In line with the key issues in the paper, the theoretical review is constituted into four parts. The first, theories of development seek to explain the differences in the level of development of countries, one of the foci of the paper. Space is provided for FDI theories as these explain the role of the key variable that influences DI. DI is the dependent variable, thus deserves space, hence the investment theories. To tie-in, the interaction of FDI and DI, the theoretical explanation of the link between FDI and DI is outlined as well.

Many theories have been propounded regarding economic development. These have been classified differently. The classification used here is modernisation, dependency, world systems and globalisation. Modernisation theory uses a systematic process to move underdeveloped countries to a more sophisticated level of development ( Reyes, 2001). This theory which stresses the importance of political development in the progress and climactic improvement of a nations’ economic standing, also acknowledges social and cultural reforms. Further, it seeks to explain inequality within or between states by identifying different values, systems and ideas held by different nations ( Martinussen, 1997). Also, it gives attention to the shift of modern technology and development institutions and labour habits complementary to industrial production ( Chase-Dunn, 1975).

Dependency theory that seeks to improve modernisation theory, combines elements from a neo-Marxist theory and adopts a revolution of underdeveloped nations model ( Reyes, 2001). This, together with the Marxist position seeks to explain the origin of surpluses, the basis of theoretical evaluation of progress, and inequality. The divide between developed and under-developed countries is well established. The reason for the difference between the divide and what interventions are required to narrow the divide engages the attention of this theory. ( Haque, 1999; Muuka, 1997; Shen & Williamson, 2001; Todaro, 2000).

The "World System" is multiple cultural systems with a single division of labour. The basic feature of this system is having a pool of labour in which different divisions and areas are dependent upon each other in exchanging the provisions of those areas ( Wallerstein, 1974). In using other levels of quantitative analysis, this theory argues that international trade specialisation and transfer of resources from less developed countries to developed countries prevents development in less developed countries by making them rely on core countries and by encouraging peripheralisation ( Szymanskiv, 1982). The theory views the world economy as an international hierarchy of unequal relations. Through the world system, a country can change its position in the global hierarchy ( Szymanskiv, 1982).

Globalisation as a theory of development uses a world mechanism of greater integration with emphasis on the sphere of economic transactions ( Reyes, 2001). Whether as an event of interdependence throughout different countries of the world in different aspects of communication, trade, and finance or continuous and widespread unification among different countries, it focuses on communications and international ties directed at cultural and economic factors in communication systems. This theory explains inequality by identifying cultural and economic factors in global connection ( Zineldin, 2002). Globalisation endangers global competition. Whilst this could hurt especially developing countries, firms in developing countries can respond by increasing production and efficiency thereby improve their economic situation. Also, rationalisation of production on a global scale and the spread of technology results from globalisation ( Intriligator, 2004; Stiglitz, 2004a; Stiglitz, 2004b; Tanzi, 2004).

Foreign direct investment theories can be viewed from three perspectives. The first perspective, which is internationalisation theory, explains why firms often prefer FDI to license as a strategy for entering a foreign market ( Hymer, 1976). In this theory, FDI is preferred for licensing and exporting. Licensing may result in a firm giving away valuable technological expertise to a potential foreign investor at a fee. This does not give a firm control over manufacturing, marketing, and strategy in a foreign country that may be required to maximise its profitability. Unfortunately, the competitive advantages of management, marketing, and manufacturing capabilities are not amenable to licensing.

The second perspective relates to the patterns of FDI. In oligopolistic industries, firms invest in other countries as a strategy by following their domestic competitors ( Knickerbocker, 1973). Related to this is the product life cycle hypothesis ( Vernon, 1966). Firms invest in other advanced countries when local demand in those countries grows large enough to support local production. Production is subsequently shifted to developing countries when product standardisation and market saturation gives rise to price competition and cost pressures.

The third perspective is the Dunnings’ eclectic paradigm. Dunning (1977); Dunning (1988); Dunning (2001) stated that the extent, geography, and industrial composition of foreign production undertaken by the multinational enterprise is determined by the interaction of three sets of interdependent variables which, themselves, comprise the components of three sub-units, namely; ownership, location and internationalisation (OLI). All other factors unchanged, the greater the competitive advantages of the investing firms, relative to those of other firms, the more they are likely to be able to engage in or increase, their foreign production. This is the ownership competitive advantage. For the location, the more the immobile, natural or created endowments, needed by the firms to use jointly with their competitive advantages, favour a presence in a foreign rather than a domestic location, the more firms will choose to supplement or take advantage of their ownership specific advantages by engaging in FDI. The multinational enterprise, thus, would undertake activities to add value to their operations. Internalisation, the final competitive advantage, offers a framework for evaluating alternative ways in which firms may organise the creation and exploitation of their core competencies. These range from buying and selling goods and services in the open market, through a variety of inter-firm non-equity agreements, to the integration of intermediate product markets and outright purchase of a foreign corporation.

