Chapter 10: Market Failures as Geometric Pathologies

Externalities as Heuristic Corruption

[Empirical.] An externality exists when the price of a good does not include the full cost on the decision manifold. A factory that pollutes a river imposes costs on dimensions d_6 (social impact — downstream communities), d_5 (trust — breach of implicit social contract), and d_8 (legitimacy — violation of environmental norms). If these costs are not reflected in the factory’s output price, the price heuristic h(n) systematically underestimates the true cost-to-go from pollution-intensive production states.

The result is predictable: the market’s A* search, guided by the corrupted heuristic, follows a trajectory that avoids the true geodesic (which would include pollution costs) and instead converges on a cheaper-seeming path that ignores dimensions d_5, d_6, and d_8. This is not market failure in the traditional sense — the market is functioning correctly on the projected manifold. It is a dimensional failure: the price heuristic operates on d_1 while the true cost lives on all nine dimensions.

Pigouvian taxes are heuristic corrections: they add a penalty to the price that approximates the missing manifold dimensions. A carbon tax adds \Delta h(n) = \beta_{\text{climate}} \cdot \text{emissions}(n) to the price heuristic, partially restoring admissibility. The geometric framework explains why Pigouvian taxes are hard to calibrate: the correct penalty depends on the covariance matrix \Sigma (how the missing dimensions interact with d_1), which varies across contexts and is difficult to estimate.

The framework also explains why some externalities are more persistent than others. A negative externality on d_6 (environmental pollution) persists because the affected parties (downstream communities, future generations) are not participants in the market transaction — they cannot bid the price up to reflect their costs. A negative externality on d_5 (trust erosion from deceptive advertising) is partially self-correcting: the eroded trust eventually reduces the firm’s customer base, feeding back into d_1. The geometric prediction: externalities are most persistent when the affected dimension is decoupled from d_1 in the covariance matrix — when the non-monetary cost does not eventually cascade back into monetary consequences. Environmental externalities are persistent because the environmental-monetary coupling is weak and delayed. Trust externalities are less persistent because the trust-monetary coupling is stronger and faster.

Information Asymmetry as Heuristic Degradation

[Empirical.] Akerlof’s market for lemons is a specific distortion of dimension d_9 (epistemic status). When buyers cannot observe product quality, their heuristic estimate h(n) for the quality dimension is unreliable — it assigns average quality to all products, causing the price to reflect average rather than actual value.

On the decision manifold, this appears as a degenerate metric along the quality axis: the Mahalanobis distance between a high-quality and a low-quality product is small (because the buyer’s \Sigma has a large variance on d_9), even though the true distance is large. Good products are undervalued and bad products are overvalued. The geodesic on the true manifold (buy good products, avoid bad ones) differs from the geodesic on the perceived manifold (treat all products as average).

Market institutions — warranties, reputation systems, regulation, certification — function as manifold calibration mechanisms: they reduce the variance on d_9, tightening the metric along the quality axis and bringing the perceived manifold closer to the true one.

The geometric framework classifies information asymmetry into three progressively severe forms:

1. Noise (\sigma_9 large but unbiased): the buyer’s estimate of quality varies around the true value. The heuristic is admissible in expectation but imprecise. Market institutions can calibrate by reducing \sigma_9. This is the case for most consumer goods where reviews, returns, and repeat purchase provide gradual calibration.

2. Bias (\sigma_9 large and systematically shifted): the buyer’s estimate is consistently wrong in one direction. The used-car market is the canonical example: sellers know the quality and buyers do not, so buyers’ estimates are biased downward (they assume average quality, which is below-average because good sellers exit). The heuristic is inadmissible — it systematically overestimates the cost of high-quality goods and underestimates the cost of low-quality ones.

3. Corruption (d_9 is actively manipulated): the seller deliberately distorts the buyer’s epistemic state. Fraudulent advertising, deceptive labeling, and insider trading are not mere information asymmetries — they are active attacks on the buyer’s manifold calibration. The heuristic is not merely inadmissible; it has been weaponized. The seller profits precisely by ensuring that the buyer’s heuristic is maximally wrong.

