- Title
- Solving models with off-the-shelf software: an example of potential pitfalls associated with the use and abuse of default parameter settings
- Creator
- Herbert, Ric D.; Stemp, Peter J.
- Relation
- 18th World IMACS Congress and MODSIM09 International Congress on Modelling and Simulation. The 18th World IMACS Congress and MODSIM09 International Congress on Modelling and Simulation: Proceedings (Cairns, Qld 13-17 July, 2009) p. 1147-1153
- Relation
- http://www.mssanz.org.au/modsim09/authorsH-K.htm#h
- Publisher
- Modelling and Simulation Society of Australia and New Zealand and International Association for Mathematics and Computers in Simulation
- Resource Type
- conference paper
- Date
- 2009
- Description
- When working with large-scale models or numerous small models, there can be a temptation to rely on default settings in proprietary software to derive solutions to the model. In this paper we show that, for the solution of non-linear dynamic models, this approach can be inappropriate. We consider a simple linear model with two stable eigenvalues (real-valued or complex-valued) and one unstable eigenvalue (real-valued). The chosen model is an extended version of a model previously described by Turnovsky (2000). It can be derived from the Sargent and Wallace (1973) extension of the Cagan (1956) model by adding a labour market defined by both employment and wages and by introducing sluggish adjustment for both these variables. We choose parameters for the model. The configuration of eigenvalues can be changed by choosing alternative parameter values. It is possible to choose parameter configurations that generate complex-valued eigenvalues and real-valued eigenvalues. For this paper we focus on the real-valued eigenvalues. Alternative linear and non-linear specifications of the model are examined. One version of the model, expressed in levels, is highly non-linear. A second version of the model, expressed in logarithms, is linear. The dynamic solution of each model version has a combination of stable and unstable eigenvalues so that any dynamic solution requires the calculation of appropriate "jumps" in endogenous variables. We can derive a closed-form solution of the model, which we use as our "true" benchmark, for comparison with computational solutions of both linear and non-linear models. Our approach is to compare the "goodness of fit" of reverse-shooting solutions for both the linear and non-linear model, by comparing the computational solutions with the benchmark solution. To make all three solutions comparable, each solution is expressed in levels, with the closed-form and computational solutions of the linear model being converted to levels by taking exp of endogenous variables. Under the basic solution method with default settings and using the "eyeball" metric, we show that there is significant difference between the computational solution for the non-linear model and the benchmark closed-form solution. We show that this result can be substantially improved using modifications to the solver and to parameter settings. A Table documents the different approaches considered in determining the best method for solving the non-linear model. We discuss how we have used a variety of approaches before finding a solution methodology that did not fail. In this paper, we have considered linear and non-linear versions of a dynamic macroeconomic model. We have shown that standard default settings in proprietary software give unsatisfactory results for solving the dynamic path of the non-linear model. Our results demonstrate how the solution of the model can be substantially improved by changes to parameter settings.
- Subject
- solving non-linear models; reverse-shooting; computational economics; computer software
- Identifier
- http://hdl.handle.net/1959.13/919481
- Identifier
- uon:8882
- Identifier
- ISBN:9780975840078
- Language
- eng
- Reviewed
- Hits: 600
- Visitors: 576
- Downloads: 2
| Thumbnail | File | Description | Size | Format |
|---|