Today: what typical households actually own, a slump in apartment building, where tax cuts pay off, why active managers play it safe, and imported ideas behind US invention.
Picked from 26 new articles across 16 sources · arrow keys for other days
Today in 30 seconds
30s left
The Fed's 2025 survey puts median US household net worth at $216,100, but only $69,200 once home equity is taken out. story 1 ↓
US housing starts fell 5.2% in September. Multi-family starts were down 24.9% from a year earlier, while single-family starts rose 2.6%. story 2 ↓
A study of 200+ countries finds corporate tax cuts pay for themselves in Canada, roughly break even in the UK, and don't in the US. story 3 ↓
Twenty years of Italian pension data show funds fire managers who lag but don't reward ones who beat the market, so managers hug benchmarks. story 4 ↓
In a study of 13 million patents, 54% of US technologies build on at least one foreign idea. story 5 ↓
Also worth reading: research suggests recessions speed up automation, so an AI investment bust could hurt workers twice.
1/5Of Dollars and Data· Nick Maggiulli· Oct 20
US household net worth by age, from the Fed's 2025 survey data
Nick Maggiulli worked through the Federal Reserve's 2025 survey of household wealth. Average net worth is $1.23 million, but the median is $216,100. Wealth rises until the early 60s, then levels off.
Why it matters · If you want to know how your savings compare with people your age, use medians and percentiles. Averages are inflated by the very rich, and much typical wealth is tied up in a house.
Know this first
Context from the editor, not the article · 32 sec
1 · Survey of Consumer Finances
The Federal Reserve runs this detailed survey of US household balance sheets every three years. It deliberately over-samples wealthy households, which makes it the standard source for measuring how wealth is distributed. The 2025 edition follows the one from 2022.
2 · Percentiles, not averages
In wealth data, a few very rich households pull the average far above what is typical. A percentile ranks households instead: the 25th percentile has more than a quarter of households and less than the rest. The 50th percentile is the median.
3 · Snapshot, not life story
A one-time survey compares different people of different ages in the same year. It does not follow any one household as it gets older. So the pattern across age groups only suggests how wealth changes over a lifetime; it doesn't prove it.
2/5Calculated Risk· CalculatedRisk by Bill McBride· Oct 20
US housing starts fell 5.2% in September as apartment building slumped
Census Bureau data show US builders began homes at a 1.221 million annual pace in September, 7.4% below a year earlier. The drop came mostly from apartments. Single-family building held roughly steady.
Why it matters · How many homes get started now decides housing supply a year or two from now, so it affects future rents, home prices and construction jobs.
Know this first
Context from the editor, not the article · 27 sec
1 · Permits, starts, completions
A home first gets a building permit, then a start when the foundation is dug, then a completion. The Census Bureau publishes all three every month. Permits tend to run ahead of starts, so they hint at what's coming.
2 · Seasonally adjusted annual rate
Building swings with the seasons. The Census Bureau strips out the normal seasonal pattern, then scales one month's pace up to a full year. "1.221 million" means September's pace, not homes actually started.
3 · Single- vs multi-family
Single-family means detached houses, mostly built to sell. Multi-family means buildings with two or more units, mostly rental apartments. The two respond to different financing conditions, so they often move in opposite directions.
3/5Klement on Investing· Joachim Klement· Oct 20
How much growth a corporate tax change costs differs widely by country
Joachim Klement highlights a 2026 study by Agostini and coauthors that estimates how strongly each country's economy reacts to corporate tax changes. In most countries, including the US, cutting corporate taxes raises the deficit.
Why it matters · Claims that tax cuts "pay for themselves" come up in every budget fight. This gives a country-by-country way to test them instead of one universal rule.
Know this first
Context from the editor, not the article · 26 sec
1 · What a fiscal multiplier is
A fiscal multiplier measures how much a country's economic output (GDP) changes when the government changes taxes or spending. A multiplier of 0.5 means output moves 50 cents for each dollar of fiscal change.
2 · The Laffer Curve idea
The Laffer Curve says that past some tax rate, raising taxes shrinks the economy so much that revenue falls. Cutting taxes could then raise revenue. Supporters of tax cuts often call this "trickle-down" economics.
3 · The UK budget
The UK government sets out its tax and spending plans in an annual autumn budget. With deficits under pressure, people expect tax rises. Klement notes corporate tax looks off the table this time.
4/5Klement on Investing· Joachim Klement· Oct 21
Pension funds fire underperforming managers but don't reward outperformance
A study of Italian pension funds from 2003 to 2023 found that lagging the benchmark gets a manager fired fast, while beating it doesn't protect them. Managers respond by taking less risk and copying the benchmark.
Why it matters · If you pay for active management through a pension or fund, the people you hire have reasons to quietly behave like an index fund while charging you active fees.