The progression from noise to bias to corruption corresponds to increasing severity of heuristic failure — and, critically, to different policy remedies. Noise is addressed by information provision (reviews, disclosures). Bias is addressed by institutional design (warranties, reputation systems). Corruption is addressed by prohibition and enforcement (fraud law, securities regulation). The geometric framework predicts that policies designed for one level of severity will fail at another — disclosure requirements (effective against noise) will not stop deliberate fraud (corruption), and fraud enforcement (effective against corruption) is overkill for honest markets with simple noise.

Moral Hazard as Manifold Decoupling

[Empirical.] Moral hazard — the tendency of insured agents to take greater risks because they do not bear the full cost — is a specific form of heuristic corruption where the agent’s perceived manifold decouples from the true manifold.

Consider a bank with deposit insurance. On the bank’s perceived manifold, the cost of risky lending (d_1 losses from loan defaults) is partially borne by the insurer, not the bank. The bank’s effective edge weight for risky transitions is lower than the true edge weight. The bank’s Bond geodesic — the path of least behavioral friction on its perceived manifold — therefore passes through regions that the true geodesic would avoid.

In the geometric framework, deposit insurance modifies the bank’s covariance matrix \Sigma_{\text{bank}} by reducing the variance weight on the risk dimension. The bank experiences a “flatter” manifold in the risk direction — transitions that involve more risk appear less costly — leading to a geodesic that is risk-seeking relative to the true manifold. The standard remedy (capital requirements, risk-based deposit insurance premiums) works by restoring the curvature: adding back the risk penalty that the insurance removed, so that the bank’s perceived manifold more closely approximates the true one.

MARIA’S COFFEE SHOP — HEURISTIC CORRUPTION

Maria experiences a mild form of heuristic corruption every time she checks the real estate listings. The listing prices — $1.2M, $1.8M, $2.5M for commercial spaces near hers — tell her the price heuristic on d_1. They tell her nothing about whether the spaces have community value (d_6), whether the landlords are trustworthy (d_5), or whether the neighborhoods have the social infrastructure that makes a coffee shop viable (d_3, d_7). Maria’s System 1 unconsciously fills in these dimensions using the price as a proxy: “expensive neighborhood = good neighborhood.” But the proxy is corrupted by gentrification dynamics: the neighborhoods with the highest prices may have the lowest community value, because the high prices reflect speculative demand, not social infrastructure. Maria’s heuristic is inadmissible on the full manifold — and the inadmissibility is generated by the price system itself.

Regulatory Capture

[Empirical.] Regulatory capture occurs when the regulator’s objective function shifts from public welfare (the full manifold) to the regulated industry’s preferences (a projected subspace). In the geometric framework, this is objective hijacking: the search is guided by a different objective than the one it was designed to optimize.

The captured regulator’s decision manifold has been contracted: dimensions d_3 (fairness to the public), d_6 (social impact), and d_8 (institutional legitimacy) are suppressed, and the regulator’s effective objective becomes alignment with d_1 interests of the regulated industry. The geodesic shifts from the public-welfare path to the industry-welfare path.

The geometric framework predicts that capture is easier when the public-welfare dimensions (d_3, d_6, d_8) are diffuse (many citizens each affected a little) and the industry dimensions (d_1) are concentrated (few firms each affected a lot). This is because concentrated interests produce steep gradients on the captured manifold, while diffuse interests produce flat gradients — the regulator’s corrupted heuristic follows the steep gradient.

Short-Termism

Quarterly earnings pressure is objective hijacking in the temporal dimension. The firm’s true geodesic optimizes over the full temporal manifold — including long-term trust (d_5), institutional legitimacy (d_8), and social capital (d_6). The quarterly proxy contracts the optimization horizon, collapsing the temporal manifold to a single period and destroying the geodesic’s long-term structure.

The geometric prediction: short-termism is most damaging for firms whose value depends heavily on evaluative dimensions (d_5–d_9) — technology companies (trust in data handling), financial institutions (trust in solvency), and community-serving businesses like Maria’s coffee shop. For commodity producers whose value is primarily d_1, short-term optimization is less distortionary because the relevant manifold is closer to the scalar projection.

