximum of
relative
time
2. The financial system governance context concept, that market
RWAMINT
considering
RWAMPAD
(standard dependant
specializes context, and also involves a monetary au-
risk RWA approach) factor
Sm factor
component
thority governing and constraining the behavior of a fi- has part
nancial institution at a specific moment in time (uses
FIBO Foundations is governed by and constrains prop- OB Market Risk
OB Foundations
contex-
tualized market reported RWAMPAD
erties). generalization monetary
amount
risk RWA market
risk RWA
components
(omitted)
custom relationship
The contextualized monetary amount concept is used to
represent amounts of cash related someway to a financial
Figure 2: Market Risk Ontology
system governance context, thus it is the major definition of
OntoBacen for addressing CQ3.
the smallest (by comparing the has amount data property);
Market Risk Ontology in the other cases, where there is only the standardized ap-
proach, this latter is the one to be reported.
The main goal of this ontology is to provide a way to eval- The market risk ontology defines the concepts men-
uate the market risk exposure by using at least one of the tioned before as specializations of contextualized monetary
approaches available. Its semantics translates the business amount. A general market risk RWA concept is defined, so
rules needed to obtain the market risk RWA (risk weighted that it encompasses any market risk evaluation approach, be-
assets by market risk exposure). ing naturally specialized by the two approaches previously
Figure 2 shows its main concepts and their relationships mentioned, and by an additional concept for the approach
using a notation similar to VOWL (Lohmann et al. 2014), that was chosen to be reported to BACEN.
where concepts are represented as circles (filling colors indi-
The intermediate results used in the evaluation of the mar-
cate modules), relationships (properties) as plain-line arrows
ket risk RWA are conceptualized as specializations of an ad-
and generalization relationships as dotted-line arrows.
ditional concept market risk RWA component. Finally, the
The first important thing to note is that there are two Sm factor concept is classified as a specialization of rel-
possible appoaches to evaluate the market risk RWA, the ative time dependant factor, defined by Factors ontology
standardized approach (called RWAMPAD) and the inter- and whose semantics define that its value is dependant of
nal model approach (called RWAMINT). Both of these ap- the amount of time elapsed from a specific date (in this case,
proaches have more specific concepts that are omitted here the market risk model transition date).
for a matter of simplicity.
In both cases, financial institutions must verify their mar-
ket risk exposure by the standardized approach, because
Capital Requirements Ontology
even the internal model approach depends on the standard- There are different capital requirements specified by the BA-
ized. In addition, they are not required to have (or to use) CEN regulations, such as minimum requirements for tier
an internal model, but in the cases that both approaches are 1 capital and reference capital. The latter is the most gen-
implemented, one of them must be chosen. eral (and relevant) definition of capital, so that this require-
To opt for the internal approach, the institution must show ments rules was chosen as the first to be implemented by
to BACEN that the proposed model actually maps the mar- the Capital Requirements ontology. For the implementa-
ket risk efficiently and that it is consistent with the regula- tion of capital requirements rules related to more specific
tions and norms. When opting for this model, things get a bit concepts, such as tier 1 capital, some efforts will be needed
messy, because in this case, some conservative constraints in the construction of ontologies for the Capital module,
for the evaluation of the RWA must be considered. described in the beginning of this section, and working in
Firstly, a market risk model transition factor (defined as progress at the moment of writing.
Sm) must be considered. This factor assumes the value of The main idea of this ontology is to compare the risk
90% in the first year of the transition, and 80% after that. weighted assets with some liabilities composition, defined in
This weighting factor is then applied to the standardized ap- this case as reference capital, and that is mainly composed
proach, producing an intermediate result, the standardized by reserves and shareholder’s equity. To define the concept
approach considering Sm factor. From the results obtained of risk weighted assets, it is necessary to consider all types
by both the application of the internal model approach and of risk (such as the market risk mentioned before), aggregat-
the standardized approach considering Sm factor, the greater ing these multiple risk types as a general and single concept.
