When the Railway Stops, the Costs Keep Moving

A railway can be running again before the economic consequences of a disruption have ended. Restoring a service does not undo a delayed arrival, recover every cancelled activity or return the time passengers have already lost.

Our assessment of the Dutch rail disruption of 15 September 2026 puts its net societal cost at approximately €680,000 under the central assumptions, in 2026 euros using August price indices. The largest component is additional passenger time, not the response on the railway. This is a modelled loss of time, resources and opportunities, not an audited bill from NS or ProRail and not an estimate of lost GDP. [1]

The railway is the starting point, not the boundary

The incident updates recorded in our report describe more than 35 faults, including repeat faults at some locations, affecting passengers, freight customers and road users at level crossings. By the late-afternoon evidence cut-off, ProRail reported that the faults then known had been resolved. That did not establish that every journey or operational consequence had returned to normal. [1]

To count the economic consequences, we compare two versions of the same day: the disrupted day and the day that would otherwise have occurred. In that second version, people would still have travelled, trains would still have used electricity and staff would still have worked. Only the difference belongs in the incident account.

Consider a passenger who loses 90 minutes and receives a €20 refund. The lost time is a welfare cost. The refund changes who holds the money: the passenger receives what the operator pays. Adding it as another national loss would count a transfer as destruction. Processing the refund, however, can consume additional resources.

A replacement taxi similarly uses vehicle and driver resources, but its fare must not be added again on top of those resources. Conversely, a cancelled train may avoid electricity use and wear. Those savings reduce the net loss. The calculation is not a running total of every payment associated with the disruption. [2]

From disrupted services to affected journeys

The operational audit examines 96 intended station departure opportunities between 07:00 and 09:00 at Zwolle, Deventer and Ede-Wageningen, plus recovery observations. Of those 96 opportunities, 78 did not depart. These are station opportunities, not 78 distinct whole-train cancellations, and the audit is a structured sample rather than a national census. [3]

The next step is to estimate the demand exposed to unavailable capacity. For each corridor and time window, the model combines intended services, the unavailable share after a background allowance, representative seating capacity and an assumed passenger load.

That produces approximately 46,585 affected corridor passages. A journey can cross more than one affected section, so passages cannot simply be counted as separate journeys. Dividing by an assumed 1.25 affected sections per journey gives 37,268 core intended journeys. Adding a separate group with smaller recovery delays brings the modelled exposure to approximately 42,900 intended journeys. None of these figures is an observed count of unique people. [3]

What happens next determines the loss. The central scenario assumes that 70% of core journeys continue or reroute by rail, 15% switch to road, 10% are postponed or replaced remotely, and 5% involve abandoned activity. These response shares are assumptions, not findings from a survey of this incident. A three-hour route closure is not automatically three hours of delay for every traveller. [4]

Why time dominates the result

For the approximately 26,087 core journeys completed by rail, the central scenario assumes 70 additional minutes from origin to destination. That produces about 30,435 passenger-hours. The report values this rail time at approximately €12.08 an hour, using an indexed transport-appraisal coefficient rather than passengers' wages, for a cost of roughly €368,000.

Adding road-switch passengers' time and smaller recovery delays brings total additional completed-journey time to approximately 33,510 hours, valued at about €405,000. Monetary values use a 2026 price basis, with August indices as the proxy for the September incident. [4]

Central scenario, € thousands. All entries are model outputs on the same 2026 price basis. [5]

The response estimate is also constructed from assumptions about activities, deployment and resource rates, not observed invoices. Other consequences, including some accessibility and longer-term effects, remain unpriced. Their omission does not establish that the reported total is a minimum. [2, 5]

What could change the estimate

Selected passenger scenarios produce net costs of approximately €0.52 million to €0.93 million. They vary passenger exposure, additional rail time and traveller responses while holding several response, freight and other assumptions fixed. The range is not a statistical confidence interval or a complete account of uncertainty; actual cost can fall outside it.

One sensitivity shows why end-to-end journey data matter. With the central number of completed rail journeys and other inputs unchanged, another 15 minutes of delay adds approximately €79,000. Better information on intended journeys and actual arrival times would materially strengthen the estimate. More decimal places cannot replace those observations. [6]

Resilience is about more than repairs

For infrastructure owners and operators, the analysis identifies costs that a repair or response budget alone does not capture. Measures that shorten recovery, preserve a useful alternative route or improve information may reduce losses even when they do not prevent the initiating event. Their benefit depends on how much disruption they actually avoid, how often comparable events occur and what the measures cost.

That is an appraisal question, not an automatic commercial opportunity. Time retained by passengers does not automatically become revenue for a supplier. A business case still needs evidence about effectiveness, customers, procurement, costs and payment. [7]

The central finding is that a limited physical intervention can transmit substantial costs through a network without requiring extensive physical damage. Understanding the consequences means following the affected journeys and resources beyond the railway, not stopping when the track is clear.

Research note. This is an event-based scenario assessment, not an audited loss total. The report's evidence cut-off is late afternoon on 15 September 2026. Its main service window is 05:30 to 17:00; later events fall outside the assessment, while some response assumptions extend beyond the observation window. The selected scenario range does not cover every unknown cost.

Explore the research

The technical report documents the service reconstruction, passenger assumptions, valuation methods and complete cost ledger. The accompanying Dutch Rail Economic Model contains the calculations and sensitivity analysis behind the scenarios.

Sources and research

[1] Incident/status: ProRail incident notice; NS 15 September update. Central €680k result: Rokin technical report + rail model.

[2] Economic boundary: CPB/PBL General Guidance for Cost-Benefit Analysis; Werkwijzer MKBA bij MIRT; Rokin report Section 1 / Appendix C.

[3] Service evidence: Rijden de Treinen archives for Zwolle, Deventer and Ede-Wageningen on 15 September, with 8 September reference-day records; Rokin report Sections 2-3 + model for normalisation.

[4] Behaviour/time valuation: Bickel et al. 2025; KiM travel-time values; RWS reachability guidance; CBS CPI and wage index; Rokin report Sections 4-5 + model.

[5] Cost composition: Rokin technical report Sections 6-7 + Dutch Rail Economic Model.

[6] Scenario range/sensitivity: Rokin technical report Section 8 + Dutch Rail Economic Model.

[7] Resilience: RWS reachability/robustness guidance; World Bank resilience paper; Rokin report Section 9.