Machine Learning and Econometric Approaches to FDI: FDI impact on Economic Growth in Georgia (1989-2023)

<p style="text-align: right;"><a href="https://search.crossref.org/search/works?q=10.52340%2Feab.2024.16.04.07&from_ui=yes">doi.org/10.52340/eab.2024.16.04.07</a> <p style="text-align: justify;">The primary objective of this paper is to analyze the impact of Foreign Direct Investment (FDI) on GDP in Georgia by applying both econometric (VECM) and machine learning (Random Forest) techniques. The paper compares the results from these two approaches, draws reliable conclusions, and offers recommendations for the government. To the best of our knowledge, no existing studies have utilized the Random Forest algorithm to analyze the effect of FDI on GDP in Georgia, and this paper addresses that gap.<br />The results from the VECM analysis show that FDI has a positive and statistically significant impact on GDP both in the short and long run. Similarly, the Random Forest model yields the same conclusion, with FDI demonstrating a positive effect, following the influence of gross capital formation. Based on these findings, the study offers recommendations to the Georgian government on strategies to attract more FDI and promote rapid economic growth. <p style="text-align: justify;"><strong>Keywords:</strong> Foreign Direct Investments, Gross Domestic Product, Machine Learning, Random Forest, mice, Econometric. <p style="text-align: justify;"><br /><strong>Introduction</strong> <p style="text-align: justify;">Data science, including artificial intelligence (AI), is advancing rapidly and becoming integral to numerous fields such as medicine, art, economics, and business. Almost every sector now relies on data, and utilizing it effectively has the potential to solve many problems across various domains. However, there is limited research that applies data science—particularly machine learning techniques—in economics and business, especially within the context of Georgia. This paper aims to fill that gap by applying machine learning methods to economic analysis, producing more reliable results, and offering valuable recommendations to the government. Notably, there are no existing studies that use the Random Forest machine learning approach to analyze the impact of foreign direct investment (FDI) on GDP in Georgia. Therefore, another key motivation of this study is to provide a foundation for future research and to encourage others to explore the use of machine learning techniques to investigate various economic topics related to Georgia.<br />One of the key effects of globalization is the movement of capital across borders. Foreign Direct Investment (FDI), defined as an investment where a foreign investor acquires at least 10% ownership in a company in another country, has become one of the most sought-after investment forms for many countries. FDI is especially crucial for developing countries, as it brings in foreign capital, technology, and expertise are considered essential drivers of rapid economic growth. As a result, policymakers often prioritize the development of favorable investment policies to attract FDI into their economies.<br />In Georgia, FDI has long been a central concern for policymakers, economists, and business leaders. There has been ongoing debate about strategies to boost FDI inflows into the country. Graph 1 below illustrates the net FDI inflows to Georgia from 1997 to 2023. <p style="text-align: justify;"><img src="http://eb.tsu.ge/img/nuca%201.png" alt="" width="434" height="245" /> <p style="text-align: justify;">The yellow line on the graph represents trend. Figures indicate that the highest levels of FDI in Georgia occurred in 2007-1.8 Bill and 2022-2.1 Bill. A significant decline was observed in 2009, largely due to the Russia-Georgia war, and again in 2020 because of the COVID-19 pandemic. While net FDI fluctuated over the years, it showed a general upward trend. It is worthwhile to analyze FDI by country and economic sectors. Graph 2 displays the FDI from the top 10 investing countries in Georgia in 2023. <img src="http://eb.tsu.ge/img/nuca%202.png" alt="" width="444" height="228" /> <p style="text-align: justify;">Based on the graph, the largest investors in 2023 were Netherlands with $386 million, followed by the UK ($364 million), the US ($182 million), Turkey ($109 million), and Russia ($103 million). It is also worth examining which sectors attracted the most investment. Chart 1 illustrates FDI by economic sectors in 2023. <img src="http://eb.tsu.ge/img/nuca%203.png" alt="" width="443" height="341" /> <p style="text-align: justify;">The largest share of FDI is directed towards financial and insurance activities (31%), followed by wholesale and retail trade (17%), manufacturing (16%), transportation and storage (9%), information and communication (6%), and real estate activities (6%).<br />While the positive impact of FDI on economic growth may seem evident, there is still considerable debate and ongoing research on the topic. Therefore, reviewing the literature on this subject is essential before proceeding with the modeling. <p style="text-align: justify;"><strong>Literature Review</strong> <p style="text-align: justify;">There are numerous studies examining the impact of Foreign Direct Investment (FDI) on economic growth in Georgia, but only a few have applied statistical analysis. Few studies have employed econometric techniques, and nearly none have applied machine learning approaches. However, econometric methods have been widely used in research on FDI and economic growth for other countries.<br />For example, a study conducted for Bangladesh applied an econometric model, specifically the Vector Error Correction Model (VECM), to investigate the impact of FDI, remittances, and foreign aid on economic growth from 1976 to 2021. The study concluded that all exogenous variables, including FDI, remittances, and aid, had a positive long-term impact on economic growth, although their short-term effects were not significant (Ummya et al., 2023).