Every number on this page is computed by the Volume II Chapter 9 notebook's own code: designs on the book's 100,000 paths (Volume I's seed 8), checks on 100,000 fresh paths drawn with the lab's seed 26209. Your notes stay on this page until you leave it; copy the memo before you go.
The board's appetite is priced by Volume I's committed package. The transformation office proposes a rule that slows the build when cash runs low, the treasury asks what an early warning of funding-window closures would be worth, and human resources has added its record of departures to Chapter 8's hiring loop. The board wants the cheapest way to keep its appetite, checked on paths the design has not seen. You support the team preparing the recommendation.
Use AXIOM to estimate chances with their errors, design packages and rules on samples, judge them on fresh paths, price information, and steer the hiring loop with outside events.
Worked Example 9.1: planned liquidity of $620 million today dips with the digital program's draw; execution and market surprises carry 60 percent into the next quarter, with $50 million of new surprise a quarter; funding windows close about once a decade, each removing $120 million on average. The appetite: at most a 5 percent chance that liquidity falls below the $350 million floor in some quarter of the program's 12. A standby line costs 3 percent of its size over the program; a quarter of committed delay costs $8 million and lifts liquidity by up to $25 million at the deepest point of the draw. Every design uses the book's 100,000 paths; every answer is judged on 100,000 fresh paths.
In the Monte Carlo engine, estimate the chance of a breach under Volume I's package in each quarter and over the program, with standard errors; compare with the Gaussian law and the sum over quarters.
Bands: the middle 50 and 90 percent of the paths, quarter by quarter. Line: the average path. Dashed: the $350 million floor.
In the chance-constraint explorer, find the cheapest committed package on the book's paths and check it on fresh paths; repeat on 1,000 and 10,000 paths, with and without a margin.
For each line, the least committed delay that keeps the appetite on the book's paths: the 95th percentile of the delays the paths need.
Each design is the cheapest committed package on its own sample, aimed at 5 percent less the margin, and judged on the 100,000 fresh paths. Bounds: the chance that a package whose true chance exceeds 5 percent passes the sample's test (Equation 9.5).
Under common random numbers, compare the package with rules that read cash once and every quarter; find the best quarter to read.
Designed on the book's 100,000 paths ($ million; delays in quarters). Slowdown: the length of the rule's slowdown and the share of paths it slows. Fresh paths: the chance of a breach (percent) and the cost on 100,000 fresh paths.
Compute the clairvoyant's bound and the value of an early warning of closures.
For the hiring instance, add departures and an uncertain yield of approvals; compare gains and costs, and run the martingale audit on the value ledger.
Departures move the bench by a standard deviation of 15 engineers over a year (σ = 15); a quarter's cohort yields 60 percent with a standard deviation of 20 points (s = 0.10). Right: guarding against a misjudged drift, charged at the rate 1/θ.
The ledger: the value earned so far plus the position booked at the best rule's value, pd² + σ²p/ρ ($ million). Lines: its expectation, in closed form. Dots: 10,000 simulated paths in antithetic pairs (the lab's seed 26209), ± 2 standard errors, half-yearly.
Chances under 1 percent in every quarter when the program-wide chance is 5 percent (Proposition 9.2); designs that look safe on their own 1,000 paths and fail on fresh ones (Theorem 9.3); a rule that saves little because cash reveals trouble late, and a warning worth $4.1 million (Proposition 9.3); a gain that stays at 4.34 a year with departures and falls to 4.16 with an uncertain yield (Theorem 9.2); and a value ledger with no drift along the best rule (Theorem 9.1(iii)).
The package or rule, its cost and its chance of a breach on fresh paths with their errors, the value of information, and the declarations the answer rests on.