Regional heterogeneity in rice prices in Indonesia: a province-level panel study of surplus and deficit areas

Abstract

Rice prices in Indonesia vary across provinces, yet empirical evidence on how rice price policy instruments operate differently across surplus and deficit markets remains limited. This study investigated whether the determinants of retail rice prices differ between surplus provinces with relatively low prices and deficit provinces with relatively high prices. Using monthly panel data at the operating-area level of Perum BULOG, the state-owned enterprise responsible for managing public rice stocks, the analysis covered 26 provinces from January 2018 to December 2022 (N = 1,560). Provinces were grouped into surplus–low-price and deficit–high-price categories based on average retail prices and a rice adequacy index. Separate fixed-effects panel models with an AR(2) error structure were estimated for each group, relating retail prices to fuel costs, policy instruments, domestic production, macroeconomic conditions, and international rice prices. This study found that the effects of rice price policy instruments differ by market type: the government purchasing price (HPP) is insignificant in surplus provinces but significantly associated with higher retail prices in deficit provinces, whereas public stock (CBP) holdings are associated with lower prices only in surplus provinces. Fuel prices, with a three-month lag, were the strongest cost-push factor in both groups. Higher rice production reduced prices across all provinces, with a more pronounced effect in deficit provinces where supplies were more constrained. These results underscore the need for province-specific rice price policies that reflect heterogeneous market structures and highlight the usefulness of the proposed panel models for short-term price monitoring and policy evaluation under normal market conditions.

1 Introduction

Rice is the main staple food and a vital agricultural commodity in Indonesia, characterized by high per capita consumption and its essential role in maintaining national food security (Warsilah and Wahyono, 2025). Fluctuations in rice prices directly affect household welfare, particularly among low-income households that allocate a substantial proportion of their expenditure to staple foods (Faharuddin et al., 2023). Sharp price increases have been documented to raise the cost of living and poverty incidence significantly, with disproportionate impacts on rural households (Misdawita et al., 2019). Consequently, rice price dynamics attract considerable political attention and have remained a central concern of Indonesian food policy (Ruspayandi et al., 2022).

Despite being a policy priority for decades, the price of medium-quality rice in Indonesia has fluctuated substantially in recent years (Arida et al., 2023; Utami et al., 2023). Monthly retail and producer prices vary markedly over time and across provinces, reflecting a combination of production shocks, cost pressures, and policy interventions (Putra et al., 2021; Diawati and Rasyid, 2025; Theresia et al., 2025). Spatial disparities are evident in Indonesia’s rice markets. Surplus production areas generally record lower retail prices, while deficit regions reliant on inflows from other provinces often face higher prices. This circumstance is consistent with distribution constraints within the country’s archipelagic setting (Aryani et al., 2021). These disparities imply that the burden of rice price shocks is not equally distributed across regions and socio-economic groups (Putra et al., 2021; Fitrawaty et al., 2023).

Due to these fluctuations and spatial disparities, rice prices are highly volatile, leading the Government of Indonesia to implement a multi-instrument rice price policy framework to manage rice prices. This framework encompasses the Government Purchase Price (HPP) and the Retail Price Ceiling (HET), both set by the government, as well as the management of Government Rice Reserves (CBP) by Perum BULOG, a state-owned enterprise responsible for managing public rice reserves and rice market operations (Ruspayandi et al., 2022; National Food Agency of Indonesia, 2025a,b). While these instruments are intended to dampen price volatility at the national level, empirical evidence suggests their effectiveness is heterogeneous and highly dependent on regional context (Putra et al., 2021). Implementation constraints, specifically related to scale, timing, and targeting, can further weaken stabilization outcomes at the local level (Ruspayandi et al., 2022; Mujihartono et al., 2023), indicating that a uniform policy design may generate significantly different impacts across structurally diverse surplus and deficit markets.

These dynamics are reflected in a substantial body of literature that has documented how supply-side factors (such as provincial rice production), demand-side conditions (including income growth and consumption patterns), and broader macroeconomic variables (such as the Consumer Price Index (CPI), fuel prices, exchange rates, and international rice prices) shape domestic rice price dynamics in Indonesia (Putra et al., 2021; Ruspayandi et al., 2022; Fitrawaty et al., 2023). Complementary work on spatial market integration shows that provincial rice markets are linked in the long run, but exhibit heterogeneous patterns of price transmission (Aryani et al., 2021; Ys et al., 2022; Ikhsan et al., 2024). This heterogeneity is consistent with evidence from other developing economies, where transaction costs and infrastructure constraints generate nonlinear price adjustments across spatially separated markets (Alam et al., 2022). In addition, descriptive evidence indicates that Java’s surplus production centers typically record lower, more stable prices, whereas deficit regions, particularly in eastern Indonesia and remote areas, tend to face higher, more volatile prices (World Bank, 2019). These structural differences suggest that national-level models may hide important provincial variation in both price behavior and policy responses (Aryani et al., 2021; Ys et al., 2022; Theresia et al., 2025).

Beyond spatial heterogeneity, broader macroeconomic and external factors also shape domestic rice prices dynamics. Several national-level studies have reported a statistically significant negative relationship between the rupiah–US dollar exchange rate and domestic rice prices, indicating that rupiah depreciation is correlated with lower domestic rice prices over the long run (Fitrawaty et al., 2023). Extreme weather anomalies, such as El Niño events and rainfall variations, significantly affect domestic rice prices by disrupting paddy production levels (Putra et al., 2021). In contrast, international rice prices seem to have a limited effect on domestic prices, owing to Indonesia’s relatively shielded rice market from global price fluctuations (Ruspayandi et al., 2022; Fitrawaty et al., 2023).

Taken together, these supply-, demand-, and macro-level determinants do not operate uniformly across provinces. Disparities in agricultural productivity, infrastructure quality, and market access across regions create differences in price levels and price volatility between surplus and deficit provinces.

In response to these heterogeneous price dynamics, the Indonesian government relies on several policy instruments to manage and stabilize rice prices. At the producer level, the HPP functions as a minimum price floor that protects farmers from steep price drops during the main harvest season and helps sustain their income. At the consumer level, the HET sets an upper limit on retail prices to prevent excessive price increases (National Food Agency of Indonesia, 2025b). In parallel, Perum BULOG manages CBP and conducts market operations by purchasing surplus rice and releasing stocks when supplies tighten. The effectiveness of these interventions is often proxied by the volume of rice released through market operations, especially when retail prices exceed the HET. However, their actual success depends heavily on how quickly stocks are distributed. Empirical studies showed that the impacts of these measures differ across regions and over time (Putra et al., 2021).

These policy differences are further complicated by the uneven spatial integration of rice markets across provinces. Spatial market integration analyses using daily retail prices reveal that many provincial rice markets exhibit long-run price relationships; however, the strength and speed of integration vary considerably across regions, particularly due to distribution disruptions during the pandemic period (Ys et al., 2022). West Java, East Java, Central Java, South Sulawesi, Riau, Jakarta, and Papua are identified as important reference markets (Ys et al., 2022). Furthermore, other studies have specifically designated the Cipinang Rice Wholesale Market (PIBC) in Jakarta, alongside Bandung, Surabaya, Palembang, and Makassar, as key benchmark markets that influence regional price formation (Sinaga et al., 2020).

Further price transmission studies reveal asymmetric adjustment patterns, in which increases in producer prices are transmitted to consumers more rapidly than corresponding decreases (Rahman et al., 2022; Kharisma and Indrawan, 2023). This combination of asymmetric transmission and spatial heterogeneity underscores the need for empirically grounded regional groupings to ensure that the impacts of policy instruments such as HPP and HET are not analyzed under the assumption of homogeneous market conditions.

Many national-level analyses still rely on aggregated data and do not directly compare policy effectiveness between surplus and deficit provinces. Yet because regional production structures can generate asymmetric policy responses, understanding how interventions work under different supply conditions is critical for effective policy design (Mujihartono et al., 2023; Diawati and Rasyid, 2025).

