% notes from group 2 breakout on semantics and contract representation % NSF Workshop on Financial Cyberinfrastructure, held 7/19-20/2012 % scribed by Benjamin Grosof, co-lead of the breakout %%%% Agenda at high-level: use cases building blocks semantic abstractions and representation - roadmap open data to make available to/by researcher community interleaved in the above: - general/misc. discussion - pragmatic implementation into deployment environments . integrating math of specialized kinds %%%% Use Cases: Big 3 business tasks: o transaction processes (eg, trade, loan, withdrawal, order, settle) - related: lifecycle (see below) o risk management (involves analytics heavily) o oversight for regulation (involves audit trail, reporting). E.g., - transparency - systemic risk - market surveillance - idea: cooperate with FINRA (tho' their data quality has problems; need sem's) %%%% Open Data to make available to/by researcher community: i.e., community resources to support research financial counterparty (-network) reference DB - part of mission of OFR financial instrument reference DB - part of mission of OFR repositories of contracts, esp. of various types - start by analyzing the original textual forms - eg Edgar, ISDA, FPML, venture capital (Oliver O), ABA e-contracts, ? CFTC, SEC prospectus, master agreements, ... mortgage loan doc's, Bloomberg, Reuters, IDC (Interactive Data) . some contacts: EDM Council, Lucian Popa, Rachel Pottinger, ? Nancy Wallace - can use ontology extraction tools OFR end: - Research and Analysis suborg - see directions in Annual Report released today 7/20/12 - still working on how to aggregate and publish the various publicly available data - idea: maybe an OFR program similar to NSF CRII below NSF Community Resource (CRII) solicitation under computer science - data and/or tools - plausible esp. for smaller proposals - deadline ~10/23 take a page from Linked OpenData Sloan Foundation wikidata / DBpedia see linkedct.org candidate data sets: - Lucian Popa: IBM crawled SEC data, pre-cleaning, in JSON . possibly also for universities: cleaned data - Nancy Wallace: contract documents . maybe: for universities (unrestricted is quite expensive) dream/vision: PubOFR (or PubFin) akin to PubMed hosting on a cloud data platform - Phil Bernstein: eg Azure probably - OFR does not have outside-the-firewall stuff yet markup of doc's with structured data eg FIBO - Annotator from Protege stuff by Mark Musen group at Stanford . Nigam Shah contact - tool being dev'd in Karsha (Louiqa Raschid) - Semantic MediaWiki+ (Vulcan's) -- popular semantic wiki develop financial-relationship network graphs for systemic network analysis - have weights %%%% Semantic "Building Blocks": (useful across many use cases and tasks) roadmap strategy of how to progress on these semantic "building blocks" - start with the "central" kinds of info: . transactions . contract terms also: . counterparties (referenced by the above) . portfolio holdings (derived from the above) . obligations and rights (derived from the above) - develop ontologies of . state attributes mentioned directly in the above . key events, eg contingencies and actions in contracts - develop other core ontologies of basic business etc. . collateral, assets, liabilities . obligations, rights - semantically represent the central info - then expand scope and iterate refine . follow steps of relationships/linkages/references %%%% More from use cases discussion: o asset management o capital formation o strategic behavior - investor objectives - supporting strategic network analysis 1. Visualization and analysis of a financial counterparty network 2. Knowledge representation of a financial contract 3. Implementation of a ???living will??? for a large financial firm 4. Fostering an ecosystem of credit analysis 5. Reasoning over financial contracts for completeness and integrity 6. Privacy and trust: multiparty sharing of confidential financial data there's also a "geek" axis of technical aspects used in the above use cases o data quality and validation - both input and output - eg trade fail rates o performance metrics wrt the above, eg: - cost - degree of automation concepts: o supply chain analysis (ie supply chain of financial info/analysis) o operational value add wrt data mgm and business processes - within firms o financial transactions lifecycle analysis - financial research, trade, clear/settle, to value, report, comply - evolvability %%%% Other General/Misc. Discussion: how to model rules (not just ontologies) -- for contracts and regulations aim for 2 or 3 use cases semantic processing architecture as part of the above repositories - standard contracts --> contingencies and state variables referenced - bankruptcy proceedings deontic: rights, obligations temporal integrating math with the logical integrating probabilistic embedding of contract instance into surrounding master agreements, law, business policies, business processes %% REA - Resource Event Actor - ontol of transactions [Mike B.] by Bill McCarthy Mich. State U. relationships [Mark Flood] ideally use cases from both equity, debt, hedging loans not nec'ly trades -- eg esp. at banks with housing strong need for defeasibility to represent contracts and change - dynamics over time, of info state - conciseness and evolvability of specification negotiated settlements and workouts AIG melt was a case of regulatory arbitrage, insuff capital reserve analytical use case - both microprudential and macroprudential - state-contingent cash flows %% extract relevant state-space dimensions from contracts and transations - eg contract depends on 30-day Libor or Deutschmark exchange rate - attributes - key events, eg associated with maturation, settlement, trouble in flavors Allan Mendelowitz: can reduce cash flow patterns to only ~30 standard contract types - 99% of contracts - with high level of precision happy path vs. trouble cases there are possible copyright issues wrt non-standard contracts - tho' there's a good argument for fair use if came fm req'd disclosure to govt - probably not a critical issue initially, but post-ers must beware could design better automated contracts - for simulation - for deployed implementation in the business etc. world subsidiaries: there are various scopes of credit support by the parent "wildcard" in practice: escalated workflows by the players efficient large-scale methods for expressive rules/ontologies bit of recap: - understand reality as it evolves - understand relationships/interconnections in overall fin ecosys - expression in semantic language - as part of the above, granular/modular development of building blocks . fundamental contract abstractions and types o eg equity = ownership; debt = obligation; hedge is compound . cash flow patterns and . state space attributes . key events . key contingencies - as part of the above, shared community repositories %%%%