FDI and sustainable GDP for select Asian countries: the role of nonlinear analysis

Foreign Direct Investment (FDI) is widely regarded as a critical driver of economic development in host countries. It provides not only capital inflows but also essential channels for technology transfer, management expertise, productivity enhancement, and integration into global markets. Among various forms of external financing, FDI is considered the most stable and less volatile during periods of global financial crisis, making it highly valued, especially for emerging economies.

Despite these theoretical benefits, the empirical evidence on the FDI-growth relationship is mixed. Several studies affirm a positive contribution of FDI to economic growth across diverse country settings (Sijabat, 2023; Yusuf et al., 2020; Okwu et al., 2020; Sengupta and Roma, 2020; Agyapong and Kojo, 2020; Akadiri et al., 2019). They emphasize the role of FDI in fostering knowledge spillovers, upgrading technology, and stimulating capital accumulation. Conversely, several empirical investigations reveal negative impacts of FDI on growth (Sothan, 2017; Rehman, 2016). Critics argue that under certain conditions, FDI can crowd out domestic investment, perpetuate dependency, distort market structures, and exacerbate income inequalities. Some studies even document the “race to the bottom” phenomenon, where host countries relax environmental and labor standards to attract foreign investors. Additionally, a third strand of the literature finds no statistically significant relationship between FDI and GDP growth, suggesting a neutrality hypothesis (Magazzino and Mele, 2022; Kathuria, 2019; Phuong and Tuyen, 2018). These findings indicate that the FDI-growth nexus is not universally positive, and its direction and magnitude may depend on various structural and institutional factors within the host economy.

Recognizing the FDI-growth relationship, scholars have increasingly emphasized that the impact of FDI may be contingent on country-specific absorptive capacities, such as human capital development, financial system maturity, technological capabilities, and institutional quality (Joo et al., 2022; Carbonell and Richard, 2018). Consequently, FDI’s contribution to sustainable economic growth cannot be assumed to be homogeneous across countries or over time. Moreover, the dynamic response of economic growth to FDI shocks may not be symmetric. Traditional linear models typically assume that positive and negative FDI shocks exert identical effects, merely differing in direction. However, real-world economic dynamics often exhibit asymmetries, particularly in the presence of structural rigidities, institutional constraints, and behavioral adjustments (Shin et al., 2014). Ignoring potential asymmetries could lead to misinterpretation of FDI’s real economic effects and, consequently, to flawed policy recommendations.

Against this backdrop, this study addresses an important gap in the literature by focusing on the nonlinear and asymmetric effects of FDI on sustainable GDP growth across seven selected Asian economies: Indonesia, Malaysia, Philippines, Singapore, Thailand, South Korea, and Japan. Thus, the primary objective of this research is to investigate whether positive and negative FDI shocks have asymmetric impacts on economic growth in both the short and long term, thereby offering more accurate insights for policy design.

The focus on Asia is especially relevant given recent FDI dynamics in the region. According to the Asia-Pacific Trade and Investment Trends Report (2021, 2022) and the World Investment Report (2022a, b), FDI inflows to Asia declined sharply during the COVID-19 pandemic. For instance, Indonesia experienced an FDI reduction of approximately $5 billion. Malaysia saw a decline of 54%. The Philippines suffered a 21.33% fall. Thailand witnessed a dramatic 247% contraction. South Korea and Japan also reported declines of 9 and 30% respectively. Although signs of recovery emerged by 2021 and 2022 (ASEAN Investment Report, 2022; UNCTAD, 2021), the fluctuations in FDI inflows highlighted the vulnerability of Asian economies to external shocks, reinforcing the need to understand how FDI shocks—positive and negative—affect sustainable GDP growth differently.

Moreover, the strategic importance of FDI for Asia’s development cannot be overstated. According to the ASEAN Investment Report (2022), the region’s share of global FDI inflows increased from 11% in 2018–2019 to 12% in 2020–2021. Asian economies have diversified beyond traditional manufacturing into sectors such as digital economy, green infrastructure, healthcare innovation, and renewable energy (World Investment Report, 2023). In 2023, FDI trends across the selected Asian economies displayed notable variation. Indonesia recorded a 4% increase in FDI inflows, reaching $22 billion. Malaysia experienced a significant rise of 39%, achieving a record $17 billion, driven largely by investments in manufacturing and greenfield projects. In contrast, the Philippines saw a 23% decline in FDI inflows, influenced by domestic acquisitions of foreign affiliates. Singapore maintained its position as a regional hub, attracting a record $141 billion in FDI and accounting for nearly two-thirds of ASEAN’s total inflows. FDI data for Thailand in 2023 was not specifically detailed in the available sources. South Korea experienced an 18% decrease in FDI inflows, totaling $18 billion. Meanwhile, Japan reached its highest-ever level of FDI inflows at $33 billion (World Investment Report, 2023).