Know this first
Context from the editor, not the article · 28 sec
1 · What a benchmark does
Active managers are judged against a benchmark, usually a market index they are meant to beat. Their performance is reported as the return above or below that index, not as the raw return.
2 · Career risk
Career risk is the danger to a professional's job, as opposed to the danger to the client's money. A manager can make a sound bet and still get fired if it looks bad for a year or two.
3 · Why studying it is hard
Public mutual funds have thousands of investors, marketing effects and fund flows, so it's hard to tell why a manager loses a mandate. A setup with one manager and one client removes most of that noise.
5/5Marginal Revolution· Tyler Cowen· Oct 21
Study of 13 million patents finds foreign ideas drive much US invention
Economists traced which older technologies show up in 13 million US patents filed since 1887. They found foreign ideas feed most American innovation, and that local immigrant ancestry steers which ideas each county absorbs and builds on.
Why it matters · Debates over trade, immigration and research funding often treat innovation as a domestic product. This study puts a number on how much US growth depends on ideas from abroad.
Know this first
Context from the editor, not the article · 28 sec
1 · Innovation as recombination
Economists often model new technology as fresh combinations of existing ideas. Under that view, the stock of ideas a place can reach, including ideas from abroad, limits what it can invent next.
2 · Correlation versus causation
Places that invent a lot may also draw foreign ideas and immigrants, so a simple correlation can't show what causes what. Researchers use an "instrument": something that shifts exposure to foreign ideas but has no direct effect on invention.
3 · Historical settlement as instrument
Earlier work co-authored by Tarek Hassan used past migration waves to predict where ancestry groups ended up in the US. Where migrants settled depended partly on the timing of arrivals and railroad expansion, not on future local innovation.
Median net worth by age group, 2025: far lower than the averages, peaking at older ages.1 / 4
The Federal Reserve has released its 2025 1, its survey of household assets and debts, run every three years. Nick Maggiulli of Of Dollars and Data has broken the results down by age. Net worth here means everything a household owns minus everything it owes, adjusted for inflation into 2025 dollars.
Average versus 2
Average household 3 is $1,231,997, up 6% after inflation since 2022. Maggiulli says that number badly overstates what is typical, because the very rich pull it up. The median household, the one in the exact middle, has $216,100. That is under a fifth of the average, and up only 2% since 2022. The average rising three times faster than the median means the biggest gains went to the top.
By age
Median net worth rises from $13,050 for households under 25 to $181,030 at ages 40–44 and $528,960 at ages 60–64. After that it levels off: $424,400 at ages 70–74 and $567,600 for those 80 and older. Two cautions apply. Each five-year group has only a few hundred surveyed households, so small bumps are noise. And the high figure for the 80+ group partly reflects 4: wealthier people live longer, so the people who remain look richer.
The survey compares different people at one point in time; it doesn't follow anyone as they age. Maggiulli still thinks the flat pattern after retirement is roughly right, because most retirees don't spend down their savings.
The spread
Gaps between households widen sharply with age. For ages 45–54, the 25th 5 holds $54,730, the median $254,670, the 75th percentile $941,970 and the 90th percentile $2.56 million. Within that group, the 80th percentile holds about $1.1 million, so moving up 10 points to the 90th percentile more than doubles net worth.
The house
Remove 6 (your home's value minus its mortgage) and the overall median falls to $69,200. Maggiulli measured the home's share directly. For middle households aged 35 and older, it makes up 55% to 70% of net worth. For the top 20% of each age group, it is 15% to 25%. In short, the less wealthy you are, the more of your wealth is in your house.
One caveat: the two medians come from separate rankings. A 40-year-old renter with $150,000 in savings is near the median on total wealth but at the 68th percentile once homes are excluded. So use one benchmark at a time, and don't read the gap between them as the typical household's home equity.
Maggiulli argues that home equity is real wealth. You can sell it, borrow against it or pass it on, and it saves rent. But he would not count it as spendable.
The Federal Reserve has released its 2025 1, its survey of household assets and debts, run every three years. Nick Maggiulli of the blog Of Dollars and Data has broken it down by age, 2 and home ownership. He measures 3 as everything a household owns (home, car, bank and brokerage accounts, rental property) minus everything it owes (mortgage, student loans, credit cards). All figures are adjusted for inflation into 2025 dollars.
The average is misleading
Average US household net worth in 2025 was $1,231,997. After inflation that is up $72,550, or 6%, from $1,159,447 in 2022. Maggiulli points out that the typical household is nowhere near a million dollars. He illustrates the problem with an example: nine people worth $100,000 each are in a bar when Warren Buffett, worth over $100 billion, walks in. The bar's average net worth jumps past $10 billion, which tells you nothing about the other nine.