of these is defined as the internal model RWA. For that, all ontologies of the RWA module must be avail-
Independently of the values for the market risk RWA ob- able; as already mentioned before, at this moment the capital
tained by different approaches, institutions must choose one requirements ontology is considering only market risk as a
of them to report BACEN. OntoBacen’s current implemen- risk type.
tation chooses, in cases where the two models are evaluated, The minimum required reference capital can be deter-
Financial Institutions Ontology Capital Requirements Ontology
reference 4.Distribute terms
ref. cap. is derived from capital
time
interval
.txt (NL) and rules in
ref. cap. product of Art.I
limit minimum requirement dependant
limit non-
compliant required factor factor 2.Self-contained, modules by
compliant
institution
institution reference
reported
Input: complete subject matter
capital
market Bacen norms sentences
is reduced risk RWA .sbvr
.sbvr
depends on by
reported
RWAs
capital
ontologies
.sbvr
depends on positive is derived from
negative (other types
margin product of
margin (dummy)
has part of risk) 1.Follow 3.Extract terms and
(dummy)
contex-
risk
has part defined permanent references rules to SBVR
margin
tualized
weighted somehow by limit .owl
.owl
over requi-
monetary
assets (dummy) ontology ref.
X .sbvr (SNL) .owl
amount
red ref. (RWA) X
capital 5.Convert SBVR
OB Foundations
has part
reference
defined
somehow by 6.Manipulate for each module to OWL
OB Capital Requirements has capital (dummy) namespaces, (steps 6-9) (formal logic)
part
OB RWA Ontologies
OB Capital Ontologies ref. capital excess of imports,
generalization for RWA
is reduced by
permanent cardinalities, 7.Generate
custom relationship
comparison assets
labels, etc. Input:
individuals Operational
Enrich semantics with custom
by adding capital limit
Figure 3: Capital Requirements Ontology application report file
SWRL rules. (Jena +
.owl SPARQL) .xml
mined from the risk weighted assets, as depicted by the de- +SWRL
pendency relationship between these concepts in Figure 3.
.rdf
This minimum requirement amount is given by applying a TBox
9.Get answers
numerical factor (reference capital requirement factor) to with new
knowledge
the RWA, whose value decreases yearly (11% before 2016, ABox
9.875% at 2016, 9.25% at 2017, 8.625% at 2018 and 8% af- 8.Infer new axioms by
using an OWL and
ter 2018). This fixed time interval dependancy is represented SWRL compatible new ?
as an specialization of the concept time interval dependant inference engine
axioms
factor of the Factors ontology.
Before this comparison between the reference capital and Figure 4: OntoBacen construction and usage steps
its minimum requirement could be done, the excess of per-
manent assets must be discounted from the capital, resulting
in an additional concept for comparison between capital and In the first step, it is needed to follow reference chains
RWA: reference capital for rwa comparation. This excess is present in such norms, to construct, in a second step, seman-
determined by additional business rules, omitted here, be- tically complete and self-contained sentences in natural lan-
cause they are in the scope of the Permanent Limit ontol- guage (Portuguese), by cutting and pasting portions of the
ogy, not yet constructed. original text.
Another important concept definition is that of margin After that, in the third step, the self-contained sentences
over required reference capital, which is given by the differ- are interpreted in order to achieve a first level of formality,
ence between the capital component, considering any nec- identifying and representing its terms and rules with SBVR
essary deductions such as the excess of permanent assets, (structured natural language). Note that there isn’t a SBVR
and the minimum required capital. From the margin over structured vocabulary for Portuguese; in this sense, relation-
required reference capital, one can determine if a financial ships and rules are translated to English so they can be based
institution is compliant with the capital requirements verify- on SBVR-SE (SBVR Structured English), but terms (con-
ing if its value is non-negative, analogous analysis could be cepts) are kept in their original form.
made to determine if the institution is non-compliant (nega- Marinos, Gazzard, and Krause 2011 proposed the creation
tive margin). of a tool for the manipulation of SBVR vocabularies with
Given these final considerations, concepts like reference auto-completion and highlighting features, work that later
capital requirement compliant and reference capital require- evoluted to SBVR Lab 2.04 , a tool freely available on the
ment non-compliant can be defined (shown as part of an ex- Web; this tool was used to help achieving the goals of the
ternal ontology in Figure 3), which indicate the compliance third step.