Given this backdrop, the methodology adopted in this paper is designed to capture the nonlinear and asymmetric nature of the FDI-growth nexus. Specifically, our research employs the Nonlinear Autoregressive Distributed Lag (NARDL) model developed by Shin et al. (2014) and the Generalized Impulse Response Functions (GIRF) by Koop et al. (1996) and Pesaran and Shin (1998), which allow for disaggregation of FDI inflows into positive and negative shocks, distinction between short-run and long-run asymmetric impacts, and more realistic modeling of economic adjustment processes following FDI fluctuations. Positive FDI shocks may strengthen productivity and investment channels more efficiently when host countries exhibit strong absorptive capacity (e.g. high human capital, developed financial systems, strong institutional frameworks). Negative FDI shocks may exacerbate vulnerabilities in economies with weaker absorptive capacity, magnifying adverse impacts through channels like reduced investment, technology withdrawal, or employment losses.

To ensure robustness and comprehensiveness, we incorporate several control variables, notably capital formation, trade openness, and population growth. These variables have been widely recognized as important determinants of economic performance in the existing literature (Topcu et al., 2020; Onyinye et al., 2017; Cleeve et al., 2015; Hasli et al., 2015; Kim, 2016; Peterson, 2017). By controlling these factors, the study isolates the net effect of FDI shocks on sustainable GDP growth more accurately.

In sum, this paper aims to contribute to the existing literature by providing empirical evidence on the nonlinear and asymmetric responses of sustainable GDP growth to FDI shocks in selected Asian countries, using a rigorous econometric framework and comprehensive control mechanisms. This distinguishes the current from previous studies, which assessed whether FDI inflow may exert a positive or negative impact on economic growth in the host economy as well as other studies that explored the causal relationship between the trends of these two macro variables. Furthermore, we explicitly define sustainable GDP growth not merely in terms of short-term output increases, but as inclusive, resilient, and environmentally sound economic expansion. This broader conceptualization of growth aligns the analysis with contemporary policy priorities in the selected Asian economies.

Our findings indicate that FDI asymmetrically affects real GDP in the long run for five Asian economies: Indonesia, Malaysia, Philippines, Singapore, and Thailand. The short-run asymmetry is confirmed for Malaysia, Thailand, and Japan. When examining the GIRF, we found that the responses of these sample countries’ $GDPt$ to the positive FDI shocks are significant in most countries. In contrast, the negative $FDI$ shock has a relatively minor impact on the GDP of these countries. Therefore, the GIRF analysis further affirms the presence of an asymmetric effect of $FDI$ shocks on $GDP$ across the sampled Asian countries. Additionally, the findings are expected to offer valuable insights for policymakers seeking to design targeted strategies to maximize the developmental benefits of FDI, especially in the context of post-pandemic recovery and long-term sustainable development ambitions.

The remaining sections of this paper are organized as follows: Section 2 reviews the related literature, while section 3 outlines the data collection and model specification. Section 4 confers the empirical findings. Finally, Section 5 concludes the paper and highlights its policy implications.

Numerous studies have been conducted to thoroughly investigate the impact of FDI on economic growth through typically utilizing the neoclassical and endogenous economic growth models (Deonanan and Daren, 2022; An and Kuo-Chun, 2021; Udeogu et al., 2021; Hagan and Anthony, 2020). The neoclassical paradigm states that FDI influences economic growth by fostering capital accumulation, technology spillover, and enhancing the short-term efficiency of the host countries’ economies (Guvercin and Adem, 2021; Yusuf et al., 2020), while the endogenous model posits that economic growth is directly linked with FDI inflow (Bambi et al., 2017). Nevertheless, despite the impact of FDI inflow on GDP growth has been extensively examined in literature; however, the results remain inconclusive and exhibit varying results due to differences in the estimation methods employed, variations in the selected samples, in addition to disparities in the time periods across studies.

