Are we overestimating inflation (again?)

Twenty years ago, a group of experts – the “Boskin Commission” – concluded that the U.S. consumer price index (CPI) systematically overstated inflation by 0.8 to 1.6 percentage points each year. Taking these findings to heart, the Bureau of Labor Statistics (BLS) got to work reducing this bias, so that by the mid-2000s, experts felt it had fallen by as much as half a percentage point.

We bring this up because there is a concern that as a consequence of the way in which we measure information technology (IT), health care, digital content and the like, the degree to which conventional indices overestimate inflation may have risen.

Measurement of prices is central to how we understand the economy. We use prices to distinguish nominal and real quantities in order to measure economic growth. We also use prices to deflate nominal incomes to measure the evolution of living standards. And, in the belief that price stability is the foundation for strong, sustainable and balanced growth, central bankers obsess over whether inflation is rising or falling, too high or too low. In economies accounting for about two thirds of global GDP, central banks have specified a level or range of inflation as a policy target.

When indices like the consumer price index (CPI) or the personal consumption expenditure price index (PCE) persistently overstate inflation, there are important consequences. So long as the upward bias is constant, central bankers can (and do) modify their inflation targets. Yet, these price indexes also are used to adjust entitlement benefits without correcting for any persistent bias. And, they can have an important impact on public discourse. In particular, upward bias means that the median real wage may have risen substantially over past decades, in contrast to reported stagnation.

If the overstatement of inflation has increased during the past decade, this also has profound consequences. For one thing, the reported slowdown in annual productivity growth – from something like 2½% in the decade prior to the crisis to about 1% today – could be more apparent than real. For another, true inflation may be even further below the Federal Reserve’s long-run objective of 2% on the PCE than current readings imply.

There is good reason to think that the price mismeasurement problem has gotten worse, but quantifying that deterioration is another thing. The impact on inflation may turn out to be small – perhaps an extra ¼% annually – leaving it well within the range of uncertainty that the Boskin Commission highlighted 20 years ago.

To help you follow our analysis, we'll summarize the big points first:

  • To measure price changes, we need to compare apples with apples. But, advances in technology (especially IT and medical care) lead to the introduction of new goods and services, as well as massive improvements in the quality of goods and services. Neither of these is likely to be accurately captured in current official price measures.
  • It is much easier to measure the impact of innovation on the quality of hardware than on the quality of software and new digital content (like Google, Facebook, Twitter, YouTube, etc.).
  • Medical care inflation has already collapsed, but perhaps not enough to reflect quality improvements.
  • Because of the low weight of IT goods and services and the high weight of medical care in overall consumption, the potential measurement of medical care costs matters far more for price bias.
  • Since computer hardware and software have a higher weight in investment than in consumption, the impact of their price mismeasurement on GDP and productivity growth is larger than their impact on consumer price inflation.

When it comes to technology, the primary issue is adjusting for the change of quality and the introduction of new products. (For a discussion of various sources of bias in price measurement, see here.) If the quality of a good or service improves but the reported price stays the same, its true cost has fallen. For example, the graphics card in an iPhone 5s (last year’s model!) has the capacity to do nearly 1,000 times the number of floating point operations per second (FLOPS) as a 1975 Cray-1 supercomputer, which (adjusting for inflation measured by the CPI basket) cost something like 100,000 times as much. That’s an almost incomprehensible price decline.

The official price changes for computer hardware (plotted as the black line in the chart below) are far less dramatic. Over the past 40 years, reported hardware deflation has averaged nearly 13% annually (compared to the 35%-plus annual decline in the estimated cost of FLOPS). And, even that reported hardware deflation has abated markedly in the past 5 years. In fact, while deflation averaged 15% annually from 1975 to 2010, since 2010, it has been less than 2%.

Personal consumption expenditure chained price index: computer hardware and software (percent change from a year ago)

Source: Bureau of Economic Analysis, Bureau of Labor Statistics and authors’ calculations. (For this chart in FRED, along with a description of the data, click    here   .)

Source: Bureau of Economic Analysis, Bureau of Labor Statistics and authors’ calculations. (For this chart in FRED, along with a description of the data, click here.)

This brings us to the first question: is the measured quality-adjusted price of computer hardware falling too slowly? Byrne, Oliner and Sichel show that semiconductor prices are falling much faster than indicated by official measures, and they have suggested that related problems could be affecting measured prices of computer hardware. If the computer hardware portion of the PCE had maintained after 2010 its pre-2005 relationship with the producer price index, PCE computer hardware deflation would have averaged about 8½% in recent years (the dotted blue line in the chart), rather than just 1¼%. We will come back to this shortly.

