Drivers of Employment Elasticities in Kenya

The relationship between output growth and employment elasticities has been of intense debate among many economists. Though there is no conflict between the two objectives, the question that arises is the rate at which employment growth responds to economic growth. The policy focus on employment in Kenya is manifested by the sheer number of employment targeted development plans and Sessional papers that have been formulated. Basically, all the policy documents developed by the government have premised employment creation on economic growth. The purpose of the study was therefore to determine the drivers of employment elasticities in Kenya. Empirical findings indicated that the first lag of employment elasticity, average wage, inflation rate, labour force participation rate, first and second lags of labour force participation rate, population density, first and second lags of foreign direct investment to be the short run drivers of employment elasticity. Empirical findings also indicated that exchange rate, foreign direct investment and population density were the long run drivers of employment elasticity in Kenya. The study recommends that policy measures to control inflation should be tightened and more efforts to attract foreign direct investment to be undertaken. The study further recommends that a stable exchange rate should be maintained. Lastly, the government should harmonize the salary scale framework to regulate the wages in the country. This could be realized through salary adjustments based on a periodical and systematic evaluation of wage parameters in the public sector and taking cognizance of the prevailing economic dynamics.

employment growth of 6 per cent yielding a total of 3.7million new jobs. However, an annual average of 511,000 new jobs were created in 2008-2012 against a target of 740,000 jobs per year (Republic of Kenya, 2013). During MTP II (2013-2017), the government targeted the economy to grow at 10 per cent and create an average of one million new jobs. However, an average of 826,600 jobs were created annually. The MTP III (2018-2022) targets to increase real GDP annual growth from an average of 5.5 per cent achieved over the 2013-2017 period to 7 per cent in order to support higher economic growth. It also aims to create over 6.5 million jobs over the Plan period. There is need therefore, to determine the drivers of employment elasticities in Kenya given the fact that the government has intensively attempted to use fiscal policy measures as instruments of employment creation.

Literature Review 2.1 Labour Demand Theory
The theory of labour demand has its roots from Marshall (1890) and Hicks (1930). The labour demand theory states that the demand for labour is a derived demand, since workers are hired for their contribution in the production of goods and services. A key feature of the theory is that price flexibility plays an important role in the correction of labour market disequilibrium and market clearing. Assuming that there are only two factors of production, the number of employee-hours hired by the firm (L) and capital (K), the aggregate stock of land, machines, and other physical inputs, the production function is written as: Where is the firm's output. The theory assumes that production exhibits constant return to scale as described by such that: > 0 , < 0 and > 0 Where represents marginal products. Elasticity of substitution is given by the rate of change in the use of to from a change in the relative price of to , holding output constant. Elasticity of substitution is by given: Where measures the ease of substituting one input for the other when the firm can only respond to a change in one or both of the input prices by changing the relative use of two factors without changing output. The constant output labour demand elasticity which is the change in demand for labour from a change in its wage is given by: is labour demand elasticity. Equation (3) implies that: = −(1 − $) < 0 (4) Where $, ($ = ) is the share of labour in total revenue. Equation (4) implies that when output requires substantial amounts of labour for production, the constant output labour demand elasticity will be smaller, because the possible change in spending on other factors is small relative to the amount of labour being used. Consequently, the constant output cross-elasticity of demand for labour describes the response to labour from a change in the price of capital is given as: = ! (5) Or, % = −(1 − $) > 0 (6) The scale effect is the factor's share times the product demand elasticity. The scale effect takes into account the possibility that output will change as a response to a change in the price of labour, and that in turn may affect the overall demand for labour. The scale effect which is the total response from a change in the wage is given as: Equation (7) is the fundamental factor law of demand. It divides the labour demand elasticity into substitution and scale effects.