Examining market traits solely at the national level can mask important structural differences across regions. On the supply side, most of Indonesia’s rice production is concentrated on the island of Java, which usually operates under surplus conditions and tends to exhibit relatively low and stable prices. In contrast, deficit regions such as Jakarta, Papua, and many provinces in eastern Indonesia rely heavily on interregional trade. These areas face higher and more volatile prices. Taken together, these contrasts mean that national averages do not capture how rice markets actually behave in different regions. Relying on such aggregates can mislead evaluations of policy performance in provinces with distinct market structures.

National-level models provide useful insights into rice price dynamics in Indonesia. However, they cannot capture how the same policy instruments may generate different price effects across structurally diverse provinces. On the other hand, single-district or city case studies offer rich local detail but lack generalizability to the full range of provincial market conditions. The large, archipelagic geography of Indonesia implies substantial differences in supply and demand conditions across provinces, which in turn shape rice price dynamics and the effectiveness of rice price policies. As a result, policymakers still lack province-level evidence on how the determinants of rice prices operate differently in surplus–low-price provinces compared to deficit–high-price provinces.

Most existing studies either focus on national-level determinants of rice prices using aggregate data (Fitrawaty et al., 2023) or examine case studies of specific regions, which may be less representative of national price dynamics (Mgale et al., 2021; Chitete et al., 2025). Several papers have documented inter-regional price linkages using cointegration-based time-series models, such as VECM or panel VECM, to analyze spatial and vertical market integration and price transmission in Indonesian rice markets (Aryani et al., 2021; Ys et al., 2022). However, they do not explicitly assess the effects of rice price stabilization policies across different regional contexts, such as structurally surplus–low-price and deficit–high-price provinces. Despite this growing body of evidence, no study to date has systematically compared the impacts of key rice price stabilization instruments between surplus–low-price and deficit–high-price provinces within a unified empirical framework.

Recent work in other rice-dependent countries has begun to explore regional heterogeneity in rice price dynamics. Antonio et al. (2025) estimated a panel vector autoregression model for the Philippines and compared impulse-response functions between rice-sufficient and rice-insufficient regions, finding differential effects of world price and climate shocks. Similarly, Dalheimer et al. (2026) assessed the relative roles of public and private rice inventories in buffering price volatility in the Philippines, demonstrating that the effectiveness of stock-based interventions varies across market contexts. However, these studies rely on VAR-based frameworks that treat all variables as endogenous and do not isolate the province-level effects of specific policy instruments such as a government purchase price or public stock holdings. Moreover, no comparable analysis exists for Indonesia, where BULOG operates a distinct multi-instrument rice price policy framework across structurally diverse surplus and deficit markets.

Existing national studies typically model rice price policies using aggregate time-series frameworks or optimization–regression models, such as ARDL specifications for national retail prices (Ruspayandi et al., 2022) or hybrid models linking retail prices to import and stock decisions (Diawati and Rasyid, 2025), estimated at the national rather than province level in a panel setting. Compared to ARDL or VAR/VECM time-series models estimated at the national or market-pair level, panel fixed-effects approaches with autoregressive error correction offer a more appropriate econometric basis for province-specific policy inference. They can jointly control for unobserved provincial heterogeneity and low-order serial correlation in monthly prices.

In this context, this study connects national aggregate analyses with local case studies using a province-level panel data approach that captures both spatial heterogeneity and temporal change. It combines province-level grouping, panel modeling, and out-of-sample forecasting to explore different policy transmission mechanisms and to develop more region-specific strategies for rice price policy design. By explicitly distinguishing surplus–low-price from deficit–high-price provinces, this study addresses these gaps and provides empirical evidence to inform more place-based rice-price policies in Indonesia.

Building on this framework, the research offers three novel contributions that extend beyond existing national-level analyses (Ruspayandi et al., 2022; Fitrawaty et al., 2023; Diawati and Rasyid, 2025) and recent regional studies in other countries (Antonio et al., 2025; Dalheimer et al., 2026). First, it delivers an empirically based categorization of Indonesian provinces into surplus–low-price and deficit–high-price groups. Second, it estimates distinct policy transmission mechanisms for each group using fixed-effects models with AR(2) error correction, explicitly accounting for unobserved provincial heterogeneity and serial correlation. Unlike panel VAR methods used by Antonio et al. (2025), this approach enables direct estimation of individual policy instrument effects across market types. Third, it provides empirical evidence supporting province-specific rice price policy strategies that recognize provincial heterogeneity in market structure and policy responsiveness, laying the foundation for more place-based policy design.

2 Data and methods

This study used monthly panel data at the BULOG operational-area level. The dataset covered 26 operational units observed over 60 months, from January 2018 to December 2022, resulting in 1,560 area–month observations. Some operational areas are identical to administrative provinces, while others combine two provinces (for example, South Sulawesi with West Sulawesi, and Papua with West Papua). Table 1 presents the definitions, units, and sources of the variables, while Table 2 reports descriptive statistics for all variables in the panel dataset. These variables cover key aspects of production, costs, macroeconomics, and policy related to rice price trends.

2.1 Provincial grouping based on price disparity and rice adequacy

Provinces were classified into two structural groups based on key characteristics of the rice market. The classification relied on two indicators: (i) the average retail price of medium-quality rice (MEDPRICE) over 60 months (January 2018–December 2022), and (ii) the rice adequacy index. The rice adequacy index was calculated as the ratio of rice production to consumption requirements (based on population and per capita consumption), averaged over the same period. An index value greater than 1 indicates surplus conditions, while a value below 1 indicates a deficit. Both variables were carefully normalized using z-scores prior to analysis to ensure comparability across provinces with different scales.

Provinces with relatively low average prices and adequacy indices above 1 were classified as surplus–low-price provinces, whereas provinces with relatively high average prices and adequacy indices below 1 were classified as deficit–high-price provinces. The difference of the two groups was assessed using one-way ANOVA on average prices and adequacy indices, confirming that the groups differ statistically in both variables. Effect size estimates confirm that these differences are substantively significant: partial η² = 0.46 for average rice prices and partial η² = 0.52 for the rice adequacy index, both exceeding the conventional threshold of 0.14 for large effects (Cohen, 1988), indicating that group membership accounts for approximately 46 and 52% of the total variance in each variable, respectively.

2.2 Model specification and estimation

For each provincial group, the empirical rice price model is estimated separately using the following fixed-effects panel specification.

The model was estimated using a panel fixed-effects approach. This specification is deemed appropriate as the dataset encompasses the entire population of provincial operational units, enabling the model to capture unobserved province-specific characteristics. To address potential serial autocorrelation in the residuals, which can arise in monthly time-series data, the error term was modeled using an autoregressive (AR) process. Specifically, the error term adheres to an AR(2) structure.

The parameters β and ρ were estimated separately for each identified provincial group to capture varying price dynamics across different provincial groups. This approach ensures that the results reflect the specific price determinants inherent in each group. Panel fixed-effects estimations were conducted using EViews version 13. Lag selection for key variables and the choice of the AR order are detailed in Sections 2.4 and 2.5.

2.3 Endogeneity considerations and panel granger causality

A key interpretive limitation of the fixed-effects OLS specification concerns the potential endogeneity of BULOG policy variables. PROC, SUPPLY, and STOCK are operational responses to prevailing market conditions: BULOG tends to procure more rice when prices are low during harvest seasons and intensifies distribution when prices rise. This simultaneity implies that the estimated coefficients on these variables may reflect associations rather than pure causal effects. Fixed-effects estimation controls for time-invariant provincial characteristics but cannot fully resolve the simultaneity between prices and BULOG’s reactive policy choices.

To provide empirical evidence on the direction of temporal precedence between rice prices and BULOG operational variables, we conducted Pairwise Dumitrescu and Hurlin (2012) Panel Granger Causality tests for each group separately. The Dumitrescu-Hurlin test was selected over the standard panel Granger test because it allows for heterogeneous causal relationships across cross-sectional units, consistent with the heterogeneous panel structure assumed throughout this study. Prior to testing, panel unit root tests (LLC, IPS, Fisher-ADF, and Fisher-PP) were conducted for all variables. Results confirmed that MEDPRICE, SUPPLY, and PROC are stationary at level I(0) in both groups. STOCK was stationary at level in Group 1 but required first differencing to achieve stationarity in Group 2; accordingly, D(STOCK) was used in the Granger test for Group 2, and results for this variable in Group 2 are interpreted in terms of changes in stock holdings rather than stock levels. The Granger causality tests used two lags, consistent with the AR(2) error structure of the main model.

