Confidence is the name of the Fed game
Wednesday we get the CPI for April. The stakes are high after the three tough months. It feels like every basis point counts, but that's not true.
We now live in the Fed’s data-driven world. This approach to interest rate policy makes some sense, given how unpredictable the economy has been during the past four years. However, it puts considerable pressure on the data and our ability to interpret it. Statistics come with some fuzziness and that calls for caution. However, the Fed and everyone else want confidence that inflation is coming down.
“Confidence” came up repeatedly at the Fed’s last press conference. Here’s one mention from Fed Chair Powell:
We have stated that we do not expect it will be appropriate to reduce the target range for the federal funds rate until we have gained greater confidence that inflation is moving sustainably toward 2 percent. So far this year, the data have not given us that greater confidence. In particular, and as I noted earlier, readings on inflation have come in above expectations. It is likely that gaining such greater confidence will take longer than previously expected. We are prepared to maintain the current target range for the federal funds rate for as long as appropriate. We are also prepared to respond to an unexpected weakening in the labor market.
Today’s post reminds us how difficult and slow it can be to build confidence with data, whether it concerns CPI, payrolls, or anything else. We want to know the truth about the economy, but the data have limits. The United States is a $28 trillion economy with over 160 million workers and over 6 million employer firms. It’s incredibly diverse and dynamic. It should come as no surprise that the statistics we pour over at 8:30 am, while informative, are imprecise.
Imprecision is inevitable.
The ‘main event’ for the month is the April CPI on Wednesday. The CPI survey collects about 94,000 prices and 8,000 rental housing unit quotes each month. That’s a lot, but it’s a tiny fraction of the prices in the US, and a sample causes variability in the estimates. The Bureau of Labor Statistics provides estimates of the truth, not the truth. Using a sample creates statistical uncertainty, but a sample is the only way to get a timely read on inflation with a reasonable budget and respondent burden.
Imprecise does not mean fake. It means we must handle the estimates with care. Unfortunately, in the hyper-data-driven world, that’s the opposite of the ritual on CPI Day. For example, last month, the percent change in core CPI in March came in at 0.4%, not the consensus of 0.3%; that was the third month with a higher-than-expected print. Treasury yields shot up immediately, and markets pushed out bets on the timing of the Fed’s first rate cut.
How big was that miss on consensus? That core reading in March was 0.36%, so it's about 0.01 percentage point of rounding down to 0.3%; the same was true of the 0.36% in February. It’s more than a ‘so close’ moment. The difference of a basis point is not different, at least statistically.
The standard error of an estimate, such as of monthly inflation, captures the imprecision of the data due to survey sampling. According to the BLS, with a high (90%) degree of confidence, we can say that the true increase in core CPI in March was between 0.29% and 0.43%. See the confidence intervals (in gray) around recent monthly changes (in orange):
Taking the fuzziness of estimates into account does not qualitatively change the pickup in monthly inflation in the first quarter, but it could improve (or worsen) the trend quantitatively. For example, the February and March estimates in 2024 were (barely) within the confidence intervals of the September and November prints in 2023. The latter two counted toward the good six months of inflation data last year.
The sampling error is only one possible reason that the estimate may differ from ‘the truth’; others include seasonal adjustment, hedonic quality adjustments, and other imputations like Owners’ Equivalent Rent.
Greater confidence in the data takes time.
The lesson is not to ignore the estimates but to understand the fuzziness and not chase basis points. A simple statistical solution is to look at longer horizons. The adage ‘three-month average is your friend’ is a good way to rein in the fuzziness. The longer the time horizon, the more precise the estimates.
The 90% confidence interval on a monthly change in core CPI annualized is +/- 0.5 percentage point (far left bar). Knowing that the true core CPI is likely between 2.5% and 3.5% would tell us little about what the Fed might do next. However, averaging multiple months helps sharpen the estimate. For example, the 12-month change (far right) has a confidence interview of about 0.1 percentage point.
Powell is not referring to shrinking the error bands when he says it will take time to gain confidence that inflation will reach 2%. However, there is a connection. The Fed knows the data are imprecise and can bounce around for many reasons. A good print for April CPI would be a relief but not a game changer. It’s only one month. A bad print, on the other hand, would add to a string of bad ones.
Stepping back, the start of the year notwithstanding, the progress on inflation is notable. The disinflation since 2022 is clear, albeit incomplete.
That big miss might not be so big.
It’s not just the CPI that’s imprecise. The numbers on Jobs Day, which come from surveys, are also imprecise. Earlier this month, we learned that payrolls rose 175,000 on net in April, well below the consensus forecast of 241,000. The estimate came from a sample of about 100,000 businesses and government agencies. Again, a tiny fraction of those in the economy.
Considering the sampling error, the actual estimate is not statistically different from the consensus forecast at the 90% level of certainty. Specifically, the confidence interval was between 45,000 and 305,000. That’s huge. Here again, smoothing out the noise month to month is useful. Yes, the April estimate was below 200,000, but the three-month average remains solid and within the range of its pace in the past year.
Every month counts, but we must be careful not to tell stories based on one month’s print alone. At the same time, we should not ignore a month.
In closing.
Powell is often asked what the Fed must see to be confident about inflation. What data? What values? How many months? Powell’s replies include vague phrases like the “totality of the data” and “more time.” Fedspeak has verbal standard errors. The 90% confidence interval around what Powell says is large, too. That’s intentional and allows flexibility, but it makes it hard to interpret the Fed.
The data are fuzzy, and the Fed is fuzzy. So, what should you do?
When the CPI release comes out this Wednesday, remember you are looking at an estimate of inflation in April, not necessarily the true inflation. The estimate will be important mainly when judged in the context of earlier estimates. That’s the Fed’s approach to interpreting data, and that informs its decision-making.
Does the Fed use this tool?
https://en.m.wikipedia.org/wiki/MIT_Billion_Prices_project
Fabulous piece. Thanks Claudia. Markets and investors pine for black and white answers. After all almost everything else is tied down to the cent...account balances, financial statements, margins and balance sheets. Therefore macro reports are also expected to be accurate with every release...few acknowledge and accept that there are shades of gray that are at best directionally useful.