Book 6 · Behavioral Finance & Digital Ethics  ·  Book 6 index · The Forced Urgency Gap (wp32) · Ethics of Algebra (wp27)
Book 6 · Working Paper 56 · GELSI · prepared for the DECS dialogue

Algorithmic Urgency

AI credit and the governance of an invisible wealth transfer — what happens when a lender can see the shadow price a regulator cannot.
Pablo Nogueira Grossi · Independent Researcher · G6 LLC, Newark NJ
ORCID 0009-0000-6496-2186 · a GELSI extension of wp32, "The Forced Urgency Gap"
Why this is here, and open
This chapter was written for the conversation the inaugural Digital Ethics Center Symposium (DECS, Oct 16, 2026) exists to have — the Governance, Ethical, Legal, and Social Implications of digital technologies. Rather than gate it behind a submission process, it is published openly and shared with those who may attend. The dialogue need not wait for anyone's permission. Claims are tagged: Verified theory + evidence · Model a formal lens · Prospective a forward governance claim · Open not yet evidenced, and flagged as such.

1 · The misdiagnosis

When people sell assets at a loss in a downturn, the dominant behavioral account — prospect theory — locates the pathology in the seller's mind: a reference point, a kink, loss aversion. The companion paper (wp32) relocates it to the seller's balance sheet. The distinction is not semantic. One diagnosis prescribes investor education; the other prescribes liquidity — and only the second moved markets when tested at scale. It matters ethically because it changes who is held responsible for the loss, and what intervention counts as just. Verified

2 · The model — urgency as a shadow price

Take agents with entirely standard preferences (no kink, no reference point) who hold an asset while carrying a non-deferrable commitment \(\bar c\) — rent, mortgage, food — whose breach carries a default cost \(\chi\) (eviction, foreclosure), and who face an income shock with probability \(f\). Define urgency as \(\lambda\), the Lagrange multiplier (shadow price) on the liquidity constraint, strictly positive exactly when

\[ \text{cash} + \text{income} + \text{credit} \; < \; \bar c \quad\Longrightarrow\quad \lambda > 0. \]

A forced sale is any sale at \(\lambda>0\) that would not happen at \(\lambda=0\) with beliefs and preferences held fixed.

Proposition 1 — the loss is not a preference
A constrained agent optimally transacts at any positive price, including below cost and below fundamental value. The realized loss therefore carries zero information about preferences — it is a fact about the constraint set, not the psyche. Model

3 · Theorem 1 — non-identification (the blind spot)

Theorem 1
On the data oversight actually observes — trades, prices, paper gains and losses, with the liquidity state \(\lambda\) latent — prospect theory and standard-preferences-under-liquidity-constraint are observationally equivalent in crisis states. Income shocks correlate with drawdowns, so forced sales cluster in the loss domain, and a fitted "loss aversion" parameter simply absorbs that clustering. Preferences are not identified without observing \(\lambda\); identification requires an instrument that shifts liquidity independently of returns. Verified

This is the ethical crux stated as mathematics: the harm is invisible in the standard data, and a plausible-but-wrong story — bad investor psychology — is always available to explain it away. Hold that sentence; the digital turn is built on it.

4 · Theorem 2 — cascade and the seed

Forced sales are not isolated. With limited arbitrage, price satisfies \(p = \mu - \kappa S\), where forced supply \(S\) itself rises as price falls (margin, collateral, refinancing). The loop yields amplification

\[ A = \frac{1}{1-\rho}, \qquad \rho = \kappa\left|\frac{\partial S}{\partial p}\right|, \]

which diverges as \(\rho\to 1\) and produces discontinuous fire-sale equilibria past a threshold. The mark-to-market loss is borne by all holders, not only the shocked fraction \(f\). Model

Corollary — the seed
An ex-ante transfer of the minimal shortfall — order \(\$10^2\!-\!10^3\) per household (one bridged mortgage payment; median eviction arrears) — can set forced supply to zero and prevent the amplified loss. With published parameters (≈27% forced-sale discount and ≈1% price spillover per nearby foreclosure; ≈$78k all-in social cost per foreclosure; peak distress ≈ 1 in 10 U.S. mortgages, 2010), the ratio of cascade cost to triggering shortfall runs to order \(10^2\) per event before spillovers. Verified (literature-parametrized)

