PBGC Trusteed Pensions: The Federal Record of Every Dead Corporate Pension

When a company with a traditional pension goes bankrupt and the plan it promised its workers turns out to hold less than it owes, a federal agency takes the keys. The Pension Benefit Guaranty Corporation steps in as trustee, assumes the plan's assets and liabilities, and pays the retirees itself—up to a statutory maximum—for the rest of their lives. The list of plans it has taken over is, in effect, a registry of dead corporate pensions: roughly 5,170 trusteed plans, one row per failed promise, each naming the plan, the former sponsor, the date the agency took over, and the number of people whose retirement now depends on the federal safety net rather than on the company that hired them.

This article covers what the trusteed-plans dataset is and how ERISA created the insurance program behind it; the difference between the kinds of plan termination—standard, distress, and involuntary—and what trusteeship actually means; the PBGC's two separate insurance programs, the single-employer program funded by premiums and failed-plan assets and the multiemployer program whose looming insolvency prompted the 2021 Special Financial Assistance program; the benefit guarantee, the annually-set maximum, and how it is reduced for early retirement; the industries the data documents leaving the defined-benefit system—steel, airlines, and auto suppliers above all; how the trusteed-plans table joins to the DOL Form 5500 filings the same plans submitted before they died; a Python workflow that pulls the trusteed plans, tallies terminations by year and sector, and sums the affected participants; and the caveats—the snapshot nature of the file, the participant-count timing, and the limits of sponsor-name analysis—that every analyst must internalize before drawing conclusions.

What the dataset is

The Pension Benefit Guaranty Corporation (PBGC) is the federal corporation that insures most private-sector defined-benefit pension plans—the traditional pensions that promise a worker a specified monthly benefit in retirement, calculated from years of service and pay, rather than the account balance of a 401(k). When a covered defined-benefit plan fails—its sponsoring employer goes bankrupt and the plan does not hold enough assets to pay the benefits it owes—the PBGC becomes the plan's trustee: it takes legal custody of whatever assets remain, assumes responsibility for the benefit obligations, and pays the participants directly. The trusteed-plans dataset is the list of every plan that has gone through this process, the agency's own record of the pensions it has had to take over.

In our database this record is stored as the table pbgc_trusteed_plans, with the grain of one row per terminated, trusteed plan—roughly 5,170 plans, drawn from the PBGC's published single-employer trusteed-plans registry. Each row identifies a single dead pension: the plan's name, its former sponsor, the date trusteeship or termination took effect, and the number of participants the plan covered. The columns capture who failed, when, and how many people were affected:

plan_name              -- the name of the terminated pension plan
sponsor_name           -- the former employer / plan sponsor
ein                    -- sponsor employer identification number
plan_number            -- the three-digit plan number (PN) for that sponsor
date_trusteed          -- date the PBGC became trustee of the plan
date_terminated        -- the plan's termination date (often near trusteeship)
participants           -- number of participants in the plan at termination
program                -- insurance program (the download is single-employer)
state                  -- state of the sponsor / plan
status                 -- trusteed, pending trusteeship, or terminated

Two pairs of columns do most of the analytic work. The first is the identity pair—the ein (the sponsor's employer identification number) and the plan_number (the three-digit plan number, or PN, that distinguishes a sponsor's several plans from one another). Together the EIN and PN form the federal primary key for an employee-benefit plan, and they are exactly the key on which the same plan's annual Form 5500 filings are filed— which, as the join section explains, is what lets an analyst trace a dead pension back through its living history. The second is the outcome pair—the date_trusteed and the participants count—which is what turns the list from a roster of names into a measurable record: when the plan failed, and how many people's retirements were pulled into the federal safety net when it did. The program field marks which of the PBGC's two insurance programs the plan belonged to, a distinction so consequential that the next sections treat it at length.

