The Fed's impossible position + regional divergence + hidden leverage

Built 2026-06-07 from models/z3/fed_policy_trap.py, models/graph/regional_leverage.py (reading data/bank_exposure.json). This is the formalization of the user's thesis: a single policy rate cannot fix a multi-dimensional (regional × sectoral × global) divergence.

1. The monetary-policy trap (Tinbergen, proven UNSAT)

To stabilize N independent targets you need ≥ N independent instruments. The Fed has ~1 (the funds rate; ~2 with the balance sheet). The targets pull in opposite directions:

Target	Rate it "wants"	Why
Inflation / dollar defense	≥4.5%	record gold, de-dollarization → tightness
Regional banks / CRE / HTM	≤2.0%	CRE unwind + securities losses → cuts
AI financial stability	~4.0%	don't reflate the capex bubble
Sovereign fiscal / debt service	~2.5%	Treasury issuance wall → low rates
Global carry (JPY)	~3.5%	avoid forcing the yen-carry unwind

Z3 results:

F1 UNSAT: no r with r≥4.5 ∧ r≤2.0 — inflation defense and bank rescue are directly contradictory.
F2 UNSAT: no r within ±1.0pt of all five targets (spread 2.0–4.5 > 2×tol).
F3 UNSAT: even two instruments (rate + QT/QE) can't independently satisfy five targets — the system is over-determined.
Best case: the Chebyshev-optimal single rate is r\ = 3.25%, and it still* leaves a target ~1.25 points mis-set. Raising toward 4.5% pushes regional banks/CRE/fiscal toward crisis; cutting toward 2.0% reflates inflation/AI/gold.

Conclusion: monetary policy cannot fix a divergence crisis — it can only choose which part of the system to sacrifice. That is the formal trap, and it is why "raise or lower the rate" has no solution.

2. Regional divergence (the empirical input)

Aggregating the top-200 banks by state (regional_leverage.py): average CRE-concentration/Tier-1 ranges 37% → 336% across states — a ~300-point dispersion (KY 351%, SC 349%, WV 339%, MT 336% at the top; coastal/large-bank states far lower). Bank stress is geographically divergent: the CRE/HTM pain concentrates in specific states/regions while the AI-driven asset inflation concentrates in others (coastal tech hubs). A single national rate is the wrong tool for a map this uneven — the cities/suburbs/rural and state-by-state divergence the user observed is real in the call-report data.

3. Hidden vs surfaced leverage (the masked risk)

reported_leverage = assets/equity; hidden_leverage = assets/(equity + HTM_loss + AFS_loss) (mark securities losses to capital). The gap is the un-marked HTM hole:

Bank	Reported	HTM-adjusted	×inflation	HTM/eq
Charles Schwab Bank	13.3×	35.9×	×2.7	−43%
Bank of Hawaii	13.5×	23.9×	×1.8	−35%
Bank of America	11.1×	16.9×	×1.5	−34%
Frost Bank	11.7×	16.6×	×1.4	−4%*

A bank can report a comfortable ~11–13× leverage while its true mark-to-market leverage is 17–36×. This is precisely the leverage that a single rate move cannot safely unwind: cutting rates relieves it (securities recover) but reflates the bubble; holding/raising rates keeps these books underwater.

Honesty caveat: the model also flags several names with negative adjusted equity — but those are US branches/agencies of foreign banks (Standard Chartered, Bank of China, BBVA, State Bank of India, Bank of Baroda, Bank Hapoalim), whose reported "equity" is a branch-accounting artifact (they're capitalized by the parent abroad). That is not a solvency signal and is excluded from the vulnerability read — flagged to avoid a false alarm.

4. Why this closes the loop

The single-instrument trap is the macro mirror of the Layer-1 finding: just as the AI core is "solvent only while external capital flows" (Z3 T4), the banking system is "stable only at a rate that doesn't exist." Both are constraint systems with no feasible interior solution — they resolve only by a transfer (more capital in, or a chosen sacrifice). The divergence spans regions, sectors, wealth classes and continents; a scalar policy rate is, provably, the wrong dimensionality of tool.

← Research index · structured data: macro-fed-trap-regional.json · macro-fed-trap-regional.md