(or not) of some Brazilian financial institution to BACEN’s The fourth step begins with the possession of the SBVR
prudential regulation at a specific time. vocabulary; its terms and rules are distributed in a set of dis-
tinct vocabularies, mainly by considering their subject mat-
ter, what inevitably demands interpretation and expertise of
Construction and Usage the domain; secondly, we take into account technical con-
The nine steps for developing and using OntoBacen are siderations, in order to achieve, for instance, a higher degree
shown in Figure 4. The six first steps are for the ontology of disjointness between vocabularies, and thus minimizing
construction, while the last three are for its usage. BACEN’s the number of relationships between terms of distinct sets.
prudential regulation is the input to this methodology, and is Other important issue involves the detection of core terms,
available as a set of .pdf files written in Portuguese, that are
4
copied to an easily editable format, such as text files. See: http://www.sbvr.co/.
those that are extensively related to other terms, and should
be organized in such a manner that could be easily shared ...
Reference Date
List of Reported Limits
between multiple modules.
Legal Person Identifier
All these initial steps involves human interpretation and Reference Capital
manual activities to be accomplished; they demand famil- ... List of Parameters (Account Code = 100) Account
Value
iarity with Brazilian legal writing, as the BACEN’s norms
are written similarly to any other Brazilian law document
List of Accounts
(legalese), and also require prior knowledge about its subject
matter, since these norms aren’t designed to be educational. ...
The fifth step is the conversion of these multiple SBVR
vocabularies to OWL 2.0, generating the ontology modules.
Some works in the literature address the conversion between
these two languages without semantics loss, however a more
detailed description is beyond the scope of this paper. Kar-
povic and Nemuraite 2011 worked in the field, resulting in
Figure 5: Example of limit report XML file (DLO)
the creation of a conversion tool freely available on the Web,
called s2o5 . The initial step for the creation of OntoBacen
OWL files used this tool6 . a numerical identification code, and an amount of money
In the sixth step, the initial OWL files obtained from s2o (Brazilian reais).
are subjected to manipulations in order to complement and A Java application was developed to do the XML deseri-
enrich their semantics. For that, the Protégé7 software was alization, that also used Apache Jena framework (with some
used to: help of SPARQL queries), for the automatic instantiation of
the individuals to be later used in conjunction with Onto-
• Add or update namepaces and ontology imports; Bacen. The final result is a set of RDF files, following the
• Include some cardinality restrictions not captured in the same modularization approach used in the ontology creation
SBVR vocabularies phase; phase.
From now on, all components required for the reasoning
• Add labels and other annotation properties useable for phase are ready. The reasoning can be done through a spe-
documenting the ontology (in Portuguese and English); cific module, subsets of modules, or even all of them. In
• Create SWRL rules to allow the inference of additional this phase, an inference engine compatible with SWRL rules
axioms, including the use of its built-in functions for math must be used; for the test cases presented in this paper, Pellet
and date operations. (Sirin et al. 2007) was used for such a task.
