For Forecasts, Bet on Monte Carlo Methods

By Jo Craven McGinty, "The Numbers" column, Wall Street Journal, S/S Aug 13–14, 2016

It is pleasing to see decision analysis applications in the news.

This article looks at forecasts for Hurricane Sandy, which devasted parts of the U.S. East Cost during October, 2012. It became the largest Atlantic hurricane of record, costing an estimated $75 billion in damages and 233 lives (in eight countries).

Forecasting the path of a hurricane is important business. And Monte Carlo simulation is suited to the task.

The program generated numerous scenarios — "random trials."" Averaging trial paths produces a mean-trajectory forecast. This is likely augmented with judgments of experienced meterologists. The U.S. reasonably knew the hurricane's course about 72 hours ahead. The Europeans, with a long legacy of weather forecasting, knew perhaps a day earlier.

The article shows numerous tracks made by American and European models. The American's mean track trajectory stayed in the Atlantic, while the Eupean's mean track made landfall about Pennsylvania. “In the case of Hurricane Sandy, meteorologists ran 50 different scenarios (trials) with more than 10 million randomized variables.”

I could not find additional trajectory forecasts to share with you. However, the tracks are similar to what we see often with scenarios projected with business and economic forecasts. Usually, because the computations are so much easier, we will have many more trajectories, and these are commonly summarized into several confidence envelopes. Some of the weather images, online, show only one confidence envelope, calling it a “cone.”

Computer model forecasts for Hurricane Sandy with multiple model tracks
and observed path — Here is an image of successive forecasts, with each trace being the mean projection at the time. The black trace (with little hurricane symbols) is the actual storm path.

Source: National Oceanic and Atmospheric Administration (NOAA)

The article includes:

A brief history of Monte Carlo simulation, crediting mathematician Stanislaw Ulum for originating the method. He was then working with John von Neumann on the Manhatten Project in WWII.
Random sampling
Other application examples in health effects from exposure to hazardous material, estimating freeway traffic congestion, new drug R&D, and predicting Superbowl and election outcomes.

Ms. McGinty closes with this advice: “As with any forecasting tool, the results of a Monte Carlo simulation can't garantee the future. But viewing myriad potential outcomes through the lens of probability statistics can help lift the cloud of uncertainty.”

—John Schuyler, August 2016, revised 2026