The Great Reversal Page 9
FIGURE 4.4 Concentration and investment gap. Annual data. We use the ten industries with the largest and smallest relative change in import-adjusted HHI indexes. The figure shows the cumulative implied capital gap (as percent of capital stock) for the corresponding industries (Gutiérrez and Philippon, 2017).
Figure 4.4 is inconsistent with the basic version of the Rise of Superstar Firms hypothesis. The historical evidence suggests that successful firms and industries maintain high levels of investment. If concentration was a sign of efficiency, then, we would expect to see more investment in concentrating industries. Figure 4.4 shows that we observe exactly the opposite across industries.
In my work with Germán Gutiérrez, we also uncover a negative relation between concentration and investment across firms. We find that industry leaders’ shares of investment and capital have decreased while their profit margins have increased. This is the opposite of what a hypothesis of superstar firms would predict. Under such a hypothesis, as leaders become more efficient, they should draw in more resources. Efficient firms typically expand by hiring more capital and more labor. In recent years, however, they have done the opposite. This is exactly what a decreasing competition hypothesis would predict. It is inconsistent with a hypothesis of superstar firms unless their investment and productivity are both badly mismeasured. Let us consider this possibility.
Intangible Investment
Our discussion so far assumes that we measure investment correctly, or at least that the quality of our measurement has not decreased over time. There are two types of investment: tangible and intangible. In Chapter 3 we noted the hypothesis of Nicolas Crouzet and Janice Eberly, that intangible investment might be partly responsible for the trends that we have discussed so far. Some firms might be really good at accumulating intangible assets. This might give them high profits and isolate them from competition.
Tangible investment is easily measured: more machines, more computers, more workers, more warehouses, more plants, more delivery vehicles. But measuring innovative investment is more difficult. Firms can invest in the development of new products and services or in the systematic improvement of existing ones. Some of these expenditures are captured under the headline “research and development.” Many are not.
When we go about measuring investment, we encounter the important distinction between tangible and intangible investment. A significant share of today’s capital is intangible. It includes patents, software, chemical formulas, databases, artistic value, special employee training, design, processes, and brand recognition. Intangible assets are not just about information technologies, however. Some intangible assets rely on computers—software and databases—but some are embedded in people, organizations, and brands. Intangible assets are also important in classic, “old-fashioned” manufacturing industries.
Economists are pretty good at measuring tangible investment. Tangible assets are usually purchased from another firm, as opposed to produced internally. If a firm needs a new truck, it buys it from a truck manufacturer; it does not build the truck itself. We therefore have a transaction price readily available to measure the cost of the investment. In addition, we have no difficulty in agreeing that it should be capitalized and not expensed. The purpose of the investment transaction is unambiguous: the truck is a long-lived valuable asset, so it falls clearly under the definition of investment.
Intangible investment, on the other hand, is harder to measure. Consider software. If a firm buys a piece of software, the transaction is similar to the truck example. It’s clearly an investment. But if the firm pays its own employees to write the software, then it is formally a labor expense. Statistical agencies are aware of this issue, so they survey firms and construct measures of in-house software investment based on the wage bill of in-house programmers. You can easily imagine why this is not as reliable a measure when compared to an outside purchase.
The impetus for improving our measure of intangible assets came in the early 2000s from a group of economists led by Carol Corrado, Daniel Sichel, Charles Hulten, and John Haltiwanger (2005). Broadly speaking, they divided intangible assets into three categories: computerized information, innovative property, and economic competencies. Computerized information that can be correctly captured as investment includes software and database development, but some database development costs may be missing from the official data. Research and development, patents, mineral exploration, and artistic IP—all falling under the rubric of innovative property—may be captured, but some design and other product development costs may escape official notice. And such economic competencies as employee training, market research, and business processes may not be captured by the data at all.
The quality of our measure of intangible investment varies widely across categories. In the domain of economic competencies, for instance, we measure very little. It is easy to see why: these are mostly in-house efforts whose contributions to the creation of long-lived assets is ambiguous. Computerized information is usually correctly measured, but internal development is difficult to capture precisely.
The data in Table 4.1 and Figure 4.1 include both tangible and intangible investment. The decline in investment applies to the sum of the two types of investment. There has been an important shift in the composition of investment toward intangible assets and away from tangible ones. If we study the two categories separately, we see that both tangible and intangible investments have been weak in recent years, but the weakness is less severe for intangible investment.c
The data suggest that intangible investment rates have declined less than tangible investment rates since 2000. The BEA fixed asset tables show a decline in the growth rate of the intellectual property products (IPP) stock since 2000, as you can see in Figure 4.5. Between 1962 and 2000, the average growth rate of the stock of IPP capital was 6.2 percent. Between 2001 and 2016 it was only 3.9 percent. The corresponding figures for structures are 4.9 percent down to 2.9 percent, and for equipment 2.6 percent down to 0.9 percent.
