Higher ROA gives traders no demonstrated buying signal in these 10 Nigerian banks. Daramola, Ogunleye and three co-authors find that more profitable banks have higher share prices at the same year-end over 2015 to 2024.

The accounts arrive too late to trade that relation.

The authors frame the study through signalling theory: management publishes profitability ratios, investors read them as news about prospects, and demand for shares increases. They select 10 deposit money banks listed on the Nigerian Exchange Group, choosing purposively from those "having the required data": First Bank, Access, Fidelity, GTBank, Stanbic IBTC, Sterling, UBA, Wema, Zenith and FCMB. Annual observations from 2015 to 2024 yield 100 bank-years. Year-end closing price per share is the dependent variable. The regressors are ROA (net income over total assets, times 100), ROE (net income over equity, times 100), size (log total assets) and age (years listed). They estimate panel feasible GLS in Stata 17.

The reported ROA coefficient is 0.474 (t = 2.08, p = 0.037). For ROE, it is 0.225 (t = 1.87, p = 0.062). The authors give an R-squared of 0.849 and an F of 21.17, then draw an investment recommendation from the regression. They advise investors to consider ROA before buying Nigerian bank shares. They also claim that "the models developed in this study can be used for predictions of SPR in the NGX."

Their contribution is an updated window through 2024. They identify two gaps in earlier research: dated samples and mixed findings. Most previous Nigerian bank studies, they argue, might miss current trends; the examples they cite end between 2017 and 2022. Extending the sample through 2024 leaves the kind of evidence unchanged. Our review of a CAMELS study on 32 Indonesian banks reached the same verdict. Significant ratios do not establish that the signal precedes the returns.

The price is recorded before the signal

The regression relates share price (SPR) in year t to ROA in year t. Its dependent variable is observed on 31 December of year t. Audited accounts for that fiscal year cannot be completed before then. Under the paper's specification, the price has therefore been fixed before the supposed signal becomes public. The model describes a contemporaneous valuation relation. It cannot forecast anything.

Nominal prices pose another problem. They range from N0.520 to N69.650 across the panel, with a mean of N14.556 and an SD of 15.435. Differences in shares outstanding may explain some of that spread. Scaling price by book or earnings would remove that effect. The paper uses neither adjustment.

What does a point of ROA buy?

Read literally, the coefficient associates one percentage point of ROA with N0.474 of share price, about 3.3% of the mean price. One SD of ROA (10.332) corresponds to roughly N4.90.

The dispersion itself raises doubts. Mean ROA is 5.189%, and the maximum is 64.952%. A deposit bank earning 65% on its assets is far more likely to represent a data error than a real bank-year. We found no winsorisation or outlier check in the paper. With only 100 observations, a single bad row can generate a t of 2.08.

The distinction between the profitability ratios is thin too. ROA and ROE have a correlation of 0.710 and VIFs of 7.561 and 7.297. ROE has the stronger raw correlation with price: 0.687 versus 0.556 for ROA. The multivariate regression puts ROA ahead, with p-values of 0.037 against 0.062. That gap carries the paper's recommendation to favour ROA. Two collinear regressors changing rank across a 5% threshold give managers too little reason to target one ratio over the other.

Fixed effects disappear from view

The Hausman test favours fixed effects (chi2 17.41, p = 0.0016). Breusch-Pagan detects heteroscedasticity (chi2 47.11), Wooldridge detects serial correlation (F = 18.328, p = 0.0020), and Pesaran detects cross-sectional dependence (7.011, p = 0.000).

The authors move to FGLS, saying that addressing all three problems "could only be achieved" that way. Their FGLS table reports a constant of -49.71 without stating whether it includes bank effects. Dropping the fixed effects favoured by Hausman would let persistent differences in banks' price levels load onto the regressors, including ROA. A N0.52 stock differs from a N69.65 stock partly for reasons such as share count. Those reasons need have no connection to profitability.

We could not find the error covariance imposed by FGLS. A full cross-bank covariance for 10 banks over 10 years has 55 free elements to estimate from 100 observations. So few observations per element can leave FGLS standard errors too small. The t of 2.08 is especially exposed.

The fit statistics add to the uncertainty. A four-regressor model would produce F near 133 and adjusted R-squared of 0.843, rather than the printed 21.17 and 0.826. The latter 0.826 matches four regressors plus nine bank dummies. Fixed effects may therefore be present without appearing in the table. Either way, readers cannot establish which model the authors estimated.

Keep the ROA hypothesis, leave the screen untraded

The claim that profitability "altogether significantly affects" price relies on the F of 21.17. That statistic tests all four regressors jointly, including size (p = 0.090) and age (p = 0.228). We found no separate Wald test for ROA and ROE. The profitability block has therefore not been tested as a block.

The authors offer two extensions: adding inflation, interest rates and exchange rates in future work, and looking beyond the banks to which these results are confined. Yet their conclusion says profitability "remains an important predictor" after controls for size and age. A contemporaneous price level and one t of 2.08 leave that claim untested.

A different test would change my view. Date each bank's ROA to the release of its annual report, then measure subsequent returns. Scale prices by book. Include banks that failed or delisted between 2015 and 2024. If lagged ROA continued to sort the returns that followed in that sample, the screen would earn a place.

The 2024 recapitalisation gives this question more urgency. The international-licence floor rose from N50bn to N500bn, and 33 banks had raised N4.65 trillion by 31 March 2026. Issuance on that scale alters share counts. Nominal price levels will consequently be even less comparable after 2024 than they were before.

We could not run these tests ourselves. Our platform has neither Nigerian bank share prices nor point-in-time Nigerian financial statements carrying publication dates. Substituting US banks would test a different market. We retain the hypothesis, but the screen has yet to earn a trade.