Theorem 10.1 (Temporal Hijacking and Evaluative Erosion). Let V_E(t) = \sum_{k=5}^{9} d_k(t) be the total evaluative value at time t. A firm optimizing d_1 on a single-period horizon will, if the off-diagonal covariance terms between d_1 and d_5–d_9 are positive, deplete V_E over time:

\frac{dV_E}{dt} < 0 \quad \text{when } \text{Cov}(d_1, d_k) > 0 \text{ and the agent maximizes } d_1^{(\text{short-term})}

Proof sketch. When short-term d_1 gains come at the expense of evaluative dimensions — which they do when the covariance structure implies that quick monetary gains require trust erosion, community degradation, or identity compromise — the evaluative dimensions decline. The positive covariance means that maintaining d_k is positively associated with d_1 in the long run, but the short-term optimizer sacrifices the long-run d_1 benefit for immediate gain. Each period, the evaluative stock V_E falls; eventually, the erosion cascades back into d_1 as customer defection, regulatory action, and reputational collapse. \square

[Empirical.] Boeing’s shift from engineering-led to finance-led management in the 2000s is a textbook case of temporal hijacking. Each quarterly decision to reduce inspection procedures (d_8: legitimacy), accelerate production timelines (d_4: worker autonomy compressed), and prioritize stock buybacks over R&D (d_9: epistemic investment sacrificed) improved short-term d_1. The cumulative evaluative erosion materialized as the 737 MAX crisis — a catastrophic boundary crossing on d_2 (safety rights), d_5 (trust with regulators and airlines), and d_8 (institutional legitimacy) that cost tens of billions on d_1 and 346 lives on d_2.

Goodhart’s Law as Objective Hijacking

[Modeling Axiom.] Charles Goodhart’s observation — “when a measure becomes a target, it ceases to be a good measure” — is, in the geometric framework, a theorem about objective hijacking. The mechanism is precise:

1. A multi-dimensional objective (the full-manifold geodesic) is measured by a scalar proxy (a specific contraction).
2. Agents are incentivized to maximize the proxy.
3. The Scalar Irrecoverability Theorem (Chapter 6) guarantees that the proxy discards information.
4. Agents exploit the discarded dimensions: they improve the proxy while degrading the unmonitored dimensions.
5. The proxy diverges from the original objective.

University rankings maximize a scalar index (combining research output, student selectivity, alumni donations). The index becomes the objective. Universities then optimize the index — recruiting international students for selectivity scores, encouraging faculty to split papers for citation counts, investing in amenities that attract high-income alumni — while the actual educational mission (d_9: genuine knowledge creation, d_4: student intellectual autonomy, d_6: community contribution) degrades. The index goes up. The education goes sideways. This is not a bug in the ranking system. It is the mathematically inevitable consequence of replacing a tensor with a scalar and then optimizing the scalar.

MARIA’S COFFEE SHOP — OBJECTIVE HIJACKING

The REIT that owns Maria’s building is managed by a property management firm whose CEO’s compensation is tied to the REIT’s quarterly distribution per unit. The CEO’s objective has been hijacked from “maximize long-term property value on the full manifold” to “maximize quarterly cash distribution.” This objective favors raising rents to market rate immediately (d_1 maximized this quarter) over maintaining tenant relationships (d_5), preserving neighborhood character (d_6), and building long-term occupancy stability (d_8). The CEO is not corrupt. The CEO is optimizing the objective placed before them. The corruption is in the objective, not the optimizer.

Asset Bubbles as Overvalued Basins

[Modeling Axiom.] An asset bubble is a local minimum in the economic decision manifold — a configuration that appears optimal on the projected manifold (d_1 shows rising values) but is suboptimal on the full manifold (d_5 trust is eroding, d_9 epistemic quality is degrading, d_8 institutional legitimacy is strained).

The bubble basin has depth: the more participants invest, the deeper the basin becomes (positive feedback). Exiting the basin requires a coordinated jump — a perturbation large enough to overcome the basin walls. But individual agents face a coordination problem: the first to exit pays the highest cost (selling at the peak of the bubble looks like leaving money on the table on d_1).

The geometric characterization: a bubble exists when the metric on the perceived manifold (where asset values look high) diverges from the metric on the true manifold (where the underlying value does not support the prices). The divergence is measurable — it shows up as increasing curvature on the true manifold, as small perturbations produce increasingly large responses. Chapter 11 develops this into a full singularity analysis.