In this regard, various studies endorse the significant positive impact of FDI on economic growth based on the FDI spillover externalities on the host economy (i.e. Wang (2019) for the Asian countries; Akadiri et al. (2019) for twenty-five African countries; Elian et al. (2020) for the BRICS (Brazil, Russia, India, China, South Africa) countries; Cakerri et al., 2020, Agyapong and Kojo (2020) for Africa, Mowlaei and Intezar (2021) for a group of 30 Islamic countries) ^[1]. Nonetheless, certain studies have determined that the positive impact of FDI on growth is statistically insignificant (i.e. Yusuf et al. (2020) for West Africa, Shahzad et al. (2019) for Brazilian economy, Gunby et al. (2017) for China). A further set of studies revealed a negative growth effect of FDI, attributed to the crowding out of domestic investments and expanding external vulnerability (Sağlam, 2017; Rehman, 2016). A key strand of literature supports the conditional effect argument, suggesting that the effectiveness of FDI in stimulating economic growth is contingent on preexisting conditions within the host economy. These preconditions, referred to as specific features, act as absorptive capacities that enable the host economy to fully harness the benefits of FDI inflows (Joo et al., 2022; Ibhagui, 2020; Dinh et al., 2019; Carbonell and Richard, 2018) ^[2].

Current studies highlight the significant role of Foreign Direct Investment (FDI) in economic growth. Phan Le et al. (2024) show that a 1% increase in FDI boosts economic growth by 9.3% in middle-income countries, with total factor productivity playing a key role. Rodríguez-Chávez (2024) noted a 3% global increase in FDI, but a 9% decline in developing countries, with Asia most affected. The UNCTAD World Investment Report (2024) reports an 11% rise in global FDI to $1.4 trillion, mainly driven by European conduit economies. Recent international reports highlight a mixed global outlook. The International Monetary Fund (IMF, 2025) reports that global inward FDI rose by 4.4% in 2023, reaching approximately $41 trillion, supported by strong inflows to emerging economies. In parallel, the Organization for Economic Co-operation and Development (OECD, 2025) forecasts a slight deceleration in global GDP growth, from 3.2% in 2024 to 3.1% in 2025, attributed to rising trade barriers and heightened policy uncertainty. Kenh and Wei (2025) emphasize that FDI is most effective when directed toward sectors with a comparative advantage, significantly boosting economic development. Korsah et al. (2025) identify natural resources, market size, trade openness, and exchange rate strength as key determinants of FDI inflows into West African countries, noting that French-speaking nations attract more FDI than their English-speaking counterparts. Arya et al. (2025) confirm a long-term relationship between trade, FDI, and economic growth, with bidirectional causality between GDP and FDI in most countries, and exports contributing notably to GDP growth in several cases.

Our paper incorporates country-specific features such as the country’s capital formation, trade openness, and population growth rate, which have been identified as significant factors in previous studies examining the FDI-GDP growth nexus. Capital formation is commonly regarded as a beneficial factor for a country’s economic growth, as evidenced by existing literature indicating its positive impact on economic growth (i.e. Ntamwiza and Masengesho (2022), for Rwanda; Pasara et al. (2020), for South Africa; Awodumi and Adewuyi (2020), for Angola and Egypt; Muhammad and Khan (2019), for 35 Asian countries; Khan et al. (2019), for 193 countries; Erum and Hussain (2019), for 43 members of the Organization of Islamic Corporation (OIC) countries; Bal et al. (2016), for India); Trade openness is another vital characteristic of host country that has received substantial attention in the literature on the relationship between FDI and economic growth. Numerous studies have validated the significant role of trade openness in promoting economic growth (Cinar and Nulambeh, 2018; Khamphengvong and Srithilat, 2017). The population growth rate is an additional crucial host country feature that plays a role in the impact of FDI on economic growth. Kim (2016) conducted an estimation of how the fluctuating age configuration of the population has affected GDP growth, ultimately concluding that demographic changes would result in a significant downward trend in economic growth. Peterson (2017) found that the influence of population growth on economic growth is contingent on the income level of the country. In high-income countries, a high population growth rate is likely to alleviate the country’s social and economic challenges, while in low-income countries may impede economic development.

