Indian bank fraud headlines offer no tradable peer signal in these tables. From the second week after each event, the peer-bank results read like the same series printed four times.

The claim and the events behind it

Dahal and Das study four fraud disclosures at listed Indian banks: IDBI Bank (26 April 2018), Punjab National Bank (18 May 2018), Bank of India (21 January 2020) and Canara Bank (18 December 2020). They began with five events, dropping one "because of event clustering and also to achieve uniformity". For each event, they estimate a market model using the BSE 30 over the preceding 120 days. They calculate daily abnormal returns from day -10 through day +30 for the victim bank, a public sector (PSU) bank series and a private bank series. They t-test the daily returns and cumulative windows from (0,1) through (0,30).

De Bondt and Thaler's overreaction and Hong and Stein's gradual diffusion supply the behavioural frame: a slow response followed by an overshoot. The paper tests negative abnormal returns for victims (H1) and negative sectoral spillover to peers (H2). A trader would care about H2, where another bank's fraud might start a drift in peer prices.

The headline says "fraud in private banks does not affect the victim bank, but is accompanied by negative spillovers to public sector banks." The Conclusion gives the spillover hypothesis only conditional support; the abstract states it without that qualification. A second-stage regression also associates greater resilience with older banks and banks with higher turnover.

Which victim was private?

The paper treats only Case I, involving IDBI, as private-bank fraud. IDBI was majority government-owned in April 2018 and entered the private-sector category only after LIC took control in early 2019. The headline's "private" side thus depends on one event whose classification is disputable.

Even accepting that classification leaves a thin spillover. The PSU series has significant negative returns on day -10 (-0.018, p 0.05), day -4 (-0.021, p 0.02) and day -2 (-0.019, p 0.04). On the event day it returns -0.009 (p 0.39). The authors attribute the earlier moves to rumour. They may be right. A trader still cannot act on a move recorded before timestamped news.

Canara's move, PNB's date

Canara supplies the clearest short-window loss: day 1 is -0.07 (p 0.00), with CAAR (0,1) of -0.09 (p 0.00). It is also the sole case with an event-day peer move, at -0.02 (p 0.01) for PSU banks and -0.01 (p 0.09) for private banks. PNB behaves differently. After returning -0.12 (p 0.00) on day -2, it rallies following the announcement; CAAR (1,5) is 0.14 (p 0.03). BOI has just one significant day, day 28 at -0.08 (p 0.00).

The Discussion says public sector victims "experienced significant and immediate negative abnormal returns beginning on the event date". The tables give event-day returns of 0.00 (p 0.96) for PNB, 0.02 (p 0.39) for BOI and -0.02 (p 0.35) for Canara. Only one case in three under the paper's classification, or one in four if IDBI counts as public, supports that description, and the move begins on day 1.

PNB's date also needs explaining. The widely reported fraud tied to Nirav Modi was disclosed in February 2018, three months before the paper's 18 May date. We found no explanation in the paper of which disclosure 18 May marks. If the date is wrong, the PNB estimates describe another event.

The testing volume adds another reason to distrust isolated stars. Across 41 days, 3 series and 4 cases, the paper runs 492 daily tests. At the 10% level, roughly 49 would flag by chance under independent tests. We found no correction for this; scattered significant days are what that volume of testing can produce.

One peer path across four events

From day 7 through day 30, peer-bank abnormal returns and p-values match in all four cases to the printed precision. On day 11, the private series returns 0.016 in Case I and 0.02 in Cases II to IV, each with p 0.00. On day 19, the PSU series returns -0.03 (p 0.01) in Cases II to IV and -0.028 (p 0.01) in Case I. Its CAAR over (0,30) is -0.05 in every case, with p-values of 0.43, 0.44, 0.46 and 0.43. The prose repeats the pattern, naming days 11 and 30 as positive for PSU banks case after case.

The event dates straddle the March 2020 crash. Market-model residuals estimated from separate 120-day windows cannot trace a shared path. The repetition bears directly on the abstract's claim that "investment shifts occur within both public and private bank segments." The case accounts draw on a positive day 3 and a negative day 4 for the private series: day 3 has p 0.01 in all four cases, while day 4 has p-values between 0.02 and 0.04. We also could not find how either peer series is constructed or whether it excludes the victim. The peer findings cannot serve as evidence until the authors rerun those series.

Four banks behind the resilience result

Age has t = 2.72 (p 0.01), while total turnover has t = 1.95 (p 0.05) and a coefficient printed as 0.0000. Section 4.6 treats the dependent variable, an abnormal return level called CAAR in the table and AAR in the text, as stability: "the greater the age and turnover, the higher the stability of the stock." A positive coefficient on a return level concerns the level of returns; steadiness requires a separate measure. The Discussion's reading of the coefficients as smaller losses is more defensible.

The Conclusion credits "Larger" banks, although firm value, its size variable, has t = -1.21. The text calls the remaining coefficients positive as well, despite negative t-values for firm value (-1.21) and number of trades (-0.32). Age barely changes within a bank, leaving its t-statistic identified off four banks. We did not find the observation count. Heteroskedasticity-consistent standard errors cannot fix a design with four clusters.

A day-long victim short

Short the victim at the close of day 0, then cover a day later. The four day-1 abnormal returns are -0.004, 0.06, 0.01 and -0.07. Their sum is -0.004, or about 0.1% per event, gross. Costs are excluded, and that calculation assumes a hedged market leg. Four events with no holdout leave no out-of-sample result to assess.

The authors acknowledge limits on "generalising the findings across different market contexts and time horizons", while saying these "do not diminish the value of the study's contributions."

We could not reproduce the test ourselves. Our data has no Indian bank equities or BSE 30, and our news history begins around 2020, after both 2018 events. US banks would lose the state-ownership split on which the paper depends.

A rerun could change our view if it produced distinct peer paths for each event, properly dated PNB, and still found a PSU-peer drop after a Canara-style disclosure using cross-sectional standard errors. That would be a small contagion finding worth testing on more events. For now, the nearest example is Canara's event day, when PSU peers return -0.02 (p 0.01): one event, tested with a t-test whose standard errors the paper does not specify.