You are on page 1of 19

Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Forecasting Foreign Direct Investment to Sub-Sa-


haran Africa using Arima Model: A Comparative
Analysis of Machine Learning Algorithms
Mukhtar Abubakar Yusuf Ph.D.
http://orcid.org/0000-0002-5521-2495

Abstract:- This study examines the factors that influence Data analytics (DA) examines data sets to determine the
individual foreign direct investment decisions and predicts information they contain, increasingly with the support of spe-
using various Artificial Intelligence (A.I.) Algorithms mod- cialized systems and software.
els. The study also gives us an in-depth insight into the dy-
namics of complex FDI decisions using those A.I. predic- ARIMA models could generate projections for varieties
tive models. We use structural equation modeling in the of time series data. It is worth noting that the ARIMA model
prescriptive strand. Return on Investment (ROI), Secu- has three parts. However, not all aspects are always required,
rity/Personal Safety, and Investment Facilitation Services but it depends on the type of time-series dataset available. The
significantly affect individual FDI decisions. On predictive three parts are the autoregressive (A.R.), the integrated (I), and
strand analysis, we used various Machine Learning models finally, the moving average (M.A.). The assumption for the
to evaluate the accuracy of predicting classes of individual A.R. part of a time series dataset is that the observed value
FDI risk decisions and the ARIMA model for prediction. varies on some linear combinations of prior observed values
We find that Random Forest and Ada Boosting Trees have up to some upper limit lags plus an error term. The assumption
substantial classification accuracies despite the "No free for the M.A. part of time series data is that the practical esti-
lunch" theorem. The result also indicates that a better pre- mate is a random error term plus some linear permutations of
diction could be made by applying multiple classes of FDI previous random error terms up to some maximum lags
inflow decisions rather than binary classes. (Stanley Jere, 2017).

Keywords:- Foreign Direct Investment, Artificial Intelligence, The forecasts by the gradient boosting model and, of
Investment Facilitation, Return-On-Investment, Investment course, the random forest model is more precise than the target
Decisions, Predictive Modeling, Random Forest, Gradient forecasts. In comparison between the gradient boosting and
Boosting random forest models, the gradient boosting model turns out
to be more specific (Yoon, 2020)
I. INTRODUCTION
Generally, forecasting outcomes play a fundamental role
Foreign direct investment (FDI) is an investment in for policymakers, economic decision-makers, and other key
long‑term affiliation with a host country. It reflects a lasting stakeholders with the view of coming up with reasonable pol-
interest being controlled by a resident entity in one economy icies and suitable strategic plans but all those depend primarily
by a direct foreign stakeholder or a parent enterprise in an en- on the accuracy of the forecasts.
terprise resident or an economy different from that of the in-
vestor (United Nations, 2007). It should, however, be noted The main objective of this study is to conduct empirical
that foreign direct investment is different from indirect (port- research on the most effective A.I. application and propose ap-
folio) investment. FDI involves establishing a substantial, propriate recommendations to promote and facilitate invest-
long-term interest in the economy of a foreign country. ment in African Sub-Region based on predictive inference.

This is in tune with the Sustainable Development Goals


Service quality is critically vital in providing companies to promote sustained, inclusive, and sustainable economic
with a competitive edge, as it influences various factors such growth, full and productive employment, and decent work for
as customer satisfaction (Osarenkhoe & Byarugaba, 2016) all, especially in Sub-Saharan African regions (Nations, 2022).
Given the foregoing, it is deemed critical that each African
It is well recognized in the economic literature that FDI country has to do much self-diagnosis in the perspective of
plays a vital role in the financial growth process in host coun- strategic thinking on how best to entice FDI and link it with
tries, and since FDI is considered a vehicle to shift new ideas, the local economy so as to develop linkages that would con-
capital, improved technology, and new skills from advanced tribute to maximizing returns at the local level (David Brady,
countries to emerging countries. Specifically, Return on In- 2010).
vestment (ROI) is broadly used for analyzing the performance
of investments in a business or investments for an individual This paper contributes to the literature on FDI and eco-
(Brauer, 2016). nomic growth and the importance of integrating investment

IJISRT22DEC444 www.ijisrt.com 164


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
promotion with investment facilitation to lure potential inves- Artificial intelligence (ML) models would be used to pre-
tors. We slightly deviate from previous studies by comparing dict the future inflow of FDI based on investment decisions by
the predictive patterns of two classes of investment against foreign investors using the FDI data obtained from an online
multiple classes in the contexts of FDI inflows decisions using survey (Qualtrics software). This study will apply FDI inflows
ML models. measured from 2006 to 2018. In this study, because of the lim-
ited sample size, no shrinkage methods algorithms have been
Our main focus as predictive analysts is to develop a de- used for variable selection to select models for predicting FDI
scriptive and then predictive research report that will help crit- decisions.
ical stakeholders better understand the predictive propensity of
the combined factors for investment policy to make informed II. RELATED STUDIES
decisions.
A. An overview of the FDI
Preliminary findings indicate that certain individual de- Foreign Direct Investment, called FDI, is an investment
cision factors of the FDI significantly impact individual FDI in a business by an investor from another country for which
decisions. Also, they are expected to be essential factors in the foreign investor controls the majority of company pur-
predicting the class of investment relative to other select fac- chases. In addition, the International Monetary Fund's Balance
tors. To gain a deeper insight into various Machine Learning of Payments Manual defines Foreign Direct Investment as an
models in the context of investor multi-layered investment de- investment that is made to obtain a lasting interest in an enter-
cision factors for FDI, we ask these questions: prise operating in an economy other than that of the share-
holder. The investor's purpose is to actively influence the en-
(1) To what extent do FDI decision factors contribute to deter- terprise's management (Moosa, 2002).
mining the ideal choice of the best A. I model for prediction,
and (2) How and to what extent does the class size of the target B. Factors influencing the FDI decisions:
FDI decision variable determine its predictability?
 Investment incentives perception.
The research questions are shaped by the purpose of the Investment incentives also include procedures designed
study and form the methods and the design of the analysis. The to influence an FDI project's size, locality, or industry by af-
research questions have enabled us to identify and produce in- fecting its relative cost. It also alters the risks attached to it
teresting hypotheses and ML models that infer and predict in- through inducements that are not available to comparable do-
dividual decisions for key stakeholders. Given its generaliza- mestic investors (OECD, 2019). Investment incentives are the
bility, R&D Units in the Investment Promotion Agencies reward for and an encouraging force behind the investment,
(IPAs) can easily replicate the method and apply new data to the object of which is usually to maximize return (McGraw,
forecast FDI inflows based on investment decisions based on 2016).
select factors.
Investment incentives are designed to minimize a firm's
However, policymakers often find modern-day scholar- tax liability. They include tax allowances such as a reduced
ship less than helpful when it employs such methods across corporate income tax rate, tax holidays; accelerated devalua-
the board for their benefit and without a clear sense of how tion allowances on capital taxes; exemptions from import du-
such a study would contribute to investment policymaking ties, and duty drawbacks on exports. However, businesses do
(Paul C. Avey, 2014). This makes us suggest strategic collab- not always succeed.
oration with academia or consultants.
 Ease-of-doing-business perception.
The proposed use of combined descriptive and prescrip- The ease of doing business index, known as EDB, ranks
tive models will be used to predict FDI inflows for years to the controlling (institutional) environment in countries around
come in the face of current global economic challenges and the world in terms of tasks for running a business (Simmons,
lingering insecurity in the African Sub-Region. The forecasts 2019). EDB includes starting a business, getting credit, getting
from this study will help governments in Africa enhance FDI electricity, paying taxes, and trading across borders, among
inflows that would revamp economic growth. The central aim other things (Nau, 2018). Corcoran and Gillanders (2014) in-
is to identify key contributing factors of Foreign Direct Invest- dicate that EDB is an inherent proxy for trade costs. A high
ment decisions in the African Sub-Region, find out their effec- level of Ease of Doing Business means the regulatory environ-
tiveness, assess the efficacy of various predictive classification ment is more conducive to starting and operating a business.
models, and forecast FDI inflow to African Sub-Region based In an empirical study on the effects of EDB, Kelley found that
on investment decisions. The findings of this study would go the EDB ranking generally shapes positive investors' percep-
a long way in formulating economic and investment policies tions of opportunities (Doshi, Kelley, & Simmons, 2019).
with a view to strategic decisions to attract beneficial FDI to However, they find no evidence that perceived EDB affects
African Sub-Region. Figure 1 depicts the hypothesized model. the FDI that a country gets (Corcoran & Gillanders, 2014).