This study examines the asymmetric relationship between foreign direct investment (FDI) and real gross domestic product (GDP) in seven Asian economies: Indonesia, Malaysia, the Philippines, Singapore, Thailand, South Korea, and Japan ^[3]. The analysis utilizes annual data spanning the period 1970–2019. Table 1 provides the definitions of the variables employed, along with their respective data sources.

In conceptualizing the long-run impact of foreign direct investment (FDI) on real gross domestic product (GDP), this study follows the recent strand of literature and adopts the following long-run specification:

where $GDPt$ is real GDP (base year is 2015), $FDIt$ is net inflows of foreign direct investment (% of GDP), $Kt$ is gross capital formation (% of GDP), $Tt$ is trade openness as exports plus imports of goods and services (% of GDP), $POPt$ is population growth (annual %), and $εt$ is the standard error term (see appendix 1). All variables are measured in logarithms.

In Equation (1), $α1$ to $α4$ denote the long-run effects of the explanatory variables—namely (⁠$FDIt$⁠, $Kt$⁠, $Tt$⁠, $POPt$⁠) on $GDPt$⁠. As demonstrated by Pesaran et al. (2001), Equation (1) can be estimated within the framework of an Error Correction Model (ECM). A major advantage of the ECM approach is that it enables the simultaneous estimation of both short-run dynamics and long-run relationships within a single equation. The ECM specification is expressed as follows:

According to Pesaran et al. (2001), Equation (2) identifies the short-run effects through the coefficients of the first-differenced variables (e.g. $∑i=0nδ2i$⁠), while the long-run effects are captured by the estimates of $θ1$ to $θ4$ normalized on $θ0$ (⁠$−θ1θ0=α1$⁠; $−θ2θ0=α2$⁠; $−θ3θ0=α3$⁠; $−θ4θ0=α4$⁠). This specification, developed by Pesaran et al. (2001), is commonly referred to as the linear autoregressive distributed lag (ARDL) model.

The specification in Equation (1) implies that the explanatory variables (⁠$FDIt$⁠, $Kt$⁠, $Tt$⁠, $POPt$⁠) affect real $GDPt$ in a symmetric, or linear, manner. Symmetry in this context indicates that increases and decreases in an explanatory variable exert effects of equal magnitude but opposite direction on $GDPt$⁠. However, Neftçi (1984) and Falk (1986) demonstrate that many macroeconomic variables exhibit nonlinear, or asymmetric, behavior over time. This suggests that fluctuations in $GDPt$ may respond differently to increases versus decreases in the explanatory variables, both in magnitude and direction, thereby reflecting the concept of asymmetric effects.

To capture such dynamics, Shin et al. (2014) extended the linear ARDL framework to the nonlinear autoregressive distributed lag (NARDL) model, which allows for the identification of asymmetric effects ^[4]. We build on this framework by integrating Generalized Impulse Response Functions (GIRF), which provides a dynamic perspective on how GDP responds to asymmetric investment shocks over time.

The NARDL approach decomposes each explanatory variable (⁠$FDIt$⁠, $Kt$⁠, $Tt$⁠, $POPt$⁠) into partial sums of positive (increasing) and negative (decreasing) changes. For instance, $FDIt$ can be expressed as the cumulative sum of its positive and negative variations as follows:

and

where $FDIt=FDI0+FDIt++FDIt−$⁠.

Similarly, $Kt$ is decomposed into $Kt+$ and $Kt−$⁠, $Tt$ into $Tt+$ and $Tt−$⁠, and $POPt$ into $POPt+$ and $POPt−$⁠. Constructing these partial sum variables enables the nonlinear adjustment of variations in the explanatory variables. Accordingly, Equation (1) can be reformulated within an asymmetric long-run framework as follows:

To estimate both the asymmetric short-run and long-run effects, the error correction model (ECM) is reformulated as follows:

Shin et al. (2014), building on the framework of Pesaran et al. (2001), demonstrate that in Equation (6) the asymmetric short-run effects are reflected in the estimated coefficients of the first-differenced variables (e.g. $∑i=0nπ2i$⁠), while the asymmetric long-run effects are captured by the coefficients $λ1$ to $λ8$ ​, normalized on $λ0$​ (⁠$−λ1λ0=β1$⁠; $−λ2λ0=β2$⁠; $−λ3λ0=β3$⁠; $−λ4λ0=β4$⁠; $−λ5λ0=β5$⁠; $−λ6λ0=β6$⁠; $−λ7λ0=β7$⁠; $−λ8λ0=β8$⁠).

