← Agent Frameworks 🕐 36 min read
Agent Frameworks

Inferring Hedge Fund LP Relationships from Public Data (2026)

Part 1 — Direct Public Evidence Sources · March 2026

Hedge fund LP lists are private by design.

Hedge fund LP lists are private by design. Partnership agreements routinely include confidentiality clauses restricting LPs from disclosing fund-level information, and GPs routinely resist disclosure. Yet the US regulatory and state-law framework creates dozens of leakage points. Public pensions are subject to sunshine laws. Foundations file 990s. Multi-employer plans file 5500s. Every hedge fund raising capital under Regulation D files a Form D. Large fund advisers file Form PF quarterly. The task is assembling these fragments into a confidence-scored inference graph — not a confirmed LP list, but a probabilistic map that gives a sales, research, or compliance team a prioritized starting point.

This document is a practitioner playbook. It covers each source class, the specific evidence each source produces, template language for FOIA requests, known exemption defenses and how to contest them, a SQLite schema for the inference graph, confidence weights for each source type, and the legal and consultant-relations constraints that bound the entire exercise.

1.1 FOIA to Public Pension Systems

Public pension systems that manage assets on behalf of state and municipal employees are government entities subject to state public records laws. Unlike corporate or private endowments, they cannot simply decline to respond. The question is not whether they must respond, but how much they can withhold under applicable exemptions — and that varies materially by state.

1.1.1 State-Level FOIA Strength for Investment Data

California — California Public Records Act (CPRA)

California is the strongest jurisdiction for LP inference work. Cal. Gov. Code § 6254.26 explicitly addresses alternative investments held by public investment funds. The default is disclosure; the carve-out is narrow. Specifically:

  • Names of investment funds and vehicles are generally disclosable.
  • Terms of individual investments (fee rates, carry structures, specific side letter terms) are exempted if the GP demonstrates competitive harm from disclosure — but the burden is on the GP, not the requester.
  • Rates of return are disclosable.
  • “Investment descriptions” — which include fund name, vintage, commitment amount, and current valuation — are disclosable unless a specific trade-secret claim is upheld.

CalPERS and CalSTRS both publish comprehensive investment reports proactively (see section 1.1.3). FOIA requests to CalPERS typically receive responses within 10 business days for routine investment schedule requests.

Washington — Public Records Act (PRA), RCW 42.56

Washington State Investment Board (WSIB) is one of the more forthcoming pension systems in the US. Washington’s PRA has a strong presumption of disclosure, and courts have consistently rejected broad trade-secret claims by GPs. WSIB proactively publishes its full investment schedule including alternative investments. FOIA requests to WSIB for the “Alternative Investment Program Schedule” are routinely fulfilled. Turnaround is typically 5–10 business days for structured records.

Oregon — Oregon Public Records Law, ORS 192

Oregon is a complex jurisdiction. The Oregon Investment Council (OIC) has partially limited disclosure after a 2003 rule change allowing greater GP confidentiality claims. However, Oregon courts have applied a balancing test: public interest in disclosure of fund-level details must outweigh the commercial harm to the GP. Fund names and commitment sizes have generally survived this test; specific fee terms often have not. OIC publishes its complete alternatives investment schedule on its website, updated quarterly. FOIA requests to Oregon are subject to a 5-business-day response timeline, with possible extension to 10 days.

Connecticut — Freedom of Information Act (CONN-FOIA), CGS § 1-200 et seq.

Connecticut uses the FOIA acronym directly. The Connecticut Retirement Plans and Trust Funds (CRPTF) manages approximately $45 billion. Connecticut’s FOIA law is moderately strong. The state has a four-part test for whether a non-governmental entity is subject to FOIA, but CRPTF itself is a governmental entity and fully subject. Connecticut courts have ruled that investment committee meeting minutes and manager selection records are public. Fee structures and side letter terms may be withheld under the “financial records” exemption (CGS § 1-210(b)(5)) if the agency can demonstrate that disclosure would harm the pension’s competitive position — a high bar.

Texas — Texas Public Information Act (TPIA), Tex. Gov’t Code § 552

Texas is a jurisdiction where GPs have achieved significant exemption wins. The Texas Teacher Retirement System (TRS) and Employees Retirement System (ERS) are subject to TPIA, but Texas law includes a broad “competitive bidding” exemption and allows agencies to request Attorney General opinions before disclosure. In practice, fund names often survive disclosure; fee terms rarely do. GP confidentiality provisions are taken seriously by Texas AG opinions. TPIA has a 10-business-day response requirement, but complex investment records often trigger the AG-opinion pathway, extending timelines by 30–60 days.

New York — Freedom of Information Law (FOIL), Public Officers Law § 84 et seq.

New York State Common Retirement Fund (NYSCRF) is subject to FOIL. New York’s approach has evolved: the state explicitly enacted legislation requiring disclosure of investment performance by asset class, and NYSCRF publishes comprehensive annual reports. The trade-secret exemption under FOIL (§ 87(2)(d)) applies to information “which if disclosed would cause substantial injury to the competitive position of the subject enterprise” — courts have applied this narrowly to specific deal terms, not fund names. Fund names and commitment amounts from NYSCRF are consistently disclosable. Turnaround is typically 20 business days with possible extension.

New Jersey — Open Public Records Act (OPRA)

New Jersey Division of Investment (NJ DoI) manages approximately $100 billion including significant hedge fund and alternatives allocations. OPRA is generally strong, with a 7-business-day response window. New Jersey has been a target of significant FOIA litigation by investigative journalists regarding private equity fees, and courts have generally required disclosure of fee and performance data. Fund names and commitment amounts are routinely produced.

North Carolina — Public Records Law, G.S. 132

North Carolina Retirement Systems is a strong FOIA jurisdiction for investment data. The state has legislated that alternative investment records are presumptively public, with a narrow carve-out for specific trade secrets proven to cause competitive harm. Fund names, commitment amounts, and current valuations are disclosable. NC publishes a detailed alternatives investment report annually.

1.1.2 FOIA Template Language

The following template language is tested and designed to maximize the scope of disclosure while remaining specific enough to avoid triggering broad-search objections. Adapt by jurisdiction.

