Finance and Labour Productivity in Nigerian Agricultural Sector: Assessing the Moderating Effect of Insecurity

This study covered the period from 1992 to 2023. The temporal framework selected for this analysis is dictated by the significant fluctuations in agricultural labour productivity attributable to heightened insecurity, which has threatened investment in the sector. This study utilised annual time series data concerning the specified variables. The data were meticulously sourced from the Central Bank of Nigeria’s Statistical Bulletin and the World Development Indicators (WDI). Labour productivity (LABP) quantifies agricultural output per worker, government expenditure (GEXP) refers to total governmental allocations to agriculture, commercial bank credit (CBCR) reflects the aggregate volume of credit extended by commercial banks to the agricultural sector, microfinance bank loans (MFBL) denotes the total value of agricultural loans disbursed by microfinance institutions, and agricultural credit guarantee scheme funds (ACGSF) accounts for the cumulative amount of loans authorised under the ACGS. To quantify insecurity, the number of fatalities (DTHS) in Nigeria resulting from security-related challenges was employed.

This research undertook an examination of the interrelationship between financial mechanisms and labour productivity, wherein capital (finance), as a component of the endogenous growth model, served as a key factor influencing labour productivity. The endogenous growth model delineates technology, along with the accumulation of physical and human capital, as pivotal elements propelling labour productivity. An increase in the quantity of equipment and machinery allocated to a specific labour unit is anticipated to enhance the marginal productivity associated with that labour unit. In a similar vein, elements that contribute to the enhancement of labour knowledge or technical acumen—such as training, education, research, and superior healthcare facilities—are regarded as essential catalysts for labour productivity. The determinants of labour productivity, as articulated by the endogenous growth theoretical framework, can be encapsulated by equation (1): (1) LABP=fK,H,A,u LABP = f\left( {K,H,A,u} \right) Where:

• K – physical capital accumulation

• H – human capital accumulation

• A – technology

• u – other factors affecting labour productivity

Equation 1 represents an articulation of the determinants of labour productivity. The accumulation of physical capital, the accumulation of human capital, and technological advancements exert a positive influence on labour productivity. Nevertheless, the supply-leading hypothesis concerning the relationship between finance and growth contends that the determinants of productivity identified by the endogenous growth model fail to manifest in the absence of sufficient financial resources allocated to productive activities within the economy. Consequently, this study emphasises that access to financial resources (capital) is a fundamental determinant of labour productivity within the Nigerian agricultural sector. In this context, the sources of financial resources (capital) pertinent to agricultural activities encompass government expenditures, credit from commercial banks, loans from microfinance institutions, and agricultural credit guarantee scheme funds. As a result, the finance-labour productivity model is delineated by equation (2). (2) LABP=β0+β1GEXP+β2CBCR+β3MFBL ++ β4ACGSF+μ \matrix{ {LABP = {\beta _0} + {\beta _1}GEXP + {\beta _2}CBCR + {\beta _3}MFBL\; + } \cr { + \;{\beta _4}ACGSF + \mu } \cr }

In alignment with the objectives of this study, the tenets of the agricultural productivity gap theory, as posited by Gollin et al. (2013) and adapted from Usman et al. (2024), were employed to modify equation (2) by recognising the possible moderating effect of insecurity on the endogenous growth framework. The “insecurity and agricultural productivity gap theory” posits that a significant factor contributing to the disparity in agricultural productivity between developed and developing nations is the prevalence of elevated insecurity levels in numerous developing regions, which directly impedes farmers’ capacities to cultivate land, access resources, and ultimately produce crops effectively, resulting in diminished overall agricultural output. In this context, the tally of deaths (DTHS) attributable to insecurity was utilised to gauge the extent to which insecurity is undermining the effect of finance on labour productivity in Nigeria. Thus, equation (2) is revised by integrating DTHS to formulate equation (3). (3) LABP=β0+β1GEXP+β2CBCR+β3MFBL ++ β4ACGSF+β5DTHS+μ \matrix{ {LABP = {\beta _0} + {\beta _1}GEXP + {\beta _2}CBCR + {\beta _3}MFBL\; + } \cr { + \;{\beta _4}ACGSF + {\beta _5}DTHS + \mu } \cr }