The Granger causality results are reported alongside the main regression findings in Section 3.4.3 and inform the interpretive framework applied throughout the discussion and policy implications.

2.4 Lag selection

The selection of suitable lag lengths for the variables in this study was carried out through a two-step process: visual identification with the Cross-Correlation Function (CCF) and statistical validation using the Akaike Information Criterion (AIC). Before the analysis, the data were transformed into monthly changes (Δ) to ensure stationarity and avoid misleading correlations.

The CCF analysis of changes in fuel prices (ΔFUEL) and rice prices (ΔMEDPRICE) revealed statistically significant contemporaneous correlations and those up to a three-month lag. As illustrated in Figure 1 (representing Group 1), the peak of statistically significant correlation occurs at lag 3 (r = 0.25), which exceeds the 95% significance threshold by ± 0.086. A similar pattern is observed for Group 2, with a 3-month lag also yielding the highest correlation (r = 0.179). The CCF between ΔPROD (or ΔSUPPLY) and ΔMEDPRICE peaks at lag 1 month, supporting a one-month lag for these variables. The regular patterns across groups indicate that energy costs and supply-side shocks typically affect retail prices within one to 3 months.

Figure 1

Formal AIC tests further validate this consistency. As presented in Table 3, the model specification incorporating FUEL_t-3 yields the minimum AIC values: 12.3403 for Group 1 and 12.5129 for Group 2, compared to other lag structures. Furthermore, for the production (PROD) and distribution (SUPPLY) variables, the CCF results confirm that a one-month lag (t-1) is the optimal specification. Implementing this lag structure ensures that the final model accurately captures price transmission patterns consistent with observed market dynamics.

2.5 Autoregressive correction and serial correlation assessment

To determine the necessity and order of autoregressive correction, the residual correlograms (ACF and PACF) of the initial specification, without AR components, were analyzed. The pattern of residual autocorrelation showed an unaddressed low-order autoregressive structure. Consequently, AR(1) and AR(2) terms were incorporated into the error process of the primary model. The adequacy of this specification was subsequently evaluated by re-inspecting the residual ACF-PACF and the Ljung-Box test; detailed autocorrelation assessment results are presented and discussed in Section 3.2.1. On this basis, the AR(1) and AR(2) error specification was adopted as the baseline model for subsequent inference.

2.6 Forecast accuracy metrics

To assess the accuracy of short-term forecasts, three common error metrics were used: root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). Let denote the actual rice price at time t, the corresponding model forecast, and T the number of out-of-sample forecasts. The three metrics are defined by the following equations.

These forecast accuracy measures were computed for the rolling out-of-sample exercise described in Section 3.3.

3 Results and discussion

3.1 Group characteristics

This subsection summarizes the main structural characteristics of the two provincial groups defined in Section 2.1. Figure 2 plots each province in a two-dimensional space using the z-scores of average retail rice prices and the rice adequacy index. A dashed line was added to show a simple separation between provinces that are surplus–low-price and those that are deficit–high-price. The pattern in the figure showed that the two groups occupy fairly distinct areas of the diagram, suggesting systematic differences in both price levels and rice sufficiency. Group 1 contains provinces with relatively low prices and high adequacy indices (surplus–low-price), while Group 2 contains provinces with relatively high prices and low adequacy indices (deficit–high-price). Table 4 presents the list of provinces in each group.

Figure 2

A one-way ANOVA was used to test whether provincial averages of rice prices and the rice adequacy index, measured on their original scales, differ significantly between the two groups, with group membership as the factor. The results showed statistically significant differences for both variables (p < 0.01), supporting partitioning the provinces into two distinct groups. Effect size estimates further confirm that these differences are substantively significant, with partial η² = 0.46 for average rice prices and partial η² = 0.52 for the rice adequacy index, both indicating large effects. Table 5 reports the summary statistics for each group and forms the basis for estimating separate price models for Groups 1 and 2 in the subsequent analysis.

3.2 Regression results for group 1 and group 2

This section presents the findings from the fixed effects panel model estimation, including an AR(2) error correction term, for both the surplus–low-price group (Group 1) and the deficit–high-price group (Group 2). Table 6 shows the estimated coefficients and standard errors for each group, indicating that both group-specific models fit well, with R-squared values around 0.98 and regression standard errors of about IDR 113 per kg for Group 1 and IDR 124 per kg for Group 2. The AR(1) and AR(2) terms were significant in both models. Durbin-Watson statistics of 1.88 and 1.89 showed effective control of first-order autocorrelation.

3.2.1 Residual autocorrelation assessment

Residual autocorrelation was assessed to evaluate whether the AR(1) and AR(2) error specifications adequately captured the dependence structure. Figure 3 shows the ACF and PACF of the residuals from the initial fixed-effects model without AR components. The ACF decreased gradually, and PACF spiked at lags 1 and 2 in both groups. This pattern indicates the presence of an unmodeled low-order autoregressive structure, and the Ljung–Box Q statistics reject the null hypothesis of no serial correlation at several lags.

Figure 3

After adding AR(1) and AR(2) terms to the error process, the residual correlograms in Figure 4 revealed small and irregular ACF and PACF values at low lags for both groups. In Group 1, the Ljung–Box tests did not reject the null of no residual autocorrelation, so the residuals can be treated as approximately white noise. In Group 2, some Ljung–Box statistics were significant at higher lags, but the associated autocorrelations were small and showed no clear pattern, indicating weak remaining serial dependence. Overall, these results suggest that the AR(1)–AR(2) error specification is adequate for drawing inference in both group-specific models.

Figure 4

3.3 Forecast performance

Short-term forecast accuracy was evaluated using three standard error measures: RMSE, MAE, and MAPE, as defined in Section 2.6. We applied a rolling out-of-sample forecasting scheme for January–December 2022: in each month, the model was re-estimated using data from January 2018 up to the previous month and then used to predict rice prices 1 month ahead for each group. Table 7 displays the monthly RMSE, MAE, and MAPE for Group 1.

To assess whether the policy and macroeconomic covariates included in the main model provide meaningful predictive value beyond pure price persistence, forecast errors from the main model were compared against those from a benchmark AR(2)-only fixed-effects model, estimated with the same panel structure but excluding all policy and macroeconomic covariates. The Diebold-Mariano (DM) test was used to assess the statistical significance of differences in forecast accuracy between the two models, with the loss differential defined as the squared forecast error of the main model minus that of the AR(2) benchmark.

For Group 1, monthly MAPE values generally remained below 1% throughout most of the year; however, the average for 2022 was approximately 1.08% due to a significant error in December. Forecast errors increased markedly from September to December 2022, coinciding with the adjustment of subsidized fuel prices, precisely the period when an early warning system would be most needed. The AR(2) benchmark model produced a similar pattern, with an average MAPE of 1.11% and average RMSE of 146.43, compared to 1.08% and 140.84 for the main model. The DM test indicates no statistically significant difference in forecast accuracy between the two models for Group 1 (t = 0.267, p = 0.790), suggesting that the policy and macroeconomic covariates do not provide additional predictive power beyond pure autoregressive dynamics in surplus provinces over this forecast horizon. For Group 2, the average MAPE was approximately 0.79%, with relatively small errors in most months and larger deviations during the price spike at the end of 2022 (Table 8). The AR(2) benchmark produced an average MAPE of 0.80% and average RMSE of 140.84, closely matching the main model. The DM test for Group 2 indicates that the main model’s squared forecast errors are statistically larger than those of the AR(2) benchmark (t = 2.165, p = 0.032), suggesting that in deficit provinces, the additional covariates introduce estimation noise that marginally reduces one-step-ahead predictive accuracy relative to the simpler benchmark. Table 9 summarizes the average forecast accuracy metrics and DM test results for both groups.

Taken together, these results point to an important distinction between the explanatory and predictive roles of the main model. The policy and macroeconomic covariates, particularly fuel prices, HPP, and BULOG operational variables, are essential for understanding the mechanisms through which rice price determinants operate differently across surplus and deficit provinces. However, they do not consistently improve short-term forecast accuracy over a simple autoregressive benchmark. Overall, the group-specific models are better suited to routine monitoring and policy analysis than to generating superior one-step-ahead price forecasts. Their primary value lies in understanding the direction and magnitude of individual variable effects, rather than in short-term prediction, particularly during periods of large structural shocks such as sudden fuel price adjustments.