What it is and the ERISA frame

The PBGC was created by the Employee Retirement Income Security Act of 1974 (ERISA), the landmark statute that, after years of well-publicized pension failures, imposed federal standards on private-sector employee-benefit plans. The catalyzing example in the public memory was the 1963 closure of the Studebaker automobile plant in South Bend, Indiana, whose pension plan was so underfunded that thousands of workers—many with decades of service—received only a fraction of the benefits they had been promised, or nothing at all. ERISA's answer to that kind of failure had two parts: substantive rules requiring employers to fund their pension promises and protecting participants' rights, and an insurance backstop for the case where the funding rules nonetheless were not enough. The PBGC is that backstop.

The PBGC is a federal corporation, not a tax-funded agency in the ordinary sense. Its single-employer guarantee is financed primarily by insurance premiums paid by the sponsors of ongoing covered plans, by the assets it recovers from the failed plans it takes over, and by recoveries from bankrupt sponsors' estates—not by an annual congressional appropriation. The premiums come in two forms: a flat-rate per-participant premium that every covered plan pays, and a variable-rate premium that underfunded plans pay in addition, scaled to the size of their funding shortfall. This structure—the healthy plans insuring against the failures of the weak—is the same risk-pooling logic that underlies any insurance scheme, and it explains why the agency's own financial position, the gap between the premiums and recoveries it collects and the benefits it must pay, is itself a closely watched number.

The plans the PBGC insures are defined-benefit plans, and the distinction from defined-contribution plans is fundamental to understanding why the insurance is needed at all. A defined-contribution plan—the 401(k) being the archetype—makes no promise about the eventual benefit; the worker bears the investment risk, and there is nothing to insure because there is no fixed obligation that can fall short. A defined-benefit plan, by contrast, promises a specific benefit, and the employer bears the risk of funding it. When the employer cannot, the promise can fail—and it is precisely that failure the PBGC exists to catch. The great structural story the trusteed-plans data documents is the decades-long migration of American retirement saving away from the defined-benefit plans the PBGC insures and toward defined-contribution plans it does not, a shift in which every trusteeship is one more traditional pension leaving the system.

Plan termination and what trusteeship means

Not every pension plan that ends does so through the PBGC, and understanding the kinds of termination is the key to understanding which plans land in this dataset. ERISA recognizes three pathways, and only some of them involve the agency stepping in to pay.

A standard termination is the orderly, fully-funded wind-down. A financially healthy employer that wants to exit the defined-benefit business terminates a plan that has enough assets to pay every benefit it owes, typically by buying annuities from an insurance company to cover the participants or paying lump sums. Because the participants are made whole, the PBGC does not take over and pay benefits—it reviews the termination to confirm the plan really is sufficiently funded, but it does not become trustee. Standard terminations are the healthy exits, and they are not what this dataset is about. A distress termination is the opposite: a sponsor in serious financial trouble—usually in bankruptcy—asks to terminate an underfunded plan it can no longer support, demonstrating to a bankruptcy court or to the PBGC that it cannot continue in business unless the plan is ended. An involuntary termination is one the PBGC itself initiates, when it concludes that a plan must be terminated to protect participants or to limit the insurance program's losses—for instance when a sponsor stops making required contributions or the plan's funding deteriorates to the point where waiting would only deepen the shortfall.

It is the distress and involuntary terminations that produce the records in this dataset, because both lead to trusteeship. When an underfunded plan terminates this way, the PBGC becomes its trustee: it takes legal title to the plan's remaining assets, assumes the obligation to pay the guaranteed benefits, and from that point sends the monthly checks to the retirees itself. The former sponsor is generally released from the plan, its remaining liability becoming a claim the PBGC pursues against the bankrupt estate. For the retiree, trusteeship is the moment the source of the pension check changes—from the company that employed them, often for a working lifetime, to a federal corporation in Washington. For the analyst, it is the moment a plan enters this dataset, and the date_trusteed is the timestamp of that transfer.

The single-employer and multiemployer programs

The PBGC runs two entirely separate insurance programs, with separate finances, separate rules, and very different histories. The program field in the data marks which one a plan belongs to, and conflating the two is one of the easiest mistakes to make with this dataset.