Subject: Public Records Request — Schedule of Alternative Investments

Pursuant to [California Public Records Act / Washington Public Records Act / Oregon Public Records Law / Connecticut Freedom of Information Act / Texas Public Information Act / New York Freedom of Information Law — select applicable], I request the following records:

  1. The complete Schedule of Alternative Investments (or equivalent document describing the pension fund’s investments in hedge funds, private equity funds, private credit funds, and other pooled investment vehicles), for the most recent fiscal year for which records are finalized, and the two preceding fiscal years.

  2. Investment committee meeting minutes and board materials for any meeting at which an alternative investment manager was approved, terminated, or subject to a performance review, for the period [specify 3-year window].

  3. Any “Manager Roster,” “Approved Manager List,” “Watch List,” or equivalent document maintained by investment staff describing current or recent alternative investment relationships.

  4. For each alternative investment identified in items 1–3: the name of the investment fund, the name of the general partner, the fund vintage (if applicable), the total committed capital, the amount drawn to date, the current net asset value as of the most recent valuation date, and the fund’s internal rate of return (or equivalent performance metric) as reported by the GP or calculated by the pension’s investment staff.

I am not requesting proprietary investment terms, side letter provisions, management fee rates, carried interest rates, or investment strategy descriptions that the GP has designated as confidential. I am requesting only the identifying information and performance metrics described above, which in [California / Washington / New York — name the specific statutory provision] are expressly subject to public disclosure.

If any portions of this request are denied, please provide a specific statutory citation for each exemption claimed, a description of the withheld information sufficient to evaluate the exemption, and whether severance and partial disclosure of non-exempt portions is possible.

Notes on this template:

  • The explicit exclusion of fee terms is strategic: it narrows the trade-secret target, forcing the GP to argue that the fund’s mere name or commitment size is a trade secret — a weak position courts have generally rejected.
  • Requesting the prior two fiscal years is important because pensions sometimes rotate out of hedge fund allocations; a current-year schedule may omit managers where inference is still valuable.
  • In Texas, this request should be accompanied by a good-faith statement that the requester is not a competitor of any described investment manager — this can reduce the likelihood of an automatic AG-opinion referral.

1.1.3 Proactive Publishers — No FOIA Required

The following pension systems publish alternative investment data proactively and are the fastest starting points for LP inference:

System AUM Publication Update Frequency URL
CalPERS ~$530B Investment Reports, Private Equity & Absolute Return portfolios Quarterly calpers.ca.gov/investments
CalSTRS ~$350B Investment Portfolio, manager listings Semi-annual calstrs.com/investment-portfolio
WSIB (Washington) ~$200B Alternative Investment Program Schedule Quarterly sib.wa.gov
NYSCRF ~$265B Annual Investment Report Annual osc.ny.gov
NJTRS/NJ DoI ~$100B Investment Division Annual Report Annual state.nj.us/treasury/doinvest
NC Retirement Systems ~$120B Alternatives Investment Report Annual nctreasurer.com
OTRS (Oklahoma TRS) ~$20B Investment Reports Quarterly otrs.ok.gov
TRS Illinois ~$70B Comprehensive Annual Financial Report Annual trs.illinois.gov
NYC Fire/Police/Teachers Combined ~$270B Investment Committee Agendas Monthly board meetings comptroller.nyc.gov

CalPERS is the single most valuable proactive publisher for hedge fund LP inference because (a) it has historically maintained an “Absolute Return Strategies” portfolio that names specific hedge fund managers and commitment sizes, (b) board meeting materials are posted 10 days before each meeting and include new manager approvals and redemptions, and © the investment staff publishes an annual review of absolute return strategies that includes strategy-level and manager-level performance attribution.

Practical note on NYC: The New York City pension systems (NYCERS, Teachers, Police, Fire, Board of Ed) hold monthly investment committee meetings whose agendas and packages are posted publicly by the NYC Comptroller’s office. These packages routinely include manager approval memos that name the specific fund, commitment size, and investment rationale. This is among the highest-signal proactive disclosure in the US — meeting packages are typically posted 5–7 days before the meeting date.

1.1.4 Turnaround Times and Common Exemption Claims

Typical FOIA turnaround by state:

State Statutory Deadline Practical Turnaround Extension Mechanism
California 10 days to respond; 14 additional if complex 10–30 days Written extension notice required
Washington 5 business days to respond 5–15 days No extension; must provide installments
Oregon 5 days; 10 if unable to respond 10–20 days One 10-day extension allowed
Connecticut 4 business days 5–15 days Hearing process available
Texas 10 business days 10–60 days AG opinion referral can extend indefinitely
New York 5 business days to acknowledge; 20 to respond 20–45 days “Unusual circumstances” extension
New Jersey 7 business days 7–20 days 7-day extension with written notice

Common exemption claims by GPs and how to contest them:

  1. “Trade secret” claim on fund name and commitment size. GPs argue that knowing which public pension has invested in their fund is commercially sensitive because competitors can use it to pitch those same LPs. Courts have generally been skeptical of this claim when the information is restricted to fund name and commitment size without fee terms. Contestation strategy: cite cases in requester’s jurisdiction where the same argument was rejected (e.g., Los Angeles Times v. CalPERS line of cases; New Jersey Appellate Division precedents). Request the AG or counsel confirm that the fund name itself — not the terms — is the alleged trade secret.

  2. “Competitive position” harm claim by the pension itself. Some pension investment offices argue that disclosing which funds they are in will cause GPs to restrict their access to future funds. This argument has generally failed in California and Washington courts. The pension’s obligation to transparency to its beneficiaries is held to outweigh speculative competitive harm.

  3. GP contractual confidentiality clauses. Partnership agreements routinely include provisions that the LP (the pension) will not disclose “fund-level information.” Courts have repeatedly held that a pension fund cannot contractually waive its statutory disclosure obligations — the contractual clause is void as against public policy. Key citation: Binkley v. Conte (CT FOIA Commission, 2008) and the Ropes & Gray survey of 2014 noting that state FOIA laws generally override LP confidentiality provisions.

  4. “Preliminary deliberations” exemption. Investment staff deliberations before a final vote are sometimes withheld under deliberative-process exemptions. Strategy: request only finalized records (post-vote investment committee minutes and commitment letters) and the investment schedule, which are operational records rather than deliberative.

1.2 Form 5500 Schedule D — Multi-Employer and Taft-Hartley Plans

Form 5500 is filed annually with the Department of Labor by ERISA-covered employee benefit plans. For multi-employer (Taft-Hartley) plans — covering union workers in industries like building trades, trucking, healthcare — Schedule D is particularly valuable.

