Traders should withhold judgment on whether rating announcements are tradable in Istanbul until Torun and Erkiliç reconcile their return tables. The authors conclude that Turkish bank stocks absorb credit rating announcements efficiently. Their published figures cannot yet support that conclusion.

What does the Turkish bank rating study measure?

The authors ask whether listed Turkish bank stocks earn abnormal returns around corporate credit rating announcements. They interpret an insignificant pooled abnormal return as evidence of semi-strong efficiency and suggest this may increase investor confidence. The premise is straightforward: a rating action is public news, so efficient prices should adjust on the announcement day and then stop. A trading opportunity would require drift before or after that date, through leakage or slow digestion of the news.

The sample contains 66 corporate rating announcements for nine Borsa İstanbul banks, collected from the Public Disclosure Platform (KAP). Those banks are Akbank, AlbarakaTürk, Garanti, Halkbank, İş Bank, Şekerbank, TSKB, Vakıfbank and Yapı Kredi. Fitch, Moody's, JCR Eurasia and Capital Intelligence are the rating agencies. Of the events, 21 fall in 2023 and 45 in 2024. The paper classifies 38 as upgrades and labels 28 Stable. None are downgrades. An announcement released after the close is assigned to the next working day.

The estimation follows a standard market model. Each bank's daily log returns are regressed on the BIST 100 index, Borsa İstanbul's main index, over (t-110, t-11). The authors describe this as seven months, although the window covers 100 trading days. They then sum abnormal returns over (t-10, t+10), a span of 21 trading days, to produce a cumulative abnormal return (CAR) for each announcement. A single t-test on the 66 CARs gives t = 0.903 and p = 0.370. The authors take this as evidence of efficiency. They also report average abnormal returns (AARs), the average abnormal return on each day of the window, for all 21 days.

Such an event study traces the average price path around an announcement. Whether that path offers a trade depends on the timing of the returns. Here, the largest daily AARs occur on days -1 and +5.

Tables 2, 3 and 5 disagree on average abnormal returns

Every CAR in Table 2 is negative. Across the 66 entries, the range runs from -0.0016 (No. 64, Şekerbank) to -0.563 (No. 21, a CI action labelled Stable on Şekerbank). Yet Table 3 gives a total of +0.917, an average of +0.0139 and a standard deviation of 0.125. Divide that positive average by 0.125 over the square root of 66 and the result is exactly the headline t of 0.903. The efficiency finding therefore rests on the positive average.

The inputs do not reconcile. We summed the Table 2 column ourselves and obtained roughly -5.9, implying a mean near -0.09. Adding the 21 daily AARs in Table 5, which are also all negative, gives roughly -0.07. Even the Table 5 sum and Table 2 mean disagree. Table 3 reports +0.0139. With the reported dispersion of 0.125, a mean of -0.07 would put the t-statistic near -4.6. We cannot determine which figure is correct. We can say that the paper's sole test statistic cannot be reconstructed from either its per-event returns or its per-day returns.

The regressions need checking too. Almost all market-model betas in Table 2 are negative, mostly between about -0.9 and -1.5; Halkbank reaches -1.501. The only positive beta belongs to Akbank, event 26, at 0.236. For Garanti, a beta near -1 against the BIST 100 would be an extraordinary economic result. A sign convention in the regression output looks more likely. The paper leaves the sign unexplained, and the regression output needs rechecking before the abnormal returns can be trusted. There is also a marking problem: every p-value in Table 5 bears the 1% marker, even 0.884 on day -5.

A flat 21-day CAR leaves the speed of adjustment unresolved

Suppose for a moment that the pooled CAR is insignificant. Prices could still drift within a 21-day window whose total is near zero. Efficiency concerns the speed of adjustment. The paper's daily results report significant AARs at 10% on days -4, -2, -1, 0, +3 and +5. On day -1, the AAR is -0.0075 (t -3.32). Day +5 has the window's largest move, -0.0091 (t -2.82).

The authors describe both 1% days as an expected situation and still conclude that prices are efficient. They argue that some announcements become public a day after earlier announcements, allowing investors to compare a rating with the previous one and earn abnormal returns. This is plausible for day -1: a Fitch action on one bank conveys news about the next bank Fitch rates. They give the same explanation for day +5, where the timing fails. A rival bank's rating released the next day arrives at day +1, not day +5. A 1%-level move five sessions after the announcement is post-announcement drift, precisely what semi-strong efficiency rules out. If the authors' reasoning holds, it also describes a cross-bank anticipation trade.

The uniformly negative sign is puzzling. All 66 CARs and all 21 daily AARs are negative, even though 38 of the 66 events are upgrades and none are downgrades. Combining upgrades with 28 events labelled Stable dilutes any upgrade effect. We did not find results split by type, agency or bank. Upgrades should be reported separately from the Stable actions.

Clustering creates another inference problem. Table 1 records eight Fitch actions on 18.03.2024, seven on 25.09.2023 and another seven on 17.09.2024. Banks rated together share market days, making their abnormal returns correlated. The t-test nevertheless treats 66 events as independent. Events for an individual bank can also have overlapping windows: Vakıfbank's JCR and Fitch actions on 24 and 26 June 2024 fall within each other's 21-day windows. The authors acknowledge confounding events, writing that "Such events may introduce confounding effects, potentially biasing the analysis results." Their inference does not address the separate problem of correlated errors across same-day announcements. Clustering standard errors by announcement date, or combining same-day events into portfolios, would address that problem.

What would change my view

First, I would want Table 3 to reconcile with the sum of Table 2 and an explanation for the negative betas, whose sign the paper does not discuss. I would then want a separate test of the post-event window (0, +5). If that return is near zero, I would accept efficiency for this sample of nine banks.

We could not test these questions ourselves. We have neither Borsa İstanbul bank price histories nor a feed of KAP rating disclosures. Substituting US banks would test efficiency in a different market and reproduce nothing here.

As published, the paper's sole test statistic cannot be reconstructed from its own Table 2.