Slides
Lecture slides for each day
The deck used in the lecture, as a PDF. Day 1 runs to 103 pages and asks when an asset loss becomes a credit contraction, following one institution from its balance sheet to its capital ratio and then to the credit it can support. Days 2 to 4 are released as the course reaches them. Each deck carries a technical appendix with the derivations in full.
The day opens on a puzzle: Bear Stearns reported $395bn of assets against $384bn of liabilities, so a loss of about three per cent would exhaust its equity. Leverage turns a small asset mark into a large equity loss, and duration decides how large that mark is. But accounting alone does not determine lending. The lecture builds an agency model in which the intermediary must stay credible to its own funders, derives a debt limit from an incentive constraint, and then closes the model with a funding-supply schedule so that lower net worth reduces equilibrium credit rather than merely reducing a ratio. It ends on capital repair: shrink, issue, or bail in, and which diagnosis each instrument actually fits.
Sections. Losses and the equity buffer; Why intermediary net worth matters; The incentive constraint and debt limit; Equilibrium amplification; Capital repair and policy.
Day 1 asked whether promised claims are supportable over the life of the assets. Day 2 changes the clock and asks whether payments due now can be met without destroying long-run value. Silicon Valley Bank received $54bn of outbound wire requests in a single day. The lecture separates solvency from liquidity as two different inequalities, derives the Diamond and Dybvig run equilibrium and its withdrawal threshold, then adds what the classical model leaves out: duration losses, concentrated uninsured funding, secured borrowing and execution speed. It closes on contagion channels and on which intervention repairs which margin, and who bears the loss.
Sections. Liquidity and maturity transformation; Liquidity insurance and the run equilibrium; Modern run technology; Contagion and systemic fragility; Liquidity policy and loss allocation.
Bank surveillance uses quantitative signals to direct scarce analyst attention. The question is whether dated public information can help allocate that attention before a material funding event occurs. The lecture fixes the empirical contract first: the population, the public-information cutoff, the target, the measurements, the evaluation and the permitted use, all before any weight is learned. It then builds one transparent probability rule from four financial measurements, derives its loss and its single signed error, and tests it in financial time against two serious baselines. The authorised output is a review priority, not a diagnosis.
Sections. The prediction decision; One dated bank-quarter; One probability rule; Future performance; Evidence review.
A probability contains an ordering, not a financial explanation. Day 4 inherits the frozen alert and asks what dated evidence supports, contradicts or qualifies the hypotheses behind it. The lecture reads one real filing by hand first, reproducing a scope-qualified uninsured-deposit ratio and reconciling carrying value against fair value, so that a manual benchmark exists before anything is automated. Only then does it introduce retrieval, narrow calculation tools, the SEC filing API and a bounded controller. Generation begins after the support set is assembled, and abstention is the correct answer when decisive evidence is absent.
Sections. Question and evidence; Worked financial record; Grounded language assistance; APIs and bounded tools; Evaluation and authority.