Empirical Literature
Crivelli, Furceri & Bernate (2012) assessed the effect of structural and macroeconomic policies on the employment-intensity of growth for 167 countries using an unbalanced panel over the period 1991-2009. The objective of the study was to provide estimates of employment-output elasticities and assess the effect of structural and macroeconomic policies on the employment-intensity of growth. The study employed two approaches. The first approach consisted of estimating elasticities using time-series regressions for each country using the equation: ln(6 1 ) = 7 + 9 : ln(6 1;: ) + < : ln(= 1 ) + > 1 (8) Where 6 1 was the level of employment at time t, and = 1 was the level of GDP at time t, 7 was the intercept coefficient, 9 : , < : were the regression coefficients and > was the error term. The main advantage of the time series regressions was to directly provide country-specific employment estimates. The second approach relied on a panel framework in which long-term elasticities were estimated using country-specific estimates for GDP slopes and employment persistence using the equation: ln(6 1 ) = 7 + 9 : ln(6 1;: ) + 9 ? @ ln(6 1;: ) + < : ln(= 1 ) + < ? D B ln(= 1 ) + > 1 (9) Where 6 1 was the level of employment for country i at time t, and = 1 was the level of GDP for country i in time t. @ was a country-specific dummy, 7 was the intercept coefficient, 9 3 , < 3 were the regression coefficients and > was the error term. The study found that point estimates of elasticities fell between 0-1, with the majority of them ranging between 0.3 and 0.8. The elasticities varied considerably across regions, income groups, and production sectors with the highest estimates recorded for the most economically developed regions, industry and service sectors. The study also found that structural policies aimed at increasing labour and product market flexibility and reducing government size had significant and positive impact on employment elasticities. Macroeconomic policies aimed at reducing macroeconomic volatility also had a positive and statistically significant impact on employment elasticities. The study recommended that in order to maximize the positive impact on the responsiveness of employment to economic activity, structural policies have to be complemented with macroeconomic policies aimed at increasing macroeconomic stability. Slimane (2015) assessed the determinants of cross-country variations in employment elasticities mainly focusing on the role of demographic and macroeconomic variables. The study used an unbalanced panel of 90 developing countries from 1991 to 2011. The equation for a country's specific elasticities was given by: ln(6 1 ) = 7 + < ln(6 1;: ) + C ln(= 1 ) + > 1 (10) Where 6 1 was the level of employment at time t, = 1 was the level of GDP at time t, < DEF C were estimation coefficients, 7 was the intercept coefficient and > 1 was the error term. The study also estimated the long-term employment to GDP elasticities for each country through the equation: γ H B = 7 + I : J + I ? K + ' (11) Where γ H B was employment elasticity for country L, J denoted macroeconomic variables, S denoted structural variables, I : and I ? were the estimation coefficients 7 was the intercept coefficient and ' was the residual term.
The results of the study indicated that the elasticity estimates varied considerably across countries while employment elasticities were higher in more advanced and closed countries. Employment elasticity comparisons across countries revealed wide variation in employment elasticities with the highest estimates found in Comoros, Gabon, Cote d'Ivoire, Niger, Algeria, Madagascar and Togo. In contrast, employment elasticities were modest in other countries such as Bosnia (0.05), Ukraine (0.09) and China (0.10). Negative estimates were found for Serbia (-0.101), Belorussia (-0.112) and Romania (-0.238). The study also revealed that the macroeconomic policies aimed at reducing macroeconomic volatility had a statistically significant effect in increasing employment elasticities. Employment intensity of growth was found to be higher in countries with larger service sector and also for the countries with a higher share of urban population.

Methodology 3.1 Theoretical framework
To meet the objective of the study, labour demand theory was used. The labour demand theory is attributed to Marshall (1890) and Hicks (1930). The firm's production function describes the technology that the firm uses to produce goods and services. The firm's objective is to maximize profits that are given by: M = − (12) Where M is the output price, is the price for labour (wage rate) and is the price for capital. The first order conditions are given by: Where λ is a Lagrangian multiplier. The ratio of the two first order conditions shows that the marginal rate of technical substitution, , equals the factor-price ratio, , for a profit maximizing firm. Assuming a Cobb Douglas production function of the form: P = Q R S (15) Where P is the level of output, Q is technical efficiency, is the labour input, is capital input and 7 and < are elasticity parameters. The firms profit function is given by:   Equation 23 is the derived labour demand function expressed as a function of output, real wage and the price level.

Empirical Model
Following the labour demand theory (Marshall, 1890 andHicks, 1930), labour demand is a function of output, real wage and the price level as shown in equation 23. Relating labour demand to employment elasticity, the study improved on the model by Mouelhi and Ghazali (2014) by incorporating demographic factors. The demographic factors were included to assess the effect of agglomeration factors on employment elasticity. The estimatable equation was expressed as: 6Ja 1 = (LE bD 1 , 6cdℎ 1 , Oa6EE6$$ 1 , Df6 1 , aOaF6E 1 , bDg O d6 1 , FL 1 , ) (24) The variable 'Ja 1 was the aggregate employment elasticity for all the sectors in period h, LE bD was the annual inflation rate, 6cdℎ was the nominal exchange rate (Kenyan shillings/US dollar), Oa6EE6$$ was a proxy for trade openness, Df6 was the average annual real wage, aOaF6E was the population density, bDg O d6 was labour force participation rate and @i was foreign direct investment. Population density and labour force participation rate also gave the effect of labour supply on aggregate employment elasticity.

