Industry: Government and Public Sector
Product: ModelRisk
Application: Disaster recovery

The Recovery Will Cost $151M and Take 33 Months — Give or Take Everything

The disaster has already struck. Roads, public housing, water systems, schools and the power grid are damaged, a federal declaration is in place, and the legislature wants two numbers: how much will the state pay to rebuild, and when will service be restored? The recovery plan answers with a single point estimate built from average damage counts and average unit costs: roughly $150M net of federal aid, restored in two years. Both numbers are medians dressed up as commitments — and both have a large chance of being wrong on the expensive side.

Simulating the rebuild 60,000 times, the median net state cost is indeed about $151M, but the distribution is heavily right-skewed: the mean is $184M and the 90th percentile is $342M — more than double the headline. There is a 19.9% chance the bill exceeds the $260M appropriation. The schedule is worse against its target: the median restoration takes 33 months and there is a 38.4% chance of breaching the 36-month statutory deadline.

A state emergency-management agency rebuilt its recovery appropriation and schedule case in ModelRisk after a prior event came in 60% over its authorized fund and a year past its deadline. The mandate: attach an honest probability to the budget and the timeline — and to the risk of missing both at once.

Why a point estimate fails

The deterministic plan multiplies an average damaged-asset count by an average unit cost and sums an average duration per category. That arithmetic is wrong in three structurally important ways:

• A worse event damages everything at once. Damage extent is driven by a single uncertain event-severity factor (modeled LogNormal): a more destructive storm inflates road, housing, utility, school and grid damage together. Treating the categories as independent lets their variation cancel — it does not.
• Costs escalate program-wide. Post-disaster labor and material price spikes (a Triangular escalation factor) hit the entire rebuild simultaneously, not asset by asset.
• Cost and schedule are correlated. The same severe events that drive cost up also clog the contractor market, stretching restoration time — so the budget overrun and the schedule overrun tend to arrive together, exactly when the agency can least afford both.

The model draws one declared disaster per realization: severity-scaled Poisson damage counts across five asset categories, LogNormal unit rebuild costs under a shared escalation factor, Triangular restoration durations on a contractor-contention critical path, and a Beta-distributed federal cost share (around 75%) that itself is uncertain at planning time.

Cost and schedule fail together

The joint view is the one a separate budget memo and schedule memo can never produce. Plotting net cost against restoration time, the probability of overrunning the appropriation AND the deadline at the same time is 11.9% — against only 7.6% if the two were independent. The shared event-severity and contractor-contention factors inflate the double-overrun corner by more than half. The realistic planning question is not "will we be over budget?" or "will we be late?" but the 46.4% chance of being over on at least one — and the one-in-eight chance of explaining both to the legislature in the same hearing.

When will service actually be restored?

The restoration S-curve converts the schedule uncertainty into a completion-probability date. There is a 10% chance of finishing within 23 months, a 50% chance by 33 months, and a 90%-confidence date of 47 months — fully two years beyond the optimistic deterministic plan. Against the 36-month statutory deadline the curve crosses at roughly the 62nd percentile, which is the same 38% breach probability seen from the cost side. A plan that commits to the deterministic finish date is committing to a date the simulation only reaches about one time in three.

What drives the cost spread

Around the P90 net cost of $342M, the dominant driver is the event-severity factor — the extent of damage — followed by the federal/aid cost share and the cost-escalation factor. Because the aid share is the second-largest lever, the agency's net exposure is acutely sensitive to how much of the bill Washington ultimately covers: a less generous share than the planning assumption can move the net appropriation need by over $100M. Unit-cost dispersion and the Poisson damage counts matter less; contractor contention drives the schedule more than the cost.

What the model changed

• The recovery appropriation was sized to the P90, not the median — the deterministic $150M plan carried a one-in-five chance of running out of money before the rebuild finished.
• A contingency tranche was structured against the aid-share uncertainty, the second-ranked driver, so a less favorable federal cost share does not force mid-recovery borrowing.
• The statutory deadline case was re-argued on the S-curve: committing to 36 months meant a 38% breach probability, and the agency negotiated milestone-based reporting tied to the P50/P90 dates instead of a single hard date.
• The joint cost-and-schedule risk was made explicit to the legislature — the 11.9% double-overrun probability, well above the 7.6% an independent reading implies — justifying a combined reserve rather than two separate, smaller buffers.

ModelRisk functionality used

• Monte Carlo simulation of 60,000 single-disaster realizations — severity-scaled Poisson damage counts, LogNormal unit costs under a shared escalation factor, Triangular durations on a contention-driven critical path, and a Beta federal cost share — producing the joint cost-and-time distribution.
• A shared event-severity common factor linking all damage categories so the aggregate tail reflects correlated, not independent, damage.
• Joint density of cost against time quantifying the double-overrun probability and contrasting it with the naive independence assumption.
• Restoration S-curve turning schedule uncertainty into a completion-by-date probability against the statutory deadline.
• Tornado sensitivity ranking event severity, the federal aid share and cost escalation as the drivers of net state exposure.

A recovery plan that quotes one cost and one date is a plan that will be wrong on the expensive side roughly half the time. ModelRisk is what shows that the honest answer is a budget at the P90 and a schedule with a confidence date attached — and that the worst case is the one where both fail at once.