The impact of remittances on household welfare: evidence from Indonesia

International migration and remittances are crucial to household welfare, particularly in developing countries such as Indonesia, which is one of Southeast Asia's largest labor-sending countries and receives remittances equivalent to roughly 1% of GDP in recent years (IOM, 2015; UN Indonesia, 2025). Globally, about 0.8% of GDP is generated through remittances, and 27 countries receive more than 10% of their GDP from remittance receipts (World Bank, 2024a). Remittances help meet immediate consumption needs while also supporting longer-term investments in education, health and productive assets (Askarov and Doucouliagos, 2020; Khoury and Tong, 2021). Despite their importance, there is limited evidence on how remittances shape specific consumption patterns and investment behavior in Indonesia. Recent evidence from Asian economies similarly shows that structural differences across household groups can generate substantial welfare and consumption gaps (Pham et al., 2024).

This study addresses this gap by examining two key dimensions of household welfare: dietary expenditure and asset accumulation. Drawing on the New Economics of Labor Migration (NELM) framework (Stark and Bloom, 1985), we analyze how remittances relax liquidity constraints, enabling households to smooth consumption and invest in long-term well-being. Understanding the effect of remittances on nutrition-related expenditure is particularly important in the Indonesian context, where food expenditures account for a large share of the household budget, particularly in rural areas, making price stability a critical factor for poverty alleviation (Faharuddin et al., 2023). This focus is further justified given persistent undernutrition in Indonesia and the relatively high cost of protein-rich foods (Statistics Indonesia, 2023). Examining asset accumulation further provides insight into household resilience and sustainable economic progress.

Using nationally representative panel data from the Indonesian Family Life Survey (IFLS), wave 3 (2000), wave 4 (2007) and wave 5 (2014), this study employs Two-Way Fixed Effects (TWFE), multi-period Difference-in-Differences (DiD), Propensity Score Matching (PSM), and quantile regression to provide robust empirical evidence. The findings show that remittance-receiving households exhibit higher food expenditures, particularly on nutrient-dense foods such as animal protein and greater investment in physical and human capital. These patterns suggest that remittances contribute to both short-term welfare improvements and longer-term financial stability.

This study examines the role of remittances in promoting sustainable household welfare in Indonesia. It contributes to the migration-and-development literature by providing new evidence on how remittances relate to dietary quality and asset ownership within an emerging Asian economy. By applying complementary econometric methods to longitudinal household data, this paper offers policy-relevant insights into how remittance flows can support more inclusive and sustainable development outcomes. The results suggest that policies aimed at reducing remittance transaction costs, improving financial literacy and expanding access to formal financial services may further enhance these benefits.

Migration and remittances are widely recognized for alleviating poverty and improving welfare outcomes in developing nations by funding essential needs such as food and healthcare and facilitating long-term investments in education, business and property (Armah and Martey, 2020; Basak and Dey, 2024). These flows are associated with higher household incomes, reduced inequality and stronger rural economies (Khan et al., 2022). For Indonesia, World Bank data indicate that remittance inflows increased substantially between 2004 and 2024, reaching about USD 16.0 billion in 2024 (World Bank, 2024b). Remittances have been linked to higher non-food consumption (e.g. education and healthcare) and improved welfare outcomes in related contexts (Hasibuan and Hartono, 2024).

Economic theories help explain how remittances may improve well-being. NELM highlights remittances as a mechanism for risk management and consumption smoothing, particularly during shocks (Gröger and Zylberberg, 2016; Taylor, 1999). Remittances may also relax credit constraints, facilitating investments in human and physical capital (Adams, 1998; Zhang et al., 2024). In addition, the Permanent Income Hypothesis suggests that stable inflows may raise permanent income, while direct income gains may reduce poverty (Azizi, 2019; Kumara et al., 2020).

Empirically, remittances stabilize consumption, support asset accumulation and enhance access to services such as healthcare and education, notably in rural areas (Byanjankar et al., 2025; Kondratjeva et al., 2022). Evidence confirms these benefits, citing income gains in Vietnam (Cuong and Linh, 2018) and nutritional improvements in Indonesia. However, impacts vary by household demographics and migration type, sometimes exacerbating inequality (Saha et al., 2022). Consistent with this heterogeneity, evidence from rural Vietnam further shows that non-farm households exhibit higher consumption expenditure across the distribution, with the largest gaps observed among lower-income groups (Pham et al., 2024).

Estimating remittance effects is challenging due to endogeneity and confounding macroeconomic conditions. Researchers, therefore, employ various econometric strategies. TWFE models control for unobserved time-invariant factors (Halder and Malikov, 2020). Propensity Score Matching (PSM) reduces selection bias by comparing similar remittance-receiving and non-receiving households, often revealing positive effects on education, health and durable goods (Bailey et al., 2017; RAND Corporation, n.d.). Difference-in-Differences methods estimate causal impacts by comparing changes across groups over time while accounting for broader economic trends (Cook and St. Clair, 2015). In addition, prior studies in Asian contexts emphasize substantial heterogeneity in household responses across the welfare distribution, suggesting that mean-based estimators may mask important differences (Hua and Erreygers, 2020).

