Measuring inflation: What’s changed over the past 20 years? What hasn’t?

Despite improvements made over the past two decades, the Consumer Price Index (CPI) still overstates the annual rate of inflation by about 0.85 percentage points, according to Brent Moulton, a 32-year veteran of the agencies that compile the nation’s economic statistics who retired in 2016 as Associate Director of the Bureau of Economic Analysis (BEA). This represents some improvement since 1996, when the congressionally appointed Boskin Commission estimated that the CPI overstated the annual rate of inflation by 1.1 percentage points.

Moulton makes these new estimates in a paper commissioned by the Hutchins Center on Fiscal and Monetary Policy at the Brookings Institution. The report—“The Measurement of Output, Prices and Productivity: What’s Changed since the Boskin Commission?”—is part of a multi-year Hutchins Center initiative to improve the measurement of productivity. 

In the decades since the Boskin Commission report, the Bureau of Labor Statistics (BLS) has made several changes to the methodology used to calculate the CPI. For example, a major criticism of the CPI was that it failed to account for consumers changing their buying patterns when prices change. When the price of Swiss cheese goes up, for instance, consumers might buy more cheddar cheese but less Swiss. As a result, the cost of living does not rise by as much as it would if consumers did not change their behavior. In 1999, the BLS altered how it measures prices for a specific category of goods, such as cheese, to account for consumers substituting within that category as prices change (called lower-level substitution).

Another source of CPI bias is that goods often change in quality over time. For instance, Apple regularly releases new MacBooks, incorporating more storage and faster processors. The BLS now does a better job of holding quality constant over time so that the CPI reflects changes in the cost of living more closely, says Moulton. For example, in January 1998 the BLS started estimating the value of certain features of personal computers, like storage and speed, to adjust the prices of new models for quality improvements.

Moulton finds that the BLS managed to substantially reduce bias from lower-level substitution and new products/quality change.

This table compares Moulton’s new estimates to those of the Boskin Commission and subsequent work by David E. Lebow and Jeremy B. Rudd, economists at the Federal Reserve Board who published a review of CPI bias in 2003. Moulton finds that the BLS managed to substantially reduce bias from lower-level substitution and new products/quality change. However, he finds that the BLS failed to address upper-level substitution bias, the bias from consumers substituting across categories as prices change. Notably, Moulton estimates that CPI bias today is only slightly lower than in 2003, suggesting that most improvements occurred in the late 1990s and early 2000s.

Updated Estimates of CPI Bias (percentage points per year)

The mismeasurement of price changes has important implications for measured output and productivity. Real output, the numerator in productivity calculations, isn’t directly observed. Instead, statistical agencies rely on price indexes “to determine how much of the change in the sales of a particular good or service is due to price changes and how much is due to changes in real output, that is the quantity or quality of that good or service,” Moulton says.

In an extensive appendix, he catalogs major changes in methodology since 1997 that have affected how output and prices are measured. For instance, the BLS now includes a wider range of services in its producer price index (PPI), and the BEA includes investment in intangible intellectual property, like software, in gross domestic product (GDP). Both changes represent advances in coverage since the 1990s that, along with changes to the CPI, improved the measurement of output.

[C]hanges in the economy have offset some of the improvement in measurement.

Moulton points out that changes in the economy have offset some of the improvement in measurement. Digital goods, the service sector, and multinational enterprises are all more important now than 20 years ago, and all are difficult to measure accurately.

He suggests tackling the remaining biases by: (1) improving the measurement of quality-adjusted prices for information and communications technology equipment and associated digital services; (2) expanding the quality adjustment process to a wider range of new goods, not just those that directly replace a disappearing good; (3) making a systematic effort to deal with the effects of globalization on the measurement of GDP.