New Economics Model Simulates U.S. Economy

Researchers at Sandia National Laboratories are developing a microsimulation model of the US economy that has the potential to significantly improve capabilities for analyzing and comparing economic policies.

The new computational tool, called Aspen, capitalizes on Sandia’s expertise in both evolutionary learning and parallel computing to provide several major advantages over more traditional economic models:

• Models the economy in a single, consistent calculation.
• Permits the impact of various legal, regulatory and policy changes such as monetary policy, tax law, or trade policies to be modeled in greater detail.
• Allows different sectors of the economy to be analyzed independently and/or integrated with other sectors to develop a better understanding of the whole economic process.
• Accurately simulates the behavior of basic decision-making segments within the economy, like households, banks, companies, and government.

Aspen uses agents to represent the various decision-making segments in the economy. Each segment may have one or more agents representing it in an analysis. One of Aspen’s key features is its ability to realistically reproduce the process used by actual economic agents to maximize utility or profit. Agents in Aspen not only can communicate with one another but also make "real-life" decisions. Through use of evolutionary learning techniques, the agents adapt their behavior according to changing economic conditions and past experience. In other words, they become smarter as they move through time.

Widespread applications

The model is designed to run on Sandia’s massively parallel computers. Through their use, researchers anticipate that the entire economy can be modeled in sufficient detail to be realistic. This would give Aspen widespread application in the following areas:

• Impact analysis of acts of terrorism.
• Economic and financial market forecasting for the government, banks, brokerage firms, or venture capitalists.
• Impact analysis of tax law or government policy changes.
• Impact studies of technology shifts for industry, consumers, or market research firms.

Technical approach

The most common tools currently used for economic modeling are either macroeconomic or computable general equilibrium (CGE). Both rely on regression analysis of aggregate data to develop parameters for use in economic forecasting. These macromodels can provide accurate forecasts, but problems arise when totally new economic policies are introduced with no past relevant data that can be used to develop modeling parameters. Also, modeling of shocks or discontinuities imposes certain limitations on the analysis. Although microsimulation could offer more modeling flexibility, it had been hampered by both a lack of essential microdata and the extensive calculation time required to micromodel something as large and complex as the US economy.

Advances in massively parallel techniques for solving complex modeling problems, coupled with new rich sources of microlevel data, have now made microsimulation a viable economic analysis tool.

Aspen microsimulates the economy using time-dependent Monte Carlo methods. These methods are ideal for investigating the behavior of complex, nonlinear stochastic systems like the economy. Agents defined in a problem start each time increment making decisions much like their real-life counterparts. Decisions on actions to take are based either on probabilities computed from actual mircoeconomic data or on results of learning models. These decisions include purchasing products, hiring workers, selling bonds, collecting welfare payments, conducting open market operations, and others. Macroeconomic variables, such as gross domestic product, inflation (CPI), and the unemployment rate are computed as aggregate results of innumerable decisions by the individual economic agents.

Development Status

A model of a market economy was completed and is available for general use. More detailed models are now under development. See the reports below for more information.

For more information contact:

Rich Pryor
Phone: (505) 844-2332
e-mail: rjpryor@sandia.gov

Image Caption: The graph shows results of a test problem that was run using the prototype model. These results demonstrate Aspen’s ability to predict business-like cycles in an economy where prices and inventories are allowed to vary.

Reports:

The following reports are available in Acrobat PDF format. A PDF reader for your computer can be found at the Adobe web site.

SAND96-0434: This report describes the prototype model.

SAND96-2459: This report describes the more detailed model and documents calculations on monetary policy.

SAND97-2093: This report documents the analysis done with Aspen on transitional economies.

Business Week article.

Sandia News article.

Newer reports and files related to this subject can be found here.