To linearise equation (3), a logarithmic transformation was applied to obtain equation (4), which is the model used for the analysis. (4) LNLABP=β0+β1LNGEXP+β2LNCBCR ++ β3LNMFBL+β4LNACGSF+β5LNDTH+μ \matrix{ {LNLABP = {\beta _0} + {\beta _1}LNGEXP + {\beta _2}LNCBCR\; + } \cr { + \;{\beta _3}LNMFBL + {\beta _4}LNACGSF + {\beta _5}LNDTH + \mu } \cr } Where:

• LN – denotes the natural logarithm of the variables

• β₀ – is the constant

• β₁ – β₀ represent the estimated coefficient

• μ – denotes the error term.

The discourse pertaining to the estimation strategy within this section is delineated into two distinct subsections. The initial subsection elucidates the methodologies employed to conduct stationarity assessments, while the concluding subsection explicates the Non-linear Autoregressive Distributed Lag (NARDL) model and the rationale underpinning the selection of this estimation strategy.

Utilising non-stationary variables in time series analysis may culminate in estimates that are inconsistent and biased. Furthermore, this practice can yield spurious regression results, which may prove unsuitable for conducting analyses and drawing inferences. For the stationarity assessment, we employed the Phillips and Perron (PP) test alongside the Augmented Dickey-Fuller (ADF) tests (Wolters and Hassler, 2005). The ADF test can be articulated as expressed in equation 5: (5) ΔXt=γ0+δXt−1+γ1∑i=1ρΔXt-i+εt \Delta {{\rm{X}}_{\rm{t}}} = {\gamma _0} + \delta {{\rm{X}}_{{\rm{t}} - 1}} + {\gamma _1}\sum\limits_{{\rm{i}} = 1}^\rho {\Delta {{\rm{X}}_{{\rm{t - i}}}} + {\varepsilon _{\rm {t}}}} Where X_t is the variable under consideration at period t. The ADF and PP tests are conducted with the null hypothesis HO: δ = 0 against the alternative hypothesis H1: δ < 0. The null hypothesis states that a unit root is present, while the alternative hypothesis states that there is no unit root. The series is considered stationary when the null hypothesis is rejected.