The 4 is the household in the exact middle, with half above and half below. It was $216,100, less than a fifth of the average. That is up only about $5,200 (2%) from $210,887 in 2022. Because the average rose three times as fast as the median, the biggest gains went to the top.
Age is the variable that matters
Maggiulli argues that you can only compare net worth fairly within an age group. Older people have worked and saved for longer, while younger people have more time left to build wealth. By five-year age group:
Age
Average
Median
Under 25
$43,566
$13,050
30–34
$201,398
$64,300
40–44
$757,910
$181,030
50–54
$1,499,654
$290,000
55–59
$1,608,679
$252,020
60–64
$2,097,867
$528,960
70–74
$1,930,205
$424,400
80+
$2,084,534
$567,600
Wealth climbs steadily until the early 60s, around retirement, and then levels off. Maggiulli warns against reading too much into single rows. The dip at 55–59 and the jump at 60–64 are probably statistical noise, because each five-year group contains only a few hundred surveyed households.
The 80+ group having the highest median is partly 5: wealthier people tend to live longer, so as poorer households die off, the people who remain look richer. There is a deeper limit too. The SCF photographs different households at a single moment. It does not follow today's 65-year-olds as they age. Maggiulli concedes that, strictly, the data can't show that retirees keep their wealth. He thinks the flat pattern is roughly right anyway, because most retirees don't actually spend down their savings.
The spread within each age group
Grouped into 10-year bands, the percentiles look like this:
Age
25th
Median
75th
90th
<35
$3,100
$32,820
$145,090
$351,200
35–44
$18,000
$156,000
$467,520
$1,233,620
45–54
$54,730
$254,670
$941,970
$2,557,390
55–64
$71,790
$411,340
$1,416,200
$3,457,640
65–74
$118,105
$431,900
$1,312,100
$3,656,000
75+
$144,060
$504,600
$1,417,000
$4,188,260
Two things stand out. First, even the 25th-percentile household has positive net worth at every age, although under 35 it is only $3,100. Second, the numbers stretch out sharply at the top. Among households aged 45–54, the 80th percentile holds about $1.1 million, while the 90th holds $2.6 million, more than double. Between ages 35–44 and 55–64, the 90th percentile gains about $2.2 million and the median gains about $255,000. Maggiulli reminds readers that pay, inheritance and luck explain a lot of where people land.
Taking out the house
Readers ask Maggiulli one question more than any other: what happens if you remove 6? Home equity here means the value of your main home minus the mortgage on it. Other real estate still counts. Without home equity, the overall median falls from $216,100 to $69,200.
He adds a caveat that is easy to miss. The two medians come from two separate rankings, so the household at the middle of one list is not necessarily in the middle of the other. His example: a 40-year-old renter with $150,000 in savings sits near the median for total net worth. But that same renter ranks at the 68th percentile once home equity is excluded, because many homeowners drop down the list. So the gap between the two medians does not tell you how much home equity a typical household has. Pick the benchmark that matches what you are measuring and use only that one.
To measure the home's share directly, Maggiulli ranked households by total net worth and calculated what fraction of each household's wealth was home equity. For middle households (the 40th to 60th percentile) aged 35 and older, the typical share is 55% to 70%. For the top 20% of each age group it is only 15% to 25%. For households under 35 the typical share is 0%, because most of them don't own a home yet. In short, the less wealthy you are, the larger the share of your wealth that sits in your house. Maggiulli thinks that explains the question: removing home equity removes most of a typical household's wealth.
Is home equity real wealth?
Maggiulli argues that it is. You can sell the house, borrow against it, or leave it to your children, and owning means you don't pay rent. Still, he would not count it as spendable wealth. A house is much harder to turn into cash than a brokerage account, and you can't spend the equity without taking on debt. His position is to count it, but not to treat it as money you can spend.
What the article leaves open
The piece benchmarks wealth; it doesn't explain it. It doesn't break down how much of the gap between percentiles comes from income, inheritance or investment returns. It also doesn't say why the average grew three times faster than the median between 2022 and 2025. Maggiulli flags the noise in small age groups and the snapshot problem himself. Treat any single row as a rough guide, not a precise target.
Remember
Average net worth grew 6% after inflation from 2022 to 2025, the median only 2%: the biggest gains went to the top.
For middle-wealth households aged 35 and up, home equity is 55–70% of net worth; for the top 20%, just 15–25%.
Compare yourself to one ranking at a time: with home equity or without it, never the gap between them.
Single- and multi-family starts since 2000, showing the bubble peak and recent apartment decline.1 / 3
The number
US builders began privately owned homes at a 1.221 million1 in September 2026, according to the 2. That means September's pace, with normal seasonal swings removed and scaled up to a full year. It was 5.2% below August's revised 1.288 million and 7.4% below September 2025.
