Two positive bank-stock reactions to ECB climate releases turn negative when the benchmark simply subtracts the sector index. Giampaoli and Renghini's headline figures offer no announcement effect anyone could trade. They measure dispersion around five releases, and the cumulative returns they emphasise average across banks selected because they moved.

The supervisory concern here is climate change. The paper distinguishes physical risk from transition risk, which supervisors bring into their oversight through tools including stress tests. Its proposed mechanism is signalling: non-binding ECB communications reveal the direction of supervisory expectations, and bank equity prices anticipate it. The authors also interpret the signs. A positive abnormal return means the market regards a bank as prepared for climate requirements; a negative return means the bank has yet to internalise climate risk. The discussion of direction below depends on that interpretation.

Which releases did the authors choose?

Between 2020 and 2022, ECB Banking Supervision published climate communications with no capital consequence. The question is whether bank equity prices responded to those signals anyway. From a wider set beginning in 2017, the authors choose five press releases:

The sample contains 178 listed banks across 25 countries. It includes 106 Eurozone banks and 35 institutions directly supervised by the ECB (SIs). Daily prices from DataStream cover 2 Jan 2017 to 24 Jun 2022. The authors call their abnormal returns RAR: residuals from a market model using the MSCI EMU Commercial Banks index. They estimate that model from 2017 until seven days before each event. The event window covers four trading days on either side of the release. Each bank-day is tested at 95% using the Boehmer et al. (1991) test.

The analysis has two outputs. One counts banks with at least one significant RAR before, on and after the announcement. The other divides the sample into a "significant" group and everyone else, then t-tests the difference between their cumulative RAR (CRAR). From this comparison, the authors report positive reactions to events 1, 2 and 5. The significant group's CRARs are +6.73%, a peak of +30.58% and a peak of +8.30%, respectively. Event 3 produces a negative reaction (−10.62%); event 4 produces nothing. The authors say some effects were priced in ahead of time and that "effects dissipated rapidly". They interpret the earlier moves as markets anticipating the ECB's direction.

Little happens on announcement day

The announcement-day shares of the 178 banks with significant RAR were 11.80%, 7.87%, 9.55%, 4.49% and 10.67%. A 5% test would flag about 5% of banks on an uneventful day. The stress-test launch falls below that threshold, and 0 of 35 SIs react. The authors acknowledge this weakness: "the daily RARs are generally not statistically significant across the observed events." Their claim rests instead on differences in cumulative CRAR.

For every event, more banks move during the four days ahead of the release than on the announcement day. In May 2020, the counts are 63 versus 21; in March 2022, they are 82 versus 19. The authors describe this as anticipation. That interpretation faces two problems.

Start with the null rate. Across four independent days, a 5% test would flag about 18.5% of banks at least once (our arithmetic). Event 5's pre-event share of 46.07% exceeds that 18.5% independent-test rate. Event 3's 19.66% pre-event share is around the null rate.

Then consider the surrounding news. The window from 8 to 18 March 2022 came about two to three weeks after Russia invaded Ukraine. The authors assume a nine-day window is short enough to exclude confounders; they include no explicit controls. March 2022's nine-day window still falls within the war. Across all nine days, chance alone would flag about 37% of banks. Event 3 records 36.52%, essentially that figure.

Independence is the generous assumption. The authors acknowledge in a footnote that event-date clustering "may lead to over-rejection of the null hypothesis." Their RAR definition cites Kolari and Pynnönen (2010), yet they use Boehmer et al. (1991) for the test. It does not correct for cross-correlation.

Selection comes before the CRAR test

The CRAR t-tests compare banks flagged as significant with those left over. Selecting on significance ensures that the selected banks moved more in absolute terms. A difference in means does not follow automatically. The group combines banks that rose with banks that fell, allowing the daily mean to cancel out. Its average sign depends on which side moved further. This explains why the daily t-tests are mostly insignificant, including event 2's announcement day (t=1.04, p=0.317).

Significance emerges when the cumulative tests add up the days. Those gaps still compare a group chosen for having moved with the remaining banks.

Some observations look like data errors. The descriptive statistics include German-bank daily returns of −8.764 and +4.71, equivalent to −876% and +471%. They also print the all-bank maximum (−0.00015) as equal to the mean. We found no rule for outliers.

Event 2 makes the problem most visible. After a non-binding guide, the reacting banks' CRAR reaches +30.58% on 1 Dec 2020. Two days later it is +11.86% (t=1.08, p=0.300). Their daily RARs range from +12.44% on 25 Nov to −12.57% on 3 Dec.

A different benchmark reverses three of four reactions

The alternative-benchmark section claims that both alternatives preserve the direction of reaction: "positive for the first and second announcements, and anticipatory for the fifth". Half the evidence supports that statement. Mean-adjusted returns remain positive, reaching cumulative +14.91% for event 1 by 26 May 2020 and +14.12% for event 5 by 18 Mar 2022.

Market-adjusted returns tell a different story. They subtract the index directly from the raw return. With this benchmark, the significant group's cumulative return is −11.50% for event 1 (t=−10.35) and −14.34% for event 5 (t=−5.21). At window end, RAR for those same events is +6.73% and +6.32%. Event 5's closing figure misses 5% significance, with p=0.065. Its +8.30% peak (p=0.001) is positive as well.

Event 3 reverses in the other direction: −10.62% under RAR becomes +7.88% under MKAR (t=8.78). Event 2 remains positive mid-window, at +14.65% on 30 Nov, then finishes at −5.18% on 3 Dec. Neither of event 2's end values differs significantly from the non-significant group. The MKAR figure has t=−0.11 (p=0.914); the RAR figure of +11.86% has p=0.300.

If the sign reverses when the index replaces beta times the index, it tells us nothing about the direction of these reactions.

The benchmark creates a broader limitation. A bank-sector index absorbs any response banks share to a banking supervisor's release. The residual captures dispersion across banks, leaving the paper unable to establish a sector-wide reaction. There is also a currency question. The authors describe the universe as euro-traded, while applying the EMU index to 24 Polish, 18 Danish and 8 Swedish banks. We could not see how they handle currency.

The paper repeatedly describes event 5 as the "publication of the climate stress test results". Its event description identifies a disclosure assessment, and its sample ends on 24 Jun 2022.

Desk context, without a trade

One finding survives for events 1, 2 and 5 only. Over nine days, at least one significant RAR appears for 53.37%, 47.75% and 59.55% of the 178 banks, compared with about 37% by chance. Correcting for clustering could reduce even that result. In each of those three events, more banks move during the four days before release than during the four days after. The paper supplies no strategy, costs or out-of-sample test. It also restricts the universe to "currently active" banks with complete histories from 2017. Failed banks therefore never enter the sample.

We could not test these results ourselves. Our equity data focuses on the US and contains neither the European bank prices nor the MSCI EMU bank index. Substituting US banks would strip out the mechanism under examination: differences in exposure to ECB supervisory expectations.

A reaction aligned with bank-level climate exposure, under a test corrected for cross-correlation, would change my view. The paper includes no bank characteristics and has only 35 names in its SI sample. It cannot establish that link yet.