The Non-linear Autoregressive Distributed Lag (NARDL) framework serves as a sophisticated tool for the examination of the asymmetric effects of financial factors on productivity. This methodology enables scholars to explore whether the influence of financial advancement (or alternative financial metrics) varies based on the directional movement of the variable, i.e., whether it is increasing or decreasing. In other terms, it can elucidate whether positive or negative fluctuations in financial metrics yield disparate outcomes for productivity (Odugbesan et al., 2021). Essentially, the NARDL model proposed by Shin et al. (2014) extends the bounds testing methodology for co-integration embedded within the Autoregressive Distributed Lag (ARDL) framework (Pesaran et al., 2001). The utilisation of such a model provides considerable flexibility, as it remains applicable regardless of the stationarity of the underlying variables. Consequently, irrespective of whether the foundational variables exhibit integration of order I(0) or I(1), or are mutually co-integrated, the NARDL model retains its applicability (Salisu and Isah, 2017). Nevertheless, this should not obviate the need for pre-testing the integration order of the variables, which is paramount to ascertain that no variables incorporated into the analysis are I(2). Furthermore, pre-testing the integration order establishes a solid foundation for validating the existence of long-term co-integration among the variables. The preference for the NARDL model over the ARDL model arises from the latter’s inability to accommodate the asymmetries inherent in the fluctuations of the variables. Nonetheless, the dynamics of agricultural finance and insecurity have demonstrated a non-linear trajectory from 1992 to 2023. Therefore, there exist intervals during which agricultural finance and insecurity escalate, alongside periods characterised by their decline. Consequently, the moderating influence of insecurity on the relationship between finance and labour productivity may be asymmetrical, potentially manifesting as detrimental when finance diminishes while insecurity escalates, and conversely, exhibiting a favourable effect when insecurity progressively declines while finance increases. Thus, the application of the ARDL model to estimate the empirical model delineated in equation (4) would likely yield spurious regression outcomes, culminating in erroneous conclusions. Similar to the ARDL model, the NARDL framework demonstrates efficiency and performs adequately with small sample sizes while concurrently accounting for endogeneity across all variables. The model incorporates both short-run and long-run asymmetries in the trajectories of the variables (specifically lending) and is applicable in the context of mixed orders of integration. The general asymmetric long-run regression model devised to scrutinise the asymmetric effects of the model is represented in equation (6). (6) LABPt=β−FINANCEt−+β+FINANCEt+ ++ β+INSECURITYt++μt \matrix{ {{\rm{LAB}}{{\rm{P}}_{\rm{t}}} = {\beta ^ - }{\rm{FINANCE}}_{\rm{t}}^ - + {\beta ^ + }{\rm{FINANCE}}_{\rm{t}}^ + \; + } \cr { + \;{\beta ^ + }{\rm{INSECURITY}}_{\rm{t}}^ + + {\mu _{\rm{t}}}} \cr } Where, LABP_t, FINANCE_t and INSECURITY_t are scalar (1) variables; FINANCE_t is decomposed as FINANCE_t = FINANCE₀ + FINANCE_t⁻ + FINANCE_t⁺ where FINANCE_t⁻ and FINANCE_t⁺ are partial sum processes of negative and positive changes in FINANCE_t. Similarly, INSECURITY_t is decomposed as and INSECURITY_t = INSECURITY₀ + INSECURITY_t⁻ + INSECURITY_t⁺ where INSECURITY_t⁻ + INSECURITY_t⁺ are partial sum processes of negative and positive changes in INSECURITY_t. Thus, the study filtered the positive changes in finance and insecurity (increase) from the negative changes in finance and insecurity (decrease) to be able to examine their individual effects on agricultural labour productivity in Nigeria. The partial sums are formally defined as displayed in equations 7 and 10. (7) LnFINANCEt+=∑i=1tΔLnFINANCEt+=∑i=1tmaxΔLnFINANCEi0 \matrix{ {{\rm{LnFINANCE}}_{\rm{t}}^ + = \sum\limits_{i = 1}^t {\Delta {\rm{LnFINANCE}}_{\rm{t}}^ + } = } \cr {\sum\limits_{i = 1}^t {{\rm{max\;}}\left( {\Delta {\rm{LnFINANC}}{{\rm{E}}_{\rm{i}}}0} \right)} {\rm{\;}}} \cr } (8) LnFINANCEt−=∑i=1tΔLnFINANCEt+=∑i=1tmaxΔLnFINANCEi0 \matrix{ {{\rm{LnFINANCE}}_{\rm{t}}^ - = \sum\limits_{i = 1}^t {\Delta {\rm{LnFINANCE}}_{\rm{t}}^ + } = } \cr {\sum\limits_{i = 1}^t {{\rm{max\;}}\left( {\Delta {\rm{LnFINANC}}{{\rm{E}}_{\rm{i}}}0} \right)} {\rm{\;}}} \cr } (9) LnINSECURITYt+=∑i=1tΔLnINSECURITYt+=∑i=1tmaxΔLnINSECURITYi0 \matrix{ {{\rm{LnINSECURITY}}_{\rm{t}}^ + = \sum\limits_{i = 1}^t {\Delta {\rm{LnINSECURITY}}_{\rm{t}}^ + } = } \cr {\sum\limits_{i = 1}^t {{\rm{max\;}}\left( {\Delta {\rm{LnINSECURIT}}{{\rm{Y}}_{\rm{i}}}0} \right)} {\rm{\;}}} \cr } (10) LnINSECURITYt−=∑i=1tΔLnINSECURITYt+=∑i=1tmaxΔLnINSECURITYi0 \matrix{ {{\rm{LnINSECURITY}}_{\rm{t}}^ - = \sum\limits_{i = 1}^t {\Delta {\rm{LnINSECURITY}}_{\rm{t}}^ + } = } \cr {\sum\limits_{i = 1}^t {{\rm{max\;}}\left( {\Delta {\rm{LnINSECURIT}}{{\rm{Y}}_{\rm{i}}}0} \right)} {\rm{\;}}} \cr }

