Manik and Akbar give four Indonesian bank stocks buy-leaning ratings without testing what happened after the calls. Their 91.63% regression fit comes largely from identifying the banks; their "undervalued" assessment relies on historical multiples the paper does not print.

What earns a buy call?

The trade is familiar: buy a bank when its multiple has fallen but earnings and profitability still look sound, then wait for a re-rating. Manik and Akbar apply that reasoning to BBCA, BBRI, BMRI and BBNI, using fundamentals to screen for value traps.

They work with annual data for 2020-2025, giving them 24 firm-years. One exercise regresses share price on EPS, ROA and ROE. The authors first estimate a pooled common effect model with one intercept for all banks, then add bank fixed effects. A separate valuation exercise compares each bank's 2025 PER and PBV with its own 2020-2025 average. The resulting "relatively undervalued" label goes alongside the path of EPS, ROA, ROE and price, with 2025 values displayed in the table. BMRI and BBRI receive buy ratings, BBCA buy/hold, and BBNI buy speculative/hold.

The regression does not rank the stocks. Later returns never enter the assessment of those calls.

From R² 0.41 to 0.92

Pooled R² is 0.4139; adding bank intercepts raises it to 0.9163. Within the fixed-effects model, EPS has t 0.140 (p 0.890), ROA t 0.525 (p 0.606) and ROE t 0.466 (p 0.647). None of the individual fundamentals clears the test. Coefficients also change sign: EPS moves from -4.011 pooled to +0.414, and ROA from -722.426 to +495.376.

The authors acknowledge what the fit measures. They caution against reading the high R² as strong individual effects for EPS, ROA and ROE, and say fixed effects likely capture structural price differences among the four banks. The body's F-test discussion likewise treats significance as belonging to the fixed-effect model as a whole. Even the abstract attributes the 91.63% to both the model and company-specific characteristics. Yet it also says "simultaneously, the three variables have a significant effect on stock prices", citing an F of 31.03.

That F includes the bank identities. With six regressors and 17 residual degrees of freedom, (0.9163/6) divided by (0.0837/17) gives 31.0. Those inputs also reproduce adjusted R² of 88.68%. The six regressors are three bank dummies plus EPS, ROA and ROE, so the reported F compares all of them together against a constant. Testing the abstract's claim requires a comparison of fixed-effects models with and without the three ratios. We did not find one. Nor did we find results for the Chow, Hausman and LM tests the paper calls customary; it describes a random effects model without estimating it.

Nominal prices range from Rp3,087.50 to Rp9,675.00 across the four banks (SD Rp2,075.83). A separate intercept for each bank can soak up much of that price-level spread before EPS, ROA or ROE contributes. We found no discussion of split adjustment. The regression has 17 residual degrees of freedom, a sample the authors themselves call very small. ROA and ROE are likely collinear, though the paper gives no diagnostic. In that setting, t-stats from 0.14 to 0.53 settle little.

There is also a mismatch in the methods section: it names PER, PBV and EPS as the tested indicators, while the regression uses EPS, ROA and ROE.

The missing PER and PBV averages

The 2025 table contains four rows without tickers. Using the text to identify them gives:

Every row says "relatif undervalued". According to the conclusion, each bank's 2025 multiples are below its own 2020-2025 average. We did not find the averages in the paper, or a stated boundary between undervalued and fair. Including 2025 in the window also lets the observation being judged influence its benchmark.

What went into those averages matters. The descriptive table puts EPS as low as Rp88, ROA at 0.50% and ROE at 2.90%. If those lows occurred during the pandemic years around which the paper frames its period, depressed earnings would have lifted early PERs. A 2025 PER below the average could simply reflect recovered earnings. PBV is less vulnerable to that effect, another reason the paper needs to show both sets of averages.

Calls without a return test

BBCA exposes a disagreement between the table and the prose. Its row, identifiable by the premium PBV of 3.53x, says "Buy". The abstract and conclusion give it buy/hold.

The rating rules are verbal. Buy calls for attractive valuation and adequate fundamentals; hold covers good fundamentals where upside looks limited or confirmation is needed. Sell/reduce applies when fundamentals cannot support a high valuation. The paper attaches no numerical thresholds, and none of the four banks receives sell. BBNI has the lowest PER and PBV, 8.14x and 0.92x, yet its ROA of 2.10% and ROE of 14% bring it the weakest rating. That judgment makes sense as an instinct. It draws, though, on the two ratios assigned p 0.606 and 0.647 in the authors' regression. The authors take the insignificant coefficients as reason to consider fundamental indicators together instead of singly. They do not test whether labels formed that way predict returns.

We found no forward returns, hit rates or costs for these calls. The authors describe the classification as "hasil analisis akademik berdasarkan data penelitian dan bukan rekomendasi investasi personal". Fair enough; the ratings remain opinions accompanied by a table. We raised a similar issue with a valuation paper that had no portfolio test behind its error metric.

We could not run this ourselves. We have no price or book-value history for these four IDX listings, and using US banks instead would mean testing different securities. Rebuilding the four 2020-2025 averages requires the bank-level price and fundamental series.

Two additions would change our view: historical PER and PBV averages for the four banks calculated without 2025, and twelve months of returns after the 2025 labels.