What Schedule D contains:

Every plan that invests in a Common/Collective Trust (CCT), Pooled Separate Account (PSA), Master Trust Investment Account (MTIA), or a 103-12 Investment Entity (103-12 IE) must attach Schedule D and list each such vehicle by name, EIN (if available), and type code. A 103-12 IE is the Form 5500 classification for a pooled investment arrangement where 10 or more unrelated plans invest together — this is the vehicle class that includes most hedge funds accessible to pension plans.

Why this matters for LP inference:

When a hedge fund registers as a 103-12 IE and files its own Form 5500 as a Direct Filing Entity (DFE), the DFE filing lists the names of participating plans. This creates a direct, legally mandated disclosure of the LP relationship — not an inference, an actual record. The DFE Form 5500 typically identifies:

  • The fund name (as registered with DOL)
  • The EIN of the fund
  • The list of participating plans (by plan name and EIN)
  • Aggregate assets

How to access Form 5500 data:

The DOL’s EFAST2 system (efast.dol.gov) allows free public search of Form 5500 filings. The DOL also publishes annual research files in structured format. ProPublica maintains a Form 5500 database accessible at projects.propublica.org/nonprofits (primarily for 990s) but the DOL data is the primary source. The Form 5500 Direct Filing Entity Research File published by DOL’s EBSA provides structured data on all DFEs including 103-12 IEs.

Practical search strategy:

  1. Search EFAST2 for the hedge fund’s known legal entity name or a variant. If the fund is accessible to Taft-Hartley plans, it may have filed as a DFE.
  2. If found, download the full filing — Schedule H lists plan assets; participating plans may be listed in attachments.
  3. Cross-reference the plan EINs found in the DFE filing with the EFAST2 plan search to identify the union or multi-employer plans invested.
  4. Cross-reference those plans with union websites, collective bargaining agreement records, and DOL plan descriptions to build an LP node.

Limitation: Not all hedge funds accessible to institutional investors are structured as 103-12 IEs for ERISA purposes. Many serve public pension plans and endowments through limited partnership structures that are outside the ERISA reporting framework entirely. Form 5500 / Schedule D is most useful for union funds, not corporate or public pension plans.

1.3 IRS Form 990 — Endowments and Foundations

Private foundations and nonprofit endowments file Form 990 or Form 990-PF annually with the IRS, and these filings are public records. They are the primary source for inferring hedge fund LP relationships among university endowments, hospital systems, and family foundations.

Where LP signals appear in Form 990:

  1. Part IX — Statement of Functional Expenses: Investment management fees are broken out. A line item reading “Investment management fees — alternative assets” with a dollar amount can signal the scale of hedge fund allocation, but does not name managers.

  2. Schedule D — Supplemental Financial Statements, Part VII (Investments — Other Securities): Foundations are required to list investments in publicly traded securities, but “other investments” (including LP interests in hedge funds) are listed in aggregate at book value. Some foundations list individual fund names; most do not, but large foundations sometimes provide a description column that names the strategy or manager.

  3. Form 990-PF — Part II, Line 13: Private foundations list investments in closely held entities separately from publicly traded securities. An LP interest in a hedge fund is a closely held investment. Some foundations disclose the fund name; others use generic descriptions. When foundations name the fund, this is high-confidence LP evidence.

  4. Related entity disclosures: Some foundations disclose related-party investment relationships in Schedule R, which can reveal investment partnerships with shared governance.

ProPublica Nonprofit Explorer:

ProPublica’s Nonprofit Explorer (projects.propublica.org/nonprofits) provides free full-text and structured access to over 1.8 million Form 990 filings since 2013, with page-image access back to 2001. The API is free to access without a key for reasonable query volumes.

Practical search strategy:

  1. Identify the target fund’s likely investors among major endowments (Ivy League universities, large hospital systems, community foundations).
  2. Search ProPublica for the institution name.
  3. Download Schedule D and Part IX from the most recent 3 years of filings.
  4. Look for LP interest descriptions, investment management fee recipients, and closely held investment entity names in Form 990-PF Part II.
  5. Cross-reference with known fund names from other sources (Form D, press coverage).

Limitation: The 990 filing reflects the fiscal year end of the filer, which for universities is often May 31. There is a 6–15 month disclosure lag from fiscal year end to public availability on ProPublica. A 2024 fiscal year filing may not appear until late 2025. This makes 990 data backward-looking but still valuable for confirming inferred relationships.

1.4 SEC Form D — Regulation D Filings

Under Regulation D (Rules 504 and 506), any private offering of securities that has not been registered with the SEC must be reported via Form D within 15 days of the first sale. Hedge funds raising capital under Rule 506(b) or 506© — which covers nearly all institutional hedge fund offerings — must file Form D on EDGAR and file annual amendments as long as the offering is ongoing.

What Form D discloses:

  • Fund legal name and GP entity name
  • Date of first sale
  • State of first sale (signals where the first LP is domiciled)
  • Total offering amount (GP’s target raise)
  • Total amount sold to date (cumulative capital raised across all annual amendments)
  • Number of investors (integer count, no names)
  • Issuer type (pooled investment fund)
  • Exemption relied upon (506(b) vs 506©)

LP inference value:

Form D does not name LPs — this is its fundamental limitation for direct inference. However, it enables several useful analytical moves:

  1. Fundraising timeline reconstruction: The sequence of annual amendments to a Form D shows the pace of capital raising. A fund that raised $0 → $200M → $800M → $1.5B → $1.5B (no change) over four amendments suggests a two-year fundraise followed by a soft close. Cross-referencing this timeline with known public pension board meeting approval dates can confirm the timing of a specific LP’s entry.

  2. State of first sale correlation: The state recorded on the initial Form D is the state where the first LP is located. If the state is Connecticut and the fund is a multi-strategy fund, the first LP is likely a Connecticut-domiciled institution (CRPTF, Yale endowment, a Connecticut-domiciled family office, or a Taft-Hartley plan). Combined with other signals, this narrows the field.

  3. Investor count bracket: Form D reports the exact number of investors (as an integer). If a fund has 23 investors and you can identify 19 through FOIA and 990 data, you know you are missing 4 LPs. This disciplines the inference process.

  4. Multiple series / parallel funds: Large managers often file separate Form Ds for each vehicle (onshore LP, offshore Cayman feeder, managed account). The number and structure of parallel Form D filings reveals whether the fund accepts managed account investors — a signal relevant for identifying large sovereign or pension allocators who typically invest via managed accounts.

