Ontology Summit 2019 Financial Explanations Track

Ontology Summit 2019 Financial Explanations Track
Session 1: Financial Industry Explanations 06 February 2019

This Presentation We will look first at a number of finance scenarios in which explanation is needed, and consider the role of ontologies in that Then we will look at some of the challenges in explaining the ontologies themselves.

Network of Financial Exposures
to counterparty

Some of these are looking a little shaky…

POP! Where does that leave the survivors?

This requires some explaining POP! Where does that leave the survivors?

Determining Financial Exposures
Some banks took days or weeks to fully determine their exposure to failing institutions They had all the data they needed… Data without meaning is not knowledge “The lack of a common industry language is a billion dollar problem” US Treasury Office of Financial Research

Financial Contexts Reporting Decision support Risk management
Investment performance attribution Credit assessment

Financial Reporting Detail
Risk reporting Risk assessment Risk management compliance e.g. BCBS239 Fair price / price discovery / investment compliance E.g. MiFID Valuation Mark to market / mark to model Models explanations

Part 1: Explaining Things with Ontology

Ontology Usages Reasoning Semantic querying Network graphs
Business visualization and presentation Derivation of dictionaries / glossaries Etc.

Ontology Issues and Considerations
Theory of Meaning Grounded versus Correspondence theory (isomorphism) in ontologies How this features in explaining the ontology What each element of the ontology represents Understanding each element of the ontology in relation to Other elements in the ontology (grounded) Contextual relations (isomorphic / correspondence)

Correspondence Theory Semantics
Graph has logical relations between elements These correspond to the relations between things in reality Automated reasoning checks the “deductive closure” of the graph for consistency and completeness

Correspondence Theory Semantics
The more detailed logic there is in the application ontology, the more confident we can be that it reflects only one set of things and their relation in reality These ontologies can do powerful processing in an application Considerations Some techniques appropriate for these ontologies would not be useful in a Concept (Reference) Ontology Decide whether to have application and reference ontologies in separate namespaces, or satisfy both sets of requirements in one namespace

Foundational Semantics: Semantic Networks
Directed Graph The meaning at each node is a product of its connections to other nodes Semantically grounded at certain points in the graph

Deep and Shallow Ontologies
Deep classification hierarchy of types of thing in the world, with relationships and sufficient logic to disambiguate Self-contained classes, properties and logical statements corresponding to some set of things in the world

Portfolio Explanations
Explain: What is the value of my portfolio? What is the outlook (projected cashflows? What is the risk profile (defaults etc.)? Use of models Valuation Mark to market Mark to model Analytics Risk models

Portfolio Explanations
What do the models look like? Valuation: Mathematical models Analytics: Mathematical calculations Derived from Net Present Value (NPV), today’s date, contract terms, maturity etc. Risk Loss given default Probability given default (complex probability calculations) Language: Mathematics, arithmetic, statistics Potential role of ontology Define ‘meaning’ of model inputs Defining ‘meaning’ of model outputs Can’t replicate the ‘language’ of mathematics but can describe the components of these in FOL (declarative) terms

Using Ontology to Explain Something
Some Proofs of Concept In outline, with a view to the explanatory aspects of these Used ontology to explain something in the world of finance How did we explain the explanations? Topics Covered Corporate relations and financial exposures Use of visualization Results of reasoning, querying etc Reporting – explaining reports content Bank of England PoC Regulatory Compliance – Reg W PoC See also last week’s slides examples from Benjamin Grosof: Explaining the ‘meaning’ of Affiliate

Some Visualizations Graphical representations of
Concepts in the ontology Individual data Results of queries Results of inference processing / reasoning (automated classification of content etc.)