<br />A 2022 study analyzed the effects of public and private investments on economic growth in 39 developing countries from 1990 to 2019. Using panel data analysis, the study found that domestic private investment had a positive impact, while FDI showed a slightly negative effect on economic growth. It also identified that gross capital formation had a positive effect, whereas government expenditure had a negative impact on growth (Faruque, 2022,8).<br />Goken (2021, 95) examined the relationship between real economic growth and net FDI inflows in Turkey using VECM, covering the period from 1970 to 2019. The study found a positive but insignificant short-term impact of FDI on GDP and a significant but small effect of GDP on FDI. In the long run, the study revealed that while FDI did not significantly impact GDP, GDP had a positive impact on FDI in the long run.<br />სიხარულიძე, ჭარაია (2018) explored the theory of FDI and its experience in Georgia. They found that while FDI had a positive impact on economic growth, its long-term effect was not stable. The study noted that foreign companies were not well integrated with local firms, and the lack of competitiveness, qualifications, and strong linkages between the two hindered more sustainable benefits from FDI. To maximize the impact of FDI, the study recommended appropriate investment policies for Georgia.<br />Hlavacek et al. (2016, 294-303) examined the relationship between FDI and economic growth in Central and Eastern Europe (CEE) from 2000 to 2012. Their panel data model showed a significant positive impact of FDI on economic growth, which was particularly pronounced as these countries joined the European Union, accelerating their integration and attracting more FDI.<br />Simionescu (2016, 187-213,) conducted a study for 28 European Union countries, focusing on the impact of FDI on GDP growth from 2008 to 2014. Using a panel data approach with Bayesian methods, the study concluded that overall, FDI had a positive effect on economic growth in the EU. However, for some countries, FDI did not contribute significantly to growth. For those countries, the study suggested the implementation of targeted economic policies, such as tax reductions and better legislative environments.<br />Pharjiani (2015, 3988-3996) investigated the sectoral impact of FDI on economic growth in 10 CEE countries from 1995 to 2012. The study used various econometric methods, including Ordinary Least Squares (OLS), Random Effects (RE), and Fixed Effects (FE), and found a positive link between FDI and economic growth in each sector, with FDI contributing to increased productivity.<br />Albulescu (2014, 507-512) researched the impact of both inward and outward FDI and Foreign Portfolio Investment (FPI) on economic growth in CEE countries using a panel Generalized Method of Moments (GMM) approach. The study, which covered 2005 to 2012, found that both inward and outward FDI, as well as FPI, positively affected economic growth.<br />Acaravaci et al. (2012, 52-67) explored the long-run causal relationships between FDI, exports, and economic growth in 10 transition European countries, including Bulgaria, Czech Republic, Estonia, and others. Using VECM and Granger causality tests, the study concluded that there were both short- and long-term causal relationships between FDI, exports, and economic growth in four of these countries.<br />Batten et al. (2009,1621-1641) studied the relationship between FDI and economic growth in 79 countries from 1980 to 2003. The study found that FDI had a positive impact on economic growth in countries with higher education levels, developed stock markets, and openness to international trade.<br />Stanisic (2008) examined FDI`s impact on economic growth in seven Southeastern European countries from 1997 to 2005. The study found no positive correlation between FDI and economic growth, attributing this to the transition period these countries were undergoing, which hindered increases in production and employment which abate the positive impact of FDI. <br />Another study by Moudatsou (2003, 689-707) analyzed FDI’s impact on economic growth in the EU from 1980 to 1996. The research concluded that while the impact of FDI on growth differed across countries, there was a positive overall effect when the data was pooled together. <p style="text-align: justify;"><strong>Data and Methodology</strong> <p style="text-align: justify;"><strong>Exploratory & Initial Data Analysis / Overview of the Data</strong><br />The data used in the modeling was collected from the World Bank covering a period of 35 years, from 1989 to 2023. Variables such as the Consumer Price Index (CPI), employment-to-population ratio, labor force participation rate, etc. were also considered, but the final selection of variables were based on their importance and correlations, also avoiding high multicollinearity. The final dataset includes the key economic variables such as:<br />• <strong>GDP</strong> - GDP per capita (USD)<br />• <strong>FDI</strong> – FDI, net inflows per capita (USD)<br />• <strong>GCF</strong> - Gross Capital Formation per capita (USD)<br />• <strong>GOVEX</strong> - General Government Final Consumption Expenditure (% of GDP)<br />• <strong>TRBAL</strong> - Trade Balance on goods and services (% to GDP)<br />• <strong>POPGR</strong> - Population Growth Rate (annual %)<br />The dataset contained some missing values, and an imputation method was applied to address this issue. Missing data analysis involves various approaches, with the choice of method depending on the patterns and mechanisms of the missingness. The pattern of missing data indicates where the missing values occur within the dataset, while the mechanism provides insight into the reasons behind the data`s absence.