Accessing Form D on EDGAR:

SEC EDGAR full-text search (efts.sec.gov) allows keyword search of Form D filings. The EDGAR company search also allows lookup by fund name. The SEC’s Division of Corporation Finance publishes aggregated Regulation D offering statistics at sec.gov/data-research/statistics-data-visualizations/regulation-d-offerings.

1.5 SEC Form PF — Private Fund Reporting

Form PF is filed quarterly (for large advisers) or annually (for smaller advisers) by investment advisers to private funds registered with the SEC. The filing threshold is $150M in private fund AUM for any filing; the “large hedge fund adviser” threshold — triggering quarterly filing with enhanced detail — is $1.5B in hedge fund AUM.

What Form PF discloses:

Form PF does not identify LPs. It is filed with the SEC and FSOC (Financial Stability Oversight Council) and is used for systemic risk monitoring. Key data available to a researcher:

  • Strategy classification: Section 1 requires advisers to categorize funds by strategy type (equity, macro, event-driven, relative value, credit, multi-strategy). This is the primary public source for classifying a fund’s strategy when press coverage is absent or ambiguous.
  • Fund-level AUM and number of investors: Aggregate assets and the number of beneficial owners (not named) per fund.
  • Leverage ratios: Gross notional leverage and net notional leverage by strategy category. This is useful for matching a fund’s risk profile to a pension’s investment policy constraints.
  • Liquidity terms: Redemption notice period ranges and lock-up structures by strategy. A fund with a 3-year lock-up and quarterly redemptions after the lock will have a different LP profile than one with monthly liquidity — this narrows the LP universe to institutions with the appropriate liquidity tolerance.
  • Concentration: Positions in the portfolio that exceed 5% of fund NAV, reported as count (not named).
  • Prime broker relationships: Listed by name in some sections (2025 amended form requirements).

Accessing Form PF:

Form PF filings are confidential and not publicly available on EDGAR. The SEC does not publish individual Form PF filings. Aggregate statistics derived from Form PF are published by the SEC in their annual Private Fund Statistics report (sec.gov/divisions/investment/private-funds-statistics). These aggregate reports provide market-level context but not fund-level data useful for LP inference.

Practical use of Form PF for LP inference:

The indirect value of Form PF is in strategy classification. By combining (a) a hedge fund’s EDGAR-registered investment adviser Form ADV (which is public and contains strategy descriptions, AUM, and number of clients), (b) press coverage of strategy focus, and © Form PF aggregate statistics for the strategy category, a researcher can classify a fund into a strategy bucket that maps to pension investment policy constraints. Many public pensions publish investment policy statements listing which strategies they are approved to allocate to — matching strategy classification to policy constraints narrows the plausible LP universe before FOIA outreach.

Note on 2024/2026 Form PF amendments: The SEC adopted amended Form PF requirements in February 2024, with original compliance by March 12, 2025. The compliance date for certain enhanced disclosures (including enhanced section 1 reporting on leverage, liquidity, and portfolio concentration) was subsequently extended to October 1, 2026. Researchers should note that fund filings between March 2025 and October 2026 may reflect partial compliance with the amended form.

1.6 Form ADV — Investment Adviser Registration

While not listed in the original brief, Form ADV (Parts 1 and 2) is a foundational source that should be queried before any other SEC form. Filed by all registered investment advisers, Form ADV Part 1 discloses:

  • Legal name, AUM, number of clients
  • Client types as percentage of AUM (e.g., “40% pooled investment vehicles; 30% pension and profit-sharing plans; 15% endowments and foundations”)
  • Ownership structure and key personnel
  • State registrations (useful for geographic LP inference)
  • Related persons and affiliated entities

The “pension and profit-sharing plans” percentage combined with total AUM gives a direct estimate of pension capital in the fund. This is a numerical anchor for LP inference — if a $5B fund reports 40% pension plan assets, approximately $2B of the fund is pension capital across all LP relationships, constraining how many and which pension LPs could account for that amount.

Form ADV is publicly available at adviserinfo.sec.gov. Annual amendments are filed within 90 days of fiscal year end.

1.7 13F Cross-Reference Analysis

Form 13F is filed quarterly by institutional investment managers with $100M+ in assets under management, disclosing all long US-listed equity positions. Both the pension fund and the hedge fund (if it holds long equity positions) may file 13Fs. The cross-reference technique exploits this overlap.

The technique:

If a public pension’s 13F shows a position in, say, 200 specific mid-cap equity names that closely overlap with the known portfolio of a quantitative equity fund (as inferred from the fund’s Form ADV strategy description, press coverage, and academic papers by the fund’s principals), and this overlap persists across 4+ consecutive quarters with a consistent lag (suggesting the pension holds a managed account that mirrors the fund’s strategy), this is circumstantial evidence of an allocation via managed account.

Practical constraints:

  • 13F filings have a 45-day lag from quarter end, so position data is always at least 6 weeks stale.
  • 13F covers long US-listed equity only — no shorts, no derivatives, no fixed income, no international equities.
  • For multi-strategy or macro funds with limited long US equity exposure, 13F cross-reference yields minimal signal.
  • For fundamental long/short equity or quantitative long-short equity managers, the technique is most applicable.

Implementation:

  1. Download the target pension’s 13F history from EDGAR.
  2. Download the Form ADV strategy description for the target hedge fund to understand expected portfolio characteristics.
  3. Use EDGAR’s full-text search to find the hedge fund’s own 13F if the fund holds long US equity above the $100M threshold.
  4. Compute portfolio overlap metrics (Jaccard similarity of position sets, sector weight correlation) across time.
  5. A stable high-overlap signal (>60% position overlap, >0.85 sector correlation over 6+ quarters) combined with a consistent quarter-lag is strong circumstantial evidence of a managed account relationship.

Part 2 — Indirect Inference Signals

2.1 RFP and Manager Search Announcements

Public pensions are required to conduct competitive manager searches for new allocations above certain thresholds. These searches are published as RFPs (Requests for Proposals) or RFQs (Requests for Qualifications) on the pension’s website and sometimes on state procurement portals.

Signal value: An RFP signals upcoming allocation, not confirmed allocation. It narrows the field to pensions actively in-market for a strategy — useful for prospecting but not for confirming current LP relationships.

