The artifact, not the headline
Start with the document itself. Pengjie Gao, Chang Lee, and Dermot Murphy circulated “Financing Dies in Darkness? The Impact of Newspaper Closures on Public Finance” as a working paper dated July 11, 2018, hosted at Brookings. The paper’s abstract states its central result plainly: “Following a newspaper closure, we find municipal borrowing costs increase by 5 to 11 basis points in the long run.”
That sentence is the load-bearing beam. Everything downstream — the advocacy reports, the state-by-state dollar figures, the policy comparisons — rests on that estimate and on the identification strategy the authors built to defend it. If you want to audit the machinery of news economics, you audit this paper first, then you audit what was built on top of it.
What the paper actually measured
The authors examined municipal bond yields in counties that lost a local newspaper, using the municipal market because its bonds are largely bought and sold by local investors — a tighter link between local information shocks and securities prices than you get in most markets.
The headline numbers, in the authors’ terms:
- In the three years after a closure, secondary-market municipal bond yields rise 6.4 basis points and offering yields rise 5.5 basis points, significant at the 1% level.
- The effect is stronger for revenue bonds, which are backed by project cash flows and carry greater information asymmetry: secondary and offering yields rise 9.9 and 10.6 basis points.
Those are measured estimates from a specific sample, not a universal law. The paper’s own framing is that the effect is causal within its identification tests — not that every closure everywhere produces the same yield move.
The identification machinery, test by test
The reason this paper survives scrutiny is that the authors did not simply correlate closures with yields. They built three separate tests, each doing distinct work.
1. Matched neighboring counties
They compared the closure county to a population-matched neighboring county that still had its own newspaper. Declining regional economic conditions would hit both counties. The paper reports a differential effect on offering yields of 7.3 basis points (p = 0.012) between closure and control counties. The control county is the point: if the economy were driving both closures and yields, the neighbor would move too.
2. Low- versus high-newspaper counties
A closure in a county with many newspapers should barely change local coverage; a closure in a county left with one or zero should change it sharply. The paper finds a closure in a low-newspaper county raises offering yields 10.3 basis points, while a closure in a high-newspaper county has no significant effect — a relative difference of 11.8 basis points (p = 0.000). This test isolates coverage loss from closure itself.
3. The Craigslist instrument
This is the most auditable piece for a forensic reader. Craigslist’s geographic rollout is a discrete, datable platform event that damaged newspaper classified revenue without directly touching municipal bond yields. The paper reports that Craigslist entry raises the probability of a newspaper closure by about 10 percentage points, and that Craigslist-induced closures raise municipal bond yields by four to six basis points. A platform event, a closure probability, a yield move — three links you can trace.
The authors also report that the closure effect does not differ significantly between high- and low-internet-usage states. Read that narrowly: online sources did not substitute sufficiently in this sample. It is not a finding that digital outlets can never substitute.
The mechanism the paper reports
The proposed channel is government inefficiency: less monitoring, more waste, higher risk premium. The paper’s own evidence for that channel includes:
- A 1.3 percentage point increase in the government wage ratio after closure — for the median county, a total government wage increase of $1.4 million.
- About four additional government employees per 1,000 residents in the long run.
- About $85 more tax revenue per capita.
- Higher likelihoods of costly advance refundings and negotiated (rather than competitive) bond sales.
These are the paper’s findings about its sample. They are consistent with the monitoring story; they are not proof that every unmonitored government behaves this way.
Where the $1.1 billion comes from
In June 2026, Matthew Baker and Dermot Murphy published a Rebuild Local News report, “Local News Shortage Leads to $1.1 Billion in Extra Borrowing Costs for Local Governments and Taxpayers.” It is a derivation, not a second study. The arithmetic is disclosed in the report itself:
- Take the paper’s 5 basis point estimate for general obligation bonds and 11 basis points for revenue bonds.
- Weight them by the reported issuance mix — 40% GO, 60% revenue — to get 8.6 basis points: (40% × 5) + (60% × 11) = 8.6.
- Multiply state-level 2025 municipal issuance from the MSRB’s Municipal Market Facts report by the state’s news-desert population share, “under the assumption that bond issuance volume scales proportionally with population.”
- Sum to a national figure: $1.1 billion in annual additional borrowing costs.
The report is explicit that this is a lower-bound estimate and that it excludes other documented costs. It is also explicit about the scaling assumption. That transparency is the reason the number is usable — and the reason it should never be quoted as a measured national cost.
The extrapolation boundary
Here is the editorial distinction that matters. The 5–11 basis point range is a peer-reviewed estimate from a defined sample with stated identification tests. The 8.6 basis point premium is a modeling choice that imports a constant 40/60 bond mix across every state and county; the report does not claim to verify that mix locally. The $1.1 billion is an aggregation built on a population-scaling assumption. Each step is disclosed. None of the later steps is a measurement.
For scale intuition only — not a claim about any real county — a jurisdiction issuing $50 million a year at an 8.6 basis point premium implies roughly $43,000 in added annual interest. That is the size of the thing being priced.
Current rate context helps frame basis points: the Federal Reserve’s H.15 release for October 21, 2026, reports a 10-year Treasury constant maturity yield of 5.13%. Against that, 5 to 11 basis points is small per bond and large in aggregate — which is exactly why the aggregation step deserves scrutiny.
What the report argues, and what it does not prove
The Rebuild report uses the $1.1 billion figure to compare against subsidy costs, arguing that support for local journalism would generate a positive return. That is the authors’ policy argument, built on their own estimate. It is not an independent audit of the subsidy’s effect. The report also uses case illustrations — Bell, California; Harvey, Illinois — which are the report’s examples, not audited findings of the bond study.
The bond-market framing does change the policy conversation: it moves local news from a cultural question to a fiscal one, with a number attached to the taxpayer. But the number’s authority comes entirely from the chain above, and the chain has a measured link and an extrapolated link. Treat them differently.
A verification habit you can apply today
When you encounter any “cost of news deserts” figure, run four checks before you repeat it:
- What is the underlying basis-point estimate, and from which paper? If it is not Gao, Lee, and Murphy (2018 working paper; the Rebuild report attributes a 2020 Journal of Financial Economics publication), ask why.
- How were bond types weighted? The 40/60 GO-to-revenue mix is a national assumption, not a local measurement.
- What issuance-scaling assumption was used? Population share is one choice; it is not the only possible one.
- Is the figure measured or extrapolated? The $1.1 billion passes the first three checks transparently and is, by the authors’ own statement, an extrapolation.
That habit is the whole method. The paper priced a mechanism; the report priced a country. Know which number you are holding.
FAQ
Did the study measure default rates?
No. The retrieved working paper measures municipal bond yields — secondary-market and offering yields — not defaults. The mechanism runs through borrowing costs, not credit events.
Is the $1.1 billion figure a measured national cost?
No. It is a derived estimate: the 8.6 basis point weighted premium multiplied by MSRB 2025 state issuance and news-desert population shares, under a stated scaling assumption. The report itself calls its figures lower-bound estimates.
Does the paper prove digital outlets cannot replace newspapers?
No. It finds no significant difference in the closure effect between high- and low-internet-usage states in its sample. That is narrower than a general claim about substitution.
Where can I read the primary documents?
The 2018 working paper is hosted at Brookings (brookings.edu/wp-content/uploads/2018/04/Murphy-et-al..pdf). The Rebuild Local News report is at rebuildlocalnews.org. The rate context is the Federal Reserve H.15 release at federalreserve.gov/releases/h15/.