7.2 Quantitative Risk Analysis & Modeling

7.2 Quantitative Risk Analysis & Modeling

Introduction to Quantitative Risk Analysis

While qualitative techniques such as HAZOP and FMEA provide vital structured screening, complex manufacturing and engineering decisions require quantitative risk analysis. Quantitative risk modeling attaches numerical probabilities, financial values, and statistical distributions to uncertain events. This rigorous approach enables Certified Quality Engineers (CQEs) to evaluate capital investments in quality, forecast project risks, and defend risk mitigation budgets with empirical precision.

Decision Tree Analysis & Expected Monetary Value (EMV)

Decision Tree Analysis is a visual and mathematical tool used to evaluate sequential decisions under conditions of risk and uncertainty. It structures choices, potential random outcomes, and financial payoffs into a branching tree model.

Expected Monetary Value (EMV) Formula

The cornerstone metric in decision tree analysis is Expected Monetary Value (EMV). EMV represents the weighted average financial outcome of a discrete probability distribution:

$EMV = \sum_{i=1}^{n} (P_i \times I_i)$

Where:

• $P_i$ = Probability of occurrence for outcome $i$, such that $\sum P_i = 1.0$.
• $I_i$ = Financial impact (positive for monetary gain/savings, negative for cost/loss) associated with outcome $i$.

Decision Tree Conventions & Nodes

A standard decision tree consists of three types of nodes:

1. Decision Nodes (Squares $\square$): Points where the decision-maker must select one alternative from available choices. The engineer controls this path; no probabilities are assigned to branches exiting a decision node.
2. Chance Nodes (Circles $\bigcirc$): Points where uncertain external events occur. Each exiting branch represents a distinct outcome with an assigned probability $P_i$. The sum of probabilities for all branches leaving a chance node must equal $1.0$ ($100%$).
3. Terminal Nodes (Triangles $\triangle$): End points displaying the net financial payoff or cost realized along that specific path.

The Backward Induction (Rollback) Method

To determine the optimal decision path, decision trees are evaluated from right to left using backward induction (also known as the rollback method):

1. Calculate the EMV at every chance node ($\bigcirc$) by multiplying each branch payoff by its probability and summing the products across all branches.
2. At each decision node ($\square$), compare the calculated EMVs (or net costs) of competing branches and select the option that maximizes EMV (or minimizes cost). Prune non-optimal branches.

Step-by-Step Worked Math Example

A quality engineering department must decide whether to purchase an Automated Optical Inspection (AOI) machine to reduce customer defect returns.

• Option 1: Purchase AOI Machine
  • Capital Investment Cost: $-$50,000$ (upfront cost).
  • Chance Node 1 Outcomes (Performance under production volume):
    • High defect prevention (Probability $P = 0.70$): Avoids scrap and warranty costs worth $+$200,000$ (Net payoff: $+$150,000$).
    • Moderate defect prevention (Probability $P = 0.30$): Avoids scrap costs worth $+$50,000$ (Net payoff: $$0$).
• Option 2: Status Quo (Manual Inspection)
  • Capital Investment Cost: $-$30,000$ (ongoing calibration/manual labor).
  • Chance Node 2 Outcomes (Baseline performance):
    • High warranty payout (Probability $P = 0.50$): Avoids costs worth $+$150,000$ (Net payoff: $+$120,000$).
    • Low warranty payout (Probability $P = 0.50$): Avoids costs worth $+$30,000$ (Net payoff: $$0$).

Step 1: Calculate Chance Node EMV for Option 1 $EMV_{\text{Option 1}} = (0.70 \times \$150,000) + (0.30 \times \$0) = \$105,000$

Step 2: Calculate Chance Node EMV for Option 2 $EMV_{\text{Option 2}} = (0.50 \times \$120,000) + (0.50 \times \$0) = \$60,000$

Step 3: Rollback Decision Node Selection Comparing Option 1 ($$105,000$ net EMV) against Option 2 ($$60,000$ net EMV), Option 1 provides an expected net financial advantage of $$45,000$. The optimal decision is to purchase the AOI system.

---

Monte Carlo Simulation Principles

When processes involve complex, interacting continuous variables, analytical formulas and static decision trees become inadequate. Monte Carlo simulation is a mathematical technique that models risk by replacing deterministic single-point estimates with continuous probability distributions and performing thousands of pseudo-random sampling iterations.

Key Simulation Steps

1. Define Parameter Probability Distributions: Assign appropriate continuous distributions to input variables based on historical process data:
   • Normal Distribution: For parameters governed by common-cause variation (e.g., shaft diameter, tensile strength).
   • Triangular Distribution: When limited data exists but Minimum, Most Likely (Mode), and Maximum bounds are known.
   • Beta / PERT Distribution: For project task durations emphasizing a smooth curve around the mode.
2. Iterative Sampling: Execute computer-generated random sampling across all input distributions simultaneously over $10,000+$ iterations.
3. Aggregate Output Distributions: Calculate the response variable $Y = f(X_1, X_2, \dots, X_k)$ for each iteration to generate an output Probability Density Function (PDF) and Cumulative Distribution Function (CDF).

Interpreting Simulation Output Metrics

• P10 (10th Percentile): Optimistic threshold; there is a $10%$ probability that the outcome will be less than or equal to this value.
• P50 (50th Percentile / Median): Baseline expected value.
• P90 (90th Percentile): Conservative risk threshold; there is a $90%$ probability the cost/duration will not exceed this value. Organizations use P90 to establish financial contingency reserves.
• Tornado Diagrams (Sensitivity Analysis): Bar charts displaying the correlation coefficient between each input variable and the total output variance, highlighting which input variable contributes the greatest risk.

---

Risk Scoring Models & Risk Boundaries

Quantitative risk screening often utilizes Probability-Impact (P-I) Matrices.

Risk Appetite vs. Risk Tolerance

• Risk Appetite: The broad level of overall risk an organization is willing to accept in pursuit of its strategic quality goals.
• Risk Tolerance: The specific, measurable maximum acceptable variation around a quality target or metric (e.g., maximum acceptable process defect rate of $0.05%$).