Counterparty Exposures PoC
And other visualizations

Semantic Information Integration Semantic Information Integration
Proposed FIBO Architecture for Institutional and Macroprudential Oversight of OTC Derivatives Regulatory Agencies Semantic Triple Store Financial Institutions Aggregated Linked Data Query Swap Database(s) Swap Data Repository Database(s) Legal Entity Data Provider(s) FpML Trading System Trading & Compliance System(s) Swap Reporting Semantic Information Integration Platform Regulatory Risk Analyst Semantic Information Integration Platform Institutional Risk Analyst Links to the Semantic Web Semantic Triple Store Linked Open Data Cloud (External) FIBO Ontologies FIBO Ontologies

Requirement #3: Electronically Express Contractual Provisions
Semantics can Represent Contractual Provisions of Swap Agreements for Risk Classification Requirement #3: Electronically Express Contractual Provisions Transaction Repository, et.al. ISDA Master Agreement Schedules Credit Support Annex Events Credit Rating Agency Downgrade Counterparty Credit FpML Capture Semantics of Contract Provisions OTC Derivative Confirm Counterparties FIBO Operational Ontologies FIBO Operational Ontologies Reduce Value of Collateral Market Reference Data Define Axioms Default Events Identify Key Contractual Events Risk Analyst Classify Contract Type by Cash Flow Termination Events Infer Counterparty Exposures Classify Counterparties into Risk Categories for Analytics Increase Collateral Identify Key Contractual Actions Transfer Payments *Report on OTC Derivatives Data Reporting and Aggregation Requirements, the International Organization of Securities Commissioners (IOSCO), August 2011 **Joint Study on the Feasibility of Mandating Algorithmic Descriptions for Derivatives, SEC/CFTC, April 2011

Requirement #4: Link Disparate Information for Risk Analysis
Semantics can Identify Positions of Legal Entities and Their Owners Across Trades and Asset Classes Requirement #4: Link Disparate Information for Risk Analysis ?entity Legal Entity type ?legal Name hasExactLegalName ?parent hasImmediateParent ?swap partyToSwap ?amount notional Amount AtRisk Transaction Repository Z ?entity Legal Entity type ?legal Name hasExactLegalName ?parent hasImmediateParent ?swap partyToSwap ?amount notional Amount AtRisk Transaction Repository Y Transaction Repository X Legal Entity type ?entity hasExactLegalName ?legal Name Risk Analyst hasImmediateParent ?parent partyToSwap swap Notional Amount ?swap ?amount type type Vanilla Interest Rate Swap subClassOf Basis Swap Interest Rate Swap subClassOf Query all Transaction Repositories to report on the sum total of aggregate exposure for all counterparties and their parents involved in all swaps associated with an interest rate swap taxonomy Data is queried using graph pattern matching techniques vs. relational joins Queries can process inferred data and highly complex and abstract data structures Queries can federate across semantic endpoints (using SPARQL 1.1) Data can be aggregated and summarized (using SPARQL 1.1)

FIBO Identifies Instrument Contractual Terms and Attributes: Signature of a CDS
Visualization of a credit default swap, and the data types and classifications of many key attributes

FIBO Financial Instrument Taxonomy Visualizations
FIBO has a Rich Multi-Tiered Taxonomy that can be used to Classify and Aggregate Data at Many Levels and Across Many Facets (subset of taxonomy shown in diagram) FIBO maps to other product taxonomies e.g. ISDA Gruff SPARQL query tool and Allegrograph Triple Store from Franz, Inc

FIBO Visualizations of Business Concepts
FIBO enables high level concepts e.g. a Fixed Float IR Swap, along with its key related concepts, to be queried for access and viewing This example is the result of a single query against the FIBO semantic glossary

Query Results (Ownership and Control Relationships)
Semantic web enables data visualizations which are more holistic and descriptive than basic columnar views FIBO aligns with LEI Legend

Ultimate Parents, their Descendents and Trading Counterparties
This capability allows for the rollup of both positions and exposures of the subsidiaries to the level of the ultimate parent for risk analysis

Visualization of Ownership Hierarchies and Exposures to Counterparties
Solid blue lines represent ownership and control relations. Violet lines represent exposures due to trading Legend

Use of Graph Analytics Software to Describe Risk
Score = eigenvector centrality of adjacent network positions SPARQL interface with R “igraph” package Line width reflects aggregate amounts at risk between individual trading parties Node size reflects grand total aggregate amount at risk for entity

Bank of England Reporting PoC

Regulatory Reporting Current State
? Reports (forms) FORMS FORMS REPORTING ENTITY REGULATORY AUTHORITY

