@ARTICLE{16370374_1990,
author = {Tauchen, George},
keywords = {algorithm, decision theory, dynamic programming, growth models, Markov analysis, statistical analysis, stochastic model, анализ Маркова, статистический анализ, статистический метод, стохастическая модель},
title = {Solving the Stochastic Growth Model by Using Quadrature Methods and
Value-Function Iterations },
journal = {Journal of Business and Economic Statistics},
year = {1990},
month = {},
volume = {8},
number = {1},
pages = {49-51},
url = {http://ecsocman.hse.ru/text/16370374/},
publisher = {},
language = {ru},
abstract = {Solution algorithm that uses value-function iterations on a discrete
state space is presented for the capital growth model set forth by
Taylor and Uhlig (1990). The grid for the exogenous process is set
using the quadrature technique, and the grid for the endogenous
capital process is set using a simple equispaced scheme in
logarithms. The discretized model is then solved with value-function
iterations. The algorithm is coded in GAUSS and run on a Compaq
386-25 computer. It appears to be very successful. When applied to a
slightly different version of the problem in which the exact solution
is known, the algorithm can approximate the exact solution with
4-digit accuracy and with a computational time of about 40 to 45
minutes. While the algorithm approximates the decision rule closely,
it still might appear to do poorly on criteria that test for
statistical violations of orthogonality conditions implied by the
Euler equation. This is because a value-function approach does not
impose the Euler equation explicitly on the discrete model. },
annote = {Solution algorithm that uses value-function iterations on a discrete
state space is presented for the capital growth model set forth by
Taylor and Uhlig (1990). The grid for the exogenous process is set
using the quadrature technique, and the grid for the endogenous
capital process is set using a simple equispaced scheme in
logarithms. The discretized model is then solved with value-function
iterations. The algorithm is coded in GAUSS and run on a Compaq
386-25 computer. It appears to be very successful. When applied to a
slightly different version of the problem in which the exact solution
is known, the algorithm can approximate the exact solution with
4-digit accuracy and with a computational time of about 40 to 45
minutes. While the algorithm approximates the decision rule closely,
it still might appear to do poorly on criteria that test for
statistical violations of orthogonality conditions implied by the
Euler equation. This is because a value-function approach does not
impose the Euler equation explicitly on the discrete model. }
}