Note that the FINANCE components are GEXP, CBCR, MFBL and ACGS, while INSECURITY is proxied by the number of deaths (DTHS).

Where the variables in equations (6), (7) and (9) are explained in the previous equations. To obtain the estimable NARDL model, equations (6), (7) and (9) were used. Thus, equations (6), (7) and (9) were substituted into the original ARDL model to arrive at the following nonlinear ARDL model for agricultural labour productivity (Shin et al., 2014), as shown in equation (11). (11) LnΔLABPt=∑t=1ρ1β1ΔLnLABPt−i+∑t=1ρ2π1ΔLnGEXPt−i ++ ∑t=1ρ3η1ΔLnCBCRt−i +∑t=1ρ4y1ΔLnMFBLt−i ++ ∑t=1ρ5τ1ΔLnACGSt−i+∑t=1ρ6φ1ΔLnDTHSt−i ++ δ1LnLABPt−1+δ2LnGEXPt−1+δ3LnCBCRt−1 ++ δ4LnMFBLt−1+δ5LnACGSt−1+δ6LnDTHSt−1+εt \matrix{ {{\rm{Ln}}\Delta {\rm{LAB}}{{\rm{P}}_{\rm{t}}} = \sum\limits_{t = 1}^{\rho 1} {{\beta _1}\Delta {\rm{LnLAB}}{{\rm{P}}_{{\rm{t}} - {\rm{i}}}}} + \sum\limits_{t = 1}^{\rho 2} {{\pi _1}\Delta {\rm{LnGEX}}{{\rm{P}}_{{\rm{t}} - {\rm{i}}}}} \; + } \cr { + \;\sum\limits_{t = 1}^{\rho 3} {{\eta _1}\Delta {\rm{LnCBC}}{{\rm{R}}_{{\rm{t}} - {\rm{i}}}}} \; + \sum\limits_{t = 1}^{\rho 4} {{y_1}\Delta {\rm{LnMFB}}{{\rm{L}}_{{\rm{t}} - {\rm{i}}}}} \; + } \cr { + \;\sum\limits_{t = 1}^{\rho 5} {{\tau _1}\Delta {\rm{LnACG}}{{\rm{S}}_{{\rm{t}} - {\rm{i}}}}} + \sum\limits_{t = 1}^{\rho 6} {{\varphi _1}\Delta {\rm{LnDTH}}{{\rm{S}}_{{\rm{t}} - {\rm{i}}}}} \; + } \cr { + \;\;{\delta _1}{\rm{LnLAB}}{{\rm{P}}_{{\rm{t}} - 1}} + {\delta _2}{\rm{LnGEX}}{{\rm{P}}_{{\rm{t}} - 1}} + {\delta _3}{\rm{LnCBC}}{{\rm{R}}_{{\rm{t}} - 1}}\; + } \cr { + \;\;{\delta _4}{\rm{LnMFB}}{{\rm{L}}_{{\rm{t}} - 1}} + {\delta _5}{\rm{LnACG}}{{\rm{S}}_{{\rm{t}} - 1}} + {\delta _6}{\rm{LnDTH}}{{\rm{S}}_{{\rm{t}} - 1}} + {\varepsilon _{\rm{t}}}} \cr }

Model (11) is an error-correction model (ECM) that is labelled as a nonlinear ARDL model. Nonlinearity is introduced through the partial sums variables in models (7) to (10). Shin et al. (2014) have demonstrated that Pearson et al. (2001) approach of estimating the linear ARDL and testing co-integration is equally applicable to the nonlinear ARDL model. The difference between these two models is that for the nonlinear ARDL, finance and insecurity changes would have a symmetric (linear) effect on agricultural labour productivity if the positive and negative coefficients in model (11) have the same size and sign. Any result aside from this outcome makes the model asymmetric.