Where to find RFPs:

  • Pension system websites (typically an “Investments” or “Procurement” section)
  • State procurement portals (e.g., BidSync, DemandStar, state-specific portals)
  • NEPC and Callan publish “manager searches in progress” summaries to their GP subscribers (paid service)
  • P&I’s manager search tracker (subscription required but some data appears in news)

FOIA for board meeting minutes: Investment committee meeting minutes are public records in most states and routinely include manager search discussions, finalist selections, and approval votes. Requesting the last 2 years of investment committee minutes from a target pension is a high-value FOIA request that often surfaces both current allocations and the decision trail.

2.2 Investment Consultant Reports and Approved Manager Lists

The five dominant institutional investment consultants — Mercer, NEPC, Callan, Cambridge Associates, and RVK — control access to the majority of public pension and endowment investment decisions. They maintain internal databases of approved, rated, or watch-listed hedge fund managers, and their ratings function as de facto gatekeepers for new allocations.

What is publicly available:

  • Annual consultant surveys by P&I (published each spring) list the consultants’ total AUA and client counts by client type.
  • Consultant RFP responses to public pensions, when they win a consulting contract, are sometimes public records — including their proposed manager rosters.
  • Some consultants publish quarterly “manager commentaries” or “market intelligence reports” that are circulated to their GP relationships and sometimes surface in SEC filings or LP due-diligence materials.
  • Mercer’s hedge fund ratings are occasionally referenced in manager search committee memos that appear in public pension board packages.

The consultant-gatekeeper dynamic:

Being on a consultant’s approved or rated list is a prerequisite for accessing their client base. NEPC has a formal “approved manager” tier that clients see in their search universe. Mercer uses a tiered rating system (A/B/C/D/U) that gates manager inclusion in client searches. Cambridge Associates maintains a “recommended” and “approved” list by strategy. Callan, by contrast, does not maintain a formal approved list — it screens all managers in its database for every search, which makes Callan clients somewhat more accessible to un-rated managers.

Inference use case: If a consultant’s report (obtained through a public pension’s FOIA production or surfaced in a board package) lists “recommended managers” in a specific strategy, and the target fund appears on that list, the consultant’s client base becomes the plausible LP universe for that fund.

2.3 Conference Speaker and Panelist Inference

Institutional investor conferences (ILPA Annual, PAAMCO Symposium, Milken Global Conference, IMN institutional investor series) sometimes pair pension CIOs with hedge fund PMs on panels focused on specific strategies.

Signal value: Weak. A co-appearance on a panel does not imply an investment relationship — conferences are organized by topic, not by portfolio. However:

  • If a pension CIO is repeatedly seen at a specific manager’s annual LP day (often invitation-only events whose speakers lists surface on LinkedIn or event pages), this is a stronger signal.
  • If a pension CIO publicly references a specific fund in a conference interview or published Q&A, this is near-confirmatory.
  • Conference appearances as a standalone signal warrant a confidence score of 0.20–0.30.

2.4 Press Coverage — Institutional Investor Trade Press

The institutional investor trade press is the highest-confidence non-FOIA source for confirmed LP relationships. Key publications:

  • Pensions & Investments (P&I): Reports mandate wins, redemptions, and RFP outcomes. P&I’s “Pension Fund Investment Trends” coverage routinely names the pension, the fund, and the commitment size when the information is sourced from a board meeting package or disclosed by a pension spokesperson.
  • FundFire (Financial Times professional): Focused on asset management, frequently reports specific mandates with named LP and named fund.
  • Institutional Investor (II): Publishes ranked lists (Hedge Fund 100, etc.) and sometimes reveals LP relationships in profiles.
  • Bloomberg News (Hedge Funds section): Reports large mandate wins and redemptions, typically for funds above $5B AUM.
  • The Wall Street Journal (Investing section): Covers high-profile hedge fund LP events (large withdrawals, blowups, major inflows).

Practical search strategy:

  1. Search P&I’s full archive (subscription required, $300/yr for individual access) for “[fund name] pension” or “[fund name] allocation.”
  2. Search Google News for “[fund name] [pension name] allocation commit invest” restricted to the last 3 years.
  3. Compile a press-sourced LP list with date, source publication, and dollar amount where available.
  4. Assign confidence 0.85–0.95 to press-confirmed relationships depending on specificity (named reporter, named pension spokesperson vs. unnamed source).

2.5 LinkedIn Job Posting Signals

Hedge funds actively raising capital from pension funds post investor relations roles with explicit LP segment language. A job posting for “Institutional Sales — Public Pension Coverage” or “Investor Relations Associate — Taft-Hartley and Corporate Pension” signals:

  • The fund is in active fundraising mode with pension LPs.
  • If the fund already has a dedicated pension IR person and is hiring another, it suggests growth in the pension LP segment.

This signal is forward-looking (fundraising intent) rather than confirmatory, and warrants a confidence weight of 0.30–0.40 for “has or is seeking pension LP relationships.”

LinkedIn’s Sales Navigator provides job posting history, which can reconstruct a timeline of fundraising activity by role posting date.

Part 3 — Building the Inference Graph

3.1 Conceptual Model

The LP inference graph is a weighted directed graph where:

  • Nodes are: HedgeFund, PublicPension, Endowment, Foundation, TaftHartleyPlan, FamilyOffice, SWF (sovereign wealth fund)
  • Edges represent IS_LP_OF relationships with attributes: confidence (0.0–1.0), source (inference method), dollar_amount (nullable), vintage (nullable), evidence_date, evidence_url
  • Evidence nodes are linked to edges via a separate SUPPORTED_BY relationship, allowing multiple evidence items to support or update a single LP inference

3.2 SQLite Schema

-- Entities
CREATE TABLE fund (
    id          INTEGER PRIMARY KEY,
    name        TEXT NOT NULL,
    legal_name  TEXT,
    gp_entity   TEXT,
    form_d_id   TEXT,        -- EDGAR accession number of Form D
    adv_crd     TEXT,        -- CRD number from Form ADV
    strategy    TEXT,        -- e.g. 'long_short_equity', 'macro', 'multi_strategy'
    aum_estimate_mm REAL,    -- USD millions, most recent estimate
    aum_source  TEXT,
    created_at  TEXT DEFAULT (datetime('now'))
);

