Economic Growth Second Edition
Download 0.79 Mb. Pdf ko'rish
|
BarroSalaIMartin2004Chap1-2
2.6.6
Speeds of Convergence Log-Linear Approximations Around the Steady State We want now to provide a quan- titative assessment of the speed of convergence in the Ramsey model. We begin with a log-linearized version of the dynamic system for ˆk and ˆc, equations (2.24) and (2.25). This approach is an extension of the method that we used in chapter 1 for the Solow–Swan model; the only difference here is that we have to deal with a two-variable system instead of a one-variable system. The advantage of the log-linearization method is that it provides a closed-form solution for the convergence coefficient. The disadvantage is that it applies only as an approximation in the neighborhood of the steady state. Appendix 2A examines a log-linearized version of equations (2.24) and (2.25) when expanded around the steady-state position. The results can be written as log[ ˆy (t)] = e −βt · log[ ˆy(0)] + (1 − e −βt ) · log( ˆy ∗ ) (2.40) where β > 0. Thus, for any t ≥ 0, log[ ˆy(t)] is a weighted average of the initial and steady- state values, log[ ˆy (0)] and log( ˆy ∗ ), with the weight on the initial value declining ex- ponentially at the rate β. The speed of convergence, β, depends on the parameters of technology and preferences. For the case of a Cobb–Douglas technology, the formula for the convergence coefficient (which comes from the log-linearization around the steady-state position) is 2 β = ζ 2 + 4 · 1 − α θ · (ρ + δ + θx) · ρ + δ + θx α − (n + x + δ) 1 /2 − ζ (2.41) where ζ = ρ − n − (1 − θ) · x > 0. We discuss below the way that the various parameters enter into this formula. 112 Chapter 2 Equation (2.40) implies that the average growth rate of per capita output, y, over an interval from an initial time 0 to any future time T ≥ 0 is given by (1/T ) · log[y(T )/y(0)] = x + (1 − e −βT ) T · log[ ˆy ∗ / ˆy(0)] (2.42) Hold fixed, for the moment, the steady-state growth rate x, the convergence speed β, and the averaging interval T . Then equation (2.42) says that the average per capita growth rate of output depends negatively on the ratio of ˆy (0) to ˆy ∗ . Thus, as in the Solow–Swan model, the effect of the initial position, ˆy (0), is conditioned on the steady-state position, ˆy ∗ . In other words, the Ramsey model also predicts conditional, rather than absolute, convergence. The coefficient that relates the growth rate of y to log[ ˆy ∗ / ˆy(0)] in equation (2.42), (1 − e −βT )/T , declines with T for given β. If ˆy(0) < ˆy ∗ , so that growth rates decline over time, an increase in T means that more of the lower future growth rates are averaged with the higher near-term growth rates. Therefore, the average growth rate, which enters into equation (2.42), falls as T rises. As T → ∞, the steady-state growth rate, x, dominates the average; hence, the coefficient (1 − e −βT )/T approaches 0, and the average growth rate of y in equation (2.42) tends to x. For a given T , a higher β implies a higher coefficient (1 − e −βT )/T . (As T → 0, the coefficient approaches β.) Equation (2.41) expresses the dependence of β on the underlying parameters. Consider first the case of the Solow–Swan model in which the saving rate is constant. As noted before, this situation applies if the steady-state saving rate, s ∗ , shown in equation (2.34) equals 1 /θ or, equivalently, if the combination of parameters α · (δ + n) − (δ + ρ)/θ − x · (1 − α) equals 0. Suppose that the parameters take on the baseline values that we used in chapter 1: δ = 0.05 per year, n = 0.01 per year, and x = 0.02 per year. We also assume ρ = 0.02 per year to get a reasonable value for the steady-state interest rate, ρ + θx. As mentioned in a previous section, for these benchmark parameter values, the saving rate is constant if α = 0.3 when θ = 17 and if α = 0.75 when θ = 1.75. With a constant saving rate, the formula for the convergence speed, β, simplifies from equation (2.41) to the result that applied in equation (1.45) for the Solow–Swan model: β ∗ = (1 − α) · (x + n + δ) We noted in chapter 1 that a match with the empirical estimate for β of roughly 0.02 per year requires a value for α around 0.75, that is, in the range in which the broad nature of capital implies that diminishing returns to capital set in slowly. Lower values of x + n + δ reduce the required value of α, but plausible values leave α well above the value of around 0.3, which would apply to a narrow concept of physical capital. Growth Models with Consumer Optimization 113 In the case of a variable saving rate, equation (2.41) determines the full effects of the various parameters on the convergence speed. The new element concerns the tilt of the time path of the saving rate during the transition. If the saving rate falls with ˆk, the conver- gence speed would be higher than otherwise, and vice versa. For example, we found before that a higher value of the intertemporal-substitution parameter, θ, makes it more likely that the saving rate would rise with ˆk. Through this mechanism, a higher θ reduces the speed of convergence, β, in equation (2.41). If the rate of time preference, ρ, increases, the level of the saving rate tends to fall (see equation [2.34]). The effect on the convergence speed depends, however, not on the level of the saving rate but on the tendency for the saving rate to rise or fall as the economy develops. A higher ρ tends to tilt downward the path of the saving rate. The effective time-preference rate is ρ + θ · ˙c/c. Because ˙c/c is inversely related to ˆk, the impact of ρ on the effective time-preference rate is proportionately less the lower is ˆk. Therefore, the saving rate tends to decrease less the lower ˆk, and, hence, the time path of the saving rate tilts downward. A higher ρ tends accordingly to raise the magnitude of β in equation (2.41). It turns out with a variable saving rate that the parameters δ and x tend to raise β, just as they did in the Solow–Swan model. The overall effect from the parameter n becomes ambiguous but tends to be small in the relevant range. 25 The basic result, which holds with a variable or constant saving rate, is that, for plausible values of the other parameters, the model requires a high value of α—in the neighborhood of 0.75—to match empirical estimates of the speed of convergence, β. We can reduce the required value of α to 0.5–0.6 if we assume very high values of θ (in excess of 10) along with a value of δ close to 0. We argued before, however, that very high values of θ make the steady-state saving rate too low, and values of δ near 0 are unrealistic. In addition, as we show later, values of α that are much below 0.75 generate counterfactual predictions about the transitional behavior of the interest rate and the capital-output ratio. We discuss in chapter 3 how adjustment costs for investment can slow down the rate of convergence, but this extension does not change the main conclusions. Download 0.79 Mb. Do'stlaringiz bilan baham: |
Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2025
ma'muriyatiga murojaat qiling
ma'muriyatiga murojaat qiling