Frontend Innovation And Top Income Inequality: Evidence From Emerging Markets


 This paper contributes to the literature on income inequality, by extending existing models to examine the effect of front-end innovation (FEI) on top income inequality. We use a fixed effect panel regression, on annual country level data for twenty four emerging markets, over a twenty four (1995–2018) year period, and find an insignificant correlation between income inequality, and FEI. The instrumental variable estimates however, shows a significant association between measures of FEI and top income shares. Further, we confirm that FEI is weakly related with broad measures of income inequality. Our instrumentation strategy, and robustness checks, suggests that this correlation partly reflects a causality, from FEI to top income inequality. Finally, we show that FEI is necessary for the survival of new ventures, in the crucial early years. Overall, our findings confirm that FEI, is a significant determinant of increases, in entrepreneurial income share.

Pollak (2014) presents a model whose central idea is that incumbent establishments, protect their competitive position, by innovating at a rate that is high enough to prohibit the entry of new businesses.
Subsequently, R&D investment now depends upon the ease of entry in individual markets rather than expected future pro ts. This development can stall the process of creative destruction and exacerbate income inequality. Similarly, Aghion et al., (2019) use cross state panel data, to demonstrate a signi cant positive association between innovativeness, and top income inequality in the US. They nd that causality runs from innovativeness to top income inequality. Further, when measured by patent per capita innovativeness, on average is responsible for seventeen percent of the total increases in the top one percent income share.
In the same vein, Zweimuller (2000) highlighted the impact of inequality on innovation driven growth. A redistribution from consumers who can afford the goods produced by the most recent innovator, to those who cannot afford it leads to an increase in the economic growth rate. Aghion et al. (2015) present a Schumpeterian growth model, which predicts that entrants, and incumbents' innovation raises top income inequality. Entrants innovation increases social mobility. Entry barriers reduces the positive effect of entrants' innovation on top income inequality. Also, Galor and Zeira (1993) developed a model where wealth distribution, plays an important role in economic growth overtime. Further, wealth distribution is a signi cant in uencer of macroeconomic adjustment to aggregate supply, and demand shocks as well as technological innovation. Gabaix and Landier (2008) propose a simple, competitive and exible model that attributes the rise in Chief Executive O cer (CEO) pay, to substantial growth in equilibrium rm size. Executives have different levels of managerial talent that are matched to rms competitively. The marginal impact of a CEO talent is assumed to rise with the value of the rm he manages. The dispersion of executive talent distribution appeared to be extremely small at the top. It is this dispersion that yields large disparities in compensation amongst CEOs, as they are magni ed by establishment size.
Lloyd-Ellis (1999) present a model where wage inequality rises when the rate of new technology introduction is not equal to the rate of absorption due to the aggressive competition for scarce labour and other resources.
This situation causes the wage of highly skilled labor to drive up the cost of innovation therefore, precipitating a decline. Likewise, Rehme (2007) asserts that increases in the number of highly skilled people nanced by a rise in taxation, rst increases inequality because of additional productivity before reducing inequality in wages and personal factor income. Additionally, internal or process innovation scales moderately faster with rm size than new product innovation. Small rms have comparative advantage in new product development over older rms as older rms already have established product lines to work on. Therefore, new entrepreneurial rms contribute disproportionately to radical innovation (Akcigit&Kerr, 2018). Caselli (1999) present a model where skill based technological revolutions means that learning investments required by new machines are greater than those needed by existing ones. Hence skill based revolutions trigger the reallocation of capital from slow to fast learning workers, thereby depressing the absolute wages of slow learning workers.

EMPIRICAL REVIEW
Several empirical papers support the link between innovation and income inequality. Pavcnik and Goldberg (2007) avow that globalization rises contemporaneously with inequality in most developing nations. The reasons why globalization increases inequality is due to trade induced skill based technical change where investment will raise the demand for skilled workers. Also, the outsourcing of intermediate goods production raises the demand for skilled labour. Equally, Herzer and Vollmer (2012) investigated the long run effect of income inequality on per capita income for forty six countries and nd that inequality has a long run negative impact on per capita income in both rich and poor countries.
Jaumotte, Lall and Papageorgiou (2013) studied the causes of rising inequality in a panel of fty one advanced and emerging markets and conclude that technological change signi cantly impacts inequality.
This result is because innovation favors high skilled workers and therefore aggravates inequality especially in developing countries where a signi cant skills gap exists due to poor access to quality education. Similarly, most aggregate productivity growth among US non farm private businesses comes from incumbents rather than entrants. Moreover, quality improvements and not new products drive most of the growth (Garcia-Marcia, Hseih&Klenow,2019). Risso and Sanchez-Carrera(2019) used a panel data set of seventy four countries over a period of eighteen years to establish a link between inequality and innovation. Long run economic growth reduces income inequality and economic growth depends upon technical change. Further, Acemoglu&Robinson(2012) assert that innovation (R&D activity) will not reach a level that can impact economic growth without the development of economic and political institutions. Wlodarczyk (2017) discovered that a higher gross domestic expenditure on R&D as a percentage of GDP increases inequality in Europe. While a high number of patent applications suppresses it. Using panel data from twenty nine European countries, Benos and Tsiachtsiras (2019) discovered that innovation reduces personal income inequality. Permana, Lantu and Suharto (2018) used a panel of twenty eight European countries to establish a positive link between innovation, patenting activity and inequality, including the top ten percent share of the richest. They also nd a positive link between technological specialization and income inequality. Equally, Akcigit, Grigsby and Nicolas (2017) a rm that innovative places are more socially mobile and successful patentees have a higher labour income. Liu and Lawell(2015) used panel data over a ve year period to examine income inequality in China. They reveal that small doses of innovation can diminish inequality while, large doses can expand it. Li, Squire and Zou (1998) analyse data from forty-nine developed and developing countries and nd that initial secondary schooling, and capital market imperfection are signi cant drivers of inequality. By making the poor unable to access credit markets and, invest in education it perpetuates a low and inequitable growth process In Italy, an average inventors earnings is strongly linked to patent activity. An inventors wage starts rising a few years before patent applications are made to the patenting o ce, it peaks in the year preceding ling and decreases again (Depalo&Addario,2014). Likewise, Bell, Chetty, Jaravel, Petkova and Van Reenen (2019) prove that the top ten percent of inventors obtain more than twenty two percent of total inventors income.
This results imply that even among innovators, income inequality exists as in the general population.

INEQUALITY AND INNOVATION
Hatipoglu (2012) a rm that a decline in inequality, may produce a rise in the number of customers who can buy new products. This change in inequality, can impact an inventor's expected pro ts and their decisions about R&D expenditure. Conversely, due to a price and market size effect Tselios (2011) avow that, an increase in inequality encourages innovation in the EU.
Although these papers con rm the potential endogeneity problem between innovation, and inequality the dynamic is different in emerging markets. What we know is that a decrease in inequality, is unlikely to in uence FEI in emerging markets. The lack of association is because higher economic growth rates, re ects a higher demand for goods and services, which creates the opportunity for innovative driven entrepreneurs (IDEs), to start a business. Subsequently, necessity driven entrepreneurs (NDEs) who form most small rms in emerging markets, are not impacted as these individuals are in uenced by poverty, and unemployment rather than demand (Van Stel, Caree&Thurik, 2005).

HYPOTHESIS
The theoretical, and empirical studies stated above highlights the fact that innovation signi cantly impacts income inequality. However, even though successful innovation largely depends on FEI (Gama,Parida&Frishammer,2019), the link between FEI, and income inequality in emerging markets is yet to be studied. The foregoing leads us to hypothesize that: There is a signi cant relationship between top income inequality and FEI.
FEI reduces top income inequality.
The activity of entrepreneurs who engage in FEI signi cantly increases top income shares.
FEI is not signi cantly correlated with broader measures of income inequality 3. Data And Methdodology 3.1 DATA Our measures of income inequality[1], Gini Index,pre tax income shares of the top ten and one percent was drawn from the standardized world income inequality database (Frederick Solt,2019), the world inequality database and the world bank database.
Our novel measures of FEI comes from the world bank database. We chose the following measures of value added for FEI: services, industry and manufacturing. The value added component of these sectors were selected because rising creation and diffusion of knowledge is a basis for boosting domestic value added.
Local creation and diffusion of knowledge is a key innovation system variable, therefore building local knowledge or innovation is critical for value addition and eventual integration into the global value chain (Egbetokun, Siyanbola&Oyewole, 2011;Lee, Szapiro&Mao, 2018;Minetti, Murro, Rotondi&Zhu, 2019;Reddy, Chundakkadan&Sasidharan,2020). Likewise, FEI enables an entrepreneur to add value to their invention. After value addition the entrepreneur will be able to reach economies of scale and access nance [2]. In the emerging markets context entrepreneurs who engage in FEI will be able to access credit via commercialization after value addition. (Pereira, Ferreira&Lopes,2017;Cosci,Meliciani&Sabato,2016;Depalo&Addario,2014;Toivanen&Vaananen,2012;Hsu&Ziedonis,2008).
We have selected adjusted savings: education expenditure (% GNI)[3] as a robustness check and an alternative measure for FEI. There are studies that con rm that educated entrepreneurs run innovative ventures (Goedhuys&Sleuwaegen, 2010;Marvel&Lumpkin,2007). However, we consider education a crude measure for FEI because, unequal access to quality education has increased inequality in developing countries, where only the rich continue to access premium education. Thus, instead of education serving as the primary engine for upward mobility it increases inequality instead (IMF,2016).
For this reason we use the value added components of the economy to measure FEI.
Manufacturing value added, (% of GDP) Services value added, (% of GDP) Industry (including construction) value added, (% of GDP) Adjusted savings:education expenditure (% of GNI) -Alternative measure for FEI In our sample we see that Brazil, Indonesia and Iran had the highest SVA, MVA and IVA (in 2017, 2002 respectively. Indonesia, Nigeria and Kenya had the lowest SVA, MVA and IVA (in 2000, 2010and 1998. Thailand extended the highest credit (166.5% in 1997) to the private sector by banks while, Ukraine (1.38% in 1996) extended the lowest credit. Malaysia had the lowest in ation rate (-5.99% in 2009)  The descriptive statistics for all the variables used in this study are presented in Table 2. From our sample, GDP Per capita has increased during our study period, while the top income shares (top1% and 10%) has not changed signi cantly. In Table 3 we see that mean income inequality (top1% and 10%) on average rose slightly between 1995 and 2015.  (Beck, Demirguc-Kunt&Maksimovic, 2008).
Thus, the banking sector signi cantly impacts rm level innovation, and helps inventors increase their incomes. The next variable is general government nal consumption expenditure (percent of GDP), in order to control for the effect of government size in each country. Empirical studies nd that government size is associated with an increase in inequality, in emerging markets (Odedokun&Round, 2004;Anyanwu, 2011).
Next we include civil war in our analysis because, emerging markets are normally characterized by civil con ict due to rapid economic growth, poverty and emerging institutions. Moreover, due to disruptions in product and factor markets, increased transaction costs, reduction in social spending, and the general disruption in economic activity, civil con icts can adversely impact income inequality (Bircan, Bruck &Vothknecht, 2017).
Next we include the globalization index, because the countries in our paper had pursued economic liberalization policies during the study period. The globalization index measures globalization along the economic, social and political dimension as well as Foreign Direct Investment, FDI as a percentage of GDP.
Empirical results regarding FDI and inequality is mixed. The IMF (2007) nds that nancial globalization especially FDI, is associated with higher inequality in developing countries. While, Jaumotte et. al., (2013) nd that trade globalization reduces income inequality, and Claus, Martinez-Vaquez and Vulovic (2012) also nd that globalization has a positive impact on inequality. We include In ation because of its relationship with inequality as discussed by Albanesi (2007) and Siami-Namini and Hudson (2019).We end up with a balanced panel of twenty four countries over a twenty four year period (1995-2018).

THE SAMPLE
The sample consists of twenty four emerging markets (Brazil, Chile, Colombia, Egypt, India, Indonesia, Iran, Kenya, Mexico, Morocco, Malaysia, Nigeria,Romania, Russia, Pakistan, Peru, Philippines, Poland, South Africa, Ukraine, Thailand, Tunisia, Turkey and Hungary) only. Our reasons for focusing solely on emerging markets are twofold. Firstly, FEI is required for small rms to access credit in emerging markets, and these countries have a nuanced basis for FEI and inventions when compared to developed countries. This nuance is due to contextual factors, such as political and economic institutions. Secondly, by focusing on emerging markets which have similar characteristics it is possible to reduce sample heterogeneity. The study period runs from 1995 -2018[7]. The analysis period provides signi cant insight about the relationship between inequality and FEI.
However, in contrast to developed countries time-series data on FEI, and income inequality is limited. The data constraint is because traditional measures of innovation such as R&D expenditure and patents are less likely to be observable in small rms (Acemoglu, Aghion&Zilibotti, 2006) and emerging markets (Gorodnichenko&Schnitzer, 2013;Kraemer-Mbula et al., 2019). Moreover, governments across middle Africa do not prioritize the collection of innovation data. There are also reservations in the literature regarding the suitability of R&D, as an innovation proxy in emerging markets. Not all innovation is caused by R&D spending, and formal innovation measures are valid for large rms only. Possible alternative quality measures for FEI such as rms using banks to nance investment (% of rms), total SME loans (% GDP), government expenditure on education, total (% of GDP) were not available for at least a quarter[8] of the countries in our sample. Also, traditional measures for innovation such as patent applications, and R&D spending were mostly nonexistent for countries in middle Africa. Therefore, analysis was con ned to countries for which income inequality, and FEI data was available.

ESTIMATION METHODOLOGY
Our estimation method is comparable to that of Aghion et. al. (2019). The resulting model is based on a xed effect panel regression with standard errors, clustered at the country level. We clustered errors at the country level to allow for the possibility of correlation between the observations within a given country. Clustering is also required to account for situations where observation within each country, are not independently and identically distributed. We take the logs of all measures of FEI and Inequality, GDP Per Capita and bank credit to private sector, and estimate the following equation: Ln (Y it ) = η +θ i + λ t + α i ln(FEI i(t-3) ) + β 2 X it + ε it Where i stands for country i, t stands for time period t , Y it is the measure of inequality (in log), η is the constant term, θ i, λ t correspond to country and year xed effects. FEI it-3 is the measure of innovativeness at the front-end (in log and lagged) and X are the control variables. The advantage of taking the logs of FEI, and inequality is that α i can be interpreted as the elasticity of inequality, with respect to innovation. Our measures of FEI were lagged by three years. All equations were estimated using auto-correlation and heteroskedasticity robust standard errors. By including country and time xed effects, we are eliminating permanent crosscountry differences in inequality, and overall changes in inequality.

THE TIME LAG BETWEEN FEI AND TOP INCOME INEQUALITY
The reason for lagging[9] FEI by three years, is because over fty percent of new rms in developing countries collapse in the rst three years (Liedholm, 2002;Beck&Cull,2014, Bowale&Akinlo,2012Mead&Liedholm,1998). Because the new rm casualty rate peaks in the third year, the few who survive the rst three years and beyond are engaging in FEI, and will access institutional nance, which is equivalent to a patent grant in advanced economies. Similarly, in the US the time lag between a patent application, and a patent grant is two and half years on average. Empirical studies also show a signi cant relationship between patents, and access to venture capital nance especially for new rms (Audretsch,Bonte&Mahagaonkar, 2012;Hsu&Ziedonis,2008;Conti, Thursby&Thursby, 2013). The process is the same in emerging markets [10], where new rm creation is equivalent to a patent application, and access to nance is equivalent to a patent grant. However, in emerging markets the journey form new rm creation till access to credit is FEI. Just like the patent application process enables innnovators to reveal information about their invention, which then makes it eligible for nance. Equally, FEI does the same thing for new rms in emerging markets, and the average time lag between new rm creation, and access to credit in developing countries is three years (Beck&Cull,2014;Liedholm, 2002). In emerging markets, it is expected that an entrepreneurs' income typically starts rising from the point of new venture creation. New venture creation, automatically puts the rm on the path to commercialization because, the innovator is adding value to his product, and readying it to scale, and access nance. This process (FEI) implies that he is prototyping/commercializing, or already running a large rm but, on a much smaller scale. For these reasons, the entrepreneurs income will rise prior to accessing credit.

TOP INCOME INEQUALITY AND FEI AT DIFFERENT TIME LAGS
In emerging markets over fty percent of new rms collapse in the rst three years (Liedholm,2002;Beck&Cull,2014;Bowale&Akinlo,2012;Mead&Liedholm,1998), and over ninety percent go bankrupt by the fth year. The few rms that remain after the third year are engaging in FEI, and will go on to scale and access nance. However, the fewer rms who still exist in the fth year, and beyond are even more likely to access credit, than those who survive the third and fourth years. Speci cally, since the few rms who survive the rst ve years, and beyond have the highest propensity to (via FEI) access nance, there is the need to extend the analysis up to a seven year lag. The seven year lag gives enough time to explain the dynamic between FEI, and the entrepreneurs income. Moreover, the new rm casualty rate declines from the third year, and onwards. For these reasons, it can be argued that the three year lag on the value added, is not enough to explain the relationship between FEI, and the entrepreneurial income share. To x this problem, and thus, test the robustness of our results we analyse the impact of alternative lags of FEI on top income inequality (top 1% only). We regress top income inequality (OLS regression) on alternative lags of FEI, and we allow the time lag to vary from three to seven years.

INSTUMENTAL VARIABLE STRATEGY
As we pointed out in section 2.2.1 inequality can also boost innovation, which underscores the potential endogeneity problems in our model. However, due to the ubiquity of NDEs in emerging markets, we argue that causality runs from FEI to income inequality. Moreover, measurement bias and omitted variable issues can be checked with the appropriate instruments [11] . Our IV strategy, relies on exogenous shocks which signi cantly raises the incentive for entrepreneurs to access credit (via FEI) for expansion. We introduce novel measures for our instrumental variables[12]: a) statistical capacity score (overall average) which is a metric that is used to assess the capacity of a country's statistical system and, b) the log of "households and NPISHs nal consumption expenditure (% of GDP)", which is the market value of all goods and services including durable products purchased by households and non pro t institutions at the country level.

THE EXOGENEITY AND TIMING OF THE INSTRUMENTAL VARIABLES
We chose the statistical capacity score because small rms, normally cannot access credit due to a lack of information about the viability of their venture. However, the literature on relationship lending, con rms a robust relationship between innovation, and the availability of information which increases credit access for small rms (Herrera&Minetti,2007;Peng,2017;Petersen &Rajan, 1994 and1995). Moreover, recent evidence from developing countries, con rm that the availability of information, via the introduction of credit registries improves the e ciency of credit allocation decisions, increases rm level access to nance and nancial sector development (Tchamyou, 2019;Tchamyou&Asongu, 2017;Ayyagari, Juarros, Martinez-Peria&Singh,2016). Therefore, the statistical capacity score, is a good metric to gauge the ability of new rms (in terms of the availability of information for lenders who seek to extend credit to new rms) to engage in FEI. The statistical capacity of developing countries is largely funded by external development partners [13], and is thus, exogenous to a nations' economic situation. We lagged the statistical capacity score by two years because, it is a metric that progresses with technological and economic development (Anderson&Whitford, 2017). The higher the level of technological advancement, the larger the score, and the greater the chances that a small rm will access credit. This process can only happen in time thus, in the emerging markets context the statistical capacity score will have a delayed impact on FEI.
Equally, we selected households and NPISHs nal consumption expenditure, as our second instrument, and a robustness check because, a change in household consumption expenditure can produce an increase or decrease in the number of customers who can buy new products (Foellmi&Zweimuller, 2016 andHaptipoglu,2012). This change in market size and price effect can impact the entrepreneurs expected pro ts, and ultimately the decision to engage in FEI. The households and NPISHs nal consumption expenditure, is largely driven by FDI in emerging markets. FDI spurs host country entrepreneurial activity which in turn boosts incomes, and then consumption via job creation (Jaumotte et al.,2013). Since FDI occurs when a business from one nation invests in another, it implies that shifts in household consumption expenditure is exogenously determined (Ghebrihiwet, 2019; Reyes,2017). Moreover, the impact of any change in household consumption spending on inequality, crystallizes via FEI and entrepreneurship (Munemo,2018;Durham, 2004). The households and NPISHs nal consumption expenditure, is lagged by two years because, entrepreneurs who engage in FEI are usually driven by demand (Hatipoglu,2012). Therefore, the household consumption, should impact the entrepreneurs decision to access credit with a lag. The timing between the start of FEI, and access to credit is also an important determinant of the lags on both instruments (Liedholm, 2002).
Although charges for intellectual property, receipts (BoP, current US$) and spillovers are also good instruments for FEI (Aghion et al., 2015 ;Chu and Cozzi, 2018; Coe, Helpman&Hoffmeister,1997) we did not use them because of their limitations [14] in developing countries. Further, the data is largely unavailable for some African [15] countries in our sample.
[1] Our main measure of inequality is the top 1% while, the top 10% and gini index are alternative measures and robustness checks.
[2] In the context of FEI in emerging markets new rm creation is equivalent to a patent application while, access to credit is comparable to a patent grant in developed countries. The entrepreneur experiences a signi cant rise in income after access to credit (Depalo&Addario,2014;Toivanen&Vaananen,2012;Hsu&Ziedonis, 2008). Unfortunately, the majority of small rms in developing counties, can not access credit due the inability to engage in FEI. Majority will collapse in the rst three years, and out of the few who survive, only a minority will go on to scale and access credit (Wennekers, Van Stel,Thurik&Reynolds, 2005).
[3] Although there is insu cient data on government expenditure on education,total (% of GDP) the authors assert that the adjusted savings: education expenditure (% of GNI) mirrors the pattern of the former and thus, is a reliable measure for education spending in these countries.
[4] Each country (Brazil, Kenya, India and Romania) was selected from a different continent(South America, Africa, Asia and Eastern Europe) in order to understand the underlying trends in FEI and inequality data regardless of location. The data show a similar trend in FEI in emerging markets.
[5] The top one percent income share rises with the value added because, because emerging markets have pursued policies of economic liberalization, which has increased the income disparity between owners of capital, and labor as well as between the skilled and unskilled (Zhuang, Kanbur&Rhee,2014;IMF,2016;Jaumotte et al., 2013) [6] Due to lack of data we used domestic credit to the private sector(% of GDP) for Egypt, India and Kenya [8] Nigeria, Pakistan, India, Kenya, Tunisia, Egypt, Morocco among others [9] The entrepreneurs income increases signi cantly after the rm has reached economies of scale and is thus, able to access institutional nance -the FEI process. Speci cally, signi cant increases in value addition and education expenditure will precede the innovators rise in income (Herrera&Minetti, 2007;Gama&Parida,2017;Depalo&Addario,2014;Toivanen&Vaananen,2012). https://www.lbs.edu.ng/lbsinsight/why-smes-fail-and-its-impact-on-national-development/ [10] Radical technological change in developing countries is mainly captured by new rm creation, and not intellectual property like patents (Srholec, 2011;Kraemer-Mbula et al., 2019) [11] The correlation coe cient between both instruments is 0.08 [12] For developing countries, the statistical capacity is usually funded by external development partners https://statisticalcapacitymonitor.org/pdf/Statistical%20Capacity%20Development%20Outlook%202019.pdf.
NPISHs are institutions which provides goods and services to households for free or below market prices. They mainly derive their income from grants and donations and are not controlled by the government.
[  [14] Radical technological change is captured by new rms entering the market rather than intellectual property like patents. Moreover, Intellectual property development is discouraged in emerging markets due to institutional factors like poor legal rights (Vivarelli&Quatraro, 2015;Srholec,2011;Kraemer-Mbula et al., 2019). Developing countries with low entrepreneurial human capital, and a poor business environment will not bene t signi cantly from spillover effects (Gorg&Greenaway,2004;Durham,2004) [15] This data is not available for Nigeria among others.

Panel Ols Regression Results
In this section, we present the results from our panel regression analysis, of income inequality on FEI. We rst report the relationship between top income inequality, (top1%) and FEI. Next, we examine the relationship between FEI, and broad measures (top 10% and gini index) of inequality. Finally, we examine the correlation between FEI and top income shares at various time lags. We have taken the logs of all measures of FEI, and Inequality so that we can interpret the coe cient of FEI as an elasticity. A one percent increase in IVA reduces inequality by 0.0718 percent. Likewise, one percent rise in the SVA, MVA and education expenditure will increase income inequality by 0.1669, 0.0788 and 0.0571 percent respectively.

FEI AND BROAD MEASURES OF INCOME INEQUALITY
We now report the results of the same regression from section 4.1 but, using broad measures of inequality as the dependent variable (top 10% and gini index). We report the results of these regressions in tables 5 and 6. We obtain similar results when we use the top 1% income share. Columns one through four reproduces the results for MVA, SVA, IVA and education expenditure for top 10%, and then the gini index (tables 5 and 6). Again for the top10%, apart from IVA which has a negative impact, the results are always positive and insigni cant. A one percent increase in IVA will reduce inequality by 0.0285 percent. Equally, a one percent increase in the MVA, SVA and education expenditure will boost income inequality by 0.0246, 0.0999 and 0.0242 percent respectively. Next for the gini index , we see that all four measures of FEI are statistically insigni cant. Again, we see that apart from SVA, all other measures of FEI has a positive correlation with inequality. A one percent rise in the SVA, decreases the gini index by 0.0145 percent. While a one percent increase in the MVA, IVA and education expenditure will increase inequality by 0.0628, 0.0158 and 0.0436 percent respectively.

FEI AND TOP INCOME INEQUALITY AT DIFFERENT LAGS
In this section, we test the robustness of our results to alternative lags for FEI.  Tables 12, 13 and 14 shows the results of the same regression when all lags are included in a single equation. The time lag between top income inequality, and FEI are allowed to vary from three to seven years. These tables shows that the coe cient on lagged FEI[2], remains signi cant up till the seventh year. Moreover, the coe cient also rises with each lag. As discussed in section 3.5, this development supports our assertion, that rms which survive beyond the rst three years, are engaging in FEI, and will most likely access credit (Liedholm, 2002;Beck&Cull,2014). Equally, in the all lags equations, we see that the coe cient on lagged FEI, although insigni cant (except for the seventh lag for SVA in table 13) also increases with time.
Again, this nding proves that rms which survive the critical rst ve years, are engaging in FEI and, will access credit.

TRUE FEI VERSUS NECESSITY ENTREPRENEURSHIP
The correlations we nd so far between our variables of interest -the top 1% and value added -are insigni cant across the board. Value added is a good proxy for FEI, for a couple reasons. First, due to the ubiquity of NDEs, a majority of rms in emerging markets remain small, and or never get to access credit (Abraham&Schmukler, 2017). Second, not all value added will impact the economy positively. Infact when rms fail, and or remain small due to the inability to engage in FEI[3] and access credit, this occurrence increases the interest rate on loans, to new rms and thus, raises inequality. For these reasons, the relationship between the value added, and inequality is weak. However, two[4] reasons leads us to conclude, that the correlation between value added and, inequality also include the activity of entrepreneurs who engage in FEI: 1) FEI when measured by MVA, SVA and education expenditure, exert a positive in uence on top income inequality; 2) the correlation between income inequality and, FEI is insigni cant for all measures of FEI. Which suggests that this association goes beyond the activity of just NDEs.

RESULTS FROM THE INSTRUMENTAL VARIABLE (IV) ESTIMATION (top 1%)
In

SUMMARY OF FINDINGS
The results of our IV, and OLS regressions are largely in line with our hypothesis. First, FEI measured by IVA, and MVA has a signi cant impact on the top 1% income share. Second, the activity of entrepreneurs who engage in FEI, increases top income inequality. Third, FEI when measured by SVA reduces inequality. Fourth, FEI is not signi cantly correlated with broad measures of income inequality. Finally, rms that survive the rst three years and, beyond are engaging in FEI and will access credit.

ADDITIONAL FINDINGS AND DISCUSSIONS
We nd evidence of the existence of Kuznets curve in our sample (see illustration from tables 4, 5 and 6 in gures 5, 6 and 7). Kuznets (1955) proposed that inequality, may rise with initial increases in GDP Per Capita but decline subsequently. Our results, indicate that the relationship between GDP Per Capita and, inequality is inverted U shaped when the top 10% and, gini index are used to measure inequality. While the relationship between both variables is U shaped, when the top 1%[20] is used to measure inequality. These results indicate that the Kuznets curve exists in emerging markets, for our broad measures of inequality, and vice versa for the top income inequality.
The results from table 8, show that civil war, globalization index, and government consumption expenditure exert the largest[21] magnitude of effect on top income shares. While, FDI and in ation has the smallest impact. We nd that markets where most entrepreneurs engage in FEI will have less con ict [22] or MEPV.
Equally, government expenditure[23] is e cient where the practice of FEI is widespread. The globalization index[24] also reduces inequality, but its impact is stronger when FEI is signi cant. FDI increases inequality in sophisticated sectors like manufacturing, and industry where it is harder and, complex for entrepreneurs to scale, and access credit. The opposite occurs in a less sophisticated segment like services [25], where the process of scaling, may be less di cult for entrepreneurs. In African emerging markets, entrepreneurs who operate in the education, healthcare sectors and wholesale, and retail trade nd it easier to access credit, when compared to their manufacturing, and industry segment counterparts. In ation has the smallest magnitude of effect on income inequality. This is partly due to the inverted U shaped relationship between both variables in developing countries, where rising incomes [26] has been able to absorb the negative in uence of in ation (Siami-Namini & Hudson, 2019). The relationship between inequality, and domestic credit to the private sector by banks is insigni cant. This incidence is because access to credit for small rms, is largely unavailable in emerging markets (Beck, Demirguc-kunt&Levine, 2007).
[1] The panel xed effects results captures the impact of entrepreneurs who engage in FEI and those who do not. The NDEs are larger than the entrepreneurs who engage in FEI hence, the statistically insigni cant result Amoros, Cristi&Minniti,2011). Likewise, the enterprise of IDEs increases inequality.
[2] The direction of effect is positive for education expenditure and SVA and, negative for MVA and IVA. These signs supports the outcome of our IV results which shows that FEI can increase or decrease inequality based on the measure.
[3] NDEs who form the majority of entrepreneurs in emerging markets usually fail and or remain small thus, their impact on inequality is temporary (Amoros&Cristi, 2008).
[4] These results are also consistent for our alternative measures of inequality, the top 10% and gini index.
[5] It is generally di cult to access credit across the board however, It is less hard to access credit for entrepreneurs in the services sector, when compared to manufacturing, construction and industry segments.
This fact implies that the services sector is broad, and impacts a wider part of the population hence, the negative sign. SVA include services attached to wholesale and retail trade, education, healthcare, nance, hospitality, banking and insurance among others which has expanded quickly in emerging markets, due to a growing middle class. The service sector, accounts for the largest share of national income in Africa, and Asia emerging markets, and they make up half of the countries in our sample. https://www.adb.org/sites/default/ les/publication/31114/developing-service-sector-engine-growth-asia.pdf [6] IVA and MVA comprise sectors like manufacturing, construction, and mining. These are fairly sophisticated sectors in the emerging markets context thus, the know how about scaling, and access to credit is rare for these segments. Hence, only innovators in the top 1% will be able to scale, and access credit in these areas (Abraham&Schmukler,2017;Stenberg & Wennekers, 2005;Amoros&Cristi,2008). Aghion et al., (2015) also reveal a positive correlation between innovation, and the share of income owned by the top one percent in the U.S. Due to the growing demand for infrastructure and, real estate most new construction globally, will happen in emerging markets (https://www.pwc.com/gx/en/assetmanagement/publications/pdfs/real-estate-2020-pwc.pdf). Likewise, the manufacturing sector in Africa emerging markets, has not expanded as expected because most rms are unable to reach economies of scale (Liedholm&Mead, 1998;Liedholm,2002) [7] Education is not a primary engine of upward mobility in developing countries, rather it increases inequality, especially in the developing countries of Asia and Africa, which make up half of the countries in our sample. (IMF,2016 andJaumotte et al., 2013). The literature on skill biased technical change (Permana, Suharto&Lantu, 2018) con rm that the dominant reason for wage inequality is education. Highly educated people, can better use new technologies, and therefore will receive higher wages.
[8] A unit increase in the statistical capacity score raises both MVA and IVA by 0.6 and 0.5 percent respectively.
[9] At the end of FEI, the entrepreneur, produces a business model based on FEI, which include information about the pro tability of the venture to bankers. On the supply side, nancial institutions must use statistical information regarding the rm, plus the business model, developed by the innovator to guide decision making. However, this process is uncommon in developing countries, who are still growing their statistical capacity. Therefore, this action is restricted to few entrepreneurs who have the ability to scale (Anderson&Whitford, 2017;Pereira, Ferreira&Lopes, 2017;Beck, Demirguc-kunt&Maksimovic, 2005;Berger&Udell, 2002).
[10] Most entrepreneurs in developing countries are motivated by necessity rather than innovation hence, their impact on inequality is temporary (Amoros et al., 2011;Wennekers et al., 2005).
[12] Environmental factors in developing countries such as the rule of law and property rights, poor infrastructure, political instability, corruption and other institutional factors which dampen FEI (Beck et al., 2008;Cenni et al., 2015 andBanerjee&Newman, 1993).The activity of NDEs also dampens FEI and increases inequality.
[13] Results from the IV regression using both instruments show a similar outcome and, is available upon request from the authors.
[14] The gini index and top 10% are broad measures for FEI compared to the top 1%, it captures the impact of a wider range of entrepreneurs who engage in FEI. A number of empirical studies con rm a less robust relationship between innovation and inequality, especially when the innovators are not in the top one percent or when broader measures of inequality like the top 10% and the gini index is used (Wlodarczyk, 2017;Aghion et al., 2019;Antonelli&Gehringer, 2017).
[15] A unit rise in the statistical capacity score raises both MVA, and IVA by 0.6, and 0.5 percent respectively.
[16] A one unit rise in the statistical capacity score, increases the top 10% income share by 0.08% and, decreases the gini index by 0.04%.
[17] Brazil, Chile, Malaysia, Mexico, Russia, Poland, Turkey, Hungary and Romania were dropped from this analysis.
[18] Due to the effect of excluding advanced emerging markets from this sample, the level of signi cance for FEI (IVA and MVA) dropped to 5% from 1% in the original sample.
[19] A one unit increase in the statistical capacity score, will boost MVA, and IVA by one percent.
[20] The activity of the top 1% will increase income inequality in the long run.
[21] The large coe cients for the civil war, and globalization index is not surprising, considering the fact that only seven countries (Tunisia, Romania, Hungary, Poland, Morocco, Chile and Brazil) out of twenty four, have not experienced any MEPV during the study period. Equally, Apart from Iran, which has been under international sanctions, every country in our sample has pursued policies of economic liberalization/glocalization. Civil war occurs less and, the globalization index and government expenditure are e cient when the majority of entrepreneurs practice FEI. This incidence implies that the development of institutions, and FEI are related (Cenni et al., 2015;Banerjee&Newman,1993).
[22] Civil war increases top income inequality. But its magnitude is smaller in the IVA, and MVA equations or when FEI is signi cant.
[23] Government expenditure increases top income inequality. But its magnitude is smaller when FEI is robust. This nding implies that government expenditure is competent when the practice of FEI is pervasive.
[24] SVA has a larger magnitude when compared to MVA and IVA. This is because access to credit is less challenging in a less sophisticated sector like services, so that it can absorb FDI.
[26] All countries in our sample have at least doubled their GDP per capita, during the study period.

Conclusion
In this paper, we present new evidence about the nature of the relationship between front-end innovation and, top income inequality in emerging markets. We found a positive, and signi cant association between top income inequality, and front-end innovation when we use manufacturing, and industry value added to measure it. We also reveal an insigni cant relationship, when we use education expenditure, and services value added to measure front-end innovation. The connection between front-end innovation, and top income inequality gets more robust with longer lags. We further discovered that front-end innovation, is not signi cantly correlated with broad measures of inequality. Firms that do not collapse in the rst ve years, has the highest propensity to engage in front-end innovation.
We submit that this association, partly re ects a causal effect from front-end innovation, to top income inequality. Our approach, was to examine the causal effect of front-end innovation, on top income inequality.
This initial step, is necessary in order to quantify the impact of front-end innovation on top income inequality. Hence, our paper complements studies in microeconomics, and entrepreneurial development, which seek to discern why only few rms can scale, and access nance in emerging markets, and thus, its impact on income distribution, using value added data. This study can be extended, by analyzing the impact of frontend innovation on poverty, and economic growth in developing countries. Cross country rm level studies, that will explore the impact of front-end innovation, on rm development, and access to nance, is another avenue for continuing this study. Examining the supply side dynamic of front-end innovation, and a theoretical inquiry regarding the front-end of innovation, are also areas of further research.
Declarations Acknowledgement: We would like to thank Lauri Elliott, the founder of Conceptualee Resources, USA. Lauri has dedicated over fteen years of her life, using the methodology for FEI she designed, to help small rms access credit in developing countries. Her experiences provided the motivation for this paper. The authors also want to thank Nathan Lupton, Julia Wlodarcyzk and Abiodun Egbetokun for their insightful reviews.

COMPLIANCE WITH ETHICAL STANDARDS
Con ict of interest: The authors declare that they have no con ict of interest

Services Value
Added (% of GDP) They include value added in wholesale and retail trade (including hotels and restaurants), transportation, government, financial and professional services such as education, health-care and real estate. Also imputed are bank service charges, import duties and any statistical discrepancy noted by national compilers, as well as discrepancies arising from re-scaling.    Table 4 Income Inequality (Top 1%)               Notes: The data was estimated using panel data OLS regressions.* denotes significance at the 10 percent level, while ** and *** denotes significance at the 5 and 1 percent levels respectively;t statistics in parentheses was computed with heteroskedasticiy and auto-correlation robust standard errors. The dependent (top 1%) and independent (manufacturing value added -lagged by three to seven years) variables are in natural logarithm. Time span:1995-2018.   Notes: The data was estimated using panel data OLS regressions.* denotes significance at the 10 percent level, while ** and *** denotes significance at the 5 and 1 percent levels respectively;t statistics in parentheses was computed with heteroskedasticiy and auto-correlation robust standard errors. The dependent (top 1%) and independent (services value added -lagged by three to seven years) variables are in natural logarithm. Time span:1995-2018.   Notes: The data was estimated using panel data OLS regressions.* denotes significance at the 10 percent level, while ** and *** denotes significance at the 5 and 1 percent levels respectively;t statistics in parentheses was computed with heteroskedasticiy and auto-correlation robust standard errors. The dependent (top 1%) and independent (industry value added -lagged by three to seven years) variables are in natural logarithm. Time span:1995-2018.