The Investment development path theory: Evidence from developing countries’ agricultural sector

The theory of investment development path

According to Kumar and McLoed (1981), the theory of the IDP was propounded by John H. Dunning in 1979. This theory holds that the net outward direct investment situation of an economy is methodically correlated with its economic progress, vis-à-vis other countries. Revisions of the theory are contained in Dunning (1981a, 1986, 1988a, 1993), Narula (1993). Essentially, patterns of the relationship between GDPs per capita (GDPPC) and outward FDI less inward FDI define five idealised stages of development (Figure 1). The FDI pattern is in turn governed by the ownership, location, and internationalisation advantages of the indigenous and foreign firms (Dunning, 1988a, 1993, 1995, 1998).

In stage I, the place-specific merits of a country are inadequate to entice IFDI (Dunning, 1988a, 1993, 1995, 1998). Those attributable to non-artificial assets are, however, exempted. Limited local markets, unsuitable economic arrangements, or political management strategies are evidence of the lack of place-bound assets. These can reflect insufficient infrastructure and poor capacity of the labour force. The theory notes that in stage I, outward direct investment is less probable (Narula, 1993).

The continual little outward investment signal of a rise in inward direct investment is symptomatic of stage II (Dunning, 1988a, 1993, 1995, 1998). Foreign firms respond to the import restriction policies adopted in response to happenings in stage I. This is reflected in foreign firms investing in import substitution industries because of either an increase in the economy's size or the citizens' purchasing power. Foreign firms take advantage of the existing level of imperceptible resources including technology, registered marks, and management skills (Narula, 1993). The export sector would respond by increasing exports of largely non-artificial resources and prime products with some level of backward and forward linkage into labour-exhaustive little technical knowledge (Narula, 1993). By implication, a country should have some looked-for place features to entice IFDI, contingent on its development strategy, and preference for technical capacity development of local businesses.

In stage III, IFDI decelerates whilst outward foreign direct investment (OFDI) accelerates moving the negative net OFDI towards zero. Building on the capacity and lessons from stage II, in stage III, the technological capabilities of the country are more and more directed to the production of standardised goods (Narula, 1993). As the rise in incomes that started in stage II continues, citizens would clamour for superior attribute goods, inspired partly by the increasing keenness amid the delivering organisations (Narula, 1993). Dunning and Narula (1996) postulate the following: 1. Relative merits in labour-exhaustive undertakings will decline. 2. Local wages will increase. 3. OFDI will be sought more and more by countries at earlier stages of the IDP. 4. The initial possession recompenses of foreign businesses also start to get worn out, as local businesses obtain the modest merits and vie with them.

In stage III, firms' ownership merits influenced by control of trademarked assets will be as those of foreign firms in the country. As domestic firms develop capacity, the role of state-engendered ownership merits is probably less important. Stage III is typical of emerging economies (Frenken and Mbuvi, 2017).

As the OFDI continues to rise from stage III, the level rises to equal to or exceeds that of the IFDI (Dunning, 1988a, 1993, 1995, 1998). This is stage IV. In this stage, domestic firms can now effectively vie for market and resources with foreign-owned firms in the home country and can enter foreign markets as well (Narula, 1993). With increased human capital and technology, capital-intensive production techniques will be employed in the production of contemporary products. In the light of ownership, location and internationalisation paradigm, the location advantages will be built largely on created assets (Dunning, 1988a, 1993, 1995, 1998). In the view of Dunning (1993), the ownership merits will be more of a ‘transaction’ than an ‘asset’. OFDI will continue to grow relative to IFDI in stage IV.

The last stage, V, is described as a home for developed countries. This stage has some features (Dunning, 1988a, 1993, 1995, 1998). 1. There is a rising tendency for international business dealings to be internalised and become intra-multinationals rather than across borders. 2. No one country has complete domination of produced assets. 3. The multinationals’ ownership merits will depend less on their economy’s non-artificial assets’ endowment but increasingly on their capacity to obtain resources and on the capacity of businesses to form their merits competently and to feat the gains of international joint control. 4. Firms become globalised, and their countries of origin become indistinct. 5. The multinationals link geo-political gulfs and integration as such they no longer function with the interests of their countries of origin. They trade, acquire resources and process these in various countries, taking advantage of created and natural assets based on their lead best interest. 6. The ownership and geographical boundaries of the organisation become unclear as they get involved in a more and more interlocking network of trans-border supportive arrangements. 7. As the place-bound assets of countries increase in similarity, the inward and outward foreign direct investments are likely to match each other. The movements in the IFDI and OFDI would lead to unusable or fluctuating net OFDI. Essentially, the peaks and troughs of net OFDI become transient (Narula, 1993).

