The Great Reversal

by Thomas Philippon

  FIGURE 2.3  Retail price index relative to consumer price index. Data sources: BEA, GDP by Industry; FRED, PCE index

  The growth of Walmart led to concentration in the retail industry. Was this good news for the consumer? Figure 2.3 shows the evolution of the price of retail trade services relative to the general consumer price index. If buying goods in a local supermarket becomes cheaper, then this index drops. There is a remarkable decline in the price of retail services over a period of twenty years, from the mid-1980s to the mid-2000s, that coincides with the expansion of Walmart. The decline means that US households saved about 30 percent on their retail shopping costs.

  It would be tempting to jump to the conclusion that this trend represents a clear improvement for the US economy. But if there is one lesson that economics teaches us, it is that there is always some confounding factor. In this case, it is the decrease in the federal minimum wage during the 1980s. A significant fraction of retail workers earn wages at or below the local minimum wage (up to a quarter in grocery stores, for instance) and the retail sector is the second largest employer of minimum wage workers in the US. As a result, when the minimum wage decreases, we can expect retail prices to decline. Economists spend time and effort measuring precisely these effects. Tobias Renkin, Claire Montialoux, and Michael Siegenthaler find in a 2017 paper that a 10 percent change in the minimum wage changes retail prices by about 0.2 percent to 0.3 percent. The real (that is, inflation-adjusted) minimum wage decreased by about one-third between 1979 and 1995. This would predict a decrease of only 1 percent in retail prices, which is small compared to the relative price decline that we observe in Figure 2.3.

  What accounts for the bulk of the decline in retail prices in this period? The retail industry had clearly become more efficient, and the cost savings were passed on to consumers. Walmart’s advanced supply chain management system was a key contributor to this evolution. Through its vendor-managed inventory system, manufacturers are responsible for managing their own inventory in Walmart warehouses. Vendors can directly monitor the inventory of their goods in Walmart stores and send additional items when the stocks are low in a particular store. This technology lowers the cost of inventory management, and the efficiency gains are passed on to consumers in the form of lower prices. Economists Ali Hortaçsu and Chad Syverson argue in a 2015 paper that superstores and e-commerce have increased productivity in the retail industry.

  The growth of Walmart provides us with an example of efficient concentration. Its profit margins remain stable or even decline, and most important, prices go down. Consumers benefit from Walmart’s expansion. It is fair to debate and challenge Walmart’s labor and management practices, but there is little doubt that Walmart has been good for US consumers.

  As I write these lines, Sears has filed for bankruptcy, showing that the US retail sector has remained competitive. Suzanne Kapner in the Wall Street Journal reported, “For much of the 20th century, Sears Holdings Corp. defined American retailing with catalogs and department stores that brought toys, tools and appliances to millions of homes.” When Sears filed for bankruptcy in mid-October 2018, it had 687 stores and about 68,000 workers. “Decades earlier, it had been dethroned by Walmart Inc. as the biggest U.S. retailer. Then it was crippled by a chief executive with unorthodox strategies, and Amazon.com Inc., an endless online catalog that sucked profits out of the business.”b

  The US retail industry shows that concentration alone is not a reliable indicator of competition. It needs to be complemented by other measures, such as profits and prices. Later on, we will also look at hiring and investment.

  Measuring Concentration with Several Large Firms

  How do we assess the concentration of an industry when there are several large firms? In Figure 2.2 we consider only the market share of Walmart, and it’s easy because we draw one line. But in an industry dominated by several large firms, this would not work. We could plot all their market shares, but that would be a rather messy figure. It would be nicer to summarize concentration with one number, even when there are many firms. This is what the Herfindahl-Hirschman index (HHI) does (see Box 2.1).

  Let’s take another look at the airline industry. Figure 2.4 shows the national concentration index (HHI). It decreases with entry of new airlines in the 1980s and increases with mergers in the 2000s. It has historically been around 0.1, but increased to 0.14 following a wave of mergers in the 2000s. The top four domestic airlines are American (18.6 percent of the domestic market between July 2016 and June 2017), Southwest (18.4 percent), Delta (16.8 percent), and United (14.8 percent). The fifth, JetBlue (5.5 percent), is a lot smaller.

  Figure 2.4 shows US airlines’ HHI at the national level. But is this the relevant market to consider? People fly from one city to another; they do not fly between averages of various cities. In the airline industry the natural definition of a market is a route between two cities. Since airlines fly only a subset of all routes, concentration at the route level is higher than at the national level. But how much higher?

  A 2014 report by the Government Accountability Office (GAO) splits routes into quintiles from high to low traffic, using data from 2012. The GAO study took place after several high-profile mergers but before the merger of American and US Airways. There were about 410 million passengers in 2012, so each quintile has about 82 million passengers. The first quintile (high traffic) includes only thirty-seven city pairs, but they all correspond to heavily traveled routes between busy airports, such as New York to Los Angeles or Washington, DC, to Boston. The third quintile (medium traffic) has 237 city pairs. The fifth quintile (low traffic) includes 9,379 city pairs and many tiny airports. HHI in the first quintile was around 0.32. HHI in the third quintile was around 0.40. If we compare it to the national HHI in Figure 2.4, which was around 0.11 in 2012, we see that the relevant HHI, at the local level, is about three times higher than the national one.