Shin et al. (2014) suggest that, once Equation (6) is estimated using the conventional Ordinary Least Squares (OLS) method, the presence of asymmetric effects of each explanatory variable on real $GDPt$ can be formally assessed through Wald tests, provided that both the positive (+) and negative (−) partial sum components are statistically significant. For instance, the null hypothesis $β1=β2$ implies no long-run asymmetry, while the alternative $β1≠β2$ indicates the existence of long-run asymmetry between real $GDPt$ and variations in $FDIt$⁠. Similarly, the rejection of the null hypotheses $β3=β4$⁠, $β5=β6$⁠, and $β7=β8$ would provide evidence of long-run asymmetry in the relationships between real $GDPt$ and gross capital formation, trade openness, and population growth, respectively. In addition, short-run asymmetries are investigated through the coefficients of the first-differenced variables in Equation (6). For example, under the Wald test, the null hypothesis $∑i=0nπ2i=∑i=0nπ3i$ implies no short-run asymmetry, whereas the alternative $∑i=0nπ2i≠∑i=0nπ3i$ indicates the presence of short-run asymmetry between real $GDPt$ and $FDIt$ fluctuations. Likewise, evidence of short-run asymmetry would arise if the hypotheses $∑i=0nπ4i=∑i=0nπ5i$⁠, $∑i=0nπ6i=∑i=0nπ7i$⁠, or $∑i=0nπ8i=∑i=0nπ9i$ are rejected, implying asymmetric short-run effects of gross capital formation, trade openness, and population growth, respectively.

The validity of NARDL inferences critically depends on the existence of cointegration, that is, a long-run equilibrium relationship among the variables. Accordingly, we test whether the variables in Equation (6) are cointegrated using the procedures of Pesaran et al. (2001) and Banerjee et al. (1998), which rely on the F-statistic and the $tBDM$-statistic, respectively. Under the Pesaran et al. (2001) bounds testing framework, evidence of cointegration is obtained when $λ0≠λ1≠λ2≠λ3≠λ4≠λ5≠λ6≠λ7≠λ8≠0$⁠. By contrast, the null hypothesis of no cointegration is defined as $λ0=λ1=λ2=λ3=λ4=λ5=λ6=λ7=λ8=0$ ^[5]. In the Banerjee et al. (1998) $tBDM$ test, cointegration is confirmed when $λ0<0$ (no cointegration if $λ0=0$⁠). Establishing cointegration is therefore a necessary precondition; only once this is confirmed can the estimated NARDL model yield valid inferences regarding both the long-run and short-run asymmetric relationships.

Along with the NARDL model, this paper employs the generalized impulse response function (GIRF) as a second practice to examine the dynamic response of $GDPt$ to the increase and decrease of $FDIt$ across the seven Asian countries. To do so, we assume that $GDPt$ is determined by $FDIt$ hikes (positive shock), and $FDIt$ falls (negative shock):

At this point, $ωk,t$ represents a shock in the kth variable, where k is the positive and negative $FDIt$ shocks. GIRF is next uncovered through estimating Eq. (7). Accordingly, GIRF offers the responses of $GDPt$ to a shock of one standard deviation in $ωk,t$ for the seven Asian countries.

We start with a visual assessment of the time path of $FDIt$ and $GDPt$ variables to verify their co-movement. Figure 1 noticeably implies a kind of co-movement among the two variables, signaling the probability of a long-run association. Nevertheless, we need to properly analyze the presence of the long-run association (cointegration) as illustrated earlier.

Previous to applying the NARDL model, we begin by checking the stationarity of the respective variables to disclose their order of integration. Regardless of the fact that the NARDL can be utilized whether the variables are of order zero, or order one, or even a mix of both (I(0), I(1)), as described before, still including I(2) variable in the NARDL model presents the possibility of spurious inferences (Pesaran et al., 2001). As a result, the first needed step for applying the NARDL is to verify that none of the series is I(2). While there are various tests available to investigate the unit root, the modified Dickey-Fuller unit root test with single break was used for each variable with the results shown in Table 2. The test assumes a null of unit root (non-stationarity). The results of Table 2 provide compelling evidence that the respective variables are not integrated of order two (I(2)), which denotes the appropriateness of using the NARDL model as the fundamental requirement is met.

