EPS accounts for about 47% of the fitted price at the paper's own averages. That relationship does much of the work in the fixed-effects regression reported by Nainggolan, Muhamat, Maryati and Marpaung.
Where is the trade?
A trader can readily imagine the strategy. Rank stocks on return on assets (ROA) and debt-to-equity (DER), buy profitable firms with little debt, and wait for prices to catch up. The premise is familiar: profitability signals asset utilisation and future cash flow, while debt brings distress risk that investors discount. Such a trade would depend on the market absorbing those signals slowly.
The authors assemble an annual panel of manufacturing firms on the Indonesia Stock Exchange from 2019 to 2023. They match each fiscal year-end closing price with the same year's ROA, current ratio (CR), DER and earnings per share (EPS). After estimating pooled, fixed-effect and random-effect models, they select fixed effects. The selection rests on a Chow statistic of 12.84 (p = 0.000) and a Hausman statistic of 9.67 (p = 0.021).
ROA has a positive coefficient of 85.42 (t = 3.75), as does EPS at 6.54 (t = 4.12). DER comes in at -45.89 (t = -2.98, p = 0.004). CR's coefficient of 12.67 is insignificant (t = 1.21, p = 0.228). The reported R-squared is 0.68, the adjusted figure is 0.64, and the F statistic is 18.75. The signs fit the investment story.
A trade remains untested.
The dependent variable is a price level. The explanatory accounts become available only after that price has been recorded.
The fit leaves questions open
The paper interprets its R-squared as meaning "approximately 68 percent of the within-model variation in stock prices is explained by the included fundamental variables, while the remaining variation is associated with factors outside the model." The term "Within-model" has no standard meaning. Readers are never told whether 0.68 refers to the within, between or overall R-squared. In a dummy-variable fit, firm intercepts could explain most of that figure. Our review of an Indonesian property study raised the same issue: its joint F included the firm effects.
EPS deserves a closer look. Table 1 gives averages for IDX manufacturers of 250 IDR in EPS and 3,500 IDR in price. The table appears in the introduction; the authors never identify it as a description of their regression sample. If it describes that sample, applying the 6.54 slope gives an EPS contribution of 1,635 IDR, about 47% of the average price. A split or rights issue moves both per-share quantities together. We found no discussion of corporate-action adjustments, nor any control for size or book value. Regressing price levels on EPS comes close to restating an earnings multiple.
A separate summary table reports means for ROA of 8.5 percent, CR of 2.1 times and DER of 1.3 times. The accompanying claim is that "the data indicate that firms with higher profitability and earnings performance tend to exhibit higher stock prices". Five averages in a column cannot establish a cross-sectional relationship.
Accounts arrive after the close
For 2019 to 2023, the regression pairs year-end prices with accounts for the fiscal year just ended. Those audited statements are released later. This design therefore measures co-movement between prices and accounts. The Method describes the purpose accordingly, "to examine the associations" between fundamentals and prices. Yet the Results says ROA and EPS "lead to higher stock valuation" and leverage "reduces stock prices". The design cannot establish whether the ratios were tradable. We found no lag structure, no out-of-sample period and no returns anywhere in the paper. Costs are absent too.
Sample selection adds another obstacle. Eligible firms must be "consistently listed during the observation period" and have "no extreme data anomalies". The latter screen is left undefined. Requiring firms to remain listed throughout all five years selects survivors, and a backtest using that universe would carry the same selection. After describing these criteria, the authors say a balanced panel "is constructed". They supply neither the firm count nor the number of firm-years. There is no way to check the degrees of freedom.
The published text flags unfinished statistical questions itself. It gives only bare diagnostic figures: normality 0.087, heteroscedasticity 0.112 and autocorrelation 1.98. It then says "the manuscript should clarify the exact test statistics and decision rules used" and recommends standard errors allowing for heteroscedasticity and serial correlation. This reads like a referee's instruction retained in the article. Similar passages appear in the Results, Discussion and Conclusion, including "The discussion would be stronger if it concentrated on...". Taken alongside that instruction, the coefficient table runs on unclustered errors.
The Hausman result is puzzling as well. With four regressors, a chi-square of 9.67 lies just beyond the 5% threshold. Its reported p of 0.021 does not match four regressors; it would match three degrees of freedom, suggesting a regressor may have dropped out of the test. Fixed effects would still be selected at 5%.
We also saw no year effects, despite a window that includes 2020 and an introduction that identifies macro conditions and sentiment as price drivers. Even the published conclusion asks for cited sources to be "accurate and verifiable". The Fama reference dated 2021 gives the volume and pages, 49(3) and 283 to 306, of his 1998 paper.
A within-firm result, with no return test
The abstract says the fundamentals "significantly explain variations in stock prices". Investors are told in the conclusion to "prioritize profitability and earnings indicators". The Discussion supplies a narrower instruction: the conclusion "should be framed as evidence from the study sample and period rather than as a universal relationship." That restriction is the right one. It also cuts against the broader advice.
Under firm fixed effects, the -45.89 coefficient means a firm's price was lower in years when its debt-to-equity was higher. This within-firm slope cannot rank firms against one another, leaving a further gap between the regression and a cross-sectional strategy. A trader needs to know whether high-DER stocks subsequently underperformed or recovered. The paper leaves that question unanswered.
We could not test it ourselves. We have neither Indonesian price and statement histories nor annual records of listed manufacturers needed to apply the continuous-listing filter. Substituting US stocks would answer a question about another market.
ROA and DER remain reasonable candidates for an IDX return study. Ratios would need to be lagged beyond their filing dates, with spreads measured after Jakarta trading costs. A positive ROA spread after costs under that design would change our view. For now, 0.68 remains a fit statistic whose type the paper leaves unspecified, estimated for the surviving manufacturers in its sample over 2019 to 2023.