Lecture The Solow Growth Model

1. The basic assumptions of the model 1.1 Inputs and output 1.2 Production function 1.3 Evolution of the inputs into production 2. The dynamics of the model 2.1 The dynamics of k 2.2 The Balanced Growth Path 2.3 The impact of a change in the saving rate 2.3.1 The impact on output 2.3.2 The impact on consumption 2.4 Quantitative implications 2.4.1 The effect on output in the long run 2.4.2 The speed of convergence 2.4.3 Growth accounting 3. Environment and natural resources

1. The basic assumptions of the model 1.1 Inputs and outputs 1. The Solow Model focuses on four variables: output (Y), capital (K), labour (L), and knowledge or the effectiveness of labour (A). (1.1) Y(t) F((K(t), A(t), L(t)),

where t denotes time. 2. Two features of the production function: a) Time does not enter the function directly but only through K, L and A. That is, output changes over time only if the inputs to production change. b) A and L enter multiplicatively. AL is referred to as effective labour. 1.2 Production function 1. Constant returns to scale in its two arguments, capital and effective labour. (1.2) F(cK, cAL) cF(K, AL) for all c 0.

The assumption of constant returns can be thought as combining two assumptions: a) The economy is big enough, so that the gains from specialisation have been exhausted. b) Inputs other than capital, labour and knowledge are relatively unimportant. In particular, the model neglects land and other natural resources. Production function in intensive form. Setting c 1/AL in equation (1.2) yields (1.3)
F( K 1 F ( K , AL) ,1) = AL AL

Here K/AL is the amount of capital per unit of effective labour, and F(K, AL)/AL is Y/AL, output per unit of effective labour. Define k K/AL, y Y/AL, and f (k) F(k, 1). Then we can rewrite (1.3) as (1.4)
y = f (k )

f(k)

FIGURE 1.1 An example of a production function

The intensive-form production function, f (k), is assumed to satisfy

f (0) 0, f (k) >0, f (k) <0.

Since F(K, AK) = AL f (K/AL), then dF(K, AL)/dK = AL f (K/AL)(1/AL) = f (k). Thus the assumptions that f (k) is positive and f (k) is negative imply that the marginal product of capital is positive, but that it declines as capital (per unit of effective labour) rises. In addition, f () is assumed to satisfy the Inada conditions (Inada, 1964): lim k 0 f (k ) = lim k f (k ) = 0 . A production function satisfying f ()>0, f ()<0 is shown in Figure 1.1. A specific example of a production function is Cobb-Douglas function, (1.5) F(K, AL) K ( AL)1- , 0 < < 1

It has constant returns: (1.6) F (cK , cAL) = (cK ) (cAL)1- = c c1 K ( AL)1- = cF ( K , AL) Intensive form: (1.7)

f (k ) F (
K = AL

K ,1) AL

= k
1.3 Evolution of the inputs into Production

the model is set in continuous time: the variables of the model are defined as every point in time. the initial levels of capital, labour, and knowledge are taken as given. labour and knowledge grow at constant rates:
L(t ) = nL(t ) A(t ) = gA(t )

(1.8) (1.9)

where n and g are exogenous parameters and where a dot over a variable denotes a derivative

with respect to time (that is, X (t ) = dX(t)dt.)

The growth rate of a variable refers to its proportional rate of change. That is, a growth

rate of some parameter X equals X (t ) / X (t ) . X (t ) / X (t ) equals d ln X (t ) / dt . To illustrate: (1.10)

d ln X (t ) d ln X (t ) dX (t ) = dt dX (t ) dt

= Thus, (1.11) (1.12)

1 X (t ) X (t )

ln L(t ) = [ ln L(0) ] + nt
ln A(t ) = [ ln A(0) ] + gt

where L(0) and A(0) are the values of L and A at time 0. Exponentiating both sides of these equations gives us (1.13) (1.14)
L(t ) = L(0)e nt A(t ) = A(0)e gt

Thus, our assumption is that L and A each growth exponentially.

output = consumption+investment. the fraction of output devoted to investment, s, is exogenous and constant. One unit of output devoted to investment yields one unit new capital. In addition, existing capital depreciates at rate . Thus,
K (t ) = sY (t ) K (t )

(1.15)

It is assumed that the sum of n, g, and is positive.