Definition and Measurement of Variables Employment elasticity (EMP):
It is the responsiveness of employment growth to economic growth and measured as the ratio of the relative change in employment to the relative change in output.

Inflation (INFLA):
It is the sustained increase in the general price level of goods and services in an economy over a period of time. It is measured by change in consumer price index (CPI). Real exchange rate (EXCH): It is the nominal exchange rates adjusted for differences in price levels between two countries. It is measured as a product of the nominal exchange rate (Kenya Shilling against the US dollar) and the ratio of the consumer price index, with year 2010 taken as the base year in both countries. Wage rate (WAGE): It is a measure of the price of labour. It is measured as the average monthly earnings per employee for each of the sectors, over time. Population Density (POPDEN): It is a measure of the intensity of land use. It is measured as the average population per square kilometre. Labour Force Participation rate (Labfoc): is defined as the section of working population in the age group of 16-64 in the economy currently employed or seeking employment. It is calculated as the labour force divided by the total working-age population. It is used as a proxy for those individuals aged between 16-64 years. Foreign Direct Investment (FDI): It is the direct investment in productive assets by an entity established in a foreign country and measured as a percentage of investment inflows to GDP. Trade Openness (OPENNESS): It is the extent which an economy is open to trade. It is measured as the ratio of the value of the total foreign (exports + imports) to the GDP

Data Type and Source
The study used annual time series data for the period 1970 to 2016. The choice of the period was primarily informed by availability of data. Data was obtained from different sources as indicated in Table 3.1.

Testing for Stationary of Data
To detect the presence of unit root in the series, the study employed the Clemente Montanes Reyes (CMR) test. The CMR test is based on the approach that allows for the possibility of having structural breaks in the mean of the series. Clemente, Montanes and Reyes (1998) extended the Perron and Vogelsang (1992) model to take care of two structural changes in the mean. The CMR test is based on a modification of the tests by Perron and Vogelsang (1992) known as the additive outlier (AO) and the innovative outlier (IO). The AO describes the break as occurring suddenly through changes in the mean and the IO views the break as sprouting slowly over time. The CMR test has a null hypothesis that the series has a unit root with structural break(s) against the alternative hypothesis that they are stationary with break(s). The CMR test hypothesis is specified as; j0: = 1 = = 1;: + I : @lm :1 + I ? @lm ?1 + n 1 (25) j1: = 1 = n + F : @o :1 + F ? @lm ?1 + ' 1 (26) Where @lm :1 is a pulse variable equal to one if h = lm + 1 and becomes zero otherwise. @o :1 = 1 if h > lm (L = 1,2) and zero otherwise. lm : and lm ? represents the time periods when the mean is being modified.

Auto Regressive Distributed Lag Model
To achieve the objective of this study, equation 24 was estimated using an Auto Regressive Distributed Lag (ARDL) model. An ARDL model is a standard least squares regressions that include lags of both the dependent variable and explanatory variables as regressors (Greene, 2008). An ARDL model for employment elasticities with the regressors identified in the functional relationship 24 was expressed as: 6Ja 1 = < q + ∑ < :  (27) This was the general ARDL model which was then rewritten as: The distributed lag form of the model that defined long run relationship was thus given as: Where 7 gave short run coefficients, λ was the speed of adjustment parameter and }~l was the residuals that were obtained from the estimated cointegration model of equation 30. The ARDL model was employed so as capture the partial adjustments and adaptive expectations in employment elasticity in Kenya as used by (Nakata & Takehiro, 2003) to estimate both partial adjustments and the elasticities of employment with respect to output and the relative wage in Japan. The partial employment adjustment model helped to explain the employment elasticity adjustment behavior and the observed employment fluctuations. The adaptive expectations model explained how the economy compensates for long-run adjustment of employment elasticity to changes in relative output by speeding up short-run employment adjustments.