Ahead of applying the NARDL analysis, we need to check for two issues. Firstly, investigate if there is support for the nonlinearity in the relationship between the engaged variables. A proper practice to employ is the Brock, Dechert, and Scheinkman test (BDS test) of Broock et al. (1996) that is often used to explore the presence of nonlinearity in the model. The BDS test postulates a null hypothesis of linearity in opposition to an alternative of nonlinearity. Secondly, test if the data suffers from shocks or unanticipated events (structural breaks). Gregory et al. (1996) report that if the analysis does not consider the structural breaks, the cointegration conclusions might be deceptive, since such breaks might produce unstable cointegration relationship, provided that the NARDL procedure requires evidence of cointegration. Thus, the test of Bai and Perron (1998) is applied to detect the structural breaks in the data. This test enables us to determine up to five breaks in addition to detecting its dates ^[6]. If structural breaks are found, the NARDL model allows, within the error correction model, to use dummy variables for the breaks. The dummy variable equals one starting from the break date and afterward and equals zero before the break date (Pesaran et al., 2001). The results of BDS test and Bai-Perron examination are reported in Tables 3 and 4, respectively.

The results represented in Table 3 indicate that the nonlinearity will be a proper representation in examining the $FDIt−GDPt$ nexus seeing that the null hypothesis of linearity is evidently rejected, seeing that the test statistics are substantially significant for all seven countries. This provides the rationale to employ the NARDL model. On the other hand, Table 4 suggests different number and dates of structural breaks for the seven countries. This suggests that the estimation of the ECM of Eq. (6) will be modified by incorporating the dummy variables as illustrated above-mentioned.

We now utilize yearly data from 1970 to 2019 to estimate Eq. (6) by the use of OLS. Ahead of examining the short- and long-run effects, we need to test for cointegration first. Table 5 describes, for all seven countries, the F-statistic and $tBDM$ statistic of the NARDL model, together with the error correction term (⁠$ECTt−1$⁠). $ECTt−1$⁠, which needs to be significant, denotes the deviation from the long-run equilibrium because of a short-run shock, and represents the adjustment speed to reinstate the long-run equilibrium. Cointegration verification requires the absolute value of the significant $ECTt−1$ to be less than one. From Table 5, the results offer confirmation for an asymmetric cointegrating relationship between the positive and negative changes of the explanatory variables (⁠$FDIt+$⁠, $FDIt−$⁠, $Kt+$⁠, $Kt−$⁠, $Tt+$⁠, $Tt−$⁠, $POPt+$⁠, $POPt−$⁠) and real GDP (⁠$GDPt$⁠). The F-statistics of all seven countries are above the 1% upper bound critical value of 3.77, and the $tBDM$-statistics are significant at different significance levels. Both test statistics signal proof of rejecting the null hypothesis of no asymmetric cointegration relationship. In addition, the $ECTt−1$ is significant and less than one in absolute value for all seven countries confirming the cointegration relationship.

As evidence of cointegration between the selected variables is found for all seven countries, now we are discussing asymmetric analysis. Let us first discuss the long-run results from Table 6, ^[7]. It is noticed that either $FDI+$ or $FDI−$ has one significant coefficient for five countries (Indonesia, Malaysia, Philippines, Singapore, and Thailand) but not S. Korea and Japan. Consequently, $FDIt$ is demonstrated to be an important factor modeling the behavior of the real $GDPt$ in the long run. What’s more, as the estimates on $FDI+$ across the five countries have different sizes, signs, and significance from the matching estimates on $FDI−$ (in Table 6), this guides us to think that there appears to be a long-run asymmetry. When performing the Wald test, the long-run asymmetry is supported in all five countries: Indonesia, Malaysia, Philippines, Singapore, and Thailand, as the $χ2$ statistic is significant at different significance levels. Thus, real $GDPt$ for the five countries seems to respond asymmetrically to the positive and negative $FDIt$ shocks. In sum, Table 6 shows that the long-run asymmetries are significant across five countries only.