Bill McBride of Calculated Risk calls the result "well below expectations." He notes that July and August were revised up, though, so the fall came from a higher base than first reported. For the year so far, starts are down only 0.5% from 2025, using raw counts that aren't seasonally adjusted.
Apartments drive the drop
The total splits into two trends:
Single-family starts (detached houses) were 859,000, down 4.6% from August but up 2.6% from a year earlier.
3 (buildings with two or more units, mostly apartments) were about 362,000, down 24.9% from a year earlier.
Rough arithmetic shows the annual decline is almost all apartments. Single-family added about 22,000 units over the year while the total fell by 98,000, so multi-family fell by about 120,000 units. The article doesn't explain the cause.
Permits point to what comes next
4 usually come before starts. They ran at a 1.400 million annual rate, nearly flat from August and 3.0% below a year earlier. Single-family permits fell 4.0% to 845,000, slightly below single-family starts, which hints at some softening ahead. Permits for buildings with five or more units were 503,000, well above the 349,000 starts in that category. That suggests approved apartment projects that haven't broken ground yet. This reading is arithmetic on the reported figures, not a forecast from the article.
Caveats
Monthly starts are noisy and often revised, so the flat year-to-date figure is a steadier signal than one month's drop. McBride says units 5 remain "slightly elevated," meaning a fair amount of supply is still in the pipeline. The details are behind his paywall. Because apartments take a year or more to build, today's drop in multi-family starts means fewer new rentals later.
The headline number
According to the 1's monthly report on permits, starts and completions, US builders began privately owned homes at a 2 of 1,221,000 in September 2026. A seasonally adjusted annual rate takes one month's activity, removes the normal seasonal pattern, and scales it up to a yearly pace. September's pace was:
5.2% below August's revised 1,288,000
7.4% below September 2025's 1,319,000
Bill McBride of Calculated Risk says the number was "well below expectations." He also points out that starts for July and August were revised up, so part of September's drop is a fall from a higher base than first reported. Measured over the whole year rather than one month, the picture is less dramatic. Starts so far in 2026 are down just 0.5% from the same period of 2025. That figure is not seasonally adjusted (3, meaning raw counts), and over a full year-to-date comparison seasonal effects largely cancel out.
Where the decline is coming from
The total breaks into two very different trends.
Single-family starts ran at 859,000, down 4.6% from August's revised 900,000 but up 2.6% from a year earlier. Detached-house building has dipped from month to month but is roughly flat to slightly higher over the year.
4 (buildings with two or more units) are what the total minus single-family leaves: about 362,000. McBride reports they were down 24.9% year over year. Buildings with five or more units, the large apartment projects, started at a 349,000 rate.
So the 7.4% annual drop in total starts comes almost entirely from apartments. Here is a rough check using the reported figures: single-family rose about 2.6% from about 837,000 to 859,000, a gain of roughly 22,000. Total starts fell by 98,000 (1,319,000 to 1,221,000). Multi-family must therefore have fallen by about 120,000 units on an annual-rate basis, from roughly 482,000 to 362,000. That is consistent with the 24.9% decline.
The article doesn't say why apartment starts are falling. Its long-run charts show that multi-family building has historically moved in large cycles, separate from single-family.
Permits: a look ahead
5 matter because they usually come before starts. In September, permits ran at a 1,400,000 annual rate:
0.2% below August's revised 1,403,000
3.0% below September 2025's 1,444,000
Single-family permits were 845,000, down 4.0% from 880,000 in August. Permits for buildings with five or more units were 503,000.
The two series are worth putting side by side, with a caution: not every permit becomes a start, and the timing varies. Total permits (1.400 million) are well above total starts (1.221 million). The gap is concentrated in large buildings, where permits ran at 503,000 against 349,000 starts. Single-family permits (845,000) sit slightly below single-family starts (859,000). That fits a single-family pipeline that is softening a little, and a multi-family segment where projects are approved but not yet breaking ground. The article doesn't draw this conclusion itself; it is simple arithmetic on its figures and should be treated as a hint, not a forecast.
Under construction
McBride titles one section "Housing Units Under Construction Remains Slightly Elevated." Homes 6 are units started but not yet completed. A high count means a lot of supply is still on its way to market, whatever starts are doing now. The detail and chart for this section are behind the newsletter's paywall, so the size of the "slightly elevated" level isn't available here.
How to read this
Three points help put a single month in context:
Monthly starts are noisy and get revised. The upward revisions to July and August show how much one month's estimate can move. The flat year-to-date figure (−0.5%) is a steadier gauge than the 5.2% monthly drop.
The segments tell different stories. Single-family is roughly holding up. Multi-family is down by about a quarter from a year ago. The aggregate hides that split.
Today's starts are future supply. An apartment building can take a year or more to finish. A sharp fall in multi-family starts now means fewer new rental units arriving later, and that matters for rent trends further out. The article doesn't quantify that effect.
