3.3.2Output and Labor Productivity Growth Contributions of Information Technology Capital

Since the productivity paradox was first identified, several empirical studies have attempted to quantify the contribution of IT capital (KIT) to output growth or labor productivity growth (Table 3.2). Even if productivity growth was sluggish during the 1970s, IT capital may still have contributed to its increase. The BLS framework is usually used to evaluate the contribution to growth of various inputs. This framework is derived from the general growth accounting techniques previously described. The equation used to calculate output growth contributions is:

gr(Y) = αKIT gr(KIT) + αKNIT gr(KNIT) + αL gr(L+q) + gr(TFP) (3.13)

where gr(.) represents growth rate, Y represents output, KIT is information technology capital, KNIT is other types of capital, L measures labor hours, q controls for labor quality and TFP is total factor productivity. The alpha coefficients (α) represent the income shares of the inputs, which under neoclassical assumptions are equal to output elasticities and sum up to one when constant returns to scale are assumed. Growth rates are expressed by gr. Thus, the output growth contribution of IT capital is measured by the product of its income share (αKIT) and the growth rate of its stock (grKIT).

The value of the income share is a crucial variable needed to evaluate the growth contribution of any input. Authors have either calculated it following the method used by BLS [Oliner and Sichel (1994, 2000), Jorgenson and Stiroh (1998, 2000)], or estimated it using regression techniques [(Brynjolfsson and Hitt (1993), Lichtenberg (1995), Lehr and Lichtenberg (1999), Gera,Gu and Lee (1999)].

Oliner and Sichel (1994) published one of the first well-recognized studies of the contribution of computer capital to output growth. Their measure of IT capital (KIT) is based on data on “Computer and Peripheral Equipment” (CPE) from the Bureau of Economic Analysis (BEA). CPE belongs to the broader category IPE (Figure 1.1). Over the period 1970-1992, the average yearly growth rate of computer equipment was estimated at 27.6%. To calculate the computer income share (αKIT), Oliner and Sichel followed the BLS formulation

α KIT = (r COMP + d - π KIT) (P KIT KIT)) / (P Y Y)(3.14)

where r COMP is the nominal rate of return common to all capital, d is the depreciation rate, π KIT is the rate of nominal computer capital loss, P Y Y is nominal output and P KIT KIT represents nominal net stock of computer capital. Using data from BEA and BLS, Oliner and Sichel found an income share for computer equipment of 0.6 percent on average over the 1970-1992 period. Therefore, the growth contribution from computing equipment is estimated around 0.16 (= 0.6*0.276) percentage points per year during the period 1970-1992, compared to a total growth rate of output of 2.77 percentage points. According to the authors, the small contribution to output growth may be due to the small income share of computer capital (0.6). Indeed, even if investment in computers has skyrocketed during the last two decades, this form of equipment still represents a relatively small share of total capital input. To answer Solow’s famous quip, Oliner and Sichel attempt to solve the productivity paradox arguing that computers are actually not seen “everywhere.”

Oliner and Sichel (1994) then considered three extensions of their study in order to explain the low contribution to growth from computers. First, computers may earn greater than competitive returns to investment for two reasons. On one hand, they might generate positive externalities as stated by Romer (1986, 1987) and De Long and Summers (1991,1992). Indeed, the “computer knowledge” that workers gain using this new technology might spread to other workers, generating positive externalities for the economy as a whole. On the other hand, computers might simply have higher private returns (even if there are no externality effects), as suggested by Brynjolfsson and Hitt (1993) and Lichtenberg (1993). Still, even when higher returns are assumed (values for r COMP up to 56%), the growth contribution of computer equipment remains small (increases from 0.16 up to 0.35 percentage points) according to Oliner and Sichel’s calculations.