Empirical Findings 4.1 Descriptive Statistics
The study analyzed the data for all the variables in order to discern its characteristics prior to estimation. This involved the determination of the mean median, maxima, minima and the standard deviation of the variables.  According to Table 4.1, the average aggregate wage employment for the period under consideration was 1,472,466 persons. The total wage employment had a maximum of 2,553,500 workers and a minimum of 644,500 workers. The growth in wage employment over the study period was as a result of the Kenyan government efforts to create employment opportunities to absorb the country's growing labour force. This was achieved through various short, medium and long-term employment creation measures like Kenyanization Programme, tripartite agreements, active labour market policies, public works programs, foreign employment, and rural development over the plan period (Omolo, 2013).
Trade openness was used to reveal the impact of international trade on employment intensity. Trade openness was given as a ratio of the sum of exports and imports to the GDP and averaged 57. Foreign Direct Investment was measured as a percentage of investments inflows to GDP. The estimation results presented in Table 4.1 shows that FDI ranged from 0.04 per cent to 2.73 per cent with an average of 0.86 per cent. The estimation results reveal that FDI inflows to Kenya have been highly volatile. This volatility could be attributed to low confidence of investors as a result of insecurity and political instability when FDI inflows were quite low as shown by a minimum value of 0.04 per cent. On the other hand, high FDI inflows could be attributed to improved investment environment through implementing various macroeconomic reforms. Population density in the country ranged between 20 people per square kilometre of land and 83 people per square kilometre of land with a mean of 47 people per square kilometre of land. The sample population density mean was low compared to the country's projected population density mean of 73.9 people per square kilometre in 2014 (Republic of Kenya, 2014).
Exchange rate was used to reveal the impact of external shocks on employment elasticity. On average the exchange rate was 43.19 Kshs/US$ with maximum and minimum values of 101.05 Kshs/US $ and 7.02 Kshs/US$ respectively. Therefore, the exchange rate over the study period was volatile. This implies that the country experienced low, mild, and high exchange rates at some points in the study period. The rate of inflation for the period under study ranged from 1.55 per cent to 45.98 per cent with an average of 12.43 per cent a standard deviation of 12.15. Overall, therefore, Kenya experienced mild, rapid and galloping inflation rates. The average rate of inflation was more than a single digit which was higher than the one envisaged in the EAC Monetary Union Protocol. According to this protocol, headline inflation rate should be about 8 per cent under macroeconomic convergence criteria (EAC, 2013). Kenya's output level was captured by real GDP. Over the period 1970-2016, the real GDP averaged Kshs.1, 638,972 million. The real GDP had a maximum of Kshs. 4,300,302 million and a minimum of Kshs. 527,290 million. The growth in real GDP over the study period could be attributed to the various development plans and strategies implemented by the government over the study period.

Unit Root Test Results
Unit root tests for all the variables were conducted so as to establish the order of integration. The test results for all the variables are reported in Table 4  The test statistic for CMR unit root is the minimum h-statistic. The estimation results of the CMR unit root test shown in Table 4.2 indicate that variables inflation rate, FDI, wage rate, and employment elasticity, were statistically significant at 5 per cent level. This is because the minimum hvalue for these variables were smaller than the critical value of -5.490 at 5 per cent significance value. Thus, according to CMR unit root test, the null hypothesis of presence of a unit root with structural break(s) for variables GDP growth rate, inflation rate, FDI, wage rate and employment elasticity was rejected and the alternative hypothesis that the series are stationary was not rejected. This implies that these variables were stationary at levels suggesting that they are integrated of order zero, I (0).
According to Table 4.2, the test statistic for variables population density, exchange rate, labour force participation rate and trade openness, were not statistically significant at 5 per cent significance level. This means that the variables were not stationary at levels. Thus, according to CMR unit root test, these variables had at least one unit root and required to be differenced to become stationary. The series were, however, stationary at first difference and therefore integrated of order one, I (1) as shown in Table 4 The structural breakpoints for key macro-economic variables in the series coincided with key economic developments in the country.