What's left open
The free portion of the post doesn't cover completions, regional breakdowns, or the full under-construction data. It also doesn't say what forecasters expected, beyond calling the result "well below expectations."
Remember
Total starts fell 7.4% from a year earlier, but year-to-date starts are down only 0.5%.
Single-family starts were up 2.6% year over year; multi-family starts were down 24.9%.
Permits for large buildings (503,000) far exceed starts (349,000), suggesting approved apartment projects that haven't started yet.
The idea that past some rate, higher taxes shrink the economy enough to reduce total tax revenue.
Estimated multipliers for corporate tax changes by country, compared with the 1.0 break-even line (Agostini et al., 2026).
The UK announces its budget next week, and tax rises are widely expected. Corporate tax reportedly isn't changing this time. Joachim Klement still uses the moment to highlight a 2026 study by Agostini and coauthors. It estimates, for more than 200 countries, how much growth a change in corporate tax rates gains or loses.
The question
When a government cuts a tax, does the faster growth that follows bring in enough revenue to cover what the lower rate gives up? Economists describe this with a 1, which measures how much economic output changes for each unit of tax change. Klement's rule: above 1.0, a corporate tax cut pays for itself. Below 1.0, it widens the deficit, and a tax rise shrinks it.
This is hard to measure. Tax changes come bundled with recessions, spending decisions and central-bank moves, and studies of single countries are hard to compare.
What the study did
The authors reviewed existing estimates from many countries. They then built one model that predicts a multiplier for every country from a set of country characteristics. Klement uses the 2 version, and the paper also offers simpler 3- and 4-parameter ones. The post doesn't list the parameters or give exact values.
What it found
US, France, Italy, Spain: below 1.0. Cuts add some growth but raise the deficit. Rises reduce it.
UK: near 1.0. A corporate tax rise would cost roughly as much growth-driven revenue as it brings in, so it barely moves the deficit.
Canada: well above 1.0. A cut could, by this estimate, raise revenue overall.
Global median: clearly below 1.0. In most countries, corporate tax cuts don't pay for themselves.
Klement argues that 40 years of US tax cuts fit this pattern: growth picked up each time, but deficits grew.
Caveats
The 1.0 line is a shortcut. Under the textbook definition, the multiplier counts GDP, and governments collect only part of any extra GDP as tax, often 30–40% (illustrative). On that basis a cut pays for itself only at a much higher multiplier. The paper may define the multiplier in revenue terms instead, but the post doesn't say.
Corporate tax only. Klement warns that personal income tax multipliers probably differ a lot by country.
Model estimates. Most country figures are predictions from the model, not direct measurements.
His bottom line: whether tax cuts pay for themselves depends on the country, and in most countries they don't.
The UK announces its budget in about a week, and Joachim Klement expects the debate to be about which taxes go up, not down. Corporate tax reportedly isn't being changed this time. Klement still uses the moment to point to a 2026 study by Agostini and coauthors. It estimates, for more than 200 countries, how much growth a change in corporate tax rates gains or loses. The headline finding is that the answer varies a lot from country to country.
The question being asked
Every argument about tax cuts comes down to one question: when the government cuts a tax rate, does the faster growth that follows bring in enough extra revenue to cover what the lower rate gives up? If yes, the cut pays for itself. If no, it widens the deficit. Tax rises work the same way in reverse. A rise reduces the deficit only if the growth it costs doesn't eat up the new revenue.
Economists describe the size of this effect with a 1: how much economic output changes for each unit of tax change. Klement uses a simple rule. Above 1.0, a corporate tax cut produces enough growth to pay for itself. Below 1.0, it doesn't.
This is hard to measure because tax changes rarely happen in isolation. They come with recessions, elections, spending changes and central-bank moves, and each country's economy, openness and tax base are different. Single-country studies give answers that vary widely and are hard to compare.
What the study did
According to Klement, Agostini et al. first reviewed existing estimates of corporate-tax multipliers from many countries. They then built a model on that evidence to predict multipliers for every country within one framework. That makes it possible to compare countries that have never been studied directly. The paper reports versions of the model with 3, 4 and 5 parameters, meaning the number of country characteristics used to predict the multiplier. Klement's chart uses the 2 version. The post doesn't say what the parameters are.
What the estimates show
The post gives no exact numbers, only where countries fall relative to 1.0:
United States, France, Italy, Spain: below 1.0. A corporate tax cut adds a bit of growth but not enough to cover the lost revenue, so the deficit grows. A rise reduces the deficit and makes debt more sustainable.
United Kingdom: close to 1.0. The growth cost of a rise roughly cancels its extra revenue. In Klement's reading, raising UK corporate tax wouldn't shrink the deficit, and cutting it wouldn't widen it much.
Canada: well above 1.0. Here a cut could, by this estimate, raise total revenue and reduce the deficit.