Regulatory Reporting Current State
? Reports (forms) FORMS FORMS REPORTING ENTITY REGULATORY AUTHORITY Change in Reporting requirements = Redevelopment effort By each reporting entity For each system and form Uncertainty Content of the reports Are we comparing like with like? Data quality and provenance

Regulatory Reporting with Semantics
! Thing IR Swap CDS Bond Contract Granular data Thing IR Swap CDS Bond Contract Common ontology Common ontology REPORTING ENTITY REGULATORY AUTHORITY Data is mapped from each system of record into a common ontology Reported as standardized, granular data Agnostic to changes in forms Receives standardized, granular data aligned with standard ontology (FIBO) Uses semantic queries (SPARQL) to assemble information Changes to forms need not require re- engineering by reporting entities

Regulation W PoC Reg W: Reg W PoC
Reg-W is a Federal Reserve regulation that established terms for transactions between banks and their affiliates Reg-W was enacted by Congress as part of the Federal Reserve Act and applies to all federally-insured depository institutions Reg W PoC PoC was formed by the EDMC that included Wells Fargo Bank, Coherent Knowledge Systems, SRI International, and the Governance, and Risk and Compliance Technology Centre (GRCTC) of Ireland, and various other members of the council

Regulation W The Federal Reserve Board’s (“FRB”) Regulation W (Transactions Between Member Banks and their Affiliates) implements Sections 23A and 23B of the Federal Reserve Act (“FRA”). Protects the financial integrity of banks: Bank affiliate includes any company that controls the bank, any company under common control with the bank, and certain investment funds that are advised by the bank or an affiliate of the bank. Limits covered transactions with affiliates that are not subsidiaries of banks (Reg W affiliates). Imposing collateral requirements on extensions of credit Prohibiting the purchase of low-quality assets by banks from their Reg W affiliates or sister banks Limits: Covered transactions with an affiliate cannot exceed 10 percent of a bank's capital stock and surplus, and transactions with all affiliates combined cannot exceed 20 percent of the bank's capital stock and surplus. Wes

Regulation W PoC Aims How to unambiguously understand and automatically comply with regulatory rules Regulation W (Reg-W) as a test case for the use of the Financial Industry Business Ontology (FIBO) in combination with advanced semantic rules (Rulelog/Flora-2) to automatically keep a bank in compliance as transactions were being processed

Reg W PoC Overview Am I in Compliance? FIBO Yes/No Why/Why not
EpistoTM engine & UI for Rulelog Regulation W FIBO Related Laws, Regulations, P&Ps, etc. Data from Legacy Systems Compliance and Risk Officers Am I in Compliance? Yes/No Why/Why not Map from Formal Logic to English

Reg W PoC Overview Explanation Am I in Compliance? FIBO Yes/No
EpistoTM engine & UI for Rulelog Regulation W FIBO Related Laws, Regulations, P&Ps, etc. Data from Legacy Systems Compliance and Risk Officers Am I in Compliance? Yes/No Why/Why not Map from Formal Logic to English Explanation

Reg W PoC FIBO and Rulelog/Flora-2 were used to make Reg-W requirements explicit and applied to sample transaction data to automate compliance assessment. Detailed explanations are provided so that humans can understand the reasoning and facts that led to the conclusions. GRCTC provided expertise in controlled natural language for rule authoring via OMG's Semantics of Business Vocabulary and Business Rules language (SBVR). Coherent Technology (Episto) and SRI technology (Sunflower) provide automated reasoning capabilities with detailed explanations in English SRI's technology provides automatic import of OWL knowledge (FIBO) into Flora-2

Covered Transactions and Exemptions
Type of Covered Transaction Asset Purchase from an Affiliate Purchase of, or an investment in securities issued by an affiliate Attribution Rule - via extension of credit Extension of Credit Exemptions Intraday Credit to Affiliates Riskless Principal Transactions Municipal securities purchases Transactions secured by cash or U.S. gov’t securities Purchasing assets, other than securities issued by affiliates, that have ready, liquid markets. Wes

Main concepts in Regulation W
Bank enters in a transaction with a Counterparty Check if the counterparty is an affiliate Check if the transaction is type is covered by regulation W Verify if the amount/ total amount is permitted Elie - On from Wes’s slides – summary of regW – 30 seconds Objective of this PoC: Am I in compliance with RegW? Yes/No Why? / Why not? Start with understanding Reg W

