1st Prize in the Complexity and Macroeconomics Competition
Poledna, S., Miess, M., and Hommes, C. (2020). Economic forecasting with an agent-based model.
Complexity and Macroeconomics Prize
The ‘Rebuilding Macroeconomics’ project, funded by the Economic and Social Research Council and hosted by the National Institute of Economic and Social Research, awarded the paper
Poledna, S., Miess, M., and Hommes, C. (2020). Economic forecasting with an agent-based model. SSRN working paper, http://dx.doi.org/10.2139/ssrn.3484768 .
the 1st Prize in the Complexity and Macroeconomics Competition:
On Friday, the 21st of May, the winning papers hold a presentation for a general audience, which is free to register:
Below is a short information about the why, what and how behind this paper.
Why:
Agent-based modelling (ABM) has been a promising technology for decades, with the hope that it would push forward innovation in macroeconomic modelling. This hope has rested on the notion that by using modern computer technology, an economy can be modelled more realistically and based on large amounts of data (“big data”). However, most agent-based models have so far remained at the stage of describing hypothetical or experimental economies. Typical for standard ABMs in the literature is a high level of instability with large, empirically unseen fluctuations - especially in the first model periods - which make their practical application highly difficult (the so-called “burn-in” phase). These standard ABMs could therefore not be used (1) for empirically-based economic forecasting or (2) for evaluating political measures or economic shocks for economies that actually exist - such as the Austrian economy. These issues resulted in a significant research gap that we have closed.
What:
As part of our modelling team, we have designed, implemented and simulated a new prototype of ABM. This ABM depicts every household and every company in the Austrian economy based on data from the national accounts. The ABM is implemented in Matlab and runs on a supercomputer. This model is more stable than the ABMs common in the literature, and is therefore suitable for forecasting and evaluating policy measures.
How:
The model is calibrated to data from the national accounts - input-output tables, demographic data for households and companies, government expenditure statistics, micro-data at company level (Sabina database), etc. - for several years. After this calibration procedure, it was used in several applications.
(1) Forecast comparison: It is shown that the ABM works just as well as standard models in forecasting the key time series for the Austrian economy (GDP, inflation, consumption, investment). As comparison models, we use established time series models - the forecasting capabilities of which are good to very good (AR, VAR models) - as well as the most widely used economic theory-based model (a DSGE model). This paper was awarded the complexity and macroeconomics prize, see https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3484768
(2) Further applications of this validated ABM:
a. Effects of the COVID-19 pandemic on the Austrian economy, see https://iiasa.ac.at/web/home/resources/publications/IIASAPolicyBriefs/pb26.html.
b. Economic effects of flooding disasters in Austria, see https://arxiv.org/abs/1801.09740.