Global median: clearly below 1.0. In most countries, corporate tax cuts don't pay for themselves.
Klement also gives his own reading of US history. He argues that across 40 years of US tax cuts, growth picked up each time but never enough, and deficits grew. That fits a multiplier below one. He warns that a country that keeps cutting under those conditions ends up with deficits and debt "worse than Italy."
Caveats worth keeping in mind
The 1.0 threshold is a simplification. Under the textbook definition, the multiplier is the change in GDP per unit of tax change. A government collects only part of any extra GDP as tax, often 30–40% (illustrative). So a multiplier of exactly 1 would recover only about that share of the cost of a cut, not all of it. Under that definition, a cut pays for itself only when the multiplier is far higher, roughly 1 divided by the 3. The paper may define its multiplier in revenue terms, which would make 1.0 the right line. The post doesn't say, so it's worth checking the paper before relying on the break-even claims for the UK or Canada.
Corporate tax only. Klement stresses that multipliers for personal income tax are probably very different in each country. These results don't tell you what an income tax change would do.
These are model predictions. Most of the 200-plus figures come from a fitted model, not from measuring each country directly. How reliable they are depends on how well the five parameters capture what differs between economies.
Growth versus revenue. Even where a rise "costs" growth, Klement presents it as worth it in below-1 countries because it makes debt more sustainable. Whether that trade is worth making is a policy judgment the numbers alone don't settle.
The takeaway
Klement's main point is that "tax cuts pay for themselves" is neither always true nor always false. It depends on the country. The study offers a way to check it country by country, and by its estimates the claim fails in most places, including the US.
Remember
Most corporate-tax multipliers are below 1, so cuts usually add some growth but still widen deficits.
The same tax change can pay for itself in Canada and fail in the US.
Check how the multiplier is defined before trusting any break-even claim about tax cuts.
Termination probability against performance vs the benchmark: it rises with underperformance and doesn't fall with outperformance.
A new study by Giacomo Nocero and Lucia Spotorno (2026), highlighted by investment strategist Joachim Klement, measures something industry insiders take for granted: fund managers worry more about keeping their jobs than about maximising returns.
Why this is hard to measure
Career risk is the danger to a professional's job, as distinct from the danger to the client's money. It's difficult to isolate in public funds, where investor flows, marketing and thousands of separate clients all affect whether a manager keeps the money.
The authors used 1 instead. An SMA is a portfolio run by an outside manager for a single client under a set of instructions called a 2. Here the clients are Italian workplace pension funds, so each case is one manager, one client and one objective. SMA terms are private, so the authors hand-collected the data, covering 2003 to 2023.
The finding: heads I win, tails you lose
The main reason a manager got fired was performance against the 3, the index the manager is meant to beat. Underperformance raised the chance of termination significantly and quickly. Outperformance did nothing to reduce it. Klement sums up the deal as "heads, I win, tails, you lose."
In that setup, a bet that might win or lose against the index has an upside that doesn't help the manager's job and a downside that can end it.
How managers adapt
Lower 4. Beta measures how much a portfolio moves with the market. Managers kept it below 1.0. As an illustration, a beta of 0.8 means losing about 16% when the market drops 20%, but also gaining only about 16% when it rises 20%.
Less 5. Active risk is how far returns stray from the benchmark. Managers kept it low, so pension funds pay active fees for portfolios Klement calls "semi-active."
More caution when mandates are split. When one mandate is shared among several managers, each is easier to replace, and the defensive behaviour gets stronger.
Performance fees don't help
A bonus for beating the benchmark doesn't change this behaviour. The fee is uncertain and the threat of being fired is not. Because managers play defence, 6 end up a small share of what they're paid.
Klement's summary doesn't give the sample size, the size of the effects, or whether the results hold outside Italy.
Joachim Klement, who spent most of his career on the 1 (the firms that invest money, as opposed to the banks that sell to them), highlights a new paper by Giacomo Nocero and Lucia Spotorno (2026). It measures how career risk shapes the way professional fund managers invest. Their conclusion, in Klement's words, is that the arrangement pension funds offer managers is "heads, I win, tails, you lose", and managers adjust their portfolios to match.
The problem: separating career risk from everything else
Most fund managers are judged by how they do against a 2 index. Klement says from experience that worrying about keeping the job often outweighs pure investment judgment. That is hard to prove. With public investment funds, many other things affect whether a manager keeps the money: investors moving cash in and out, marketing, fund mergers, and the behaviour of thousands of separate clients.
The authors avoid this by studying 3. An SMA is a portfolio run by an outside manager for a single client under a written 4, a set of instructions covering what to invest in and how. Here the clients are Italian occupational pension funds, meaning workplace retirement schemes. Each relationship involves one manager, one client and one clear objective. When a pension fund replaces a manager or closes the SMA, the cause is much easier to identify.