The descriptive statistics outline the fundamental characteristics of the data set employed in this study. Table 2 presents the results of the summary statistics.

Table 2 shows the descriptive statistics for the variables used in our study. Labour productivity had a minimum value of 28.93 and a maximum value of 30.59, with an average value of 29.87, while its deviation from the mean was 0.60. Government expenditure averaged 9.66 with a minimum value of 6.12 and a maximum value of 11.31, having a standard deviation of 1.48. For commercial bank credit, the maximum and the minimum values were 14.33 and 8.85, respectively with an average of 11.76 with a deviation from the mean of 1.53. Additionally, microfinance bank loans averaged 7.80, with values ranging from 3.38 to 10.05 and a standard deviation of 1.85. Agricultural credit guarantee scheme funds had a mean score of 7.60, with minimum and maximum values of 4.39 and 9.47, respectively, alongside a standard deviation of 1.70. The number of deaths due to insecurity averaged 10.31, with values ranging from 7.87 to 13.92, while its standard deviation was 0.42. Regarding the skewness and kurtosis, all the variables are normally skewed, with values being less than 1 with fatter tails at the end. The Jarque-Bera shows that the natural logarithm of the variables was normally distributed; thus, the null hypothesis of a normal distribution was accepted.

The results of the Philip-Perron (PP) and Augmented Dickey Fuller (ADF) tests are presented in Table 3.

Table 3 displays the results attained. It was found that labour productivity (LABP), agricultural credit guarantee scheme funds (ACGSF) and number of deaths (DTHS) were all integrated of order one, I(1), while government expenditure (GEXP), commercial bank credit (CBCR) and microfinance loans (MFBL) were integrated of order zero, I(0). These results have two implications. First, they provide evidence in support of the choice of the NARDL model, which allows for the inclusion of I(1) and I(0) independent variables in the same empirical equation. Second they validate the application of the bounds testing procedure to examine the existence of a long-run relationship between components of finance and labour productivity in Nigeria.

The main aim of using the unit root approach of Zivot and Andrews (2002) in this section is to test for structural breaks in the series. A structural break is an unexpected change over time in the parameters of a regression model, which can lead to huge forecasting errors and undermine the reliability of the model. The results displayed in Table 4 empirically confirm the presence of a structural break in all variables included in the empirical model. However, for labour productivity (LABP), agricultural credit guarantee scheme funds (ACGSF) and the number of deaths (DTHS), stationarity is attained at level, while for government expenditure (GEXP), commercial bank credit (CBCR), and microfinance bank loans (MFBL), stationarity is attained after first differencing, which is consistent with the results of the Augmented Dickey-Fuller (ADF) and Phillips–Perron unit root tests. The presence of structural breaks and the mixed order of integration (none exceeding order one) confirms the suitability of the Nonlinear ARDL (NARDL) model for examining the nonlinear relationships among the variables under consideration.

The study followed the bounds test approach to co-integration to examine the long-run relationship between the dependent and independent variables, as detailed in Table 5.

From Table 5, the result of the bounds test suggests that the F-statistic of 13.798 is greater than the upper bound critical value of 3.000 at the 1% and 5% levels of significance. It can therefore be concluded that there is a long-run association or co-integration between the dependent and independent variables. The long-run asymmetric effect of finance on labour productivity in Nigeria was then estimated while controlling for insecurity.

The main goal in this section is to estimate the asymmetrical effect of finance on labour productivity amidst long-run insecurity. The long-run estimates are presented in Table 6.

