EC4010 Notes, 2005 (Karl Whelan)

Topic 2: AK Models
While the Solow model is widely used as a baseline model of economic growth, it is still considered by many to be unsatisfactory as a description of the process leading to economic growth. This is because the model views improvements in total factor productivity (technological progress) to be the ultimate source of growth in output per worker, but does not provide an explanation as to where these improvements come from. In the language of economists, long-run growth is determined by something that is exogenous in the model. In the next two handouts, we will examine two dierent theories in which growth is endogenous, meaning determined by the actions of the economic agents described in the model. The two models achieve this quality using two very dierent routes: The one we look at now does so by getting rid of the assumption of diminishing marginal returns to capital accumulation, the second by endogenising the rate of technological progress.

The AK Model Recall that our production function in the Solow model was of the form
Yt = At Kt L1 t

0<<1

(1)

and the parameter being less than one implied diminishing marginal returns to capital accumulation, which had important implications for the model. This suggests that one route to endogenising the growth rate is to dispense with the dimishing marginal returns assumption altogether. The simplest such model just sets = 1 which gives Yt = At Kt (2)

For obvious reasons, this class of modelsrst discussed in Sergio Rebelos 1990 JPE paperare known as AK models. We can use similar methods to those used before to think about what a steady-state growth path might look like in this economy. Taking logs and derivatives, the growth rate of output is determined by Yt At K t = + Yt At K t Capital accumulation is still given by Kt = sYt Kt (4) (3)

EC4010 Notes, 2005 (Karl Whelan) So the growth rate of capital input is Kt Kt = sYt Kt = sAt

(5) (6)

Thus, the growth rate of output is Yt At = + sAt Yt At (7)

Note that in this model, the growth rate of output depends not only on the growth rate of A but also on the level. This means tha an increasing A will lead to an explosive path for output, so steady growth requires consistent with the AK model is Yt Yt
ss At At

= 0. For this reason, the steady-state growth rate

= sA

(8)

This simple model shows that getting rid of diminishing marginal returns to capital accumulation has a dramatic eect on the models predictions for the sources of growth. The steady-state growth rate depends positively on the savings rate and negatively on the depreciation rate, neither of which had any eect on long-run growth in the Solow model. Also, the fact that the level of technological eciency has an eect on the growth rate has important implications. While it may be impossible (if this model were the correct one) for A to grow without bounds, it may still be possible for government policy to aect the level of A, for instance through regulatory policies. Thus, the AK model presents a dramatically dierent picture of growth, and one in which the link between government actions and growth is much more obvious than in the Solow model.

A More Realistic AK Model There is an obvious criticism of the simple AK model: Where are the people? Clearly one needs human input to create the goods and services that constitute GDP. The following model shows, however, that one can extend the AK model to a case in which there is labour input as well as physical capital. The key here, however, that will distinguish this model from the Solow model is that the eect of labour input is determined by the stock of what is termed, human capital. Thus, a more skilled individual will be assumed to produce more output than an unskilled individual, and the total stock of such skills is called human

EC4010 Notes, 2005 (Karl Whelan)

capital. Crucially, human capital can be accumulated through education. Thus, both types of capital can be accumulatedthis turns out to imply that the model has similar properties to the AK model. Formally, this model views output as determined by
Yt = At Kt Ht1

(9)

where Kt is physical capital and Ht is human capital. So, growth is determined by At Kt Ht Yt = + + (1 ) Yt At Kt Ht We will assume that each type of capital accumulates according to Kt = sK Y K Ht = sH Y H (11) (12) (10)

Note that Ive assumed the same depreciation rate for both types of capital. This simplies some of the calculations for the model without changing any of the relevant conclusions. As before, we can dene capital output ratios xK xH = = K Y H Y (13) (14)

and, as before, we can write the growth rates of the two capital inputs as functions of the relevant capital-output ratios K K H H = = sK xK sH xH (15) (16)

With the simple AK model, we showed that it was not possible to have steady-state growth unless A was constant. The same turns out to be true for this model, but demonstrating it is a bit more subtle. First, note that one can re-write the production function as Y = A (xK Y ) (xH Y )1 Dividing across by Y this becomes 1 = Ax x1 K H (18) (17)

EC4010 Notes, 2005 (Karl Whelan) Taking logs and derivatives of this, we get x x A K H = (1 ) A xK xH

(19)

So, positive steady-state growth in A is only possible, if a weighted average of the growth rates of the capital-output ratios was negative. But, falling values of xK and xH would imply exploding values for the growth rates of K and H (see equations 15 and 16) so this is not consistent with steady-state growth. Equations 15 and 16 imply that the steady-state consistent with constant growth in K and H features both xK and xH being constant, and thus both type of capital growing at the same rate as output. This implies sK sH Ht Kt = = = Kt xK xH Ht This tells us that sK sH xK sK K sK = = = = = xK xH xH sH H sH So, along a steady-state growth path, the stock of human capital can be written as H= sH K sK (20)

(21)

(22)

We can use this fact to re-write the level of output on the steady-state path as Y = AK = A and the growth rate of output is Y Y = K K sK Y = K sH 1 = AsK sK = As s1 K H (25) (26) (27) (28) sH sK sH K sK
1 1

(23) K (24)

So, we see that allowing for both types of inputsphysical and human capitalto be continuously accumulated produces a model that has the same long-run implications as the

EC4010 Notes, 2005 (Karl Whelan)

basic AK model. The steady-state growth rate is now As s1 instead of As , so K H we have replaced the single savings rate with a geometric average of the two savings rate in the two-factor model. But substantively, the models implications are unchanged.

Discussion of AK Model The preceding model showed that AK models cannot be dismissed as easily as one might rst have thought. That said, there are still some reasons to doubt the predictions about long-run growth generated by this class of models: Non-Accumulable Factors: In the real world, there are factors of production that are in xed supply, such as land, or that cannot simply be accumulated indenitely such as energy. Remember that the AK model results are of a knife-edge variety: Any move away from all factors being accumulable, and we move back to the Solow model results. Treatment of Human Capital : The strict parallel between human capital and physical capital in the model just described is probably not completely accurate. For instance, not all expenditures on education will produce the same eect on output. The marginal boost to aggregate output of teaching someone how to read and write is presumably greater than that of a masters in economics! Thus, there may be limits to which one can increase growth just by boosting educational enrollment. It may also be reasonable to assume some role for non-skilled labour in the production function. For instance, in a well-known paper, Greg Mankiw, David Romer, and David Weil (Quarterly Journal of Economics, May 1992) put forward a model in which the production function is
Yt = At Kt Ht L1 t

0<<1

0<<1

(29)

In this case, there is a role for human capital, H, which can be accumulated but also for unskilled labour, L. Not all factors in this model can be accumulated, so its long-run growth rate is determined by the growth rate in At , as in the Solow model. However, the model behaves more like a Solow model with a higher capital share parameter, (i.e. a higher value of the parameter in the last handout). This implies larger level eects of changes in saving on output per worker, and also slower convergence speeds. The moral of this type of model is that the AK approach may be strictly wrong about savings rates aecting long-run growth rates, but they may have larger level eects than

EC4010 Notes, 2005 (Karl Whelan)

in the basic Solow model. Also, while the AK models may be wrong about there being no convergence dynamics towards a steady-state level of output, these dynamics may be slower than the Solow model predicts.