Poverty Traps as Low-Income Basins

[Empirical.] A poverty trap is a basin of attraction on the economic manifold from which the escape energy exceeds what agents trapped within it can accumulate. The basin is defined by low d_1 (low income), constrained d_4 (limited autonomy — can’t invest in education, can’t move to opportunity), degraded d_9 (limited information about opportunities), and high boundary penalties on the paths leading out (education costs, relocation costs, credential requirements).

The geometric prediction: poverty traps are deepest when multiple dimensions reinforce each other — low income (d_1) constrains autonomy (d_4), which limits information (d_9), which prevents discovery of escape paths. Breaking a poverty trap requires simultaneous intervention on multiple dimensions, not just income transfer (d_1 alone). This is precisely what the empirical evidence on effective anti-poverty programs shows: the most successful interventions combine cash transfers (d_1), skills training (d_9), asset provision (d_2), and social support (d_5, d_6).

Theorem 10.2 (Multi-Dimensional Poverty Trap Depth). The depth D(s^*) of a poverty trap at state s^* is a function of the number of simultaneously constrained dimensions. If k dimensions are constrained (below their boundary thresholds), the basin depth scales superlinearly:

D(s^*) \propto k^2 \cdot \bar{\beta}

where \bar{\beta} is the average boundary penalty for crossing the constraint boundaries. The quadratic scaling arises from the Mahalanobis structure: when multiple dimensions are constrained, the covariance matrix amplifies the effective distance because displacements on constrained dimensions interact.

[Empirical.] This prediction is testable. Banerjee and Duflo’s (2011) evidence on “graduation programs” — multi-dimensional anti-poverty interventions — shows that simultaneous intervention on income (d_1), assets (d_2), training (d_9), and social support (d_5) produces persistent escape from poverty, while single-dimension interventions (cash alone, training alone) produce only temporary improvement. The geometric explanation: single-dimension interventions move the agent along one axis of the manifold but do not escape the basin, because the other constrained dimensions pull the agent back. Multi-dimensional interventions move the agent diagonally across the basin, exploiting the covariance structure to escape with lower total energy.

The Mahalanobis geometry is key. If the covariance matrix \Sigma has large positive off-diagonal terms between d_1 and d_9 (income and information are correlated — rich people have more information), then a diagonal move (simultaneously improving income and information) covers more Mahalanobis distance than two sequential single-dimension moves of the same total magnitude. The diagonal move escapes the basin; the sequential moves do not. This is why integrated anti-poverty programs work and single-instrument policies often fail.

Herding as Heuristic Convergence

[Empirical.] Asset bubbles are sustained by herding — the tendency of agents to follow the crowd rather than their private information. In the geometric framework, herding is heuristic convergence: agents replace their individual heuristic h_i(n) with the market consensus heuristic \bar{h}(n), even when h_i and \bar{h} disagree.

On the decision manifold, herding flattens the heuristic field: instead of n independent heuristic estimates pointing in different directions (which, by the wisdom-of-crowds effect, tend to cancel out individual errors), all agents follow the same estimate. If the consensus estimate is wrong — if \bar{h}(n) systematically underestimates the cost of the asset-price path — then all agents simultaneously follow a suboptimal trajectory, creating a coordinated deviation from the geodesic that individual agents could not produce.

The bubble pops when a perturbation (a margin call, a credit downgrade, a piece of bad news) pushes some agents’ g(n) high enough that the sum f = g + h exceeds the cost of exiting the basin. These agents exit. Their exit raises the cost for remaining agents (falling prices increase g). More agents exit. The cascade is the A* search, restarted with updated costs, converging on a different geodesic — one that leads away from the bubble rather than deeper into it.

The Sunk Cost Fallacy as Path-Dependence Violation

[Empirical.] The sunk cost fallacy — the tendency to continue investing in a losing project because of what has already been spent — is a gauge violation of a specific type: it violates re-description invariance along the temporal dimension.

The BIP requires that the evaluation of a decision depend only on the current state and the available future paths, not on the path that led to the current state. A decision between “continue” and “abandon” should be evaluated from the current node forward (g(n) measured from v_0 to the current state is fixed and the same for both options). The sunk cost fallacy introduces g_{\text{past}} — the cost already incurred — into the evaluation of future paths, where it does not belong.

