trust, privacy (Breakout session 8)
sec'y Benjamin Grosof; moderator Leora Morgenstern

Richard Anderson:  apparently a topic of research?:
- rights/shareab for data that OFR will collect
- San Cannon:  what public-use subsets / obfuscations for researchers
- Benjamin Grosof:  need to operationalize the


Benjamin:
privacy "inference problem" (San Cannon:  a.k.a. "mosaic")
- eg opp to reverse engineer a firm's investment strategies from
     trades data, eg OFR will get data by trader desk not just firm
- basic approach is to actually try extracting answers to queries
  of interest over an integrated repository

Louiqa:
previous efforts to learn from, that have dealt with similar issues:
- Census
- health care, HIPAA

Louiqa: research topic:
new kinds of data sources,
- eg crowdsource mortgage data (Zillow-ish?)

Pete Kyle:
issue:  property rights to data, eg by exchanges or dealers
- Leonard Nakamura:  there is murky law, eg wrt use of URIs;
                     "sweat of brow" case precedents

Leonard Nakamura:  topic for research:
economic anal of value of these property rights

Charles Taylor:  OFR has broad powers to define its relevance scope
of data to demand/collect
- but they cannot collect from financial infrastructure, eg
  DTCC -- whose members however have agreed to since they have to
  provide individually
- bank examiners already work with very restricted rules
- unclear what powers OFR will have to share its confidential data
  more broadly with other regulators
- unclear if OFR collects directly whether it will cut out private
  data providers (in a way that's prohibited or legally actionable)

Louiqa: research topic:  ramifications of the fact
- pro forma definitions -- therefore ontologies -- can themselves be
confidential
- Pete:  or a new (kind of) derivative, with pro forma valuation
- Benj:  distinguish voluntary portion of the disclosure/reporting

Andrei Kirilenko: research topic:  map of data feeds
construct a map/"geography" of who provides data,
where do they get it from and how, who owns it;
incl. aggregation aspect
- very immediate issue in conn. w/ data repositories, data formats
  . upcoming event:  rules soon to be issued in 90-180 days by
    CFTC (Commodities Future Trading Corp.);
    sim. process by SEC in progress

Michael Wellman:  research topic:  incentives
incentives of value-added info products that there might be a market for
- Chester Spatt:  this is impt, eg use automated (econ) mechanism design
  . also:  incent disclosures in form that's more useful to others
- Michael Donnelly:  eg data is binding in some sense
- Louiqa:  can approach via multiple sources of data
- San Cannon: research topic:  study correlations of accuracy / hiding
  to when info is submitted, eg liars report early vs. late

Michael Wellman:  research topic: control and audit of how info is used,
that it's used only for its intended/permitted purpose(s)
- more about usages of data than authority to collect
- it's about accountability
- eg bookie discloses their illegal income to IRS, since knows that
  they are institutionally constrained from sharing that with the FBI

Leonard Nakamura:
- stress tests in banks/etc. -- need to be able to drill down to
compare regulator #s to the bank #s, not just take the bank #s

Pete Kyle:  research topic:
need to have meta-data about trust level of accuracy of the data in
sense of correspondence to reality,
- eg liar loans without proper documentation
- eg need provenance on prices that come from dealer valuation model,
  or where dealer is suspect in some other way
- Leora M:  research topic:  how things can go wrong -- is a big area
- Louiqa:  can borrow from life sci's work on mult ways to categorize
  a prediction eg about biological function there
- Andrei Kirilenko:  timestamps are often kinda meaningless or bogus,
  . eg some from backend clearance providers have been aggregated,
  and are off by say 30-40 sec;
  . eg lack of synchronization of clocks among order receipt and execution
    steps
  . Louiqa:  can approach lags via comp sci techniques
- Leonard Nakamura:  inaccurate data is of prime importance to regulators
  . bad data often indicates their processes are broken/failing
  . ... or that it's suspicious, eg when departs a lot from some avg
  . research topic:  how to help data entry be more accurate, eg detect
    inconsistencies at that time

Charles Taylor:  research topic:  extend accuracy and incentives to
rest of how the financial system works, not just capital markets
- eg very important:
  underwriting standards decline underlay much of the credit crisis
  . residence appraisals, mortgage-backed ratings
- what can one do to detect; delays; broken processes
- how to get beyond impressionistic and anecdotal
- Benjamin:  trust is the other side of leverage

% recap of research topics:

confidentiality
- masking and aggregation
- privacy inference problem
  . investment strategies
  . sandbox for investigations
- policy repn
  . for rights/trust mgmt
  . eg use rule-based with deontic, similar to repn of regulation
- ontologies themselves can be confidential -- what are the implications?
  . eg pro forma, valuation models, new kinds of derivatives

accuracy
- incentives for accuracy
- types of inaccuracy
- examples
  . underwriting; eg liar loans, rater corruption
  . stress tests
  . timestamping of trades
- forensics
  . data mining -- learn patterns of the deceivers or the troubled
  . eg wash trades to create apparent bogus price to report to risk mgers

map of data feeds
- provenance
- aggregation

property rights
- value of info
- incentives for usefulness of info
- economics-heavy more than KR-heavy

%%