Applying Latin hypercube sampling to agent-based models

Andrew J. Collins (Virginia Modeling, Analysis, and Simulation Center (VMASC), Old Dominion University, Norfolk, Virginia, USA)
Michael J. Seiler (Old Dominion University, Norfolk, Virginia, USA)
Marshall Gangel (Virginia Modeling, Analysis, and Simulation Center (VMASC), Old Dominion University, Norfolk, Virginia, USA)
Menion Croll (Virginia Modeling, Analysis, and Simulation Center (VMASC), Old Dominion University, Norfolk, Virginia, USA)

International Journal of Housing Markets and Analysis

ISSN: 1753-8270

Publication date: 30 September 2013

Abstract

Purpose

Agent-based modelling and simulation (ABMS) has seen wide-spread success through its applications in the sciences and social sciences over the last 15 years. As ABMS is used to model more and more complex systems, there is going to be an increase in the number of input variables used within the simulation. Any uncertainty associated with these input variables can be investigated using sensitivity analysis, but when there is uncertainty surrounding several of these input variables, a single parameter sensitivity analysis is not adequate. Latin hypercube sampling (LHS) offers a way to sample variations in multiple parameters without having to consider all of the possible permutations. This paper introduces the application of LHS to ABMS via a case study that investigates the mortgage foreclosure contagion effect. This paper aims to discuss these issues.

Design/methodology/approach

Traditionally, uncertainty surrounding a single input variable is investigated using sensitivity analysis. That is, the variable is allowed to change to determine the impact of this variation on the simulation's output. When there is uncertainty about multiple input variables, then the number of simulation runs required to undertake this investigation greatly increases due to the permutations that need to be considered. LHS, which was first derived by McKay et al., offers a proven mechanism to reduce the number of simulation runs needed to complete a sensitivity analysis. This paper describes the LHS technique and its applications to an agent-based simulation (ABS) for investigating the foreclosure contagion effect.

Findings

The results from the foreclosure ABS runs have been characterized as “good”, “bad” or “ugly”, corresponding to whether or not a property market crash has occurred. As the only thing that can induce a property market crash within our model is the spread of foreclosing properties, these results indicate that the foreclosure contagion effect is dependent on how much impact a foreclosed property has on the price of the surrounding properties.

Originality/value

This paper describes the application of LHS to an agent-based foreclosure simulation. The foreclosure model and its results have been described in Gangel et al. Given a certain output “boundary” found within these results, it was highly appropriate to conduct an extensive sensitivity analysis on the simulation's input variables. The outcome of the LHS sensitivity analysis has given further insight into the foreclosure contagion effect thus demonstrating it was a beneficial exercise.

Keywords

Citation

J. Collins, A., J. Seiler, M., Gangel, M. and Croll, M. (2013), "Applying Latin hypercube sampling to agent-based models", International Journal of Housing Markets and Analysis, Vol. 6 No. 4, pp. 422-437. https://doi.org/10.1108/IJHMA-Jul-2012-0027

Download as .RIS

Publisher

:

Emerald Group Publishing Limited

Copyright © 2013, Emerald Group Publishing Limited

Please note you might not have access to this content

You may be able to access this content by login via Shibboleth, Open Athens or with your Emerald account.
If you would like to contact us about accessing this content, click the button and fill out the form.
To rent this content from Deepdyve, please click the button.