The many theories of investment include accelerator theory, cash flow theory, neoclassical theory, modified neoclassical theory and Q theory. Chenery (1952); Clark (1917) and Koyck (1954) among others, emphasised the accelerator model that sought to explain business cycles. Eisner & Strotz (1963); Eisner & Nadiri (1968) incorporated profit and investment equation into the accelerator theory.

Cashflow, the resources left after paying dividends to shareholders determines investment from internal sources. Past profits drive future investment decisions ( Kuh, 1963; Meyer & Kuh, 1955; Strong & Meyer, 1990). The decision to invest is informed by the cost of capital which in turn is driven by the profit maximising behaviour of the firm ( Jorgenson, 1963; Jorgenson, 1971; Modigliani & Miller, 1965). Biscoff (1971) extended the neoclassical theory to show that it is possible to alter the capital-output ratio so that later, the substitution effect of output goes to zero. Consequently, the investment can be effectively induced by changes in the capital-output ratio.

The Q theory of investment notes that, if investors seek to maximise the market value of the firm, they will add to their capital stock whenever the marginal addition to the firm’s market value exceeds the replacement cost of the capital stock ( Brainard & Tobin, 1968; Tobin, 1969). This theory is driven by the ratio of the market value of capital to its replacement cost, unlike the neoclassical theory that is driven by the cost of capital.

The fourth theoretical perspective is the interaction between FDI and DI. Markusen & Venables (1999) theorised the relationship between multinational enterprises (MNEs) and domestic firms on the entry of MNEs to be a competition effect and a linkage effect. Regarding the competition effect, the entry of MNEs increases competition in the final product industry and reduces the profitability of domestic firms in the same industry. Consequently, domestic firms do exit the market. For the linkage effect, the entry of multinational corporations (MNCs) could cause the demand for domestic production of intermediate inputs to rise. This tends to create an increased number of domestic firms in the intermediate inputs industry. Barro & Sala-i-Martin (2004) have also acknowledged these effects.

Barrios et al. (2005) posit that as the number of MNEs increases in the host country, the number of domestic firms might drop first and then rise. Assume that MNEs enter the downstream (final product) market, an increase in the total number of firms (both foreign and domestic) in the downstream industry will decrease the price level for the final product. The possibility is the reduction of profit for all firms and thereby forcing some domestic firms to exit. This would happen if the MNEs are more productive than domestic firms. On the other hand, the possible linkage effect can increase the number of upstream domestic firms and then reduce the cost of production in the downstream industry for both MNEs and domestic firms. Therefore, the number of domestic firms in the host country will eventually go up.

The effect of FDI on DI could be complementary, substitutional or neutral. In the developing country context, Agosin & Machado (2005); Ahmad et al. (2018); Djokoto (2013); Josue et al. (2014); Oualy (2019) and Wang (2010) have reported the neutral effect of FDI on DI in developing countries, in the short run. Kim & Seo (2003) found a significant crowd-in effect of FDI on DI for the whole of Korea. In the long run, whilst Agosin & Machado (2005); Morrissey & Udomkerdmongkol (2012) and Mutenyo & Asmah (2010) reported crowding out effect of FDI on DI for Latin America, developing countries and sub-Saharan Africa respectively, Djokoto et al. (2014) and Farla et al. (2016) found crowding-in effect of FDI on DI for Ghanaian agriculture and developing countries respectively. Djokoto et al. (2014) attributed the positive effect to technology diffusion and spillover of management know-how by MNEs and vertical inter-firm linkages with domestic firms.

Mileva (2008) found a crowding-in effect of FDI on DI for economies in transition using data from 1995–2005 in the short run and long run, despite the presence of less developed financial markets and weaker institutions. Cooperation instead of competition can be adduced for the crowd-in effect. Jude (2019) and Kejžar (2016) reported switching effects of FDI on DI for transition economies; short-run crowding-out effect to long run crowding-in effect. The magnitude of the possible linkage and spillover effects over time overcomes the initial competition effect. Also, the period for the long run is such as would allow completed plants to run for some time to generate output and associated benefits. Technology, knowledge transfer, employment and expenditure on social responsibility would come to fruition.