Intangible investment has not declined as much as tangible investment. Moreover, researchers using broader definitions of intangible capital than the official BEA tables find even weaker declines. It is not likely, however, that this bias would overturn our results. One reason is that statistical agencies have been improving their measurement of intangible investment over time. We certainly measure IPP capital better now than twenty years ago. Moreover, if you think that government agencies are too conservative in their treatment of intangible investment, you can choose your own definitions at the firm level. Gutiérrez and I have done that, and even using the most aggressive measures, we still find an investment gap.d
FIGURE 4.5 Growth rate of intangible capital stock: intellectual property products
We have thus found support for the Intangible Assets hypothesis. There has been a shift toward intangible assets, and the investment gap is smaller for intangible investment than for tangible investment. The great boom of intangible investment, however, was during the late 1990s. In recent years, intangible investment has been weak—perhaps not as depressed as tangible investment but definitely not strong enough to pull the economy forward.
Productivity
Perhaps the most direct prediction of the Rise of Superstar Firms hypothesis is that concentration should be linked with strong productivity growth. According to this hypothesis, concentration reflects an efficient increase in the scale of operation. A key prediction, therefore, is that concentration leads to productivity gains at the industry level, as high productivity leaders expand. This has happened before. In Chapter 2 we discussed the example of retail trade during the 1990s. The retail trade industry became substantially more concentrated and more productive during that decade.
But the 1990s are long gone. Are rising superstar firms the main driver of concentration over the past twenty years, as hypothesized by David Autor, David Dorn, Lawrence Katz, Christina Patterson, and John Van Reenen (2017)? To test this idea, Matias
Covarrubias, Germán Gutiérrez, and I (2019) study the relationship between changes in concentration and changes in total factor productivity (TFP) across industries during the 1990s and 2000s. We use our trade-adjusted concentration measures to control for foreign competition and for exports.
Box 4.2 and its table summarize our results and discuss the interpretation of the various numbers in statistical models. We find that the relationship between concentration and productivity growth has changed over the past twenty years. During the 1990s (1989–1999) this relationship was positive. Industries with larger increases in concentration were also industries with larger productivity gains. This is no longer the case. In fact, between 2000 and 2015, we find a negative (but somewhat noisy) relationship between changes in concentration and changes in productivity.
This pattern holds for the whole economy as well as within the manufacturing sector, where we can use more granular data (NAICS level 6, a term explained in the Appendix section on industry classification). The relationship is positive and significant over the 1997–2002 period but not after. In fact, the relationship appears to be negative, albeit noisy, in the 2007–2012 period.
Box 4.2. Statistical Models
Table 4.2 presents the results of five regressions, that is, five statistical models. The right half of the table considers the whole economy; the left half focuses on the manufacturing sector.
TABLE 4.2
Regression Results
Productivity growth Years
(1)
(2)
(3)
(4)
(5)
Manufacturing
Whole economy
97–02
02–07
07–12
89–99
00–15
Census CR4 growth
0.13*
0.01
−0.13
[0.06]
[0.05]
[0.17]
Compustat CR4 growth
0.14*
−0.09
[0.06]
[0.07]
Data set & granularity
NAICS-6
KLEMS
Year fixed effects
Y
Y
Y
Y
Y
Observations
469
466
299
92
138
R2
0.03
0.00
0.02
0.07
0.09
Notes: Log changes in TFP and in top 4 concentration. Standard errors appear in brackets below the coefficients. 97–02 means that the sample spanned 1997–2002. See Covarrubias, Gutiérrez, and Philippon (2019) for details.
Let us look at the right side and explain all the numbers: (4) means it is the fourth model. It covers the whole economy over the period 1989–1999. The coefficient 0.14 means that, over this sample, a 1 percent increase in the market share of the top four firms is associated with an increase in productivity of 0.14 percent. The number below, in brackets, is the standard error, which measures the precision of our estimate. A standard error of 0.06 for a coefficient of 0.14 means that the effect could really be anywhere between 0.08 (0.14 − 0.06) and 0.20 (0.14 + 0.06). We put a star (*) next to the coefficient when it is more than twice the standard error to signify that we are pretty confident that the coefficient is meaningfully positive. In the jargon of empirical economics, we say that the coefficient is statistically different from zero. In column 2 you see a coefficient of 0.01 with a standard error of 0.05: this means that there is no statistical connection in that sample between concentration and productivity.