IJISRT22DEC444 www.ijisrt.com 165


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
 Foreign exchange rate manipulation1. technical and emotional aspects (Geetika, Ghosh, &
Exchange rates are the value of one exchange relative to Chowdhury, 2015). Therefore, to a prospective investor, as in
another. In theory, it reveals the relative suitability of a cur- the medical field (Kostelnick, 2014), safety is a high priority
rency pair, with the simple economic principle of expecting in FDI decisions.
money to flow towards a stronger currency. It is simply the
proportion of exchange for two currencies (Loftus, Leo, Brown and Hibbert (2017) analyzed the impact of violent
Daniliuc, & Boys, 2020). It is the amount of exchange that one crime on foreign direct investment (FDI) inflows in a sample
needs to pay to buy one unit of a different currency of 62 countries from 1997 to 2012. Their findings indicate that
(Zakamulin, 2017). violent crime acts as a deterrent to FDI because criminal ac-
tivities have the potential to increase the cost of doing busi-
A previous study indicated that the regression effects of ness. They also reduce the demand for goods and services
exchange rate depreciation (EXC) have a considerable posi- (Brown & Hibbert, 2017).
tive impact on FDI inflows into Nigeria (Nurudeen, Wafure,
& Auta, 2011). However, the coefficients of error correction  Investment promotion perception.
of FDI flow and foreign exchange rate are substantially nega- Narrowly defined, investment promotion consists of ad-
tive, while that of natural resource outflow and GDP are sig- vertising, direct mailing, investment seminars, and investment
nificantly positive. Foreign exchange appreciates with FDI in- missions (Kitchin & Thrift, 2009). An investment promotion
flow and resource outflow (Dinda, 2008). agency (IPA) is often a government agency whose mission is
to attract investment to a country, State, region, or city. Gen-
 Return on investment perception (ROI). erally, IPA services have four core functions: image building
ROI is the total anticipated discounted stream of net of FDI hosting country, investment generation, project man-
profits divided by the costs of the investment (PMI, 2017). Re- agement, and aftercare services (wikipedia, 2018). While IPAs
turn on investment (ROI) is broadly used for evaluating the play an essential role in enticing investment to developed
execution of investments in a company or investments for an countries, some IPAs have additional advocacy functions.
individual (Brauer, 2016). A different study by Wensveen
(2016) discovers that return on investment (ROI) is a vital Essential investment promotion services include attract-
measure of an industry's and individual firms' capabilities to ing foreign direct investment (FDI) or enhancing domestic in-
attract capital for sustained growth and replacement of existing vestment, which requires a wide range of efforts. (World
assets. In addition, an empirical study by Alavinasab (2013) Bank, 2010).
used an econometric simulation to reveal significant positive
effects of return on investment on FDI. He also contends that Morisset argues that the perception of investors of the ef-
the viability of an investment is one of the significant contrib- ficacy of IPAs is highly reliant on the quality of the investment
uting factors to FDI (Alavinasab, 2013). climate and the level of development of the country (Morisset
& Andrews-Johnson, 2004). He concludes that when the in-
 Corruption perception. vestment climate is poor, excess resources have to be spent on
Certain countries remain unable to attract FDI despite convincing potential investors. However, a study by Harding
high levels of corruption (Kolstad & Villanger, 2008). Ufere and Javorcik (2011) suggests that to some investors, invest-
Perelli, Boland, and Carlsson (2012) indicate that local execu- ment promotion does in developing countries but not in tech-
tives are active perpetrators of bribery, which is another form nologically advanced economies.
of corruption rather than casualties of the bribe-demanding
government. Greppin, Carlsson, Wolfberg, and Ufere (2017)  Investment facilitation perception.
also found that when executives give way to extortion, they Karl (2018) and Sauvant (2015) define investment facil-
commonly do so because they feel they have no choice. itation not as a matter of promotion-as-usual but as a means of
Additionally, it has been established that only by paying bribes finding and diffusion of new approaches and applications, a
would they be able to get resources out of institutions, get new process that needs nurturing and support (Sauvant & Hamdani,
business, avoid retaliation, enable their company to flourish, 2015). Investment facilitation usually takes place through an
or prevent their company from being shut down (Greppin et investment promotion agency (IPA). Typically through a gov-
al., 2017). ernment agency whose mission is to attract investment to a
country, State, region, or city (Korinek & Sourdin, 2011)
 Security/personal safety perception. (OECD, 2011).
Security is the State of being or at least feeling safe, and,
in particular, the safety of a country or an organization against C. Modeling FDI Prediction using Machine Learning (ML)
illicit activities such as extremism or espionage, and other po- models
tential dangers (Badie, Berg-Schlosser, & Morlino, 2011). In Not enough literature is known about how machine learn-
a broad sense, security encompasses the protection of physical ing could be used to predict specific individual Foreign Direct
and digital assets, which may be physical, personal or build- Investment (FDI) decisions in the Sub-Saharan African region.
ing, or informational (Nemati, 2017). Safety factors are influ-
enced by perceived reality and are a subtle combination of

1
Exchange rate changes due to manipulation increase the risk and
uncertainty for investors

IJISRT22DEC444 www.ijisrt.com 166


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
In reality, it is hard to predict future values of FDI be- coverage of the predictions, and the averaging was also con-
cause linear models sometimes can't capture the complex de- firmed to be the best model under the forecast error distribu-
signs in the data. It is anticipated that estimation of FDI tion (Netshivhazwaulu, 2018). Len (2020) concludes that the
through the deep learning process would not only be able to prediction indexes selected after quantification based on the
capture such unpredictability more efficiently. Efficiently but random forest could improve the prediction accuracy (Lei
also be able to forecast future inflows of FDI more accurately Wen, 2020).
than traditional forecasting procedures (Roy, 2020).
III. METHODS
By using the dataset from the period 1970-2009, Pradhan
(2010) finds that Artificial Neural Network (ANN) is an effec- The explanatory models would be built with the use of
tive tool for forecasting FDI. This supports the evidence that it SPSS-AMOS regression analytical applications. The focus
is possible to extract information hidden in FDI and make fu- would be on direct effect regression of the predictors on the
ture forecasts regarding FDI inflows (Pradhan, 2020). target variable. The second part would be identifying the most
effective models for predicting FDI decisions and their appli-
A study by Stanley (2017) found that in comparison of cation to making a forecast.
the three methods used shows that the ARIMA (1, 1,5) is the
best fit model to predict FDI because it has the minimum error.  Description of the design.
He concludes that decision-making on coming up with reason- This is a two-level exploratory sequential method re-
able policies and suitable strategic plans depends on the accu- search design where we begin by exploring the quantitative
racy of forecasts (Stanley Jere, 2017). data and analyzing it, then build models to be validated
(Creswell & Creswell, 2018). The next level of the study
Another study that combined ARIMA and ANN models builds on the results of the initial findings to make predictive
in linear and non-linear modeling indicates that the combined inferences (Creswell & Creswell, 2018: 224). The first part of
model could effectively improve forecasting accuracy the study strand is motivated by the desire to examine the sig-
achieved by either of the models used separately nificance of individual investors' FDI perception factors that
(G.PeterZhang, 2003). shape investors' risk preferences. The second strand is inspired
by the need to have a better understanding of whether the pre-
An article by Javier Arroyo (2009) proposed adapting the dictive models are good enough to predict decisions and the
k-NN method to forecasting HTS and yielded promising re- third strand is motivated by the desire to make predictions of
sults. The extension of the proposed method allows the direct FDI decisions over the next ten years.
inclusion of lagged values of explanatory time series. It could
be beneficial in some applications (Javier Arroyo, 2009).  Source data for Descriptive Analysis
In this study, we used preexisting data obtained from var-
A study by Akbar (2014) finds that N.N. approaches bet- ious FDI-relevant stakeholders both in Nigeria and in the dias-
ter explain FDI determinants' weights than traditional regres- pora using Qualtrics that had some forms of investment in Ni-
sion methodologies. However, their preliminary findings offer geria from the year 2006 to 2018 (Yusuf, 2020).
essential and novel implications for future research in this area
(Akbar & Yusaf, 2014). The sampling and data collection, participants, proce-
dure, constructs, and measures are explained in the previous
The accuracy of any forecast is measured by mean abso- studies (Yusuf, 2020).
lute percentage error and root squared mean error. Another
empirical study indicates that the gradient boosting and ran- A. Hypotheses for Structural Equation Model
dom forest models' predictions are more accurate than the
benchmark forecasts. Between the gradient boosting and ran-  Individual FDI Perception Factors
dom forest models, the gradient boosting model turns out to be The investment climate is routinely viewed to be the pri-
more precise (Yoon, 2020). mary (and sometimes the only) FDI determinant. However,
FDI allocation lies beyond investment climate alone since in-
A different study by Wu (2021) conducted three experi- vestors from different countries estimate investment climates
ments by testing the predictive ability of the decision tree al- differently (Panibratov, 2017). Ease of doing business is a ma-
gorithm, testing the decision tree algorithm with performance jor determining factor that attracts FDI into a country. That is
improvements, and determining the best decision tree forecast especially true for those potential investors that rely heavily on
rate comparison. And using the logistic regression model indi- the World Bank's ease of doing business country ratings. The
cates that the random forest has the highest and best prediction study by Hossain, Hassan, Shafiq, and Basit (2018) found that
rate in contrast with the logistic regression model (Wang, Jen- ease of doing business indicators significantly and positively
Hsiang, & Pei, 2021). impact Inward FDI. The study posits that the ease of doing
business enables inward FDI through better contract enforce-
However, Nyawedzeni (2018) used the shrinkage selec- ment, getting credit, and registering property. I assert that:
tion methods of elastic net, least absolute shrinkage, and selec- Hypothesis 1. Ease of Doing Business Perception positively
tion operator (Lasso) for the choice of best variables. He dis- affects Foreign Direct Investment (FDI) inflow.
covered that linear quantile regression averaging was the best
model to predict foreign direct investment since it had 100%