The data has a weakness. SMA terms are private, so the authors hand-collected records covering 2003 to 2023. Klement says he usually distrusts the phrase "hand-collected dataset", but he accepts that nobody could study SMAs any other way.
The finding: an asymmetric firing rule
The main reason managers got terminated was performance against the benchmark, which is what you'd expect. The pattern was lopsided, though:
Underperformance raised the probability of termination significantly, and the effect showed up quickly.
Outperformance did not lower the probability of termination. Beating the benchmark bought no extra job security.
The paper's chart (below) plots termination probability against relative performance and shows this shape. Klement's post doesn't give the exact probabilities, so none are quoted here.
To see why this matters, here is an illustrative example that is not from the study. A manager could take a bet that beats the index by 3% or trails it by 3%, with even odds. For the client, the expected value of that bet may be positive. For the manager, the upside does nothing for their job and the downside could end it. A rational manager avoids the bet.
How managers respond
According to Klement's summary, the study finds managers protect themselves in three ways.
1. They lower 5. Beta measures how much a portfolio moves with the market. A beta of 1.0 moves one-for-one with the benchmark. Managers in the study ran portfolios with beta below 1.0. As an illustration, a beta of 0.8 means that when the market falls 20%, the portfolio falls about 16% and beats the benchmark in a bear market. The cost comes in rising markets: if the market gains 20%, the portfolio gains about 16%. Managers accept that trade because losses relative to the benchmark get them fired and gains don't save them.
2. They cut 6. Active risk, often measured as tracking error, is how far a portfolio's returns stray from its benchmark. Managers keep it low so they never fall far behind. Klement's point is that pension funds then pay active-management fees for portfolios that are only "semi-active", which leaves them close to an index fund at a higher price.
3. They get more defensive when the mandate is split. Some pension funds divide one mandate among several managers. In that case each manager is easier to replace, because the pension fund can hand the portfolio to a manager already doing the same job with little disruption. The study finds the defensive behaviour is stronger in these cases.
Why 7 don't fix it
A common answer to timid managers is a performance fee, an extra payment when the manager beats the benchmark. The study finds it changes nothing. The fee is uncertain, while the threat of being fired is concrete, so career risk wins. Because managers run their SMAs defensively and with little active risk, they rarely earn much in performance fees, and those fees end up as only a small share of what managers are paid in total.
What it means, and what's left open
The incentive problem sits with the client, not the manager. A pension fund that fires for underperformance and ignores outperformance is teaching managers to hug the benchmark, and it still pays active fees for that. Klement presents this as confirming what insiders already suspect.
The summary leaves several questions open. It doesn't say how many pension funds or mandates are in the sample. It doesn't give the size of the termination effect, or say how long underperformance has to last before a firing. It doesn't measure how much lower beta or active risk actually is. It doesn't say whether the results apply outside Italy or to other kinds of institutional client. It also doesn't say how pension funds could redesign their contracts, for example with longer evaluation periods or explicit credit for outperformance. A reader wanting those details would need the original paper.
Remember
When managers are punished for lagging but not rewarded for leading, the rational move is to copy the benchmark.
Clients can end up paying active-management fees for portfolios that behave much like a cheaper index fund.
Performance fees don't fix this, because the threat of being fired is certain and the bonus is not.
A new working paper by economists Tarek Alexander Hassan, Aakash Kalyani and Juan Cruz Lopez Del Valle argues that foreign ideas are a major, and causal, driver of American invention. Tyler Cowen, who flagged it on 1, calls it more evidence that "globalization is once again better than many people realize." His post quotes only the abstract, so details such as effect sizes aren't available yet.
The problem
Goods crossing borders get recorded. Ideas don't. Even if foreign know-how clearly shows up in US innovation, showing that it causes more innovation is hard. Inventive places also tend to attract immigrants and foreign contacts, so a correlation could run either way.
How they measured it
The authors used the full text of 13 million US patents filed from 1887 to 2024 to build a "2" of technologies. For each new technology they found its 3 and checked which earlier technologies those patents drew on. The result is an 4, a map of which technologies were combined to make which new ones. It also shows which of those inputs came from abroad.
What they found
Foreign ideas are pervasive. 54% of US technologies build on at least one foreign input. The most fertile US technologies, the ones that spawn the most descendants, rely on foreign inputs most heavily.
To rule out reverse causation, the authors use where migrant groups historically settled as an , a source of variation unrelated to a county's later inventiveness. Counties with more ancestry from a given country absorb substantially more of that country's technologies. They then invent significantly more US technologies that build on them.
Caveats
Patents are an imperfect proxy for innovation, since much invention is never patented. The causal claim depends on historical settlement patterns being unrelated to other lasting county traits. Ancestry could also carry ideas through several channels, such as family ties, trade or migrating inventors, and the abstract doesn't separate them. The policy reading on immigration and openness is Cowen's, not the abstract's.
