Federal Reserve Economic Data

Measures of inflation are some of the most popular data series on FRED. Two of the most important ones are the consumer price index (CPI), constructed by the Bureau of Labor Statistics (BLS) in the Department of Labor, and the personal consumption expenditures price index (PCE), constructed by the Bureau of Economic Analysis (BEA) in the Department of Commerce.

The CPI is probably the most widely watched measure of inflation and is used for many purposes, such as indexing Social Security payments. The Federal Open Market Committee (FOMC) has been looking at the PCE price index since the 1990s, however, and made that index the measure for its official inflation target in January 2012, when they introduced an official target.

To understand how the CPI and PCE inflation rates differ, consider a stylized representation of a price index for the year 2023 as the weighted sum of the prices of three types of goods, which we will call goods 1, 2, and 3.

P₂₀₂₃=w_1,2023 p_1,2023+w_2,2023 p_2,2023+w_3,2023 p_3,2023

The weights, w, represent how much money is spent on each good in the consumption basket. For example, gasoline would get a higher weight in a consumer price index than shoelaces. As relative prices, technology, or people’s tastes change, the weights in the baskets change.

Both the CPI and PCE are constructed in this general way, as the weighted averages of prices of various goods, but they differ.

1. The two price indices measure somewhat different baskets of goods: The CPI is designed to measure the cost of living for an urban consumer, while the PCE measures a broader cost of living. Because of this difference in emphasis, the weights in the baskets differ and the CPI famously places more emphasis on the cost of housing.
2. The weights in the CPI basket aren’t revised as often as those in the PCE basket, meaning that the PCE probably better measures consumer responses to rapidly changing relative prices.

The FRED graph above shows that monthly CPI (blue line) and PCE (red line) inflation move closely together, but the CPI generally exceeds the PCE. Over the full sample since 1960, the arithmetic average of 12-month CPI inflation was 3.77% and the standard deviation—a measure of volatility— of that series was 2.83%. The analogous figures for the PCE price index inflation were 3.31% and 2.43%. That is, the CPI inflation rate was 0.46% higher on average and somewhat more volatile. The fact that the PCE weights are revised more often than the CPI weights helps explain the higher average CPI inflation because consumers tend to substitute away from products whose prices rise sharply and the PCE index more quickly reflects such behavior.

The purple line on the FRED graph shows the difference between the CPI and PCE inflation rates, with an average value of 0.46%. This difference is almost always positive but small, usually in the range of 0 to 1%, but it does increase with overall inflation rates.

One component that may also help depict the difference between these two price indexes is annual inflation in “imputed rental of owner-occupied housing.” Shown by the green line, this is basically what a homeowner would have to pay to live in the house if they were renting it. Since CPI has higher weights for housing, this imputed rent should contribute much more to the CPI than the PCE. And it does seem to be correlated with the overall difference between the CPI and PCE inflation rates (purple line), but saying any more would require more careful analysis.

How this graph was created: On FRED, search for and select “CPI.” From the “Edit Graph” panel, change “Units” from “Index” to “Percent Change from Year Ago.” With the “Add line” option, search for “PCE” and select “Personal Consumption Expenditures: Chain-type Price Index,” then click “Add data series.” Repeat for “imputed rent,” selecting “Imputed rental of owner-occupied housing.” Add a line again, with the CPI and PCE series and apply formula a-b. Toward the top of the editing box, select “Percent Change from Year Ago” in the “Units” box and select “Copy to all.” Adjust the sample period to start on 1960-01-01.

Suggested by Christopher Neely.

In early 2020, the COVID-19 pandemic severely reduced the number of miles Americans drove their cars, as we described in a previous blog post. The pandemic also decreased public transit ridership in a similar way: Many people started working from home, and others took the safety precaution of avoiding travel that would put them in close proximity to others.

However, the use of public transit has not rebounded from the pandemic as quickly as the number of vehicle miles traveled has. In fact, public transit ridership appears to be leveling off at 30% below pre-pandemic levels.

The reasons for the quick rebound in total vehicle miles driven can explain some of the lag in public transit ridership: A study by the Cleveland Fed showed that large numbers of residents in large metro areas migrated to small and midsize metro areas. These residents must now drive farther to reach those large metro areas, which has boosted vehicle miles driven.

But large metro areas also include well-established public transit systems—think, New York City, Chicago, and Washington D.C. Fewer residents in these large metro areas means fewer riders that could use those public transit systems. (Many midsize and small metro areas are less densely populated and have less-extensive public transit systems, so their populations rely more on motor vehicles.)

Fewer people available to use public transit also means less money for it, since there are fewer paying customers. A decline in public transit revenue can reduce hours and available routes, which can further reduce the opportunities for ridership.

There also has been an increase in remote work, which further reduces demand for daily trips. This movement of people and where they can work helps explain why, as of December 2023, public transit ridership numbers have not recovered to levels seen before the pandemic as total vehicle miles has.