Gallova (2011) reported that FDI exerted no effect on DI for Bulgaria and Romania but crowding in effect for Croatia and Slovenia. For the Balkans, for the period 1993–2009, the effect was crowding out. Gallova (2011) explained that foreign companies do employ the services of the same suppliers as their parent companies that are not in the host country. MNEs tend to also bring with them to the host country, foreign producers from which they take the components necessary for their production. Additionally, foreign-owned companies are so strong in individual sectors and effectively functioning. When domestic firms fail to assert themselves effectively and establish cooperation with MNEs they tend to be crowded out of the market. In the long run, however, based on the panel data from 1993–2009, Gallova (2011) found no long-run effect for the Balkans and the individual countries. For the Czech Republic and Hungary, Mišun & Tomšík (2002) concluded on a long run crowding in effect but the crowding-out effect for Poland. Other studies on developed countries also concluded on mixed outcomes regarding FDI and DI. Pilbeam & Oboleviciute (2012) found a short and long-run crowd-out effect for EU-15 (Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain, Sweden and the United Kingdom) but no effect for EU-12 (Bulgaria, Czech Republic, Estonia, Cyprus, Latvia, Lithuania, Hungary, Malta, Poland, Romania, Slovenia, and Slovakia). In the work of Polat (2017), the effect of FDI on DI was neutral for Organisation for Economic Co-operation and Development (OECD) countries for data covering 2006 to 2013, just as Wang (2010) for developed countries in the long run. In the short run, however, Wang (2010) found the crowding-out effect for developed countries.

From the empirical review, some studies dwelled on whole economies across countries whilst others covered one or single economy. One single economy study addressed agriculture. Food manufacturing, the offtaker of agricultural production and the product supplier for marketing in the agribusiness value chain, has not been given attention. This paper focuses on this.

Some previous studies used time-series data ( Chen et al., 2017; Djokoto, 2013; Djokoto et al., 2014; Gallova, 2011; Kim & Seo, 2003) whilst others employed panel data ( Agosin & Machado, 2005; Ahmad et al., 2018; Budang & Hakim, 2020). This study used panel data. The data consist of 49 countries for the period 1993 to 2016 ( Table 1). The description of the data and measurement are outlined in Table 2. All the data was obtained from FAOSTAT (2020). The panel structure enables pooling the observations for each unit over time resulting in more observations and smaller standard errors. Further, the data structure allows for analysing the dynamic behaviour of units and dynamic models; using lagged outcome variables as explanatory variables ( Greene, 2003; Gujarati & Porter, 1999; Wooldridge, 2002).

Agosin & Machado (2005) developed a theoretical model based on a neoclassical investment model using partial adjustment of capital stock and adaptive expectations of economic growth in which FDI has been included. This model has been applied in the study of the relationship between FDI and DI ( See for example, Budang & Hakim, 2020; Oulay, 2019; Pilbeam & Oboleviciute, 2012). The Agosin & Machado (2005) model is adapted and specified in Equation 1.

D I i , t = β 0 + β 1 D I i , t − 1 + β 2 D I i , t − 2 + β 3   F D I i , t + β 4   F D I i , t − 1 + β 5   F D I i , t − 2 + β 6   G R i , t − 1 + β 7   G R i , t − 2 + β 8   T I M E i , t + ε i , t 1

Where ε _i,t is the error term.

To account for the levels of development of the countries in the data, for which reason the effects of FDI on DI may differ, Equation 1 is re-specified:

D I i , t = β 0 + β 1   D I i , t − 1 + β 2   D I i , t − 2 + β 3   F D I i , t + β 4   F D I i , t − 1 + β 5   F D I i , t − 2 + β 6   G R i , t − 1 + β 7   G R i , t − 2 +   β 8   F D I _ D V P i , t + β 9   F D I _ D V P i , t − 1 + β 10   F D I _ D V P i , t − 2 + β 11   D I _ D V P i , t − 1 +   β 12   D I _ D V P i , t − 2 + β 13   F D I _ T R S i , t + β 14   F D I _ T R S i , t − 1 + β 15   F D I _ T R S i , t − 2 + β 16   D I _ T R S i , t − 1 + β 17   D I _ T R S i , t − 2 + β 18   T I M E i , t + γ i , t 2

All the variables and their sources are described in Table 2. The classification of countries based on the level of development reflects the basic conditions ( United Nations, 2020). It does indirectly capture some macroeconomic variables that could otherwise have been included as control variables.