IJISRT22DEC444 www.ijisrt.com 167


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
The attitude towards inward foreign direct investment get materials out of customs, get new business, avoid retalia-
(FDI) has changed considerably over the last few years. Many tion, enable their company to thrive, or prevent their company
host countries have liberalized their economic policies to at- from being shut down. I posit:
tract foreign multinational corporations (MNCs) investment. Hypothesis 4. Corruption Perception has a negative effect on
Nigeria is no exception. These include fiscal incentives such Foreign Direct Investment (FDI) inflows.
as tax holidays and lower taxes for foreign investors; financial
incentives such as grants and preferential loans to MNCs. And Over the past decade or more, Sub-Saharan Africa has
measures such as market preferences, infrastructure, and witnessed unprecedented security challenges occasioned by
sometimes even monopoly rights (Blomström, Kokko, & the activities of militants in the south-south region, kidnappers
Mucchielli, 2003). A study by Roberts (1993) on determinants in the southeast, violent armed robbery in all parts of the coun-
of FDI incentive preferences of MNEs finds that government try, political assassination, ritual killings, and more recent ac-
programs waiving import duties were deemed most important tivities of Boko Haram in some parts of the northern Nigeria
by their targeted investors (Rolfe, Ricks, Pointer, & McCarthy, region, especially northeast. When put together, these social
1993). I thus posit that: menaces impinge on the security of lives and property of Ni-
Hypothesis 2. Investment (Business) Incentives Perception gerian citizens and foreigners living or even trying to invest in
positively affects Foreign Direct Investment (FDI) inflow. the country (Udeh & Ihezie, 2013). These menaces trigger a
problematic sense of insecurity that challenges Nigeria's ef-
Return on investment (ROI) is an essential measure of an forts toward national economic development and, conse-
industry's and individual firms' ability to attract capital for quently, its Vision 20:2020. It also reduces the attractiveness
continued growth and replacement of existing assets of foreign investment and its contributions to economic
(Wensveen, 2016). No investor would like to venture into a growth in Nigeria (Udeh & Ihezie, 2013).
business that has a low ROI. However, some investors from a
particular region of the globe do not give it a priority; instead, This indicates a profound negative impact of insecurity
they look at other factors like security and ease of doing busi- on FDI inflows to the country. Thus, I posit that:
ness. High ROI can make it much easier for host nations to Hypothesis 5. Security/Personal Safety Perception has a neg-
experience an influx of investors, and they are expected to ative effect on Foreign Direct Investment.
sharpen their skills in handling business registration and other
processes. An empirical study by Alavinasab (2013) used an Despite the ambiguous evidence on the benefits of FDI,
econometric model to detect significant positive effects of re- investment promotion has become an active area of policy, and
turn on investment on FDI. He argues that the profitability of a growing number of nations are offering services and incen-
an investment is one of the essential determinants of FDI tives to attract investment from multinational firms. Andrew
(Alavinasab, 2013); thus, I posit that: (2015) finds that the positive effect of investment promotion
Hypothesis 3. Return on Investment Perception has a positive on FDI inflows is robust across various empirical specifica-
effect on Foreign Direct Investment (FDI) inflow. tions (Charlton & Davis, 2017). In support, a previous study
by Harding and Javorcik (2011) finds that investment promo-
A study on the effect of corruption on foreign direct in- tion leads to higher FDI flows to countries in which red tape
vestment inflows in Sub-Saharan Africa by Omodero (2019) and information asymmetries are likely to be severe. The study
finds that Nigeria's corruption ranking position has an insig- suggests that investment promotion works in developing coun-
nificant positive impact on FDI. The implication is that Ni- tries but not in industrialized economies. Essential investment
geria's poor institutional and legal framework qualities are promotion services include attracting foreign direct invest-
helping corruption to thrive in all areas of Nigeria's economy ment (FDI) or enhancing domestic investment; (World Bank,
(Akinlabi & Hamed, 2011). A similar study on corruption, for- 2010). The apex investment promotion agency, NIPC in Nige-
eign direct investment, and economic growth in Nigeria by ria, contributes to minimizing capital flight by providing ex-
Akinlabi and Hamed (2011) found that there is a long-run re- isting investors with aftercare services at no cost. By so doing,
lationship between FDI inflow and a low level of corruption. it helps reduce the problems of low Ease of Doing business for
It is suggested that for Nigeria to attract a large volume of FDI investors; Thus, I posit that:
inflow, corruption at all levels of governance must be drasti- Hypothesis 6. Investment Promotion Service Perception has a
cally reduced and checkmated. However, some countries con- positive effect on Foreign Direct Investment.
tinue to attract FDI despite high levels of corruption (Kolstad
& Villanger, 2008). Cuervo-Cazurra (2006) argues that cor- Foreign Exchange refers to uncertainty and risk associ-
ruption results in a reduction in FDI and a change in the com- ated with the manipulation of the exchange rate. Foreign ex-
position of the country of origin. change rate uncertainty affects managers' and risk-neutral
multinational firms' foreign direct investment decisions
From a different perspective, a study by Ufere et al. (MNCs). An empirical study by Sung and Lapan (2000) on
(2012) found that local managers are active perpetrators of Strategic foreign direct investment and exchange rate uncer-
bribery rather than victims of the bribe-demanding govern- tainty finds that the firm can increase expected profits with
ment. Greppin et al. (2017) found that when executives decide sufficient exchange-rate volatility.
to succumb to extortion, they generally do so because they feel
they have no choice: only by paying bribes will they be able to A previous study indicated that exchange rate deprecia-
tion (EXC) significantly impacts FDI inflows into Nigeria

IJISRT22DEC444 www.ijisrt.com 168


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
(Nurudeen et al., 2011). However, the coefficients of error cor- area of investment facilitation to attract foreign direct invest-
rection of FDI flow and foreign exchange rate are significantly ment, including regulatory transparency, streamlining admin-
negative, whereas that of natural resource outflow and GDP istrative processes, and dispute prevention (International
are quite positive. FDI flow and resource outflow directly in- Centre for Trade and Sustainable Development, 2020).
fluence the foreign exchange rate.2 Foreign exchange appreci-
ation with FDI inflow and resource outflow (Dinda, 2008). A study that investigates private investment facilitation
Stevens (1998) found statistically significant evidence of the strategy finds that the primary mechanism to market or "sell"
implied negative relationship between exchange rates and FDI the State as a prime location for private sector investment
inflows (Stevens, 1998); thus posits that would significantly impact FDI inflow if not because of its in-
Hypothesis 7. Foreign Exchange Rates Uncertainty3 Percep- visibility. The State does not typically appear on the long list
tion has a negative effect on Foreign Direct Investment (FDI) of location alternatives for active consideration by investors
inflow (Effiom & Etim Edet, 2019). We, however, know that at the
federal level, when the One-Stop Investment Center (OSIC) at
Although a clearer picture of what investment facilitation Nigerian Investment Promotion Commission was established
means has yet to be developed, facilitating FDI flows is essen- in 2006, it recorded an unprecedented influx of FDI inflows
tial to mobilizing resources for development. Governments are (Nigerian Investment Promotion Commission, 2017); thus, I
increasingly concerned with investment facilitation (Hees & posit that:
Mendonça Cavalcante, 2017). Many countries have recog- Hypothesis 8. Investment Facilitation Perception has a posi-
nized the importance of focusing on domestic reforms in the tive effect on Foreign Direct Investment (FDI) inflow.
The SEM hypothesis is depicted in Figure 1

Fig 1: SEM Model Hypothesis

B. Analysis for Structural Equation Modelling (Descriptive) & Anderson, 2010). Bartlett's Test of Sphericity was signifi-
cant (Chi-square = 4924.914with df = 171, p.<0.001), imply-
 Measurement model. ing significant intercorrelations. The five components had a
We conducted the case and variable screening (Yusuf, total variance of 71.245% (Hair et al., 2010). The total vari-
2020) and then analyzed the factor loadings and cross-loadings ance explained can be obtained in Table 1. For the final pattern
of the items. As a rule, items were maintained if (a) they had matrix that integrates Cronbach's alpha and the percentage of
high loadings on their primary factor. Typically, l >.30. And if variance explained, refer to Table 1
(b) they had minimal cross-loadings on some other factor (i.e.,
cross-loadings were less than half of their primary loadings We find that the factor Cronbach's alphas are robust;
(Hinkin, 1998). Kaiser-Meyer-Olkin (KMO) statistic was.659 Ease-of-Doing-Business is 0.959, Investment Promotion is
(Kaiser Meyer, 1970). The commonalities are acceptable; they 0.895, Social Desirability is 0.957, security is 0.841, and For-
are all above the minimum limit of 0.300 (Hair, Black, Babin, eign Exchange Rates is 0.804, respectively. The Correlation

2
This implies that the quantum of FDI inflow to a host country cou- of escalating the rates become higher which affect the investor’ de-
pled with finished manufactured products contribute to more influx cision to continue with investment or expansion.
of foreign currency into the host country which helps stabilize the
3
exchange rates, and on the other hand, if the rates are allowed to be FOREX rate is a high risk factor associated with investment uncer-
freely manipulated by the government of parallel market, then risk tainty, where sufficient exchange rate volatility usually increases ex-
pected profit.

