French linkers tied to domestic CPI draw local buyers while other euro investors lean away. Boermans and Swinkels can see that split in bond-level holdings: France sells linkers tied to two price indices, allowing a comparison within the same issuer. For a trader, that contrast is more compelling than the paper's broader claim about who owns inflation-linked debt.
Who holds the linkers?
The paper asks which government linkers investors use to hedge inflation, and which measure of inflation they choose. The authors interpret the holdings as preferred-habitat demand: investors seek bonds suited to liabilities that are nominal or real. Their data come from the ECB's Securities Holdings Statistics by sector, quarterly from 2013-Q4 to 2023-Q4. They cover holders in 20 euro area countries, grouped as insurers, investment funds, pension funds, banks, and smaller sectors. The bond universe spans 74 issuer countries. Bonds with less than EUR 500 million outstanding or fewer than 300 days to maturity are excluded, leaving 12,988 bonds, including 381 linkers. Just 63 linkers are euro area issues, from Germany, France, Italy and Spain.
The authors estimate a gravity-style demand equation using log market-value holdings for each bond, holder country and sector. Alongside an inflation-linked dummy, the controls cover bond size, investor-issuer distance, currency, yield and maturity; holder country, sector and time fixed effects also enter. A positive dummy means greater holdings than in a comparable nominal bond. Liability matching supplies the proposed mechanism. Life insurers largely write nominal contracts: the paper cites USD 209 billion in nominal US annuities in 2023 against USD 47 billion inflation-linked. Pension funds, by contrast, target real income.
At end-2023, euro area investors owned 13% of global government bonds and 20% of linkers. With controls, the global dummy is 0.303, equivalent in the authors' calculation to a 35.4% overweight. For euro area bonds, pension funds register 0.538 (5% level), while insurers register -0.352, significant only at 10%. Investment funds (0.328) and banks (0.012) are indistinguishable from zero.
How firm is the insurer result?
The abstract calls the split "pronounced market segmentation". The insurer result gives less reason for that confidence. In the global interaction model, insurers are 0.761 below other investors (standard error 0.317). Among euro area bonds, the underweight is concentrated at the long end: -0.922 beyond 10 years, compared with -0.005 at 0-5 years. The authors then match 54 of the 63 euro area linkers to similar nominal bonds. In that sample, the insurer coefficient moves to -0.163 and loses significance; the pension-fund coefficient rises to 0.947. The authors call insurer demand "absent among insurance companies" in the matched result, a narrower finding than underinvestment.
Pension-fund demand holds up better. Its 0.947 matched-sample overweight is joined by coefficients of 0.473 and 0.751 in the short and long maturity buckets. The 5-10 year bucket is the exception, at an insignificant 0.418. For issuers, timing matters too: the pension tilt vanishes for bonds issued within the quarter (interaction -1.025). This clientele appears to build positions in the secondary market.
One issuer, two inflation indices
France issues linkers indexed to French CPI and to euro area inflation. Holding the issuer fixed makes the contrast less vulnerable to ordinary home bias. French investors show a 0.721 tilt toward linkers, plus an extra 0.840 toward the French-CPI line. Non-French euro investors lean away from that line, at -0.846. The sharpest division is among French households. Relative to other French investors, their holdings are 1.845 lower for euro-indexed linkers and 1.302 higher for French-CPI linkers.
A separate test across all euro area bonds finds no additional home bias for linkers: the linker-times-home interaction is -0.123 with a distance control and -0.079 without one. The French comparison gives direct support to Arnold's argument, cited by the authors, that a euro area index is irrelevant to most investors. Its reach remains limited to one issuer and one country's investors.
The Dutch reform test
In June 2019, the Netherlands agreed the WTP (Future Pensions Act). It links benefits to investment returns and requires a minimum nominal funding ratio during the transition. Annual reports show that ABP and PFZW, the two largest Dutch pension funds (the civil-service fund and the healthcare fund), held more than EUR 30 billion of inflation-linked debt in 2018 and less than EUR 1 billion by end-2021. Across euro area pension funds, the linker share fell from about 14% in early 2019 to around 6% by late 2020.
The headline difference-in-differences estimate is -0.812 against all euro area investors and -0.756 against other euro area pension funds. Its post period runs from 2019-Q2 through 2020-Q1. Other columns extend the post dummy from 2019-Q2 or 2019-Q4 to the end of the sample, rather than ending it in 2020-Q1. Against other pension funds, three of five columns are significant: -0.756, -0.665 and -0.472. The remaining estimates, -0.209 and -0.315, have standard errors near 0.2 and are insignificant. The -0.209 starts in the same 2019-Q2 as the significant -0.472. The text's claim of consistency across specifications therefore applies to three of the five pension-fund comparisons. That control group matters most: the all-investor group includes insurers and banks, each with its own tilt.
The authors acknowledge the baseline problem. Dutch pension linker holdings spiked between 2018-Q4 and 2019-Q1, and no control group tracks that move. Parallel trends hold only from 2017-Q1 to 2018-Q3. Unusual buying immediately before the announcement makes the subsequent decline look larger. The authors also consider a coordinated change in inflation views, though they find it puzzling given ECB surveys showing expectations falling across the euro area. The abstract says demand "may be shaped by regulatory requirements". That degree of caution fits the evidence; the annual-report figures persuade more than the regression.
Holdings cannot price the trade
No regression has a price or yield on its left side. The paper therefore leaves the clientele's effect on real yields and breakevens unmeasured. Demand was flat during the 2021-23 inflation spike, with a coefficient of 0.310 against 0.318 in the low-inflation years and 0.370 during Covid. The authors note the flat response and attribute it to anchored long-run inflation expectations.
We could not replicate any of these results. The analysis requires bond-level holdings by investor sector and country from restricted Eurosystem data. We also lack bond-level linker prices with which to test a pricing consequence.
The ownership evidence still matters for allocation. Pension funds show the clearest euro area linker overweight, and they build it in the secondary market. The two largest Dutch funds, in Europe's largest pension sector (over EUR 1.5 trillion), reduced linker holdings from more than EUR 30 billion to under EUR 1 billion. Across the sample, Dutch pension positions fell by more than half. The paper also notes that the Eurosystem held up to 38% of the euro area government market in 2022 but is absent from these data. Every reported overweight is measured against holdings that exclude it. A breakeven trade would still need evidence linking these holdings to yields.