Economic Growth From A Classical Perspective

Economic Growth from a Classical Perspective by Duncan K. Foley and Adalmir Marquetti Department of Economics, Barnard College, Columbia University, New York, NY 10027 USA ([email protected]) and Department of Economics, Graduate Faculty, New School for Social Research, 65 Fifth Avenue, New York, NY 10003 USA

The efficiency schedule for an economy, a line having for its vertical intercept output per worker, or labor productivity, and for its horizontal intercept output per unit of capital, or capital productivity, is a method of visualizing patterns of economic growth and technical change over time. Efficiency schedules constructed from the long data series of Duménil and Lévy for the U.S. economy reveal a pattern of labor-saving, capital-using (Marx-biased) technical change, corresponding to the pattern underlying Marx's explanation of the falling rate of profit, punctuated by a period of pure factoraugmenting (Hicks-neutral) technical change. Data from the Penn World Tables supplemented by our capital stock estimates reveal a similar pattern of Marx-biased technical change over many economies in many time periods. The Marx-biased pattern of technical change appears in over half of the sample observations. There appears, however, to be no strong quantitative correlation in the data between the magnitudes of labor-saving and capital-using technical progress. Keywords: economic growth, technical change, Marx-bias.

2

Economic Growth

Foley-Marquetti

Introduction The purpose of this paper is to explore the existence of patterns of biased technical change in economic growth. The basic tool of analysis we use is the efficiency schedule, a version of Piero Sraffa's (1961) wage-profit rate relation.The efficiency schedule for an economy in a given time period, developed in detail in the next section, is determined by real labor productivity and the output-capital ratio, which we will, abusing language slightly, often refer to as capital productivity. The comparisons of efficiency schedules for a given economy over time reveals the pattern of technical change the economy has experienced. Using extensive data from the Penn World Tables (Summers and Heston, 1991) and long-run data for the U.S. economy (Duménil and Lévy, 1993), we look for systematic tendencies toward bias in the pattern of technical change. The data reveal a striking pervasive tendency for rises in labor productivity over time to be accompanied by declines in capital productivity. There are, however, significant and interesting exceptions to this pattern, in particular, periods in which both labor and capital productivity either rise or decline together. These periods of uniform technical progress and regress appear to be associated with historical watersheds in economic development which separate epochs of technical change biased toward labor productivity and against capital productivity. Economists have recognized this basic tendency for labor productivity to rise and capital productivity to fall for a long time, and have explained it from a variety of perspectives. The two broad approaches are the Classical interpretation, which sees these movements as the reflection of a bias in the adoption of technical changes induced by systematic incentives in the capitalist economy, and the Neoclassical interpretation, which sees these movements as occurring along the isoquant of a historically stable production function. We lean toward the Classical interpretation, and will comment below on this issue. But our main purpose here is to document the underlying tendency itself, rather than to offer explanations for it. The data we use are aggregated national output measures based on the market valuation of outputs, aggregated labor inputs, and aggregated capital measures based on the market valuation of different capital goods. The simple analytical framework we employ assumes a single measure of output, labor and capital to parallel the structure of the data. There are important issues of aggregation to be addressed in the use of each of these measures. They implicitly neglect changes in labor skills and the composition of labor inputs by skill, for example. They also cannot distinguish between changes in capital measure due to changes in price and composition of the stock of capital goods from changes due to uniform changes in the quantity of capital goods of each type, an issue raised sharply in the Cambridge Capital Debates. We will not address these issues explicitly in this paper, except to point out here that the existence of a pervasive pattern in the aggregate data poses a problem of explanation for any theoretical approach, and to note that our analytical apparatus, based as it is on Sraffa's conceptions, can in principle be extended to accomodate disaggregated capital data.