IJISRT22DEC444 www.ijisrt.com 169


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Matrix Table was inspected in Table 2, with none of the
amounts surpassing the threshold of 0.7 (Hair et al., 2010).

TABLE 1: Pattern Matrix Table with Cronbach's Alpha Table 3: CFA Model Validity Analysis

Significance Indicators: p < 0.100, * p < 0.050, ** p <


0.010, *** p < 0.001 Validity Concerns
⁂Correlation is not specified in the model. No validity con-
cerns here

The chi-square difference test results indicate the uncon-


strained model, chi-square=576 DF=278, and another set of
chi-squares from a completely constrained model (zero and
equal; chi-square=600.5, DF=297). The p-value is significant;
0.040 significant; 0.040 significant; 0.040 significant; 0.040.
The results suggest that we are 90% confident that Common
Method Bias (CMB) exists. We now controlled the method
bias by applying common method variance (CMV) or Com-
mon Latent Factor (CLT) statistical techniques.
TABLE 2: Factor Correlation Matrix
 Structural Analysis.
We imputed the constructs centered on the final CFA
model, which comprised method bias adjustment by integrat-
ing the unmarked variables of the CLF factor score in the im-
puted variables. The structural model combines computed var-
iables (latent factors) following the CFA and calculated mean
variables from the formative constructs. The analyzed research
models for the study are depicted in Figure 1.
Significance Indicators: p < 0.100, * p < 0.050, ** p < 0.010,
*** p < 0.001
 Multivariate assumptions.
On linearity, the results imply some robust linear rela-
We applied the pattern matrix obtained from the EFA to
tionships between the variables with the highest "Fs." Still, the
build a CFA model structure and then examined it with AMOS
connections are significant; that is, p=.001, and the F value is
software v25.0. The sample size of 250 was adequate, and the
the highest in the equation (10.860). For Ease of Doing Busi-
commonalities were sufficient. For model fit, the chi-square
ness to Investment Promotion and p=0.000, Return on Invest-
CMIN value was 336.976 with 139.00 degrees of freedom, and
ment and Investment Promotion, they are sufficiently linear to
CMIN/DF is 2.424, which is between 1 and 3 and, therefore,
be tested in covariance-based SEM algorithms. We did not ob-
considered good. CFI=0.96, SRMR=0.060, RMSEA is 0.076,
serve VIFs greater than 3.1.
and the p-value was significant .000, indicating we can reject
the null hypothesis as the model is statistically significant
Cronbach's alphas (Tavakol & Dennick, 2011) suggest
(Hair et al., 2010).
decent reliabilities of the reflective variables, as shown in the
EFA earlier; refer to Table 5. We assessed the model fit and
Composite reliability (C.R.) results for all the variables
observed that the chi-square CMIN value was 21.6629, with
exceed the threshold of 0.7 (Hair et al., 2010), with ease of
11.00 degrees of freedom. CMIN/DF is 1.9694, considered ac-
doing business at 0.965. I retained all. Convergent validity
ceptable. CFI=0.9728 and, SRMR=0.0615, RMSEA is 0.0615.
(AVE) values were >0.5 (MacKenzie, Podsakoff, &
Pclose=0.274 and the p-value is 0.0274, which is significant
Podsakoff, 2011), with social desirability highest at 0.892. The
(Hair et al., 2010) as shown in Table 4
constructs have excellent validity; all convergent validity
(AVE) values are more significant than 0.5 (Mackenzie et al.,
2011). Composite reliability (C.R.) scores for all the variables
surpass the threshold of 0.7 (Hair et al., 2010), as shown in
Table 3.

IJISRT22DEC444 www.ijisrt.com 170


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Table 4: Model Fit

We now created and analyzed the algebraic table contain-


ing the inter-correlations among the various study variables.
The outcome suggests the statistical significance of the corre-  Logistic Regression (glm) Model
lations at p<0.01 and p<0.05 in most independent variables ex- Logistic regression is a model for binary classification
cept between Corruption and Investment Incentives; this is predictive modeling. The parameters of a logistic regression
shown in Appendix G. model can be estimated by the probabilistic framework called
maximum likelihood estimation (Brownlee, 2019). Logistic
We have generated the underlying model using the Sta- regression yields as good performance as ML models to make
tistical Package for the Social Sciences (SPSS) and Analysis predictions, but it has been established that logistic regression,
of Moment (AMOS) version 26 analytical tool. In the process, gradient boosting machines, and neural networks were system-
we uploaded eight exogenous FDI perception factors, control atically ranked among the best models (Simon Nusinovicia,
factors, and FDI endogenous factors from the variable data set. 2020).
We then link the exogenous elements to the endogenous fac-
tors of FDI inflow and compute the estimates. We evaluated  Naïve Bayes Model
the direct effect by calculating the estimates and then examin- Naïve Bayes is a simple learning algorithm that applies
ing the results to ascertain the model fit, regression weights, Bayes' rules together with a strong assumption that the attrib-
beta-estimations, p-values, and R-square. Years of Investment, utes are conditionally independent given the class. Naïve
Education, and Gender control FDI inflow decisions. Bayes nonetheless often delivers competitive classification ac-
curacy. Naive-Bayes is a generative model in which we model
C. Machine Learning Models for Prediction the conditional probability of input x given the label (Fred J.
Damerau, 2010). We performed a couple of visualizations to
 Source of dataset take a better look at each variable before normalization, and
The source of this FDI dataset is described under the mul- this stage is essential to understanding the significance of each
tilevel list number 4.2, and specifically, that of the target vari- predictor variable, as depicted in Figure 2
able is depicted in Table 5
Fig 2.: Visualized Data of a few variables
Table 5: Structure of Classified Data

To predict using the Naïve Bayes algorithms, we ran the


Naive-Bayes model and predicted the default status on the test
set.

 Decision Tree Model

IJISRT22DEC444 www.ijisrt.com 171


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
A decision tree is a type of Algorithm in machine learn- creating new FDI inflow bootstrap samples, fitting, and adding
ing that employs decisions as the elements to represent the re- trees to the sample is repeated k times until no further devel-
sult in the form of a tree-like formation. It is a graph that uses opment is seen in the ensemble's performance on a validation
a branching method to illustrate every possible output for a dataset. This results in better performance than a single well-
specific input (Contributor, 2012). configured decision tree algorithm.

To predict using "rpart" Decision Tree algorithms, we ran  Random Forest:


the Decision Tree model and forecast the default status on the To have an attribute with the highest information gain
test set using the ANOVA method. I plotted the tree with reg- and split based on decision attributes, a high-performant en-
ular observation by setting the control part "min split" to 60 semble model with a sufficiently diverse group of individual
and predicting the FDI inflow classified amounts. To improve base learners, we applied random forest with a penalty when
the model performance, We tuned the parameters to check if the trees start to pick a particular attribute at a given level more
we could improve the model over the default value by control- than certain times. Random Forest models decide where to
ling the min split to 4, "minbuckey" to round5/3, and max split based on a random selection of features, unlike boot-
depth to 3. We anticipate an accuracy higher than 0.548. strapped, whose data broadly breaks off at the same features
throughout each model. We performed a random forest tuning
 K-Means (KNN) Model with the use of OOB so that R.F. could be fit in one sequence,
Machine learning techniques have been widely used in with cross-validation so that once the OOB error stabilizes, the
many scientific fields, but their use in the medical literature is training can be terminated. We now created the training and
limited partly because of technical difficulties. K-nearest validation datasets, then extracted an OOB & validation errors,
neighbors (KNN) is a simple method of machine learning and compared the error rates by changing the number of trees
(Zhang, 2016) and with different vi as indicated in Figure 3.

K-Nearest Neighbor or K-NN is a Supervised Non-linear Fig 3: Error Rates by Trees number
classification algorithm. K-NN is a non-parametric algorithm; Result:
i.e. Algorithm; i.e., it doesn't assume underlying data or its dis-
tribution. It is one of the simplest and most widely used algo-
rithms, which depends on its k-value(Neighbors’-value)

We scaled the dataset and clustered the FDI inflow based


on eight decision factors. The graph indicates the proximity of
distance between related decision factors. Thereafter, we ran
the k-means Algorithm to cluster the factors with the choice of
the initial value of k = 4 and the number of restarts at 25.

