Open Market Operations at Light Speed: How AI Converted Fekete's 1922 Warning into a Closed-Loop Capital Destruction Engine

In an interview published in March 2014, Antal Fekete described the policy of Federal Reserve open market operations as "a check-kiting scheme, pure and simple." ¹ He argued that the mechanism, which he dated to 1922, had been illegal from inception. His contention, in his own words, was that open market operations were "introduced illegally by the Fed in 1922" in "clear violation of the F.R. Act of 1913 which purposely excluded government bills, notes and bonds from the list of eligible paper." The claim is his and it is contested: Section 14(b) of the original Act of December 23, 1913 does on its face authorize the Reserve Banks to buy and sell "bonds and notes of the United States," and Fekete's argument turns on reading the eligible-paper provisions of Section 13 as the controlling limit. ² Congress amended Section 14(b) in the Banking Act of 1935 — the retroactive legalization, as Fekete read it — after the damage from the 1929 crash, which he attributed in large part to the 1922 innovation, was already done.

The specific mechanism Fekete objected to was not the monetization itself, though he objected to that too. It was a second-order effect: open market operations, once they became routine, made bond speculation risk-free. When the Fed pre-announces its bond purchases, speculators can front-run the Fed, buying bonds before the Fed does and selling them to the Fed at a guaranteed profit. The profit is not the result of economic judgment. It is a predictable consequence of a transparent policy.

Fekete predicted two things would follow from this arrangement. First, the bond market would develop a class of speculators whose entire business model was riding central-bank flows, rather than evaluating the underlying credits. Second — and this was the deeper point — the simultaneous manipulation of the interest rate (down) and the bond price (up) by the central bank would produce a form of capital destruction that standard monetary economics could not detect, because it worked by inflating the reported capital of financial institutions while deflating their actual solvency.

Both predictions have been borne out. And in the last five years, the mechanism Fekete described has been accelerated to the physical limits of latency and automated beyond any human capacity to monitor or correct.

Fekete's proposition

The theoretical core of Fekete's critique can be stated in a single sentence, and it is the sentence he himself acknowledged was the controversial one: a falling rate of interest — not a low rate, a falling one — destroys capital across the board ³. The arithmetic premise underneath it is unremarkable. The rate of interest cannot be pushed down without the price of bonds being pushed up, because the two are inversely related by definition. What Fekete built on that premise is the contested part.

A bond's price and its yield move in opposite directions. When the central bank buys bonds to push yields down, bond prices rise by arithmetic necessity. Any financial institution that holds bonds on its balance sheet — which is essentially every bank, every insurer, every pension fund — appears to experience a capital gain. Reported capital increases. Apparent solvency improves.

But the productive capacity of that capital has been impaired in exactly the same arithmetic proportion. The bond that now trades at a higher price yields less future income. Every institution that holds it has just been handed a mark-up that will reverse, with perfect certainty, as the bond matures at par. In the meantime, the institution must replace its maturing bonds with new bonds yielding less. Its future earning power has been quietly destroyed while its current balance sheet has been flattered.

Fekete used the language of both capital erosion and capital destruction for this, without drawing a formal distinction between them — the terms move together in his writing rather than marking successive stages. His point about where the damage sits is the one that matters: a rate cut raises the liquidation value of debt, so the sum a bond issuer must produce to retire its obligation ahead of schedule rises while the income stream servicing that obligation falls. The gap is the erosion. When it becomes too large to paper over, the institution experiences what Fekete called "sudden death syndrome" ⁴ — his phrase, applied to the insurance industry in 2014: apparent solvency one quarter, insolvency the next, with no intermediate warning in conventional accounting.

The mechanism cannot be fixed by the tools that created it. Further interest rate cuts deepen the erosion. Rate increases reveal it, by reversing the bond-price mark-up that had concealed it. In either direction, capital is destroyed. Under a gold standard, Fekete argued, this pathology is impossible, because the central bank cannot simultaneously suppress rates and inflate bond prices without losing gold. Under fiat, nothing prevents the operation, and nothing reveals the damage until the institutions begin to fail in sequence.

The 2023 proof case

Silicon Valley Bank's March 2023 failure was the clearest recent illustration of Fekete's mechanism, though almost no mainstream commentary framed it that way. SVB had accumulated a large portfolio of long-dated securities during the ZIRP and QE era, at prices the Fed's policy had directly inflated — predominantly agency mortgage-backed securities rather than Treasuries, it should be said: of the $91 billion held to maturity at the end of 2022, some $86 billion was mortgage-backed paper maturing beyond ten years, against roughly $16 billion of Treasuries across the book. ⁵ On the regulatory balance sheet, the portfolio appeared sound. When the Fed reversed policy and rates rose, the mark-to-market losses on the portfolio — never realized on the regulatory balance sheet thanks to the held-to-maturity accounting treatment — became a deposit-run vulnerability. The apparent capital that QE had created disappeared the moment the policy reversed, and a solvent-looking institution became insolvent in days.