Let β _STk be the short run effects, where k=1,2, 3,4 are developing country, economies in transition, developed countries and all development groups, respectively. Then,

β ^ S T 1 = β 3 + β 8 = 0 3

for developing countries.

In the case of transition economies,

β ^ S T 2 = β 3 + β 13 = 0 4

Whilst for developed countries,

β ^ S T 3 = β 3 = 0 5

For the three country groups effects, the null hypothesis for the short run is

β ^ S T 4 = β 3 + β 8 + β 13 = 0 6

Failure to reject these null hypotheses with a chi-square test implies FDI has no short-run or contemporaneous effect on DI. Rejection of the null hypothesis and that β _STk > 0 means a contemporaneous crowd-in effect of FDI on DI. Alternatively, β _STk < 0 means a contemporaneous crowd-out effect of FDI on DI.

The long-run effect is represented by β ^ L T 1 , where k=1,2,3,4 as previously. Then, the null hypotheses for the long-run effects ² are

β ^ L T 1 = ∑ j = 1 5 β ^ j + ∑ j = 8 12 β ^ j = 1 7

for developing countries. For transition economies

β ^ L T 2 = ∑ j = 1 5 β ^ j + ∑ j = 13 17 β ^ j = 1 8

In the case of developed countries,

β ^ L T 3 = ∑ j = 1 5 β ^ j = 1 9

For all development groups, in the long run,

β ^ L T 4 = ∑ j = 1 5 β ^ j + ∑ j = 8 17 β ^ j = 0 10

a. With a chi-squared test, if it is not possible to reject the null hypothesis that β ^ L T 1 = 1 , then, in the long run, an increase in FDI of one currency unit, becomes one currency unit of additional total investment. Stated differently, investment by MNEs simply matches one-to-one to investment by domestic firms, and no macroeconomic externalities are stemming from FDI.

b. If the null hypothesis, β ^ L T K = 1 is rejected and β ^ L T K > 1, this is evidence of the long-run crowd-in effect of FDI by DI. Thus, in the long run, one additional currency unit of FDI becomes more than one additional currency unit of total investment.

c. Now, consider the case that the null hypothesis, β ^ L T K = 1 is rejected and β ^ L T K < 1, there is then a long-run crowd-out effect. One additional currency unit of FDI leads to less than one currency unit increase in total investment. This means there is a displacement of DI by FDI.

The pertinent literature used the ordinary least squares (OLS) estimator ( Chen et al., 2017; Djokoto, 2013; Djokoto et al., 2014; Gallova, 2011; Kim & Seo, 2003). Agosin & Machado (2005); Pilbeam & Oboleviciute (2012) and Polat (2017) used a generalised method of moments (GMM). GMM has some weaknesses. First, in the presence of the lagged dependent variables on the right-hand side of the equation and the time-invariant country-specific factors, the fixed-effects estimator would yield inconsistent estimates because of the correlation between the lagged dependent variable and the error terms. Taking the first differences to remove the time-invariant country-specific factors still results in a correlation between the error term and the DI. Second, FDI inflows are likely to be endogenous and determined jointly with the DI. Impliedly, there is a two-way relationship between DI and FDI ( Al-sadiq, 2013; Arellano & Bond, 1991; Arellano & Bover, 1995). The system-GMM estimator enables controlling for the unobserved country-specific factors and solves the correlation problem by using a set of internal instruments based on the assumption of no second-order serial correlation in the first-differenced idiosyncratic errors, and the independent variables are weakly exogenous. In this way, the estimated coefficients would then not be subject to bias from an omitted variable. Using a series of internal instrumental variables based on lagged values of the dependent and independent variables, the system-GMM estimator solves the endogeneity problem as well. All these are accomplished by combining in one system, the regression in differences with the regression in levels under the assumption that there is no correlation between the differences of the right-hand side variables and the unobserved country-specific effects ( Al-sadiq, 2013; Arellano & Bond, 1991; Arellano & Bover, 1995; Blundell & Bond, 1998). The assumption of no serial correlation is testable. Whilst first-order autocorrelation (AR(1)) of the errors is permissible, that of the second-order (AR(2)) is inadmissible. As internal instruments are used, the Sargan test tests whether the overidentifying restrictions are valid. Rejection of the null hypothesis would indicate that the instruments are not valid, therefore, the estimates are not reliable.