Challenges in understanding RegW
Unstructured Text size Federal Reserve System Final Rule 12 CFR Part pages of text Summary: 19 pages (comprehensive review) Reference chains Implements the Federal Reserve Act (references it and others) Definitions to identify, delimit and flesh out Complex sentences: Legalese and NL ambiguities Exceptions/ exemptions Elie – 1.5 mins – (Elie’s total 2mins up to here) - flash examples here

Current techniques SMEs (both Business and Legal)
Link to regulation (unreliable, difficult to maintain, hard to navigate) Arbitrary categories, lack structure and accurate definitions SME’s understanding of the regulation Traceability? Reusability? Can share? SMEs (both Business and Legal) Handcraft guidance Partial Coverage often limited to recurrent activities Based on non-documented/ non-formalised process Accuracy at the discretion of the SMEs Thoroughness, Diligence depends on the SMEs Lacks transparency, disconnected… reusability? More problems when the rule is complex Lack of structure i.e. taxonomy of transaction types, exemptions, collaterals, etc… Link to other regulations, new spreadsheets??? Elie – 2 mins (Elie’s total 4mins up to here)

Use Structured Natural Language
Following GRCTC methodology to interpret regulation in SBVR SBVR OMG Specification for business Vocabularies and Rules Vocabulary: Captures the business domain Terms referring to business concepts, links/relationships between concepts, definitional constraints on these relationships Rules: Capture the business behavioural constraints Obligations, prohibitions, etc. Elie – 30 seconds - (Elie’s total 4.5mins up to here)

GRCTC methodology to interpret regulation in SBVR
Follow reference chains and produce self-contained sentences Define terms iteratively until all confusions are clarified Identify, Describe and Constrain links/relationships between terms Capture regulatory requirements using the interlinked vocabulary elements from previous steps Elie – 30 seconds (Elie’s total 5mins up to here) – use process representation

Limiting PoC Scope using SBVR SE
Each is defined Body Corporate Source: CFR II § 223.2 Elie - 1min (Elie’s total 6 mins up to here) FIBO Concept

Elie 30 – seconds (Elie’s total 6.5 mins up to here)

Not only terms but relationships too: Fleshing out the links between concepts Elie 15 – seconds (Elie’s total <7 mins up to here)

Elie 1 minute (Elie’s total <8 mins up to here) Reconstruct the rule from previously defined “building blocks” to ensure confusion is removed

Elie – 10 to 30 seconds (Elie’s total 9 mins up to here) Other examples of regulatory requirements captured in SBVR

Reg W PoC Outcomes Benefits to this semantic approach include:
Improved confidence about correctness of compliance checks Both for bank domain experts and regulators Largely because understandable explanations are provided. Reduces cost and risk due its ability to adapt more easily and quickly to changes in regulations Because it utilizes a common financial language to ensure alignment with standards.

Semantics for rules Rules have predicates
Program rules have programmatic predicates Rules exist in the real (business) world also Business rules should have business predicates Otherwise they are not really business rules You can’t define rules at the business level by reference to data model elements in some logical data model desin See e.g. efforts to define MBS Waterfall models in Python Examples: Reg W Proof of Concept

Part 2: Explaining Ontology with Things

Explaining Ontologies
A difficult research problem Issues: Explaining to business SMES: Context Explaining to Implementers: Depth Some of the techniques looked at Tables (term, definition, synonyms Uses ontology to disambiguate the concepts, hidden from viewer Diagrams (boxes and lines) Explained in set theoretic terms

How we Explain Semantics
We start with some kind of thing Some kind of thing

We ask just two questions about this kind of thing: What kind of thing is it? What distinguishes it from other things? Some kind of thing

Animal What kind of thing is it? Vertebrate Invertebrate Bird Mammal Fish Waterfowl Some kind of thing

Animal What distinguishes it from other things? Vertebrate Invertebrate Bird Mammal Fish Waterfowl Walks like a duck Some kind of thing Swims like a duck Quacks like a duck

Animal It’s a Duck! Vertebrate Invertebrate Bird Mammal Fish Waterfowl Walks like a duck Swims like a duck Quacks like a duck

Possible classes of Thing

Example “Thing”: Equity
Real world definition of Equity: "An equity is a financial instrument setting out a number of terms which define rights and benefits to the holder in relation to their holding a portion of the equity within the issuing company".