The efficiency schedule We begin by constructing the efficiency schedule as a framework for analyzing technical change in the course of economic growth. In any period we can measure the gross real output per year of an economy, X, which will be represented by the Penn World Table estimate of real GDP, its aggregate labor input, N, represented by the Penn World Table estimate of employed labor, and the capital stock net of depreciation, K, represented by our estimate of the real value of net capital, constructed by the methods described in the Appendix. It is convenient to work with ratios that are independent of the absolute scale of the economy, x = X/N, the ratio of real output to labor input, is a measure of labor productivity, and r = X/K, the ratio of real output to capital, which is a measure of capital productivity. Labor productivity has the units of real output per worker-year, and capital productivity has the units % per year, like an interest rate or growth rate. The ratio x/r = k is the capital-labor ratio for the economy.

Foley-Marquetti

Economic Growth

The efficiency schedule for an economy in (x,r) space is a straight line with vertical intercept equal to x and horizontal intercept equal to r. The capital-labor ratio is the negative of the slope of this line.

output per worker x slope = -k

% per year r Figure 1: The efficiency schedule represents labor productivity, x, capital productivity, r, and the capital labor ratio, k, as a straight line connecting (0,x) to (r,0) with slope = -k.

The efficiency schedule is also the basis for a convenient representation of the national income accounts. Taking the basic national product identity X = C + I, where C is consumption, I is gross investment, and government purchases of goods and services, and net exports have been consolidated into consumption and gross investment respectively, we can represent the product account on the efficiency schedule as a vertical line of height x at the gross growth rate g + d = I/K, divided by the efficiency schedule into social consumption (including the consumption of non-workers), c = C/N, and gross investment per worker, i = I/N. Here d represents depreciation charges as a fraction of the stock of capital. Similarly, we can represent the income side of the national accounts, X = W + Z, where W is the wage bill and Z = X - W is aggregate non-wage income and depreciation, as a vertical line at the gross profit rate, r + d = Z/K divided by the efficiency schedule into the wage per worker, w = W/N, and cash flow per worker, z = Z/N.

3

4

Economic Growth

Foley-Marquetti

output per worker x i z

c w

g+d

r+d

r

% per year

Figure 2: The efficiency schedule can also represent the output and income sides of the national accounts in per-worker terms.

Technical Change and the Efficiency Schedule The movement of the efficiency schedule over time gives a vivid representation of the pattern of technical change in a growing economy. Technical change that raises labor productivity while keeping capital productivity constant (purely laborsaving, or labor-augmenting, or Harrod-neutral technical change) corresponds to a clockwise rotation of the efficiency schedule around its horizontal axis intercept. We measure the degree of labor-saving technical change by the percentage increase in labor productivity between two periods, which we denote by g = Hx+1 /x) - 1. A negative value for g indicates declining labor productivity. Technical change that raises capital productivity while keeping labor productivity constant (purely capitalsaving, or capital augmenting technical change) corresponds to a counter-clockwise rotation of the efficiency schedule around its vertical axis intercept. In what turns out to be the common case of declines in capital productivity this type of technical change is called capital-using. We measure the degree of capital-saving technical change by the percentage increase in capital productivity between two periods, which we denote by c = Hr+1 /r) - 1. In many historical cases it turns out that c < 0, corresponding to capital-using technical change. If c = 0, we have the case of Harrod-neutral technical change, and if c = g the efficiency schedule shifts out parallel to itself, which is the case of Hicks-neutral technical change. Any arbitrary pattern of technical change can be decomposed into a combination of labor-saving and capitalsaving changes, corresponding to the pair (c,g), representing the degrees of capital-saving and labor-saving change experienced, and the corresponding movements of the horizontal and vertical intercepts of the efficiency schedule. The pattern of technical change can also be observed by looking directly at the pair (c,g).