Next, we turn our attention to the short-run findings. Table 7 reports the short-run estimates of $ΔFDI+$ and $ΔFDI−$ across the seven countries. We notice that at least one of $ΔFDI+$ and $ΔFDI−$ includes a significant coefficient at the 5% level or lower for three countries: Malaysia, Thailand, and Japan. In addition, the $χ2$ statistics confirm the short-run asymmetric effects of $FDIt$ on real $GDPt$ (Table 7). As a result, it turns out that $FDIt$ seems to have a significant asymmetric role in adjusting the real $GDPt$ of these three countries in the short-run. However, in the case of Japan the short-run effects do not last in the long-run as compared to Malaysia and Thailand where the short-run effects do last in the long-run. In the Japan case, the short-run effect vanishes in the long-run due to the possibility of the restricted environmental laws of technology transfer in Japan. Such laws could lead to symmetric FDI effects rather than asymmetric.

The soundness of the findings from the NARDL estimations rely upon the degree to which these estimations are statistically robust. Accordingly, we run several diagnostic tests that investigate the stability and robustness of the NARDL model. The tests are Lagrange Multiplier (LM), Ramsey’s RESET test, adjusted R², Jarque-Bera (JB) normality test, cumulative sum of recursive residuals (CUSUM), and cumulative sum of squares of recursive residuals (CUSUMSQ). The LM test examines the serial correlation of the residuals (null hypothesis: no serial correlation in the residuals). The Ramsey RESET test examines if the model is well specified (null hypothesis: functional form is correctly specified). Jarque-Bera (JB) normality test examines the error normality (null hypothesis: normal distribution). CUSUM and CUSUMSQ tests of Brown et al. (1975) examine the stability of the long-run parameters. The findings of the diagnostic tests are reported in Table 8.

Table 8 shows no indication of serial correlation of the errors needed for an optimal model. The LM statistics (⁠$χ2$ statistics) from Table 8 reveal that we cannot reject the null hypothesis for all countries, implying that the residuals are not suffering from serial correlation. For the Ramsey RESET test, which has a null hypothesis assuming the model is well specified, the statistics from Table 8 show that we cannot reject the null, indicating that the functional form of the NARDL model is suitable. As for the adjusted R², Table 8 shows notable goodness of fit for all countries. Additionally, the Jarque-Bera statistics (J-B) for error normality are insignificant implying normality. Finally, Figure 2 verifies the stability of the parameters as both CUSUM and CUSUMSQ lie within the 5% critical bounds across all countries. Overall, the CUSUM and CUSUMSQ tests give rationale for employing the NARDL model and its outcomes.

In this part, we proceed with the responses of the sample countries’ $GDPt$ to the positive and negative $FDIt$ shocks using GIRF, as shown before (Figure 3). The impulse response will be statistically significant, at the 5% level, once both the mean and confidence interval values are falling above or below zero. We first start by exploring the responses of the sample countries’ $GDPt$ to a positive $FDIt$ shock. For Indonesia, the positive $FDIt$ shock is significant and $GDPt$ rises over the 10 periods, and the same thing applies for Philippines. Whereas for Malaysia, the significant positive shock causes $GDPt$ to rise for the first four periods, but then $GDPt$ falls. In the case of Philippines, the positive $FDIt$ shock is significant in the first two periods where $GDPt$ falls, and significant from the fifth period again seeing that it vanishes. For the Thailand case, the positive shock is significant starting from the fourth period and causes $GDPt$ to rise. On the other hand, for S. Korea and Japan the positive $FDIt$ shock is not significant for the entire 10 periods.

Next, we proceed to the effects of the negative $FDIt$ shock on the sample countries’ $GDPt$⁠.

The negative $FDIt$ shock is not significant for Singapore and Japan. However, for Indonesia and Malaysia, the negative $FDIt$ shock is significant and causes $GDPt$ to fall over the entire 10 periods. For Philippines and S. Korea, the negative $FDIt$ shock is significant but causes $GDPt$ to rise. Finally, in the case of Thailand, the GIRF shows that the negative $FDIt$ shock is significant but causes the $GDPt$ to drop in first two periods and then stays almost steady after that.