A "train control()" function is used, and the "tune length"


is set at 20 to select the optimal model and plot the outcome of
repeated Cross-Validation with the application of RMSE. We More trees are generally better in performing rf so long
scaled the data frame to obtain a z-score using the total Within as we can support training the Algorithm. The classification
the Sum of Square "wss" to determine the best number of clus- error is lowest with Out-of-Bag Error.
ters k. We now fitted the KNN model by training the FDI da-
taset using the function "knn" and generated the confusion ma- The hyperparameters of the random forest model are
trix and ROC-AUC classifier. Before that, we evaluated the searched for using the Out-Of-Bag (OOB) errors. Recall that
model to choose K with the best classification accuracy. We the number of attributes selected randomly at each split node
used K=1, 3. 5, and 15. indicates that the optimal value of the m parameter "mtry" is
four, as depicted in Figure 4
 Random Forest, Bagging, and Boosting Trees
A decision tree is a non-parametric supervised learning Fig 4: Out-Of-Bag (OOB) errors
algorithm for classification and regression problems. It is also
often used for pattern analysis in data mining. A decision tree
works well with data without much preprocessing. It can use
categorical values and numerical values as it is. It can also han-
dle missing features and large-scale differences among differ-
ent components in the dataset (Ping, 2022)

 Bootstrap Aggregation (Bagging):


This general procedure can be used to reduce the vari-
ance for algorithms with high variance by first selecting ran-
dom samples of a training dataset with a replacement. The FDI
decision classification could make predictions from this weak
learner, combined to make a single prediction. This process of

IJISRT22DEC444 www.ijisrt.com 172


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
We searched for a hyperparameter grid by attempting dif- investments of up to $500, whereas "1" represents the invest-
ferent values of hyperparameters of "mtry" values between 20 ments made from $500 and above. In model B, the investments
to 30 in step of 2 for the ranger model. The combination of are classified as classes 1 to 13 in Billions of USD, as depicted
these hyperparameters creates a grid search with a size of 96. in Table 5 above. To get better results, we have normalized the
By measuring the model prediction error for each combina- datasets in the two ARIMA models (Date exclusive) with the
tion, I find the index combination that results in a minimum use of the "range" method to ensure it represents a data frame.
OOB RMSE where "mtry" is set to 28, the node size is 3, and
80% of data is sampled during the bagging process.  Time Series Stationary:
To perform any successive modeling on the FDI time se-
We make FDI inflow classification investment decisions ries, our time series must be stationary; that is, the mean, var-
on the Ada Boost dataset. The number of trees is specified us- iance, and covariance of the series should all be constant with
ing the" iter" parameter, which is set to 500 in this study. the given time. We also converted the data frame to time series
and double-checked with the Dickey-Fuller Test of Station-
 Neural Network arity, using the relevant packages to plot the "acf", "pacf", and
Typical neural networks, which are the basis of "deep the "adf." ACF stands for Auto-Correlation Function. ACF
learning" approaches, use many layers of interconnected neu- gives us values of any autocorrelation with its lagged values.
rons that transform the complex, high-dimensional input sig- In essence, it tells us how the present value in the series is re-
nal into a classification or regression output (Rifai, 2022). Us- lated in terms of its past values. PACF stands for Partial Auto-
ing the "neural net, "we regressed the dependent variable "FDI Correlation Function. Instead of finding correlations of the
inflow" variable against the other independent variables and present with lags like ACF, it finds a correlation of the residu-
set the number of hidden layers to (2,1) based on the hidden als with the next lag value.
(2,1) formula. IV. RESULTS

The linear output variable is set to FALSE, considering A. Hypothetical Model Results for SEM
the impact of the independent variables on the dependent var- Our findings indicate that the hypotheses; Return on In-
iable is assumed to be non-linear. I also set the threshold to vestment has a positive effect on FDI inflow
0.01, connoting that if the change in error during an iteration (=P) (H1c), and Security/Personal Safety
was less than 1%, then no additional optimization would be Perception has a negative influence on FDI inflow (=
carried out by the model. Applying a (2,1) configuration 0.0187**) (H1). And Investment Facilitation Percep-
yielded 0.629% classification accuracy for this model tion has a positive effect on FDI inflow (90%, =
 These are supported with significant P-
We then generated the error of the neural network model, Values <0.05. Nevertheless, the other theories (H1a, H1b,
along with the weights between the inputs, hidden layers, and H1d, H1f, and H1g) are not supported. Even though the meas-
output. We now predicted the rating using the neural network urements were not statistically significant, they were positive.
model. The expected rating would be scaled and transformed
to make a logical comparison with the actual rating. Sequel to These results also suggest the strongest (95% Confidence
that, we created a confusion matrix to compare the number of Interval) meaningful relationships among the Variable quanti-
true/false positives and negatives. ties of ROI on FDI inflow decisions (95%, = 0.9819), Se-
curity on FDI inflow decisions (95%, = 0.4902), and Invest-
 Gradient Boosting ment Facilitation on FDI inflows (90%, = 1.0465). Refer to
Boosting is a sequential technique where each new model Figure 5 and Table 6 for results.
is built from learning the established errors of the previous
model, i.e., each predictor is trained using the residual errors Figure: 5: SEM Analysis Result
of the predecessor as labels.

Boosting, which is analogous to the bagging method, ex-


cept the trees are grown sequentially: each succeeding tree is
developed using information from earlier grown trees to min-
imize the error of the earlier models.

D. Forecasting FDI Inflow Decision with ARIMA


ARIMA is the abbreviation for Autoregressive Inte-
grated Moving Average. Auto-Regressive (A.R.) terms refer
to the lags of the difference series; moving Average (M.A.)
terms refer to the lags of errors and are the number of differ-
ences used to make the time series stationary.

In model A of this ARIMA model, We used classified


"0" as the investments made by investors before 2006 to 2018,
which range from $0.00 M to above $6B, and classified "0" as

IJISRT22DEC444 www.ijisrt.com 173


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

B. Machine Learning model Results

 Logistic Regression (glm) Model

Table 7: Significance of the Variables

Table: 6: Summary of Hypotheses Test (Descriptive


Analysis)
We can see from R-generated Table 6 that Security,
Ease-Of-DoingBus, Corruption, Investment Incentives, In-
vestment Incentives, and FOREX are not statistically signifi-
cant. As for the statistically significant variables, Return-On-
Investment has the lowest p-value suggesting a strong associ-
ation of the ROI with the probability of having an FDI inflow
decision. The negative coefficient for this predictor indicates
that all other variables, being equal, security and corruption,
are less likely to have influenced FDI inflow decisions.

In the output above, the first thing we notice is the call;


this is R reminding us what the model we ran was and what
options we specified. Deviance residuals are a measure of
model fit. This part of the output shows the distribution of the
deviance residuals for individual cases used in the model.

For every one-unit change in InvestFacilitation, the log


odds of FDI inflow in $0-$500m (versus above $500m) in-
creases by 2.0022.For a one-unit increase in ROI, the log odds
of FDI inflow in $0-$500m (versus above $500m) increase by
1.7311.

IJISRT22DEC444 www.ijisrt.com 174


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
For a one-unit increase in Investment Promotion is, Pro- deviances are significantly reduced. AdditionallyAlsoAddi-
motion, the log odds of FDI inflow in $0-$500m (versus above tionally, changing the parameter improves the predictive
$600m) increases by 1.5771. power of the model (AUC)

The ANOVA test is performed, and the difference be-  K-Means (KNN) Model
tween the null and residual deviance depicts how the model is
doing against the null model (a model with only the intercept). Fig 8: The graph indicates the proximity
The more significant this gap, the better it becomes. There is a
drop in deviance when we add each variable one at a time.

Besides, increasing ROI and Investment Promotion fac-


tors significantly reduce residual deviance. The other variables
seem to improve the model less. This is depicted in Appendix
H.
In making a prediction, we used "FDI_rf_model_glm" to
assess the predictive ability.

We evaluated the fitting of the model to determine how


the model is doing when predicting you on a new set of data
by setting the parameter type "response" with the expectation
that R would output probabilities in the form of P(y=1|X). Our
decision boundary would be 0.5 with the condition that if We now have 4 clusters of variables
P(y=1|X) > 0.5, then y = 1; otherwise, y=0. We now re-ran the Algorithm using other distances. We
obtained a new cluster size which indicates the cluster of the
We now plotted the ROC curve and calculated the AUC 250s observations in the combinations of 73, 84, 26, and 67.
(area under the curve), which are typical performance meas- ResaM.pling results across tuning parameters used k=11 to ob-
urements for a binary classifier tain the minimal (best) RMSE of 0.4772065

We also checked on the model misclassification. Mis- Table 8: Resampling minimal RMSE error
classification errors came out to be 41.93%. We could use re-
gression techniques with categorical variables to compare var-
ious other data. Although, as a rule of thumb, a model with a
sound predictive ability should have an AUC closer to 1 (1 is
ideal) than 0.5; this, and any other models, ranging from 0.52
to 0.63 because more than half of the dependent variables don't
have a significant P-value. We changed the parameters to im-
prove the AUC of this model from 0.5524 to 0.5767.

