Holding-company layer methodology

The form

The FR Y-9C is the Federal Reserve’s quarterly consolidated financial statement for holding companies. Consolidated means the parent plus every subsidiary it consolidates, not the insured bank alone. That is the point of this module: /banks reads the call report, which stops at the insured bank, and this reads the form above it.

FinObservatory carries two tables. bhc_financials holds 181,114 filings, one row per holding company and quarter, over 139 quarters from Q3 1986 to Q1 2021. bhc_institutions holds one row for each of the 4,842 holding companies, with its first and last report dates and the number of quarters it filed. The financial table carries exactly four measures: TOTAL_ASSETS_THOUSANDS_USD, TOTAL_LIABILITIES_THOUSANDS_USD, TOTAL_EQUITY_CAPITAL_THOUSANDS_USD and TIER1_CAPITAL_THOUSANDS_USD. Everything on these pages is built from those four columns, and anything that needs a fifth is not built.

Units. All four are in thousands of US dollars, the same unit as the FDIC parquets. A figure rendered as $25.07T is 25,073,559,860 as stored. No column on these pages is in millions, and none is in whole dollars.

The double count, and why the panel total starts at Q1 2006

Before Q1 2006, lower-tier holding companies filed the FR Y-9C alongside the parents that owned them. Both filings are consolidated, so both contain the same subsidiary bank. Summing the panel therefore books those assets twice. The size of the artifact: 2,310 filers report $16.18T at Q4 2005; 1,009 filers report $11.53T at Q1 2006, a 28.7% fall in one quarter. 1,303 entities carrying $5.18T, 32.0% of the earlier total, report at Q4 2005 and not at Q1 2006.

The largest entity to stop filing, N B HOLDINGS CORP (RSSD 2173092), reported $1.28T. The National Information Center records it as a subsidiary holding company of Bank of America Corp, which reported a balance sheet of similar size in the same quarter, kept filing, and grew the following quarter. The overview page prints both entities’ figures side by side.

The rule this imposes. bhc_financials has no parent-child column, so the duplicate filings cannot be identified and netted out. So the duplicates can only be excluded wholesale, by starting every system-level aggregate at Q1 2006. That is what this module does: 61 quarters of summable panel, and 78 earlier quarters that are used entity by entity and never added up. The break quarter is not typed in as a constant; it is derived as the largest quarter-on-quarter fall in the filer count anywhere in the panel, so it stays correct as the source is extended.

Assets do not equal liabilities plus equity

The accounting identity fails in 22,182 of the 181,114 filings, and that is the form working as designed: TOTAL_EQUITY_CAPITAL is common and preferred equity capital and excludes noncontrolling (minority) interests in consolidated subsidiaries. The residual is real money. On BANK OF AMER CORP at Q4 2010, assets of $2.27T minus liabilities of $2.03T minus equity capital of $228.25B leave $8.97B. Across the filings where the identity does not close, the gap is 0.35% of assets at the median, 2.8% at the 90th percentile and 76.2% at its widest.

Consequences. Book debt is the reported liabilities column, never assets minus equity, which would silently absorb the minority interest into debt. And the capital ratio on these pages is labelled equity capital / assets, not book equity: it is a narrower numerator than the accounting residual.

Tier 1 is absent, not zero

TIER1_CAPITAL_THOUSANDS_USD is null in 60,274 filings. Of those, 56,150 sit in the 56,150 filings dated before Q1 1996, the first quarter in which any filer reports the field at all; the other 4,124 are scattered through the later years. Separately, 329 filings report a genuine zero.

A null and a zero are different facts, and the difference is the whole point. Coalescing the null to zero would turn “the form did not ask” into “this firm held no capital”. Nulls render as n/a. Aggregate Tier 1 ratios are summed over reporting filers only.

The panel is a moving regulatory window

The FR Y-9C is filed by holding companies above a size threshold, and that threshold was raised repeatedly. The three largest quarter-on-quarter falls in the filer count are Q4 2005 to Q1 2006 (2,310 filers to 1,009, median filer $372.2M to $827.3M); Q4 2014 to Q1 2015 (1,129 filers to 677, median filer $1.07B to $1.89B); Q2 2018 to Q3 2018 (650 filers to 372, median filer $2.30B to $5.53B). A median filer that jumps in size while the count collapses is a threshold change, not an exit wave.

