Lawler on Demographics: Observations and Updated Population Projections

For Housing "demographics are not nearly as positive" as previous projections

CR Note: This is a technical post, but the key take away for housing is: “the “demographics” are not nearly as positive over the next several years as the Census 2017 projections would have suggested.”

Way back in 2015, I projected a surge in homebuying in the 2020s based on demographics. The surge has happened, but the demographics are not as positive as originally predicted. Many analysts are repeating my earlier analysis without adjusting for less net international migration (NIM), and more deaths.

From housing economist Tom Lawler:

Analysts who rely on US population projections (total and by characteristics) to project other economic variables (including those housing related) have been frustrated by the lack of any credible official projections.  The last official intermediate and long term population projections from Census was done in 2017, and (which I’ve written about before) those projections significantly over-predicted US births, significantly under-predicted deaths, and significantly over-predicted net international migration (NIM).  On the latter score it’s not clear the extent of the over-prediction of NIM in the Census 2017 projections, as Census has not released updated NIM estimates for the 2010-2020 period.  The last such NIM estimates were in the “Vintage 2020” population estimates, which did NOT incorporate the Census 2020 results, and the Vintage 2020 population estimate for April 1, 2020 was about 2.05 million below the Census 2020 tally.  Presumably most of the Vintage 2020 “miss” reflected higher NIM over the past decade than that shown in the Vintage 2020 estimates, though the yearly differences in NIM are unknown, and Census has not yet released updated intercensal population estimates for the 2010-2020 period that are consistent with the Census 2020 results.

On this last point, the latest Census population estimates for each year from 2011 to 2019 do NOT incorporate the Census 2020 results, and as such are not comparable to the Vintage 2021 population estimates for 2020 and 2021.  Updated intercensal population estimates for last decade are currently not scheduled for release until 2023, and I have no idea when Census might release a new set of intermediate and long term population projections. 

To be sure, it is understandable why Census may be delaying the release of updated population projections.  First, because of the pandemic Census 2020 was “challenged,” and many key releases have been delayed.  Second, trying to project deaths by age as well as net international migration by age is especially challenging given the uncertain path of the pandemic.

But analysts who use population projections need to realize that the latest official Census population projections are dated, inconsistent with Census 2020 results, and used what are completely unrealistic assumptions about the key drivers of population growth. 

The purpose of this piece is to provide analysts with an updated set of US population projections by age through 2025 that (1) take into account Census 2020 results, and (2) use what I believe are more realistic assumptions about US births, US deaths, and NIM.

Before going on to that, however, on the next page is a comparison of the Vintage 2021 population estimates by age (which take into account Census 2020 results) with the Census 2017 population projections by age for July 1, 2021.  As the table indicates, there are substantial differences not just in the totals but also in the age distributions.

But now on to the updated population estimates. In terms of deaths, for the 12 month period ending 6/30/2022 I use the CDC deaths reported through the end of March of this year (reporting delays materially understate later months) by age group and assume the YOY growth rates for Q2/2022 are similar to that of Q1/2022. While CDC does not show monthly deaths for single year of age it does show deaths by single year of age since 1/1/2020, so I use the distribution of deaths by single year of age within each age group to produce estimated deaths by single year of age (which are necessary to produce population estimates/projections).

For the 12-month period ending 6/30/2023 I assumed that death rates by age equaled the average of death rates for 2022 and death rates for 2019. This may be optimistic. For 2024 and 2025 I assumed that death rates by age would equal those in 2019. Again, this may be optimistic.

My assumptions by year for NIM (net international migration) are shown on the next page (2021 is the Vintage 2021 Census estimate.) In terms of assumptions about the age distribution of NIM, I used past (and unpublished) Census assumptions from previous vintages, though I did a little rounding and assumed zero NIM for ages 85 and over.

Finally, I used the following assumptions for births.  US births have been way below previous projections, but the monthly data through the end of last year suggested that the trend may have reversed.  These assumptions, though, do not affect the age groups that would impact housing, labor force, etc.  (2021 in the Vintage 2021 Census estimate)

Below are the resulting population projections by 5-year age groups out to July 1, 2025.  (I also show the Vintage 2021 estimates for 2020 and 2021).  Also shown are the Census 2017 projections for those years.

I do have several caveats.  First, the Post-Enumeration Survey (PES) results for Census 2020 suggested some statistically significant over and undercounts for selected age groups (and some really large undercounts for certain racial/ethnic groups).  However, Census has no plans to revise the Census 2020 figures, and future Census population estimates will assume the Census 2020 numbers are correct.

Second, assumptions both about death rates and NIM (total and by age group) could be off, and by quite a bit.  By request I can generate different projections using different assumptions.

However, I figured that it might be useful for some analysts to have alternative population projections that they could use for intermediate-term forecasts of other variables.

For example, some analysts like to look at projections of the number of 30 to 39 year olds, as that age group would likely include a large number of potential first-time home buyers.  The Census 2017 Projections predicted that from July 1, 2020 to July 1, 2025 the number of 30-39 year olds would increase by 2,571,433, with a 697,200 increase in 2021.  This updated projection has the number of 30-39 year olds increasing by only a third of that amount (864,670), with a 291,209 increase (from Vintage 2021) in 2021.  The reasons for the big differences are (1) a different starting age distribution; (2) significantly higher deaths; and (3) significantly lower NIM.   Obviously, the “demographics” are not nearly as positive over the next several years as the Census 2017 projections would have suggested.

I’ll have more on these topics later.