The single-employer program insures the pension plans of individual companies—the classic case of one employer sponsoring one plan for its own workers. It is the larger program by plan count and is funded by the premiums those single-employer plans pay plus the assets and recoveries from the ones that fail. When a single-employer plan is trusteed, the mechanics are as described above: the PBGC takes over and pays the guaranteed benefits directly, replacing the bankrupt sponsor as the payer. Most of the high-profile pension failures of recent decades—the large industrial and airline bankruptcies—ran through the single-employer program, and it is the single-employer trusteed plans that the PBGC publishes as the downloadable registry behind this dataset.

The multiemployer program is structurally different. A multiemployer plan is a single pension plan jointly maintained by many employers— typically unionized employers in the same industry and region—under a collective bargaining agreement, so that a worker can move between participating employers without losing pension coverage. These plans dominate industries like trucking, construction, mining, and retail food. The PBGC's role here historically differed: rather than taking over an insolvent multiemployer plan outright and paying participants directly, the program generally provided financial assistance to keep an insolvent plan paying guaranteed benefits, and the guarantee level for multiemployer participants is set on a different, generally lower basis than the single-employer guarantee. For years the multiemployer program faced a slow-motion solvency crisis: a handful of very large plans—most prominently in the trucking sector—were projected to run out of money, and the multiemployer insurance program itself was projected to become insolvent, threatening drastic benefit cuts for millions of participants. The response came in the American Rescue Plan Act of 2021, which created the Special Financial Assistance (SFA) program: a one-time, government-funded infusion to financially troubled multiemployer plans, designed to let them pay full benefits for decades without cutting them. The SFA program was a watershed, a direct federal appropriation into a program that had been built to be self-funded, and it reshaped the outlook for the multiemployer side even as the single-employer program continued on its premium-funded footing.

The benefit guarantee and the maximum

The single most consequential fact for a participant whose plan is trusteed is that the PBGC guarantee is a capped guarantee, not a promise to pay whatever the failed plan promised. The agency pays guaranteed benefits, and the guarantee is bounded by a statutory maximum. For a participant whose promised pension was modest, the guarantee usually covers the full benefit—the cap never bites. But for a participant who was promised a large pension—a long-tenured executive, a senior pilot, a skilled worker with decades of service in a generous plan—the cap can mean a meaningful cut from what the plan would have paid had it survived.

The single-employer maximum guarantee is set by statute and adjusted annually, and it is expressed as a maximum monthly (or annual) benefit for a participant retiring at age 65 in straight-life-annuity form. Two adjustments make the practical guarantee lower than that headline figure for many people. First, the maximum is reduced for retirement before age 65: a participant who is already drawing a pension at, say, age 55 when the plan is trusteed faces a substantially lower cap than the age-65 figure, because the guarantee is actuarially reduced to reflect the longer payout period. Second, the maximum is reduced for benefits paid in forms that protect a survivor, such as a joint-and-survivor annuity. The cap applicable to any given participant is therefore a function of the maximum in effect in the year the plan terminated, that participant's age, and the benefit form—which is why the headline maximum, while a useful reference point, is not the number any particular retiree necessarily receives.

There is a further wrinkle that matters for understanding why some retirees feel the loss of a trusteeship acutely. The guarantee covers vested benefits up to the maximum, but it phases in benefit increases: a benefit improvement adopted shortly before a plan fails may not be fully guaranteed, on the principle that a sponsor heading for trouble should not be able to load up promises the insurance program must then absorb. The combined effect of the dollar cap, the early-retirement reduction, and the phase-in of recent increases is that trusteeship can leave a high-earning, early-retiring participant—again, the airline pilot is the canonical example—receiving materially less than the bankrupt plan had promised, even as a lower-paid colleague in the same plan is made entirely whole. The participant count in this dataset records how many people were pulled into the safety net; it does not, by itself, say how many of them landed below the cap.

The industries that shed their pensions

Read in aggregate, the trusteed-plans dataset is a documentary record of which corners of the American economy abandoned the defined-benefit pension, and the concentration is striking. A few industries account for a disproportionate share of the participants pulled into the PBGC, and each tells a piece of the larger story of industrial decline and competitive upheaval.