Despite these advances, important gaps remain. While it has been established that household welfare in Indonesia is highly sensitive to price shocks in essential food groups—specifically rice, vegetables and fish (Faharuddin et al., 2023)—empirical evidence on how remittances specifically shape these nutrition-related expenditures remains limited. Furthermore, little is known about how such impacts differ across household and contextual characteristics within the Indonesian longitudinal context. This study addresses these gaps by examining remittance associations with food expenditure and asset accumulation using complementary econometric approaches, providing evidence relevant for policy design.

This study examines the impact of remittances on food expenditure and asset ownership using panel-data methods. A TWFE regression model is used to control for unobserved, time-invariant household characteristics such as preferences and location. However, recent work highlights potential biases in TWFE models under heterogeneous treatment effects (Goodman-Bacon, 2021). To address this limitation, we complement TWFE with multi-period DiD estimators following Callaway and Sant’Anna (2021).

The analysis uses data from the Indonesian Family Life Survey (IFLS), a nationally representative longitudinal survey covering approximately 83% of Indonesia's population across 13 provinces (RAND Corporation, n.d.). We use wave 3 (2000), wave 4 (2007) and wave 5 (2014), because earlier waves lack sufficient employment information. These waves form a balanced panel aligned to pre-, treatment- and post-periods in a multi-period DiD framework.

After merging waves, we retained one observation per household by selecting the household head as the reference respondent, yielding a household-level panel. From an initial merged sample of approximately 38,000 household-wave observations, the analytical sample is 10,983 household-wave observations (3,661 households observed in each of the three waves). This restriction ensures inclusion of households observed in all waves and provides complete information on remittance receipt and expenditure/asset outcomes. Households with missing or inconsistent information were excluded to maintain comparability. The analysis focuses on households with working-age household heads (15 years and above) and uses household-level outcomes.

Although IFLS is not designed as a dedicated migration survey, it contains detailed information on overseas remittances and household expenditures (Adams and Cuecuecha, 2010). Remittances are defined as funds received from abroad and households are classified as recipients or non-recipients based on reported inflows.

The primary outcome variables are food consumption (⁠$y1$⁠), measured as total household expenditure on staple foods and asset ownership (⁠$y2$⁠), measured as total household assets including property and savings. Control variables include household-head characteristics (gender, age, education, marital status), household size, property value, urban–rural location and land ownership. Descriptive statistics by survey wave are presented in Tables A1–A3 ^[1]. Table A4 [1] reports means and standard deviations for continuous variables and frequencies and percentages for categorical variables. Figure A1 ^[1] displays the distributions of these variables.

Quantile regression is applied to examine heterogeneous welfare responses to remittances. While FE and DiD models estimate average effects, they may mask distributional differences predicted by NELM and the Permanent Income Hypothesis, particularly stronger impacts among liquidity-constrained households. Prior evidence shows that the effects of household characteristics often vary substantially across quantiles, making mean-based estimators potentially misleading (Hua and Erreygers, 2020). Quantile regression complements the causal estimators by capturing variation across the outcome distribution.

Three limitations remain. TWFE can be biased under heterogeneous treatment timing/effects; PSM adjusts only for observable characteristics; and DiD relies on the parallel trend assumption, which is limited by having only one pre-treatment period. Accordingly, findings are interpreted cautiously and the use of multiple complementary methods is intended to strengthen robustness.

While fixed effects control for time-invariant heterogeneity, they do not address time-varying confounders such as economic shocks, policy changes, or local disasters that may affect both remittances and welfare. To reduce this risk, PSM is used to improve comparability between recipient and non-recipient households based on observed characteristics. Nevertheless, PSM cannot eliminate bias from unobserved time-varying factors.

Endogeneity arises when household characteristics influence both migration decisions and welfare outcomes. The baseline model is:

Where $yit$ is the outcome (log food expenditure or log assets) for household i in year t. Remittance_it is a binary indicator of remittance receipt. $αi$ captures household fixed effects, γ_t controls for year fixed effects and X_it consists of (1) household-head characteristics (gender, age, education, marital status), (2) household size, (3) property value, (4) urban–rural residence and (5) ownership of agricultural and/or non-agricultural land. The coefficient β captures the within-household association between remittance receipt and the outcomes, conditional on model assumptions.

Fixed effects estimation reduces bias by controlling for unobserved, time-invariant household traits such as preferences and productivity Wooldridge (2010). Identification relies on within-household changes over time (Arellano, 2003). However, time-varying unobservables remain a concern, motivating the use of DiD and PSM as complementary strategies.