3.4 Interpretation of the model and economic mechanisms

The central question of this study is whether the same price determinants and policy instruments have different effects across structurally distinct provincial markets. The group-specific estimates above suggest that several key variables do, in fact, behave differently in surplus–low-price and deficit–high-price provinces. Before discussing individual coefficients, it is useful to recall that PROC, SUPPLY, and STOCK are policy variables that BULOG adjusts in response to prevailing market conditions. Fixed-effects OLS controls for time-invariant provincial characteristics but cannot fully resolve possible simultaneity between prices and BULOG’s policy choices. Consequently, the coefficients on these variables are interpreted as associations rather than strict causal effects in the discussion that follows.

3.4.1 Group 1 (surplus–low-price)

The estimation results for Group 1 explained 98.3% of the variation in rice prices (R² = 0.983), with a standard error of approximately IDR 113 per kilogram and a Durbin–Watson statistic of 1.88, indicating well-controlled first-order serial correlation. For comparison, a fixed-effects specification without autoregressive errors yielded an R² of approximately 0.94 and a regression standard error of roughly IDR 218 per kilogram. Therefore, the improvement in fit primarily reflects strong dynamic persistence in monthly rice prices, captured by the AR(1) and AR(2) terms. Consequently, the high R² primarily captured province-specific heterogeneity and serial dependence, while the policy and macroeconomic covariates explained the remaining within-province variation over time. The AR(1) and AR(2) coefficients were significant, indicating that the autoregressive error specification adequately captures low-order serial dependence in monthly rice prices. Furthermore, Ljung–Box tests up to lag 12 revealed no significant residual autocorrelation, confirming that the residuals are approximately white noise.

From an economic perspective, four variables were significantly associated with retail rice prices in surplus–low-price provinces: fuel prices (lagged 3 months, FUEL_t-3), BULOG distribution (lagged 1 month, SUPPLY_t-1), rice production (lagged 1 month, PROD_t-1), and CBP stocks (STOCK). In contrast, the Government Purchase Price (HPP), Consumer Price Index (CPI), international rice price (IRP), exchange rate (EXCH), and BULOG procurement (PROC) showed no significant correlation with retail prices within this group.

The fuel price, a three-month lag (FUEL_t-3), had a relatively large impact on retail rice prices in surplus–low-price provinces. An increase of IDR 1,000 per liter in fuel prices was associated with a rise of approximately IDR 158 per kilogram in retail rice prices, roughly 3 months later (coefficient 0.158, p < 0.01). This lag structure appears to reflect the time needed for higher transportation and milling costs to spread through the supply chain and finally affect retail markets. The coefficient, the largest among the significant variables, indicates that fuel prices represent the strongest cost-push factor in short-term rice price dynamics in surplus provinces.

Rice production with a one-month lag (PROD_t-1) had a statistically significant, though economically modest, effect on retail rice prices in surplus–low-price provinces. An increase of one ton in provincial rice production per month was associated with a decline of approximately IDR 0.061 per kilogram in rice prices one month later (coefficient −0.000061, p < 0.05). The small magnitude of this effect suggests that supply shocks exert only limited downward pressure on retail prices in these already well-supplied and competitive markets.

BULOG’s rice distribution, a one-month lag (SUPPLY_t-1), exhibited a small, yet statistically significant, association with retail prices in surplus–low-price provinces. An increase of one ton in BULOG distribution in the previous month was associated with a rise of approximately IDR 0.0007 per kilogram in rice prices (coefficient 0.00069, p < 0.05). This positive sign indicates a reactive distribution pattern, corroborated by Granger causality tests showing that price movements temporally precede distribution decisions, where BULOG tends to release more rice after prices have risen, rather than proactively in anticipation of future shocks. The small size of this association indicates that, even with large distribution volumes, BULOG’s distribution does not meaningfully offset the price pressures already present in surplus markets.

The level of CBP stocks (STOCK) exhibited a similarly small association with retail prices. An additional ton of CBP stocks held by BULOG was associated with a decrease of approximately IDR 0.0007 per kilogram in rice prices (coefficient −0.00070, p < 0.01). The small effect size and the observed relationship may partly reflect reverse causality, as BULOG typically accumulates stocks during harvest seasons when prices are already low. The stock variable is therefore more plausibly viewed as a reflection of plentiful market conditions during harvest time, rather than as an external policy tool that independently pushes prices downward.

The interpretation of BULOG policy variables as associations rather than causal effects is supported by Dumitrescu-Hurlin panel Granger causality tests conducted for Group 1. Results consistently indicate that MEDPRICE Granger-causes STOCK (p < 0.001), SUPPLY (p = 0.004), and PROC (p < 0.001), whereas none of the three BULOG variables Granger-cause MEDPRICE (p = 0.359, p = 0.863, and p = 0.218, respectively). This unidirectional pattern confirms that price movements temporally precede BULOG operational responses in surplus–low-price provinces, consistent with a reactive rather than proactive operational mode.

Several variables did not exhibit statistically significant effects on retail rice prices in surplus provinces, namely HPP, CPI, IRP, EXCH, and PROC. The insignificant HPP is particularly noteworthy, as it suggests that in competitive, well-supplied surplus markets, the government-set producer floor price does not automatically translate into higher retail prices. The lack of a CPI effect indicates that rice prices in these provinces are mainly influenced by local supply and demand conditions and cost-related factors rather than broader macroeconomic inflation. The insignificance of IRP and EXCH further underscores the heavily regulated and domestically oriented nature of Indonesia’s rice market, where imports are tightly controlled and domestic policy interventions play a more prominent role in shaping prices.

3.4.2 Group 2 (deficit–high-price)

The model for Group 2 explained 98.6% of the variation in rice prices (R² = 0.986), with a standard error of approximately IDR 124 per kilogram and a Durbin–Watson statistic of 1.89. Similar to Group 1, much of this fit reflects province-specific effects and strong dynamic persistence in monthly prices, captured by the AR terms. Policy and macroeconomic variables accounted for the remaining within-province variation over time. The residual diagnostics were generally satisfactory, although Ljung–Box tests at higher lags remained significant at low levels of autocorrelation, suggesting weak serial dependence in the residuals unlikely to affect inference materially. In deficit–high-price provinces, four variables were statistically significant: the fuel price with a three-month lag (FUEL_t-3), the Government Purchase Price (HPP), BULOG’s rice distribution with a one-month lag (SUPPLY_t-1), and rice production with a one-month lag (PROD_t-1). Conversely, CPI, IRP, EXCH, PROC, and STOCK were not significant. This pattern differs markedly from Group 1, particularly in the significance of HPP and the lack of significance of CBP stocks.

The fuel price, a three-month lag (FUEL_t-3), had a positive and statistically significant impact on retail rice prices. An increase of IDR 1,000 per liter in fuel prices was associated with a rise of approximately IDR 93 per kilogram in rice prices 3 months later (coefficient 0.093, p < 0.01). The direction of this effect matched that in Group 1; however, its magnitude was smaller, making up about 59% of the effect in the surplus–low-price group. This difference reflects the basic characteristics of deficit provinces. These areas rely heavily on trade with other provinces. In these provinces, logistics costs, such as fuel, accounted for a large share of the final retail price. Nevertheless, due to the availability of supplies from multiple origin provinces and the presence of competing suppliers, the pass-through of fuel costs to consumers is only partial. This is evidenced by the coefficient in deficit provinces (0.093) being approximately 41% smaller than that in surplus provinces (0.158).