Steel is the archetype. The integrated steel producers of the Rust Belt carried enormous defined-benefit obligations to a heavily unionized, long-tenured workforce, accumulated over generations when the industry dominated American manufacturing. As foreign competition, mini-mills, and overcapacity hollowed the integrated producers out, a wave of steel bankruptcies in the late 1990s and early 2000s sent some of the largest pension plans the PBGC has ever taken over into trusteeship, pulling hundreds of thousands of steelworkers and retirees into the federal safety net. Airlines are the second great cluster. Following deregulation and then the financial shock of the early 2000s, several major legacy carriers entered bankruptcy and terminated their underfunded pension plans—plans whose generous promises to pilots and other long-service employees made them among the most expensive terminations on record, and the cases that most vividly illustrated the bite of the early-retirement cap on highly-paid participants.

Auto suppliers form a third heavy concentration. The tiered structure of the automotive industry—a handful of large assemblers above a deep base of parts suppliers—meant that when the auto sector contracted, it was frequently the suppliers, with their own unionized workforces and legacy pension plans, that failed and handed their plans to the PBGC, the supplier bankruptcies of the late 2000s being the most acute episode. Older clusters appear too—textiles and apparel as that industry moved offshore decades earlier, and scattered failures across other heavy and consumer manufacturing. The common thread is the legacy-cost problem: mature, capital-intensive, unionized industries that promised rich pensions when they were dominant and could not sustain those promises once their competitive position eroded. The dataset is, in this reading, a map of where the post-war industrial pension bargain came undone—though, as the caveats section stresses, attributing a plan to an industry from its sponsor name alone is only a first approximation of that map.

Joining to Form 5500 and the wider pension data

The trusteed-plans table is most powerful not in isolation but as the terminal record in a plan's life history, and the join that unlocks that history is to the DOL Form 5500. Every covered defined-benefit plan files a Form 5500 annual report with the Department of Labor while it is operating, disclosing its assets, liabilities, funding status, participant counts, and—on the actuarial Schedule SB—the detail of its funding shortfall. Those filings are keyed by the sponsor's EIN and the plan's plan number (PN), the very same two identifiers the trusteed-plans table carries. The EIN+PN pair is therefore the bridge between a plan's living disclosures and its death record.

That bridge enables the most analytically valuable work this dataset supports: reconstructing the run-up to failure. By joining a trusteed plan back to its final years of Form 5500 filings, an analyst can watch a plan's funded ratio deteriorate in the years before trusteeship, observe whether the sponsor kept making required contributions, and quantify the underfunding the PBGC inherited. The question of whether plan failures were foreseeable from the funding disclosures—and how many years of warning the Form 5500 record gave—is answerable only across this join. It also supports the inverse question: of all the underfunded plans visible in the Form 5500 data, which ones went on to be trusteed and which survived, a natural setup for understanding what distinguishes a plan that fails from one that merely struggles.

Beyond the Form 5500, the trusteed-plans data sits alongside the PBGC's own companion datasets—the insured-plans data describing the still-living plans that pay premiums into the system, and the agency's premium and financial reporting—so that the trusteed file can be read against the universe of plans that have not (or not yet) failed. It connects, too, to bankruptcy and corporate records: a trusteeship is almost always the pension consequence of a corporate bankruptcy, and the sponsor named in this dataset can frequently be matched to the bankruptcy case, the enforcement record, and the broader story of the company's collapse. The dead pension is one artifact of a corporate failure; the trusteed-plans table is the place that artifact is cataloged.

Analytical uses

A national, dated, participant-weighted registry of failed pensions supports a distinctive set of analyses about the decline of the traditional pension and the health of the federal insurance system that backstops it.

Termination trends over time is the most immediate use. Because every row carries a trusteeship date, the failures can be bucketed by year to trace the waves of plan terminations—the steel cluster, the airline cluster, the auto-supplier cluster—and to relate them to the business cycle, since pension terminations spike with the corporate bankruptcies that drive them. Weighting those failures by the participant count, rather than counting plans, is what reveals the human scale: a single large industrial plan can pull more people into the safety net than hundreds of small ones, so plan-count and participant-count trends tell genuinely different stories and both are worth computing.