Foley-Marquetti

Economic Growth

The Classical/Marxian theory of the falling rate of profit Adam Smith, following a long tradition before him, identified a falling rate of profit with the accumulation of capital and the growth of an economy as a pervasive tendency of economic growth. David Ricardo explained this tendency in terms of diminishing returns due to the scarcity of natural resources (for example, fertile agricultural land and easily mined mineral deposits), leading to a decline in labor productivity with capital accumulation and population growth, a rise in rents, and a fall in the rate of profit. Ricardo recognized that technical changes which economized on scarce natural resources could temporarily raise labor productivity and the rate of profit, but foresaw the eventual cessation of capital accumulation as a result of rents rising, and the profit rate falling to zero. Karl Marx criticized Ricardo's explanation of the tendency of the rate of profit to fall on the grounds that it ignored the powerful incentives to technical progress inherent in the capitalist mode of production. Marx saw capitalist economies as systematically generating technical change to overcome diminishing returns to scarce factors of production, and rejected Ricardo's explanation of the falling rate of profit as due to declining labor productivity and rising rents to scarce resources. Marx argued that the tendency for the rate of profit to fall identified and accepted by Smith, Ricardo, and the long tradition of political economic writing that preceded them had to be explained in conjunction with rising labor productivity due to induced technical change. Marx argued that individual capitalists would seek out and adopt technical changes that lowered costs of production at current levels of real wages (called viable technical changes in modern theory), in order to reap "super-profits" by continuing to sell their output at prices determined by the higher costs of their less technically advanced competitors. Marx saw this process as a powerful engine of technical revolution of capitalist production. If real wages rise in parity with rises in labor productivity (which has been the actual historical experience of capitalist economies, and corresponds to a constant wage share in national income, or to a constant value of laborpower in Marx's terms (see Foley, 1986)), the mechanization process can generate a falling rate of profit. (As Okishio, 1961, points out, viable technical changes cannot lower the rate of profit without some increase in the real wage.) Marx summed up this vision of the long-term development of the capitalist mode of production in his theories of relative surplus value and falling rate of profit. Putting these ideas in modern terms, Marx saw a systematic bias toward labor-saving and capital-using technical change as the typical pattern of capitalist development. In contrast to Harrod-neutral technical change, which is labor saving but neither capital-saving or capital using and corresponds to a clockwise rotation of the efficiency schedule around its horizontal axis intercept, and Hicks-neutral technical change, which is equally labor and capital saving and corresponds to a parralel movement of the efficiency schedule, what we might call Marx-biased technical change is labor-saving and capital-using, and corresponds to a clockwise rotation of the efficiency schedule around a point in the positive quadrant. Thus Marx-biased technical change has g > 0 and c < 0. The point of intersection of the new and old efficiency schedules represents a real wage at which the two are equally profitable, and is called the switchpoint. From this point of view, the rate of profit in a capitalist economy is determined by two factors: the pattern of technical change and the evolution of the wage share in income. A combination of Marx-biased technical change with a constant (or slowly falling) wage share in income can (but need not necessarily) lead to a falling rate of profit. In this paper we consider the pattern of technical change, without examining the evidence on the wage share in income or the rate of profit.

5

6

Economic Growth

Foley-Marquetti

output per worker x+1 x

w+1 w

% per year

r+1 r

r+d r+1 +d

Figure 3: A falling rate of profit with Marx-biased technical progress and a constant wage share in income.

Long-run patterns of growth in the U.S. Gérard Duménil and Dominique Lévy (1993) have compiled aggregate statistics for the U.S. economy from 1869 to 1992, which they have kindly made available to us. In particular these statistics include careful estimates of the real value of gross output, employed labor, and the real value of the capital stock which allow us to plot efficiency schedules for the U.S. economy for these years, and to visualize the long-run tendencies of technical change in the U.S. economy. Figures 4 and 5 show rather typical 15-yr shifts in the U.S. efficiency schedule, estimated from the DuménilLévy data.

output per worker 30

U.S. 1970-1985

25 1985 20 15 10

1970

5 0.2

0.4

0.6

0.8

1

% per year

Figure 4: U.S. efficiency schedule shift from 1970 to 1985 (Duménil-Lévy data).