Fig 6: AUC for Changed Parameter

The chart shows that elbow point 2 provides the best


value for k. While WSS will continue to drop for larger values
of k, we have to make the compromise between overfitting.
Here, the elbow point provides that compromise where WSS,
while still decreasing beyond k = 2, drops at a much lower rate.
In other words, adding more clusters beyond 2 brings less im-
provement to cluster homogeneity

To confirm that, we also applied the Silhouette Method


In the graph, the x-axis is the false positive rate, and the to determine the number of clusters. Again, we see that 2 is the
y-axis is the true positive rate. We can see each of the points ideal number of clusters. Here we look for large values for the
represents a confusion matrix that we don't have to evaluate Silhouette Width (Y-Axis)
manually. The points also represent the tradeoff between true
positive and false positive, as depicted in Figure 7. ROI has
the lowest p-value suggesting a strong association of the ROI Fig 9: The best K with WithiN.N.N. Sum of Square "wss"
with the probability of having an FDI inflow decision, and by
increasing ROI and Investment Promotion factors, the residual

IJISRT22DEC444 www.ijisrt.com 175


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Result:

Alternatively, I applied the Silhouette Method to deter-


mine the number of clusters. Again, we see that 2 is the ideal
number of clusters. Here we look for large values for the Sil- Fig 11.: Neural network with weights
houette Width (Y-Axis). K= 3 has the best accuracy and so
fitted the KNN model with K at three and generated confusion
Matrix and ROC-AUC. The summary of accuracy metrics is
shown in Table 14

 Random Forest, Bagging, and Boosting Trees


It can be seen that the random forest model has 500 trees,
which is the default setting, and two variables were tried at
each split; it is our m parameter. The model seems to have a
relatively low R-squared value of 2.28% because a few varia-
bles out of eight have significant p-values. The plot of the
training error rate against the number of trees indicates a con-
siderable improvement for adding the first 100 trees and virtu-
ally a flat error rate after that, as shown in Figure 10.
It can be seen that each predictor variable has one neuron.
Fig 10: Error Rates by Trees # (improved) The first layer has eight neutrons. The second layer has two
neurons, and the out variable has one neuron.

The model generates seven true negatives (0's), 31 true


positives (1's), and 27 false positives (0 s), while there are 17
false negatives. Ultimately, we yield an 54.84% accuracy rate
in determining whether an FDI investment decision is above
$600m or not, with an AUC of 0.5013. The misclassification
error came out to be 13.67%. We can further increase the ac-
curacy and efficiency of our model by increasing or decreasing
nodes and bias in hidden layers. The strength of machine learn-
ing algorithms lies in their ability to learn and improve every
The MSE is at its minimal when the number of trees is time in predicting an output.
101, and the RMSE of this optimal random forest is
0.4849.0.4849.  Gradient Boosting
The predicted numbers using six trees are shown in Table
The confusion matrix output classified eight false posi- 11
tives and 17 false negatives. A summary of the result is de-
picted in Table 26 and ROC-AUC in Appendix P Table 11.: Predicted numbers

 Neural Network
Given the output, we may conclude that both repetitions
converge. However, we will use the output driven in the sec-
ond repetition because it gives less error (97.40946) than the
error (100.43106) from the first repetition derives. We now
generate the error of the neural network model, along with the
weights between the inputs, hidden layers, and outputs, as
shown in Table 28 and Figure 16

Table 10: Error of the neural network Fig 12.: ROC-AUC for Gradient Boosting

IJISRT22DEC444 www.ijisrt.com 176


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
The first part of the output above shows the ratios of
FDIInflow above $500m (yes) and FDIInflow below $500 m
(no) in the training set (called a priori probabilities), followed
by a table giving for each target class, mean and standard de-
viation of the (sub-)variable. Also, note that the Naive Bayes
algorithm assumes a normal distribution for the independent
variables using numeric predictors. The conditional probabili-
ties for each attribute level given the FDIInflow status are
missing because all the predictors are non-categorical.

Of the variables, FOREX, Security/safety, and Ease off- The AUC and Accuracy of the model are not that
Doing-Business are on top of the list strongly in tune with the model's algorithms and the predictors'
significance.
 Decision Tree Model
Security has the highest variable importance, as depicted C. Consolidated Results Summaries
in Table 12.
Table 13: Consolidated summary of other ML models
Table 12: Split errors and Variable Importance

We now used Tableau to visualize the summary of all


other ML models (Accuracy only) and the combined metrics
result (AUC inclusive) as depicted in Figure 14
We tried to tune the parameters and see if we could im-
prove the model over the default value. But need to get an ac- Fig 14.: Visualized summary of other ML models (combined
curacy higher than 0.548. We now have an improved variable metrics)
of importance with the controlled part, as shown in Table 13.

Table 13: Split errors and Variable Importance (improved)

From the tree, it is clear that those who have an Invest-


ment Promotion decision of less than 0.73, Ease of Doing busi-
ness of less than 0.72, and Security decisions equal to or
greater than 0.69 are those that invested in more than $500m,
as depicted in Figure 13
Analysis of the models indicates that at 0.72, K-nearest
neighbors had the highest accuracy and were 40.07% higher
Fig 13: Decision Tree
than Naïve Bayes, which had the lowest accuracy at 0.52. Ac-
curacy and total AUC are positively correlated with each
other. K-nearest neighbors accounted for 17.43% of accu-
racy. Across all 7 Models, Accuracy ranged from 0.52 to 0.72,
AUC ranged from 0.50 to 0.72, and F.I. ranged from 0.54 to
0.76.

The visualization suggests that Random Forest and Ada


Boosting Trees have effective classification accuracies despite
the context of "the No free lunch" theorem

D. Forecasting FDI Inflow Decision with ARIMA


 Naïve Bayes Model

IJISRT22DEC444 www.ijisrt.com 177


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig 15a: Model A

After converting the data and then checking the partial A


correlation function, We confirmed that the ACF of the resid-
uals shows no significant autocorrelations.

In model A, the ACF of the residuals shows no signifi-  FDI decisions Prediction
cant autocorrelations, as shown in Appendix P. Even after con- Now we make a forecast at 95% CI and forecast for ten
version, the data remains stationary, with no difference after years using the ARIMA Model as depicted in Figures 16a and
being controlled. The Dickey-Fuller test returns a p-value of 16b.
0.8611, resulting in the rejection of the null hypothesis and ac-
cepting the alternative hypothesis that the data is stationary, as Fig 16a: Forecast for 10 Years
shown in Figure 15b Results:

Fig 15b: Model B

Fig 16b: Forecast for ten years

In model B, the ACF of the residuals shows no signifi-


cant autocorrelations. After conversion, we have improved
stationary data, with no difference after being controlled. The
Dickey-Fuller test returns a p-value of 0.2174, as in Figure 21,
resulting in the rejection of the null hypothesis and accepting
the alternative hypothesis that the data is stationary, as shown
in Appendix Q.Q. The two outputs have fluctuating patterners
but are stationary with P-values of more than 0.05, so we
checked the best model

 Choice of best models


To create an FDI model, we used the "auto. Arima()"
function in R that uses a combination of unit root tests and The forecasts are shown as a blue line, with the 80% pre-
minimization of the AIC and MLE to obtain an ARIMA diction intervals as a dark-shaded area and the 95% as a light-
model. The best ZRIMA in model A is ARIMA(0,0,0), and in shaded area. Validate the forecast with "Box. Test" to check if
the model, B is ARIMA(1,0,0), as shown in Table 15 there are correlation problems. We checked with lags 2, 5, and
7. None of the results is significant, i.e..,>0.05, indicating no
correlation issues. This is illustrated in Appendix R

Table 15: ARIMA(0,0,0) and ARIMA (1,0,0)