It is obvious from Figure 1 that a rise in GDPPC is connected to a rise in the position of the IDP. To some extent, this supports the common practice of using GDP per capita as an indicator of development (Abd Hakim et al., 2022; Macek, 2014; McNabb, 2018; Minh Ha et al., 2022; Neog and Gaur, 2020; Ranis, 2004; Radulović and Kostić, 2020; Santiago et al., 2020), measured as the human development index (HDI). Recently, Djokoto (2021b) found that the IDP explains human development (HD). Further, using the GDP per capita as an indicator for IDP, Djokoto (2022) showed that GDP per capita is positively related to human development. It is worth noting that although GDP per capita is an ingredient in the construction of the HDI, it is not the only ingredient.

Empirical evidence

The empirical literature is summarised in Table 1. Since the seminal paper of Dunning, the earliest empirical work was published in 1996 and the latest in 2022. The scope of the studies included single countries, regions, and development groups. The data structure varied from time series through cross-sectional to panel. Charts have been the main analytical tool. Others have been tables and regression analyses. There is evidence of all stages of the IDP. Whilst some empirical evidence confirmed the theoretical stages, others were inconsistent with the theory.

One of our previous research projects studied the IDP of Small States and segregated Small States into both developed and developing countries (Djokoto, 2021a). The exchange rate influenced the net outward foreign direct investment per capita (NOFDIPC) for the developing, the developed and the combined sample. Inflation significantly influenced NOFDIPC for developing and developed Small States. Whilst the effect of the latter was negative, that of the former was positive. In the combined sample, the effect of inflation was neutral. Regarding human capital, the effect was significant for developing Small States and the combined sample but not for developed Small States. In the case of trade openness, the coefficient was negative and statistically significant for developed Small States but negative and statistically insignificant for developing Small States and the combined sample.

It can be observed from Table 1 that no IDP study focused on agriculture. Their assessments avoided higher-level statistical analysis. Our study goes beyond the diagrammatic illustration of the IDP to include econometric analysis with control variables. Our data covers developing countries’ agriculture.

Study design

The quantitative research approach is employed by relying on secondary data. This has a cross-sectional dimension (countries) and a time dimension (years). Hence, panel data was used. The data were obtained from public sources.

Model

Before the statistical analysis, we created charts of the IDP for agriculture. This involved a scatterplot of NOFDIPC on the vertical axis and GDPPC on the horizontal axis and fitted with a polynomial trend line. Although the origins of the IDP lie in the chart depiction of the nexus between NOFDIPC and GDPPC (Figure 1) (Dunning and Narula, 1996), statistical assessments tend to provide a more rigorous outcome (Dai, 2021; Djokoto, 2021a; Djokoto and Pomeyie, 2021; Fonseca et al., 2016; Satoglu, 2017).

The theory of IDP supposes the relationship between NOFDI and GDPPC (Boudier Bensebaa, 2008; Djokoto and Pomeyie, 2021; Dunning, 1981a, 1986; Durán and Ubeda, 2001, 2005; Frenken and Mbuvi, 2017; Gorynia et al., 2012, 2019; Iacovoiu and Panait, 2014), we specify model 1,

NOFDIPC is the stock of OFDI less than the stock of IFDI divided by the total population (male and female) of a country for the corresponding year. The data on OFDI and IFDI for agriculture was reported as flows and not stocks. Consequently, the first observation for every country was considered the initial flow. The subsequent flows were added consecutively to build the stock. The NOFDIPC was used to reduce the magnitude of the values to be comparable especially to the GDPPC as with in existing studies (Djokoto, 2021a; Djokoto and Pomeyie, 2021; Djokoto, 2022). The use of logarithms is avoided in NOFDIPC as it would mask the nonlinearity that we seek to measure. GDPPC is nominal GDP divided by the total population. We follow Dunning (1981a, 1986, 1988a, 1993) in this regard.

However, other factors could explain NOFDIPC other than GDPPC (Andreff, 2003; Djokoto, 2021a; Durán and Ubeda, 2001, 2005; Frenken and Mbuvi, 2017). These can be discussed from two perspectives: the host country and the home country (Paul, 2014). We focus on the home country determinants based on the data employed. In a review of IDP studies, Sawitri and Brennan (2022) identified factors including international trade, exchange rate, human capital and inflation as other determinants of OFDI.