This was not a failure of bank supervision. It was the predicted outcome of the mechanism Fekete described in 1922 terms and spent the better part of four decades warning about, from his turn to monetary economics in the early 1980s to his death in 2020. SVB held bonds the Fed had spent a decade inflating, and was caught when the Fed let them deflate. Unrealized losses on securities across the FDIC-insured system stood at $515.5 billion in the first quarter of 2023 — the quarter SVB failed — and peaked at $683.9 billion in the third, an appropriately Fekete-scale number for capital that was never truly there. ⁶

The institutions that escaped SVB's fate did so by having access to deposits stickier than SVB's (and by regulatory forbearance that let them hold the underwater bonds to maturity rather than realize the losses). Neither is a structural solution. The erosion remains in the system. The reversal remains incomplete. The next occasion will not be kinder.

The AI acceleration

What is new in 2026, and what Fekete did not live to see, is that the speculation he described in 1922 terms has been fully automated and compressed to the physical limits of information propagation.

The Fed now pre-announces not only the fact of its bond purchases but the specific timing, volume, and maturity distribution of its operations, in advance and in public. This is not an inference: the New York Fed's Open Market Trading Desk publishes a tentative operation schedule that names the operation dates, the settlement dates, the security types, the maturity range of eligible issues, and an expected size range for each operation. ⁷ Dealer-bank research desks decode Fed communications in real time. Algorithmic trading systems trained on those communications — and, increasingly, on the full corpus of central-bank speech, testimony, and committee minutes — adjust positions within microseconds of any new signal.

The market-making function has concentrated into a handful of principals — Citadel Securities, Virtu Financial, Jump Trading, XTX Markets, Tower Research Capital, Hudson River Trading — and the concentration is not seriously disputed, though it resists a single clean number. No published statistic measures their combined share of displayed depth across equities, fixed income, and foreign exchange together; what can be said is that Citadel Securities alone is commonly estimated to handle on the order of a quarter of U.S. equity volume and Virtu something approaching a fifth. High-frequency strategies as a class are commonly estimated to execute above 60% of U.S. equity volume, on figures that come from industry and vendor research rather than from any regulator. ⁸ Commercial market-research estimates put the automated algorithmic trading market somewhere between $20 billion and $27 billion in 2026, growing at roughly 13% annually — a range wide enough to be worth quoting as a range. ⁸ Execution algorithms have in fact been given over to machine learning at the largest dealers; JPMorgan's LOXM platform is the well-documented case. The specific performance gains attributed to those systems circulate widely and are almost entirely vendor-sourced, and no figure of that kind is relied on here. ⁸

Read alongside Fekete's 1922 critique, these figures describe a specific system: the front-running that Fekete identified as the core pathology of open market operations is now performed by a small number of firms, at speeds that exclude any human participation, using machine-learning models trained to anticipate Federal Reserve actions before they occur. The "risk-free profit" Fekete described has been industrialized. The profit is no longer scattered across thousands of traders exercising individual judgment. It is concentrated in a small number of firms executing algorithmic strategies at hardware-latency limits.

The closed loop

The feedback structure that results is what deserves the label closed-loop capital destruction engine. The loop has four stages.

Stage one: the Fed commits to a path of bond purchases or rate suppression, directly or through forward guidance.

Stage two: algorithmic trading systems front-run the committed path in nanoseconds, moving bond prices and derivative hedges to the anticipated end-state before the Fed's actual operations take place.

Stage three: financial institutions mark their bond holdings to the inflated prices, reporting capital gains that are, in Fekete's sense, putative rather than real. Apparent bank capital grows. Reported regulatory ratios improve. Equity analysts upgrade.

Stage four: the institutions' underlying earning power — the yield on their bond portfolios, the net interest margin on their loan books — silently deteriorates. The reported improvement in capital masks a deterioration in the cash-flow engine that actually services the institution's liabilities. The gap between apparent and real solvency widens.

The Federal Reserve then reviews the financial system, observes that "bank capital has strengthened," and concludes that its policy has been successful. The Working Group on Financial Markets (the Plunge Protection Team) monitors for disorderly conditions and, finding none in the inflated-price regime, takes no action. ⁹ The loop repeats.