<br />There are three primary mechanisms for missing data: Missing Completely at Random (MCAR), Missing at Random (MAR), and Missing Not at Random (MNAR). The nature of the missing data mechanism influences the selection of appropriate analysis methods. (Little et.al 2019). <br />Common imputation techniques include: Unconditional mean imputation, where missing values are replaced by the overall mean of the observed values; Conditional mean imputation, or Regression imputation, where missing values are predicted from other variables using regression models; Stochastic regression imputation, which refines regression imputation by adding random noise to predicted values; Hot deck imputation, where missing values are filled using values from similar individuals in the dataset; Cold deck imputation, which uses previously collected data to fill in missing values; Last Observation Carried Forward (LOCF) imputation, often used in longitudinal studies, where the last observed value is carried forward to fill in missing data. (Little et al. 2019). <br />The R statistical software provides various packages for multiple imputation, each using different algorithms. One widely used package is mice (Multiple Imputation by Chained Equations), which generates multiple completed datasets by imputing missing values according to a specified method, applies the substantive statistical model to each dataset, and combines the results using Rubin’s rules. (Van Buuren, et al. 2011, 1-67).<br />Within the mice framework, Random Forest (RF) is one of the available imputation methods. Random Forest Multiple Imputation is particularly effective for time series data under certain conditions. It excels at capturing and relationships between variables, datasets and to or in the data. the data is or Random Forest especially when are or the dataset is />In this study, the relationship between the and variables is and with the missing data mechanism and variables with one Therefore, the Random Forest method within the mice package was applied for imputing missing data, and the final time series dataset is in Graph <p style="text-align: justify;"><img alt="" /> <p style="text-align: justify;">The on the the data while the yellow line indicates the trend. the 35 years, key and economic have Georgia, including when the Union when the to the of the when the and 2012, when the following a in and FDI per capita have the same both The Gross Capital Formation per capita followed the of GDP per and the was to the and it for modeling. Government Expenditure as a of GDP, showed an upward from 1995 to but 2012. The trade as a of GDP, was negative and increased which it showed a positive a in values on The growth between and 1995 due to the of the Union, to and from <p style="text-align: <p style="text-align: and <br />The time series is crucial for econometric modeling. A time series is its mean and across time series to correlations, relationships between variables that not which in about data to over it to the results and the <br />There are two primary for and methods This paper a which is a approach. The the in a time the to the is to the time series a that the data is it is not the data is et al. <br />Common the which the that a is in the the which the by including to for and the which is similar to the but for and in the et al. This paper on the and to the of the data. <p style="text-align: <p style="text-align: each time series may not their that the series and long-term the time series data is found to it that the series and used for In such such as Error Correction or Vector Error Correction (VECM) methods are particularly in fields financial markets, and econometric time series analysis, for example, when examining the long-term relationships between variables such as GDP, and <br />There are three of This method the relationship between the two time series data using Ordinary Least Squares regression and the of the regression for using the the are the series are considered A method uses a more general approach the method and uses to the multiple time series data. The method is the this method is similar to the but uses a slightly different approach for the <br this study involves multiple variables, the was to the before applying the Vector Error Correction Model <p style="text-align: Error Correction Model <p style="text-align: Error Correction (VECM) are used to model time series data. They are of Vector and are to both short-term and long-term more VECM is particularly for multiple variables with for for from long-term and these over to the model, in the variables, with the short-term in the The the long-term relationship and indicates variables to a VECM of the time that and are into the />The general of the VECM <p style="text-align: justify;"><img alt="" /> <p style="text-align: a VECM, the is to the time series data are using the variables in the the appropriate of using model selection such as or - Bayesian the VECM is by both long-term and short-term with the for short-term and the <br />In conclusion, the VECM is a econometric for modeling and time series data and it is particularly valuable in financial and economic where long-term relationships are Therefore, this model has been applied in this <p style="text-align: <p style="text-align: Forest is a machine learning algorithm used for and regression It by multiple their results, and final is on a random of the data, and at each a random of is This to and the to applied to time series data, Random Forest other learning methods, but it to the of the data to future information from into the model and to data et al. <br to such as values of existing used to short-term or the time series strong or or For and long-term the data is another effective approach. the are the model on the data and on the data. for time series Error Error Error and