The HPP had a statistically significant effect on retail rice prices in deficit–high-price provinces. An increase of IDR 1,000 per kilogram in HPP was associated with a rise of approximately IDR 68 per kilogram in retail rice prices (coefficient 0.068, p < 0.01). This pattern contrasted with Group 1, where HPP was not significant (coefficient 0.011, p = 0.801). Economically, deficit provinces depend heavily on interprovincial shipments, with an increase in HPP raising suppliers’ costs in these markets. Given limited local production capacity and dependence on imported supplies from other provinces, suppliers have little choice but to pass the increased HPP-induced costs onto consumers. This pass-through pattern is consistent with evidence of asymmetric price transmission in rice markets, where upstream cost increases are transmitted more readily and rapidly through intermediary traders to retail consumers than cost decreases (Rahman et al., 2022; Arida et al., 2023; Kharisma and Indrawan, 2023). The partial pass-through rate of approximately 6.8% (IDR 68 per IDR 1,000 increase in HPP) reflects the indirect nature of HPP’s influence on retail prices, moderated by competition along the supply chain. In contrast, in surplus provinces, higher HPP does not necessarily lead to higher retail prices, as plentiful local production can meet demand without depending on more costly inter-provincial inflows. In surplus markets, HPP mainly serves as an income support for producers, while in deficit markets, it becomes a significant factor affecting costs.

Rice production with a one-month lag (PROD_t-1) had a negative and statistically significant effect on retail rice prices in deficit–high-price provinces. An increase of one ton in provincial rice production per month reduced rice prices by approximately IDR 0.100 per kilogram 1 month later (coefficient −0.000100, p < 0.05). This effect was roughly 64% larger than that observed in Group 1, reflecting a distinct transmission mechanism operating in deficit markets. Unlike surplus provinces where additional production adds to an already adequate supply base, production increases in deficit provinces are more likely to reduce reliance on interprovincial inflows. Such inflows typically carry additional transportation and logistics costs, so reduced dependence on them generates a larger price-reducing effect at the retail level.

BULOG’s rice distribution with a one-month lag (SUPPLY_t-1) was positively and significantly associated with retail prices in deficit–high-price provinces (coefficient 0.00112, p < 0.05). The positive sign of this coefficient should not be interpreted as distribution causing prices to rise. Rather, it reflects the reactive nature of BULOG operations: distribution volumes increase in response to prices that have already risen in the previous period, producing a positive statistical association between lagged distribution and current prices. This interpretation is directly supported by the Granger causality results reported below, which confirm that price movements precede distribution decisions rather than the reverse. The larger coefficient relative to Group 1 (0.00112 vs. 0.00069) reflects the more volatile price environment in deficit markets, where larger price spikes trigger stronger BULOG distribution responses.

Panel Granger causality tests for Group 2 yield an identical directional pattern. MEDPRICE Granger-causes D(STOCK) (p < 0.001), SUPPLY (p = 0.017), and PROC (p < 0.001), while none of the BULOG variables Granger-cause MEDPRICE (D(STOCK): p = 0.982; SUPPLY: p = 0.200; PROC: p = 0.929). The use of first-differenced STOCK for Group 2 reflects its I(1) properties in deficit provinces, and the result is interpreted as changes in stock holdings not preceding price movements. Taken together, these results confirm the reactive nature of BULOG operations in deficit–high-price provinces and reinforce the associative interpretation of coefficients throughout this study.

The variables without statistically significant effects in Group 2 were CPI, IRP, EXCH, PROC, and STOCK. The insignificance of STOCK in deficit provinces indicates that, in the data, changes in CBP holdings in these provinces were not systematically linked to local rice prices, which may reflect logistical and implementation constraints on using stocks for price stabilization. The insignificance of the international rice price (IRP) and the exchange rate (EXCH) confirms that, even in deficit provinces, domestic rice prices are largely shielded from global market shocks. Additionally, the insignificance of PROC shows that BULOG’s procurement mainly responds to low prices, rather than serving as an external policy tool that independently affects market outcomes.

3.4.3 Comparative analysis of key heterogeneities between groups

The regression results revealed several important differences in price transmission mechanisms between surplus provinces with low prices and deficit provinces with high prices. A first notable heterogeneity concerns the effectiveness of HPP. HPP was statistically significant only in Group 2 (deficit–high-price). This asymmetry highlights the market structure in deficit provinces, where inter-provincial supplies cannot be easily replaced by local production, leading to a higher pass-through of embedded costs into retail prices. Conversely, surplus provinces are characterized by more competitive supply conditions, which dampen the pass-through of HPP adjustments to retail prices. Therefore, HPP adjustments should be carefully designed to act as an effective price-shaping tool in deficit markets, while mainly functioning as an income-support mechanism in surplus markets.

A second key heterogeneity concerns the sensitivity of prices to local production. Production sensitivity was evident, as the price-reducing effect of increased rice production was approximately 64% greater in deficit provinces (coefficient −0.000100) than in surplus provinces (−0.000061). This difference reflects the limited local supply capacity in deficit provinces. Therefore, increasing production in deficit markets results in a larger increase in available supply than in surplus markets, which are already well-stocked. Consequently, strengthening local rice production capacity in deficit provinces is essential for easing price pressures in those areas.

A third important difference arises in the response to fuel price shocks. Fuel price shocks showed a positive and statistically significant coefficient in both surplus and deficit provinces. However, the magnitude of the effect was greater in surplus provinces (0.158) than in deficit provinces (0.093), representing approximately a 41% difference. This pattern diverges from international evidence indicating that elevated fuel and transportation costs are generally more pronounced in deficit provinces dependent on long-distance trade. In East Africa, Dillon and Barrett (2016) found that the elasticity of local food prices with respect to world oil prices tends to increase in markets far from seaports, while Li et al. (2025) demonstrated that higher local gasoline prices widen interprovincial food price disparities through transport-cost channels in China. The finding in this study, that fuel price shocks are reflected more strongly in rice prices in surplus–low-price provinces, can therefore be viewed as a distinctive empirical pattern that differs from existing evidence in other countries. One tentative explanation is that surplus provinces may function as major milling and distribution hubs rather than final consumption points. However, this mechanism cannot be verified without disaggregated supply chain data at the district level. The counter-intuitive direction of the fuel price differential should therefore be interpreted with caution and treated as an empirical anomaly that warrants further investigation rather than a confirmed mechanism. Future research incorporating district-level price data, inter-provincial trade flow data, or a spatial lag specification, in which prices in neighboring surplus provinces enter the deficit province model directly, could help disentangle the transmission channels more rigorously.

Regarding BULOG’s operations, the estimates indicate distinct roles for procurement, distribution, and stockholding. The effect of BULOG’s procurement (PROC) on prices was not significant in either group, fitting with its mandate to buy rice mainly during harvest periods when prices are relatively low, rather than to move market prices directly. Rice distribution (SUPPLY_t − 1) was positively associated with prices in both groups, suggesting a largely reactive pattern in which BULOG increases distribution once prices are already high. CBP stocks (STOCK) were negatively associated with prices only in Group 1 but this association was not significant in Group 2. This result indicates that higher stock holdings are associated with lower prices in surplus provinces but show no significant price association in deficit provinces. Overall, BULOG’s interventions seem to work more as an income-stabilization mechanism for producers through counter-cyclical procurement than as a strong consumer price-stabilization tool at the retail level across provinces.

The reactive nature of BULOG’s operations is further corroborated by Dumitrescu-Hurlin panel Granger causality tests conducted separately for each group. Across both Group 1 and Group 2, the direction of Granger causality runs consistently from MEDPRICE to each of the three BULOG variables, namely STOCK, SUPPLY, and PROC, with no causality running in the reverse direction. Specifically, in Group 1, MEDPRICE Granger-causes STOCK (p < 0.001), SUPPLY (p = 0.004), and PROC (p < 0.001), while STOCK (p = 0.359), SUPPLY (p = 0.863), and PROC (p = 0.218) do not Granger-cause MEDPRICE. An identical pattern holds in Group 2: MEDPRICE Granger-causes D(STOCK) (p < 0.001), SUPPLY (p = 0.017), and PROC (p < 0.001), whereas D(STOCK) (p = 0.982), SUPPLY (p = 0.200), and PROC (p = 0.929) do not Granger-cause MEDPRICE. This consistent unidirectional pattern across both market types provides empirical support for treating the estimated coefficients on BULOG variables as conditional associations rather than causal effects, and confirms that BULOG’s procurement, distribution, and stockholding activities respond to price signals rather than independently driving retail price outcomes.