IJISRT22DEC444 www.ijisrt.com 178


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
We now look at the residuals of the model. The residuals V. CONCLUSIONS – DISCUSSIONS
will tell us if the model was able to capture all of the infor-
mation provided by the data. Three graphs, one story: 1) Time A. General
Series graph of the residuals, 2) ACF Graph for the first five Investment decision-makers always quest to predict the
lags, 3) Distribution of the residuals values of foreseeable foreign direct investment and its fore-
most factors to ascertain its needs for the funding required for
For model A, the model left information in the residuals. investment in general on the one hand and, on the other hand,
The first and last graphs show us that the residuals don't appear to identify the approaches of attracting the beneficial FDI to
to be in significant white noise but have some correlation the host country. In this study, we used machine learning al-
among them, and the second graph confirms it. None of the gorithms to predict potential investors' foreign investment in-
first four lags go above the threshold established by the auto- flow decisions and their fundamental determinants. Sequel to
correlation function. using numerous algorithms for accuracy tests, it appears that
the foreign direct investment inflows into Nigeria for the next
For model B, the first and last graphs show that the resid- decade, 2018–2028, would be on the decline, especially given
uals don't appear to be in significant white noise but have some the insecurity challenges.
correlation, and the second graph confirms it. None of the first
five lags go above the threshold established by the autocorre- This study addresses fundamental problems related to
lation function. forecasting inflows of FDI based on individual decisions and
to better understand the efficacy of ML models for making
We analyzed time series on the FDI inflow decisions such decisions on multi-classes of investments. To this end,
modeling in R. We dwelled on how to "rationalize" our data, we attempted to gain a deeper understanding of how individual
determine order parameters from ACF/PACF, and ultimately ML models predict FDI decisions.
how to build our ARIMA model and make a prediction with
the two models. Time series modeling is a complex aspect, es- To this end, we ventured to gain a deeper understanding
pecially in the context of FDI decisions that determine the of how individual perceptions drive FDI decisions and, more
level of investments. importantly, how the role of ML models in FDI inflow fore-
casting. We have addressed the two researched questions (1)
From the results, we may infer that: To what extent do significant FDI decision factors contribute
to determining the choice of the best A.I. model for prediction,
 An all-inclusive set of specific factors. and (2) How and to what extent does the class size of the target
We found considerable differences in individual experi- FDI decision variable determine its predictability?
ences of the institutional environment, beneficial business en-
vironment, and safety concerns for FDI decisions. Some of On question (1), the findings support the hypothesis that;
these factors have a peculiar influence on various Machine Return on Investment has a positive effect on FDI inflow, and
Learning Models in terms of classification accuracy. The Security/Personal Safety Perception negatively influences FDI
fewer the number of significant variable predictors, the weaker inflow. And Investment Facilitation Perception has a positive
the predictive efficacy effect on FDI inflow and is supported by significant P-Values
<0.05. Nevertheless, the other theories (H1a, H1b, H1d, H1f,
 Distinctive internal algorithms designed for each model and H1g) are not supported.
Although the "No Free-Lunch theorem" states that all op-
timization algorithms perform equally well when their perfor- With regards to question (2), given the ARIMA with
mance is averaged across all possible problems, this implies models A and B, it is suggested that model B, with 13 different
that there is no single best optimization algorithm; we discover classes of the FDI target variables (compared to just 2), has a
that specific models have the potential to make a more accu- higher AIC, BIC. And Dickey-Fuller test returns a p-value
rate prediction on individual FDI decisions depending on the which signifies a more reliable prediction/forecast.
efficacy and size of the data collected. This also answered our
first question. We now know that the research questions have shaped
the purpose of our study and formed the methods and the de-
 The effect of the Length of Target classification sign of our analytics. Further to analytics of various ML mod-
The length of the vector class plays a critical role in de- els in the contexts of the "No Free-Lunch theorem," the en-
termining its predictive efficacy, such that the longer/number semble method can be applied and often performs much better
of the categories of the target variable, the more feasible and than any single classifier.
better the effectiveness of its forecasting ability. However, in
the quest for the accuracy of any predictive models, when re- B. Summary of Conclusions - Practical Implication
searchers provide themselves with the correct data, they are These findings imply that the main FDI decision drivers
sure to have a clearer view of the future, leading to more ef- and other key stakeholders have to look inwards to identify
fective use of resources. other significant individual FDI determinants to make more
effective predictions of what FDI inflows within the next dec-
ade would be like. The ROC-AUC, Accuracy, and other met-
rics established under this study indicate average results (0.5
to 0.6). Still, ideally, it should be over 7.0.

IJISRT22DEC444 www.ijisrt.com 179


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
REFERENCES
The concerned IPAs in the African Sub-Region should
empower and dedicate various R&D units to use the latest da- [1]. Akbar, D. P., & Yusaf, H. 2014. Neural Network
taset and ML software for business analyses for a better FDI Approaches to Estimating FDI Flows: Evidence from
policy. Central and Eastern Europe.
http://dx.doi.org/10.2753/EEE0012-8755440302.
IPAs must be innovative to attract beneficial FDIs to [2]. Akinlabi, A. O., & Hamed, B. 2011. Corruption, foreign
their regions. For IPAs to perform effectively in this 21st cen- direct investment and economic growth in Nigeria: An
tury, while they hold onto the status quo, they must transform empirical investigation. Journal of Research in
from investment intelligence to investment analytics with a International Business Management, 1(9): 278-292.
view to making predictions. [3]. Alavinasab, S. M. 2013. Determinants of foreign direct
investment in Iran. International Journal of Academic
With a number of Sub-Saharan African countries having Research in Business and Social Sciences, 3(3): 258-
electricity infrastructure challenges, among others, it is im- 269.
portant that IPAs predict a surge or plummet in electricity gen- [4]. Badie, B., Berg-Schlosser, D., & Morlino, L. (Eds.).
eration and transmission investment projects by enhancing the 2011. International encyclopedia of political science.
enabling environment to attract FDI. For example, recently, in Thousand Oaks, CA: SAGE Publications.
August 2021, global foreign direct investment (FDI) an- [5]. Blomström, M., Kokko, A., & Mucchielli, J. 2003. The
nouncements surged as software and financial firms expanded economics of foreign direct investment incentives. In H.
their international footprints, and renewable developers out- Herrmann, & R. Lipsey (Eds.), Foreign direct
lined major green electricity generation projects. Strategic col- investment in the real and financial sector of industrial
laborations among the key FDI stakeholders and between aca- countries: 37-60. Berlin, Heidelberg: Springer.
demia and industry are highly suggested to achieve effective [6]. Brauer, R. L. 2016. Safety and health for engineers (3rd
investment synergy. ed.). Hoboken, NJ: John Wiley & Sons.
[7]. Brown, L., & Hibbert, K. 2017. The Effect of Crime On
We now also know that this study contributes to the lit- Foreign Direct Investment: A Multi-Country Panel Data
erature on individual FDI decisions and economic growth. The Analysis. Journal of Developing Areas, 51(1): 295-307.
importance of integrating investment promotion with invest- [8]. Brownlee, J. 2019. Probability for Machine Learning -
ment facilitation has the potential to attract potential and ex- Google Books: Machine Learning Mastery.
isting investors from targeted regions. We suggest that FDI [9]. Charlton, A., & Davis, N. 2017. Does investment
policymakers will find a modern-day scholarship, especially promotion work? The B.E. Journal of Economic
the application of A.I. more-and-more-helpful when they em- Analysis & Policy, 7(1): 1-21.
ploy the method across the board for informed decisions and [10]. Contributor, T. 2012. What is decision tree?:
policymaking. WhatIsDotCom.
[11]. Corcoran, A., & Gillanders, R. 2014. Foreign direct
C. Limitations of the Study investment and the ease of doing business. Review of
We have a limited number of observed data in size and World Economics, 151(1): 103-126.
independent variables. In predicting FDI inflows based on in- [12]. Creswell, J. W., & Creswell, J. D. 2018. Research
vestment decisions, we lacked monthly, quarterly, and semi- design: Qualitative, quantitative, and mixed methods
annual data, so we were left with no options but to use yearly approaches (5th ed.). Thousand Oaks, CA: Sage
data in the study. We had challenges generating a confusion Publications.
matrix for some models because of their design or internal [13]. Cuervo-Cazurra, A. 2006. Who cares about corruption?
structures. In designing the model layout using Analysis of Journal of International Business Studies, 37(6): 807-
Moment (AMOS) for assessing the direct impact of the I.V.s 822.
on DV, we had to use a surrogate variable, "Business Environ- [14]. David Brady, M. S. 2010. Leadership and Growth
ment," as a mediator for "FOREX" to get a good model fit. (illustrated ed.): World Bank Publications, 2010.
[15]. Dinda, S. 2008. Factors determining FDI to Nigeria: An
D. Future Research empirical investigation: MPRA Paper 28097, University
Our proposed future research would use quarterly preex- Library of Munich, Germany. https://mpra.ub.uni-
isting FDI time-series data (to boost the number of observa- muenchen.de/28097/.
tions) and also explore the effect of FDI decisions in some eco- [16]. Doshi, R., Kelley, J. G., & Simmons, B. A. 2019. The
nomic sectors bordering on manufacturing, entertainment, and power of ranking: The ease of doing business indicator
services sectors. It is anticipated that future research should and global regulatory behavior. International
also integrate other predictor variables that were missing in Organization, 73(3): 611-643.
this empirical stud [17]. Effiom, L., & Etim Edet, S. 2019. Facilitation of foreign
direct investment: Evidence from Cross River State,
Nigeria. International Journal of Accounting and
Finance (IJAF), 8(2): 75-95.
[18]. Fred J. Damerau, N. I. 2010. Handbook of Natural
Language Processing: CRC Press.