2. The dynamics of the model 2.1 The dynamics of k

Define k=K/AL (1.16)

K (t ) K (t ) A(t ) L(t ) + L(t ) A(t ) k (t ) = A(t ) L(t ) [ A(t ) L(t ) ]2

K/AL = k. From (1.8) and (1.9), L/ L and A/ A are n and g, respectively. K is given by (1.15). Substituting these facts into (1.16) yields

(1.17)

k (t ) =

sY (t ) K (t ) k (t )n k (t ) g A(t ) L(t )

=s Since Y/AL=f(k) (1.18)

Investment per unit of effective labour

Y (t ) k (t ) nk (t ) gk (t ) A(t ) L(t )

k = sf ( k ) ( n + g + ) k (t)

(n + g + )k

sf ( k )

FIGURE 1.2 Actual and break-even investment

Equation (1.18) is the key equation of the Solow model, where sf(k), is the actual investment per unit of effective labour; (n+g+ )k is break-even investment, or the amount of investment that must be done just to keep k at its existing level, see Figure 1.2.

Since f(0)=0, (n+g+ )k = sf(k) at k=0. In view of the Inada conditions f (k ) is large and thus that the sf(k) line is steeper than the (n+g+ )k line. Thus for small values of k, actual investment is larger than break-even investment. In view of the Inada conditions also at some point, the slope of the actual investment line falls below the slope of the break-even investment line. With sf(k) line flatter than the (n+g+ )k line, the two must eventually cross. Finally, since f (k ) <0, the two lines intersect only once for k>0 at k .

0
k

FIGURE 1.3 The phase diagram for k in the Solow model 2.2 The Balanced Growth Path

K / K = n+g. Y /Y= n+g

K/L Y /L = =g K/L Y /L

2.3

The Impact of a change in the saving rate

2.3.1 The impact on output

The increase in s shifts the actual investment line upward, and so k rises, see Figure 1.4
Investment per unit of effective labour !

(n + g + )k

NEW

f (k )

OLD

f (k )

kOLD

kNEW

FIGURE 1.4 The effects of an increase in the saving rate on investment

Growth rate of Y/L

lnY/L

FIGURE 1.5 The effects of an increase in the saving rate.

2.3.2 The impact on consumption

Whether consumption eventually exceeds its level before rise in s is not immediately clear from Figure 1.5 . Let c denote consumption per unit of effective labour on the balanced growth path. ( c = f (k ) - s f (k ) ). On the balanced growth path, s f (k ) = (n + g + )k . Thus, (1.19) c = f (k ) ( n + g + )k

k is determined by s and the other parameters of the model, n, g and ; we can therefore write k = k (s, n, g, ). Thus, (1.19) implies
(1.20) dc dk ( s, n, g , ) f (k ( s, n, g , )) (n + g + ) = ds ds

If f (k ) is less than n+g+ , then the additional output from the increased capital is not enough to maintain the capital stock at its higher level. Thus, consumption must fall to maintain the higher capital stock. If f (k ) exceeds n+g+ , on the other hand, there is more than enough additional output to maintain k at its higher level, and so consumption rises. f(k)
Output and investment per unit of effective labour

c
(n + g + )k

sf(k)

k
f(k)
Output and investment per unit of effective labour

(n + g + )k

sf(k)

f(k)

Output and investment per unit of effective labour

(n + g + )k

sf(k)

FIGURE 1.6 Output, investment, and consumption on the balanced growth path

f (k ) can be either smaller or larger than n+g+ , see Figure 1.6. Top panel: f (k ) < n+g+ , so the increase in the saving rate lowers consumption even when economy has reached the new balanced growth path. Middle panel: an increase in s raises consumption in the long run. Bottom panel, f (k ) = n+g+ , i.e. f(k) and (n+g+ )k lines are parallel at k= k . In this case, a marginal change in s has no effect on consumption in the long run, and consumption is at its maximum possible level among balance growth paths.. This value of k is known as the golden rule level of the capital stock.

