"When you can measure what you are speaking about, and express it in numbers, you know something about it; but when you cannot measure it, when you cannot express it in numbers, your knowledge is of a meagre and unsatisfactory kind." William Thomson, Lord Kelvin, 1883.
Policy, especially monetary policy, is about numbers. Is inflation close to target? How fast is the economy growing? What fraction of the workforce is employed? And, what is the relationship between the policymakers’ tools and their objectives? Answering all of these questions requires measuring a broad array of economic indicators, with consumer prices high on the list. In this post, we discuss some of the pitfalls in measuring prices.
Price indices of the sort that we use today have been around since the late 19th century. In the United States, near the end of World War I, the National Industrial Conference Board starting constructing and publishing a cost-of-living index. This work was eventually taken over by the Bureau of Labor Statistics (BLS). Over the past century, the theory of price indexes (see, for example, here and here) and the means of measurement have both moved forward substantially.
With the advent of inflation targeting, price indices have taken on a new prominence. If monetary policymakers are going to focus on controlling inflation—setting numerical targets for which they are then held accountable—then the construction of the price index itself becomes an issue. What is included and how can become critical to the way policy is conducted and to the achievement of the stated objective, namely price stability.
There is a broad array of interesting and important challenges facing statisticians in the construction of price indices: how to include new goods, how to treat quality improvements, how to handle changes in consumption patterns, and the like. As we discussed in an earlier post, some of this is related to how technology is treated. And economists have expended enormous amounts of energy trying to find ever better ways of removing high-frequency volatility from inflation data to ensure that policy responds to signals, not to noise (see here and here).
Our focus in this post is somewhat different: we highlight how conventional approaches to price measurement influence policy’s ability to achieve the ultimate goal of price stability. For this purpose, we will look at two important components of consumer price measures: owner-occupied housing and health care.
Starting with housing, there are three common ways to include housing in price indices. The first is the cash-flow approach that simply takes owner’s out-of-pocket expenses. The second is the net acquisition approach based on the cost of constructing a new home (excluding the land). And the third is the rental equivalence approach that estimates the amount owners would have to pay if they were to rent the home.
Before continuing, it is worth noting that Europe’s Harmonized Index of Consumer Prices (HICP) currently excludes owner-occupied housing. Since this is the index targeted by the ECB, changes in the cost of living arising from higher or lower house prices will affect euro-area monetary policy only insofar as they give rise to changes in rent. In practice, that relationship—between house prices and rent—is often tenuous and lagging, not least to due to widespread rent controls.
In the United States, we have experience with the first and third of the measurement methods. Prior to 1983, the BLS used the cash-flow approach. Since then, the CPI has been based on the rental equivalence approach. National accounts data—personal consumption expenditures (PCE) and the associated price indices—are based on the latter of these. That is, owners are assumed to rent their homes from themselves, so there are adjustments made both to income and to consumption.
There are some pretty big pitfalls here. Starting with the cash-flow approach, the inputs into the calculation include house prices, mortgage interest rates, property taxes, insurance, and maintenance costs. Consider the consequences of interest rates being explicitly included in a price index: when interest rates rise, mortgage costs rise, so owners’ payments increase, and measured inflation goes up.
How does this matter for policymakers? In the following chart, we plot U.S. CPI inflation over the next 12 months on the vertical axis against the three-month Treasury Bill rate on the horizontal axis. The red dots are for the 1969-to-1982 period and use 1983 vintage data. The black dots are for 1990 to 2004, using current vintage data. The difference is striking: in the earlier period, higher interest rates are associated with higher measured inflation a year later. Unless an inflation-targeting central bank accounts for this pattern, it will be prone to chasing its tail, potentially amplifying true inflation cycles.
CPI inflation and interest rates: 1969 to 1982 and 1990 to 2004
Note: The dashed lines are based on simple fitted regressions. The slope declines from 0.41 in the first period to 0.09 in the latter period. Sources: FRED and ALFRED.
A number of countries use the acquisitions approach. New Zealand and Australia are examples. At first, this may seem quite sensible: it follows the method used for consumer durables, where the purchase of new washing machines or television sets is what is included in the price index. But, like the cash-flow approach, the acquisitions approach presents a big pitfall, especially for a small open economy.
As an example, consider the challenge that New Zealand faces today: The latest reading on annual CPI inflation is 0.2 percent, well below the 2 percent objective, even though housing prices in Auckland have surged nearly 50 percent over the past 18 months. Received wisdom tells us that under such circumstances, an inflation targeting central bank should cut interest rates. And, in August, the Reserve Bank of New Zealand (RBNZ) did just that, lowering the official cash rate that they control by 25 basis points to 2 percent. What will this do? One normally strong component of the monetary transmission mechanism is that an easing drives up asset prices and encourages lending. But in New Zealand house prices are booming, with Auckland prices approaching 10 times income and rental yields below 3 percent. In addition, household debt exceeds 160 percent of income.
This is where measurement comes in. Because New Zealand uses the acquisitions approach—which focuses on construction costs—when lower interest rates drive up the price of existing homes, measured consumer prices do not rise! Were a rental-equivalence price measure available, it might show a substantially higher inflation rate now and over the horizon relevant for setting monetary policy.
Beyond housing, the second big issue that arises in price index construction is the treatment of goods and services that individuals do not pay for directly. In most countries, the biggest such item is health care. As the chart below shows, in the euro area and the United States, out-of-pocket health care outlays are only a fraction of total health care expenditure. The difference is provided and paid for by a combination of public- and employer-provided insurance.
Euro Area and the United States: Out-of-pocket health expenditure (percent of total expenditure on health care), 2014
Source: World Bank.
Driven by improvements in technology, health care inflation has been quite low in recent years. The most recent U.S. reading is less than 1 percent at an annual rate. Should health care inflation drop further, say to zero, the impact on the HICP and CPI indices (which reflect out-of-pocket costs) would be far smaller than the impact on broad price measures based on the national accounts. Given the substantial changes in health care inflation in recent years, the way these measures respond can have a material impact on policy.
Reflecting on these issues helps us to understand why the Federal Reserve’s Open Market Committee has chosen to frame its inflation objective in terms of the PCE price index rather than the CPI. In addition to including total health care costs, the PCE index uses the rental equivalence method for measuring owner-occupied housing. In the end, this approach will give us monetary policy that is more consistent with policy’s ultimate objective of price stability. As far as core measures of PCE inflation go—currently 1.7 percent—the Fed now is only modestly below its 2-percent target.
The bottom line is that careful quantitative measurement still matters greatly. Even where government statisticians have broadly implemented the best-available approaches, there remains considerable room for improvement, no least in accounting for quality and the introduction of new goods and services. Lord Kelvin didn't faces these challenges—the physical world doesn't have an incentive to innovate the way firms and households do—but he knew how meager our understanding would remain without innovation in measurement.