The behavior of macroeconomic variables and international prices further highlights how insulated Indonesia’s rice market is. The Consumer Price Index (CPI), the exchange rate (EXCH), and the international rice price (IRP) are insignificant in both groups, suggesting that domestic rice prices are only weakly linked to broader macroeconomic and global shocks. This pattern is consistent with findings from other heavily regulated rice markets. Strong government interventions such as robust public stockholding (Dalheimer et al., 2026), combined with low import dependence (Ruspayandi et al., 2022; Qi et al., 2025), have been shown to limit international price pass-through in comparable settings. It reflects a price formation process driven mainly by domestic production and local supply–demand conditions rather than by movements in general prices or in world rice markets.

Overall, the group-specific models illustrate a key principle: the same policy instruments can be linked to different price outcomes depending on market structure. HPP is associated with higher retail prices only in deficit markets where supplies are tight, and increases in production have their strongest price-reducing effects precisely in deficit provinces with limited local supply. In contrast, BULOG’s stocks are associated with lower prices only in surplus provinces and show no detectable price effect in deficit markets, underlining the uneven spatial reach of stock-based interventions. Fuel price shocks, meanwhile, remain a common cost-push driver in both groups. These patterns support a location-specific approach to rice price policy rather than a single, uniform national strategy.

3.4.4 Systemic interactions among price determinants

The regression results reveal that rice price formation is not a product of isolated variable effects, but rather reflects a complex interplay among cost-push drivers, local supply conditions, and policy instruments. Three systemic interactions are particularly noteworthy.

First, the joint behavior of HPP and CBP stocks across groups reveals an unintended policy consequence that is particularly pronounced in deficit markets. HPP was originally designed as a producer price floor to protect farmer incomes during harvest seasons, not as a retail price instrument. However, in deficit provinces where retail rice is sourced predominantly from interprovincial inflows, HPP increases are significantly associated with higher retail prices (coefficient 0.068, p < 0.01), indicating that the cost of producer price support is effectively passed through the supply chain and borne by consumers. Critically, this upward pressure is not offset by public stock holdings, which show no significant price association in deficit provinces (coefficient −0.00101, p = 0.167), suggesting that stock-based interventions have not been effective in cushioning consumers against HPP-induced price increases in these markets. The contrast with surplus provinces is notable: in Group 1, HPP shows no significant association with retail prices (p = 0.801), and CBP stocks are negatively associated with prices (coefficient −0.00070, p < 0.01), indicating that under surplus conditions the two instruments work in a complementary fashion, where HPP supports producer incomes without raising retail prices, while stocks help moderate price levels. In deficit provinces, however, this complementarity does not hold: HPP raises retail prices while stocks fail to provide a countervailing buffer, producing a net adverse outcome for consumers, an effect that falls outside the stated stabilization objectives of the policy framework.

Second, the interplay between fuel prices and BULOG distribution reveals a systematic temporal coordination gap with practical policy implications. Fuel price shocks take 3 months to fully transmit to retail prices, a stable lag structure that, in principle, creates an identifiable window for anticipatory policy intervention before price increases materialize at the retail level. However, the positive association between BULOG’s lagged distribution and current prices, confirmed by Granger causality tests showing that price movements precede distribution decisions, indicates that BULOG consistently begins intensifying distribution only after retail prices have already risen. Rather than utilizing the three-month transmission window to deploy stocks proactively, BULOG’s operational pattern suggests that distribution responses are triggered by realized price increases rather than anticipated cost pressures. This means that by the time BULOG responds, the fuel-driven price increase has already been absorbed by consumers, rendering distribution a lagging rather than a buffering instrument against energy cost shocks.

Third, a comparison of the relative magnitudes of supply-side and cost-push effects reveals the dominance of energy costs over local production in shaping retail prices. To assess their relative importance on a comparable scale, we evaluate the retail price impact of a one-standard-deviation change in each variable. In Group 1, a one-standard-deviation increase in fuel prices (IDR 1,087 per liter) is associated with a retail price increase of approximately IDR 172 per kilogram. By contrast, a one-standard-deviation increase in monthly production (227,582 tons) reduces prices by only approximately IDR 14 per kilogram. The cost-push effect is thus roughly 12 times larger than the supply-side effect in surplus provinces. In Group 2, the corresponding figures are IDR 101 per kilogram for fuel and IDR 11 per kilogram for production, yielding a similar ratio of approximately 9 to 1. These magnitudes indicate that even in well-supplied surplus markets, the price-stabilizing benefits of additional production are systematically outweighed by fuel-driven cost pressures. This pattern is consistent across both market types.

3.5 Policy implications

The empirical heterogeneity documented in Section 3.4 shows clear differences in rice price dynamics and policy transmission between surplus–low-price and deficit–high-price provinces. This section draws out the main policy implications, focusing on fuel price lags, HPP, BULOG’s operations, and production sensitivity as inputs for more province-specific rice price policies.

3.5.1 Fuel price shocks and anticipatory coordination

We found that fuel prices, with a three-month lag, were a dominant cost driver of rice prices in both groups, with larger effects in surplus–low-price provinces. This pattern aligns with the significance of transport and milling costs throughout the rice supply chain. The relatively stable lag structure, with detectable correlations from 0 to 3 months and a particularly strong effect at a three-month lag, indicates that fuel price shocks tend to pass through to retail rice prices over several months rather than instantaneously. This delayed transmission provides a practical window for coordinated policy interventions.

In practical terms, fuel price adjustments should not be viewed in isolation from food policy. When the government plans or announces fuel price changes, food-related agencies can prepare support measures in advance, such as sharpening plans for releasing rice reserve to vulnerable provinces and coordinating public communication on price expectations. Such anticipatory coordination may help attenuate the magnitude of rice price spikes once fuel cost shocks are fully transmitted.

3.5.2 Province-sensitive HPP design

The group-specific estimates provide a basis for evaluating the effectiveness of HPP as a dual-purpose instrument, simultaneously protecting producer incomes and maintaining affordable retail prices. In surplus provinces, HPP performs its income support function without generating statistically detectable retail price effects (coefficient 0.011, p = 0.801), suggesting that the instrument achieves its producer protection objective without imposing consumer costs in these markets. This outcome reflects the competitive supply conditions in surplus provinces, where abundant local production limits the pass-through of producer price floors to retail consumers.

In deficit provinces, however, HPP’s effectiveness as a consumer-friendly instrument is compromised. The estimated pass-through rate of approximately 6.8%, meaning that each IDR 1,000 per kilogram increase in HPP is associated with an IDR 68 per kilogram increase in retail prices, indicates that producer price support is being partially transmitted to consumers through interprovincial supply chains. This transmission pattern is consistent with broader evidence that upstream cost pressures in agricultural supply chains tend to be passed through to retail consumers more readily than cost decreases (Rahman et al., 2022; Arida et al., 2023; Kharisma and Indrawan, 2023). Critically, as shown in Section 3.4.3, this HPP-induced upward pressure on retail prices in deficit markets is not offset by CBP stock interventions, which show no significant price association in Group 2.

These findings suggest that HPP, as currently implemented, functions asymmetrically across market types: as an effective producer income support tool in surplus provinces, but as an inadvertent cost-push instrument for consumers in deficit provinces. Policy optimization in this regard could take two complementary directions. First, HPP revision schedules could be designed to account for provincial market structure, specifically, revisions in deficit provinces could be accompanied by automatic triggers for BULOG stock releases or targeted consumer subsidies to offset the pass-through effect. Second, the 6.8% pass-through rate provides a concrete basis for calibrating the scale of such compensatory measures: for every IDR 1,000 per kilogram increase in HPP, deficit province consumers face an estimated IDR 68 per kilogram increase in retail prices, which could inform the design of targeted relief mechanisms for vulnerable households in these markets. This quantitative approach moves beyond descriptive differentiation toward an evidence-based calibration of policy responses, where the estimated 6.8% pass-through coefficient provides a concrete parameter for designing compensatory measures proportional to HPP adjustments in deficit markets.

