Sensitivity Analysis

Analyze Impacts of Different Variables on Uncertainty

orange triangle

Prioritize Risks by Ranking their Impacts

Analyzing uncertainty, and specifically the key inputs that drive that uncertainty, is at the heart of risk analysis. Which variables actually impact your outputs the most? When you prioritize your key risks, you can efficiently and optimally assign controls and mitigations across your entire business.

What is Sensitivity Analysis?

Rank Your Inputs in Order of Importance

Deterministic sensitivity analysis is a method of analyzing models that allows you to rank your inputs in order of importance. It’s an advanced yet accessible practice that informs decisions on how to effectively allocate your organization’s limited resources to addressing the risks you face. By itself this is a critically important result, but the process can also be used as an interim step prior to the creation of a focused, probabilistic model that utilizes Monte Carlo simulation.

Communicate Results with Graphs

A sensitivity analysis generates quantitative data based on the behavior of outputs in response to changing inputs. This data allows the creation of tornado and spider graphs, giving a visual representation of the inputs’ relative impact on your key outputs. Together, these graphs and data provide communication tools and hard numbers to validate your business decisions.

How Does Sensitivity Analysis Work?

Sensitivity analysis operates directly on your preexisting model. Generally, the software will identify all inputs affecting an output you specify (NPV, Total Project Cost, Return, anything at all). Then these inputs are stepped through a meaningful range of values (such as +/- 10%), indicative of the uncertainty in each. For every one of these values the entire model is recalculated, with new data recorded for all identified outputs. This data represents the direct impact that each input has on the calculated output value. The magnitude of this range is the metric by which the inputs are ranked, and conveniently displayed in tornado charts and spider graphs. Greater impact means an input is more important, requiring mitigation or further investigation and modeling.

A testing report assessing the accuracy of a newly-created neural network.

A tornado graph showing the results of a sensitivity analysis. Large bars on top have the most impact.

Local and Global Sensitivity Analysis

In general, sensitivity analysis falls into one of two categories: local and global.

Local Sensitivity Analysis

This method is appropriate for simple models and involves adjusting input variables one at a time across a defined, but generally limited, range. For instance, you might vary all inputs +/- 10%, as noted earlier. As any one given input variable is adjusted, all others are fixed. Local sensitivity analysis is an excellent precursor for further analysis, as it saves tremendous time and energy by helping to focus the project.

Global Sensitivity Analysis

Global sensitivity analysis, by contrast, adjusts all input variables at the same time. In addition, the ranges of values sampled for each variable is much broader and is meant to represent the entire range of possible values each variable could take. Effectively, global sensitivity analysis is performed using Monte Carlo simulation.

Sensitivity Analysis Software from Palisade

Palisade’s TopRank software enables any Excel user to quickly and easily identify the most important factors in any spreadsheet model. TopRank intelligently identifies all cells which influence whichever output cell(s) you specify, and then varies them automatically or according to your preferences. The resulting graphs and data are highly effective communication tools, enabling you to take mitigation steps to reduce variability. More commonly, TopRank is used to identify which variables should be further defined with probability distributions in @RISK to enable a subsequent, more advanced Monte Carlo simulation analysis.

Join decision-makers around the world who