Cost of Saving

What happens when one part is starved

The check-up was tried on real history: public company filings, and US manufacturing industries. Part of it held, and part of it did not. Both are here.

In short

Public companies

On a phone, swipe the tables sideways to see every column.

18,791 company years from 3,052 companies, 2011 to 2023, from SEC filings. Income is revenue, running costs are selling, general and admin, and direct costs are cost of revenue. Each group is compared with balanced companies in the same year with similar revenue change, margin and size. Figures are points more (+) or less (-) often than those companies, with the range the result could reasonably fall in.

PositionYearsRevenue falls next yearRevenue falls 2 years outOperating loss next yearHeld?
Running costs starved; income and direct costs ahead2,161+8.0 (5.2 to 10.8)+12.6 (8.9 to 15.4)+5.4 (3.0 to 7.5)Yes
Both costs starved; income ahead1,274+7.6 (0.7 to 13.4)+15.1 (9.2 to 19.5)+8.8 (5.6 to 12.6)Yes
Direct costs starved; income and running costs ahead532+3.6 (-2.8 to 9.2)+5.4 (-1.3 to 10.6)+9.6 (5.4 to 14.9)Profit only
Income and running costs starved; direct costs ahead1,488-0.6 (-5.7 to 3.6)+3.9 (-1.4 to 7.5)+10.2 (6.9 to 14.0)Profit only
Income and direct costs starved; running costs ahead3,463-4.2 (-8.4 to -1.2)-4.9 (-8.0 to -1.0)+5.9 (3.3 to 8.7)Revenue better, profit worse
Income starved; both costs ahead474-10.6 (-18.2 to -3.3)-13.7 (-21.4 to -5.9)-1.8 (-9.3 to 5.5)No, the opposite

What counted as holding was set before the test ran: revenue falling more often, clearly above zero, in both 2011 to 2016 and 2017 to 2023. The raw rates of the seven positions had already been seen by then, though the rules did not change.

US manufacturing, with a real production figure

21,359 industry years from 364 industries, 1959 to 2017, from the NBER-CES database. Input is materials, work is production hours, output is shipments, with price changes taken out. Each group is compared with balanced industries in the same year with similar output change and size. The claim to test and what would count were set before any result was seen.

PositionYearsOutput falls next yearOutput falls 2 years outHeld?
Work starved; input and output ahead (the claim tested)1,344+4.5 (1.3 to 7.1)+6.6 (3.7 to 9.8)Yes
Input and work starved; output ahead745+7.7 (3.5 to 11.6)+5.2 (1.6 to 9.8)Yes
Work and output starved; input ahead1,028+1.8 (-1.0 to 4.9)+1.6 (-2.0 to 4.7)No clear change
Input starved; work and output ahead418+1.7 (-3.2 to 6.4)+1.8 (-3.1 to 6.3)No clear change
Input and output starved; work ahead790-0.3 (-3.9 to 3.0)+0.4 (-3.2 to 4.4)No clear change
Output starved; input and work ahead175-8.6 (-16.1 to -1.3)-7.1 (-15.3 to 0.4)No, the opposite

Both groups that held did so in both 1959 to 1988 and 1989 to 2017. Fewer hours can also mean better tools rather than too little work, and these figures cannot tell the two apart.

The three readings

Reading (public companies)YearsRevenue fell next yearRevenue grew next yearOverhead share up 10% or moreDebt ahead of revenue
Stable7,09820.4%58.1%13.9%23.0%
Stressed3,74828.1%57.6%29.3%28.8%
Unstable7,94536.2%52.1%26.5%31.3%

For ranking which companies would see revenue fall, the reading scored 0.593, where 0.5 is a coin toss. That beat "did they lose money this year?" at 0.546 and was about level with "how did revenue change this year?" at 0.601. So it is a modest signal. Its added value is saying which part is being starved, not out-predicting a simple trend.

Where it fell short, and what changed

What has not been tested

Free for anyone to use: no ads, no cookies, no sign-in and no personal data needed. Personal portfolio / skills demonstration: not a commercial product.
For potential employers and recruiters: see my online portfolio at butterflyfx.us for how it was made and the technologies and skills behind it.