3.5.3 BULOG operations: beyond reactive distribution

The panel estimates provide a basis for evaluating the operational effectiveness of BULOG’s three primary instruments across market types. Procurement (PROC) was not statistically significant in either group, consistent with its counter-cyclical mandate to purchase rice when prices are low during harvest seasons rather than to actively move market prices. This finding confirms that procurement functions as a producer income buffer rather than a consumer price stabilization tool. Distribution (SUPPLY_t-1) was positively associated with prices in both groups, and Granger causality tests confirm that price movements precede distribution decisions rather than the reverse. This reactive pattern means that BULOG’s distribution consistently arrives after retail price increases have already materialized, limiting its buffering capacity. CBP stock holdings (STOCK) were negatively associated with prices only in Group 1 (coefficient −0.00070, p < 0.01) but showed no significant price association in Group 2 (coefficient −0.00101, p = 0.167), indicating that stock-based interventions are effective in surplus provinces but have not translated into measurable price moderation in deficit markets. These province-level findings are broadly consistent with national-level evidence showing that BULOG’s market operations have not had a significant effect on retail price formation, while stock management has shown only limited short-term effectiveness (Ruspayandi et al., 2022).

Taken together, these results indicate that BULOG’s current operational mode is predominantly reactive across all three instruments, with effectiveness concentrated in surplus provinces. In deficit provinces, where price stabilization needs are greatest, none of the three BULOG instruments demonstrate statistically significant price-reducing effects on their own. This represents a critical effectiveness gap that the current operational framework has not resolved.

A key insight from Section 3.4.3 is the temporal coordination gap between fuel price shocks and BULOG’s distribution responses. Fuel price increases take 3 months to fully transmit to retail prices, creating an identifiable intervention window. However, as confirmed by the Granger causality results, BULOG’s distribution responses are triggered by realized price increases rather than anticipated cost pressures, meaning that the intervention window goes unused under the current operational pattern.

Policy optimization for BULOG operations could therefore focus on three specific directions grounded in the empirical findings. First, distribution targeting should be shifted from price-reactive to cost-anticipatory. Fuel price movements could serve as a leading indicator for pre-emptive stock deployment in deficit provinces during the three-month transmission window, before retail price increases materialize. Second, stock placement decisions should be differentiated by provincial group. In surplus provinces, where stocks are associated with lower prices, current placement strategies appear broadly effective and should be maintained. In deficit provinces, where stocks show no significant price association, the logistical and implementation constraints on stock deployment should be systematically reviewed and addressed. Third, the scale and timing of distribution responses could be guided by province-specific price risk indicators derived from the group-specific panel models developed in this study, allowing for more anticipatory and proportional stock deployment rather than reactive responses.

3.5.4 Local production and price stabilization

Rice production exhibited a statistically significant negative association with prices in both groups. However, the magnitude of this effect was substantially larger in deficit–high-price provinces. This pattern reflects structurally scarce local supply in deficit provinces, where additional production reduces dependence on more expensive interprovincial inflows and consequently has a more pronounced impact on retail prices. Conversely, in surplus provinces, where local supply is already abundant, marginal production increases naturally have smaller price effects.

Price stabilization in deficit provinces depends on a combination of interprovincial distribution and efforts to increase local production capacity. Relevant policy options include investments in irrigation and on-farm infrastructure, reductions in post-harvest losses, improved access to technology, and enhanced local market connectivity, measures consistent with provincial evidence linking infrastructure and domestic production to lower cereal prices (Aryani et al., 2021; Antonio et al., 2025). Such interventions can help reduce price vulnerability and potentially strengthen local income bases in provinces that have historically functioned primarily as consumption markets.

3.5.5 Continuous monitoring and adaptive policy design

The group-specific panel models developed in this study achieved reasonable short-term forecasting accuracy under normal conditions, with average MAPE values of approximately 1.1% in surplus provinces and 0.8% in deficit provinces; however, performance declined during large structural shocks. This result highlights the potential of econometric frameworks not only as ex-post analytical tools but also as inputs to ongoing monitoring and policy evaluation.

In future applications, a more systematic approach utilizing such models, including periodic re-estimation, scenario analysis of alternative policy combinations, and comparative evaluations between surplus and deficit provinces, could assist policymakers in fine-tuning the intensity and mix of instruments with greater precision. Improving data quality, particularly at sub-provincial levels and along the supply chain, is a critical precondition for fully realizing the benefits of this evidence-based, adaptive approach.

4 Conclusion

We analyzed monthly rice prices in 26 Indonesian provinces by grouping provinces into surplus–low-price and deficit–high-price markets and estimating group-specific fixed-effects panel models with AR(2) errors. The grouping separates provinces into two statistically distinct sets based on average retail prices and rice adequacy indices, and the models explain about 98% of the variation in rice prices. One-step-ahead forecasts for 2022 produce average MAPE values of around 1.1% in surplus provinces and 0.8% in deficit provinces under normal conditions, with lower accuracy during major fuel-price shocks. Comparison against an AR(2)-only benchmark model reveals that the additional policy and macroeconomic covariates do not significantly improve forecast accuracy in surplus provinces (DM test: p = 0.790) and marginally reduce it in deficit provinces (DM test: p = 0.032), underscoring that the primary value of the model lies in policy interpretation rather than short-term price prediction. These results show that market structure and location matter for how rice prices behave across provinces. To our knowledge, this is the first province-level panel study for Indonesia that jointly estimates the effects of multiple rice price policy instruments across structurally defined surplus and deficit markets.

Three core findings emerge. First, policy transmission differs by market type: HPP is linked to higher retail prices only in deficit provinces, while CBP stocks are associated with lower prices only in surplus provinces. Second, fuel prices, with a three-month lag, act as a common cost-push factor in both groups. Third, higher rice production is associated with lower prices across provinces. However, the price effect is much stronger in deficit provinces, where local supply is structurally tight. This finding means that the same policy instruments can yield very different price responses across provincial market structures.

These findings highlight the need for rice price policies that are both evidence-based and place-based. HPP design, BULOG’s stock management and distribution, and production support should be differentiated between surplus and deficit provinces, with a stronger focus on local production and targeted inflows in deficit areas. The panel model framework can support such policies by providing routine short-term price monitoring and province-level risk indicators under normal market conditions.

This study has several limitations. The analysis was conducted at the operational-provincial level using monthly data, limiting the capture of sub-provincial heterogeneity and higher-frequency market dynamics. Furthermore, BULOG policy variables are endogenous responses to market conditions, implying that the estimated coefficients reflect associations rather than pure causal effects. This interpretation is empirically supported by Dumitrescu-Hurlin panel Granger causality tests, which consistently show that price movements temporally precede BULOG operational responses—in both surplus and deficit provinces—rather than the reverse. Additionally, the province-level panel framework treats each provincial unit as analytically independent, and does not explicitly model the spatial interdependencies between surplus and deficit provinces through interregional trade flows. This limitation is particularly relevant for interpreting the counter-intuitive finding that fuel price shocks are more strongly associated with rice prices in surplus provinces than in deficit provinces—a pattern that may partly reflect the role of surplus provinces as primary milling and distribution hubs, but which cannot be fully resolved without more disaggregated supply chain data and explicit spatial modelling. The 2018–2022 sample reflects a specific policy regime and several exceptional events, so parameter estimates may change under different policy or macroeconomic conditions, pointing to the need for future work that tests the model in other periods and regimes.

Overall, the study shows that rice price volatility in Indonesia is closely tied to structural heterogeneity across provinces. By demonstrating that the same policy instruments operate through fundamentally different mechanisms in surplus and deficit markets, this study advances the understanding of place-based rice price policy beyond aggregate national frameworks. These findings provide an empirical foundation for designing province-specific interventions that better reflect the diversity of market structures across Indonesia’s rice economy.

