Indonesian food and beverage stocks reacted to the free-meal pledge at p = 0.005 when measured against the market. Against their own historical returns, the reaction disappears. Putri and Butar Butar report both results for MBG, the Free Nutritious Meal program. Yet their abstract says "the announcement of the MBG program did not trigger a meaningful market reaction." The benchmark determines which answer a trader gets.

Prices at the pledge, sales after launch

MBG (Makan Bergizi Gratis, the Free Nutritious Meal program) reached the public as a Prabowo-Gibran campaign pledge on 28 November 2023. Policy implementation began on 6 January 2025. The demand argument is straightforward: school kitchens, mostly operated by small businesses, buy rice, meat, milk and processed food, and listed manufacturers supply those inputs. A national feeding program therefore represents a demand shock for those manufacturers. Belief in that story should show up in prices at the announcement, followed by sales after rollout.

The authors examine both. From 84 IDX food and beverage firms, they exclude 2 suspended names, 1 corporate action, 2 beer makers and 6 firms missing interim reports. That leaves 73. The corporate action was JAWA's issuance of 12.45 billion new shares on 22 November 2023, which the authors place inside the window.

For daily abnormal returns, they use a mean-adjusted model, taking each stock's average return over 270 trading days (t-272 to t-3) as its expected return. The five-session event window spans t-2 to t+2 around 28 November 2023. Trading volume activity, daily volume divided by listed shares, is averaged over three days before and three days after. The main fundamentals test compares quarterly sales growth and net profit growth in Q1 and Q2 2025 with the corresponding quarters of 2024. A sensitivity check instead compares them sequentially with Q3 and Q4 2024. Kolmogorov-Smirnov rejects normality throughout, so the authors use Wilcoxon signed-rank tests for every comparison.

None of the mean-adjusted days is significant; at t0, p = 0.968. Volume also shows no change (p = 0.673), with its mean declining from 0.003 to 0.002. Sales growth in Q1 2025 exceeds Q1 2024 (p = 0.048), while Q2 does not (p = 0.343). Neither quarter has a significant YoY difference in profit growth (p = 0.446 and 0.151). The authors attribute the price null to investors discounting an uncertain campaign promise. For the weak fundamentals, they point to kitchens buying locally under a directive from the National Nutrition Agency, Badan Gizi Nasional.

What changes when the benchmark is IHSG?

Subtract the Jakarta Composite Index, IHSG, return instead, and the same 73 stocks produce significant results at t-2 (p = 0.016) and t0 (p = 0.005). This is the market-adjusted model. The authors publish these findings and cite Brown and Warner (1985) for its greater power. They then write "Meskipun demikian, penerimaan hipotesis tetap menggunakan hasil pengujian yang utama": the hypothesis decision remains tied to the main test.

The label "utama" is the paper's only reason for preferring the mean-adjusted result. The market-adjusted finding appears in neither the abstract nor the conclusion. The conclusion instead reports no significant abnormal return difference over the entire event window.

That preference settles H1. Expected returns under the mean-adjusted model average -0.00008, with an SD of 0.002. The main test consequently comes close to asking whether raw returns differed from zero. The market-adjusted version asks whether F&B stocks moved relative to the index. For a desk using an index hedge, the latter question matters.

The information-leakage claim runs into a separate inconsistency. The authors treat significance at t-2 as evidence of leakage, drawing on the very test they set aside for H1. The intervening day, t-1, has p = 0.954. Ten day-by-model tests at the 5% level would produce 0.5 false positives on average under independence. Two hits are suggestive. A leakage explanation still needs more than a day whose neighbour shows nothing.

An 800% day in the return data

These returns need cleaning before anyone trades them. At t+2, mean actual return is 0.164, the SD is 0.990 and the maximum is 7.996. That maximum represents a one-session gain of 799.6%. The minimum at t-2 is -0.899. According to the authors, a handful of stocks account for the t+2 spike. A food stock rising 800% in a session is either a data error or an untradeable name. The original 84-firm population already included plantation companies suspended for 64 and 23 months.

The advertised sample size also overstates what enters the tests: the paper cites 219 observations (73 firms times 3 years), while every test uses N = 73.

Direction is another problem. A Wilcoxon's sign depends on rank sums, calculated as count times mean rank. The tables give only mean ranks. Tabel 11, the market-adjusted sensitivity table, illustrates the consequence: at t-1, the positive mean rank is 47.86, the negative mean rank is 30.24, and p is 0.954. Far fewer pairs must have been positive for that combination to occur.

The authors infer a positive reaction because positive mean ranks exceed negative ones at t-2 (41.55 vs 35.04) and t0 (42.30 vs 35.00). Those comparisons alone cannot establish the direction. At t+2, mean ranks of 9.73 and 14.17 imply that only about 19 to 27 non-zero pairs entered the test out of 73. We did not find an explanation for the missing pairs.

Sales acceleration without a procurement link

Only one of the four sales comparisons supports H3 (p = 0.048). The sequential sales comparisons are significant at p = 0.002 and 0.000, with the authors interpreting their direction as negative. Sequential profit growth from Q4 2024 to Q1 2025 is also significant (p = 0.027); the authors interpret that direction as positive. YoY profit remains insignificant. Both directional readings face the same mean-rank problem.

The authors favour YoY comparisons because they remove seasonality, a defensible choice. But the variables being tested are already growth rates: the mean is 0.223 in Q1 2024 and 0.273 in Q1 2025. The significant result therefore measures acceleration.

Attribution is the larger obstacle. All the listed food companies pass through 2025, and the authors themselves cite cost pressure and commodity inflation to explain the profit finding. Without an untreated group, one quarter's 0.048 cannot be assigned to MBG procurement. Their recommendations acknowledge the gap: focus on direct suppliers (rice, meat, chicken, eggs, milk) and observe more than two quarters after launch. The profit data also calls for cleaning. Its Q2 2025 mean of 16.325 is driven by a maximum of 1178.513, with an SD of 137.920.

A tradeable test starts with suppliers

We could not rerun this.

Our data contains neither IDX prices nor Indonesian fundamentals. Substituting US food stocks would lose the mechanism being tested, Indonesian state procurement.

The authors' recommendations provide the starting point for a tradeable version: identify firm-level exposure through direct suppliers. Cumulate abnormal returns across the window against IHSG or a market model, and replace daily Wilcoxon calls with standardized cross-sectional tests. Names with daily moves of the 7.996 kind would also have to go. Across the six sessions around the announcement, mean turnover fell from 0.3% of listed shares before to 0.2% after. A pre-event SD of 0.014 and maximum of 0.110 show how much a few names contribute to that average.

A market-adjusted cumulative return for the supplier subset, surviving the outlier filter, would change our mind. Until then, this paper supports a benchmark-dependent price reaction and a sales effect confined to one quarter at p = 0.048.