Focus
February 24, 2023 | 14:19
U.S. Inflation: A Tale of Two Indices
Economics
Economics
Economics
Fixed Income Strategy
Economics
Economics
Economics
Economics
Economics
Fixed Income Strategy
Economics
Economics
U.S. Inflation: A Tale of Two Indices |
|
There are two official measures of inflation facing American households, the Consumer Price Index (CPI) and the Personal Consumption Expenditures price index (PCE price index or PCEPI). This week, the PCE price index increased 0.6% in January, lifting the annual change a tenth to 5.4% y/y. Despite drifting up, it is still down from June's 40-year high of 7.0% (Chart 1). Last week, the CPI rose 0.5% in January, lowering the annual change by a tenth to a 15-month low of 6.4% y/y. This is down from the four-decade high of 9.1% also reached last June. These two measures display different inflation rates, as they typically do, and we discuss the reasons why below. The PCE price index has been the Fed’s preferred price metric since 2000 (before, it tracked the CPI), with a 2% annual change in the PCEPI becoming the FOMC’s formal inflation target in January 2012. Despite the policy status, the CPI remains more popular; its results regularly make the media headlines. The affinity partly reflects the fact that the CPI is released earlier (e.g., by 10 days this month), has been around longer and is more comparable with international inflation measures. The CPI is also widely followed because it’s used for making cost-of-living adjustments (COLAs) to wages and benefits, and for escalating other contract prices and costs. As mentioned above, the CPI and PCEPI show different inflation rates, 6.4% and 5.4% respectively in January, for a divergence of 1.0 ppts. Since (monthly) PCEPI inflation figures started in 1960, the difference has ranged from 3.9 ppts in June 1980 to -1.7 bps in August 1983 with a median of 39 bps (Chart 2). The latest spread has narrowed from the 40-year high of 2.1 ppts hit in July. Casual observation notes that the most extreme divergences occur when the inflation rates themselves are at extremes, which is a good segue into the discussion on why these two metrics generate different results. The CPI and PCEPI are constructed differently and any consequent divergence in calculated inflation rates can be dissected into a formula effect, a scope effect, a weight effect and an ‘other effects’ remainder. In Table 1, we show the reconciliation between the annual changes in the CPI and PCE price index for 2022Q4. |
|
Formula effect: Calculated by the Bureau of Labor Statistics (BLS), the CPI measures the prices of items in a representative basket of goods and services for urban households. When comparing two periods and determining an inflation rate, the basket may not change. Technically, this is a Laspeyres index formula. Note that beginning January 2023, the basket is being updated annually instead of biennially but the BLS smooths the impact of basket updates to avoid distorting year-on-year inflation rates. Calculated by the Bureau of Economic Analysis (BEA… the GDP folks), the PCEPI includes the prices of all goods and services purchased or used by all households. This time, when comparing two periods and determining an inflation rate, what was bought or used is the ‘basket’ and it changes each period. Technically, this is a Fisher Ideal index formula. |
|
For both the CPI and PCEPI, these are essentially weighted averages of prices with the relative importance or weights of items based on surveys. When prices rise, households tend to purchase less of the more expensive goods and services and substitute them for more affordable alternatives, given their budget constraints. By construction, this gets reflected in the PCEPI but not in the CPI (that’s why the PCEPI is considered to be closer to a true COLA index). The lack of substitution biases up the CPI relative to the PCEPI. In 2022Q4, the formula effect alone contributed 11 bps to the divergence between CPI and PCEPI yearly inflation rates. And, it typically averages around 16 bps (Chart 3). Scope effect: The PCE price index is a much broader concept than the CPI. It includes the prices of all items, not just the expenditures paid for ‘out of pocket’. The CPI includes only the latter. As such, the PCEPI includes the prices of goods and services bought on behalf of households by governments and businesses, along with final consumption expenditures of non-profit institutions serving households (think of a service club). For example, along with the prices of medical goods and services paid directly by consumers, the costs of such things such as Medicare and employer-sponsored health insurance are also included. Education prices include the cost of public education along with tuition. About 25% of PCE outlays are not covered by the CPI. However, a more complete and comprehensive spending picture comes with a consequence. Roughly 20% of PCEPI prices or costs are unobservable (e.g. financial services furnished without payment), or are infrequently observed, and must be estimated by the BEA on a monthly basis. In 2022Q4, the scope effect contributed -35 bps to the discrepancy between CPI and PCEPI yearly inflation rates, and it typically stays on the negative side of the ledger (it tends to bias up PCEPI inflation relative to the CPI). Weight effect: The source for the CPI’s relative importance of items, or spending weights, is the BLS’s Consumer Expenditure Survey. However, the source for the PCEPI’s weights are all business surveys such as the Quarterly Services Survey and the Census Bureau’s other surveys on retail sales. Different surveys can give rise to different weights, but it’s the broader scope of the PCEPI that impacts weights the most. The weights of most items that both indices share will necessarily be smaller in the PCEPI. And, in an area like medical goods and services, that includes large outlays not covered in the CPI, the PCEPI weights will be noticeably larger. Meantime, in determining the contribution to the discrepancy between CPI and PCEPI inflation rates, the largest-weighted CPI items end up having the most influence, such as rent of shelter and energy. In 2022Q4, the weight effect alone (151 bps) could account for all the discrepancy. Other effects: While item weights are derived from different sources, usually the CPI and PCEPI use the same item price. However, in some case, different prices are used. For example, the CPI figure for airline fares is based on a sampling of routes. The PCEPI figure is based on airlines’ passenger revenues and miles flown. Data revision is another issue. The seasonally unadjusted CPI is nearly never revised, with only the seasonal adjustment factors getting re-estimated annually (this had a noticeable impact this year in turning the ‘tenth dial’ on some monthly moves). The BEA doesn’t publish seasonally unadjusted PCE price data, and the published figures get revised often (monthly back two periods, annually for up to five years). In 2022Q4, these other effects contributed 18 bps to the divergence between CPI and PCEPI yearly inflation rates. |
As a final note, markets were recently focussing more than normal on how the CPI and PCEPI were constructed as they scrambled to create a CPI version of the Fed’s new favourite inflation metric, the PCEPI’s core services excluding housing costs, also known as ‘supercore’. In January, it increased 0.6% which lifted the annual change by four tenths to 4.6% y/y. The version the market seems to have settled on is the CPI’s core services excluding tenant and owners’ equivalent rents. In January, it increased 0.4% which kept the annual change at 6.2% y/y. From a technical perspective, this could have been made even more comparable. While food services are included in the food component of the CPI, they are not in the PCEPI. So, excluding food, energy and other goods still leaves food services in the PCEPI but not the CPI. Adding food services back into the CPI version has it increasing 0.4% in January which lifted the annual change by a tenth to 6.6% y/y. However, the refinement does little to change the general conclusion… underlying inflation is still appearing stubborn no matter how you measure it (Chart 4). |
|