From the long-run estimates reported in Table 6, it was found that both positive and negative finance (lnGEXP, lnCBCR, lnMFBL, and lnACGS) variables and the moderating effect of insecurity measured by lnDTHS carry significant coefficients with different signs and sizes, which supports the presence of long-run asymmetric effects of finance components and insecurity on labour productivity in the Nigerian agricultural sector. Thus, the estimated long-run NARDL empirically confirms an asymmetric effect of finance amidst insecurity on labour productivity, because positive changes in all the finance components exerted a significant positive (increase) effect on labour productivity while negative changes (decrease) in finance components exerted a significant negative effect. Specifically, the results revealed that a one per cent increase in lnGEXP generated approximately a 0.258 increase in agricultural labour productivity, while a one per cent increase in lnCBCR generated approximately a 0.410 increase in labour productivity; a one per cent increase in lnMFBL resulted in approximately a 0.175 increase in agricultural labour productivity, while lnACGS generated approximately a 0.157 rise in agricultural labour productivity, ceteris paribus. On the other hand, all else being equal, a one per cent increase in lnDTHS led to a 0.521 decline in agricultural labour productivity in Nigeria. These findings are similar to those of Yola et al. (2024), Magaji et al. (2023), Toheeb and Dabo (2022), Agbana and Lubo (2022), Obialor et al. (2022), Afolabi et al. (2021), and Iderawumi and Ademola (2015), who examined the impact of various finance sources on agricultural productivity in Nigeria. The findings also align with those of Tank et al. (2024), Falade et al. (2024), Abdulaziz et al. (2022), Babajide and Badru (2022), and Illesanmi and Odefadehan (2022), who observed an inverse relationship between insecurity and agricultural productivity, such that the negative effect of insecurity could inhibit the potential positive effects of finance on agricultural productivity over time. On the other hand, some studies, like Nuhu et al. (2022), Amakor and Anyamaobi (2022), Danladi et al. (2021), and Ndubuaku et al. (2019), have shown that finance components such as government expenditure, bank loans and agricultural credit guarantee scheme funds failed to significantly impact agricultural productivity in Nigeria.

The economic implications of these findings are that positive changes in finance components increase the availability of investable funds, which encourages new agricultural projects, thereby increasing investment – a key component of economic productivity. Increased financing enables farmers to purchase better inputs, modernise farming practices, and expand their operations, ultimately leading to higher yields, increased incomes, reduced poverty and increased productivity in the agricultural sector. Financial resources allow farmers to invest in improved seeds, fertilisers, and irrigation systems, which can significantly boost crop yields and livestock productivity. Access to credit and loans enables farmers to purchase and use modern machinery and technologies, leading to more efficient farming practices and reduced labour costs. Through funding, higher agricultural incomes generated through increased productivity can improve the livelihoods of farmers and their families, contributing to poverty reduction in rural areas, which further contributes to overall economic growth by increasing food supply, reducing food import dependence, and generating more revenue for the country. However, insecurity can be seen to hamper the positive effect of finance on labour productivity in the Nigerian agricultural sector. Insecure environments, especially those with banditry and farmer-herder conflicts, negatively impact agricultural productivity by deterring investment, reducing capital stock, and diverting funds to security, hindering infrastructure development and job creation. This implies that the availability of funds for agriculture can be a key driver of productivity, but if the environment is insecure, it can be difficult for farmers to invest, cultivate, and reap the benefits of those funds.

Theoretically, the supply-leading hypothesis suggests that finance and productivity are intrinsically linked. Well-developed financial systems, with efficient allocation of capital and sound institutions, can significantly boost productivity growth by enabling firms to access resources for investment and innovation. The relationship between finance and productivity is observed across various sectors, including agriculture, where access to credit is vital for adopting new technologies and improving yields. Access to finance is crucial for boosting productivity, while efficient financial systems can also contribute to higher productivity. Conversely, financial friction and inefficiencies can hinder productivity growth. The findings of this study align with the supply-leading hypothesis of finance and productivity.