Second, Oliner and Sichel (1994) measured the growth contribution of computers correcting for mismeasurement errors. In order to do so, they assumed that one dollar of measured output from computer capital corresponds to one other dollar of unmeasured intangible output. This is equivalent to assuming a return of 50% to computer capital, and as seen earlier, this would not change significantly the growth contribution of computer equipment.

Finally, the authors explored the effects of considering not only computer equipment, but also other communication devices. They found that information processing equipment (IPE) contributed 0.31 percentage point annually to output growth over the period 1970-1992. Thus, even if computers represent a small share of the stock of information processing equipment (1/6), they still account for most of its growth contribution (about 50%).

Using essentially the same framework, Oliner and Sichel (2000) found a greater contribution of computers to output growth for several reasons. First of all, the stock of computer equipment boomed during the 1990s and is now relatively much more important than a decade ago. Furthermore, Oliner and Sichel have enlarged their definition of computer capital, which now includes software and communication equipment. The return to computer capital is also suspected of having increased this last decade. Thus, the income share of the computer capital equipment was 5.3% and 6.3% respectively for the periods 1991-1995 and 1996-1999. For those periods, the contributions to output growth of information technology capital are respectively 0.54 and 1.08 percentage points. The growth contribution of information processing equipment doubled between the 1996-1999 and 1974-1995 periods. The contribution of different inputs to growth in labor productivity is calculated simply by subtracting the growth rate of total hours from both sides of equation (3.13), yielding:

gr(Y/L) = αKIT gr(KIT/L) + αKNIT gr(KNIT/L) + αL gr(q) + gr(TFP) (3.15)

In this framework, the growth in labor productivity is decomposed into capital deepening, gr(K/L), change in labor quality, gr(q) and change in TFP, gr(TFP).

Oliner and Sichel reported a substantial increase of 1.05 percentage point in labor productivity between the first and second half of the 1990s. This increase was mostly due to due growth in TFP (+0.68) and capital deepening (+0.50), with a negative contribution of labor quality (-0.13). The authors also compared the growth contribution from the use and from the production of information technology, which is embedded in the growth of TFP. They divided the nonfarm business sector into three sectors: a sector s produces semiconductors for the computer manufacturing sector (sector c) and all other industries (sector o). Using a framework based on the work of Hulten and Schwab (1984), Triplett (1999), Stiroh (1998) and Whelan (1999), they showed that MFP growth could be decomposed according to the following equation:

gr(MFP) = μc gr(MFP)c + μo gr(MFP)o + μs gr(MFP)s(3.16)

where parameters μc and μo represent the shares of output for each of these sectors, and μs is the value of semiconductors used by others. The output of the computer sector is estimated using the sum of computer spending by U.S. business, households and all level of government, plus net exports of computers, published by BEA. The semiconductor output is estimated using data from the Federal Reserve Board.

Finally, sectoral MFP growth rates are estimated using the “dual” method from Triplett (1999) and Whelan (1999). Oliner and Sichel’s results indicate that the 1.05 percentage point gain in labor productivity between the first and second half of the 1990s is decomposed into: (1) 0.46 point from the growth in information technology capital per hour – capital deepening, (2) 0.04 from other capital deepening, (3) – 0.13 point from labor quality decline, (4) 0.26 from MFP growth in the computer-producing sector, (5) 0.11 from MFP growth in semiconductor producing sector and (6) 0.32 point MFP growth in all other industries. Jorgenson and Stiroh (1999) used a framework based on the work of Christensen and Jorgenson (1973). They attempted to quantify the output contribution of IT equipment as both an input used by firms to produce as well as a consumption good for households. They started with a production function of the form:

g(I,C,S) = f(K,D,L,T)(3.17)

where I represents investment goods, C consumption goods and services, S flow of services from consumers’ durable goods, K inputs of capital services, D consumers’ durable services, L labor input and T technology. Then, the distinction is made between computer (c) and non-computer (n) portions of those inputs and outputs:

g (Ic, In, Cc, Cn, Sc, Sn) = f (Kc, Kn, Dc, Dn, L, T)(3.18)