In A* terms, the fallacy is a bookkeeping error: the agent includes the sunk cost in g(n) for the “abandon” path but not for the “continue” path, making abandonment appear more costly than it actually is. The correct A* comparison evaluates f(n) = g(n_{\text{current}}) + h(n) for both options, using the same g(n_{\text{current}}). The sunk cost is in g(n_{\text{current}}) and cancels out of the comparison.

The geometric reading: the sunk cost fallacy is a violation of gauge invariance under temporal re-description. The same current state, described as “we have invested $2M and need $1M more” versus “we need $1M to complete a project,” produces different decisions. The first description activates a path-dependent evaluation; the second does not. The economic content is identical. The violation is measurable: agents who are given the “sunk cost” framing are approximately 35% more likely to continue than agents given the “fresh start” framing, even when the projects are identical from this point forward (Arkes and Blumer, 1985).

Currency-Denomination Effects

[Empirical.] Raghubir and Srivastava (2002) demonstrated that tourists spend more when prices are denominated in a foreign currency with a large numeric value relative to their home currency. A Japanese tourist in Indonesia, seeing prices in thousands of rupiah, perceives them as “large numbers” even though the amounts are trivial in yen terms. The evaluation depends on the denomination — a pure gauge transformation that should leave the evaluation invariant.

The geometric characterization: the agent’s heuristic function h(n) is not gauge-invariant. It responds to the numeral attached to the price, not to the real value the numeral represents. The gauge violation tensor V_{ij} (Chapter 8) can be measured directly: present the same basket of goods in two currencies and compare purchase decisions. The difference is the gauge violation. It is large for unfamiliar currencies and small for familiar ones — suggesting that the gauge invariance of the heuristic is learned through experience, not innate.

Why Single-Instrument Policies Fail

The four-pathology taxonomy explains why single-instrument economic policies often fail to address market failures. A carbon tax addresses heuristic corruption (correcting the price signal on d_6) but does not address objective hijacking (if the regulators setting the tax rate have been captured by the fossil fuel industry), local minima (if the economy is locked into carbon-intensive infrastructure), or gauge breaking (if agents confuse nominal and real costs of the transition).

Effective policy addresses all active pathologies simultaneously. For climate change:

• Heuristic repair: Carbon pricing (tax or cap-and-trade) corrects the price signal.
• Objective protection: Independent climate agencies with multi-dimensional mandates (not just GDP growth) resist capture.
• Basin escape: Green infrastructure investment overcomes the lock-in of carbon-intensive capital.
• Gauge correction: Real-cost accounting that includes the social cost of carbon in all government projections.

The geometric framework does not prescribe specific policies. It provides the diagnostic taxonomy that identifies which pathologies are active and therefore which policy instruments are necessary. The diagnosis precedes the prescription — and the diagnosis requires the full manifold, not the scalar projection.

Market Failures as Information About the Manifold

A final and perhaps counterintuitive observation: market failures are informative. They reveal the shape of the decision manifold.

When the price system fails to allocate efficiently, the failure is diagnostic. An externality reveals a dimension (d_6: social impact) that the price heuristic does not capture. A bubble reveals a local minimum whose basin structure can be characterized. Regulatory capture reveals a dimension (d_8: institutional legitimacy) whose boundary penalties have been suppressed. Money illusion reveals a gauge violation whose magnitude can be measured.

Each failure is a data point about the manifold. A perfectly efficient market — one where the price system captures all relevant information and the A* search follows the geodesic — would tell us nothing about the manifold structure beyond d_1. Market failures, by revealing the gap between the price projection and the true manifold, map the dimensions that the price system misses. The program of behavioral economics — cataloguing “anomalies” — is, in this light, an empirical mapping of the decision manifold, conducted without the geometric vocabulary to describe what is being mapped.

This book provides the vocabulary. The catalog of market failures becomes a map of manifold structure. The anomalies become data. And the data points toward the same conclusion as the conservation laws, the gauge invariance, and the BGE: the economy is not a scalar. It is a nine-dimensional geometric object, and the scalar projection — useful as it is — is not the economy. It is its shadow.