This paper gives traders no usable signal yet. Its main result depends on a variable labelled a dividend payout ratio whose median is 1,989.

Reysya, Paleni, Kesuma and Anggraini examine whether dividend payout ratio (DPR) and year-on-year sales growth explain realized returns among consumer-goods firms on the Indonesia Stock Exchange. They propose profitability (ROE) as the channel connecting both variables to returns.

Their sample contains 34 firms from a population of 131, observed over 2021 to 2024: 136 firm-years. Returns measure realized capital gains or losses. The authors estimate panel regressions in EViews 13 and report negative associations. In separate regressions, DPR produces t = -2.611 (p = 0.0104), while sales growth produces t = -3.393 (p = 0.0010). The joint F-test has a reported p = 0.008595. Adding ROE to either regression leaves it insignificant, which the authors interpret as evidence against mediation. They advise firms seeking optimal returns to reduce their payout ratio.

Who gets through the dividend filter?

Only 26% of the population qualifies: the 34 firms paid dividends every year from 2021 to 2024 and supplied complete reports for all four years. The authors apply both conditions across the entire study window. That creates survivorship in any forecasting interpretation. In 2021, nobody knew which firms would still be paying through 2024.

The filter also excludes every zero-payout observation. Claims about payout levels consequently apply only to differences among consistent dividend payers. We encountered the same combination of surviving firms and same-year returns in our review of eighteen LQ45 stocks.

The units need explaining

Start with the distribution of X1:

If these are percentages, the median firm distributed about twenty times its earnings. Something has been mis-scaled or mis-measured. The written equation gives the X1 coefficient as -6.0250617994; the table gives -6.03E-08. The conclusion then names the variable a "Debt Payout Ratio."

Using the table's coefficient, a move from median DPR to mean DPR changes predicted return by about -0.021. A move to the maximum changes it by about -1.8. The tail carries whatever economic content the variable has.

Returns are heavily tailed too. Their maximum is 17.248 (roughly 1,725%), against a median of -0.014, with skewness of 8.0 and kurtosis of 75.5. The residual Jarque-Bera statistic reaches 31,019.57, with p = 0.000000. In the results section, the authors acknowledge that the model has yet to satisfy the basic assumptions. Their conclusion says all classical assumptions were satisfied and therefore calls the estimates valid and reliable. Those statements contradict each other. With only 136 observations and kurtosis near 75, a few firm-years could generate a t of -2.6. We found no reported winsorizing or outlier-resistant inference.

Sales growth loses significance alongside dividends

The sales-only regression supplies the claimed significance: its sales-growth coefficient is -0.6006 (p = 0.0010). In the joint regression printed in the paper, that coefficient shrinks to -0.0912, with t = -1.376 and p = 0.1720. The authors use this joint model for their F-test and R-squared, yet the conclusion continues to call sales growth significant.

The abstract quietly drops significant for sales growth while retaining it for dividends. In discussing the joint result, the paper cites earlier studies finding no significant sales-growth effect. Wahyudi (2022) appears on both sides of the argument. This comes close to conceding the point.

An 85% coefficient decline cannot be explained by collinearity here: the correlation between X1 and X2 is -0.0286. The tables probably use different estimators. The constant has a standard error of 0.142 in the sales-only regression, compared with 0.021 in the joint regression.

The F-test supplies another clue. With two regressors and about 133 residual degrees of freedom, an F of 1.866 yields p near 0.16. The reported value is 0.008595. Meanwhile, the gap between R-squared of 0.395 and adjusted R-squared of 0.183 implies about 36 estimated parameters across 136 observations. That parameter count reconciles the F-test: F = (0.395/35)/(0.605/100) is about 1.866, matching the reported F. And F(35,100) at 1.866 yields p near 0.009, matching the reported 0.008595.

The reported F therefore tests firm intercepts together with DPR and sales growth. Its scope extends beyond the two regressors. This fits firm fixed effects, although the paper never identifies its estimator. The table carries the label "Weighted Statistics," and we found neither a statement of the panel estimator nor a Chow or Hausman test. Elsewhere, the text gives adjusted R-squared as -0.038395 and says 99.12% of variation remains unexplained. Neither figure agrees with the table.

An unmeasured indirect effect

The mediation tables add ROE to a return regression containing each driver separately. Beside DPR, ROE has a coefficient of -0.1068 (p = 0.1130). Beside sales growth, it has -0.0967 (p = 0.3875). The indirect effect would multiply the driver-to-ROE path by the ROE-to-return path. We found no regression of ROE on DPR or sales growth, and no Sobel or bootstrap estimate. An insignificant leg, estimated from 136 noisy observations, cannot establish that the product equals zero.

The DPR coefficients do support little mediation. Adding ROE barely changes the coefficient, taking it from -6.10E-08 to -6.09E-08. Sales growth behaves differently: its coefficient declines from -0.6006 to -0.1540. Interpreted as a coefficient change, that decline would suggest partial mediation, contrary to the authors' conclusion. Another inconsistency between specifications is the more likely explanation.

When could a trader act?

Each firm contributes one observation per year (34 x 4 = 136). We found no lag separating fundamentals from returns, nor any account of when returns are measured relative to annual report publication. Without that timing, these regressions describe co-movement. Whether a trader could have acted on the relationship remains unknown. The regressions include no market, size or beta controls. The window spans four years, 2021 to 2024, with nothing held out.

We could not test this ourselves.

Our coverage extends only to US equities. We lack IDX price histories and consumer-goods fundamentals, including data for the 97 firms excluded by the sample criteria.

One result would make the dividend finding worth trading. Rebuild DPR within a range of zero to one, measure returns after report publication, retain the non-payers, and check whether a t near -2.6 survives. Until then, the negative payout coefficient describes 34 surviving dividend payers and a variable whose units we cannot identify.