Empirical Results
Unit root tests results in Table 4.2 indicated that the variables used in the study mixed orders of integration, that is I (0) and I (1) suggesting that ARDL was the appropriate model for estimation. These variables included employment elasticity, average wage, exchange rate, inflation, labour force participation rate, population density, trade openness and FDI. The ARDL model was preferred due to its ability to estimate the long and short-run parameters of the model simultaneously for the avoidance of the problems posed by non-stationary time series data. Pesaran and Shin (1999) showed that cointegrating systems can be estimated as ARDL models, with the advantage that the variables in the cointegrating relationship being either I(0) or I(1), without needing to prespecify which are I(0) or I(1). Pesaran and Shin (1999), also note that unlike other methods of estimating cointegrating relationships, the ARDL representation does not require symmetry of lag lengths; each variable can have a different number of lag terms.
Before estimation of the ARDL, a bound test was estimated to determine whether the independent variables had a long-run relationship with the dependent variable. The bound tests were estimated using the approach proposed by Pesaran and Shin (1999) for testing long run relationship among the variables expressed in equation 24. The -statistic tests for the joint significance of lagged variables. If the -statistic falls below the lower critical value (Lower bound), the null hypothesis of no long-run relationship is accepted irrespective of the orders of integration but if the -statistic falls above the upper critical value (upper bound), the null hypothesis of no longrun relationship is rejected (Pesaran and Shin, 1999). However, if the -statistic falls between the lower and upper critical values, any inference would be inconclusive and knowledge of the order of integration of the variables will be needed before conclusive inferences are made. The bound test results are shown in Table 4.4.

Source: Derived from collected data
The estimation results in Table 4.4 show an -statistic value of 10.12 which is more than the upper bound value of 4.43 at one (1) per cent significance level thus rejecting the null hypothesis and concluding that there exists a long run relationship among the model variables. The results, therefore, justify the use of ARDL and ECM version of ARDL to derive the long run and short run relationships of the variables.
Before estimating the short and long run relationships of the variables, diagnostic tests were conducted to establish the ARDL statistical appropriateness. Equation 24 was employed to establish the optimal lag length and the goodness of fit. The adopted optimal lags by the Akaike's Information Criterion (AIC) automatic lag selection were (1, 2, 2, 0, 2, 0, 0, 2). The R-squared value was 95 percent. The estimated model had a -statistic of 11.349 with a corresponding a-value of 0.0023. Since this a-value (0.0023) was less than the critical value of 0.5 at 5 per cent significance level, the null hypothesis of joint significance of the explanatory variables being equal to zero was rejected. The LM serial correlation test and Breusch Pagan -Godfrey tests were used to test for the presence of serial correlation and heteroscedasticity respectively. The estimated results gave an observed •squared value of 1.0542 with a corresponding a-value of 0.28. The null hypothesis that there is no serial correlation was thus accepted at five per cent level. The a-value (0.4332) for the observed •-squared in Breusch Pagan Godfrey test also led to the acceptance of the null hypothesis of no heteroscedasticity at 5 per cent level.
The objective of the study was therefore realized by estimating the Auto regressive Distributed Lag (ARDL) model given in equation 24 for long run and an Error Correction Model (ECM) version of ARDL for short run sources of employment elasticities. The short run estimation output are given in Table 4.5.