  Box 2.1. Measuring Concentration with the Herfindahl-Hirschman Index

  The Herfindahl-Hirschman index (HHI) is a measure of market concentration. Imagine an industry with N firms. The market share of the first firm is s1, of the second firm s2. HHI is then computed as the sum of the squared market shares.

  HHI = (s1)2 + (s2)2 + … + (sN)2

  Why squared? The simple sum of market shares would always be one, by definition, so that would not be informative. If all the firms have the same market share s, then you can see that HHI = s. An industry with 10 identically sized firms has an HHI of 0.1. More generally, 1 / s would be the number of firms and 1 / s × s2 = s, so the HHI is always s when the firms are identical. When the firms are different, squaring the market shares means that we put more weight on the larger firms, which makes sense since we are interested in market power.

  Let us consider an example. First, a monopoly (one firm with s = 1) has an HHI of 1. That is the maximum value. An industry with one large firm controlling one half of the market (s = 0.5) and a bunch of tiny firms producing the rest would have an HHI close to 0.25. This would be the same HHI as with 4 identical firms. HHI therefore allows us to compare industries with different configurations.

  If you read legal documents, you will see an HHI of 0.25 written as HHI = 2,500. That’s just because we often measure HHI in basis points by multiplying the natural value (0.25) by 10,000. So instead of saying HHI = 1, we often say HHI = 10,000.

  For purposes of antitrust enforcement, the US Department of Justice defines a competitive market as one with an HHI score of 1,500 or lower. When the HHI is between 1,500 and 2,500, the market is classified as moderately concentrated. Above 2,500, the DoJ considers a market highly concentrated. Naturally, antitrust concerns are greatest in highly concentrated markets, and the DoJ sees any merger that increases the HHI in a highly concentrated market by more than 200 points to be a potential violation of antitrust regulations. (Now you see the convenience of the basis points scale: it’s easier to say 200 points than 0.02.)

  FIGURE 2.4  HHI in US air transport
industry. Data source: US firms in Compustat

  As explained in Box 2.2, this means that we need to be careful when we interpret national concentration indexes. This is an old controversy in the branch of economics that studies industrial organizations. Carl Shapiro (2018) is unconvinced by industry-level evidence: “it is extremely difficult to measure market concentration across the entire economy in a systematic manner that is both consistent and meaningful.” The problem of estimating HHI at the correct level of granularity is a difficult one. As we saw with the airline industry, Shapiro argues that many industry sectors that economists treat as participants in a national marketplace are often local (think of restaurants, supermarkets, wired telecommunications, and hospitals). Shapiro argues that the rise in concentration might simply “reflect the fact that large, national firms have captured an increasing share of overall revenue during the past 20 years.”

  Box 2.2. What Is the Relevant Market?

  If we consider the US retail sector, we see concentration rising from 1995 to 2008. At the national level, HHI increases from 0.03 to 0.06. As far as competition is concerned, however, the US retail sector is not one market. It is made of many local markets that correspond to the places where people actually go shopping. Clearly, the concentration of retail stores in downtown Chicago is irrelevant to the shopping experience of households in Tampa.

  Imagine a country with two regions and four independent retailers, two in each region, with equal market shares. Imagine also that people shop only in the region where they live. Clearly, then, the region is the right market to consider. Each region has two equally sized retailers, so each has an HHI of 0.5. The national HHI is 0.25, however, accounting for all four equally sized retailers. This national measure is not meaningful, and the correct HHI is 0.5, the regional one. Now suppose that one retailer of the first region merges with one in the second region. Local competition does not change because the merger happens between firms that did not compete in the first place. Each region still has an HHI of 0.5. The national HHI, however, increases to 0.375 since there is now one firm with a share of 0.5 and two firms with shares of 0.25. Looking at the national index, we could wrongfully conclude that market concentration has increased.

  That’s a fair criticism. Antitrust economists need to work with extremely granular data because they need to overcome a difficult burden of proof. We should take their concerns seriously. On the other hand, we cannot simply stop because we do not have a perfect measure. My approach in this book will be never to draw conclusions from only one measure. If we find signs of rising concentration, we will immediately look at other indicators, such as profits and prices. There are additional criticisms of concentration measures that we will discuss later in the book, notably when we consider foreign competition.