Industry and geographic concentration exploits the sponsor and state fields to map where the defined-benefit system contracted—which sectors and which regions shed the most plans and participants—though, as noted, robust industry attribution requires joining out to a SIC or NAICS code rather than relying on the sponsor name. PBGC exposure and program health aggregate the participants and, where the data permit, the underfunding the agency inherited, feeding the perennial policy question of the insurance program's long-run solvency and the adequacy of the premiums that fund it. And the foreseeability questionalready described—joining trusteed plans back to their Form 5500 funding history to ask how predictable each failure was—is the analysis that turns a static registry of dead plans into a study of how pension failures actually happen.

Python workflow: trusteed plans by year, sector, and participants

The script below pulls the PBGC trusteed-plans dataset from the agency's public data resources on pbgc.gov, then computes three of the core metrics: terminations by year (the wave structure of pension failure), the total and median participants pulled into the safety net, and the largest dead pensions by participant count—plus a deliberately crude sector tally from sponsor-name keywords. No API key is required; the file is a single Excel (.xlsx) spreadsheet download. Because PBGC release column names vary, the script resolves the plan, sponsor, date, and participant columns defensively rather than hard-coding them; any production use should be validated against the current PBGC data dictionary and should treat the sector keyword pass as a placeholder for a proper SIC/NAICS join through the Form 5500.

import requests, io
import pandas as pd
from collections import Counter

# PBGC publishes its "Single-Employer Plans Trusteed by PBGC" dataset
# through pbgc.gov data resources -- a single Excel (.xlsx) spreadsheet
# download, no API key required. It is one row per single-employer plan
# the PBGC has taken over since 1974: the plan name, the former sponsor,
# the trusteeship / termination date, and the participant count.
#
# Column names vary across PBGC releases, so the script resolves them
# defensively rather than hard-coding them, and any production use
# should be confirmed against the current PBGC data dictionary.
TRUSTEED_URL = "https://www.pbgc.gov/sites/default/files/trusteedplans.xlsx"


def load_trusteed(url=TRUSTEED_URL):
    r = requests.get(url, timeout=300)
    r.raise_for_status()
    # The published file is an .xlsx workbook; openpyxl reads it.
    return pd.read_excel(io.BytesIO(r.content), engine="openpyxl")


def col(frame, *candidates):
    # Return the first column whose lowercased name matches a candidate.
    lower = {c.lower().strip(): c for c in frame.columns}
    for cand in candidates:
        if cand.lower() in lower:
            return lower[cand.lower()]
    # Fall back to a substring search on the candidate tokens.
    for cand in candidates:
        for low, orig in lower.items():
            if all(tok in low for tok in cand.lower().split()):
                return orig
    raise KeyError(f"none of {candidates} in {list(frame.columns)[:12]}...")


df = load_trusteed()
print(f"Trusteed plans loaded: {len(df):,}")

c_plan  = col(df, "Plan Name", "Plan")
c_spon  = col(df, "Sponsor", "Company", "Sponsor Name")
c_date  = col(df, "Date Trusteed", "Trusteeship Date", "Termination Date",
              "Date of Plan Termination")
c_part  = col(df, "Participants", "Number of Participants",
              "Total Participants")

# --- 1. Terminations by year ------------------------------------------
df["_dt"] = pd.to_datetime(df[c_date], errors="coerce")
df["_year"] = df["_dt"].dt.year
by_year = df["_year"].dropna().astype(int).value_counts().sort_index()
print("\nTrusteeships by year (last 15 shown):")
for year, n in by_year.tail(15).items():
    print(f"  {year}: {n:>4,} plans")

# --- 2. Participants pulled into the PBGC safety net ------------------
df["_part"] = pd.to_numeric(df[c_part], errors="coerce")
total_part = int(df["_part"].sum(skipna=True))
print(f"\nTotal participants across trusteed plans: {total_part:,}")
print(f"Median plan size (participants): "
      f"{df['_part'].median(skipna=True):,.0f}")

# --- 3. Largest dead pensions by participant count -------------------
biggest = df.dropna(subset=["_part"]).nlargest(15, "_part")
print("\nLargest trusteed plans by participants:")
for _, row in biggest.iterrows():
    name = str(row[c_plan])[:40]
    print(f"  {name:<42} {int(row['_part']):>8,}")

