Collins Conference Room
Working Group
All day
Our campus is closed to the public for this event.
What caused the current economic crisis, how could it have been prevented, and what is the best way to get out of it? The crisis has been blamed on many factors, including the housing bubble, credit default swaps, too much leverage, too much concentration in the banking sector, inconsistent government interventions in the banking sector, lack of regulation, excessive CEO compensation, agency problems (e.g. by the firms rating bonds), and greed in general. There are just as many provisions for solving it, including Keynesian style economic stimuli, lowering taxes, providing guarantees to banks so they will lend more, and doing nothing. Similarly, there are many provisions for avoiding crises in the future, including stronger collateral requirements, market transparency requirements, restrictions on derivatives, and better monitoring of markets and leverage in particular.
To address the challenge of understanding this we propose to build an agent-based model in which all the major market institutions that played important roles in the recent crisis are represented: housing and mortgage markets, credit and capital markets, banks and other financial institutions, securitization processes and investors, consumers, firms, and regulatory agencies. Agent computing techniques permit representation of these different types of agents and gives us the ability to let them interact directly with one another. The rules specifying the behaviors and interactions of such agents will be based on detailed domain knowledge derived from conversations with experts, and will be guided whenever possible by advances in behavioral economics. Thus, instead of pre-specifying from the top-down the kinds of economic crises that can emerge, with agent modeling we stipulate the micro-level and use computers to find out how the actions of economically purposive agents leads to crisis or stability at the aggregate level. Having such a model will make it possible to explore alternative policies with much higher degrees of verisimilitude than conventional representative agent models and at much lower cost than actual policy experimentation with the real economy. It also provides a means to solve a long-standing problem in economics by unifying microeconomics with macroeconomics.
To get started we are focusing on a model of housing prices as a function of various factors, such as the availability of credit. This week's group is focused on producing a first cut of this model by identifying the right level of abstraction. This is being done to find good compromise between research goals and availability of data. This is very nitty-gritty - we are focused on questions about how the housing market works, how buyers and sellers make the decisions that set prices, and what aspects of human behavior in this narrow domain we can calibrate from the data available, which comes from a variety of different sources.
To address the challenge of understanding this we propose to build an agent-based model in which all the major market institutions that played important roles in the recent crisis are represented: housing and mortgage markets, credit and capital markets, banks and other financial institutions, securitization processes and investors, consumers, firms, and regulatory agencies. Agent computing techniques permit representation of these different types of agents and gives us the ability to let them interact directly with one another. The rules specifying the behaviors and interactions of such agents will be based on detailed domain knowledge derived from conversations with experts, and will be guided whenever possible by advances in behavioral economics. Thus, instead of pre-specifying from the top-down the kinds of economic crises that can emerge, with agent modeling we stipulate the micro-level and use computers to find out how the actions of economically purposive agents leads to crisis or stability at the aggregate level. Having such a model will make it possible to explore alternative policies with much higher degrees of verisimilitude than conventional representative agent models and at much lower cost than actual policy experimentation with the real economy. It also provides a means to solve a long-standing problem in economics by unifying microeconomics with macroeconomics.
To get started we are focusing on a model of housing prices as a function of various factors, such as the availability of credit. This week's group is focused on producing a first cut of this model by identifying the right level of abstraction. This is being done to find good compromise between research goals and availability of data. This is very nitty-gritty - we are focused on questions about how the housing market works, how buyers and sellers make the decisions that set prices, and what aspects of human behavior in this narrow domain we can calibrate from the data available, which comes from a variety of different sources.
SFI Host:
Doyne Farmer