CREATE TABLE lp_entity (
    id          INTEGER PRIMARY KEY,
    name        TEXT NOT NULL,
    entity_type TEXT NOT NULL CHECK (entity_type IN (
                    'public_pension', 'endowment', 'foundation',
                    'taft_hartley', 'corporate_pension', 'swf',
                    'family_office', 'other')),
    state       TEXT,        -- US state abbreviation, NULL for non-US
    country     TEXT DEFAULT 'US',
    aum_estimate_mm REAL,
    form_5500_ein TEXT,      -- EIN from Form 5500 if applicable
    form_990_ein  TEXT,      -- EIN from Form 990 if applicable
    created_at  TEXT DEFAULT (datetime('now'))
);

-- LP relationships with confidence scoring
CREATE TABLE lp_relationship (
    id              INTEGER PRIMARY KEY,
    fund_id         INTEGER NOT NULL REFERENCES fund(id),
    lp_entity_id    INTEGER NOT NULL REFERENCES lp_entity(id),
    confidence      REAL NOT NULL CHECK (confidence BETWEEN 0.0 AND 1.0),
    status          TEXT NOT NULL CHECK (status IN ('inferred', 'confirmed', 'disputed', 'redeemed')),
    commitment_mm   REAL,       -- USD millions committed, NULL if unknown
    vintage         INTEGER,    -- year of first allocation
    redemption_date TEXT,       -- ISO date if redeemed
    created_at      TEXT DEFAULT (datetime('now')),
    updated_at      TEXT DEFAULT (datetime('now')),
    UNIQUE(fund_id, lp_entity_id)
);

-- Evidence items linked to LP relationships
CREATE TABLE evidence (
    id                  INTEGER PRIMARY KEY,
    lp_relationship_id  INTEGER NOT NULL REFERENCES lp_relationship(id),
    inference_method    TEXT NOT NULL CHECK (inference_method IN (
                            'foia_direct',          -- FOIA production from pension
                            'proactive_disclosure', -- pension's own published schedule
                            'form_5500_schedule_d', -- DOL Form 5500 Schedule D / DFE
                            'form_990',             -- IRS Form 990 / 990-PF
                            'form_d_state_corr',    -- Form D state of first sale correlation
                            'form_adv_client_type', -- Form ADV client type % inference
                            '13f_cross_reference',  -- 13F position overlap analysis
                            'press_confirmed',      -- Named in P&I, FundFire, Bloomberg
                            'board_meeting_minutes',-- From public pension board minutes
                            'rfp_outcome',          -- RFP/RFQ award record
                            'consultant_report',    -- Named in consultant's approved list
                            'conference_copanel',   -- Conference co-appearance
                            'linkedin_ir_signal',   -- LinkedIn IR job posting signal
                            'peer_inference',       -- Inferred from similar LP peer
                            'other'
                        )),
    confidence_delta    REAL,   -- additive contribution of this evidence item
    evidence_date       TEXT,
    evidence_url        TEXT,
    raw_excerpt         TEXT,   -- verbatim excerpt supporting the inference
    notes               TEXT,
    created_at          TEXT DEFAULT (datetime('now'))
);

-- Audit log
CREATE TABLE inference_log (
    id              INTEGER PRIMARY KEY,
    relationship_id INTEGER REFERENCES lp_relationship(id),
    old_confidence  REAL,
    new_confidence  REAL,
    changed_by      TEXT,
    reason          TEXT,
    changed_at      TEXT DEFAULT (datetime('now'))
);

-- Indexes
CREATE INDEX idx_lp_rel_fund ON lp_relationship(fund_id);
CREATE INDEX idx_lp_rel_lp ON lp_relationship(lp_entity_id);
CREATE INDEX idx_evidence_method ON evidence(inference_method);
CREATE INDEX idx_evidence_relationship ON evidence(lp_relationship_id);

3.3 Confidence Scoring Framework

The following weights represent the epistemic value of each inference method. These are not additive probabilities — they represent the confidence assigned to the LP relationship if this is the sole evidence type. When multiple independent evidence items exist, confidence is updated using a Bayesian update: P(LP | E1 ∧ E2) = 1 − (1 − P(LP | E1)) × (1 − P(LP | E2)).

Inference Method Base Confidence Notes
foia_direct 0.95 FOIA-produced investment schedule naming the fund
proactive_disclosure 0.95 Pension’s own published schedule naming the fund
board_meeting_minutes 0.90 Approval vote in public board minutes
press_confirmed 0.85 Named journalist, named pension spokesperson
press_confirmed_unnamed_source 0.70 Press report citing unnamed source
form_5500_schedule_d 0.90 DFE filing lists the plan directly
form_990 0.80 990-PF Part II names the fund specifically
rfp_outcome 0.75 RFP award record published; fund named as awardee
form_adv_client_type 0.45 Client type % suggests pension AUM; no specific LP
13f_cross_reference 0.50 High position overlap over 6+ quarters
form_d_state_corr 0.35 State of first sale matches pension domicile
consultant_report 0.40 Named on consultant’s approved list for strategy
peer_inference 0.30 Similar-profile LP in same state confirmed; peer inferred
linkedin_ir_signal 0.25 Fund hiring IR for pension segment
conference_copanel 0.20 CIO and PM co-appeared on strategy panel

Applying the Bayesian update:

If a fund has a Form D with a first sale in Connecticut (confidence 0.35) and P&I reported the CRPTF is reviewing the fund (confidence 0.70 for “reviewing” vs “allocated”), and then a board meeting minutes FOIA produces an approval vote (confidence 0.90):

  • After Form D signal: P = 0.35
  • After P&I review signal (treating “reviewing” as 0.50 probability of eventual allocation): P = 1 − (1 − 0.35) × (1 − 0.50) = 1 − 0.325 = 0.675 (tentative)
  • After FOIA approval vote: P = 1 − (1 − 0.675) × (1 − 0.90) = 1 − 0.0325 = 0.97 → round to confirmed

3.4 Peer Inference Heuristic

Peer inference applies when a confirmed LP relationship exists and a similar institution in the same state and strategy-approval profile has not yet been directly confirmed. The heuristic:

Conditions for peer inference:

  1. Institution A (confirmed LP) and Institution B (candidate LP) are both public pension plans in the same US state.
  2. Both institutions share the same investment consultant (if determinable from public records).
  3. The target fund’s strategy falls within Institution B’s published investment policy statement (IPS) strategy approvals.
  4. Institution B’s AUM is within 3x of Institution A’s AUM.
  5. Institution B has not been confirmed as a non-LP (e.g., no public redemption notice, no “manager not approved” record in board minutes).