Hence,

Where the other variables are controls, namely trade openness (AGTO), the exchange rate (EXRATE), human capital (HC) and inflation (INFLA). AGTO is defined as the sum of exports and imports to the ratio of the gross domestic product for the agricultural sector. Both inward and outward FDI engenders trade. Accessing new markets by multinationals (MNEs) could start with exports of finished products to the to-be host country. Whilst in the host country, the AM could export intermediate finish products to the home country as well as other countries. Where resources are the attraction for the AM, trade could involve imports from the to-be host country. Thus, trade influences both outward FDI and inward FDI (Buckley et al., 2007; Djokoto, 2021a; Frenken and Mbuvi, 2017; Tolentino, 2008). EXRATE is the official exchange rate captured as the annual average of the local currency per US dollar. The increase in the value of the local currency will cause a decline in local currency resources to invest abroad (Buckley et al. 2007; Paul, 2014) which enhances the level of outward FDI. Also, the increase in value of the local currency, makes products and services more expensive. This decreases the attractiveness of exports relative to FDI; hence, positively influences FDI going out. Thus, the effect of the local currency’s value may harm FDI going out (Bhasin and Jain 2013; Djokoto, 2021a, Paul, 2014).

HC was captured as the secondary school enrolment as a percentage of gross enrolment following Djokoto (2022). A skilled labour force is an ownership advantage that firms must possess to engage in outward FDI and to support inward FDI. Thus, human capital influences both outward and inward FDI (Djokoto, 2021a; Stoian, 2013; Tolentino, 2008).

INFLA was defined as the annual growth rate of the consumer price index following Djokoto (2023), Amal and Tomio 2012; Djokoto, 2021a; Paul, 2014 and World Bank (2023a). In the presence of low macroeconomic stability, businesses will probably seek stable economic environments outside the home country. Proxying economic stability by inflation, a less volatile or more volatile inflation rate points to a positive business environment, that encourages a firm's outward relocation (Amal and Tomio 2012; Djokoto, 2021a; Paul, 2014). We specified equation 2, thus:

Equation 3 was estimated for the appropriate powers of the GDPPC that were established econometrically.

Data

The data consists of a panel of 55 developing countries in Table 2 from 1991 to 2021 based on United Nations (2022). The data include a total of 885 observations. Data to construct NOFDIPC and GDPPC were obtained from the Food and Agriculture Organization Corporate Statistical Database (FAOSTAT) (2023a, b) whilst data for EXRATE, HC and INFLA were obtained from the World Development Index of the World Bank (2023a, b). The countries included in the data were based on the availability of data from the sources. Further, to ensure consistency, all data was extracted from the United Nations data system.

Estimation procedure

First, we established the polynomial order of the GDPPC using the information criteria (Akaike, 1974; Schwarz, 1978). Second, we expressly tested for the existence of endogeneity between our key variables, NOFDIPC and GDPPC. Thirdly, we estimated equation 3 using panel fixed effects (FE) and random effects (RE) estimators, selected the appropriate specification based on the Hausman test (Hausman, 1978) and tested for violations of the classical regression; serial correlation (Wooldridge, 2002) and heteroscedasticity using the modified Wald test for heteroskedasticity (Greene, 2000). The third step was applied to each model during the robustness check of the estimates of the GDPPC, GDPPC² and GDPPC³ to the control variables.

Descriptive Statistics

The minimum NOFDIPC is -180.1437 (Uruguay, 2008) with a maximum of 32.5600 (Malaysia, 2011) (Table 3). The mean of -4.4210 is close to that of Mozambique in 2008. Based on the standard deviation of 17.2928, the variance is more than the mean suggesting overdispersion of the data. Similar overdispersion can be observed with GDPPC². Except for GDPPC and HC for which the standard deviation is less than the respective means, for all other control variables, the standard deviation exceeds the mean.

Determining the order of the polynomial

As the observation in Figure 2 is based on the scatter plot of NOFDIPC and GDPPC, it may well be that incorporating other control variables could change this. Thus, we estimated models 1 – 3 to test this. The AIC and BIC for model 3 are the lowest (Table 4). This suggests that model 3, with the cubic functional form, a polynomial of order 4 is appropriate for the curve based on the estimations. This conforms to the cubic curve shown as the trend line in Figure 2. This is in line with our specification of equation 3 to be modelled.

Figure 2. Chart of investment development path for developing countries' agriculture.

Test for endogeneity

In an earlier study with other colleagues, we found that agricultural FDI explained agricultural GDP (Djokoto et al., 2022). Other studies made similar findings ((Djokoto, 2013; Gunasekera et al., 2015) whilst the reverse is also true (Djokoto, 2012; Kassem and Awad, 2019; Lv et al., 2010), hence there can be an endogeneity problem. Rather than anticipate and model accordingly, we proceeded to test for endogeneity (Table 5). First, we modelled GDPPC as the dependent variable with all others as exogenous variables. We predicted the errors GDPPC_ue and did the same for GDPPC ² and obtained GDPPC²_ue as well as GDPPC³ and obtained GDPPC³_ue. In Model 7, the predicted terms were introduced as additional explanatory variables. We then tested the significance of the GDPPC_ue, GDPPC²_ue and GDPPC²_ue with a chi-square test. Failing to reject the null hypothesis suggests that the error terms are not correlated with the NOFDIPC, hence the suspicion of endogeneity is not borne out by the data. Following the absence of endogeneity, equation 3 was estimated with FE and RE estimators. The Hausman test showed that the hull hypothesis that the differences between the matched coefficients are not systematic could not be rejected. Consequently, the RE estimation was preferred to fixed effects. The errors were found to be serially correlated and non-constant (heteroscedastic). We addressed this with robust standard errors. The serial correlation was resolved by introducing a first lag of NOFDIPC, d.NOFDIPC as appropriate.

