Bernd Johannes Wuebben
Asset Pricing Theory · AI · Mathematics · Theoretical Physics
A price is a claim about the future that somebody was willing to fund. Most of my work asks what can be known about such claims: whether a model that earns is a model that explains, where compensation for risk ends and compensation for absorbing other people’s constraints begins, and why the part of the cross-section that pays is so often the part nobody can name.
These are one question approached from two sides. Performance and identification are governed by the same covariance spectrum, and they are complementary rather than aligned: a high-Sharpe stochastic discount factor need not identify priced risk, and the alpha tends to live in the statistically unnameable bulk, at a floor set by statistical power rather than by anyone’s skill.
That pushes outward rather than inward — toward who is on the other side of a trade and what they are solving for, toward markets as aggregation mechanisms that sometimes aggregate belief and sometimes only aggregate attention, and toward the possibility that the discretionary–quantitative divide is not a matter of technique but of what a person is willing to accept as warrant for a belief.
Separately, and for its own sake, I work in geometric topology and algebraic geometry — four-manifolds and instanton Floer theory, mirror symmetry, and the geometry of toric Fano varieties and Calabi–Yau threefolds — and on black hole information, where the same random-matrix universality that governs high-dimensional return panels reappears in the Page curve. A newer strand asks what large-scale artificial intelligence does to markets, institutions, and the balance of power between states.
Day to day, I run the systematic investing business at one of the world’s largest asset managers.
For in the end, what we call awareness may simply be the most efficient way for entropy to remember itself.
Research interests over the years
- Asset pricing theory — stochastic discount factors and the statistics of high-dimensional return panels; the spectral structure of the cross-section, and whether a high-Sharpe model identifies priced risk or only describes it; the equilibrium level of factor compensation; information-theoretic foundations, from Kelly and Grinold to portfolios as codes; the demand side — inelastic markets, intermediary capital, who is on the other side; and how factors are born, crowd, and decay.
- AI and the machine economy — institutional cognition as competitive advantage, prediction markets as informational infrastructure, and what disciplined discovery looks like once hypotheses are free and only validation is scarce.
- Geometric topology — exotic smooth structures on four-manifolds, instanton knot homology and the Atiyah–Floer program, and turning classical existence questions into finite, machine-certified computations.
- Gauge theory and moduli theory — Donaldson and Seiberg–Witten invariants, and the architecture beneath them: when the passage from an equation to a moduli space to a compactification to an algebra can be run in reverse.
- Algebraic geometry — mirror symmetry, slope stability of tangent bundles on toric Fano varieties, deformations and smoothability of Calabi–Yau threefolds, and open FJRW theory.
- Theoretical physics — quantum gravity, holography, black hole thermodynamics and the information paradox, quantum information, random-matrix theory and free probability, critical phenomena and universality.
- International relations — how large-scale AI deployment re-centres physical constraint as a source of power, and how delegating decisions to machines erodes deterrence by blurring authorship and intent.
Papers
Markets and quantitative finance
- The Validation Bottleneck: Alpha Discovery When Hypotheses Are Free What happens to discovery when generating hypotheses costs nothing and only validation is scarce. Two pre-registered experiments on public data. SSRN, July 2026 · replication code
- The AI-Native Asset Manager: Institutional Cognition as Competitive Advantage As frontier AI commoditizes information and analysis, the scarce resource shifts from intelligence to institutional epistemology — the architecture by which a firm generates, challenges, sizes, and learns from its investment beliefs. Returns decompose into narrative beta, cross-sectional factor alpha, and idiosyncratic issuer alpha, each demanding a different way of knowing, and the durable advantage proves to be a velocity rather than a position. SSRN, April 2026
- Beyond De Prado and Cotton: Hierarchical and Iterative Methods for General Mean-Variance Portfolios Hierarchical Risk Parity and its Schur-complement generalization are signal-blind: both solve only the minimum-variance problem and cannot carry an arbitrary expected-return forecast. Three methods lift the framework to the general case at the same cost — HRP-μ, HRP-Σμ, and CRISP, a Gauss–Seidel shrinkage solver proved unconditionally convergent. arXiv:2604.23833 · replication code
Mathematics
- An exotic S²×S² and an exotic ℂP²#ℂ̄P² Upgrading Lidman–Piccirillo’s slicing-distinguished four-manifold pair from equal cohomology rings to homeomorphic is equivalent to producing an exotic S²×S². The paper closes that gate by proving π₁ = 1 for one explicitly parametrized piece, via coset enumeration and Knuth–Bendix completion with complete certificates. arXiv:2608.17267 · scripts and certificates
- The instanton homology of the (−2,3,q) pretzel knots and Maurer–Cartan deformations in the two-arc algebra of the pillowcase For the pretzel family P(−2,3,q) the rank of instanton knot homology I♮ is q+2, and naive pillowcase Floer homology misses it by exactly one differential, with the sign set by the determinant. The bounding cochains that repair the deficiency are computed — the first explicitly nonzero ones in pillowcase Floer theory, realizing both directions of the conjectured correction inside a single family. arXiv:2607.26096 · code
- Traceless SU(2) characters and ℤ/4 instanton gradings for two-bridge and (3,n)-torus knots The traceless SU(2) character varieties and their ℤ/4 instanton gradings, computed explicitly for the two-bridge and (3,n)-torus families. These are the input on which the pretzel-knot bounding-cochain computation rests. arXiv:2607.26095 · code
- Slope stability of tangent bundles of smooth toric Fano varieties A Klyachko-criterion sweep with exact certificates over all 8,630 smooth toric Fano varieties in dimensions three through six. Every Kähler–Einstein variety in the census has polystable tangent bundle, but not conversely — 102 of the 109 five-folds with stable tangent bundle are not Kähler–Einstein. arXiv:2608.20411 · certificates and data
- Smoothability of Calabi–Yau threefolds A five-paper series on when a singular Calabi–Yau threefold admits a smoothing, in Batyrev's setting of anticanonical hypersurfaces in Gorenstein toric Fano fourfolds. A sweep of the full Kreuzer–Skarke classification finds that 8.27% of all 473,800,776 reflexive four-polytopes carry a non-smoothable face, and that exactly 590 leave both the threefold and its Batyrev mirror isolated-singular — among them a unique mirror pair which a vanishing-cycle obstruction then proves non-smoothable on both sides, one for local and one for global reasons. In the positive direction, an explicit deformation of the ambient toric pair smooths what can be smoothed, and a combinatorial locking condition on a facet says when that mechanism cannot reach a germ at any degree. The two halves meet at first order; what separates them is integrability. Papers in preparation · code, data, and the five papers
- The central-charge threshold for wall-crossing in two-variable open FJRW theory The open FJRW invariants of xʳ + yˢ are canonical — no primary wall-crossing — exactly when the central charge falls below one, which is to say exactly for the simple singularities A, D₄, E₆ and E₈. The first walls of every non-simple pair are classified, and each wall-crossing generator is identified as a Hamiltonian vector field.