Given that this study spans multiple periods with variation in the timing of remittance receipt, we employ the generalized DiD framework developed by Callaway and Sant’Anna (2021). The primary estimand is the group-time Average Treatment Effect on the Treated (ATT):

Where $Yt(1)$ and $Yt(0)$ represent the potential outcomes for treated and untreated households at time $t$⁠, respectively. $Gg$ is a binary indicator for households that first received remittances in period $g$⁠.

The $ATT(g,t)$ is calculated by comparing the change in outcomes for the treated group with the change in a “never-treated” (⁠$NT$⁠) comparison group:

This framework allows estimation of dynamic treatment effects and assessment of the parallel trend assumption. A key assumption within this framework is the irreversibility of treatment, meaning that households remain treated once they begin receiving remittances. In our data, only 182 households stopped receiving remittances, indicating limited switching.

To address observable selection bias, PSM is used to improve comparability between remittance recipients and non-recipients. Propensity scores are estimated using logistic regression based on demographic and economic characteristics, which is the standard approach for summarizing observable covariates into a single balancing score. Nearest-neighbor matching with a caliper is then applied to pair recipient and non-recipient households with similar propensity scores, ensuring better balance and reducing poor-quality matches (Morgan, 2018). Covariate balance was assessed using standardized mean differences and t-tests. Treatment effects from the matched sample complement the TWFE and DiD results.

Baseline balance tests examine whether systematic differences existed in 2000 between households that later received remittances and those that did not. Two-sample t-tests compare logged food expenditure and logged asset ownership before treatment (Table A5 ^[1]). Recipient households show significantly higher baseline food spending (mean log expenditure: 13.19 vs. 9.94, p < 0.01) and asset ownership (18.87 vs. 17.28, p < 0.01). These differences indicate potential endogeneity, as remittance recipients appear better off before receiving transfers. To address this concern, TWFE, multi-period DiD and PSM are used to reduce bias from both unobserved and observed characteristics.

Table A6 ^[1] reports baseline covariate comparisons between remittance recipients and non-recipients, including age, education, marital status, household size, location, homeownership and land ownership. Initial comparisons reveal imbalances. PSM improves comparability by estimating propensity scores through logistic regression and matching households using nearest-neighbor matching with a caliper. Post-matching diagnostics show substantial improvements: most covariates display bias below 10%, t-tests are largely insignificant and variance ratios are acceptable. These results suggest that matching reduces observable selection differences. Estimates from the matched sample continue to show positive associations between remittances, food expenditure and asset ownership.

Table A7 ^[1] reports the Average Treatment Effect on the Treated (ATET). Remittances are associated with higher logged food expenditure by approximately 3.16 (p < 0.01) and log asset ownership by approximately 1.63 (p < 0.01), confirming the significant welfare benefits of remittance receipt.

Table 1 compares Ordinary Least Squares (OLS) and Fixed Effects (FE) estimates. Columns (1)–(2) use total assets as the dependent variable, while columns (3)–(4) examine food expenditure. In both outcomes, remittance receipt is positively associated with better welfare measures. The FE estimates differ from OLS, consistent with omitted-variable bias in cross-sectional models due to unobserved household characteristics. Higher adjusted R-squared values in FE models indicate improved explanatory power for within-household variation.

Table A8 ^[1] reports DiD estimates from Equation (3). Coefficients represent ATT estimates using a doubly robust DiD approach combining outcome regression with stabilized inverse probability weighting. Table 2 presents the overall ATT for both outcomes, using wild bootstrap standard errors to address heteroskedasticity; results are positive. Event-study estimates in Table A9 ^[1] present ATT over a seven-year pre-period (2000–2007), the treatment year (2007) and a seven-year post-period (2007–2014). Figure A2 ^[1] visualizes these estimates for assets and food expenditure, showing ATT coefficients with 95% confidence intervals across the pre-period (T–1), treatment period (T0) and post-period (T+1). Pre-treatment estimates are close to zero and statistically insignificant, supporting the parallel trends assumption. A non-significant pretrend test (chi-square = 0.4262, p-value = 0.5138) further supports this interpretation, consistent with Kahn-Lang and Lang (2020). Having established average effects through TWFE and DiD, we next examine heterogeneous impacts using quantile regression.

Building on Callaway and Sant’Anna (2021), we assess robustness across alternative estimators. The doubly robust method with stabilized inverse probability weighting (DRIPW) is used as the baseline specification. Additional estimators include inverse probability weighting DiD (IPW), stabilized IPW (STDIPW) and an improved doubly robust estimator (DRIMP). These approaches address concerns regarding heterogeneous treatment effects and selection bias (Sant’Anna and Zhao, 2020) by reducing reliance on any single estimator. Table A10 ^[1] summarizes the ATT findings. Estimates using IPW and STDIPW remain positive and significant though smaller than the baseline (DRIPW). The DRIMP estimator yields positive but more conservative coefficients. Overall, the results are consistent across specifications.