How this graph was created: In FRED, search for and select the seasonally adjusted “public transit ridership” series. Select the timeline of 2010 to the latest data for 2023 to focus on the decline of ridership below 2019 levels.

Suggested by John Fuller and Charles Gascon.

Background

The overall unemployment rate is a key macroeconomic indicator that economists, policymakers, and the media follow closely. The Bureau of Labor Statistics calculates it from a monthly survey of households that tallies the number of those not employed but looking for work divided by the total labor force, which itself is the sum of employed workers plus the number of those not employed but looking for work.

A separate unemployment measure that receives less attention is the insured unemployment rate: It is the number of workers currently claiming unemployment insurance benefits relative to the entire pool of workers covered by unemployment insurance. A worker is “covered” if they could claim unemployment insurance benefits after losing their job—and meeting other eligibility requirements.

The insured unemployment rate serves as a switch that activates state-level extended unemployment insurance benefits when it rises above a specified threshold. Some argue that extending benefits for unemployed workers can help stabilize the macroeconomy during downturns. While many states adopt additional criteria tied to the overall unemployment rate, the insured unemployment rate is the one criterion that all states use.

What the data show

The red line in the FRED graph above shows that the share of workers covered by unemployment insurance has grown since the Department of Labor began tracking these data in 1971: Back then, only 70% of workers were covered. Today, about 90% are.

Yet, over the same period, the insured unemployment rate (IUR) has declined relative to the overall unemployment rate, as shown by the blue line in the graph. For the third quarter of 2023, an IUR of 1.3% is divided by an overall unemployment rate of 3.9% to achieve the value of 0.3 shown in the graph—a big drop from the nearly 0.7 value back in 1971.

Note that in 2020, both values in the graph briefly rise above 100%: During the pandemic, the US government expanded unemployment benefits to individuals who would typically be ineligible and paused the requirement that individuals receiving benefits be actively looking for work.

Discussion and analysis

If more people than ever work in jobs covered by unemployment insurance, why aren’t more unemployed individuals claiming benefits?

Individuals separated from jobs covered by unemployment insurance may still be ineligible for benefits under additional state eligibility requirements: for example, enrolling in educational programs; picking up part-time temporary work; earning too low an income while employed; and working for too short a period in their last job. (See Nicholson and Needels, 2006.)

But eligibility alone cannot explain the trend, as many eligible workers don’t file for unemployment insurance. Birinci and See (2023) find that, on average, about 57% of unemployed individuals are eligible for unemployment benefits, but only 61% of the eligible unemployed workers claim these benefits.

Let’s consider several possible reasons why the “take-up” rate (the share of unemployed individuals receiving benefits) is relatively low, including worker characteristics, policy changes, and regional patterns.

Those eligible for unemployment benefits may be less likely to need them: Eligible unemployed individuals who choose not to take-up these benefits tend to have greater wealth and shorter unemployment durations. (See Birinci and See, 2023.) Current Population Survey data identify the most popular reason eligible unemployed workers do not apply for benefits: They do not expect to be unemployed for long. (See Wandner and Stettner, 2002.)

The national average take-up rate was mostly stable throughout the 1960s and 70s, but it declined throughout the 1980s. Blank and Card (1991) argue this sharp decline was driven by composition effects: Regions with persistently low and decreasing take-up rates also tended to have higher rates of unemployment due to sectoral and demographic shifts. At the same time, the share of covered jobs expanded to include more government employees, who are less likely to become unemployed.

Around the same time, unemployment benefits became taxable income, reducing the after-tax benefits filers received. By 1987, these benefits were “fully taxable for all recipients,” which Anderson and Meyer (1997, p. 915) argue was also an important reason for the decline in take-up. More-recent research points out that states and regions with larger benefits and longer maximum benefit durations tend to have higher take-up rates (Bell et al., 2023; Nicholson and Needels, 2006) and that individuals are less likely to apply for benefits when what they can receive is small relative to prior income. (See Birinci and See, 2023.)

Finally, the percent of lost income that can be recovered through benefits and eligibility requirements may be crude proxies for true accessibility. (See O’Leary et al., 2021.) Social institutions can also have an impact, albeit one difficult to quantify. For example, by educating their members about unemployment insurance and helping them file, unions may bolster take-up, leading to lower overall take-up rates as unionization has declined. (See Blank and Card, 1991 and Nicholson and Needels, 2006.) Similarly, difficulty navigating the benefits system due to limited technology or English proficiency may keep some populations from applying, although earlier analysis finds relatively little support for this being widespread.

How this graph was created: Search FRED for and select the insured unemployment rate series (IURSA). From the “Edit Graph” panel, use the “Customize data” option to search for and add the unemployment rate. In the “Formula” bar below, type a/b and click “Apply.” Next, use the “Add Line” option to search for and add the covered employment series. Use the “Edit Line” tab to select Line 2 and the “Customize Data” option again to add the employment level series and apply the formula a/(b*1000). (We multiply the denominator here by 1000 because the employment level is reported in thousands while covered employment is not.)

Suggested by Bill Dupor and Marie Hogan.