Measuring the growth contribution shows that computer investment goods (Ic) made the largest contribution with 0.26 percentage points during the 1990-1996 period. Computer equipment and services (Cc) made a contribution of 0.13 percentage points. Taken together, computer inputs contributed 0.16 percentage points to output growth of 2.4% per year for the period 1990-1996, and are directly due to substitution toward IT equipment. Jorgenson and Stiroh (1999) concluded:

‘The resolution of the Solow paradox is that computer-related gains, large returns to the production and use of computers, and network effects are fundamentally changing the U.S. economy. However, they are not ushering in a period of faster growth of output and total factor productivity. Rather, returns to investment in IT equipment have been successfully internalized by computer producers and computer users.’

Using a similar approach, Jorgenson and Stiroh (2000) found results slightly different than Oliner and Sichel (2000). Using recent data from BEA (1999), they considered information technology as investment in computers, software and communication equipment, as well as consumption of computers and software as outputs. IT is again considered as both an input and an output. Assuming constant returns to scale and competitive product and factor markets, the model starts from a Hicks neutral production function of the form:

Y (I, C) = A f (K, L)(3.19)

where I and C represent investment and consumption goods respectively, and will be decomposed into sub-components. K and L stand for capital services and labor inputs respectively, also decomposed into sub-components. The share-weighted growth of outputs is then expressed as

wiΔln(I) + wcΔln(C) = vkΔln(K) + vlΔln(L) + Δln(A)(3.20)

where w and v represent the shares of nominal output and income respectively. Therefore, it is possible to estimate the growth contributions of different inputs as well as outputs (computers, software and communication equipment distinctively).

Considering average labor productivity (ALP) as the ratio of output (Y) to hours worked (H), and the ratio of capital services to hours (k), labor input (L), the growth in average labor productivity is then expressed as:

Δln(ALP) = vk Δln(k) + vl Δln( L) - Δln(H) + Δln(A)(3.21)

Hence, average labor productivity growth is a function of capital deepening (k), labor quality (L-H), and TFP (A).

Jorgenson and Stiroh (2000) then considered the computer-producing and computer-using sectors separately. They argued that on one hand rapid technical progress in the computer-producing sector will increase TFP and therefore labor productivity at the aggregate level. On the other hand, computer equipment accumulation in the computer-using sector will only increase aggregate labor productivity through capital deepening according to equation 3.21. It will not affect aggregate TFP. The usefulness of this framework also lies in the consideration of substitution between outputs and between inputs. First, the distinction is made between capital services and capital stock. Capital stocks are measured using the perpetual inventory method on investment series from BEA, and aggregated using rental prices as weights. Jorgenson and Stiroh also identify rental prices with marginal products of different types of capital. Those prices incorporate differences in asset prices, service lives and depreciation rates, and the tax treatment of capital incomes. The difference between growth of capital services and capital stocks reflects the growth in capital quality, which represents the substitution towards assets with higher marginal products. Using this methodology, the authors found that information technology equipment had an output growth contribution of 0.17 and 0.36 percentage points for the periods 1973-1990 and 1996-1998 respectively.

Jorgenson and Stiroh (2000) also used an original framework to evaluate the contribution of individual industries to aggregate TFP growth. They argued that

‘Aggregate TFP gains – the ability to produce more output from the same inputs - reflects the evolution of the production structure at the plant or firm level in response to technological changes, managerial choices, and economic shocks. These firm- and industry- level changes then cumulate to determine aggregate TFP growth.’

Output is considered as “gross output” and therefore inputs include capital (K), labor (L), energy (E) and materials (M) such as:

Qi = Ai. Xi (Ki,Li,Ei,Mi)(3.22)

The growth accounting equation becomes:

Δln(Qi) = Δln(Ai) + wk Δln(Ki) + wl Δln(Li) + we Δln(Ei) + wm Δln(Mi)(3.23)

where w represents the average share of the subscripted input in the ith industry, and Δln(Ai) represents industry productivity based on “gross output” measures. This productivity is often referred to as multifactor productivity (MFP) and is analogous to TFP, which is based on a value-added concept.