Finally, by using the set of inferred axioms, one can
At this point, each OWL file is the final version of an on- get the answers to the proposed competence questions, for
tology, concluding the terminological component (TBox) of instance: a Brazilian financial institution complies to BA-
the system. CEN’s prudential regulation in a determined context, if it is
The main reason to adopt such approach is because ma- classified as capital requirement compliant in that context.
nipulating a large vocabulary such as the BACEN’s pruden- Notice that each module can answer some limited and
tial regulation, is made quickly by just listing its terms and specific questions, for instance: the Market Risk ontology
rules in text files than formalizing all them in ontology de- can only determine the degree of exposure to market risk,
velopment tools. In this sense, for OntoBacen’s initial de- and that isn’t sufficient to verify the compliance or not to
velopment phase, described in this paper, only small subsets BACEN’s prudential regulation; the presented competence
of the identified concepts were subjected to the formal on- questions have a higher level of complexity, and need some
tology engineering process (step 6). reasoning over all modules of the ontology so that they can
For the construction of the assertion component (ABox), be answered.
the input artifact for the definition of instances of the on-
tology classes (individuals) is the DLO8 , a XML report file Test Cases
that Brazilian financial institutions must submit to BACEN,
briefly depicted in Figure 5. For testing the proposed knowledge system capabilities, a
This file, containing the data required for the possible standard example of the DLO was used, as provided by BA-
assertions which one can infer by the use of OntoBacen, CEN. From there, three additional test cases were created
is available in accordance with a XML Schema similar to by altering elements of the XML file, so that their instantia-
accounting, where each structure is composed basically of tion could be automatically done by the custom application
mentioned in the construction and usage section.
5
See: http://s2o.isd.ktu.lt/about.php. Table 1 shows the relevant input data of the four test cases,
6
It was necessary to apply a simple syntax transformation algo- in millions of Brazilian reais, and for a matter of simplicity,
rithm in this step, whose details are beyond the scope of this paper. the market risk RWA models were consolidated by its com-
7
See: http://protege.stanford.edu. ponents, as it was shown in Figure 2.
8
DLO is an acronym for Demonstrativo de Limites Opera- According to the information shown in Table 1, OntoBa-
cionais in Portuguese, and stands for operational limit statement. cen can apply business rules by using an inference engine
compliant with OWL and SWRL rules, in order to answer if tive margin over the required reference capital and therefore
the institutions in each case are compliant or not with BA- the conclusion that it is not adherent to the norms.
CEN’s capital requirements regulations, considering their Case 3 has both models implemented too, but the internal
exposure to market risk. approach indicates the lower risk exposure (being reported),
not sufficiently low to break the barrier of the standardized
Case 1 2 3 4 approach weighted by the Sm factor (90% in this case be-
Reference Date 2015 2015 2016 2016 cause the model transition was done in less than a year).
Jan 1 Jan 1 Jan 1 Jan 1 Additionally, the reference date is now at the year of 2016,
Mkt. Risk Model N/A 2014 2015 2014 when the reference capital requirement decreases from 11%
Transition Date Jul 1 Jul 1 Jan 1 to 9.875% of the RWA, and finally results in a handily pos-
Reference Capital 2139 1000 1000 1000 itive margin over the required capital, concluding that this
Excess of perma- 0 500 0 100 case is compliant.
nent assets In the last case, number 4, the market risk approaches dif-
RWAMPAD 382 5000 5000 15000 fer greatly from one another, in such a manner that even the
RWAMINT N/A 6000 4750 10000 Sm factor being of 80% (since the model transition hap-
pened more than a year before the reference date), the in-
Table 1: Test cases input data ternal model broke the barrier of the standardized approach
weighted by Sm, so that the final RWA value by the inter-
Case 1 2 3 4 nal model will be the same of the barrier. As in this case
Sm Factor N/A 90% 90% 80% the internal model will assume 80% the value of the stan-
RWAMPAD (con- N/A 4500 4500 12000 dardized approach, OntoBacen will choose it for being the
sidering Sm) lower one. The minimum required reference capital is then
RWAMINT (con- N/A 6000 4750 12000 evaluated from the resulting RWA and compared with a mi-
sidering Sm) nor value, which brings to a negative margin and the conclu-
Reported Market RWA RWA RWA RWA sion of non-compliance to BACEN’s capital requirements
Risk RWA MPAD MPAD MINT MINT regulations.