Foley-Marquetti

Economic Growth

7

output per worker 5

U.S. 1870-1885

4 1885 3 2 1870 1 0.2

0.4

0.6

0.8

1

% per year

Figure 5: U.S. efficiency schedule shift from 1870 to 1885 (Duménil-Lévy data).

Despite the century that separates the two periods, the qualitative pattern of change is remarkably consistent, and reflects the labor-saving and capital-using Marx bias. A comparison of the two Figures as a whole, however, reveals that both labor productivity and capital productivity rose substantially between the two periods. Labor productivity increased almost 10-fold, while capital productivity rose on the order of 25%. The rise in capital productivity over the century we have just examined is shows that the Marx-bias pattern is not uniform in the U.S. data. Indeed, if we look at the period 1929-1949, as in Figure 6, we see a strikingly different pattern of technical change that appears to have augmented the productivity of both factors almost equally (that is, to be essentially Hicks-neutral.)

output per worker 20 17.5 15 1949 12.5 10 7.5 5 1929 2.5 0.2

U.S. 1929-1949

0.4

0.6

0.8

1

% per year

Figure 6: U.S. efficiency schedule shift from 1929 to 1945 (Duméenil-Lévy data).

As Duménil and Lévy argue, using somewhat different methods of analysis, the period around 1919-1949 in the U.S. economy was a sharply anomalous period in terms of the patterns of technical change accompanying capital accumulation. The Marx bias that characterizes both the previous and succeeding periods in the U.S. gives way to a period of uniform factor-augmenting technical change that increases both labor and capital productivity. This finding supports the view that the 1919-1949 period was a watershed in U.S. economic history, a period during which there were deep and lasting structural changes in the organization and technique of production.

8

Economic Growth

Foley-Marquetti

Figure 7 shows the efficiency schedules at 5 year intervals for the period 1949-1989, and confirms the view that the Marx-biased pattern returned.

output per

U.S. 1949-1989

worker 30

5 year intervals

25 20 15 10 5 0.2

0.4

0.6

0.8

1

% per year

Figure 7: U.S. efficiency schedules at 5 year intervals from 1949-1989 (Duménil-Lévy data).

The efficiency schedule in Brazil, 1959-1990 We can also use the Penn World Tables data together with our own reconstruction of national capital stocks to estimate efficiency schedules for a wide range of countries over the period 1959-1992 which the PWT covers. In some cases we have created our own estimates of missing capital stock data for particular countries and years in the PWT, using the method described in the Appendix. Before we turn to an analysis of the whole of this data set, we will examine the experience of one country, Brazil. Figure 8 shows the evolution of the efficiency schedule for Brazil from 1959 to 1979. This period exhibits the pattern of Marx-biased (labor-saving, capital-using) technical change we have seen arising already in the U.S. data.

output per worker

Brazil 1959-79

14000 12000 10000

1979

8000 6000 4000 1959 2000 0.2

0.4

0.6

0.8

1

1.2 % per year

Figure 8: Brazilian efficiency schedules 1959-1979 at 5-yr intervals (Penn World Table data: our reconstruction of national capital stock).

Foley-Marquetti

Economic Growth

9

This classic pattern was interrupted, however in the 1980s, as Figure 9 shows. In the 1980s the Marx-bias in technical change in Brazil gives way to fluctuations of a Hicks-neutral type, in which capital and labor productivity move together. Furthermore, as Figure 10 indicates, the strong upward movement of labor productivity in the period 1959-1979 changes in the 1980s to a pattern of fluctuation around a constant level. Clearly the 1980s in Brazil, like the 1929-1949 period in the U.S., represent a period of structural change in the economy. In the U.S. the 1929-49 period marked a substantial advance in both labor and capital productivity, setting the stage for a period of renewed Marx-biased technical change. In Brazil the 1980s marked a period of stagnation in both capital and labor productivity.

output per worker

Brazil 1980,83,88,91

14000 12000 10000 8000 6000 4000 2000 0.2

0.4

0.6

0.8

1

1.2 % per year

Figure 9: Brazilian efficiency schedules 1980-1990 (Penn World Table data: our reconstruction of national capital stock).

output per worker

Brazil

11000 10000 9000 8000 7000 1962

1967

1972

1977

1982

1987

year

Figure 10: The evolution of labor productivity in Brazil, 1959-1990 (Penn World Table data: our reconstruction of national capital stock).