The bottom of the table offers more background information, such as the data used and the number of observations. The inclusion of year fixed effects (Y = yes) means that the regression controls for any common shock that would move all the industries in the same direction in any given year. This is important because the US economy was not (and is not) static over this period: there are booms and busts, a stock market bubble, a terrorist attack, a housing bubble, and a financial crisis. We thus want to make sure that our results are driven by the comparison of industries within the US. Finally, the R2 measures the goodness of fit of the model: 0.07 means that it captures about 7 percent of the changes we see in the data. Not surprisingly, there are many other factors that affect productivity growth beyond concentration, and it is also likely that there is quite a lot of measurement error in the data.
To summarize, the 1990s supported a hypothesis of superstar firms, but the 2000s rejected it. There is, however, one last issue to worry about: Are we measuring productivity correctly?
A popular explanation by techno-optimists for the slowdown of the economy (slowdown in productivity and investment) is that we mismeasure the free goods provided by firms like Google and Facebook, and the intangible investments that sustain them. The story rings true but the evidence suggests that it is a rather small effect. The best available research concludes that mismeasurement is unlikely to be the explanation for our current lackluster economic growth. As Chad Jones (2017) explains, “Byrne et al. (2016) and Syverson (2017) conclude that the slowdown is so large relative to the importance of the ‘free’ sector that mismeasurement is likely a small part of the explanation.”e I would add that nothing is free. Remember the Silicon Valley adage given in Chapter 2: “If you are not paying for it, you’re not the customer; you’re the product being sold.” We will dig deeper into the business models and growth contributions of Apple, Amazon, Google, Facebook, and Microsoft in Chapter 13. Nonetheless, the measurement of the digital economy is an active area of research, and we are likely to obtain better estimates in the near future. For instance, Erik Brynjolfsson and his co-authors (2019) have recently argued that properly accounting for Facebook’s free services could add between 5 and 10 basis points (0.05 to 0.1 percent) to our measure of growth for the US economy.
Weak Investment and Weak Productivity
The pattern of investment and productivity growth is inconsistent with the hypothesis of rising superstar firms, which holds efficiency gains to be the root cause of increasing concentration. If concentration were reflective of increasing efficiency, then we should see more productivity growth in places where concentration increases. We saw some of it during the 1990s, but the opposite happened during the 2000s. The evolution of productivity is consistent with the investment choices that firms make. Industry leaders’ shares of investment and capital have decreased, and their profit margins have increased. Given that leaders in concentrating industries do not feel the urge to invest and choose to increase their payouts to shareholders, it is hardly surprising that productivity growth is lackluster.
* * *
a Net investment. In a stationary environment (constant profits, constant wages, no growth), firms replace the depreciated capital each year. That would be between 5 and 10 percent, depending on the industry and the kind of capital. In Table 4.1, the depreciation rate in 2014 was 1,583 / 20,983 = 7.5 percent. Although important, this is not what we are interested in. We want to understand why and how firms grow. We therefore focus on net investment.
b Gutiérrez and Philippon (2017) test eight possible explanations for the weakness of investment, ranging from measurement errors to financial constraints, and find consistent support for only three: rising concentration in product markets (the hypothesis that domestic competition is decreasing); tightened governance and increased short-term thinking; and rising intangible capital (which itself is a complex explanation involving measurement problems, efficiency gains, and barriers to entry).
c See Haskel and Westlake (2017) for more on intangible assets.
d This is consistent with figure 5.6 in Haskel and Westlake (2017): “it turns out that the effect of including previously unmeasured intangibles is to raise the investment / GDP ratio, but not to greatly affect its trend.” They conclude that “the mismeasurement
of intangible investment does not explain most of the investment problem.”
e David Byrne, John G. Fernald, and Marshall B. Reinsdorf (2016) explain that “mismeasurement of IT-related goods and services also occurred before the slowdown and, on balance, there is no evidence that it has worsened.” Moreover, innovations—such as free internet services—cannot explain the business sector TFP slowdown.
CHAPTER 5
The Failure of Free Entry
WE HAVE SEEN in previous chapters that, since the late 1990s, most US industries have become more concentrated, industry leaders have become more entrenched, and leaders’ profit margins have increased.
But how did this concentration actually occur? Concentration comes from the equilibrium of two main forces: entry and growth of new firms versus exit and mergers of existing firms. The relationship between concentration and entry and exit can be easily explained.
We are going to study these two forces. In doing so, we will continue our analysis of antitrust, in particular of merger reviews. On the entry side, suppose n firms are created each year. On the exit side, suppose a fraction d of existing firms is destroyed each year. Now ask yourself: after the dust settles, how many firms should we observe in this economy? Or in the jargon of economics: what is the number of firms in steady state? The steady state is reached when there are as many firms entering as there are firms exiting in any given period (say, a year). Let’s call N the number of active firms in steady state. For entry and exit to cancel out, we must have entry (n) equals exit (d × N). Since balancing entry and exit requires n = d × N, the number of firms in steady state must be N = n / d.