by Calculated Risk on 4/03/2017 05:11:00 PM

From housing economist Tom Lawler (note: this is work in process, but here are some initial findings)

In assessing the intermediate and/or long-term outlook for the US economy, housing, entitlement spending, tax revenues, and a host of other variables, a Major input to analysts’ forecasts are projections of the US population – not, of course, just the totals, but the composition of the population, especially by age. Analysts often use “official” Census projections, which include detailed projections by age, gender, and race/ethnicity, to project other key variables such as labor force growth, household formations, tax revenues, etc. Unfortunately, the last official Census population projection available to the public was released in late 2014 (and its “starting point” was the “vintage 2013” population estimate), and since then (1) “actual” population growth has been significantly lower than these projections, and (2) the outlook for population growth has changed drastically. As such, the latest available “official” Census population projections are not very useful to analysts or policymakers, and any economic/housing/other forecasts based on these outdated population projections is of little use as well.

One of the major reasons the 2014 Census population projections are of limited value relates to (1) the “unusually” high projections for net international migration (which were way above the latest estimates for the last few years) ; and (2) the implications of Donald Trump’s election for net international migration over the next several years.

On the first point, in its 2014 population projection Census assumed that net international migration from 2014 to 2016 would average 1.241 million a year, a sizable increase from the annual average of about 876,000 in the previous three years. According to the latest estimates (“vintage 2016,” released late last year), net international migration averaged 1.004 million from 2014 to 2016. (The vintage 2016 estimates incorporated an improved methodology for estimating foreign-born emigration, which reduced previous estimates of net international immigration. (See LEHC, 2/7/2017)). As such, the cumulative net international migration assumption in Census’ 2014 projection for the 2014 to 2016 period was about 710,000 higher than the latest estimate for that period.

Click on graph for larger image.

On the second point, in the 2014 Census population projections the average annual net international migration assumption from 2017 to 2020 was 1.264 million. With the election of Donald Trump, however, virtually no competent analysts believe that such a projection is plausible. While it is extremely difficult to translate various Trump officials’ statements into actual projections of net international migration, it seems to safe to safe that a “best guess” projection from now through 2020 would be massively lower than the Census 2014 projections.

Another reason (not nearly as important) why one might think that the Census 2014 projections need to be updated related to its assumptions on the number of deaths in the US. In the 2014 projections, the projected number of deaths per year was 2.619 million in 2015 and 2.650 million in 2016. The most recent estimates are than deaths in the US totaled 2.687 million in 2015 and 2.746 million. The 2016 jump surprised and worried many analysts, and folks are still “digging into” the numbers, However, there is good reason to believe that the 2014 Census projection for the number of deaths from 2017 to 2020 (death projections increase each year, mainly reflecting the aging population) are too low.

Finally, the Census 2014 projections for the number of births was 3.999 million in 2015 and 4.027 million in 2016, what the latest estimates for these years are 3.983 million and 3.978 million, respectively. (These differences have virtually no impact on projections of, say, the labor force or the number of households through 2020.)

Below is a table showing Census 2014 projections for 2016 (July 1), compared to the “vintage 2016” estimates.

July 1, 2016 US Resident Population, Current Estimate vs. Census 2014 Projection
	Current Estimate	Census 2014 Projection	Difference
Total	323,127,513	323,995,528	-868,015
18 Years or older	249,485,228	250,293,421	-808,193

Census has not yet released “vintage 2016” estimates on the detailed characteristics of the population (including more detailed age data, as well as revisions for previous years.) Attempting to estimate the vintage 2016 age distribution of the population is actually quite challenging, because there were downward revisions to previous years, mainly reflecting downward revisions in net international migration. Using Census based assumptions on the age distributions of immigrants, as well as using Census based on assumptions of death rates by age (adjusted to equal newly revised estimates, I have attempted to estimate the “vintage” 2016 population estimates by age. My estimates, compared to Census 2014 projections for 2016, are shown below.