Due to the unbalanced nature of the panel, the number of observations for the variables is not the same. The observations ranged from 611 to 1,344. The growth rate (GR) for food manufacturing value-added averaged about 3% ( Table 3). The means of DI_1 for developing, transition and developed economies were respectively, 0.0369, 0.0079 and 0.1083. These are statistically distinguished ( Table 4). Thus, developed countries on average accumulate about 11% of food manufacturing domestic investment as GDP whilst developing countries manage about 4%. Economies in transition posted less than 1%. The positive value of FDI implies that there has been a transfer of FDI from one economy to the other. This is symptomatic of international capital transfer ( Dunning, 1977; Dunning, 1988; Dunning, 2001; Reyes, 2001). The means of FDI_1 of the development groups is likewise statistically distinguished. However, the values constitute less than 1% of food manufacturing value-added. As in the case of DI, developed countries post a higher value of 0.99%. Transition economies, however, seem to attract more FDI dollars of food manufacturing GDP than developing economies.

As the key variables are FDI and DI, linear correlation coefficients were computed (second panel of Table 4). There is a significant positive correlation between FDI and DI for each developing country group. For the country groups together, however, the positive relationship is not statistically distinguishable from zero. Similarly, in the long run, there are positive significant relationships between FDI and DI but no significant relationship for all country groups. Although the linear correlation coefficients could give an early indication of the effect between the variables, the influence of other variables during the estimation process could change the effect shown by the linear correlation coefficient.

Up to 15 lags were estimated for the system GMM. This is because lag length is known to influence the size of estimates especially, coefficients and standard errors, which are crucial in the calculation of the short and long-run effects of FDI on DI. Based on the number of significant coefficients and the level of significance lag 4 was selected. The selected system GMM results are reported in Table 5 as model 4.

To establish the consistency of the estimates of model 4, four other models; 1–3 and 5, were estimated. Model 1 includes only the key variables; the lags of DI and the FDI and its lags. In model 2, GRs are added to model 1. In model 3, the GRs are dropped, and the TIME variables added. Model 5 includes all variables as in model 4, but without some observations. Data of large economies could influence the overall results. Thus, for developed countries, the observations for India and Brazil were dropped. For economies in transition, Russia was dropped whilst the United States of America was dropped from the observations of developed economies ( Table 1). The only departures are FDI_TRS1 for model 1, FDI_TRS for model 2, and FDI_2 and FRI_TRS for model 5 ( Table 5). For model 1, the coefficient of the variables is weakly significant at 10% whilst those for all other models are statistically insignificant. For model 2, the coefficient of the variable is statistically insignificant together with that of model 5 but the coefficients of the variable for all models are statistically significant at either 1% or 10%. For model 5, the coefficient for FDI_2 is statistically insignificant, whilst one of the lag 2 of FDI is statistically significant at 5% whilst the other three are significant at 10%. It is instructive to note that, despite these minor departures, the size and sign of the magnitudes of the coefficients are comparable. Thus, generally, the estimates are not sensitive to the changes in variables and observations and are consistent and robust.

As the goal of the estimations is to ascertain the short and long-run effects of FDI on DI, the estimates of model 1 – 5, were used to compute the effects ( Table 6). The short and long-run effects are consistent across all five models for developing and developed economies and for the long run for economies in transition. The only departures are for the short run in the case of transition economies; statistically significant effects for model 1 and model 3. This is not surprising. Transition economies in the data are made up of three countries; Armenia, Kazakhstan and the Russian Federation ( Table 1). Thus, the results of the group are more likely to be responsive to the deletion and inclusion of variables and observations than those country groups with larger numbers. Combining all the country groups, the short-run effects for model 1 differ from those of other models. Also, the statistics of the long-run effect of model 5 is statistically insignificant, unlike others that are statistically significant, although the magnitude is like others. Model 5 has two coefficients that were statistically insignificant more than the other models, suggesting high standard errors. This certainly contributed to the statistically insignificant Wald statistic. Overall, however, the results of the short and long-run effects are generally consistent and robust to the estimations of model 4, the standard.

Data have been obtained from the Food and Agriculture Organization of the United Nations: http://www.fao.org/faostat/en/#data

Figshare: Foreign Direct Investment crowding out Domestic Investment in Food Manufacturing, https://doi.org/10.6084/m9.figshare.13591235.v2 ( Djokoto, 2021).

Data are available under the terms of the Creative Commons Zero "No rights reserved" data waiver (CC0 1.0 Public domain dedication).