>>160543, >>160544
- because NFP is a survey, it tries to account for gains/losses from businesses who opened/closed (aka birth-death). so BLS adds this adjustment in the birth-death bad math
- the math behind this is simplistic and is weighed based on older data (moving average regression)
- if you compare Q1 2025 to Q2 2023 (when peak fed rates hit) there are now more quarterly job losses than gains from birth-death (-150,000 vs +500,000 then)
- the overly optimistic B-D weighting from the past skews the real negative change today
how off is it?
- englander: "there's about a hundred thousand jobs a month that are ... basically just not there" and it swings +/- with the business cycle
- Bloomberg's @AnnaEconomist, @MishGEA, @MBjegovic, along w @EconguyRosie @DonMiami3 @profplum99 @onechancefreedm were some of the first to raise this flaw over the past couple years, in fact NFP was also undercounting jobs in 2022 during the boom
- @MishGEA mentions another wrinkle in auditing these numbers: the birth-death numbers are not seasonally adjusted, so it's hard to separate the B-D # from seasonal adjustment
- tl;dr the count is off and no one at BLS has seen The Wire
so what's actually accurate?
- the accurate numbers come from BLS's Quarterly Census of Employment and Wages (QCEW) which pulls from state unemployment records, and covers over 95% of U.S. jobs
- englander: "QCEW is basically the universe, it's not a sample... it's very authoritative"
- but the problem is the long lag: 5-6 Months vs NFP's week
- englander also suggests looking at another BLS source: the prime Employment-Population Ratio (25-54 Yrs), it's a separate monthly survey of 60K households and has little revisions
how do we fix this?
- englander's main suggestion is stop B-D from using historical averaging and leverage incoming dynamic data more