What is an Equity? Or to put it another way… Equity Equity security
Instrument Terms Financial Instrument Is a kind of Has rights defined in In relation to

What is an Equity? Using OWL to define the classes of real things in the world, and the facts about those things Modeled in TopBraid Composer

Financial Semantics in OWL
Pizza approach “Everything is a Thing” What about common terms? accounting terms for equity, debt, cashflow Places, time concepts Legal terms (securities are contracts) Better partitioning needed

The Explanation Issue Consider this diagram for Bonds

Examples: Bonds

Examples: Bonds Foundational concepts for Contracts
Terms common to all debt Interest, Principal terms common to Bonds Terms specific to Municipal Bonds

The Explanation Issue What the business user expects to see:
Definition of what it means to be a Municipal Bond Not expecting to have to understand inherited stuff What the implementer expects to see Lots of context-specific data elements Swap interest first payment date Many of these have unchanged semantics from higher level abstractions Words Maturity: Original Maturity or Current Maturity Data: Current doesn’t even ‘mean’ anything in data

Another Example: Ratings

Ratings Model Explanation
The example above uses Financial Instrument as the rated entity. This same model also applies to: Financial Institutions Accounts Customers Geographic Units Benchmarks Retail loans and mortgages Property Rating Agency – such as Moody’s Geographic Unit – there’s a rating set that has different values in Italy than in the rest of Europe Rating Set Definition – general information about the rating set. Rating Values – these are the values such as AA, AA+, etc. Rating Qualifier – some rating agencies add a prefix or a suffix to further explain the rating. For example, on some of their rating sets Moody’s adds a qualifier of RUR to indicate the rating is under review.

Explaining Ontologies – Research Questions
Contextual ontologies Contextual diagrams (of a broader ontology)

Concepts Extraction: Ontology
Context-specific concept models = Context N-dimensional content Shown as 4D for simplicity Various extracts from that hypercube in lower dimensionality

Concepts Extraction: Diagrams
Context-specific diagrams = Context N-dimensional content Shown as 4D for simplicity Various diagrams of content in that hypercube in lower dimensionality

Vocabulary Layer by Context
Context-specific concept models Vocabulary Local Context Vocabulary Ontology N-dimensional content Shown as 4D for simplicity Various extracts from that hypercube in lower dimensionality

Explanation Application Areas Business Understanding
Visualization Reasoning AI and NLP Business Understanding Valuation models Lending Credit ratings Audit: Explanation’s up-tight cousin

Application Areas Visualization Reasoning AI and NLP
Use of network graphs and visualization in risk and exposures Reasoning Use of reasoning in exposures, risks, counterparties, contagion Along with network viz of the results AI and NLP Use of NLP in determining exposures to New regulations (conformance requirements) Contractual commitments

Understanding Ontology of valuation models
Challenge: language of models is math; ontology as FOL represents a different view Lending decisions accountability (retail lending and credit) Reporting, reasons for rejection Automated recommender systems Credit ratings (rating agencies) Retail, wholesale lending / finance Instrument rating

Summary Ontologies help in explanation in many financial scenarios (lending, credit, risk and exposures) Reasoning with ontologies adds a further aspect that itself needs to be explained Ontology concepts still need to be presented to end users in an explainable way Not all things to be explained are first order Rules, mathematical constructs etc.

Conclusions Explanations involve
Language Context Language: That in which you explain what you explain Context: Most human understanding is contextual and this needs to be taken into account in presenting explanatory material

Questions?

Ontology Summit 2019 Financial Explanations Track

Similar presentations

Presentation on theme: "Ontology Summit 2019 Financial Explanations Track"— Presentation transcript:

Similar presentations

About project

Feedback

Log in

Auth with social network:

Ontology Summit 2019 Financial Explanations Track

Similar presentations

Presentation on theme: "Ontology Summit 2019 Financial Explanations Track"— Presentation transcript:

Similar presentations

About project

Feedback