Theoretical physics
- The Page Curve Through the Transition: Testing the Diagonal Approximation Against the Exact Replica Spectrum Both sides of the diagonal approximation are computable in the west-coast model, so this computes them. It reproduces the shape of the broadened crossover to about one per cent of the curve’s amplitude but overstates the entropy deficit at the transition by roughly ten per cent, and summing all non-crossing permutations rather than the two dominant saddles yields a Marchenko–Pastur law for the radiation spectrum. SSRN, March 2026
- The Page Curve in Higher Dimensions: Phase Transitions, Critical Phenomena, and Universal Broadening The width of the Page transition is set by the square root of the heat capacity — a result derived away from criticality, which an AdS black hole violates, since it has a phase diagram. The width proves independent of spacetime dimension for large Schwarzschild–AdS, is parametrically enhanced near the Hawking–Page transition, and diverges with a different exponent at the Reissner–Nordström critical point — so the broadening discriminates between the two transitions. SSRN, March 2026
- The Page Curve is not Enough: Universality, Ensemble Dependence, and Microscopic Ambiguity in Black Hole Information Recovery Reproducing a Page curve does not identify the microscopic resolution of the information paradox. A four-level universality hierarchy separates what is forced from what is diagnostic: any two scrambling dynamics forming approximate unitary designs with matching thermodynamics agree on everything through level three, so fuzzballs, islands, random unitary evolution, state-dependent interiors and ER=EPR are indistinguishable there and diverge only at the fine-grained fourth level. SSRN, April 2026
- Informational Singularities and the Completion of Gravity: Replica Wormholes, Indeterminism, and Cosmic Censorship Replica wormholes restore a coherent map from initial to final states without removing any curvature singularity — informational completion without geometric completion. Making that distinction precise yields the notion of an informational singularity, logically independent of the geometric one and coming apart from it in both directions, and motivates recasting strong cosmic censorship as a claim about coherent state maps rather than about Cauchy horizons. SSRN, August 2026
AI and the machine economy
- Prediction Markets as the Informational Substrate of the Machine Economy The three institutional gaps blocking autonomous machine-to-machine commerce — adjudication, reputation, and liability — share one root: there is no market price for agent performance risk, deliverable quality, or failure correlation. Five prediction-market mechanisms are designed to supply those signals, with the instrument delimited as carefully as it is developed: an informational deficit can be priced, but a primitive needing an agreed vocabulary rather than a resolving fact needs coordination instead. SSRN, April 2026
- The Institutional Plumbing for a Machine Economy: A Readiness Assessment of Building Blocks, Missing Infrastructure, and Commercial Opportunities Software agents can now pay, but paying is not economic agency: open exchange between agents with no prior relationship also needs identity, delegated authority, bonding, verification, adjudication, settlement, portable reputation, and somewhere for residual loss to land. Auditing that stack against what actually shipped through 2026 — where roughly two-thirds of settlements moved inside linked clusters and a fifth were fictitious — gives the conclusion that machine commerce expands along a verification-and-risk frontier, not a payment frontier. SSRN, March 2026
International relations and politics
- Material Deployment Power in the Age of Artificial Intelligence: How Physical Constraints Reshape International Order Prevailing accounts treat AI as an informational technology; deployment at scale instead re-centres physical constraint, since it depends on energy, transmission, cooling, and critical materials that expand slowly and unevenly. The paper defines material deployment power — the capacity to mobilize those physical systems under rivalry — and shows it shapes trade policy and alliance behaviour independently of innovation leadership. SSRN, April 2026
- Algorithmic Plausible Deniability: The Structural Erosion of Deterrence in the Age of Artificial Intelligence Delegating decisions to machines generates persistent uncertainty over authorship and intent, weakening deterrence even between rational states that prefer stability; unlike strategic ambiguity it is not chosen, but emerges from the delegation itself. The effects are asymmetric — hostile acts become more deniable while signals of restraint become less credible, which undermines escalation control. SSRN, April 2026