Every number on this page is computed by the Volume II Chapter 13 notebook's own code: the desk's record (seed 1300), Chapter 9's 100,000 paths (seed 8) and 100,000 fresh paths (seed 26209), the fresh programs' readings drawn with the lab's seed 26213. Your notes stay on this page until you leave it; copy the memo before you go.
The market desk proposes to sell the board a learned warning of funding-window closures. The audit committee will not pay for it until it knows what the warning can be worth, at which threshold it should be read, and whether its promise survives fresh data. You support the team preparing the answer.
Use AXIOM to learn the warning from the desk's record, price it by the packages it allows, check it on fresh paths, and certify it.
Worked Example 13.1: the desk reads five funding-market indicators when the program starts, in standard units: Meridian's bond spread, the high-yield spread, equity volatility, banks' lending standards, and the bank-funding spread. A common financial-conditions factor drives them; programs that will see a closure start higher. The treasury and the transformation office will sign two packages, one for "warning" and one for "all clear", with the 5 percent appetite counted over both; the audit committee certifies at one-sided 95 percent confidence. A perfect warning would be worth $4.12 million (Chapter 9).
In the record panel, compare the readings of windows with and without a closure, and compute the ceiling: the most a warning on these readings can be worth.
The best warning reads the declared law's log-likelihood ratio of a closure; each threshold splits programs into warned and all clear, and each group gets its own package with the 5 percent appetite counted over both. The curve scales the signal in all five readings.
In the workbench, fit a logistic model and a boosted model; compare their scores on the record, on held-out folds, and on fresh programs.
In the integration bench, sweep the threshold; mark the threshold of best accuracy and that of best value.
The logistic warning on the book's 100,000 paths, each group's closure share as the law gives it; accuracy and the check are on the fresh paths, each program with its own readings.
In the evaluation suite, check the trusted, held-out, and certified designs on fresh paths, and repeat on 200 records.
Warned: share of the 100,000 fresh programs warned; packages: line ($ million), delay (quarters); promised value: Volume I's $10.44 million less the expected cost the design expects; fresh: the true chance of a breach and the value on the fresh paths ($ million). ᵃ Each group's closure share as the law gives it; the other logistic and boosting rows estimate it from the desk's record. A fresh value is shown only where the fresh breach is near the appetite.
In the components and attribution panels, price a warning on one index and on subsets of the readings, and compare the interventional and conditional attributions of one warned reading.
A ceiling well below a perfect warning (Theorem 13.2(iii)); a valuable threshold that is less accurate than never warning (Theorem 13.2(ii)); a perfect record score that breaks the appetite (Theorem 13.1); certified designs that keep it (Theorem 13.3); and an index that keeps the variance and loses the warning (Proposition 13.3(ii)).
What the warning can be worth, the threshold and packages it should use, its certified value and fresh check, and what the desk should record next.