The amplification is the crisis; the shortfall is only its seed. The same mechanism supplies a concrete micro-foundation for \(r > g\): those forced to sell into the trough transfer wealth to those liquid enough to buy it, every cycle. Model

5 · The digital turn — what AI changes

Everything above is a documented, pre-digital mechanism. The reason it belongs at a digital-ethics forum is what algorithmic credit does to it.

Modern AI credit and marketing systems ingest high-frequency behavioral and financial signals — cash-flow volatility, missed-payment precursors, app and search behavior, the calendar of obligations. In the model's language, they estimate \(\lambda\) in real time, per person. The state that Theorem 1 proves is latent to the regulator becomes an observable to the lender. That single asymmetry — \(\lambda\) visible to the pricing system, invisible to oversight — converts a diffuse, ambient wealth transfer into a targeted, timed, and deniable one. Prospective

Governance. Offer, price, and timing are driven by an estimated internal state that leaves no audit trail separating "priced for risk" from "priced for desperation." Oversight cannot see the variable doing the work.
Ethical. Extraction is conditioned on vulnerability itself. Proposition 1 says the constrained accept any terms; a system that detects \(\lambda>0\) and acts then is not meeting a willing counterparty but a cornered one. Consent under \(\lambda>0\) is compromised by construction.
Legal. Fair-lending doctrine polices protected classes and disparate impact on observable attributes. \(\lambda\)-conditioned pricing discriminates on a latent state correlated with hardship, and the lender's defense — "this is risk-based pricing" — is precisely the observational equivalence of Theorem 1. The law has no handle on a variable it cannot see and the defendant can always re-describe.
Social. The mechanism concentrates the \(r>g\) transfer and aims it at the liquidity-poor, at a scale and precision earlier lenders could not achieve. A distributional engine with an accelerator.

6 · Governance by observability

The model names its own remedy, and it is not "educate borrowers." Two levers.

Make \(\lambda\) observable to oversight. Because the entire harm rides on an asymmetry of visibility, the governance target is auditability: mandated logging of the features and the timing that drive an offer, so distress-conditioning can be told apart from risk-pricing after the fact. This connects to a broader program of mine on verifiable, auditable AI — machine-checked guarantees about what a decision actually depended on — as a governance instrument rather than a compliance afterthought. Prospective

Bridge liquidity — the seed. If small, timely transfers set forced supply to zero, the just and efficient intervention sits upstream of the sale, not downstream of the loss. The policy object is the shortfall, not the "irrational" seller. Model

7 · Contribution, status, and an honesty line

This is a summary of completed work, deposited and citable, grounded additionally in the author's years on the sell side as a registered representative — where the demand this machinery serves ("the best return, in the fastest time, with the least risk") is stated in a sentence and satisfied by no asset, only by payoff shapes engineered to appear to satisfy it.

Two lines I hold, and that a digital-ethics venue should demand. The base mechanism — non-identification, cascade, the seed, the \(r>g\) micro-foundation — is argued from theory and macro evidence (1971–2026) and is the settled contribution. Verified The algorithmic-amplification claim — that AI credit systems already estimate and price on \(\lambda\) — is a prospective governance argument: the capability plainly exists; comprehensive public evidence of its deployment is still Open, and I flag it rather than assert it. Separating the two is itself the method: the paper models the exact blind spot — a real harm a plausible wrong story can hide — and then refuses to commit the same overreach in its own claims.

Full working paper: "The Response Gap" — the formal version, with the evidence-and-identification section and the measurement agenda — is deposited at 10.5281/zenodo.21752834.
Read next: wp32 · The Forced Urgency Gap (the theorems in full) · wp27 · The Ethics of Algebra.
Keywords: algorithmic credit · liquidity constraints · identification · wealth transfer · auditable AI · GELSI · fair lending · r>g.