Results are consistent across estimation methods, with positive and statistically significant associations between remittance receipts and both food expenditure and asset accumulation. Fixed effects models account for time-invariant unobserved heterogeneity, while multi-period DiD provides stronger identification under standard assumptions. The log-linear specification facilitates interpretation, but percentage translations must be handled carefully because effects are multiplicative in levels.

Findings suggest that remittances ease liquidity constraints, enabling higher food consumption and greater asset accumulation. Additional income allows households to purchase more food and durable goods while supporting asset building, particularly in resource-constrained settings. These results align with the view that remittances help households smooth consumption and buffer against economic shocks.

Although the study focuses on welfare impacts, reverse causation cannot be fully ruled out: households with higher baseline resources may be more likely to finance migration and later receive remittances. PSM reduces bias from observable selection, and FE controls for time-invariant unobserved factors, but unobserved time-varying shocks (e.g. inflation or natural disasters) may still affect both remittance receipt and welfare outcomes.

The estimated 350% increase in food consumption for remittance-receiving households reflects the log-linear model and the heavy-tailed distribution of food expenditure. After trimming the top and bottom 1% of observations and re-estimating using log(1+y), the effect remains positive but is reduced to approximately 200%, suggesting that the original magnitude should be interpreted cautiously.

Tables 1 and 2 show that remittances have a significant positive impact on asset accumulation. Within-household changes in remittance receipts are associated with a 94% increase in total assets. Based on average asset values in 2014, this corresponds to an estimated gain of approximately $29,422 relative to non-recipient households. In contrast, living in rural areas and having a larger household size are negatively associated with asset ownership, reducing investment by about 6%.

Remittances have a stronger effect on food spending than on assets. Recipient households spend substantially more on food than non-recipients, averaging about $44 per week in 2014 compared to the national average of $9.80 (Statistics Indonesia, 2014). The increase is most pronounced for animal protein consumption, indicating improvements in dietary quality. Spending on vegetables and staple foods is similar across groups, but remittance-receiving households allocate significantly more to protein-rich foods. Figure A3 ^[1] illustrates these patterns through bar charts of average weekly expenditures by food type. Food expenditure also rises with the age of the household head, household size and rural residence. Table A11 ^[1] presents summary statistics of average weekly food expenditure and total assets.

Quantile regression results (Table A12 ^[1]) indicate that remittances significantly increase food expenditure across the distribution, with the largest proportional differences among lower-expenditure households. Education and household size are positively associated with food spending, whereas rural households tend to spend less. The effect of age is positive but diminishes at higher quantiles.

Remittances generate the greatest benefits at the lower end of the welfare distribution. Households in the 10th percentile of food expenditure increase spending by approximately 350% relative to non-recipients. The effect gradually declines to 275% at the 25th percentile, 231% at the median and 195% at the 90th percentile. These patterns highlight the important role of remittances in improving food security, particularly for the poorest households. This result is consistent with recent evidence from Asian economies documenting larger welfare differentials at the lower end of the expenditure distribution (Pham et al., 2024) and with prior findings that household responses tend to be stronger at lower quantiles (Hua and Erreygers, 2020).

This study demonstrates that remittances are positively associated with household welfare in Indonesia. Using complementary econometric approaches – including TWFE, multi-period DiD and PSM – we find that remittances correspond to increased food expenditure, particularly on nutrient-dense items such as animal protein. In addition, remittance-receiving households exhibit higher investment in education and physical assets, supporting both immediate well-being and longer-term financial stability. These benefits are most pronounced among lower-expenditure households, consistent with remittances helping to overcome liquidity constraints. Future research should explore regional and subgroup heterogeneity to tailor policy responses more effectively.

Results indicate that remittance-derived welfare gains are concentrated among lower-expenditure households. Consequently, targeted household-level measures may be more effective than broad macro-level initiatives in facilitating sustained welfare improvements. Lowering formal remittance fees is critical to increasing disposable income for households reliant on frequent, small transfers. Furthermore, implementing micro-savings and financial literacy programs can help channel remittances into productive assets, such as livestock and education. To improve dietary quality, pairing inflows with nutrition-focused education or vouchers, especially in rural areas, is recommended. Ultimately, these measures align with the observed quantile heterogeneity, ensuring that remittance-related gains are both inclusive and durable.

The authors used RStudio and StataSE 17 for data analysis and visualization. The manuscript was professionally proofread by a certified language-editing service, with no generative AI or AI-assisted writing tools used. The author reviewed and revised the content as necessary, assuming full responsibility for the final publication.

The supplementary material for this article can be found online.