# --- 4. Crude industry tally from sponsor-name keywords --------------
# A keyword pass is only a first approximation; a rigorous study would
# join the sponsor to a SIC / NAICS code from the Form 5500 filings.
SECTORS = {
    "steel":    ["steel", "iron", "metal"],
    "airline":  ["air", "airways", "airline", "aviation"],
    "auto":     ["motor", "auto", "automotive", "parts", "axle"],
    "textile":  ["textile", "mills", "fabric", "apparel"],
}
sector = Counter()
for raw in df[c_spon].fillna("").astype(str).str.lower():
    for label, keys in SECTORS.items():
        if any(k in raw for k in keys):
            sector[label] += 1
print("\nApproximate sector mix (sponsor-name keyword match):")
for label, n in sector.most_common():
    print(f"  {label:<10} {n:>5,} plans")

Two practical notes apply. First, the sector tally at the end is intentionally a first-pass heuristic: matching a handful of keywords against the sponsor name will both miss plans whose industry is not in the name and mis-classify plans whose name coincidentally contains a keyword, so it is a sketch of the industry mix, not a measurement of it. The rigorous version replaces the keyword pass with a join, on EIN and plan number, to the Form 5500 filings the same plan submitted—which carry the sponsor's industry code—and that join is the natural next step for anyone building a serious analysis. Second, the participant count summed here is the count recorded at termination; it is a stock at a moment in time, not a live headcount of how many people the PBGC is paying today, which drifts as retirees pass away and beneficiaries are added. The two numbers answer different questions, and the dataset answers the first.

Limitations and analytical caveats

The trusteed-plans dataset is the authoritative public record of failed private-sector defined-benefit pensions, but it carries structural limitations that an analyst must internalize before drawing conclusions from it.

It is a registry of failures, not of the whole system. The dataset records only the plans that failed and were trusteed—the deaths, not the births or the survivals. The plans that terminated in a healthy standard termination, the plans that froze their benefits but kept paying, and the vast universe of still-living plans paying premiums do not appear here. Any statement about “pensions” drawn from this dataset alone is a statement about failed pensions, and treating the trusteed list as if it described the defined-benefit universe—rather than its mortuary—is the single most common misreading. The insured-plans data and the Form 5500 universe are the denominators this numerator needs.

Participant counts are timed to termination and are a stock.The participant figure is the count as of the plan's termination, which is the right number for measuring how many people a given failure affected but the wrong number for almost anything else. It is not the number the PBGC pays today, which changes continuously; it is not separable, in the raw count, into retirees already drawing benefits versus deferred vested participants not yet collecting; and summing participants across plans of very different vintages mixes counts measured years and decades apart. Recency and headcount questions need the agency's current participant accounting, not the termination-time stock recorded here.

Sponsor-name analysis is fragile, and the two programs must not be mixed. Industry and corporate-family analysis built on the free-text sponsor name inherits all the hazards of name matching: a single corporate group may appear under several legal names, a name may not reveal the industry, and similarly-named but unrelated firms collide. Serious industry attribution and corporate-family rollups require resolving the sponsor to a stable identifier and a classification code, which is exactly what the EIN-based join to the Form 5500 provides. And because the single-employer and multiemployer programs differ so fundamentally—in how trusteeship works, in how the guarantee is set, and in how the program is funded—any aggregate that pools them without respecting the program field will produce a number that means nothing for either program.

It is a snapshot, and statuses evolve. The published file is a point-in-time extract: a plan can be in the process of being taken over (pending trusteeship) before it is fully trusteed, and the status of a plan can change between releases. The dataset is authoritative for the established record of completed trusteeships and for multi-decade trends; it is not a real-time monitor of every plan currently sliding toward failure, and the most recent failures may be incompletely reflected while the trusteeship process runs its course.

Held with these caveats in mind, the pbgc_trusteed_plans table is a uniquely valuable resource: a dated, sponsor-named, participant-weighted record of every private-sector defined-benefit pension the federal government has had to catch when its sponsor could no longer pay—the registry of the dead corporate pension, and through its join to the Form 5500, a window into how each of those pensions came to die.