How about software and digital content? We find carrying printed maps an awkward way of finding our way around on foot in new cities. Google Maps solved this problem, and this is just one of many ways in which IT has improved our welfare. Yet, Google’s measured contribution to consumer prices is nearly zero: we don’t pay for Google Maps. In fact, the contribution of Google’s products to GDP is like that of an advertising company that produces only intermediate products. When we compute GDP as expenditure on final goods, which is the common way to do it, Google doesn’t appear. As a result, when Google (or Facebook or Twitter or YouTube) improves services, there is no registered price decline. The same has been true for years of broadcast television, but the “freebies” have spread with digital content.

That said, if we look back at the numbers in the previous chart, we see that the measured index for software prices has fallen on average by less than 1% per year since 1975. Can that really be right?

We doubt it. Recent work suggests that we are dramatically underestimating the quality improvements associated with digital content – things like on-demand movie streaming services. (Think of Amazon Prime, where for a $99 per year subscription you can watch any of the tens of thousands of titles whenever you want, in addition to getting two-day delivery of your Amazon orders.) Put differently, the improved quality of household internet services (in terms of content and speed) is unlikely to be fully reflected in the measured price, which rose by about 0.9% annually in the years after 2007.

In a sense, this hardware and software mismeasurement is less important than it seems. The reason is that household expenditure on IT goods and services – which includes digital content – is small. According to the most recent expenditure detail (BEA Table 2.4.5U), IT accounts for less than one percent of personal consumption. As a result, even a 10-percentage-point overstatement of IT inflation year after year would only result in only a 0.1% overstatement of overall inflation. But, this result is largely a consequence of the way in which digital content is included in GDP. Aside from government services, things for which there is no market price generally are not counted in expenditures. We could change this (see, for example, here), but doing so would require a complete overhaul of the national accounts.

What about medical care? Health care accounts for a whopping 17.1% of personal consumption expenditures, which include payments by insurance companies and the government. So, if we get price measurement wrong here, the error would be much more important than with IT. (The CPI includes only out-of-pocket expenses, so the health-care weight is a more modest 7.8%.)

Yet, quality changes in health care are particularly tough to convert into prices. Hospitals and clinics now offer us online medical histories, including nearly instantaneous test results, allowing doctors to proceed rapidly with life- and limb-saving steps. Many of these procedures also take less hospital and recovery time (think laparascopic surgery). Economists have shown that quality-adjusted price measures are materially different in the case of medical care (for an early example, see here). As one economist friend recently noted, today we speak of “bad cancers” because medicine has advanced to the point where we can cure many types of cancer. Not long ago, all cancers were bad.

Personal consumption expenditure chained price index: health care (percent change from a year ago)


 NYU Stern 
 /* Style Definitions */
	{mso-style-name:"Table Normal";
	mso-padding-alt:0in 5.4pt 0in 5.4pt;
   Source: Bureau of Economic Analysis. (For this chart in FRED, along with a description of the data, click    here   .)

Source: Bureau of Economic Analysis. (For this chart in FRED, along with a description of the data, click here.)

The chart above shows the annual percent change in the price index for health care. It has slowed significantly in the past few years. Comparing the 30 years from 1975 to 2005 with the past five years, 2010 to 2015, average annual health care inflation dropped from 6.1% to 1.1%. With an expenditure share of 17%, this drop means that overall PCE inflation was 0.9 percentage points lower on average than it would have been had medical care cost inflation not slowed.

This decline surely reflects a combination of factors that are specific to the United States. If it were primarily driven by technology, which diffuses quickly across advanced economies, one would expect the price decline to be roughly in line with that in other countries. However, a cursory look elsewhere does not show a matching pattern. So, perhaps quality gains are still insufficiently reflected in health care prices.

Our analysis to this point has centered on the implications of price mismeasurement for consumer price inflation. Much of the recent discussion has focused instead on the fall in measured productivity growth. Because IT plays a much larger role in investment than in consumption, mismeasuring the price declines will have a bigger impact on growth and productivity than on consumer price inflation. To see the point, note that domestically-produced IT equipment and software is currently 16% of private fixed investment which in turn is 16.1% of GDP. (That’s not a typo.) This means that a 10-percentage-point overstatement of IT prices will cause investment growth to be underestimated by 1.6 percentage points and GDP growth (and productivity growth) by 0.256 percentage points. Adding in the impact on consumption expenditure would raise this to 0.32 percentage points. This is a bigger deal.

So, what’s the bottom line? We have little doubt that inflation has been overstated for decades. That means that the rise of U.S. real output, real income, productivity, and living standards has been understated materially over the long run. In recent years, IT price mismeasurement probably has worsened this growth and productivity bias significantly. But the potential impact of IT mismeasurement on measures of consumer price inflation – which has been the source of much discussion – is small compared to what a worsening bias in health care prices would imply.