IJISRT22DEC444 www.ijisrt.com 180


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[19]. G.PeterZhang. 2003. Time series forecasting using a [39]. Morisset, J., & Andrews-Johnson, K. 2004. The
hybrid ARIMA and neural network model. 50. effectiveness of promotion agencies at attracting
[20]. Geetika, Ghosh, P., & Chowdhury, P. R. 2015. foreign direct investment. Washington, DC: FIAS
Managerial economics (3rd ed.). India: McGraw-Hill Occasional Paper;No. 16. Washington, DC: World Bank.
Education. https://openknowledge.worldbank.org/handle/10986/15
[21]. Greppin, C., Carlsson, B., Wolfberg, A., & Ufere, N. 073.
2017. How U.S. Executive expatriates work in [40]. Nations, U. 2022. THE 17 GOALS | Sustainable
environments of pervasive corruption. Development, Vol. 2022: Department of Economic and
[22]. Hair, J. F., Black, W. C., Babin, B. J., & Anderson, R. E. Social Affairs Sustainable Development.
2010. Multivariate data analysis (7th ed.). Upper Saddle [41]. Nau, H. R. 2018. Perspectives on international relations
River, NJ: Pearson Prentice Hall (6th ed.): C.Q. Press.
[23]. Harding, T., & Javorcik, B. S. 2011. Roll out the red [42]. Nemati, H. 2017. Information Security and Ethics:
carpet and they will come: Investment promotion and Concepts, Methodologies, Tools, and Applications
FDI inflows. The Economic Journal, 121(557): 1445- (illustrated ed.): IGI Global, Sep 30, 2007.
1476. [43]. Netshivhazwaulu, N. 2018. Forecasting Foreign Direct
[24]. Hees, F., & Mendonça Cavalcante, P. 2017. Focusing on Investment in South Africa using Non - Parametric
investment facilitation - Is it that difficult?, No. 202 ed., Quantile Regression Models. South Africa: University of
Vol. 2017. Columbia Center on Sustsinable Investment. Venda.
[25]. Hinkin, T. R. 1998. A brief tutorial on the development [44]. Nigerian Investment Promotion Commission. 2017.
of measures for use in survey questionnaires. Invest Nigeria home page.
Organizational Research Methods, 1(1): 104-121. [45]. Nurudeen, A., Wafure, O. G., & Auta, E. M. 2011.
[26]. Hossain, M. T., Hassan, Z., Shafiq, S., & Basit, A. 2018. Determinants of Foreign Direct Investment: The Case of
Ease of doing business and its impact on inward FDI. Nigeria. IUP Journal of Monetary Economics, 9(3): 50-
Indonesian Journal of Management and Business 67.
Economics, 1: 52-65. [46]. OECD. 2019. OECD.org - OECD.
[27]. International Centre for Trade and Sustainable [47]. Omodero, C. O. 2019. Effect of corruption on foreign
Development. 2020. Investment facilitation: ICTSD. direct investment inflows in Nigeria. Studia
[28]. Javier Arroyo, C. 2009. International Journal of Universitatis "Vasile Goldis" Arad – Economics Series,
Forecasting(25): 192-207. 29(2): 54-66.
[29]. Kaiser Meyer, O. 1970. Kaiser-Meyer-Olkin measure for [48]. Osarenkhoe, A., & Byarugaba, J. M. 2016. Service
identity correlation matrix. Journal of the Royal Quality Perceptions of Foreign Direct Investors.
Statistical Society, 52: 296-298. http://dx.doi.org/10.1080/10496491.2016.1185492.
[30]. Kitchin, R., & Thrift, N. (Eds.). 2009. International [49]. Panibratov, A. 2017. International strategy of emerging
encyclopedia of human geography. (Vol. 1). U.K.: market firms: Routledge.
Elsevier. [50]. Paul C. Avey, M. C. D. 2014. What Do Policymakers
[31]. Kolstad, I., & Villanger, E. 2008. Determinants of Want From Us? Results of a Survey of Current and
foreign direct investment in services. European Journal Former Senior National Security Decision Makers.
of Political Economy, 24(2): 518-533. International Studies Quarterly, 58(2): 227-246.
[32]. Korinek, J., & Sourdin, P. 2011. To what extent are [51]. Ping, D. 2022. The Machine Learning Solutions
high-quality logistics services trade facilitating? Paris: Architect: Packt Publishing.
OECD Trade Policy Papers. No. 108. [52]. Pradhan, R. P. 2020. Forecasting Foreign Direct
https://doi.org/10.1787/5kggdthrj1zn-en. Investment in the Asian Economy: An Application of
[33]. Kostelnick, C. 2014. Mosby's textbook for long-term Neural Network Modeling. International Economic.
care nursing assistants, 7th ed. St. Louis, MO: Elsevier [53]. Project Management Institute. 2017. PMI guide to
Health Sciences. business analysis: Project Management Institute.
[34]. Lei Wen, X. Y. 2020. Forecasting CO2 emissions in [54]. Rifai, N. 2022. Tietz Textbook of Laboratory Medicine:
Chinas commercial department, through B.P. neural Elsevier Health Sciences.
network based on random forest and PSO. [55]. Rolfe, R. J., Ricks, D. A., Pointer, M. M., & McCarthy,
ScienceDirect, 718. M. 1993. Determinants of FDI incentive preferences of
[35]. Loftus, J., Leo, K., Daniliuc, S., & Boys, N. 2020. MNEs. Journal of International Business Studies,
Financial reporting (3rd ed.): John Wiley & Sons. 24(2): 335-355.
[36]. MacKenzie, S. B., Podsakoff, P. M., & Podsakoff, N. P. [56]. Roy, S. S. 2020. Predictionof Foreign Direct Investment:
2011. Construct measurement and validation procedures An Application of Long Short-Term Memory. Volume
in MIS and behavioral research: Integrating new and 57. No. 2, 2020(Vol. 58 No. 2 (2021): Volume 58 No. 2
existing techniques. MIS Quarterly, 35(2): 293-334. (2021)).
[37]. McGraw, T. 2016. Financial institutions & markets (5th [57]. Sauvant, K. P., & Hamdani, K. 2015. An international
ed.): McGraw-Hill Education. support programme for sustainable investment
[38]. Moosa, I. A. 2002. Foreign Direct Investment: Theory, facilitation: The E15 Initiative. International Centre for
Evidence and Practice - I. Moosa - Google Books: Trade and Sustainable Development (ICTSD).
Palgrave Macmillan UK, 2002. http://ccsi.columbia.edu/files/2015/08/KPS_KH-SIFU-
published-July-15.pdf.

IJISRT22DEC444 www.ijisrt.com 181


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[58]. Simmons, B. 2019. The impacts of the World Bank ease
of doing business rankings: University of Pennsylvania.
[59]. Simon Nusinovicia, Y. C. T. M. Y. Y. 2020. Logistic
regression was as good as machine learning for
predicting major chronic diseases. ScienceDirect, 122:
56-69.
[60]. Stanley Jere, B. K. O. C. 2017. Forecasting Foreign
Direct Investment to Zambia: A Time Series Analysis -
OJS_2017022816111501.pdf.
[61]. Stevens, G. V. G. 1998. Exchange rates and foreign
direct investment: A note. ScienceDirect, 20(3): 393-
401.
[62]. Sung, H., & Lapan, H. E. 2000. Strategic foreign direct
investment and exchange-rate uncertainty. International
Economic Review, 41(2): 441-423.
[63]. Tavakol, M., & Dennick, R. 2011. Making sense of
Cronbach's alpha. International Journal of Medical
Education, 2: 53-55.
[64]. Udeh, S. C., & Ihezie, U. 2013. Insecurity and national
economic development implications for Nigeria's vision
20: 2020. International Journal of Development and
Management Review, 8(1): 93-109.
[65]. Ufere, N., Perelli, S., Boland, R., & Carlsson, B. 2012.
Merchants of corruption: How entrepreneurs
manufacture and supply bribes. World Development,
40(12): 2440-2453.
[66]. United Nations. 2007. World investment report 2007:
United Nations Conference on Trade and Development.
[67]. Wang, H.-C. W., Jen-Hsiang, C., & Pei, W. 2021. Cash
Holdings Prediction Using Decision Tree Algorithms
and Comparison with Logistic Regression Model.
https://doi.org/10.1080/01969722.2021.1976988.
[68]. Wensveen, J. G. 2016. Air transportation: A
management perspective (8th ed.): Routledge.
[69]. wikipedia. 2018. Design - Wikipedia.
[70]. World Bank. 2010. Innovation policy: A guide for
developing countries (Illustrated ed.): World Bank
Publications.
[71]. Yoon, J. 2020. Forecasting of Real GDP Growth Using
Machine Learning Models: Gradient Boosting and
Random Forest Approach. Computational Economics,
57(1): 247-265.
[72]. Yusuf, M. A. 2020. PhD thesis. Case Western Reserve
University, OhioLINK-EDT Center.
[73]. Zakamulin, V. 2017. Market timing with moving
averages: The anatomy and performance of trading
rules: Palgrave Macmillan.
[74]. Zhang, Z. 2016. Introduction to machine learning: k-
nearest neighbors - PMC. Ann Transl Med: 218.

IJISRT22DEC444 www.ijisrt.com 182

You might also like