10

Economic Growth

Foley-Marquetti

World patterns of economic growth The remarkable comprehensiveness of the Penn World Table data allows us to look for world patterns of technical change as well. If a labor-saving, capital-using Marx bias is typical of capitalist economic development, we would expect to see a strong downward-sloping relation between x and r over the whole world economy. Figure 11 plots (r,x) observations for all 126 countries and years for which data is available for any part of the period 1959-1990. The data is fitted using a robust non-parametric method (Cleveland, 1993) called robust loess. The loess technique calculates a weighted least-squares fit to the data at each point on a grid, with weights that decline sharply with the distance of the data point from the grid point. The loess fit is made robust by calculating robustness weights that decline sharply with the size of the residual for each data point from the loess fit, and then iterating the loess fit with these robustness weights.

x 50000

Full Sample

40000 30000 20000 10000

2

4

6

8

10

12

r

Figure 11: Full sample of (r,x) points, 1959-1990 (Penn World Table data: our reconstructions of national capital stock).

The existence of powerful pattern of negative correlation between capital productivity and labor productivity in the course of economic development is unmistakable in this data. There are some striking exceptions (represented by the sprinkling of data points to the northeast of the main cluster). It is equally clear that there are substantial variations in the exact paths that national economies follow in the course of economic development, as shown by the wide scattering of the points around the sharp turning point of the fitted curve. But the dramatic clustering of the points around the pattern of negative tradeoff and the sharp identification of the monotonic relation between r and x by the robust loess fit, leave little doubt that there is a powerful tendency for national economies to follow a path of declining capital productivity and rising labor productivity in the course of economic development. The explanation for this striking pattern lies beyond the scope of this paper, which is concerned with an exploration of the gross features capitalist economic development. We will briefly mention two hypotheses which have been put forward in the economics literature. The enormous literature on the neoclassical growth model (running from Solow, 1970 through Mankiw, Romer and Weil, 1992) attempts to interpret this pattern as arising at least partially from the existence of a stable production-function relationship between capital and labor inputs. There are serious quantitative anomalies in this line of explanation, which remains immensely influential. The much smaller literature putting forward a Classical/Marxian alternative to the neoclassical production function (Duménil and Lévy, 1995 is a leading example) suggests that these patterns are the results of biases in induced technical change, rather than movements along a stable production function isoquant.

Foley-Marquetti

Economic Growth

11

Patterns of technical change The evidence we have presented up to this point supports the hypothesis that capitalist economic development typically but not universally follows a pattern of combined labor-saving and capital-using technical change. In this section we look at data directly on rates of change in labor productivity and capital productivity, c and g. Figure 12 plots c and g measured over 5 year periods (that is, for each country and year we calculate the rate of change in r and x between that year and 5 years later) for the 126 countries in our sample. There is, as we would expect, a strong tendency for the points to cluster in the northeast quadrant, corresponding to negative c and positive g, though there is a scattering of points in all the quadrants. The loess curve fitted to the data, however, reveals no negative quantitative correlation between c and g. In the northeast quadrant the curve is basically flat, indicating no systematic correlation between the magnitude of changes in c and the corresponding period's changes in g. This seems inconsistent with the hypothesis of a stable production function isoquant along which national economies are moving, since such an isoquant would introduce some negative correlation between c and g.

Full Sample

g5 15 10 5

-30

-20

-10

10

c5

-5 -10 -15 Figure 12: Full sample (c,g) points, measured over 5 year intervals, 1959-1990 (Penn World Table data: our estimates of national capital stock).

