Consumer Price Indexes

The purpose of a price index is to summarize information on the prices of multiple goods and services over time. Consumer spending accounts for about two thirds of the U.S. gross domestic product (GDP). The Consumer Price Index (CPI) and the Personal Consumption Expenditure deflator (PCE) are designed to summarize information on the prices of goods purchased by consumers over time. In a hypothetical primitive society with only one good—say, one type of food—we would not need a price index; we would just follow the price of the one good. When there are many goods and services, however, we need a method for averaging the price changes or aggregating the information on the many different prices. The rate of change of prices—inflation—is important in both macro- and microeconomics. Estimating inflation and real growth, for example, requires measures of price changes, and in a flexible, dynamic modern market economy, obtaining accurate measures is complicated. A single large superstore may contain more than fifty thousand separately priced items. Within that individual store, new items are continually introduced and old items discontinued. The quality of many items improves in some objective way—greater energy efficiency, more durability, less maintenance, to name a few. Of course, many more items claim to have improved. When quality increases but the price stays the same, the real price has fallen. Even with modern scanner technology, summarizing what happened to prices in just one store over a period as short as one month is complicated. Doing so for the entire economy is vastly more complex.

To obtain information on various prices requires not only measuring the prices but also weighting the various components in the index. Weighting each price change equally would be simple but not very revealing. For example, if the price of red delicious apples fell by 5 percent and rent rose by 5 percent, such an index would suggest that there had been no change in the overall price level. But that would be silly. We need to “weight” the goods on which consumers spend more of their income more heavily than those on which they spend less.

When economists try to measure the “true” inflation rate—the rate of change of prices—it is to answer the question, “How much more income would consumers need to be just as well off with a new set of prices as the old?” Thus, a cost-of-living concept is at the core of proper measures of prices and of changes in prices. This clearly involves tracking “substitution”—that is, how consumers respond to the changes in the relative prices of various goods. It also requires measuring quality-adjusted prices. One would not want to count as inflation a major improvement in quality that resulted in a tiny price increase.

Most traditional consumer price indexes, including the CPI in the United States, measure prices with a fixedweight system, taking the expenditure weights from some base period as given. Table 1 reports the most recent weights on very broad categories of goods from 2002; the Bureau of Labor Statistics (BLS) derives these weights from expenditure surveys that report how much consumers spent on different types of goods and services. For example, at a very broad level of aggregation, those weights are 15.6 percent for food, 6.0 percent for medical care, 40.9 percent for housing, 17.3 percent for transportation, and so on. Within each category, of course, are thousands of specific goods; for example, red delicious apples of a certain size and quality are a component of the apples subcategory, which is a component of fresh fruits, which in turn is a component of fresh fruits and vegetables.

With these expenditure weights at hand, it still takes a high-quality, expensive operation to track the prices. And whose prices? For commodities purchased where and how? In the United States, there are two closely related consumer price indexes. One measures the change in a weighted average of consumer prices, with the base year expenditure weights, for a typical urban family, the socalled CPI-U. The other, not quite identical, construct is the CPI-W, which measures prices for urban wage and clerical workers. I focus here on the more widely cited CPI-U. Neither of these fixed-weight indexes accounts for substitution, the fact that consumers substitute away from goods whose prices increase more and toward goods whose prices increase less.1

The CPI serves, and should serve, many purposes. For example, the CPI is used to measure consumer inflation on a monthly basis; to make cost-of-living adjustments in Social Security, income tax brackets, and other government programs; to provide price data as inputs to the National Income and Product Accounts (although the Commerce Department now uses its own set of weights and methods to construct its PCE deflator from these raw data).

Figure 1 provides recent data on the U.S. CPI-U. The CPI-U sets the index = 100 for the years 1982–1984. As the figure shows, the pace of measured consumer inflation has slowed considerably relative to the 1970s and 1980s, has recently been running in very low single digits, and has had considerably less variation than in the high-inflation 1970s and early 1980s.

People change their spending patterns over time, and do so specifically in response to changes in relative prices. When the price of chicken increases, for example, people may buy more fish, and conversely. Hence the weights change, and a price index that fails to account for that—as does the fixed-weight base period CPI—overstates the true change in the cost of living.