Tyler Cowen at 1 highlighted a new working paper by economists Tarek Alexander Hassan, Aakash Kalyani and Juan Cruz Lopez Del Valle. It asks how much US innovation depends on foreign ideas. Their answer is a lot, and they say they can show the link is causal, not just a correlation. Cowen sums it up in one line: "Globalization is once again better than many people realize." The post quotes only the paper's abstract, so what follows covers the design and headline results. It can't cover the full set of estimates or robustness checks.
The problem: ideas leave few traces
Goods crossing borders show up in customs records. Ideas don't. A US inventor might draw on a German chemical process or a Japanese electronics design without anyone recording the transfer. Two questions are hard to answer for that reason: how much American invention rests on foreign work, and whether exposure to that work actually causes more invention. The second is harder. Innovative places tend to attract immigrants, capital and foreign contacts all at once. Any simple link between foreign exposure and patenting could run in either direction.
Building a family tree of technologies
The authors use the full text of 13 million US patents filed between 1887 and 2024 to build what they call a directed "2" of technologies. The idea is to identify each new technology, find its 3, and see which earlier technologies appear in them. If a new technology's founding patents rely on an earlier one, that earlier technology counts as an input. "Directed" means the links point one way, from parent to child.
Put together, this gives an 4 of technological recombination. The term borrows from economics, where input–output tables show which industries supply which others. Here the network shows which technologies were combined to make which new ones. It also shows where each input came from. That lets the authors flag the foreign technologies that fed US invention. The abstract doesn't say exactly how a technology's origin country is assigned.
Finding 1: foreign ideas are everywhere
54% of US technologies build on at least one foreign input. The most "fertile" US technologies, meaning those that go on to spawn the most later technologies, lean on foreign inputs the most. So foreign ideas aren't on the edges of American innovation. They are concentrated in its most productive branches.
Finding 2: the effect is causal
To get past correlation, the authors use the historical settlement of migrants as an 5. An instrument is a factor that changes exposure to foreign ideas but has no other route to affect local invention. The reasoning goes like this: where a given ancestry group settled long ago was shaped by when migrants arrived and where the country was growing at that moment. It wasn't shaped by which counties would later invent particular technologies. Hassan has used this kind of ancestry-based design in earlier work on foreign investment.
As an illustrative example, not one from the paper: take two similar counties, one settled heavily by Swedish migrants in the late 1800s and one not. The method predicts the first will take in more Swedish technologies, and then produce more new US technologies built on them. The paper reports both steps. Counties more exposed to knowledge from a given country absorb substantially more of that country's technologies, and then invent significantly more US technologies that build on them. The abstract gives no effect sizes, only the words "substantially" and "significantly."
Finding 3: foreign ideas steer direction, not just volume
US counties are far more likely to pioneer the specific technologies whose foreign inputs already sit in their 6. A county doesn't just innovate more. It innovates along lines set by the foreign ideas it already holds. The authors also link the changing origins of foreign ideas to a geographic shift in American innovation, away from its historical heartland and toward the coasts. The abstract doesn't name the countries or the timing behind that shift.
Finding 4: lasting effects on jobs
Decades later, counties that absorbed more of a given origin's ideas have more jobs working with those ideas. They have even more jobs working with the US technologies descended from them. That links idea flows to concrete labor-market results, not just patent counts.
What to weigh
Patents are only a proxy. Much innovation is never patented, and how often inventors patent varies by industry and era. A text-based genealogy picks up ideas that are written down and cited in language. Tacit know-how passed between people may be undercounted.
The instrument rests on an assumption. Historical settlement counts as random only if it is unrelated to other lasting county traits that also shape innovation. The abstract doesn't say how the authors test this.
Ancestry bundles several channels. Ancestry can carry ideas through family ties, trade, language or migrating inventors. The abstract doesn't separate these.
Policy implications are left to the reader. The results bear on immigration and openness to foreign research, but the abstract makes no policy claims. Cowen's framing is his own reading.
The full paper should have the effect sizes, the method for assigning a country of origin to each technology, and the account of the coastal shift. None of these appear in the excerpt.
Remember
The most productive US technologies, not marginal ones, are the ones that lean hardest on foreign inputs.
Immigrant ancestry shapes which technologies a county invents, not only how many.
The effects show up decades later in local jobs, not just in patent counts.
Ideas steer direction. Counties are far more likely to pioneer the specific technologies whose foreign inputs are already in their local stock of ideas. Changes in where foreign ideas came from accompany US innovation's shift from the historical heartland toward the coasts.
The effects reach jobs. Decades later, counties that absorbed more of a country's ideas have more jobs working with those ideas, and even more working with the US technologies descended from them.
