Notebook · not a forecast tab
A forecast often gives one number. But 70% utilization on Saturday can describe steady corporate demand, a leisure surge, an event with expected cancellations, a storm, or a competitor price change. Each situation may need a different action. The number alone does not explain the week.
The week is a mixture
Forecasting completes an unfinished week. RateHighway’s pickup model starts with on-the-books reservations and adds the expected remainder of a seasonal booking pattern. This works when the rest of the curve behaves like similar past weeks. It does not by itself detect that the type of demand has changed.
Demand sources behave differently. Walk-up, wedding, leisure, online travel agency (OTA), and corporate bookings arrive at different times and cancel at different rates. Melt means bookings that cancel or do not arrive. Prepaid and pay-later reservations should not use the same melt assumption. Walk-up demand also has its own timing at the rental counter.
This notebook proposes simulating many realistic weeks. Each version can vary by customer type, pickup time, rental length, vehicle class, sales channel, willingness to pay, and melt. Possible futures means a range of weeks and their probabilities, not one forecast line.
Policy under uncertainty
Revenue-management models often evaluate decisions across many possible demand paths. People who set rates rarely see that full range. Scenario tools usually show only three cases: base, strong, and weak. Three lines are still a small summary of the same forecast.
To be useful, the model must identify demand patterns that lead to different actions: corporate, leisure, OTA, event, cancellation, or disruption. The output should show which patterns are plausible and help test a policy across them. It should not simply produce one expected utilization number.
The simulated weeks must reproduce booking curves, melt, and length-of-rent mix that RateHighway already measures. They must also reflect the location’s current bookings and known events. In a stable corporate location, a better single forecast may be more useful than a large range of scenarios. In that case, the research idea should not be used.
A direction, not a paper
The mathematics we are exploring
Model a set of possible weeks {ω}. Each week contains arrivals, melt, and walk-up demand. Choose a policy by its expected result across those weeks, not only by its result at one expected utilization value.
{ω}
A probability distribution across possible weeks. Proposed actions can be tested on weeks that have not happened yet.
π
A policy the operator can inspect, limit, and reject. Better intelligence supports the strategy; it does not take control away from the operator.
μ(ρ | x)
The probability of each demand pattern after current evidence is observed. A pattern is useful only if it leads to a meaningful difference in action. The simulated range must also match the booking curves and melt already measured by the platform.
This is a notebook, not an announced product or delivery commitment. The operator still decides.