There are two obvious approaches to weighting the prices. The first uses a fixed-base period weighting: quantity or expenditure weights remain fixed at their base period levels, and then we see what happens to the weighted average of prices as prices subsequently change. An alternative possibility is to use the expenditure weights or quantities in the second period, after the substitution. Economic theory strongly supports the idea of taking an average of these two numbers, a point originally made by the great American economist irving fisher (1922). Since 2002, the BLS has computed a closely related measure called the chained-CPI; it has been rising much less rapidly than the traditional CPI-U, suggesting that the failure to account for consumer substitution explicitly is a serious weakness of the official CPI.

Similarly, where people make their purchases changes over time. Discount stores and online sales have become more important relative to traditional small retailers. Because price data are collected within outlets, the shift of consumer purchasing from discounters does not show up as a price decline, even though consumers reveal by their purchases that the price decline more than compensates for the potential loss of personal services. Thus, in addition to substitution bias among commodities, there is an outlet substitution bias.

Even when purchases are made can become important. We typically measure prices monthly, during a particular week. But if, for example, consumers get wise to post-Christmas discounts and start buying a lot more holiday items after Christmas, surveys that look solely at prices in the second week of December will miss this.

Another problem is that price data tend to be collected during the week. In the United States, about 1 percent of price quotes are collected on weekends, despite the fact that an increasing share of purchases is made on weekends and holidays (probably reflecting the increase in prevalence of two-earner couples). Because some outlets emphasize weekend sales, there may be a “when” bias as well as “what” and “where” biases. This phenomenon may explain, in part, recent research suggesting that prices rise less rapidly in data collected by scanners on actual transactions than in that collected by BLS employees gathering data on prices on shelves and racks.

Finally, an additional bias results from the difficulty of adjusting fully for quality change and the introduction of new products. In the U.S. CPI, for example, VCRs, microwave ovens, and personal computers were included a decade or more after they had penetrated the market, by which time their prices had already fallen 80 percent or more. Cellular telephones were not included in the U.S. CPI until 1998.

The CPI currently overstates inflation by 0.8–0.9 percentage points: 0.3–0.4 points are attributable to failing to account for substitution among goods; 0.1 for failing to account for substitution among retail outlets; and 0.4 for failing to account for new products. Thus, the first 0.8 or 0.9 percentage points of measured CPI inflation is not really inflation at all. This may seem small, but the bias, if left uncorrected for, say, twenty years, would cause the change in the cost of living to be overstated by 22 percent.

The U.S. CPI is one of the few economic statistics that is never revised, even if subsequent data reveal that the published statistic is wrong. This is done because many contracts and other government programs are expressly indexed or adjusted to the CPI, and revisions would cause practical and legal complexities.

We know that different sets of consumers have different expenditure weights because they spend different fractions of their income on the various commodities: renters versus homeowners, the middle aged versus the elderly, and so on. Interestingly, most analyses find only modest differences in inflation rates across groups with different expenditure weights.

What about differences across groups in prices and rates of change of prices? For example, do the prices paid by the elderly differ from those paid by the general population? And if they do differ, have the differences changed over time? Economic theory suggests the prices will not differ much for most items, but we do not have serious empirical evidence on this score.

Thus, inflation—the rate of change of prices—is hard to measure accurately. Government statisticians in all countries, especially those at the U.S. Bureau of Labor Statistics, have made numerous important improvements over the years. Yet, new products are introduced all the time, existing ones are improved, and other products leave the market. Relative prices of various goods and services change frequently, causing consumers to change their buying patterns. Literally hundreds of thousands of goods and services are available in rich, industrialized economies. As we have become richer, our demands have shifted toward services and away from goods, and toward characteristics of goods and services such as enhanced quality, more variety, and greater convenience. But all these factors mean that a larger fraction of what is produced and consumed in an economy today is harder to measure than it was decades ago, when a larger fraction of economic activity consisted of easy-to-measure items such as tons of steel and bushels of wheat. Thus, how to obtain information on who is buying what, where, when, why, and how, in an economy, and then to aggregate it into one or a few measures of price change raises a host of complex analytical and practical problems.

Price index research and measurement—at one time considered staid and boring—has undergone a renaissance in recent years. Price index research in academia, think tanks, and government agencies, plus practical improvements in real-time government statistics, will be an ongoing effort of major importance and immense practical consequence for many years to come.