Tuesday, November 14, 2017

Summary of Findings: Monte Carlo Simulations (3.5 out of 5 Stars)

Note: This post represents the synthesis of the thoughts, procedures and experiences of others as represented in the articles read in advance (see previous posts) and the discussion among the students and instructor during the Advanced Analytic Techniques class at Mercyhurst University, in November 2017 regarding Monte Carlo Simulations as an Analytic Method, specifically. This technique was evaluated based on its overall validity, simplicity, flexibility and its ability to effectively use unstructured data.

Description:
Monte Carlo simulations are mathematically-based algorithms that use random sampling to obtain estimates. Typically, Monte Carlo simulations are used for physically or mathematically complex problems that would be difficult to solve by other means. Many forms of Monte Carlo simulations are designed to be run with many test iterations. The results of these interactions is a range of probabilities, distributions, or possible outcomes.

Strengths:
• Allows decision makers to determine a range of possible outcomes  and the probability that an outcome may occur rather than a single-point estimate
• Flexible in its application
• Proven effectiveness for increasing forecasting accuracy
• Can take into account and use several variables
• Has vast amounts of evidence giving it credibility
• Effective when used in conjunction with decision trees/complex scenario based forecasting
• Time and cost-effective

Weaknesses:
• Assumptions need to be fair; if a number is derived from unrealistic assumptions, then they possess no real value
• Requires a very in depth knowledge of math; for many that can be extremely challenging
• Validity dependent on the variables, if the input is inaccurate then the output will be inaccurate.
• Many samples may be required to obtain an acceptable precision in the answer

How-To:

The exercise used was a simplistic version of a Monte Carlo simulation in order to show the method in a short amount of time. We attempted to forecast the number of turns it would take us to kill a fictional monster in an RPG if we were rolling dice.

1. Set the “Monster’s” health, ours was 20
2. Role die and subtract that number from the “Monster’s” health
3. Role until monster dies
4. Run simulation over several iterations. (10,000 is suggested but for time we ran 5)
5. Take average and build a distribution model to evaluate findings and scenarios

Application of Technique:
The class was given a graph with the x-axis numbering up to 24. They were required to start at the number 20 and after rolling a six sided dice, subtract the number rolled from 20 and keep rolling until they got to zero. They plotted the numbers they rolled on a graph to see how many turns it took to get to zero. This was done five times for each student. The number of attempts was collected from everyone in class and average was calculated for the number of attempts required for a dice to reduce 20 to zero. The method used in class was a simplistic process of Monte Carlo, in a real life scenario the method is repeated 1,000 or more times to get a better output. Our exercise launched a discussion over the actual value of a monte carlo simulation compared to other forms of regression. We came to the conclusion that Monte Carlo does offer different kinds of answers that could be more valuable in the uncertain world of intelligence analysis.

For Further Information: