Non-Banking Sector development effect on Economic Growth. A Nighttime light data approach

This paper uses nighttime light(NTL) data to measure the nexus of the non-banking sector, particularly insurance, and economic growth in South Africa. We hypothesize that insurance sector growth positively propels economic growth due to its economic growth-supportive traits like investment protection and optimal risk mitigation. We also claim that Nighttime light data is a good economic measure than Gross domestic product (GDP). We used weighted regressions to measure the relationships between nighttime light data, GDP, and insurance sector development. We used time series South African GDP data collected from the World Bank for the period running from 2000 to 2018, and the nighttime lights data from the National Geophysical Data Centre (NGDC) in partnership with the National Oceanic and Atmospheric Administration (NOAA). From the models fitted and the reported BIC, AIC, and likelihood ratios, the insurance sector proved to have more predictive power on economic development in South Africa, and radiance light explained economic growth better than GDP and GDP/Capita. We concluded that nighttime data is a good proxy for economic growth than GDP/Capita in emerging economies like South Africa, where secondary data needs to be more robust and sometimes inflated. The findings will guide researchers and policymakers on what drives economic development and what policies to put in place. It would be interesting to extend the current study to other sectors such as micro-finances, mutual and hedge funds.


Introduction
Before drawing the line between our contribution and the existing literature, we give our main contributions to this study as follows: I.
Measuring economic growth contribution from the insurance industry using satellite imagery and nightlight data is new, to the best of our knowledge.

II.
The approach is convenient because all the daytime satellite images and nightlight intensity data are evenly accessible from the open-source online interfaces linked to South Africa.

III.
The study approached the problem more robustly.We matched all the grid images of different economic sectors (insurance, micro finances) to South African country codes to generate corresponding features.Thus, ensuring less variation and noise within our data, analytics, and results.
The role of the non-banking sector in economic growth and development is considerably growing and is paid attention to across the globe.Nevertheless, scant research has been conducted to unravel the relationship between the insurance sector and real-time economic growth.It is vital to indicate that the insurance sector plays an imperative role in any economy by providing indemnity or investment protection.Moreover, its ability to pool funds in the form of premiums enables it to be a crucial institutional investor in emerging economies like South Africa.From another position, measuring economic activities and output is primarily done using traditional means of averaging Gross domestic product (GDP).However, this carries some bias, leading to incorrect and less accurate results and policies.Therefore, this study aims to measure the insurance-growth nexus in South Africa using nighttime light data (NTL) rather than Gross domestic product.Our conjecture is that; insurance sector businesses positively drive economic growth due to their economic growth-supportive Key Performance Indicators (KPIs) like investment protection and labor (life) insurance, among others.We also claim that Nighttime light data is an excellent economic growth measure than Gross domestic product.
Of further interest, the studies on the subject have been mainly of a cross-sectional or panel nature, for instance, Azman-Saini and Smith (2011) and Moshirian, et al. (2010).Most research done on the subject mainly employed a panel data analysis approach.The most significant drawback of this method is losing specific effects in analysis, which, if combined with inaccurate GDP data, needs to be more accurate.Therefore, it is essential also to interrogate the relationship between insurance sector development and economic growth using nighttime light data on a time series basis.In this study, we are using South Africa among all other emerging economies because of its stage of development and financial stability.Establishing the nature of the relationship between the insurance sector and economic growth in South Africa being our first intention, the performance comparative hypothesis testing for GDP and NTL data makes up the novelty of this paper.There is substantial literature explaining the role of the insurance sector in economic growth in South Africa.
Sibindi and Godi (2014) discussed the relationship between insurance growth and economic growth in South Africa.They used insurance density as the proxy for insurance market development and real per capita growth of domestic products for economic growth.They tested for cointegration amongst the variables by applying the Johansen procedure and then tested for Granger causality based on the vector error correction model (VECM).They found at least one co-integrating relationship and indicated that the direction of causality runs from the economy to the long-term insurance and from the economy to the entire insurance sector.Ward and Zurbruegg (2000) examined the relationship between economic growth and growth in the insurance industry for nine OECD countries.Using annual data, they conducted a bivariate cointegration analysis and tested for causality using the real GDP and the total real premiums in each country from 1961 to 1996.They found that in some countries, the insurance industry Granger causes economic growth; in others, economic growth Granger causes the development of the insurance sector.This agrees with the works of Haiss and Sümegi (2008), who investigated the impact of insurance investment and premiums on GDP growth in Europe using a cross-country panel data analysis for 29 European countries from 1992 to 2009.
Chi-Wei, Hsu-Ling, and Guochen (2013) also apply the bootstrap Granger causality test to examine the relationship between insurance development and economic growth in seven Middle Eastern countries.They as well used insurance density as the indicator for insurance development.
Their results found evidence for bi-directional causality between the life insurance sector and economic growth in higher-income countries such as United Arab Emirates, Kuwait, and Israel.All these works were based on investigating the role and causal links between insurance and economic growth, and their results proved to move in the same direction.They all found a positive link between economic growth and insurance businesses.However, despite all the interesting results from the literature, we found no direct application of nighttime light data as a proxy measure of economic growth when assessing the role of insurance in economic development.However, the correlation is weaker concerning wages.All these related works proved that nighttime light data (NTL) is a proxy measure of economic activities in an economy.However, they loosely looked at the association of NTL data as a proxy of economic growth and insurance businesses in South Africa.We present examples of satellite images for the period 1992 and 2020, respectively, in Africa.
Figure 2 gives a clear clue about the satellite and nighttime light data about South Africa-our country case.Different satellite systems are used to record the datasets of nighttime lights for extended analysis and comparative analysis with the real GDP.

Per Capita Gross domestic product (GDP/Capita)
It is a measure of total output produced within the borders of an economy per head, OECD (2023).
It is considered a good measure of the economic growth of a nation.It is typically measured using three distinct methods: output, expenditure, and income.All of these give reasonable estimates if used correctly and appropriately, and the integrated use of all these is our recommendation.GDP is a good measure of economic growth only in countries where statistics are well-enriched and well-defined in a structured way.This is a common phenomenon in developed or well emerged economies.As such, we conjectured that nighttime light data is instead of GDP a good economic measure in developing and emerging economies like South Africa.GDP per capita is an important indicator of economic performance and a valuable unit for cross-country comparisons of average living standards and economic well-being.It is usually a non-static variable, meaning it varies with time.This may come from new resource discoveries or exhaustion, technological advancement, or a change in any factor that affects production.For South Africa, Figure 3 depicted below shows the GDP growth patterns.
Figure 3:we present the GDP values (%) using World Bank data (2022).Ideally, the GDP fluctuates over the entire period considered.Interestingly, there was a sharp increase between 2019 and 2021 due to Covid19.
Concerning GDP per capita, there are other measures of economic growth, such as Gross national income (GNI), Real GDP, and Gross national product (GNP).These are all suitable measures of economic growth in countries with excellent and well-organized statistical figures, unlike in developing countries like South Africa, where we have the opposite case.Therefore, alternative methods, such as nighttime light data, are ideal in such countries.

Nighttime light data and Economic growth
The application earthly satellite images at night (night lights data) was initially invented to study human activities and natural events.However, the idea was extended to economics, arguing that night lights can be good measures of economic growth for any active economy.Additionally, night lights can be used to measure poverty, and social inequalities.Also, night lights data helps in answering questions that previously had no answers especially in places with insufficient data.To deeply understand the role of night lights data on economic growth, we briefly look at the composition of the satellite image pixels and their economic relevancy.Ideally, each pixel of a satellite image represents an area of less than one square kilometer on earth.It is comprising of a digital number measuring the brightness at night.Intuitively, the higher the number the brighter the spot.When we integrate over all pixels for a given country (South Africa), it becomes an indicator that measures the activities of that country at night.In this study, we integrate these numbers over all pixels for industrial sectors with insurance businesses for South Africa.This gives us a barometer of economic development and fluctuations in South Africa.Ideally, night lights help in measuring the dynamics and developmental changes of one or more economies over time, yet GDP cannot capture this phenomenon in depth.

Data and variables
We used time series South African GDP data collected from the World Bank for the period running from 2000 to 2018.Our choice for South Africa is entirely driven by the motive to assess the economic growth patterns and rate for emerging economies.Additionally, we used a finegrained geo-coded industrial full sample micro-data set for South Africa and match it with radiance and saturated light emissions.We collected nighttime lights data from the National Geophysical Data Centre (NGDC) in partnership with the National Oceanic and Atmospheric Administration (NOAA).The data is used to test our set of hypotheses.Overall, our analysis consulted the following variables:

Variable Description
Insurance

Matching the Nighttime lights data (NTL) and Economic Data
To compare the light data with the Statistics of South African demographic data, we used a generated grid based on the coordinates of the provided data on the concentration of the insurance companies and matched geographical locations and respective economic outputs.The light data were then summarized in the grid to allow for analysis.Using ESRI's ArcGIS software, we created a grid of squares based on the lower left coordinates provided by the Statistics department of South Africa.Separate grids were generated for the insurance industry and population data, resulting in two grids overall (the coordinates provided were based on the available data).This was to refine our analysis and to make them precise.In addition, we regressed real GDP/Capita and nighttime lights (see results).

Correlation analysis
We now turn to the empirical analysis, connecting the light-emission data, GDP data, and the selected insurance indicators.We begin with a fundamental correlation analysis to identify any bivariate relations.Separate analyses were done on (1) the correlation of activity light intensity and insurance premium density and (2) GDP/capita and insurance premium density.We employed Radiance light and Saturated Light, GDP/capita, GDP, and insurance indicators in the correlations for robust comparisons.We also computed correlations for specific industry and occupational structures, with all correlation coefficients significantly weaker than for total people or industry values.However, the correlation between people and light is generally more robust when we use radiance light emissions instead of saturated light.Also, the correlation between GDP, GDP/capita, and insurance indicators are weaker than for nighttime data.

Weighted Regression(WR) Analysis
Weighted regression is a technique that allows us to examine possible spatial non-stationarity by using weighted sub-samples of the data.This implies that we can produce locally linear regression estimates for every point in space.In other words, we generate, which is a local estimate for every observation.This implies that the  will be there to capture any effect on the systematic estimates.The effect can be geographical such as latitude changes.Such changes can, in some way, affect economic activity.These disparities will then be captured by the Beta coefficients in the selected regions of South Africa and visible in the maps generated from the analysis.The methodology makes it possible to compare the unstandardized  coefficients from the OLS estimation of the involved parameters.In simpler terms, the idea is that the WR estimation produces information about parameter variation over space.The difference between WR and spatial autocorrelation techniques is that the latter identifies spatial dependence through the residual, while the former addresses spatial non-stationarity directly through the estimated parameters.In a WR, we assume the regression model to be: The estimator of our parameters is given by; ̂() = (  ()) −   (), , , , Where () is a weight specific matrix to location , observations near to  are given greater weight than distant observations.However, in our empirical analysis below, all estimations will be in single regressions.This is done to comply with the strong multicollinearity between the independent variables. ̂()is the vector of parameters.These parameters are sensitive and form the basis of our WR models. is the vector specifying our covariates incorporated in this study, and they explain South African economic development.This study incorporates variables like insurance premiums and insurance density. is the dependent economic development variable, which we seek to explore its movements against insurance sectorial developments.  is the error term.Ideally, the error term captures the residuals from the WR model, which are other factors contributing to economic development but excluded from our current model.
We aim to examine if we can find any strong interdependencies between economic development and the insurance sector where the economic growth measurements are compared.In case of failure of the radiance emissions generated somewhat more robust correlation coefficients, the radiance light emission is used instead of the saturated light data as the dependent variable or our economic measure in all regressions below.Also, GDP and GDP per capita shall be used as control variables.Our regressions are run in the log-log functional form to cater to contemporaneous correlations.

Insurance Contribution(%)
which in this case is Gauteng Province.The table provides us with necessary information on the role of the insurance sector on economic growth before running our in-depth regressions.

Results and Findings
We start by summarizing statistics on the main variables included in this study.We reported each variable's mean, standard deviation, and minimum and maximum values.the skewness of the data either positively or negatively, and it deviates from normality and is true otherwise.Thus the data variable with a small mean is considered with a small corresponding standard deviation.GDP/Capita and Radiance light are used and compared accordingly.

Correlation analysis
Below is a matrix for the correlation coefficients across all the variables used in this paper.When measuring insurance presence and density against either GDP/Capita and Radiance light (our core variables of interest), we observe that insurance density and presence are more highly correlated to radiance light than GDP/Capita, indicating its proxy efficiency in measuring economic activity.Although GDP/Capita and radiance light tend to move in the same direction, we further investigated it using the scatter plot presented below.A simple informative scatter plot above depicts some contemporaneous correlations between GDP/Capita and Nighttime light intensity in South Africa.Although there is a positive correlation between the two, nighttime light intensity (radiance light) will soon prove to be a proxy when measuring economic growth.Meanwhile, the nighttime lights are related to the true latent GDP/Capita through an unknown production function (•) and an additive error term.

Nighttime lights data versus GDP per Capita
Mathematically, this can be presented as below: , = ( , * ) +  ,  (  ) (3) The error term distribution above varies with the geographic locations; see Hu and Yao (2019).
The data inform the specification of the production function (•).Logarithmic transformation is applied to the GDP data to enhance a good fit.We regress GDP/Capita on nighttime lights for South Africa (its provinces) with year dummies.We conjectured that Nighttime lights real GDP per capita.

Weighted Regression analysis
We ran weighted regressions, incorporating all our variables, with radiance light as the dependent variable.Ordinary least squares (OLS) add less value to geographically weighted data and are more or less similar to the results obtained from the above correlations.Also, weighted regressions help control multicollinearity problems among our variables.For example, the variable insurance presence is omitted and dropped out of the model due to collinearity.In this study, we assume our WR model to be Where () is a weight-specific matrix to location , so that observations near to  are given greater weight than the remote ones,  is the matrix of our regressors/independent and predetermined variables. is the dependent variable matrix.The results are in the table below.
Table 5: All regressions are in the log-log format.
Only GDP was used to measure economic growth, which forms our paper's novelty.Application of nighttime light data appeared in literature but in other versions.For example, Zhang and Guo et al. (2019) compared the LJ1-01 nighttime light data with Visible Infrared Imaging Radiometer Suite (VIIRS)data in modeling socio-economic parameters.They used ten parameters from the four aspects of the gross regional product (annual average population, electricity consumption, and area of land in use) were selected to build linear regression models from the selected regions of China.The results showed that nighttime light data offered a better potential for modeling socio-economic parameters than the comparable VIIRS data; the former can be an effective tool for establishing models for socio-economic parameters.In addition, Mellander and Lobo et al. (2016) used a combination of correlation analysis and geographically weighted regressions to examine if light can function as a proxy for economic activities at a finer level.They used a fine-grained geo-coded residential and industrial full sample micro-data set for Sweden and matched it with radiance and saturated light emissions.Their results showed a correlation between NTL and economic activity and are strong enough to make it a relatively good proxy for population and establishment density.
We extracted a photograph illustrating the nighttime lights in Africa, as infigure1.The figure is clearly showing the light concentrations on a geographical basis.The light concentration is used to trace and measure economic activities.People use light to produce, construct and live in the modern world.The patterns of light, and the movement in these patterns over time, can guide us as to where people choose to live and produce and how their choices change.See below.Light concentration and patterns on a geographical basis in Africa.

Figure 4 :
Figure 4: Scatter plot for Radiance light intensity and GDP/Capita based on log transformations.

Table 2 :
Proportion of variation in the insurance industry at the provincial level in South Africa

Table 3
shows the summary statistics of all the variables used in this study.N is the number of sample size.Relative to radiance light, saturated light has high mean and standard deviation values of 18.95 and 21.34, respectively, whereas the corresponding values for radiance light are 12.5 and 17.31.Radiance light and GDP/Capita shall be compared since GDP has a greater mean and standard deviation of 2.13 and 5.91, respectively, than GDP/Capita.A significant mean implies

Table 4 :
Correlation matrix for the variables used in this study.CommentsWe examined if there exist any relationships among all our variables of interest.We employ both Radiance light as well as Saturated light in the correlations to be able to examine possible differences.From table 2 above, the pair correlations across all the variables are strongly positive.