Sudarmi's 60 observations give traders no evidence about which bank will rise next year. Across 20 Indonesian banks from 2023 to 2025, higher profitability goes with higher share prices. Even the negative finding on PER depends on which of two tables you trust.

What does Sudarmi's 20-bank IDX study measure?

Novice investors are the paper's intended audience. Its motivation is FOMO buying that ends in lost capital. It proposes three familiar ratios to help assess a bank's condition: return on equity, earnings per share and the price-to-earnings ratio (PER). The reasoning follows ordinary fundamental analysis. Better-run banks should fetch higher prices, and the ratios should therefore move with price. The sample covers banks listed on the Indonesia Stock Exchange (Bursa Efek Indonesia, BEI, also IDX) over 2023 to 2025.

Earlier findings disagree, which gives the paper its starting point. Mustaqim (2022) found EPS significant, and Kartiko and Rahchmi (2021) found ROE significant. In a study of banks, Felia (2024) found ROE insignificant and EPS significant. Evidence on PER for banks remains thin, the paper adds. Its practical claim concerns assessment: investors can use the ratios as supporting information while weighing factors outside the model. It claims no ability to forecast returns.

From 48 listed financial firms, Sudarmi selects 20 banks. The exclusions comprise 12 non-banks, 2 banks with stock splits, 6 with partial delistings, 2 without profits in all three years and 6 labelled outliers. Annual ROE, EPS, PER and share price for 2023, 2024 and 2025 yield 60 pooled firm-years. Price in rupiah is the dependent variable. Its mean is Rp1,648.77, and it ranges from Rp50 to Rp9,675. Sudarmi estimates OLS in SPSS 25. We found no firm or year effects, and we could not find the date within each year used to measure price.

In the final table, ROE has B = 142.028 (t 3.205), EPS has 7.345 (t 3.964), and PER has 5.513 (t 1.172, Sig. 0.246). The joint F is 41.329; R² is 0.693. The conclusion is restrained. ROE and EPS can be one consideration ("dapat menjadi pertimbangan") when analysing bank stocks. Investment decisions should still account for other financial and market factors.

Price on both sides

PER divides price by EPS. Price therefore equals PER times EPS exactly, and an additive regression of price on EPS and PER fits a straight line to that product. PER ranges from 2.20 to 228.80. The maximum is far above the mean of 19.96 and SD of 30.53, despite the removal of six "outlier" banks.

From the insignificant PER coefficient, Sudarmi suggests ("dapat diasumsikan", it can be assumed) that PER "bukanlah indikator utama yang dipertimbangkan oleh investor" (is not the main indicator investors weigh) during the period. The explanation offered is that a rising PER can indicate either a healthy bank or financial trouble. The identity cannot support that interpretation. Alongside EPS, the PER slope describes a straight-line fit to a product. Its t-stat captures the shape of that identity in this sample and tells us nothing about what investors consider.

EPS has a related weakness. Both EPS and price are rupiah per share, so the same share count scales each of them. Nominal unit size accounts for part of the 7.345 slope: BBCA (Bank Central Asia) traded near Rp9,400 in 2023, compared with roughly Rp1,150 for BJTM (Bank Jatim). We raised the same issue with PBV in a single-stock IDX study in an earlier review.

The two PER results disagree

The paper contains two different regressions. In the multicollinearity table, the constant is -1040.857, ROE is 140.776 (t 3.460), EPS is 8.112 (t 4.659), and PER is 16.253 with t 3.209 and Sig. 0.002. In the regression table, the constant is -453.945 and PER has Sig. 0.246. One table makes PER significant. The other makes it insignificant.

No explanation connects the two tables. A Cochrane-Orcutt correction is the only transformation mentioned. The paper places it under autocorrelation, although one sentence describes it as a remedy for heteroskedasticity. Total ANOVA df of 58 implies 59 observations, compared with 60 in the descriptive statistics. Losing one observation would be consistent with Cochrane-Orcutt. If applied to the 60 stacked rows, however, the lag would connect one bank's 2025 figure with the next bank's 2023 figure. At 1.691, the reported Durbin-Watson exceeds dU of 1.6889 by just 0.0021.

The abstract's "PER does not have a significant effect" appears to depend on that transformation, named twice and never specified.

The discrepancies continue elsewhere. PER is described in the discussion as "berpengaruh positif dan signifikan" (has a positive and significant effect), beside Sig. 0.246. For mean ROE, the text reports 9.3587 and the table reports 9.2470. Normal residuals are accepted using a Monte Carlo Sig. of 0.081, immediately after the preceding sentence quotes 0.143. That table also contains a Lilliefors-corrected p-value of 0.001. Sudarmi acknowledges that the study "mungkin terdapat kesalahan" (may contain errors) and invites criticism and suggestions. The candour deserves credit. These are the corrections.

What R² of 0.693 can support

R² of 0.693 measures variation in price levels across 59 or 60 pooled bank-years, after screening within the window being explained. The stated outlier criterion applies to 2023 to 2025 itself. The profitability screen likewise requires results for all three years. Those screens remove 8 of the 28 banks remaining after the split and delisting exclusions. Both select on the data the regression seeks to explain. Clustering errors by bank would likely widen the standard errors and reduce these t-stats; we found no clustering in the paper.

The coefficient interpretation also mixes units. The paper describes EPS as changing price by 7.345 for a "1%" increase. A levels model assigns that slope to each rupiah of EPS.

Asking investors to consider other factors leaves the central gap untouched. The regression answers a contemporaneous question, yet its ratios are offered to novices losing capital on FOMO buys. Those readers need a forward test. A bank trading below its fitted value in 2024 would need to be tracked into 2025 returns before anyone could call it cheap. We found no forward-return test, no return variable and no cost assumption anywhere in the paper.

We cannot rerun the IDX bank regression

Our data covers US equities. We have neither IDX bank statements and prices nor Indonesian split or partial-delisting histories to reconstruct the 20-bank screen. A regression on US banks would examine a different market and would not reproduce this result.

One result would change our view: a positive slope from regressing next-year returns for these 20 banks on current ROE and EPS, with errors clustered by bank.