Each iteration of this loop widens the gap between reported and real capital. Each iteration increases the systemic exposure to a rate reversal. Each iteration produces a set of SVB-like vulnerabilities, distributed across thousands of institutions, that will all be triggered at once by the first large rate move in the reverse direction.

AI has not created this pathology. Open market operations created it in 1922. But AI has accelerated it to a speed at which human oversight is not merely lagging — it is categorically impossible. The Federal Open Market Committee meets eight times per year. The algorithmic systems reacting to its every communication execute millions of adjustments per second. The mismatch in timescales is not an engineering problem to be optimized. It is a structural feature of the current monetary regime.

The insurance and pension dimension

Fekete was particularly concerned with the fate of the insurance industry under a protracted regime of suppressed interest rates, and addressed it specifically in his 2014 essay How the Fed Bankrupted the Insurance Industry by Destroying Its Capital Efficiency. His mechanism is not the familiar one about actuarial assumptions and reserve deficiencies, and it is worth reproducing as he put it. The insurer is hit twice by a rate cut. Its float earns less than it did before — "the account carrying insurance premiums will be compounding at a reduced rate," as he wrote, and "it will increase more slowly." And the capital efficiency of the premium stream itself is destroyed, because the present value of that stream is what the insurer effectively pays to collect it, and a halving of the rate doubles that present value with no offsetting reduction in the risk underwritten. The insurer is, in his image, "forced to skate on ice twice as thin." His diagnosis was that the industry was "a dead man walking"; his prognosis, the sudden death syndrome quoted above.

The pension sector is structurally identical. Defined-benefit plans calibrated to long-term returns near 8% — the average investment return assumption across the public plans NASRA surveys was 7.95% in 2007 — cannot meet those returns in a regime that has held long-term yields below 4% for most of two decades. ¹⁰ The assumptions have since been marked down rather than met: the same survey average had fallen to 6.91% by 2025, below 7% for the first time in more than forty years. ¹⁰ The underfunding accumulates silently until the liabilities mature, at which point the sponsor must either recapitalize the plan, cut benefits, or default. All three outcomes have been visible in the American pension landscape over the last decade, most severely at the state and municipal level.

The AI-driven algorithmic dimension does not exempt insurers and pension funds. It affects them in a specific way: the apparent mark-to-market gains on their fixed-income portfolios during the QE era were visible on their reports. The actuarial reserve shortfalls driven by the same policy were not. The asymmetry in visibility — gains that flow to current earnings, losses that accumulate in future obligations — is precisely the Fekete erosion dynamic, distributed across the industries that are systemically responsible for retirement security.

What breaks first

The closed loop cannot run indefinitely. It breaks when the rate-reversal dynamic begins to reveal the accumulated capital erosion, and when the pace of reversal exceeds the system's capacity to absorb it without forced liquidation.

The 2022–2023 rate cycle was a mild stress test of this dynamic. It produced SVB, First Republic, and Signature Bank. It produced acute stress in the UK gilt market, requiring the Bank of England to intervene directly — buying gilts from September 28 to October 14, 2022, at up to £5 billion a day and then £10 billion — to prevent pension-fund liability-driven-investment strategies from unwinding disorderly. That was the Bank acting alone in its own market, not a coordinated international operation, and it was gilts rather than Treasuries: the comparable dysfunction in the U.S. Treasury market belongs to March 2020, not to the 2022 cycle. ¹¹

These were previews, not the event. The accumulated capital erosion across the U.S. banking system, the insurance sector, and the defined-benefit pension universe is far larger than the mild 2022–2023 test revealed. The next episode of material interest-rate volatility — which the inflationary pressures from the Iran war and the accompanying energy shock may supply — will stress the system along the same lines, but with a larger accumulated imbalance to unwind.

The predicted response will be consistent with the closed loop: the Fed will inject liquidity, compress the most visible stress spreads, and allow the deeper marketability impairment to persist. Balance sheets will be papered over. The underlying productive capacity of the capital base will continue to decline. The loop will resume at a slightly lower steady-state level of real productive capacity.

This is, in Fekete's precise formulation, the mechanism by which the civilization quietly decapitalizes itself.

The uncomfortable corollary

The corollary that follows is one that Fekete stated explicitly but that most policy discussion refuses to accept: the monetary authority cannot solve this problem using the tools of the monetary authority. Every available Fed response — rate cuts, rate increases, forward guidance, balance-sheet expansion, balance-sheet contraction — operates through the same mechanism that produced the problem. The closed loop cannot be interrupted from inside itself.