The mean and standard deviation of the c and g data are presented in Table 1. The means confirm the general hypothesis of Marx-biased technical change, but the wide scattering of the points makes the standard deviations large relative to the means.

mean sd

c - 1.80931 3.37411

g 1.40838 3.02809

Table 1: Mean and standard deviation for full sample of (c,g), measured over 5 year intervals, 1959-1991 (Penn World Table data: our estimates of national capital stock).

A chi-square test of the distribution of the sign patterns of c and g based on Table 2 rejects the hypothesis of an equal likelihood of among the different patterns of technical change with a P (probability significance level) indistinguishable from 0.

12

Economic Growth

g > 0 g < 0

c < 0 0.52907 0.192202

Foley-Marquetti

c > 0 0.21238 0.0663475

Table 2: Frequency of sign patterns for full sample (c,g), measured over 5 year intervals, 1959-1990 (Penn World Table data: our estimates of national capital stock).

Conclusions The Classical political economists, Smith and Ricardo, identified a tendency for the rate of profit to fall with capital accumulation. Marx associated this tendency with a more fundamental bias in the patterns of technical change in capitalist societies, toward labor-saving and capital-using technologies, a pattern we have dubbed "Marx-biased" technical change. Long data series compiled for the United States economy by Duménil and Lévy show the Marx-bias pattern dominating the 1869-1919 and 1949-1989 periods in the U.S. economy, but with a watershed period of Hicksneutral technical progress in the intervening years. The Penn World Tables supplemented by our estimates of national capital stock also reveal a Marx-biased pattern of technical change for the Brazilian economy in the 1959-1980 period, followed by a stagnation of technical progress in Brazil in the 1980s. An exploration of the evidence for Marx-bias in the patterns of technical change in the Penn World Table data supplemented by our estimates of national capital stocks on national economic growth from 1959-1990 confirms the predominance of the Marx-biased pattern, but reveals a substantial minority of cases with other patterns of technical change. While the gross tradeoff of declines in capital productivity with increases in labor productivity is consistent with the neoclassical production function, the details of the statistics raise substantial quantitative problems for the production function view. These preliminary findings suggest a number of avenues for further research. It would be useful to categorize the "anomalous", non Marx-biased instances to understand which countries and years these represent, and to try to identify those political, social, and economic factors that might explain their anomalous status. It would also be useful to study the empirical dependency of the degree of labor-saving and capital-using technical change on other factors, such as the size of national economies and their degree of absolute development as measured by their labor productivity. Such studies could lead to a deeper understanding of the inward character of capitalist economic growth.

Acknowledgments This paper was prepared for the International Colloquium convened at the University of Brasília, April 2-4, 1997. We would like to thank Thomas R. Michl for extensive conversations on the efficiency schedule, and for access to unpublished papers on this subject.

Foley-Marquetti

Economic Growth

References Blades, Derek 1993. Comparing capital stocks. In Explaining economic growth: essays in honor of Angus Maddison, eds. Adam Szirmai, bar Van Ark, and Dirk Pilat. Amsterdam: North-Holland. Cleveland, William S. 1993. Visualizing Data. AT&T Bell Laboratories. Duménil, Gérard, and Dominique Lévy. 1993. The Economics of the Profit Rate. Edward Elgar. Duménil, Gérard, and Dominique Lévy. 1995. A Stochastic Model of Technical Change: An Application to the U.S. Economy (1869-1989). Metroeconomica 46 (October), 213-45. Foley, Duncan. 1986. Understanding Capital: Marx's Economic Theory. Harvard University Press. Mankiw, N. Gregory, David Romer and David N. Weil. A Contribution to the Empirics of Economic Growth. Quarterly Journal of Economics 152 (May), 407-37. Okishio, Nobuo. 1961.Technical Change and the Rate of Profit. Kobe University Economic Review 7, 86-99. Solow, Robert. 1970. Growth Theory. Oxford University Press. Sraffa, Piero. 1961. Production of Commodities by Means of Commodities. Cambridge University Press. Srinivasan, T. 1995. Long-run growth theories and empirics: anything new? In Growth theories in light of the East Asian experience, ed. Takatoshi Ito and Anne Krueger. Chicago: University of Chicago Press. Summers, Robert and Heston, Allen. 1991. The Penn World Table (Mark 5): An expanded set of international comparisons, 1950-1988. Quarterly Journal of Economics 106:327-68.