Overall, the generalized impulse response function (GIRF) shows that the responses of the sample countries’ $GDPt$ to the positive and negative $FDIt$ shocks do not seem to be mirror images of one another. This can be interpreted as a sign of the asymmetric impact of $FDIt$ on $GDPt$⁠. This conclusion can be supported further by the plots of the Accumulated Response to Generalized one standard deviation Innovations in Figure 3. The plots do show that positive and negative $FDIt$ shocks provide different effect on $GDPt$⁠.

This paper enhances the existing body of literature by investigating whether $FDI$ asymmetrically affects $GDP$ among seven selected Asian countries: Indonesia, Malaysia, Philippines, Singapore, Thailand, South Korea, and Japan. We employed the NARDL model in conjunction with GIRF analysis. From a methodological standpoint, the NARDL model enables the examination of potential asymmetry in how GDP responds to positive and negative $FDI$ shocks, revealing insightful findings that may remain hidden when using a symmetric (linear) analysis. Concurrently, the GIRF analysis allowed us to evaluate the dynamic responses of $GDP$ to both positive and negative $FDI$ shocks.

The NARDL model findings indicated that in the long term, $FDI$ shocks asymmetrically affect the $GDP$ of five Asian countries: Indonesia, Malaysia, Philippines, Singapore, and Thailand. Conversely, the presence of short-run asymmetry was only confirmed for Malaysia, Thailand, and Japan. The GIRF estimations, on a general note, demonstrated that a positive $FDI$ shock appears to significantly influence the $GDP$ of most sample countries, while a negative $FDI$ shock has a relatively minor impact. Therefore, the GIRF analysis provides further additional evidence supporting the asymmetric influence of $FDI$ shocks on $GDP$ across the selected Asian countries. In conclusion, when examining the repercussions of $FDI$ fluctuations on a country’s $GDP$⁠, it is essential to account for asymmetry in the analysis; otherwise, the empirical procedures may lack proper specification and raise doubts concerning the reliability of the findings. Overall, the results reveal nonlinear and context-specific responses in both advanced and emerging markets, indicating that FDI impacts are neither uniform nor linear. This underscores the importance of designing economic policies that consider each country’s structural characteristics and institutional context. Our analysis further demonstrates that FDI shocks can produce heterogeneous outcomes, depending on the direction of inflows and the country’s absorptive capacity.

Several policy implications would emerge based on our empirical results. First, we propose that to maximize the benefits of positive FDI shocks, host countries should strengthen absorptive capacities through targeted investments in human capital development, innovation ecosystems, and financial sector reforms. Specifically, we now recommend that governments prioritize policies such as improving the quality of education and workforce skills, investing in innovation and research infrastructure, creating a more stable and attractive environment for long-term foreign investment, strengthening financial market development to facilitate investment absorption, and enhancing institutional transparency and governance to retain high-value foreign investors. Second, to mitigate the adverse effects of negative FDI shocks, we suggest diversifying FDI sources, promoting domestic investment as a buffer, supporting domestic entrepreneurship and SMEs to reduce dependency on foreign capital, and establishing contingency policies such as countercyclical fiscal measures to sustain growth momentum during downturns and to cushion against external FDI volatility. Third, our findings are relevant to governments that embrace open economic policies, underscoring the need for them to allocate added consideration to the FDI-sector-based analysis (i.e. education, health, industry, infrastructure, and telecommunication). Fourth, the findings from our control regressors indicate the importance of the government’s monetary, fiscal, and societal standings committed to such robust key factors. These factors encompass the development of national human capital, adjustments to trade-openness policies, and careful consideration of the country’s demographic shifts—such as population growth rate, population age structure, and per capita income levels.

For future research, although FDI appears to have an asymmetric impact on the real GDP on the aggregate level of data across Asian countries, however, this dynamic may differ when conducting a sectoral-specific analysis among these countries. Forthcoming research could adopt an analysis approach utilizing detailed sectoral-level data or may incorporate FDI stock data to further enrich the understanding of FDI’s asymmetric effects on sustainable GDP growth across Asian countries. Similarly, future research could highlight whether the empirical long- and short-run FDI asymmetric results vary by utilizing diverse types of population growth. Furthermore, future research could incorporate a wider array of control variables such as a country’s income level, population size, and population age structure which could certainly contribute to more robust assessments of the FDI-growth relationship.

The supplementary material for this article can be found online.