Synthesis of three disciplines AgentBased Modeling System Dynamics

Synthesis of three disciplines ● Agent-Based Modeling ● System Dynamics ● Economics

Foundations ● Steve Keen’s Debt Deflation Model ● Hyman Minsky’s Financial Instability Hypothesis ● Goodwin’s Class Struggle Model ● Graziani’s Monetary Circuit Theory ● Kalecki’ Markup Pricing Theory

Minsky’s Financial Instability Hypothesis ● Hedge Financing ● Speculative Financing ● Ponzi Finacing

Steve Keen’s Debt Deflation Model ● Goodwin’s Class Struggle Circuit ● Kalecki’s Markup Pricing Circuit ● Minsky’s Financial Instability Circuit ● Graziani’s Endogenous Money Circuit

Versions of Keen’s Models

Definition Examples Real Output: Labor Force: Productivity: Nominal Output: Employment: Population: Price Level:

Functions Wage Change / Employment Investment / Profit Rate Keen: “Firms invest more than their profits during a boom, less than their profits during a slump. ”

Keen’s model in action

Schum. PTR ● Custom-made System-Dynamics framework ● System and models built in C#. NET ● Reflection-based variable gathering ● Runge-Kutta-based approximation algorithm ● Alerts ● UI-based diagnostics

Experiment: varying the investment/profit ratio Debt/Output Wage Share Inflation

Experiment: varying bank consumption rate account drawdown rate x 50 Bank Consumption Debt/Output

Experiment: varying investment convexity Investment/Profit Functions Debt/Output Profit Rate (STD)

Conclusions (technical) ● Consumption rates largely irrelevant to stability ● Investment / profit ratio strongly influences stability ● Higher I/P Ratio destabilizes Keen’s models without pricing system ● Lower I/P Ratio destabilizes Keen’s models with pricing system ● More concave investment functions destabilize both kinds of models

Conclusions (qualitative) ● Keen’s models have many imperfections ● Debt Deflation Model does not represent Minsky’s FIH well ● Model is not suitable as a foundation for an agent-based Minsky model ● Another hypothesis on what the model represents is called for

The Collusion Hypothesis ● Instability caused by a low investment rate ● Firms’ collective profit share rises as investment rate falls ● Contrast between individual and collective incentives ● Potential for collusion ● Collusion cause the instability

Agent-Based System-Dynamic Modeling

Aggregate-individual separation

Strategy ● Object Parameters ○ Investment / Profit Ratio ● Meta Parameters ○ Conformity Factor ○ Winner Copying Factor

Ethics ● Individual Interest ● Class Interest ● Universal Interest

Training Mechanism ● Evolutionary Algorithm ● Strategy Mutation ● Training Pool > Participants ● Ranking based on Ethics ● Pair-wise Merging/Cloning Winners of

Results: Full Class Interest = 1 Full Evaluation Debt / Output Individual Interest = 0 Universal Interest = 1 Investment / Profit

Results: Some Individual Interest Class Intererest = [0. 9, 0. 8, 0. 7] Individual Interest = [0. 1, 0. 2, 0. 3] Universal Interest = [1. 0, 1. 0] Investment / Profit Debt / Output Inflation

Multiple Training Sets

Results: Multiple Training Sets Debt Ratio Investment/Profit MA 25 Profit Share 10 th Agent’s Share

Conclusions on the Collusion Model ● Collusion Hypothesis corroborated in the strictest sense ● … but disproven in a practical sense ● Enforcing agency may be required to bring about the effect ● eg. a cartel or government

Conclusions from modeling efforts ● Collusion Model is by no means realistic ● Model illustrates a set of modeling principles: ○ The separation between aggregates and individuals ○ The separation between Ethical, Strategical and Logical variables ○ Training of agent sets by evolutionary means ○ Construction and diagnosis of models in a C#. NET environment ● Experiments demonstrate the power of Schum. PTR ● High potential for further development along these lines

Discussion on the Rationality Principle ● Model is based on model-aware, rationally optimizing agents ● Model does employ some epistemic limitations: ○ ○ ○ Agents can see a limited number of time units ahead Agents assume that other agents are like them, unaware of individual differences Agents can only act on a reflex-based reaction to the profit rate ● Alternative starting point: zero rationality ● Both approaches should be encouraged

Products of this work ● A thorough, 62 page analysis of Keen’s models ● The C#. NET based System-Dynamics Library Schum. PTR ● 5 model reproductions and +/- 40 model extensions in Schum. PTR ● 2 original agent-based system-dynamic models (1 working) ● Complete thesis length: 86 pages + 39 appendix pages

Future Development ● Modeling Schumpeter’s business cycle theory ● Releasing Schum. PTR as an open-source project ● Helping other students or academics use Schum. PTR

Fin Questions?