Short Run Sources of employment elasticities in Kenya
The short run estimation was done in two stages. The first stage involved estimating the reduced cointegrating ARDL equation 24. The residuals got from estimation were then lagged once (ECT-1) and were used in the second stage to estimate the ECM version of ARDL model. The short-run results represent the coefficients of the differenced explanatory variables and they give short-run growth effects. The coefficients describe short-term growth in dependent variable resulting from previous period's growth in the independent variables. The results are presented in Table 4.5.   Table 4.5 indicates that only short-run coefficients for variables first lag of employment elasticity, average wage, inflation rate, labour force participation rate, first and the second lag of labour force participation rate, population density and the first and the second lag of FDI were statistically significant. This reflects presence of short-run relationship between employment elasticity and these variables.
Employment elasticity was the dependent variable and denoted the overall growth employment elasticity at time h. The annual employment elasticity was calculated by dividing the percentage change in employment by the corresponding percentage change in GDP during a given period to provide a time variability which is not possible with OLS estimates. The estimation results presented in Table 4.5 shows that the coefficient of the first lag of employment elasticity change is positive and statistically significant at one per cent level of significance. The change in employment elasticity in the current period impacts positively the change in employment elasticity after one period.
Inflation rate is viewed as a proxy for the level of economic stability in an economy and it is theoretically expected to have a negative effect on employment yield of economic growth. The coefficient of inflation was -0.11 and statistically significant at 1 per cent level. This implies that a one per cent increase in inflation rate will reduce employment elasticity by 0.11 per cent. The study results concur with Kapsos (2005), Crivelli et.al (2012) and Ghazali (2014) who found the rate of inflation to be negatively related with employment elasticity. Inflation can be expected to have negative effect on employment yield of output growth. This is because high rates of inflation mean high production costs in terms of high prices for inputs, raw materials and even labour. Instability in macroeconomic variables would thus imply limiting investment opportunities that would in turn create employment opportunities.
The population density variable had a coefficient of 0.031. The coefficient was positive and statistically significant at 1 per cent level. Therefore, a one per cent increase in the population density will cause a 0.031 increase in employment elasticity. This implies that increase in the population density growth rate enhanced employment yield of output growth. The study results concur with Adegboye, Egharevba & Edafe (2017), but contradict with Crivelli et.al (2012) who found population density to be negatively correlated with employment elasticity. The implication of the results is that in areas with high population density, the level of unemployment is also very high. This could mean that any change in employment growth would have a bigger impact than in areas which are densely populated.
The annual average wage had a negative sign for its coefficient in the current period and statistically significant at 5 per cent level. The magnitude of annual average wage was -0.0012; this indicates that a one per cent increase in annual average wage will lead to a decrease in employment elasticity by 0.0012 per cent. The implication is that average wages have an indirect relationship with employment elasticity. The results concur with Ghazali (2014) that higher average annual real wages reduce employment elasticity. However, as the period moves into the future, the coefficient becomes insignificant. A one-unit growth of average wage affects employment elasticity by 0.004 one period latter and 0.001 two periods after. The study findings imply that higher average annual real wages reduce employment-growth elasticity within the current year. This is because higher wages and non-wage benefits increases the cost of production and could constrain growth-induced employment opportunities.
Labour force participation rate was included in the model to assess the effects of labour supply on employment elasticity. The coefficient of labour force participation variable was found to be positive with a magnitude of 0.045 and statistically significant at 5 per cent level in the current period. The coefficients of the first and second lags of labour force participation variable were also positive and statistically significant at 5 and 1 per cent levels respectively. The magnitude of effect increases as the variable move into the future as indicated by higher value of the coefficient as compared to the previous value. This implies that an expanding supply of labour leads to a more employment-intensive growth. This concurs with the classical theoretical prediction that higher labour supply will lead to lower average wages and ultimately lead to an increase in demand for labour input (Ambrosi, 1986). The results are consistent with Chan (2001) who found a positive and significant relationship between labour force participation rate and employment elasticity. The estimation results, however, contradict Crivelli (2012), who found labour force participation rate to be negatively correlated with employment-output elasticities for advanced countries.
A probable explanation for the relationship between labour force participation rate and employment elasticity could be the existence of a dualistic economy in Kenya where there exists a big difference between formal sector and informal sector wages due to the existence of labour market imperfections. The rates of rural to urban migration have tended to exceed the absorptive capacity of the modern sector, leading to the growth in the informal sector. This means that employment growth would be responsive to any slight change in the growth of labour force. Another probable explanation could be that population growth stimulates technological progress and makes possible the realization of economies of scale that provide incentives for the adoption of more efficient techniques and institutional arrangements (Kawagoe, Hayami and Ruttan, 1985). This would thus imply that population growth promotes development by increasing the economy's productive capacity which in turn create employment elasticities.
The coefficient of foreign direct investment variable was found to be negative with a magnitude of -0.718 and was statistically insignificant at all three levels in the current period. The magnitude of the first lag of FDI was 1.179 and statistically significant at 5 per cent level. The magnitude of the second lag was 0.896 and statistically significant at 10 per cent level. This implied that a unit increase in foreign direct investment increases employment elasticity by 0.1179 in the first year and 0.896 in the second year respectively. The increase in FDI in the current period affects employment elasticity after one period and also two periods later. The study results concurs with Adegboye, Egharevba, & Edafe (2017), who found that lagged values of FDI had a positive impact on employment elasticity in Sub-Saharan Africa. A probable explanation could be the bureaucratic procedures involved in the release of funds from abroad. Even if the funds were to be released and invested immediately, the production in the affected sectors of the economy will only be realized with a lag since it takes time to establish the investments. Foreign direct investment also generates employment through forward and backward linkages with domestic firms. Table 4.5 reveals that the coefficient of the trade openness variable though positive with a magnitude of 0.0009, was not statistically significant at all the three levels of significance. The results concur with (Bruno, Falzoni, & Helg, 2003) who did not find any statistically significant relationship between trade openness and labour demand elasticity. The results, however, contradict Crivelli (2012), who found that employment elasticities tend to be higher in more advanced and closed economies. Despite the insignificant relationship between trade openness and employment elasticity, the positive relationship among the two variable is consistent with the trade theory. Wood (1999), ascertains that trade openness can lead to an increase in labour demand in labour-abundant countries due to comparative advantage. This in turn is expected to increase labour-demand elasticities as labour comes under pressure due to stiffer competition in the goods and labour markets.
Just like inflation, exchange rate was also used in empirical analysis as an indicator for assessing macroeconomic stability. The coefficient of exchange was found to be negative and statistically insignificant at all levels in the current period. The coefficients of the first and second lags of exchange rate were, however, positive but also statistically insignificant at the three levels of significance. The study results contradict Ghazali (2014) who found a negative and highly statistically significant relationship between exchange rate and employment elasticities in Tunisia.
The co-efficient of the error term or the adjustment factor had a magnitude of -1.921 and was statistically significant at 5 per cent level. The fact that the error correction term was negative provides evidence about the existence of a long-run association among the variables. The coefficient of the error term showed the proportion of the current changes in employment elasticity that were explained by the disequilibrium error in the previous period.