  For instance, when we look at Figure 2.4, we see a sharp increase in concentration in the airline industry after 2010. That is enough to trigger our interest, but not enough to conclude that competition has weakened. We must first check that concentration has also increased at the route level. We find that it has. We can further show that it came together with higher prices and higher profits. We can then conclude that this concentration was probably bad for US passengers. Even that is not entirely obvious, however, because—in theory—the quality of services could have improved so much as to justify the price increase. Most readers will laugh at this idea in the case of airlines (I am also laughing as I write these lines), but there are other industries where this could happen. Another explanation could be that safety regulations have increased fixed operating costs. That could also explain a rise in concentration. In that case, however, we can look at what happened in Europe, where safety regulations are just as tight, but there we do not see higher concentration or higher prices. In short, the case of US airlines is a rather straightforward example of weak antitrust with negative consequences for consumers, and this discussion might sound overly cautious. As a rule, however, it is healthy to consider alternative explanations.

  Amazon vs. Walmart

  Let us now return to the retail sector and focus on the growth of Amazon. We have seen that, as Walmart’s market share increased, retail prices decreased sharply. But the improvement stops in 2005. This coincides with the development of online shopping, and in particular with the growth of Amazon. When we think about the retail industry now, we still think of Walmart, Home Depot, and Target, but we also think of Amazon.

  The growth of Amazon coincides with constant prices. This means that Amazon’s expansion is not about cutting prices. Instead, Amazon is all about improving the shopping experience. This can lead to higher or lower effective prices, so how can we compare prices in the world of Amazon?

  Suppose you want to buy a tool for $50. You can drive to your local hardware store and get some expert advice before buying. You could buy similar tools for $45 on Amazon, but you are not sure which one would be best for you. If you are indifferent between these two options, it means that you value the expertise of the sales people in your local store at $5. Equivalently, the store is not selling just the tool, it is selling a bundle of tool and expert advice. The price on Amazon is not really cheaper because Amazon is not selling the bundle.

  It can also go the other way. If you prefer to buy a $20 toy from Amazon instead buying it for $17 in a brick-and-mortar store, you could ask yourself: how low would the price have to be at the store to convince me to drive there? Let’s say it would have to be $15. That means that you value the convenience and the time that you save at $5. In that sense, Amazon effectively saves you $5, and it’s equivalent to being able to buy the toy in the store for $15. Since the store sells it at $17, one could argue that Amazon is in fact $2 cheaper than the brick-and-mortar store.

  These examples illustrate a complicated issue in the measurement of prices. It is difficult to compare prices because goods vary in quality and come with add-ons. Box 2.3 explains how statistical agencies construct quality-adjusted prices.

  Digital platforms create new challenges in measuring prices because they often provide some services for free (for example, Google Maps). Of course, these services are not really free; they are simply part of a new bundle, where the most valuable component is the acquisition of personal data. As they say in Silicon Valley, “If you are not paying for it, you’re not the customer; you’re the product being sold.”

  If we adjust prices for the convenience of shopping online, it is likely that Amazon has also contributed to lower prices. There are differences, however, in how these gains are shared. Walmart created more value for lower-income consumers. Amazon is more valuable for upper middle-class households whose disposable income and opportunity cost of time are relatively high. In our previous example, the quality adjustment would be the amount of time saved multiplied by the hourly wage. A team of economists has recently quantified the gains from e-commerce (Dolfen et al., 2019). E-commerce spending reached 8 percent of consumption by 2017. The researchers estimate that gains from e-commerce are equivalent to a 1 percent permanent boost to consumption, which represents about $1,000 per household. Some of the gains arose from saving travel costs. Higher income cardholders gained more, as did consumers in more densely populated counties.

  Box 2.3. Adjusting Prices for Quality Changes

  Economists love prices. We love to estimate them, compare them, and bundle them into indexes.

  Why do we care so much about price indexes? Because they are supposed to tell us about the evolution of the cost of living. If both your income and your cost of living go up by 10 percent, you really have not experienced any improvement in your living standards, no real growth.

  Constructing accurate price indexes is a surprisingly important job. Any number of widely used (and expensive) government programs use price indexes as a guide to setting budgets and payment rates, from Social Security to Medicare to the per-mile reimbursement for employee travel. If we mess up the computation of price indexes, we draw the wrong conclusions and implement the wrong
policies.

  In 1996, the Boskin Commission concluded that the US Consumer Price Index (CPI) overestimated inflation by about 1 percentage point, notably because of quality changes and the introduction of new goods. Considering that the annualized rate of inflation at the time was 2.9 percent, this was a rather large mistake. Why is it so hard to construct price indexes?

  The CPI is constructed by the Bureau of Labor Statistics (BLS). The BLS collects prices every month on tens of thousands of goods and services in thousands of retail outlets. It then computes an “average” price change. This is relatively easy to do when the goods available this month are the same that were available last month. We can compute the percentage change in the price of each good and take an average, weighted by the share of spending on each good. But what happens when a new product appears? Or when an old one disappears? Or when an existing one changes significantly?

  The accurate measurement of quality changes is a difficult task. The idea behind the CPI is to compute over time the cost of buying the same basket of goods and services. If you paid $100 for a particular basket last month, and you would have to pay $101 for the same basket this month, then inflation is 1 percent. In reality, products disappear, products are replaced with new versions, new products emerge, and often you cannot buy exactly the same basket.