Following Domar (1961), Jorgenson and Stiroh (2000) then decomposed aggregate TFP as a weighted average of industry productivity:

Δln(A) = Σi=1...37 wi. Δln(Ai) (3.24)

where PiQi is current dollar gross output in sector i, PyY is current dollar aggregate value-added and wi is the “Domar weight.” Data come from BLS and BEA, for 37 industries. The Domar weight can be expressed as:

wi = ½ * [(PitQit / Py, tYt) + (Pit-1Qit-1 / Py, t-1Yt-1)] (3.25)

Jorgenson and Stiroh (2000) found that productivity (TFP) has increased in both sectors, but it seems that this increase was not due to investment in IT for the IT-using sector. Indeed, more investment is correlated with lower TFP growth.

Whelan (1999) used a slightly different approach for measuring the output growth contribution of IT capital. Instead of using the Solow vintage model to construct capital stocks, Whelan stressed the importance of using “productive” instead of “wealth” stocks. The productive stock accounts for “technical obsolescence,” which occurs when computers are retired while they still retain productive capacity. Whelan found output growth contribution of IT capital of 0.39, 0.33 and 0.82 percentage points during the 1980-89, 1990-1995 and 1996-1998 periods respectively. Following Oliner and Sichel (2000) and Jorgenson and Stiroh (2000), the productivity growth contribution of IT capital is decomposed between the computer-producing and computer-using sector. According to Whelan, TFP growth in the computer-producing sector and capital deepening in the computer-using sector account for almost all of the recent increase in labor productivity during the 1996-1998 period (valued at 2.2% per year).

Kiley (1999) augmented the traditional growth accounting framework by including a common specification of investment adjustment costs. Earlier, Morrison and Berndt (1992) also included these costs in their analysis. Investment adjustment costs require that increases in investment lower the productive capacity of the firm and the economy. Typically, in a steady state framework, investments are small relative to the capital stock and do not influence the firm’s productive capacity. However, according to Kiley, the recent massive increase in IT capital investment has made investment adjustment costs more important and the Solow growth accounting framework becomes inappropriate when studying the productivity effects of IT capital. Using this augmented framework and data on computer capital stock from BLS, Kiley estimates that the investment adjustment costs lower MFP growth by 0.50 percentage points since 1974. These adjustment costs include “costs of reorganizing plant layouts to incorporate new machinery, managerial costs stemming from alterations to production plans consistent with the installation of new capital, and other costs associated with the interruption of normal work activity to install (or disinstall) capital.” Thus, he found a negative output growth contribution of computer capital of –0.34 and –0.27 annual percentage points during the 1974-84 and 1985-1998 periods respectively.

Adopting a firm-level analysis, Brynjolfsson and Hitt (1993) found a more optimistic contribution of computers to growth. With data on 367 business units for five years (1988-1992), the authors estimated a yearly output growth contribution of 1% for IT capital. Lau and Tokusu (1992) found an even greater contribution of 1.5%. On the other hand, studying 60 business units, Loveman (1994) found the output growth contribution of IT capital to be not significantly different from zero. His results were robust to numerous variations in the formulation of the basic framework. Gordon (1999) has a different view about the productivity paradox, when he argued:

‘There has been no productivity growth acceleration in the 99 percent of the economy located outside the sector which manufactures computer hardware, beyond that which can be explained by price remeasurement and by a normal (and modest) procyclical response.’

Thus, the problem of the productivity paradox remains because authors have not unanimously dismissed it yet. Even after the increase in productivity observed in the last 4 years, researchers still need to understand the reasons for the productivity paradox for information technology capital in prior years. This is what I intend to do in this dissertation, using a model that I describe in the next chapter.