2.4

Quantitative implications

2.4.1 The effect on output in the long run

The long-run effect of a rise in saving on output is given by (1.21) dy dk ( s, n, g , ) = f (k ) ds ds

where y = f (k ) is the level of output per unit of effective labour on the balanced growth path. Thus to find dy / ds , we need to find dk / ds . To do this, note that k is defined by the condition that k =0; thus k satisfies (1.22) s f ( k (s, n, g, ))= (n + g + )k ( s, n, g , ).

Equation (1.22) holds for all values of s (and of n, g, and ). Thus, (1.23) sf (k ) dk dk + f ( k ) = (n + g + ) ds ds

(arguments of k are omitted for simplicity). Rearranged: dk f (k ) (1.24) = ds (n + g + ) sf (k ) Substituting (1.24) into (1.21) yields (1.25)
dy f (k ) f (k ) = ds (n + g + ) sf (k )

Convert (1.25) to an elasticity by multiplying both sides by s/ y . Use sf (k ) = (n + g + ) k to substitute for s. We get: (1.26)
s dy s f (k ) f (k ) = y ds f (k ) (n + g + ) sf (k )

k f (k ) / f (k ) 1 [k f (k ) / f (k ]

k f (k ) / f (k ) is the elasticity of output with respect to capital at k = k . Denoting this by K (k ) , we have

(1.27)

K (k ) s dy = y ds 1 K (k )

If capital earns its marginal product, the share of total income that goes to capital on the balanced k f (k ) , or K (k ) . growth path is f (k )
2.4.2 The speed of convergence

Our goal is thus to determine how rapidly k approaches k . We know that k is determined by k: recall that the key equation of the model is k = sf (k ) (n + g + )k (see 1.18). Thus we can write k = k (k). When k equals k , k is zero. A first-order Taylor-series approximation of k (k) around k= k therefore yields

(1.28)

d k (k ) (k k ) dk k =k

That is, k is approximately equal to the product of the difference between k and k and the derivative of k with respect to k at k= k .

d k (k ) Let denote dk

k =k

With this definition, (1.28) becomes

(1.29)

( k (t ) k (t ))

Since k is positive when k is slightly below k and negative when it is slightly above,

d k (k ) dk

is negative. Equivalently, is positive.

k =k

Equation (1.29) implies that in the vicinity of the balanced growth path, k moves toward k at a speed approximately proportional to its distance from k . That is, the growth rate of k (t ) k (t ) is approximately constant and equal to - . This implies (1.30) k (t ) k + e t k (0) k

evaluating the resulting expression at k = k yields d k (k ) dk

k =k

(1.31)

= sf (k ) (n + g + )

= (n + g + ) sf (k ) = (n + g + )
(n + g + )k f (k ) f (k )

= 1 K (k ) (n + g + ) ,

where the third line again uses the fact that sf (k ) = (n + g + )k to substitute for s, and where the last line uses the definition of K . Thus, k converges to its balanced growth path value at
1 K (k ) (n + g + ) . In addition, one can show that y approaches y at the same rate that k approaches k . That is, y (t ) y e t y (0) y
2.4.3 Growth accounting

In the Solow model, long-run growth of output per worker depends only on technological progress. But short-run growth can result from either technological progress or capital accumulation. Growth accounting, which was pioneered by Abramovitz (1956), Solow (1957) provides a way of tackling the subject. To see how growth accounting works, consider again the production function Y (t ) = F ( K (t ), A(t ) L(t )) . This implies (1.33)
Y (t ) =

dY (t ) dY (t ) dY (t ) K (t ) + L(t ) + A(t ) dK (t ) dL(t ) dA(t )

dY / dL and dY / dA denote [dY / dAL] A and [dY / dAL]L respectively. Dividing both sides by K (t ) L(t ) A(t ) , , Y(t) and rewriting the terms on the right-hand side yields (also multiplying by ) K (t ) L(t ) A(t )

(1.34)

Y (t ) K (t ) dY (t ) K (t ) L(t ) dY (t ) L(t ) A(t ) dY (t ) A(t ) = + + Y (t ) Y (t ) dK (t ) K (t ) Y (t ) dL(t ) L(t ) Y (t ) dA(t ) A(t )