13

14

Economic Growth

Foley-Marquetti

Appendix on Data Sources and Methodology Appendix on Data Source and Methodology This appendix presents the data source and a brief description of the methodology used to calculate the series employed in the article. The data source utilized is the Penn World Table (Mark 5.6) ( PWT v. 5.6). The PWT v. 5.6 displays for 152 countries a basic set of national accounts, relative prices, and demographic data which allows the comparisons between countries and over time. For some countries the capital stock data is also reported for the period 1965-1992. For the list of variables and the exposition of the PWT methodology, see Summers and Heston (1991). The PWT v. 5.6 covers the 1950-1992 period for some countries and for others it starts after 1950 and/or finishes before 1992. In the present article all countries with the first observation after 1960 were eliminated. The result is a sample of 126 countries with the following distribution per continent: Africa, 48; America, 27; Asia, 22; Europe, 25; Oceania, 4. The real Gross Domestic Product utilized is the Chain Index expressed in 1985 purchasing power parity (PPP). It was obtained by multiplying the total population and the real GDP per capita. The labor productivity is presented in the PWT v. 5.6 measured as the real GDP per worker. However, this variable has its last observations in 1990. Thus, this is the last year covered in the present paper. The productivity of capital is calculated using our estimated capital stock data for 126 countries and our "benchmark data", the PWT v. 5.6 estimates. Our stock of capital is obtained by the Perpetual Inventory Method (PIM) using the investment series computed from the variable real investment share of GDP presented in the PWT v. 5.6. An asset life of ten years and, consequently, a depreciation rate of ten percent was employed. Depreciation is assumed to follow a straight-line basis. Furthermore, all the assets were considered to have the same life span. Initially, the investment data was properly accumulated to generate the first observation, then the capital stock was computed. It is the cumulated, depreciated sum of the past aggregate investment, while our benchmark is the "cumulated, depreciated sum of past gross domestic investment in producer durables, nonresidential construction, and other constructions" (Summers and Heston, 1991, p. 347) augmented by the stock of capital of the residential construction. Certain problems are inherent in this attempt to extend the PWT data. First, there is the problem of data quality on investment of the PWT table. Srinivasan (1995) calls the attention for this problem. Second, as the investment data is not presented by categories of gross fixed capital formation and they are reported for a short time, not only are all categories of gross capital formation assumed to have the same asset life, but also the asset life is very short. The PIM procedure understates the size of capital stock and its relative bias may change considerably for countries with very different composition in the gross fixed capital formation when the 'true' asset life is considered. In terms of the growth rate of the capital stock the effect of a short service life is to reduce it in an economy with rapid investment expansion, and vice versa. But, as Blades (1993, p. 404) remarks the "use of erroneous service lives does not introduce any systematic bias into capital stock growth rates". The estimation of the stock reduces the period of time coverage in the present work in relation to the PWT v. 5.6. The first observation for the countries analyzed extend from 1959 to 1969. Despite the limits of the information and the limits of our procedure, the output-capital ratio and its growth rate computed from our estimates are similar to those calculated with our benchmark PWT data when the PWT capital stock data is available. The average correlation coefficient between our estimation of capital stock and the PWT estimation is 0.97. First, for each country that the PWT v.5.6 reports a estimation of capital stock the correlation coefficient was computed, then their average was calculated. This indicates a very high linear association between the estimates.

Foley-Marquetti

Economic Growth

15