When all five conditions are met, assign peer_inference at confidence 0.30. If the same investment consultant is confirmed (conditions 1, 2, 3, 4, 5), boost to 0.40.

Example: CalPERS is confirmed LP of a multi-strategy fund via proactive disclosure. CalSTRS has the same investment consultant (hypothetically), similar strategy approvals, and is 2/3 the size. No CalSTRS FOIA has been returned yet. Peer inference → edge weight 0.35, status inferred.

3.5 Strategy Classification from Public Sources

Mapping a fund to a strategy category enables matching against pension investment policy constraints. Classification pipeline:

  1. Form ADV Item 5 “Advisory Activities”: The principal business activity description often includes strategy language.
  2. Form ADV Part 2A Brochure: Required to describe investment strategies; available on adviserinfo.sec.gov.
  3. Academic and press fingerprinting: Published papers by fund principals, media interviews, SEC examination notices in public enforcement orders.
  4. Form D Item 4 “Type of Fund”: Categories include “Hedge Fund” (undifferentiated) but some GPs add detail in the “Other” field.
  5. Form PF aggregate statistics (for market-level classification only): published by SEC at sec.gov/divisions/investment/private-funds-statistics.

Strategy taxonomy for the inference graph (maps to typical pension investment policy categories):

strategy Value Description Typical IPS Category
long_short_equity Fundamental or quant equity L/S Equity Alternatives
global_macro Discretionary or systematic macro Macro / Diversifying
systematic_futures Managed futures / CTA Diversifying / Real Assets
event_driven Merger arb, distressed, activism Event Driven Alternatives
multi_strategy Internal PM multi-strat Absolute Return
credit_ls Long/short credit Credit Alternatives
relative_value Fixed income arb, vol arb Relative Value
quantitative_equity Quant long/short, stat arb Quantitative Strategies

Permitted activities:

  • Filing FOIA/public records requests with any government agency, including public pension systems. This is a legal right under federal and state law. There is no restriction on volume, frequency, or purpose of public records requests unless the requester engages in demonstrably bad-faith conduct (extreme volume, harassment, requests for records that clearly do not exist).
  • Downloading and analyzing publicly filed SEC documents (Form D, Form ADV, 13F) from EDGAR. The SEC’s EDGAR system is a public database; scraping is permitted under the SEC’s EDGAR Terms of Use, subject to rate limits (10 requests/second as of 2024; the SEC has implemented a policy requiring a User-Agent header identifying the requester).
  • Downloading and analyzing publicly filed DOL Form 5500 data from EFAST2. The DOL explicitly makes this data available in bulk for research purposes.
  • Searching and analyzing IRS Form 990 data via ProPublica Nonprofit Explorer or the IRS’s own public database. 990s are public records by statute (IRC § 6104).
  • Reading and analyzing publicly available pension investment reports, board meeting minutes, and investment committee materials posted on pension websites.
  • Monitoring and analyzing publicly available press coverage, conference programs, and LinkedIn public profiles.

Permitted with care:

  • Automated bulk downloading from EDGAR or EFAST2. Both systems have rate limits and require User-Agent identification. Automated access is permitted within published limits. Exceeding rate limits constitutes a terms-of-service violation and may result in IP blocking.
  • LinkedIn data. LinkedIn’s Terms of Service prohibit scraping via automated means. However, public profile review and manual research is permitted. The hiQ v. LinkedIn litigation established that scraping publicly accessible data does not violate the CFAA (Computer Fraud and Abuse Act), but this is an evolving legal area. Conservative practice: use LinkedIn for manual research; do not build automated scrapers.

Not permitted:

  • Obtaining pension investment data through deception — e.g., impersonating a journalist or beneficiary to obtain records that would otherwise be withheld.
  • Purchasing stolen or improperly obtained data (e.g., from an insider who violated confidentiality obligations).
  • Using FOIA as a competitive intelligence tool against a pension’s investment staff in a way that amounts to harassment (repeated, clearly bad-faith requests with no legitimate research purpose).

4.2 Consultant Protocol — The Gatekeeper Risk

The most important practical constraint in this entire exercise is the consultant-gatekeeper dynamic, and it operates at the sales/IR layer, not the research layer.

The five major investment consultants (Mercer, NEPC, Callan, Cambridge Associates, RVK) collectively influence the majority of public pension and endowment alternatives allocations. Each consultant maintains an internal database of approved or rated managers. If a hedge fund’s IR team approaches a pension directly — bypassing the consultant — the consultant may:

  1. Note the bypass as a red flag about the manager’s understanding of institutional processes.
  2. In some cases, explicitly communicate to the pension that the manager made unsolicited direct contact, which undermines the manager’s chances of being included in the next formal search.
  3. In extreme cases (aggressive, repeated direct solicitation), put the manager on an informal “do not recommend” internal list.

Protocol for LP inference as a precursor to sales activity:

  • LP inference research is a pre-call research activity. The output should inform who to contact, not replace the proper channel.
  • The proper channel for most public pension allocations is: (1) build a relationship with the pension’s investment consultant; (2) get rated or approved; (3) when the pension runs a search in your strategy, the consultant will include you.
  • Direct outreach to pension trustees (board members) rather than investment staff is particularly sensitive. Trustees are the fiduciaries but are rarely the decision-makers in manager selection. Contacting a trustee directly signals a lack of understanding of how institutional allocation works and may be escalated to the investment staff as a compliance concern.
  • Direct outreach to investment staff (CIO, portfolio managers) is the correct channel if you have no consultant relationship. However, do this only after confirming the pension is in-market (via RFP or public board minutes signaling a search). Cold outreach to pension investment staff is low-yield; warm outreach (referral from a known LP, referral from the consultant) is the only reliable path.

Exception: For very large sovereign wealth funds and the largest pension systems (CalPERS, CalSTRS, NYSCRF, OTRS), direct institutional coverage (a dedicated sales person who manages the relationship) is appropriate and expected. These systems have internal investment staff large enough to evaluate managers independently of consultant recommendations.