K (t ) K (t ) + L (t ) L(t ) + R(t ) K (t ) L(t )

Here L (t ) is the elasticity of output with respect to labour (

dYK ) at time t, K (t ) is again the dKY

A(t ) dY (t ) A(t ) L(t ) . Subtracting from elasticity of output with respect to capital, and R (t ) Y (t ) dA(t ) A(t ) L(t ) both sides and using the fact that K (t ) + L (t ) =1 gives an expression for the growth rate of output per worker:

(1.35)

Y (t ) L(t ) K (t ) L(t ) L(t ) = K (t ) + L (t ) + R(t ) Y (t ) L(t ) K (t ) L(t ) L(t )

K (t ) L(t ) = K (t ) (1 L (t )) + R(t ) K (t ) L(t ) K (t ) L(t ) = K (t ) + R (t ) K (t ) L(t )

The growth rates of Y, K and L are straightforward to measure. And we know that if capital earns its marginal product K (t ) can be measured using data on the share of income that goes to capital. R (t ) can be then measured as the residual in (1.35). Thus 1.35 provides a way of decomposing the growth of output into the contribution of growth of capital per worker and a remaining term, the Solow residual. The Solow residual is sometimes interpreted as a measure of the contribution of technological progress.
Convergence

Do poor countries tend to grow at faster than rich countries rate? There are at least three reasons that one might expect such convergence. - The Solow model predicts countries converge to their balanced growth paths. - The Solow model implies that the rate of return on capital is lower in countries with more capital per worker. - Differences shrink as poor countries gain access to state-of-art methods. Baumol (1986) examines convergence from 1870 to 1979 among 16 industrialised countries using Maddisons (1982) data. Baumol regresses output growth over this period on a constant and initial income; that is, he estimates (he considers output per person rather than output per worker): Y Y Y (1.36) ln ln = a + b ln + i N i ,1979 N i ,1870 N i ,1870

Here ln(Y / N ) is log per person, i is an error term, and i indexes countries. If there is convergence, b will be negative: countries with higher initial incomes have lower growth. A value for b -1 corresponds to perfect convergence: higher initial income on average lowers subsequent growth one-for-one, and so output per person in 1979 is uncorrelated with its value in 1870. A value for b of 0, on the other hand, implies that growth is uncorrelated with initial income and thus there is no convergence. The results are: (1.37) Y Y Y ln ln = 8.457 0.995ln , (0.094) N i ,1979 N i ,1870 N i ,1870

R 2 = 0.87, s.e.e = 0.15

where the number in parentheses, 0.094, is the standard error of the regression coefficient. The regression suggests almost perfect convergence. The estimate of b is almost exactly equal to -1, and it is estimated fairly precisely; the two-standard-error confidence intervals is (0.81, 1.18). In this sample, per capital income today is essentially unrelated to per capita income 100 years ago. De Long (1988) demonstrates, however, that Baumols finding is largely spurious: - Sample selection. Since historical data are constructed retrospectively, the countries that have long data series are generally those that are the most industrialised today. - Measurement error. Estimates of real income per capita in 1870 are imprecise.
Saving and investment

Fieldstein and Horioka (1980) examine the association between saving and investment rates. They find that, contrary to this simple view, saving and investment rates are strongly correlated. Specifically, they run a cross-country regression for 21 industrialised countries of the average share of investment and GDP during the period 1960-1974 on a constant and average share of saving in GDP over the same period: (1.40)
I S = 0.035+ 0.887 , Y i (0.018) (0.074) Y i

R 2 = 0.91

Strong correlation. Possible explanations: - Existence of significant barriers to capital mobility. - There are underlying variables that affect both saving and investment. - Government policies that offset forces that would otherwise make saving and investment differ.

3.

Environment and natural resources

The amounts of oil and other natural resources on earth are fixed. This could mean that any attempt to embark on a path of perpetually rising output will eventually deplete these resources and must therefore fail. The fixed supply of land may become a binding constraint on our ability to produce. An ever-increasing output may generate an ever-increasing stock of pollution that will bring growth to a halt.

