Our CAIO, Tom Roderick PhD, has published a follow-up article, about building a Monte Carlo simulation model to forecast FEMA emergency deployments. This model is constructed using three key random variables, based on historical FEMA data:
Number of people deployed weekly for emergency events, where events are grouped by their weekly declaration dates.
The time delay between an event’s declaration and actual deployment, which can vary based on logistical factors.
Duration of deployment, indicating how long individuals remain in the field once deployed.
Each of these variables is represented using histograms to visualize their distributions, which serve as the basis for the simulation. The Monte Carlo simulation randomly selects values from the probability distributions of each variable, iterating over many scenarios to provide a forecast for how many people will be deployed and when.
A key insight is the independence of the second and third variables (time to deploy and deployment duration), simplifying the simulation. The model also accounts for seasonality in the weekly deployment needs, as certain types of emergencies, such as hurricanes, are more likely to occur at particular times of the year.
The simulation proceeds in three steps:
Randomly select the number of people deployed in a week from within the historical data’s interquartile range (5th to 95th percentile) to avoid outliers.
For each individual deployment, randomly select the time between an event’s declaration and the actual deployment, as well as the duration of the deployment.
Repeat the process over many iterations (e.g., 10,000) to generate a comprehensive view of expected deployments and resource needs.
The outputs from the simulation include predictions about total people deployed in a given time frame, such as a year, allowing FEMA to better prepare for resource allocation and budgeting. The simulation can either cut off at a specific time frame or continue running to let deployments "decay," simulating long-term resource needs.
The article provides an example of a single Monte Carlo draw, showing how deployments accumulate over time. It highlights one extreme event that occurs in the third year of the simulation, leading to a significant spike in weekly deployments. The visual representation of this single iteration emphasizes the importance of running multiple simulations to get a more reliable forecast, as a single iteration may produce an outlier.
In conclusion, the article outlines the next steps in refining the model and warns against drawing conclusions from any one simulation. The authors will cover further analysis in future posts, including extensions of the Monte Carlo framework and the limitations to be considered when interpreting the results.
For more information see the original article.