4.3 GDPR and Non-US Institutions

For non-US institutional investors — European pension funds, Canadian pension plans (CPP Investments, Ontario Teachers’, OMERS, HOOPP), Australian superannuation funds — the inference methodology changes:

European pension funds:

Most European public pension systems are not subject to FOIA-equivalent laws as strong as US state sunshine laws. Key exceptions:

  • Swedish Pension Funds (AP1–AP7): Sweden has the world’s oldest freedom-of-information law (Tryckfrihetsförordningen, 1766). Swedish AP funds disclose their full investment schedules publicly. This is one of the best non-US sources for LP inference.
  • Danish ATP: Subject to Danish public records law; investment schedules are partially disclosed.
  • Dutch ABP and PFZW: Netherlands has a moderate public records law (Wet open overheid, WOO). Direct investment schedules are not routinely disclosed, but ABP and PFZW publish annual reports with significant alternatives detail.
  • UK LGPS (Local Government Pension Scheme): LGPS funds are subject to the UK Freedom of Information Act. Annual reports and investment strategy statements are public; individual manager names are sometimes disclosed.

GDPR considerations:

When building an LP inference database that includes personal data (e.g., names of pension CIOs, investment committee members, LP contacts), GDPR applies to data about EU residents regardless of where the data processor is located. Key requirements:

  • Lawful basis for processing: “legitimate interest” (Article 6(1)(f)) is typically the applicable basis for sales intelligence databases. Legitimate interest requires a balancing test — the data subject’s privacy rights vs. the controller’s commercial interest.
  • Publicly available professional information (LinkedIn profiles, public conference appearances, published board meeting minutes) is generally within scope of legitimate interest processing.
  • Do not store sensitive personal data (political opinions, union membership) in the inference database.
  • Maintain a data retention policy — LP relationship data more than 5 years old without a refresh should be reviewed for continued relevance.
  • If building a commercial product that includes EU individual data, consult with a GDPR-qualified attorney.

Canadian pension funds:

The large Canadian pension funds (CPP Investments, Ontario Teachers, OMERS, AIMCo, PSP, Caisse de dépôt) are not fully subject to provincial or federal freedom-of-information laws in the same way US public pensions are. However, they publish extremely detailed annual reports, sustainability reports, and investment strategy updates that often name asset managers in specific sub-categories. CPP Investments, for example, publishes an annual report with a full list of “fund investments” including fund names and commitment amounts in some cases. These proactive disclosures are the primary source for Canadian LP inference, not FOIA.

Part 5 — Operational Implementation Checklist

Phase 1 — Seed the graph (week 1–2)

  1. Download CalPERS, CalSTRS, WSIB, NYSCRF, and NC Retirement Systems’ most recent investment schedules for absolute return / hedge fund allocations (all proactive; no FOIA required).
  2. For each named fund, run an EDGAR Form D search and Form ADV lookup. Populate fund table.
  3. For each pension, populate lp_entity table. Create lp_relationship edges at confidence 0.95 (proactive_disclosure).
  4. Cross-reference Form D investor counts against the number of confirmed LPs per fund. Flag funds where confirmed LP count is significantly below investor count — these have unknown LPs to find.

Phase 2 — FOIA outreach (week 2–4)

  1. Prioritize FOIA requests to: Connecticut CRPTF, New Jersey DoI, North Carolina Retirement Systems, Oregon OIC, and Texas TRS (acknowledging Texas complexity). Use the template in Section 1.1.2.
  2. For NYC pension systems, monitor the NYC Comptroller’s website for upcoming investment committee meeting packages (monthly cadence).
  3. Request investment committee meeting minutes for the last 2 fiscal years from each jurisdiction.

Phase 3 — Form 5500 sweep (week 3–4)

  1. For each target fund, search EFAST2 and the DOL DFE research file for fund legal name variants.
  2. Where 103-12 IE filings are found, download and extract participating plan names.
  3. Cross-reference plan EINs with EFAST2 plan search to identify union/Taft-Hartley LPs.

Phase 4 — 990 sweep (week 4–5)

  1. For each target fund, search ProPublica Nonprofit Explorer for university endowments, hospital systems, and foundations with AUM suggesting alternatives capacity (>$500M endowment).
  2. Review Form 990-PF Part II disclosures and Schedule D for named fund references.
  3. Update inference graph with 990-sourced edges at confidence 0.80.

Phase 5 — Press archive sweep (week 5–6)

  1. Search P&I, FundFire, Bloomberg, and Google News archives for “[fund name] + [allocator name]” combinations.
  2. Compile a press-confirmed LP list with source citations.
  3. Update edges where press confirmation supports existing inferences (confidence boost) or adds new edges.

Phase 6 — Gap analysis and peer inference (week 6–7)

  1. For each fund with known LPs, identify peer pension funds in the same state with similar AUM and strategy approvals.
  2. Apply peer inference heuristic (Section 3.4) to generate candidate LP nodes at confidence 0.25–0.40.
  3. Tag these nodes as inferred status pending future confirmation from FOIA or press.

Summary: Signal Quality by Source

Source LP Named Typical Lag Coverage Confidence
FOIA investment schedule Yes 0–30 days US public pensions 0.95
Proactive pension disclosure Yes 0–90 days Large US pensions 0.95
NYC board meeting packages Yes 0–7 days NYC pensions 0.90
Form 5500 DFE schedule Yes 6–18 months Taft-Hartley plans 0.90
Press (named spokesperson) Yes 0–7 days Large funds 0.85
Form 990-PF Part II Sometimes 6–15 months Foundations 0.80
Swedish AP fund reports Yes 3–6 months Swedish pensions 0.95
Canadian pension annual reports Sometimes 3–6 months CA pensions 0.75
Form ADV client type % No (aggregate) 90 days Any RIA 0.45
13F cross-reference No (indirect) 45 days Long equity funds 0.50
Form D state correlation No 15 days Any Reg D fund 0.35
Peer inference No n/a Any 0.25–0.40
Conference co-appearance No 0–30 days Any 0.20

The highest-ROI starting point is the set of proactive publishers (CalPERS, CalSTRS, WSIB, NYSCRF, Swedish AP funds) combined with NYC Comptroller board packages. These sources require no FOIA request, update regularly, and name specific funds with commitment amounts. FOIA requests to the secondary tier (CT, NJ, NC, OR) are the next lever. Form 5500 and Form 990 sweeps are most productive for union plans and foundations respectively. Press archive sweeps confirm and date inferences derived from structural sources.

Research completed 2026-06-18. Form PF compliance dates reflect SEC extension to October 1, 2026. FOIA turnaround times are representative based on statutory deadlines and practitioner experience; actual response times vary.