Reported RWA 382 5000 4750 12000 All test cases executed the mentioned inferences in few
Reference Capital 11% 11% 9.875% 9.875% seconds, by using Pellet reasoner within Protégé, so that
Req. Factor computational performance was adequate, since this isn’t a
Min. Required 42 550 469 1185 real time problem. The time for loading required ontologies
Reference Capital from the Web was approximately thirty seconds.
Ref. Capital for 2139 500 1000 900 OntoBacen currently deals with very general and aggre-
RWA comparison gated concepts, so that performance was not a great con-
Margin over Req. 2097 -50 531 -285 cern yet. However, as the level of detail (and data) increases,
Reference Capital to deal with all data at once could be a performance issue,
given the exponential nature of reasoning systems; this is
CQ4: Ref. Capital Yes No Yes No
another reason to consider modularity, alongside with dis-
Limit Compliant?
tributed reasoning, as a primary design requirement for mod-
Table 2: Test cases inference steps and results eling this domain.
Case 1 is the default example provided by BACEN9 . In Conclusions
this case, only the standardized market risk approach is used, This paper presented an alternative technical approach for
so that it is necessarily the reported one, and the reference dealing with risk management: instead of the standard spe-
capital is more than five times higher than the reported RWA, cialized systems, we propose to use ontology-based tech-
with a highly positive margin over the required reference nologies together with a knowledge-based system.
capital, resulting in the inference of such institution as com- Our proposal provides an open knowledge reference to
pliant to the capital requirements. address issues of this domain, freely available in the Se-
In case number 2, both models are implemented, but the mantic Web10 ; its semantics provides a formal representa-
standardized model indicates a lower risk exposure than the tion of rules established by BACEN’s prudential regulation,
internal model, and as OntoBacen is configured to choose and also a computational artifact that along with automated
the minimum exposure approach (this isn’t obligatory), the reasoning can provide answers to the user.
standardized approach is chosen to be reported. This institu- By the use of such approach, any Brazilian financial in-
tion has a high amount of permanent assets, so that the refer- stitution that needs to report accounting and risk indicators
ence capital is halved when reduced by the excess of perma- to BACEN by using an XML operational limit statement
nent assets, making the minimum required reference capital (which has only syntactical and structural constraints, with-
higher in comparison with this amount, resulting in a nega- out formal semantics), could verify its semantic consistency
before submission, or even generate semantically consistent
9
The operational limit statement XML files are available at:
10
http://www.bcb.gov.br/fis/pstaw10/leiaute limitesDLO.asp. URI: http://lti.pcs.usp.br/⇠filipe.polizel/OntoBacen/.
content, avoiding mistakes and delays that can bring even- [Abi-Lahoud, OBrien, and Butler 2013] Abi-Lahoud, E.;
tual penalties and losses. OBrien, L.; and Butler, T. 2013. On the road to regulatory
Finally, for the effective success of such solution, modu- ontologies: Interpreting regulations with sbvr. AICOL.
larity must play a central role in its design from the begin- [BACEN 2014] BACEN. 2014. Regulação prudencial. Re-
ning to the end, allowing financial institutions to take advan- trieved from https://www.bcb.gov.br/?REGPRUDENCIAL,
tage of computational features such as distributed comput- last access on 2015 Apr 12.
ing and avoiding to handle massive amounts of knowledge [Berners-Lee, Handler, and Lassila 2001] Berners-Lee, T.;
at once, given the exponential computational complexity of Handler, J.; and Lassila, O. 2001. The semantic web.
logical reasoning algorithms. Scientific American.