Natural resources and Land: a baseline

To extend our analysis using Cobb-Douglas production function to include natural resources and land yields: 1 (1.41) Y (t ) = K (t ) R(t ) T (t ) [ A(t ) L(t ) ] ,

> 0, > 0, > 0, + + < 1

Here R denotes resources used in production, and T denotes the amount of land. The dynamics of capital, labour and the effectiveness of labour are: K (t ) = sY (t ) K (t ) ,

L(t ) = nL(t ) , A(t ) = gA(t ) .

New assumptions concern resources and land: 1) since the amount of land on earth is fixed, the long-run quantity in the production cannot be growing. Thus: (1.42)

T (t ) = 0

2) since resource endowments are fixed and that resources are used in production imply that resource use must eventually decline. Thus: (1.43)

R(t ) = bR(t ), b > 0

The presence of resources and land in the production function means that K/AL no longer converges to some point. As a result, we cannot use our previous approach of focusing on K/AL to analyse the behaviour of this economy. A useful strategy is to ask whether there can be a balanced growth path and, if so, what the growth rates of the economys variables are on that path. By assumption, A, L, R and T are each growing at a constant rate. Thus what is needed for a balanced growth path is that K and Y each grow at a constant rate. The equation of motion for capital, K (t ) = sY (t ) K (t ) , implies that the growth rate of K is

(1.44)

K (t ) Y (t ) =s K (t ) K (t )

Thus for the growth rate of K to be const, Y/K must be const. That is, the growth rates of Y and K must be equal. Using (1.41) to find when this can occur (take logs of both sides of (1.41)): (1.45)
ln Y (t ) = ln K (t ) + ln R (t ) + ln T (t ) + (1 ) [ ln A(t ) + ln L(t ) ]

Differentiate both sides of this expression with respect to time. (1.46)

gY (t ) = g K (t ) + g R (t ) + gT (t ) + (1 ) [ g A (t ) + g L (t ) ]

where g X denotes growth rate of X. The growth rates of R, T, A and L are b, 0, g and n respectively. Thus (1.46) simplifies to (1.47)
gY (t ) = g K (t ) b + (1 ) [ n + g ]

We can now use our finding above that gY and g K must be equal if economy is on a balanced growth path. Imposing gY = g K on (1.47) and solving for gY gives us (1.48)
BGP gY =

(1 )(n + g ) b , 1

BGP where gY denotes the growth rate of Y on a balanced growth path. But does the economy, in fact, converges to this balanced growth path? From (1.47), we know that if g K exceeds its balanced-growth-path value, gY does as well, but by less than g K does. Thus is g K exceeds its balanced-growth-path value, Y/K is falling. Equation (1.44) tells that g K equals s(Y/K)- . Thus, if Y/K is falling, g K is falling as well. That is, if g K exceeds it balanced-growth-path value (which occurs when Y/K exceeds its balanced-growth-path value), it is falling. Similarly, if it is less than its balanced-growth-path value, it is rising. Thus, g K converges to its balanced-growth-path value, and so the economy converges to its balancedgrowth-path. Equation (1.48) implies that the growth rate of output per worker on the balanced growth path is

(1.49)

BGP BGP BGP gY / L = g Y g L

(1 )(n + g ) b n 1 (1 ) g b ( + )n 1

The declining quantities of resources and land per worker are drags on growth. But technological progress (g) is a spur to growth.
An illustrative calculation

To calculate growth drag, Nordhaus (1992) to ask first how much greater growth would be if resources and land per worker were constant. Thus instead of T (t ) = 0 and R(t ) = bR(t ), use

T (t ) = nT (t ) and R (t ) = nR(t ) . 1 % BGP (1.50) gY / L = (1 ) g 1

The growth drag is the difference between growth in this hypothetical case and growth in the case of resource and land limitations. (1.51)
BGP % BGP Drag= gY / L gY / L

(1 ) g (1 ) g b ( + )n 1

Thus, the growth drag is increasing in resources share ( ), lands share ( ), the rate that resource use is falling (b), the rate of population growth (n) and capitals share ( ).
A complication

b + ( + )n = 1

The stock of land of land is fixed, and resource use must eventually fall, it may still appear that those limitations must eventually become a binding constraint on the ability to produce. The reason that this does not occur in our model is that the production function is CobbDouglas, which has constant returns.