Table 7 reports the short-run estimates from the NARDL model, which empirically confirm the presence of asymmetric effects of finance and insecurity on labour productivity. This is evident from the fact that the coefficient of the independent variables differ in both sign and magnitude. Thus, the results indicate that positive and negative changes in the independent variables have differing dimensional effects on labour productivity, ceteris paribus. In addition, the short-run estimates reveal several noteworthy insights.

The error correction term [ECM (−1)] reflects the speed of adjustment and captures the endogenous response of the dependent variables to short-run shocks. The coefficient of the ECM is −0.200, indicating the existence of co-integration and stability among the variables in the model. This suggests that approximately 20% of the deviation from long-run equilibrium is corrected annually following a short-run disturbance. The coefficient is statistically significant at the 1% level. The adjusted R-squared indicates that the explanatory variables account for approximately 83.7% of the total variation in agricultural labour productivity. The Prob. (F-statistics) value of 0.000 reported in Table 7 further confirms that the estimated model is well specified and that the independent variables collectively explain a significant proportion of the variance in the dependent variable.

It can be observed that the finance components took different lag orders, supporting adjustment asymmetry. This implies that the time it takes for agricultural labour productivity to respond to positive changes in finance and insecurity is different from the time it takes to respond to negative changes in finance and insecurity. Specifically, the coefficients of lnGEXP revealed that a one per cent increase in government expenditure generates a 0.118 increase in agricultural labour productivity, while a one per cent decrease in lnGEXP generates a 0.244 decrease in agricultural labour productivity, all other things being equal. Likewise, a one per cent decrease in lnCBCR caused approximately a 0.286 fall in labour productivity, while a one per cent increase resulted in a 0.098 increase in agricultural labour productivity. Likewise, a one per cent decrease in lnMFBL caused about a 0.046 decrease in agricultural labour productivity, while a one per cent increase in lnMFBL led to a 0.122 increase in agricultural labour productivity. A one per cent increase in lnACGS led to a 0.217 increase in labour productivity, while a one per cent decrease in lnACGS caused agricultural labour productivity to decrease by 0.212. A one per cent increase in lnDTHS caused agricultural labour productivity to decrease by 0.136, while a decrease in lnDTHS resulted in a 0.380 increase in agricultural labour productivity. The economic intuition behind these estimates is that positive changes in finance led to an instantaneous increase in agricultural labour productivity, while a negative change led to an immediate reduction in agricultural labour productivity. Also, a rise in insecurity was accompanied by a significant decline in agricultural labour productivity, while a decrease brought about a spontaneous increase. Simply put, short-run increases in insecurity diminish the potential positive effect of finance on agricultural labour productivity. In the short term, insecurity significantly negatively impacts finance-led agricultural labour productivity by deterring investment, disrupting supply chains, and reducing the availability of labour. This leads to a decline in agricultural labour output, increased post-harvest losses, and further exacerbates existing food security challenges.

The study carried out a diagnostic and reliability test to ensure that our estimations from the NARDL are reliable. The results are reported in Table 8.

Table 8 reports various statistical and econometric tests. From the diagnostics and reliability tests, it can be concluded that the long-run and short-run dynamic estimates from the NARDL model are free from econometric and statistical problems. Specifically, the estimated results from the NARDL model are free from heteroskedasticity, serial correlation, and functional form, and they are also normally distributed since all the probability values are greater than 0.05. In addition, the CUSUM and CUSUMSQ graphs (see Fig. 2 and 3) also reveal that agricultural labour productivity was stable in the period from 1992 to 2023, because the plots of the cumulative sum and cumulative sum of squares lie within the 5 per cent critical bound. However, the Wald test estimates for both the short and long runs are statistically significant, supporting the long-run and short-run asymmetric effect of finance amidst insecurity changes. Thus, the observed asymmetries between finance and labour productivity in the Nigerian agricultural sector support the analysis of insecurity as a moderating factor.