et al. <br time series such as Random Forest not a relationship between and future It also missing data the for imputation and not the data to This Random Forest a for modeling time series data, as it relationships between multiple variables and provide the variables in this relationships Random Forest model was applied to and relationships the <p style="text-align: <p style="text-align: and it is to the of the time series data. The and were used for this specifically the and in the value of the the the that the data is In other the is the the of is The were conducted at a The results of the that and were at the while the other variables were However, all variables the In the revealed that none of the variables were at the but all were the The results from the were considered more as the for all variables at the were below were at the Therefore, the results were used as the primary that the time series data was at the but at the a and VECM analysis was />The study multiple variables, before VECM, the was and the and values are in <p style="text-align: justify;"><img alt="" /> <p style="text-align: justify;">The results indicate that for the is which the value of Therefore, the that no is For the of is the value of to the of the that at most one However, for the of is the value of the that at most two to the results that there are at most two a long-run relationship the a Vector Error Correction Model (VECM) provide a more of these />The appropriate of 2 were using model selection such as and and only variables were in the The results of VECM are in which only the significant variables in the <p style="text-align: justify;"><img alt="" /> <p style="text-align: />The Random Forest were to GDP using FDI, government gross capital and other economic variables were in the model, and model as well as the of and the most variables were were one and one with <p style="text-align: Forest model not To a of four variables were and GDP was on FDI, and The model a strong with an of and an of The importance results are in Graph <p style="text-align: justify;"><img alt="" /> <p style="text-align: Forest with <p style="text-align: the model, were to the data, and the trade and government expenditure variables were to GDP was on FDI, and their as well as the of GDP The model showed a strong with an of and an of The importance results are in Graph <p style="text-align: justify;"><img alt="" /> <p style="text-align: and <p style="text-align: justify;">The study examined the impact of Foreign Direct Investment (FDI) per capita on GDP per capita in Georgia over a period from 1989 to 2023. econometric and machine learning were key economic variables such as Gross Capital Formation per capita Government Expenditure as a of GDP Trade Balance as a of GDP and Population Growth Rate FDI per other variables were also considered, only the most were in the final />For the econometric analysis, the data was for followed by a A Vector Error Correction Model (VECM) was The results from the VECM that the of GDP per capita has a positive and statistically significant effect on GDP per and both the and of FDI per capita have a positive and significant impact on GDP per This that FDI per capita positively influences per capita economic growth in the long However, the for other variables, such as and were found to that the VECM may not have explore these findings, a machine learning approach was specifically using the Random Forest model, both with and The some to the of of the model with an and a Error between and The Random Forest model that the largest impact on GDP per capita from per capita followed by per capita net FDI government expenditure (% of and the trade (% of This that FDI has a significant positive impact on economic growth, but its effect is to that of gross capital />In the model, government expenditure and trade were due to The FDI, and their with the of The results showed that gross capital formation had the largest impact on GDP per capita followed by GDP FDI and FDI As of other studies, the results of this paper also indicate that FDI has both a short-term and long-term positive effect on GDP per on these findings, the government prioritize an to attract more foreign direct investment The to FDI in Georgia />• capital and />• on financial />• and economic />• A small that is to />• A of education and a />• over the and of the and of />• particularly in <br address these and encourage FDI the government on these which economic growth in Georgia.<br econometric (VECM) and Machine learning (Random Forest) approaches, the more and results for this analysis, but the of econometric and machine learning techniques, there is potential for research using econometric such as Error Correction and Granger In the of machine future studies explore techniques including and of using variables and econometric and machine learning when used provide more valuable to the analysis. the most results are by applying both approaches, on <p style="text-align: justify;"><br <br />• ჭარაია <br />• Direct and Georgian in />• & Foreign Direct and from EU of />• & the of in and />• of and on Growth of />• Foreign Direct and Portfolio Growth in Central and Eastern and />• & Analysis of the between Foreign Direct Investment and />• The between Foreign Direct Investment and A of />• & Regression with Missing a of based on Random of and />• & of Foreign Direct Investment and Growth in Central and Eastern European />• & the of of />• & Analysis with Missing Data & />• Error Correction to Multiple />• Foreign Direct Investment and Growth in the European of />• />• Direct Investment on Growth by in Central and Eastern European />• & Foreign Capital and from />• The between Growth and Foreign Direct Investment the in the European />• of with of <br />• Foreign Direct the Growth of Southeastern European Europe of />• & in Using Random />• & Imputation by Chained in of