[BIS 1997] BIS. 1997. Core principles for ef-
Further Work fective banking supervision. Retrieved from
This work intends to further explore the BACEN regulations http://www.bis.org/publ/bcbs30a.pdf, last access on
in order to define business rules for other risk types, such 2014 Aug 8.
as credit and operational risk. Besides, the ontologies for [BIS 2012] BIS. 2012. Core principles for ef-
capital definition will be created so that the reference capi- fective banking supervision. Retrieved from
tal concept can be decomposed in more detailed definitions. http://www.bis.org/publ/bcbs230.pdf, last access on
Therefore, the Capital Requirements ontology could be 2014 Aug 8.
enriched in order to consider all possible types of risk, and [BIS 2013a] BIS. 2013a. Principles for effective risk data
capital requirements rules could be evaluated in additional aggregation and risk reporting.
levels, such as tier 1 capital requirements. [BIS 2013b] BIS. 2013b. Regulatory consistency assess-
Other issue concerns the integration capabilities of the ment programme (rcap) assessment of basel iii regulations
proposed ontology. In order to address that, additional se- in brazil.
mantic alignments can be done between OntoBacen and
other ontologies, while the work on financial ontologies [Engel et al. 2013] Engel, P.; Hamscher, W.; Shuetrim, G.;
(FIBO, FIRO, etc) progresses. Another possible approach vun Kannon, D.; and Wallis, H. 2013. Extensible business
for this question would be consider additional ontologies as reporting language (xbrl) 2.1.
possible semantic foundations, such as a mathematics ontol- [Grüninger and Fox 1995] Grüninger, M., and Fox, M. S.
ogy to express some of the relations identified in OntoBa- 1995. Methodology for the design and evaluation of on-
cen (derivation of concepts by differences, maximums, etc). tologies.
As these goals are achieved, additional efforts can be done [Guarino 1995] Guarino, N. 1995. Formal ontology, concep-
in such a way to deepen the concepts definition for each sub- tual analysis and knowledge representation. International
domain that composes the governance of a financial system Journal of Human and Computer Studies 43:625–640.
risk management, increasing the proposed system granular- [Karpovic and Nemuraite 2011] Karpovic, J., and Nemu-
ity and thus allowing it to treat additional risk management raite, L. 2011. Transforming sbvr business semantics into
questions (or even accounting). To this end, a lot of work in web ontology language owl2: Main concepts. In Proceed-
the interpretation of an extensive set of BACEN regulation ings of the 17th international conference on Information and
documents should be done, and as the ontology evolve in Software Technologies (IT 2011), 231–238.
depth of details, new modular considerations must be taken
[Lohmann et al. 2014] Lohmann, S.; Negru, S.; Haag, F.; and
into account, such as those discussed in this paper.
Ertl, T. 2014. Vowl 2: User-oriented visualization of ontolo-
After the construction of the modules proposed in the gies. In Proceedings of the 19th International Conference
specification section, this work also aims to evaluate the ap- on Knowledge Engineering and Knowledge Management,
proach using a more realistic case study, by applying Onto- EKAW ’14, 266–281. Springer.
Bacen rules with the data of a real Brazilian financial insti-
tution, that is publicly available via balance sheets and risk [Marinos, Gazzard, and Krause 2011] Marinos, A.; Gazzard,
reports, or even by collecting more detailed data using the P.; and Krause, P. 2011. An sbvr editor with highlighting and
operational limit statement XML sent to BACEN by some auto-completion.
institutions. [Niles and Pease 2001] Niles, I., and Pease, A. 2001. To-
wards a standard upper ontology. In Proceedings of
the 2nd International Conference on Formal Ontology in
Acknowledgements Information Systems (FOIS 2001). ACM. See also
Jaime Sichman partially financed by CNPq, Brazil, proc. www.ontologyportal.org.
303950/2013-7. [OMG 2008] OMG. 2008. Semantics of business vocabulary
and business rules (sbvr), v1.0.
References [Sirin et al. 2007] Sirin, E.; Parsia, B.; Grau, B. C.; Kalyan-
pur, A.; and Katz, Y. 2007. Pellet: A practical owl-dl rea-
[Abi-Lahoud et al. 2013] Abi-Lahoud, E.; Butler, T.;
soner. Web Semantics: Science, Services and Agents on the
Chapin, D.; and Hall, J. 2013. Interpreting regulations with
World Wide Web archive 5:51–53.
sbvr. In Proceedings of RuleML (2)’13.