Indonesia's JCI study treats five annual readings as 225 observations

A trader should distrust p-values of 0.0000 from a regression whose dependent variable has five values. Setyowati, Sriyono and Oktaviyanti report them for the Jakarta Composite Index (JCI, also IHSG), the Indonesia Stock Exchange's all-share index. Their claim is that firm profitability moderates the index's response to macro shocks. The effective count of index readings leaves that claim on thin ground.

Watch the macro series, buy high ROA

The proposed trade follows the paper's macro story. Its introduction says a higher Bank Indonesia policy rate (the BI Rate) makes low-risk instruments more attractive and suppresses interest in stocks. Rupiah moves against the dollar alter exporters' and importers' margins; foreign capital flows change liquidity. The authors add signaling theory at the firm level: companies with higher return on assets (ROA) send a positive signal and should absorb macro shocks better. That leads to a practical reading of the study: watch the macro series and tilt toward high-ROA names.

The authors use 45 LQ45 companies observed annually from 2019 to 2023, producing 225 rows. We did not find a statement saying whether the 45 firms form a fixed list or were point-in-time constituents. JCI is the dependent variable, with the BI Rate, capital flow, the USD/IDR rate and ROA as regressors. Chow, Hausman and Lagrange multiplier tests lead the authors to random effects. Their Moderated Regression Analysis (MRA) adds a product of ROA and each macro series to test moderation. Every slope is positive, with p below 0.05. The interaction estimates are BI_ROA 0.8900 (p 0.0350), CF_ROA 0.7900 (p 0.0240) and KURS_ROA 0.1200 (p 0.0082). KURS is the paper's label for the rupiah exchange rate.

The index has only five readings

JCI supplies one value per year, as do the BI Rate, the rupiah and aggregate foreign flow. Each annual value is copied across 45 company rows. In the descriptive table, JCI has a mean of 6,596.20 and ranges from 5,979 to 7,272 across "225" observations. The BI Rate ranges from 3.5 to 6.0. Capital flow averages exactly 6,000.000, with a range of 2,000 to 12,000; the units are unstated. The exchange rate has both a mean and a median of exactly 14,800.00. Those round figures and the matching mean and median are what repeated annual values produce. The abstract calls the observations "secondary time-series data". The methods say the panel can "simultaneously capture inter-unit and inter-period dynamics", although JCI has no inter-unit variation.

The diagnostics bear this out. The Chow test reports a cross-section F of 0.024220 on (44, 176) degrees of freedom, p 1.0000. Firm dummies explain nothing when the outcome is identical across firms. In the Breusch-Pagan test, the statistic for time effects is 4,812.040, versus 26.96 for cross-section. The residual structure lies almost entirely at the year level.

Count coefficients against the variation available. A constant plus three macro slopes makes four coefficients for five annual index values, leaving one degree of freedom before the three interactions enter. Yet the t-statistics treat the 225 rows as independent draws. Shared annual shocks make conventional standard errors for year-level regressors too small.

Can firm ROA moderate a shared index?

ROA ranges from -4.90 to 14.87 across the 225 firm-years. Within any year, though, all 45 firms have the same JCI outcome.

The cross-firm ROA spread therefore has no cross-firm index variation to explain. Identification of the interactions can only come from the year's average ROA multiplied by each macro series and aligned with five index readings. The claim that "companies with stronger profitability are more resilient" calls for a company-level outcome, such as each firm's return. The paper never uses one.

Two specification problems add to the concern. The stated equation includes ROA as a main effect (β4); the MRA table leaves it out. Each interaction can then absorb variation otherwise attributed to ROA alone. The interpretation of the signs also shifts. The abstract's account of greater resilience implies an interaction opposite in sign to the main effect. The conclusion says profitability "strengthens the influence" of all three variables. Positive interactions alongside positive main effects indicate amplification. Those readings cannot both hold.

The BI Rate sign is especially telling. The introduction says rate hikes suppress equities, while the estimated coefficient is positive in both models: 0.155472 in the basic model and 0.0250 in the moderated one. The authors suggest hikes "can be viewed as a monetary policy signal to maintain economic stability, control inflation, and maintain investor confidence." They conclude that "increases in the BI Rate during the study period were not always negatively responded to by the stock market." Five annual points leave that account dependent on what happened in those five years.

Inflation and the reported tests

The abstract and conclusion say inflation has no significant effect on JCI and that profitability does not moderate its effect. Only the abstract's summary lists inflation as a regressor. The Methods section specifies the BI Rate, capital flow and exchange rate as independent variables; inflation appears neither in the equation nor in Tables 9 and 10.

The basic-model table has arithmetic discrepancies. BI_RATE is 0.155472 with a standard error of 0.030985, yielding t ≈ 5.02 rather than the reported 5.167350. CAPITAL_FLOW is 0.055472 over 0.020985, or about 2.64 against the reported 2.750450. All five coefficients end in 5472, and 0.120985 appears twice as a standard error. The heteroscedasticity table assigns BI_RATE and KURS the same coefficient, 0.060695. The moderation table's arithmetic does reconcile: 0.89 over 0.42 gives the printed 2.119. The authors call classical assumption tests "optional because the characteristics of panel data already address most violations". Their multicollinearity check uses pairwise correlations, reaching 0.603451 between the BI Rate and the rupiah, without including the interaction terms.

Where is the forecast?

Every equation term has the same time subscript t. The estimates describe a contemporaneous annual association, though the paper presents them as useful for predicting JCI movement. We did not find a lagged specification, a holdout year or a portfolio test. The year's rupiah level, for example, is fully known only at year-end, when the index level it "explains" is also printed.

We could not run this ourselves. Our data has no JCI history and no Indonesian macro series.

A rerun using each firm's annual return as the outcome would change my view. Year fixed effects could absorb the macro levels, leaving the ROA interactions to identify the proposed moderation. The ROA main effect would also need to be restored.