Long Run Sources of Employment Elasticities in Kenya
The long run coefficients of the determinants of employment elasticities showed how employment elasticities reacted to a permanent change in the independent variables. The long run results are given in Table 4.6.  Table 4.6 indicates that only long run coefficients for variables exchange rate, FDI, and population density were statistically significant. This reveals presence of long run relationship between employment elasticity and these variables. The coefficient of exchange rate variable was found to be negative with a magnitude of 0.143 and statistically significant at 5 per cent level. Therefore, a one per cent increase in the exchange rate will cause a 0.143 decrease in employment elasticity in the long run. This implies that increase in the rate of exchange rate (depreciation of the domestic currency) deterred employment yield of economic growth. A probable explanation is that depreciation of the domestic currency contracts the growth of real output due to high dependence on import of inputs and capital goods. Alexandre, Bacao, Cerejeira & Portela (2010) finds similar results; a negative relationship between exchange rate and employment elasticity for 23 OECD countries. The results are also consistent with Mouelhi and Ghazali (2014) who found that nominal exchange rate was negative and highly statistically significant to employment elasticity across all specifications in Tunisia.
The FDI variable had a statistically significant coefficient of 0.061 at one per cent level of significance. Therefore, a one per cent increase in FDI led to 0.061 per cent increase in employment elasticity. This implies that an increase in FDI enhanced employment elasticity in the long run. Foreign direct investment brings investable financial resources, provides new technologies and improves the efficiencies of existing technologies and therefore acts as a stimulant for employment growth. The results contradict Akinkugbe (2015) who found that the surge in FDI inflows has not manifested in significant formal sector job creation and reduction in unemployment levels in the long run.
The coefficient of population density variable was found to positively influence employment elasticity. The coefficient was also statistically significant at 5 per cent level with a magnitude of 0.053. Therefore, a one unit increase in population density increases employment elasticity by 0.053 per cent. This meant that an increase in population density also led to a high employment elasticity in the long run. The estimation results contradict Crivelli et.al (2012), who found population density to be negatively correlated with employment-output elasticities, for 167 countries.

Conclusions and Policy Implications
This study concludes that the first lag of employment elasticity, average wage, inflation rate, labour force participation rate, first and second lags of labour force participation rate, population density, first and second lags of FDI were the short run drivers of employment elasticity. Empirical findings also indicate that exchange rate, FDI and population density are the long run drivers of employment elasticity in Kenya. The study recommends that policy measures to control inflation should be tightened. This could be realized by devising strategies to increase the tax base and improve the tax compliance in the country. The study also recommends that efforts to attract more foreign direct investment should be undertaken. Government through relevant agencies should increase the ease of doing business in the country. This could be achieved by enhancing infrastructural development in the country which is a key driver of FDI. Another measure to enhance FDI in the country could be through enhancement of foreign direct investments incentives. These measures can be directed by the National Treasury. A stable exchange rate should be maintained. This could be achieved by growing and diversification of exports. In addition, tourism which is a major source of foreign currency can be promoted by investing in product diversification in the tourism sector. Lastly, the government should harmonize the salary scale framework to regulate the wages in the country. This could be realized through salary adjustments based on a periodical and systematic evaluation of wage parameters in the public sector and taking cognizance of the prevailing economic dynamics.