References

• 1

  AlamM. J.McKenzieA. M.BegumI. A.BuysseJ.WailesE. J.SarkarM. A. R.et al. (2022). Spatial market integration of rice in Bangladesh in the presence of transaction cost. Agric. Econ.10:20. doi: 10.1186/s40100-022-00228-5

  • CrossRef
  • Google Scholar

• 2

  AntonioR. J.ValeraH. G.MishraA. K.PedeV. O.YamanoT.VieiraB. O. (2025). Rice price inflation dynamics in the Philippines. Aust. J. Agric. Resour. Econ.69, 440–452. doi: 10.1111/1467-8489.70012

  • CrossRef
  • Google Scholar

• 3

  AridaA.MasbarR.MajidM. S. A.IndraI. (2023). Does vertical asymmetric price transmission exist in the rice markets?Agric. Resour. Econ.9, 69–90. doi: 10.51599/are.2023.09.01.04

  • CrossRef
  • Google Scholar

• 4

  AryaniD.NatawidjajaR. S.NoorT. I.MulyanaA. (2021). The integration of rice market in Indonesia as an archipelago country (vector error correction model analysis). Int. J. Adv. Sci. Eng. Inf. Technol.11:1599. doi: 10.18517/ijaseit.11.4.11264

  • CrossRef
  • Google Scholar

• 5

  ChiteteM. M. N.Pangapanga-PhiriI.PhiriH.MgomezuluW.KawayeP.FoliasL. E.et al. (2025). Asymmetric price transmission and market integration in Malawi’s selected rice markets. Agrekon64, 398–413. doi: 10.1080/03031853.2025.2581679

  • CrossRef
  • Google Scholar

• 6

  CohenJ. (1988). Statistical Power Analysis for the Behavioral Sciences. New York: Psychology Press.

  • Google Scholar

• 7

  DalheimerB.FosterK.ShivelyG.PedeV. O.FiankorD.-D. D.Ricker-GilbertJ.et al. (2026). Stocks and shocks: assessing the relative roles of public and private inventories in buffering rice price volatility in the Philippines. Food Policy139:103052. doi: 10.1016/j.foodpol.2026.103052

  • CrossRef
  • Google Scholar

• 8

  DiawatiL.RasyidA. S. (2025). Mitigating retail rice price volatility for sustainable supply chains: an optimization and regression-based approach. F1000Res14:311. doi: 10.12688/f1000research.161723.1

  • CrossRef
  • Google Scholar

• 9

  DillonB. M.BarrettC. B. (2016). Global oil prices and local food prices: evidence from East Africa. American J Agri Economics98, 154–171. doi: 10.1093/ajae/aav040

  • CrossRef
  • Google Scholar

• 10

  DumitrescuE.-I.HurlinC. (2012). Testing for granger non-causality in heterogeneous panels. Econ. Model.29, 1450–1460. doi: 10.1016/j.econmod.2012.02.014

  • CrossRef
  • Google Scholar

• 11

  FaharuddinF.YaminM.MulyanaA.YunitaY. (2023). Impact of food price increases on poverty in Indonesia: empirical evidence from cross-sectional data. JABES30, 126–142. doi: 10.1108/JABES-06-2021-0066

  • CrossRef
  • Google Scholar

• 12

  FitrawatyH.HermawanW.YusufM.MaipitaI. (2023). A simulation of increasing rice price toward the disparity of income distribution: an evidence from Indonesia. Heliyon9:e13785. doi: 10.1016/j.heliyon.2023.e13785,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 13

  IkhsanS.SantosoD.IlmiY.AnandaD. E. (2024). Analyzing the spatial integration of Rice market in South Kalimantan Province, Indonesia: using vector error correction model (VECM). ujar12, 262–271. doi: 10.13189/ujar.2024.120206

  • CrossRef
  • Google Scholar

• 14

  KharismaB.IndrawanZ. M. S. (2023). Analysis of rice price transmission in West Java, Indonesia. Cogent Food Agric.9:2266198. doi: 10.1080/23311932.2023.2266198,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 15

  LiQ.YangJ.YangX.MuE. (2025). Decompose food price disparities in China: evidence from wholesale markets. Food Policy131:102817. doi: 10.1016/j.foodpol.2025.102817

  • CrossRef
  • Google Scholar

• 16

  MgaleY. J.YunxianY.DimosoP. (2021). Cointegration and spatial price transmission among rice markets in Tanzania: implications for price stabilisation policies. Agrekon60, 157–175. doi: 10.1080/03031853.2021.1920436

  • CrossRef
  • Google Scholar

• 17

  MisdawitaH.HartonoD.NugrohoA. (2019). Impacts of food prices on the economy: social accounting matrix and microsimulation approach in Indonesia. Rev Urban & Reg Develop31, 137–154. doi: 10.1111/rurd.12099

  • CrossRef
  • Google Scholar

• 18

  MujihartonoS.HwangH. S.ShinD. H. (2023). Analyzing factors and developing strategies for Rice Price stabilization policy in Indonesia. J Int Dev Coop18, 3–27. doi: 10.34225/jidc.2023.18.2.3

  • CrossRef
  • Google Scholar

• 19

  National Food Agency of Indonesia (2025a). Keputusan Kepala Badan Pangan Nasional Nomor 14 Tahun 2025 tentang Perubahan atas Keputusan Kepala Badan Pangan Nasional Nomor 2 Tahun 2025 tentang Perubahan atas Harga Pembelian Pemerintah dan Rafaksi Harga Gabah dan Beras. Jakarta: Badan Pangan Nasional Republik Indonesia. Available at: https://jdih.badanpangan.go.id/doc/40704480-06f4-4bb3-bde3-50045d06ae9d (Accessed January 21, 2026).

  • Google Scholar

• 20

  National Food Agency of Indonesia (2025b). Keputusan Kepala Badan Pangan Nasional tentang Penetapan Harga Eceran Tertinggi (HET) Beras Medium dan Premium. Jakarta: Badan Pangan Nasional Republik Indonesia. Available at: https://badanpangan.go.id/storage/app/media/2025/NOV/20250822_%20Kepbadan%20HET%20Beras%20Medium%20dan%20Premium_HET.pdf (Accessed January 21, 2026).

  • Google Scholar

• 21

  PutraA. W.SupriatnaJ.KoestoerR. H.SoesiloT. E. B. (2021). Differences in local rice price volatility, climate, and macroeconomic determinants in the Indonesian market. Sustainability13:4465. doi: 10.3390/su13084465

  • CrossRef
  • Google Scholar

• 22

  QiC.MengC.YanS. (2025). Transmission effect of international grain prices on China’s grain prices. Front. Sustain. Food Syst.9:1483424. doi: 10.3389/fsufs.2025.1483424

  • CrossRef
  • Google Scholar

• 23

  RahmanM. C.PedeV. O.BaliéJ. (2022). Welfare impact of asymmetric price transmission on rice consumers in Bangladesh. Rev. Dev. Econ.26, 1600–1617. doi: 10.1111/rode.12882

  • CrossRef
  • Google Scholar

• 24

  RuspayandiT.BantacutT.ArifinB.FahmiI. (2022). Market-approach-based policy to achieve rice price stability in Indonesia—can it be a complement?Economies10:296. doi: 10.3390/economies10120296

  • CrossRef
  • Google Scholar

• 25

  SinagaJ. P.FirdausM.ArsantiI. W.FauziA. (2020). Medium rice market integration in Indonesia: regional disaggregation analysis. IJSSER5, 2480–2502. doi: 10.46609/IJSSER.2020.v05i09.005

  • CrossRef
  • Google Scholar

• 26

  TheresiaA.IkhsanM.KacaribuF. N.SumartoS. (2025). Spillover effect of food producer price volatility in Indonesia. Economies13:256. doi: 10.3390/economies13090256

  • CrossRef
  • Google Scholar

• 27

  UtamiA. D.HariantoH.KrisnamurthiB. (2023). Exploring the pattern of price interdependence in rice market in Indonesia in the presence of quality differential. Cogent Econ. Finance11:2178123. doi: 10.1080/23322039.2023.2178123

  • CrossRef
  • Google Scholar

• 28

  WarsilahH.WahyonoA. (2025). “Building local food systems through non-rice food banks for nutrition and food sustainability: an Indonesian case study,” in Navigating Local Sustainability in Food, Community, and Innovation, eds. YoshizumiM.ArataM.MiyaguchiT. (Singapore: Springer Nature Singapore).

  • Google Scholar

• 29

  World Bank (2019). Food Market Integration and Price Differential Across Provinces in Indonesia. Washington: World Bank.

  • Google Scholar

• 30

  YsN. A. A.SaharaS.AsmarantakaR. W. (2022). Rice market spatial integration during Covid-19 in Indonesia. jeas3, 29–38. doi: 10.32996/jeas.2022.3.3.4

  • CrossRef
  • Google Scholar