Deeper Insights from System Dynamics Models

1
Deeper Insights from System Dynamics Models

• Mark Paich
• Lexidyne Consulting
• 10/9/08

2
Observations from (too much) experience

• Decision makers are hungry for policy insights
• Forecasts are not believed
• Many SD models leave insight on the table
• Modeling project time allocation to much time
  modeling not enough time on analysis
• Techniques for analyzing models and extracting
  insights are not well developed
• Simple techniques can yield significant insights
• Techniques are know but are not standard in SD
• Techniques used in analyzing agent and discrete
  event models

3
Concepts that often generate useful policy
insights

• Synergy
• Implementing policy A and policy B together
  produces more improvement that the sum of policy
  A and policy B individually
• Multiple changes are necessary to achieve
  significant improvement
• Timing the timing and ordering of policies
  significantly change their effectiveness
• Robustness the policy is effective enough
  across many uncertainties
• Value of flexibility and policy rules
• Real options

4
Example Process Improvement Project

• Project description
• Synergy analysis
• Generate multiple simulations (design of
  experiments) for
• Decision variables
• can be discrete or continuous
• Useful to categorize them into low medium , high
  buckets
• Uncertain parameters
• External scenario variables. e.g. product demand
• Outcome metrics what is important
• Concluded that there are significant synergy
  effects. Policy changes must be implemented as a
  total package to get breakthrough results.
  Leaving elements out greatly increases the
  likelihood of failure

5
Example Process Improvement Project

• Export data to Excel and import into a standard
  statistical package
• Plot relationships between policy variables and
  outcome metrics
• Estimate simple statistical models that relate
  outcome metrics to policy and uncertain
  parameters
• Direct effects
• Synergy interaction effects
• Analyze relationships and outlying data points

6
Extensions

• Robustness
• How do uncertain values change the effectiveness
  of policy? Are there strong interaction effects
• Many other techniques for Robust Adaptive
  Planning Steve Bankes Rand Corporation
• Value of flexibility Real options and SD.
  (David Ford Texas AM)
• Data analysis tools
• Non linear estimators (neural nets)
• Rule induction