CYCLES IN CASUALTY:
Balancing Loops in the Insurance Industry
Kawika Pierson
MIT Sloan PhD Candidate
PRESENTATION OUTLINE
The Insurance Industry Past Research
Economics Control Theory System Dynamics
The Model Boundary Causal Loop Diagram Important Structures PID Control Behavior
How You Can Help
THE INSURANCE INDUSTRY
Basic Idea Two Sides to the Business
Insurance Investing
Insurance Cycle – What is Cycling? Underwriting Loss Ratio or Combined Loss Ratio
Loss Ratio – Adjustments/Premiums Expense Ratio – Expenses/Premiums Combined Ratio – Loss + Expense = (A + E) / P
A VIEW TO A CYCLE
A VIEW TO A CYCLE
THE INSURANCE INDUSTRY
Insurance Cycle – What Causes It? Industry View:
“The next stage is precipitated by a catastrophe or similar significant loss, for example Hurricane Andrew or the attacks on the World Trade Center.” – “The Insurance Cycle” wikipedia
Academic View: “Using quarterly data from 1974 through 1990, we
provide evidence of a long-run link between the general economy and the underwriting performance as measured by the combined ratio.” – Grace and Hotchkiss, 1995 J o Risk and Insurance
“Fluctuations in the supply of property-liability insurance may be exacerbated by regulation.” Winter, 1991 Economic Inquiry
PAST RESEARCH IN ECONOMICS
Early 1980’s through Mid 90’s Three Main Schools of Thought Cycle Caused by Interest Rate Fluctuations
Doherty and Kang (1988) – Insurance Prices Change in Lagged Response to Interest Rates
Grace and Hotchkiss (1995) – “External Impacts on the Property-Liability Insurance Cycle”
Cycle Caused by Limits to the Supply of Insurance Winter (1988, 1991, 1994), Gron (1989, 1994)
Cycle Caused by Feedback Processes Brockett and Witt (1982) – Loss expectations from the
past inform current premiums, causing autocorrelation
PAST RESEARCH IN CONTROL THEORY
If a Cycle Exists we Will Create a Lagged Negative Feedback Loop to Explain It
Balzer and Benjamin 1980 – “Dynamic Response of Insurance Systems with Delayed Profit/Loss Sharing Feedback…” Journal of the Institute of Actuaries
Zimbidis and Haberman 2001 – “The Combined Effect of Delay and Feedback on the Insurance Pricing Process: a Control Theory Approach” Insurance: Mathematics and Economics
PAST RESEARCH IN SYSTEM DYNAMICS
The Claims Game and Hanover Insurance“claims management, quality and costs”Quality = Claim Adjustment Quality
Daniel H. Kim Learning Laboratories Peter Senge – “The Fifth Discipline” Moissis 1989 Masters Thesis (Sterman) Focuses on Determining Decision Rules Cavaleri and Sterman (1997) “Towards evaluation
of systems thinking interventions: a case study” Improved Manager’s Mental Models
PAST RESEARCH IN SYSTEM DYNAMICS
Insurance Cycle…
Are There Really no SD Articles on the Insurance Cycle?
Thomas Beck Co-President of Swiss SD Society Works for Large Swiss Reinsurer No Published Articles on Insurance Cycle
THE MODEL – BOUNDARY
Endogenous Variables Premiums Underwriting Quality (Risk) Claims Employees Administrative Costs
Exogenous Variables Desired Profit Margin Size of the Total Market Some Components of Administrative Costs
THE MODEL – BOUNDARY
Many Feedbacks Excluded Size of the Insurance Market Investments and Interest Rates Free Capital’s Influence on Underwriting Effect of Time Pressure on Claim Settlement Competitive Effects on Profit Margins Random Claim Incidence Employee Productivity
Is this Too Far Towards “Negative Loop w/ Delay”
THE MODEL – CASUAL(TY) LOOP DIAGRAM
THE MODEL – STRUCTURES
THE MODEL – STRUCTURES
THE MODEL – STRUCTURES
THE MODEL – STRUCTURES
THE MODEL – STRUCTURES
THE MODEL – STRUCTURES
THE MODEL – STRUCTURES
THE MODEL – PID CONTROL
Translating Equations to SD isn’t Always Easy Proportional Control = Standard Structure Integral Control = No Steady State Error
Reasonable that People Use IC Derivative Control = Less Overshoot
Less Likely that People Use DC
THE MODEL – PID CONTROL
THE MODEL – PID CONTROL
THE MODEL – PID CONTROL
THE MODEL – BEHAVIOR
Displays Decaying Oscillation to Step Input
Combined Ratio
0.95
0.925
0.9
0.8750
0.851950 1956 1962 1968 1974 1980 1986 1992 1998 2004 2010
Time (Year)
dmnl
Perceived Combined Ratio : Current
THE MODEL – CASUAL(TY) LOOP DIAGRAM
THE MODEL – BEHAVIOR
Instability A Function of Largest Source of Costs
Combined Ratio
0.95
0.925
0.9
0.8750
0.851950 1956 1962 1968 1974 1980 1986 1992 1998 2004 2010
Time (Year)
dmnl
Perceived Combined Ratio : No Admin Cost Change
THE MODEL – CASUAL(TY) LOOP DIAGRAM
THE MODEL – BEHAVIOR
Loop Gain Very ImportantCombined Ratio
0.95
0.925
0.9
0.8750
0.851950 1956 1962 1968 1974 1980 1986 1992 1998 2004 2010
Time (Year)
dmnl
Perceived Combined Ratio : No Premium Change
Combined Ratio
0.95
0.925
0.9
0.8750
0.851950 1956 1962 1968 1974 1980 1986 1992 1998 2004 2010
Time (Year)
dmnl
Perceived Combined Ratio : No Premium Change Underwriting Standards Overreaction
THE MODEL – POTENTIAL SOLUTIONS
Derivative Control of Premiums? Careful Tuning Is Necessary Managerial Implementation Industry Wide Application
Why Do Quality Standards Change? Can This Loop Be Cut Life Insurance
The Kalmanuclear Option? Optimal LINEAR Filter Just Build a Really Good Model Instead