Sensitivity analyzer

Sensitivity Analyzer

Overview

The Sensitivity Analyzer skill provides comprehensive capabilities for identifying critical inputs and understanding how model outputs respond to parameter changes. It supports both local (one-at-a-time) and global sensitivity analysis methods, enabling robust decision-making under uncertainty.

Capabilities

• One-at-a-time (OAT) sensitivity
• Global sensitivity analysis (Sobol indices, Morris screening)
• Tornado diagram generation
• Spider plot creation
• Parameter importance ranking
• Threshold identification
• Breakeven analysis
• Scenario comparison

Used By Processes

• Monte Carlo Simulation for Decision Support
• Multi-Criteria Decision Analysis (MCDA)
• Prescriptive Analytics and Optimization
• What-If Analysis Framework

Usage

One-at-a-Time (OAT) Analysis

# Define OAT analysis
oat_config = {
    "base_case": {
        "price": 100,
        "volume": 10000,
        "cost": 60,
        "fixed_costs": 200000
    },
    "variations": {
        "price": {"range": [-20, 20], "step": 5, "unit": "%"},
        "volume": {"range": [-30, 30], "step": 10, "unit": "%"},
        "cost": {"range": [-15, 15], "step": 5, "unit": "%"},
        "fixed_costs": {"range": [-10, 10], "step": 5, "unit": "%"}
    },
    "output_variable": "profit"
}

Global Sensitivity (Sobol Indices)

# Define Sobol analysis
sobol_config = {
    "parameters": {
        "price": {"bounds": [80, 120], "distribution": "uniform"},
        "volume": {"bounds": [7000, 13000], "distribution": "uniform"},
        "cost": {"bounds": [50, 70], "distribution": "uniform"}
    },
    "sample_size": 10000,
    "calculate_second_order": True
}

Morris Screening

Efficient screening method for many parameters:

• Identifies parameters with negligible effects
• Distinguishes linear vs. non-linear effects
• Detects interaction effects

Sensitivity Indices

Index	Meaning
S1 (First-order)	Direct effect of parameter
ST (Total)	Direct + all interaction effects
S2 (Second-order)	Pairwise interaction effect

Visualization Types

1. Tornado Diagram: Horizontal bars showing impact range
2. Spider Plot: Lines showing output vs. % change in each input
3. Scatter Plot: Output vs. single input with trend line
4. Sobol Bar Chart: First-order and total indices comparison
5. Morris Plot: Mean vs. standard deviation of elementary effects

Input Schema

{
  "analysis_type": "OAT|sobol|morris|breakeven",
  "model": "function or expression",
  "parameters": {
    "param_name": {
      "base_value": "number",
      "range": ["number", "number"],
      "distribution": "string"
    }
  },
  "options": {
    "sample_size": "number",
    "output_variable": "string",
    "calculate_interactions": "boolean",
    "confidence_level": "number"
  }
}

Output Schema

{
  "analysis_type": "string",
  "parameter_rankings": [
    {
      "parameter": "string",
      "importance_score": "number",
      "effect_direction": "positive|negative",
      "first_order_index": "number",
      "total_index": "number"
    }
  ],
  "breakeven_points": {
    "parameter": {
      "breakeven_value": "number",
      "current_distance": "number"
    }
  },
  "interactions": [
    {
      "parameters": ["string", "string"],
      "interaction_index": "number"
    }
  ],
  "tornado_data": {
    "parameter": {
      "low_output": "number",
      "high_output": "number",
      "swing": "number"
    }
  },
  "visualization_paths": ["string"]
}

Best Practices

1. Start with Morris screening for many parameters (>10)
2. Use Sobol indices for detailed analysis of top parameters
3. Include parameter correlations when they exist
4. Report confidence intervals for sensitivity indices
5. Consider non-linear effects (total vs. first-order indices)
6. Communicate results using tornado diagrams for executives
7. Document parameter ranges and their justification

Interpretation Guidelines

Sobol Index Interpretation

• High S1, High ST: Important direct effect
• Low S1, High ST: Important through interactions
• High S1, Low ST-S1: Few interactions
• Low ST: Parameter can be fixed at nominal value

Breakeven Analysis

Identifies the parameter value where:

• NPV = 0
• Profit = 0
• Decision changes
• Threshold is crossed

Integration Points

• Receives model from Monte Carlo Engine
• Feeds into Decision Visualization for charts
• Supports MCDA methods for weight sensitivity
• Connects with Real Options Analyzer for volatility impact