A genuine resolution would require the tools Fekete specified and the broader Austrian tradition has defended: an external anchor, operated outside the discretion of the central bank and the Treasury, that resists the compounding erosion by making it immediately visible. The historical form of that anchor was gold. The digital-era form may be different. But the structural requirement — an instrument whose saleability is determined by market forces rather than by policy and whose price cannot be simultaneously manipulated against bond prices — has not changed.

Until that structural change is made, the closed loop will continue to run. AI has made it faster, more automated, and less visible. It has not changed the destination.

---

Next in this series: AI compute — the API credits, GPU-hour reservations, and model-access contracts that coordinate the modern software economy — as a spontaneously emerging form of what Fekete called "real bills" or "gold bills," and why this may be the first genuine monetary-adjacent clearing instrument to appear in the commercial world since the Bank of England opened its Manchester branch ¹².

---

Sources

Footnotes

1. Fekete, Antal E., interviewed by Anthony Wile. "Antal Fekete's Neo-Misesian Revisionism and Why He Believes It Is Necessary." The Daily Bell, 30 March 2014. https://thedailybell.com/all-articles/exclusive-interviews/anthony-wile-antal-feketes-neo-misesian-revisionism-and-why-he-believes-it-is-necessary/ — Source of the check-kiting quotation, the 1922 dating, the eligible-paper contention, the risk-free-speculation and front-running argument, and the Depression link. ↩

2. 12 U.S.C. 355, credit line "Dec. 23, 1913, ch. 6, §14(b), 38 Stat. 264." https://uscode.house.gov/view.xhtml?req=(title:12+section:355+edition:prelim) — The original Act authorises purchase of U.S. bonds and notes; the amendment history records the Banking Act of 1935 change. ↩

3. Fekete, Antal E. The Mechanism of Capital Destruction. Address to the Committee for Monetary Research and Education, New York, 16 October 2008. https://professorfekete.com/articles/AEFCMRE.pdf — For the falling-rate thesis he acknowledged was the controversial one. ↩

4. Fekete, Antal E. How the Fed Bankrupted the Insurance Industry by Destroying Its Capital Efficiency, 2014. https://professorfekete.com/articles/AEFHowFedBankruptedInsInd.pdf — "Sudden death syndrome," "a dead man walking," "skate on ice twice as thin," and the float-compounding argument. ↩

5. Federal Reserve Board Office of Inspector General. Material Loss Review of Silicon Valley Bank, September 2023. https://oig.federalreserve.gov/reports/board-material-loss-review-silicon-valley-bank-sep2023.pdf — $91bn held to maturity, ~$86bn agency MBS beyond ten years, ~$16.2bn Treasuries, 6.2-year HTM duration. ↩

6. FDIC Quarterly Banking Profile, Q1 2023 (https://www.fdic.gov/news/press-releases/2023/pr23043.html) and Q3 2023 (https://www.fdic.gov/news/press-releases/2023/pr23099.html) — $515.5bn and $683.9bn of unrealized securities losses. ↩

7. Federal Reserve Bank of New York. FAQs: Reserve Management Purchases and Reinvestment Purchases. https://www.newyorkfed.org/markets/reserve-management-reinvestment-purchases-faq — The Desk publishes operation dates, settlement dates, security types, eligible maturity ranges and expected size ranges in advance. ↩

8. Commercial market-research estimates; no regulator publishes a consolidated figure. — Market-structure figures for HFT volume share and algorithmic-trading market size are vendor and market-research output, not regulatory data, and providers disagree by wide margins. JPMorgan's LOXM is the documented case of machine-learning execution at a major dealer. ↩ ↩² ↩³

9. Executive Order 12631, 18 March 1988. https://www.presidency.ucsb.edu/documents/executive-order-12631-working-group-financial-markets — The order's own title is "Working Group on Financial Markets.". ↩

10. National Association of State Retirement Administrators. Public Pension Plan Investment Return Assumptions. https://www.nasra.org/returnassumptionsbrief/ — Average assumption 7.95% (2007) to 6.91% (June 2025 survey, 131 funds). ↩ ↩²

11. Bank of England. An anatomy of the 2022 gilt market crisis. https://www.bankofengland.co.uk/-/media/boe/files/working-paper/2023/an-anatomy-of-the-2022-gilt-market-crisis.pdf — Intervention 28 September to 14 October 2022, £5bn daily raised to £10bn. For the March 2020 Treasury analogue: https://home.treasury.gov/system/files/136/IAWG-Treasury-Report.pdf. ↩

12. Bank of England. "Branches of the Bank of England." Quarterly Bulletin, 1963. https://www.bankofengland.co.uk/-/media/boe/files/quarterly-bulletin/1963/branches-of-the-boe.pdf — Manchester opened September 1826, the second branch after Gloucester. ↩