This is why the concentration section recomputes the top-5 share on a fixed population of the largest 349 filers. A top-5 share measured on a shrinking denominator rises without anything concentrating.

Guarded divisors

814 filings report negative equity capital. An assets-over-equity leverage ratio flips sign on those rows. They are rendered n/a, not clamped to a floor.

Vintage: this is a historical module

The source ends. The last quarter in bhc_financials is Q1 2021, report date Mar 31, 2021, while fdic_financials and call_reports behind /banks run to the current quarter. Nothing on these pages is a statement about the holding-company system today. The cutoff is read from the data at build time, never typed into a query or into prose, so a refreshed source moves the pages rather than stranding them.

The two perimeters at Q1 2021

At the last quarter of the panel, 349 holding companies consolidate $25.07T, and 5,044 FDIC-insured institutions hold $22.73T. The ratio is 1.103.

These are not additive.A holding company’s consolidated total already contains its insured bank, so the two populations overlap by construction. Nor can the difference between them be read as the size of the nonbank sector: the FDIC universe also contains insured banks owned by no FR Y-9C filer, so the residual mixes what the holding companies own outside their banks with what the banks outside the holding companies own. FinObservatory reports the two levels and the ratio, and stops there.

The parent-child key: it exists, and it still does not close the comparison

The design note for this module claimed no parent-child key existed anywhere in the estate. That is wrong, and this section is the correction. bhc_financials.RSSD_IDis the holding company’s identifier and call_reports.idrssd is the bank’s, and those two sets barely intersect (1 of 4,842 holding-company RSSDs appear in the call report). But the FDIC Summary of Deposits, held here as sod_branches, carries rssdhcr, the RSSD of a bank’s high-holder holding company, alongside rssdid, the bank’s own. Joining on it matches 310 of the 349 filers in the panel’s last quarter, covering 89.1% of their consolidated assets.

What that key does not deliver is a defensible holdco-versus-bank number, and the failure is loud enough to measure: for 32 of those 310 holding companies, the subsidiary banks’ call reports sum to more than the parent consolidates. A subset of a balance sheet cannot exceed it.

FinObservatory therefore publishes no holdco-versus-bank ratio and no “nonbank share” of a holding company.

What this module still cannot do

No double leverage.Double leverage is the parent’s equity investment in its subsidiaries over its own equity. The numerator is BHCK2130on the form and it is not in this table, which carries four measures and no fifth. The ratio is not computed, not proxied and not estimated. The crosswalk above does not help: it maps ownership, not the parent’s carrying value of what it owns.

No dollar SRISK, and precisely why. SRISK needs book debt, a long-run marginal expected shortfall, and the market value of equity. This module unblocks the first: consolidated book debt is TOTAL_LIABILITIES_THOUSANDS_USD, a reported FR Y-9C field, present in all 181,114 filings from Q3 1986 to Q1 2021, with 0 nulls and 4 zeros. The second exists in the estate: systemic_series carries LRMES for 28 listed banks from Jan 3, 2020 to Jul 6, 2026, which overlaps the book-debt panel for 5 quarters. The third does not exist anywhere in FinObservatory: equity_returns carries date, ticker, log_return, volume, so it has no price level and no share count, and there is no market-capitalisation table and no ticker-to-RSSD crosswalk to build one. Book debt is sourced; market cap is not; SRISK stays unbuilt. It is not filled in with a book-value stand-in, because a SRISK computed on book equity is not SRISK.

Entity pages

There is a page for every holding company that filed in the panel’s last quarter, and for every holding company whose consolidated assets ever reached $10.00B: 570 entities. Each shows the filed balance sheet quarter by quarter, with capital ratios where they can be computed and n/a where they cannot. Filers below both bars are in the aggregates on the overview page but have no page of their own.

Sources

The FR Y-9C data behind bhc_financials and bhc_institutionsis the Federal Reserve’s holding-company reporting series; the identity of N B HOLDINGS CORP as a subsidiary holding company is from the National Information Center register; the insured-bank comparison at Q1 2021 is from the FDIC quarterly financial data already used by /banks.