Data Measurement and Monetary Policy

Instead, economic data contain imperfect or noisy information about these central elements relevant to policymakers. For example, the Fed's monetary policy framework has long acknowledged the challenges of estimating the level of maximum employment. While inflation has an exact definition and a measured target, the concept of price stability is also prone to measurement challenges, as the quality of products and services changes over time. Economists attempt "hedonic" adjustments to control for these changes as the basket of goods consumed by households evolves, but this process is far from straightforward. (The 2025 film Superman may not be a hedonic improvement over the 1978 version, although our ability to watch both movies wherever and whenever surely is.)

Policymakers are also being confronted by lower response rates to economic surveys, which makes it harder to collect timely and accurate data. Response rates have declined for almost all surveys collected by the federal government to inform economic statistics. This includes nonfarm payrolls, unemployment, the consumer price index (CPI), the employment cost index, and job openings. Goldman Sachs researchers recently estimated that declines in response rates by between 10 and 30 percentage points have resulted in 26 percent higher standard errors relative to 2015-2019 collections alone. Since 2019, the number of price changes the Bureau of Labor Statistics imputes rather than manually collects when calculating its estimate of CPI has also risen, which may increase measurement error depending on the magnitude of consumption in the missing geographies and product categories.

For monetary policymakers, the impact of a decrease in measurement precision depends on the type of error. Is it the case that the variability has increased but on average the change is neutral? Sometimes called classical measurement error, in this case, the optimal response to any new data generally remains the same, as policymakers can see through the measurement error over time and the data still represent the best approximation of the economic variable of interest. However, in models with gradualism, where frequent changes to the policy interest rate have additional costs, this may lead to slower adjustment in rates, particularly in the face of extreme data realizations when policymakers don't know whether the change in data arose from measurement error and whether that error was positive or negative.

On the other hand, if we know that an economic variable is biased, or systematically mismeasured, then the optimal response is to respond more (or less) than suggested by data realizations. For example, if policymakers know that unemployment is measured too high by 20 basis points, they can simply subtract 20 basis points from the observed data. However, when the nature and extent of the bias is unknown — or worse, if the bias is related to cyclical factors — it can be harder to adjust for. In this case, where the bias is unknown, the optimal policy response may look similar to the case with no bias, but outcomes may be worse.

One clear implication of increased noise is the increased importance of collecting additional informative signals. At the extreme, if we had a perfect measure of any economic variable, there would be no point in gathering other signals. Without such perfect measures, it has become increasingly valuable to take advantage of additional information about economic outcomes such as we can find from new surveys and qualitative outreach.

To be sure, the challenges of making optimal policy under uncertainty extend beyond the questions related to measurement error in data collection. Regime shifts and possible structural breaks as technology evolves make it exceedingly difficult to find and establish the existence of stable econometric parameters. This is an old problem, and old solutions continue to be relevant today: gather more data. Indeed, the Bureau of Labor Statistics began collecting data on retail wages and prices in the early 1890s to better understand the effects of new tariffs. Collecting more data will help us in the present as well.