July 1 2016 US Resident Population by Selected Age Groups (000's)
	Based on Vintage 2016*	Census 2014 Projections	Difference
Total	323,128	323,996	-868
0-14	60,963	61,037	-74
15-24	43,468	43,613	-145
25-34	44,630	44,865	-235
35-44	40,512	40,578	-66
45-54	42,840	42,864	-24
55-64	41,519	41,619	-100
65-74	28,635	28,747	-112
75+	20,560	20,673	-113
*LEHC Estimate. Totals may not add up due to rounding

Looking ahead to 2020, the net international migration assumptions imbedded in the Census 2014 US population projections through 2020 are unreasonably high. Trying to gauge what more “realistic” assumptions under a Trump administration is, unfortunately, a rather difficult exercise. E.g., some statements on potential actions, if taken at face value and implemented, would effectively imply net international out-migration over the next few years!

Rather than attempting to make specific forecasts about net international migration through 2020, I am instead show what population projections by age would look like under two different scenarios, shown below, and compare them to projections from the Census 2014 projections. (Note: each year’s forecast is from July 1st of the previous year to June 30th of the following year, so 2017 only has three months left).

Net Immigration: Three Scenarios
	Census 2014 Projection	Sorta Trumpy	Really Trumpy
2017	1,256,000	850,000	750,000
2018	1,262,000	600,000	0
2019	1,267,000	600,000	0
2020	1,271,000	600,000	0

Below are estimates of the US resident population by selected age groups for each scenario. (Note: The starting point for the “Trumpy”scenarios are based on Vintage 2016 estimates, while the “Census 2014 Projection” starting point reflects that release’s projection for 2016. Also, the “Trumpy” scenarios use updated estimates for annual deaths and births).

US Resident Population by Age Group Under Various Scenarios, 2016-2020 (million's)
		Total	0-14	15-24	25-34	35-44	45-54	55-64	65-74	75+
Census 2014	7/2016	324.0	61.0	43.6	44.9	40.6	42.9	41.6	28.7	20.7
	7/2020	334.5	61.6	43.1	46.9	42.6	40.8	43.0	33.1	23.4
	Dif	10.5	0.5	-0.5	2.0	2.1	-2.0	1.4	4.3	2.7
Sorta Trumpy	7/2016	323.1	61.0	43.5	44.6	40.5	42.8	41.5	28.6	20.6
	7/2020	330.3	60.7	42.5	45.9	42.2	40.6	42.7	32.8	23.1
	Dif	7.2	-0.2	-1.0	1.3	1.6	-2.3	1.2	4.1	2.5
Really Trumpy	7/2016	323.1	61.0	43.5	44.6	40.5	42.8	41.5	28.6	20.6
	7/2020	328.5	60.5	41.9	45.4	41.9	40.4	42.6	32.7	23.1
	Dif	5.3	-0.5	-1.5	0.7	1.4	-2.4	1.1	4.0	2.5

Assuming no material change either in labor force participation rates of headship rates by age, here is what annual growth rates in select measures would look like under these scenarios.

	Population Growth	Labor Force Growth	Household Growth
Census 2014	0.80%	0.51%	1.19%
Sorta Trumpy	0.55%	0.26%	1.00%
Really Trumpy	0.41%	0.09%	0.86%

Obviously, a projection for overall economic activity under the “really Trumpy” scenario would probably be significantly slower than projections based using Census 2014 projections, unless either (1) labor force participation rates increased significantly, or (2) productivity growth increased materially. For those who prefer numerical rather than % change numbers for households, a “unchanged” headship rate projection using Census 2014 population projections translates into about 1.48 million net household formations per year from 2016 to 2010, while the “Really Trumpy” scenarios translates into about 1.06 net household formations per year over than same period.

Of course, these “Trumpy” projections are based on “seat of the pants” projections of net international migration, combined with incorporation of updated “actuals” for 2016 as well as updated projections for births and death, and different net international migration projections would produce different population/labor forece/household projections. For entities who produce such forecasts and cite them for various purposes and whose latest forecast was based on the Census 2014 population projection report, however, they should either (1) send out a note that their latest forecasts are no longer valid, or (2) develop new projections based on more realistic assumptions.