A desk should keep asking both venues for a price: the reported 6 bps rise in CCP borrowing costs does not identify its next fill. In the ten trading days around the WHO pandemic declaration, borrowing a given bond through a euro-area CCP became about 6 bps dearer relative to borrowing it bilaterally. The estimate holds across the four fixed-effect sets in the baseline table from Danisewicz, Dieler, Mancini, Mazzari and Metzler. Getting back to those tables requires more judgment calls than their tidy presentation suggests.

Same bond, different prices

A bilateral security lender can see the borrower and charge for that borrower's risk. A CCP novates the trade, making it anonymous and mutualising losses through a default fund; the lender instead prices a pooled borrower. The authors argue that pooling dampens specialness in ordinary markets and steadies repo prices under stress. They use MMSR (Money Market Statistical Reporting) trades from 5 to 18 March 2020 to measure the 6 bps relative rise in CCP costs. They read the rise as stabilisation. We read the measured result as a relative price increase at CCPs, and we did not find a dispersion measure in the tables.

Specialness is the price a bank pays to obtain a particular security through a reverse repo. The ECB deposit facility rate (DFR) was −0.5%. Trades below the DFR count as security-driven, and borrowing cost is the absolute distance from the repo rate to the DFR.

The model starts with a lender whose rate must compensate for the shadow value of leaving the bond unencumbered. In bilateral trades, that lender sees borrower type and sets a rate for each type. At a CCP, novation and mutualised losses leave the lender facing a single pooled borrower, with a success probability averaged across types. Break-even rates are convex in default risk. As the authors put it, "averaging borrower-specific prices is not the same as pricing the average borrower." The average OTC rate is therefore more negative. Under a condition involving pH and α, Proposition 2 says that a rise in α, the perceived share of bad borrowers, compresses the CCP-OTC gap. That condition matters.

MMSR covers transactions reported by the 46 largest euro-area banks, roughly 80% of the market. The authors retain one-day tenors, bank or CCP counterparties, and banks trading in both venues on either side of the shock. The window is 5 to 18 March 2020 and ends at the PEPP announcement. Before the shock, the CCP coefficient is −5.403 bps (s.e. 1.816). Covid×CCP is 5.891 bps in the sparse specification and 5.857 to 6.189 with richer fixed effects. Compression is weaker among higher-NPL borrowers: the triple interaction ranges from −0.00839 to −0.00516. The authors put that at about 1 bp per 130 bp standard deviation in the NPL ratio. For German, French and Dutch government bonds it is stronger, with Covid×CCP×SAFE at 5.670 to 5.993 bps.

Which trades enter the comparison?

The venue dummy, window, fixed effects and bank-pair clustering are stated. Four sample choices remain open to us:

  1. Which side of the DFR. The main tables leave unclear whether trades above the DFR are excluded. Section 3.1 says "most transactions are security-driven, and we therefore focus on reverse repos." The event-study and Gini discussions, though, distinguish a full sample from security-driven trades. The event-study passage says the findings "hold both for the full sample and when splitting transactions into cash-driven and security-driven segments." The Gini discussion reports results "for the full sample and for security-driven transactions," with no response in the cash-driven segment. This suggests that the headline sample retains cash-driven trades, for which distance from the DFR measures funding cost. An implementer examining specialness would exclude them.
  2. The event date. Covid becomes one after 11 March in the text; the Table 2 caption says after 12 March. We did not find a discussion of the ECB measures announced on 12 March. They fall within the post window under either dating choice.
  3. What ΔCDS measures. The description calls it a day-on-day change. Table 1, however, gives means of 76.11 bps (CCP) and 102.62 bps (OTC), figures that look like levels. The triple interaction has a different meaning depending on which reading is right.
  4. CCP identities. MMSR makes CCP legs anonymous. The matched sample infers counterparties using time, volume, rate, haircut and collateral, but we did not find the matching tolerances.

Figure 3 needs a unit check as well. It is labelled in bps, though its values (−.65 to −.45) are percent. Its caption gives the DFR as −0.5 bps, while the text gives −0.5%. The German repo rate of −66.26 bps on 16 March matches Figure 3 when that figure is read in percent.

The model needs mostly risky borrowers

Proposition 2 requires (1−2α)(1−pHq)+α²(pL−pH)q < 0 for CCP-OTC compression. The paper calls the condition sufficient. Equation (11) makes it necessary as well: all the other factors have fixed signs. With α below 1/2, both terms are positive and the condition fails. Lenders must already regard a majority of borrowers as the risky type for the model's gap to compress. Even α above 1/2 does not suffice on its own. The condition in the main text also omits a q from the denominator that appears in the proof. Propositions 3 and 4 rely on "mild regularity restrictions" and a "sufficiently risky" pool. The empirical question beneath that theory is whether lenders in March 2020 saw most of the euro area's largest banks as the bad type. Across CCP transactions, those banks had a mean NPL ratio of 212.54 bps.

Core sovereigns carry much of the move

Once SAFE enters as an interaction, the base Covid×CCP estimate falls to 2.318 to 3.180 bps. Core sovereigns add 5.670 to 5.993 bps, roughly twice the base term for other collateral (2.318 to 3.180). Government bonds from any euro-area issuer make up 99.88% of CCP collateral, versus 68.83% OTC. After the announcement, shorting 1 to 10 year Bunds became persistently profitable, net of volume-weighted repo financing, at roughly 12 to 150 bps.

A flight into core collateral, concentrated in the flow already going through CCPs, could give the same sign without pooling. The paper treats default fund contributions, netting benefits and CCP anonymity as effectively unchanged over the window and captured by fixed effects. It does not treat CCP margin changes inside the window that way. A CCP-specific change during the window, including a margin change, would load directly on Covid×CCP.

The Gini table reports a base Covid×CCP of −20.30 to −28.72. This is the effect at a Gini of zero; the paper does not give the mean Gini needed to interpret the net effect.

The borrower preference result deserves attention. With the borrower preference index included, Covid×CCP remains at 7.172 to 7.542, so the authors are right that the venue interaction survives. The BPI triple interaction is significant too, at −3.855 to −5.214. A borrower's concentration on one lender (BPI) predicts less compression; the frequency-based measure does not. Figure 8 plots the pre-period estimates. The authors describe them as small and insignificant, and we found no tabulated coefficients.

A desk trades the quote it can get

The raw mean distance is 76.69 bps OTC (sd 111.47, median 35.00), compared with 8.65 bps at CCPs (median 6.00). Before the shock, the adjusted gap is −5.403. Average CCP tickets are 30.72 million euro; OTC tickets average 4.56 million. Collateral mixes differ too, so composition plausibly explains much of the raw distance. The −5.403 estimate comes from the sparse column, which controls for collateral sector, issuer country, rating, haircut and tenor but has no ISIN or bank fixed effects. The CCP level disappears from the richer columns. In column 1, −5.403 + 5.891 leaves a post-shock adjusted gap of about half a basis point. A desk asks whoever will quote its size. This estimate describes an average transaction during one fortnight, without estimating the fill that desk would receive.

We could not run the study ourselves. MMSR is confidential, and we have no repo transactions carrying OTC or CCP venue identifiers, nor bank NPL ratios with which to reconstruct the borrower-risk split.

The 6 bps is a measured repricing of CCP specialness during one event. It is a reason to keep asking both venues for a price.