Every number on this page is computed by the Chapter 8 notebook's own code. Worked Example 8.1's 100,000 paths are the book's draws (seed 8); the other paths and histories are fresh draws with the lab's random seed 26108, not Meridian records. Your notes stay on this page until you leave it; copy the memo before you go.
Meridian's board has asked for a risk view of the digital program that it can trust: the probability of breaching the liquidity floor, what drives it, and what each mitigation buys. The chief risk officer also wants the planning pack's multi-year fans and event provisions checked for consistency. You support the chief financial officer.
Use AXIOM to see how fans form and settle, price the company's event risks, measure the flaw of averages, and put an honest probability on the floor breach under each mitigation.
Worked Example 8.1: the digital program's quarterly liquidity plan (Chapters 1 and 5), a persistent execution and market deviation, and funding-window closures; Sections 8.6–8.7: Industrial Systems' revenue deviation and regulatory interventions.
In the distribution evolver, start Industrial Systems' revenue deviation after a strong quarter and watch its distribution drift and widen; then sweep the persistence through one in the variance saturator and watch the band stop settling.
Rk+1 = 0.7 Rk + wk, shocks of standard deviation $15 million, from R0 = +30 (Section 8.6). Band: mean ± 2 standard deviations of the exact law (Theorem 8.3(i)); grey: fresh paths. Right: the distribution in that quarter, 100,000 fresh paths against the exact normal law.
The half-width of the 2-standard-deviation band, quarter by quarter. Below one it settles at 2 × 15/√(1 − a²); at one there is no restoring force and it grows like 15√k (Figure 8.2).
Enter a one-year and a three-year fan from the planning pack in the Chapman–Kolmogorov checker and test whether they compose.
The program office's one-quarter view: from on plan, on plan next quarter with probability 0.9; from off plan, back on plan with probability 0.3. The odds over j + k quarters pass through every state after j: Pj+k = PjPk (Theorem 8.2(ii)).
Industrial Systems' revenue deviation: the one-year fan applied three times against the twelve-quarter fan of the same law.
In the jump lab, generate regulatory interventions at 0.15 a year, read the plan line and the risk line, then condition on five quiet years and compare next year's chance.
Interventions arrive on a Poisson clock at λ = 0.15 a year and cost on average $40 million (standard deviation $30 million). Steps: fresh ten-year histories, waiting times first and then sizes (Section 8.9). Line: the plan line λE[J]t. Band: plus or minus two standard deviations, √(λE[J²]t) (Theorem 8.4; Figure 8.3).
In the convexity demonstration, bend a cost function and watch the gap between the average cost and the cost at average volume.
Chapter 5's start-up near zero, with demand noise: after one year X = X0 exp((0.9 − σ²/2) + σW), an exponential, so convex. With no spread the outcome at the average input and the average outcome coincide; the more the function bends the result, the wider the gap (Theorem 8.5(ii); Online Appendix 8.B.7). 200,000 fresh draws.
Worked Example 4.1's Unit A with random attrition restored: each engineer's year drawn, 100,000 fresh workforces. The mean follows the deterministic roll-up exactly (Theorem 8.5(i)).
On the Monte Carlo bench, reproduce Worked Example 8.1: vary the number of paths and watch the error bar shrink, switch each channel off in turn, and test the standby line and the phased build.
100,000 three-year quarterly paths, the book's draws (seed 8); every scenario reuses them, so the levers are compared on the same futures. Left: the 10th–90th percentile band, the median, the plan, and twenty paths. Right: each path's lowest quarter (Figure 8.4).
A Monte Carlo estimate's standard error is √(p(1 − p)/N). The runs of 100, 1,000 and 10,000 paths use the book's streams, seeds 100 + N (Section 8.9); the curve is a fresh stream with the lab's seed, its estimate ± 2 standard errors as the paths accumulate.
A stable company's fan stops widening, while an accumulator's never does (Theorem 8.3); quiet years do not make an event “due” (Theorem 8.4); and the average trough lies well below the planned trough (Theorem 8.5).
State the breach probability with its error bar, the share due to each channel, the effect of each lever, and a recommendation against the 5 percent appetite.