Robustness

The estimates of GDPPC, GDPPC² and GDPPC³ in 8 are consistent with those of models 9 – 13 (Table 6). The estimates of AGTO, EXRATE, HC and INFLA are also consistent across models 9 – 12 and model 13.

We acknowledge the role of multicollinearity especially when using powers of the same variable as explanation variables. An important consequence of multicollinearity is the inflation of the standard errors leading to the invalidation of the hypotheses tests. Our hypothesis tests are not appreciably invalidated, moreover, the estimates are robust and consistent with theory, thus, we did not consider the impairment of our results due to multicollinearity.

Discussion of the IDP for agriculture

We focus on model 13 for discussions. The negative and statistically significant coefficient of GDDPC confirms the curve declines initially as in Figure 2. The statistically insignificant coefficients of GDPPC² and GDPPC³ suggest points of inflexion. The positive and negative signs respectively show the rise and fall before and after the points of inflexion. The estimates correctly depict Figure 2 notwithstanding the introduction of the control variables. It is informative to note that notwithstanding the cubic function in Figure 2, confirmed by the model selection in Table 2 and estimates in Table 5, the curve remains below the GDPPC axis. Indeed, the curve does not show any significant rise after its first decline. Juxtaposing Figure 2 and Figure 1, developing agriculture is in stages I and II of the IDP. The stage I and II found is in line with the theory of the IDP proposed by Dunning (1981a, 1986, 1988a, 1993) and Narula (1993). This is consistent with the finding for the total economy studies for developing countries (Andreff, 2003; Satoglu, 2017) and other regions (Boudier-Bensebaa, 2008; Borowicz, 2021; Dai, 2021; Gorynia et al., 2009, 2010a,b, 2012; Zhubikenov, 2022). However, total economy studies of other regions found IDP stages other than I and II (Djokoto, 2021a; Iacovoiu and Panait, 2014; Kuzel, 2017; Trąpczyński et al., 2019).

The statistical significance of the GDPPC is in line with the evidence that the size of the economy determines FDI in agriculture (Djokoto, 2012; Djokoto, 2021c; Husmann and Kubik, 2019; Rashid and Razak, 2017). It also confirms the outcome of early studies of Dunning (1981a, 1986, 1988a, 1993) and Narula (1993). The statistical significance of the GDPPC also confirms the existence of the IDP. This should be viewed within the framework of the evidence that GDPPC, IDP and HD are positively correlated (Djokoto, 2021a,b,c, 2022; Djokoto and Pomeyie, 2021).

The continual low or negligible OFDI and rise in IFDI are observable. The outcome means that the location-specific advantages of developing countries are inadequate to attract IFDI. These can be attributable to inadequate infrastructure and poor capacity of the labour force. Foreign firms may be interested in investing in import substitution industries because of either an increase in the economy's size or the citizens' purchasing power. The weakness of the developing countries encourages foreign firms to take advantage of the existing level of imperceptible resources including know-how, registered marks, and managerial expertise. Foreign firms may also invest in export sectors increasing exports of largely non-artificial resources and prime commodities with some level of backward and forward linkage integration into labour- exhaustive low know-how.

Discussion of control variables

The positive and statistically significant coefficient of AGTO implies that trade enhances NOFDI. Aside from accessing new markets with new products at the start of internationalisation, home, and foreign affiliates engage in the trade of raw materials and finished and semi-finished products. Where divestments occur, the markets could be filled with products from foreign affiliates of the home country. In all these, the constituents of trade openness, imports and exports take place, hence, this is a positive sign. Djokoto (2021a) found a negative sign of the estimates of trade that were statistically insignificant for both developing and the combined Small States except for developed Small States. Djokoto (2021a) did not provide reasons for the results on the trade variable.

The negative sign of the coefficient of EXRATE is consistent with that of Djokoto (2021a), there is a departure regarding the statistical significance. The coefficient of INFLA is also negative but statistically insignificant. Our finding is contrary to that of Djokoto (2021a) who found positive and statistically significant coefficients for developing Small States as well as the combined sample, but negative and statistically significant coefficients for developed Small States. As in the case of the other control variables, Djokoto (2021a) did not provide reasons for the significant coefficients.

Regarding the control variables, in summary, agricultural trade openness enhances agricultural net foreign direct investment whilst the exchange rate has the opposite effect. Inflation does not affect agricultural net foreign direct investment.