The exploration of the linear relationship between exchange rates and agricultural exports is a significant topic in trade and can impact trade dynamics globally. Various studies have identified both the direct and indirect effects of exchange rate fluctuations on export performance, particularly in developing countries where agriculture plays a vital role in economic activity. Using the conventional ordinary least squares (OLS) models, several researchers have looked at the close link between agricultural exports and exchange rates. For example, Gizaw et al. (2022) used an extended generalised Cobb–Douglas production function model to offer important insights into the relationship between coffee exports and the economic growth of Ethiopia. The researcher investigated the connection between Ethiopia’s economic growth and coffee exports between 1980 and 2017 using a vector error correction model (VECM). According to the researchers’ findings, Ethiopia’s economy has grown significantly because of coffee exports. The researchers recommend that improving the coffee industry’s efficiency and enhancing coffee beans’ quality prior to exporting could help Ethiopia raise more money to support its economic expansion. A similar study by Poku (2017) indicated that exchange rates significantly affect domestic prices and influence the supply and export levels for both cocoa and maize in Ghana.

Mpungose, Ngubane and Sekome (2023) revisited the effects of exchange rates on imports and exports in SA with the use of the vector auto-regression (VAR) method. The study revealed that fluctuations in exchange rates significantly impact exports, imports and inflation, ultimately influencing the trade balance and economic growth. Therefore, this research analysed agricultural export flows from SA to SADC economies, providing insights into the multifaceted effects of exchange rate dynamics on trade performance in the agricultural sector.

Pasricha (2020) analysed trade data and the country’s foreign exchange rate annual values from 1991 to 2019 to investigate how currency rate fluctuations affect India’s international trade. Using the OLS regression model, the study found a connection between India’s foreign trade and exchange rates. Furthermore, utilising quarterly data spanning from the first quarter of 2002 to the first quarter of 2019, Karahan (2020) discovered an inverse casual of exchange rate fluctuations on economic growth using the Granger causality test, the Johansen cointegration test and innovation accounting techniques. The findings support the arguments of structuralist economists.

In their article, Edwards and Hlatshwayo (2020) argue that South African exporters have relatively lower responses to exchange rate fluctuations, which means that a decline in trade may only deepen this insensitivity and reduce export volumes, even when currency depreciation offers a price advantage. This is supported by a study by Aye et al. (2015), which shows that exchange rate volatility has a negative effect on export performance in SA, meaning that reduced trade might further enhance the risk aversion of exporters. Also, exchange rate volatility is equally damaging to intra-SADC, as Pamba (2023) observes that exchange rate volatility has a large impact on the competitive position of exports, meaning that a decrease in intra-SADC trade may well be associated with increased volatility and uncertainty. In the same vein, Seti (2023) identified constraints that may affect intra-African agricultural trade and therefore may slow down the capacity of SA to unlock its agricultural export opportunities, which will, in turn, deepen the effects of reduced trade on the region’s economy, regional integration and growth. Therefore, based on the theoretical framework of regional integration, one can assume that the elimination of trade barriers in the Southern African Development Community-Free Trade Agreement (SADC-FTA) as well as the enhancement of cooperation among member countries might lead to an improvement of the trade and economic results (Moyo, Kwarambo & Nchake 2020). The fluctuations in exchange rates cannot be overlooked, as they significantly influence the trade balance of SA and other SADC nations, subsequently impacting the export competitiveness of agricultural products.

These studies have added to the vast number of linear studies in the literature that showed that, in most economies, exchange rates should be stable to maintain and enhance exports. It is significant to indicate that this relationship is affected by various factors such as domestic economic conditions, regional conditions, market dynamics, macroeconomic factors, trade agreements and political contexts. Therefore, there exists an extensive body of work within the literature review of the linear relationship of exchange rates on agricultural exports in developed and developing economies, which has been presented in studies by Pamba (2023), Edwards and Hlatshwayo (2020), Ali (2020), Karahan (2020), Burakov (2016) and Aye et al. (2015).

This study examines the effect of exchange rates on agricultural export from SA to SADC. The study adopts and modifies the model of Machado and Silva (2019), where as they introduced an augmented panel quantile regression (QR) known as the MMQR technique considering fixed effects. Their estimation of the conditional quantiles can be written as Equation 1:

In equation 1, Q_Y (τ│X_it element is the dependent variable that denotes the scale, while Xit′β+Zit′γq(τ) represents the scale estimate of the fixed effect. Equation 1 of the MMQR is adopted and modified as equation 3.2 to examine the objective of the study as Equation 2:

where Q_Y (τ│X_it is agricultural exports denoted as LGEX_ij, the real effective exchange rates is REER_ij, SA economic activity is GDP_it, SADC gross domestic product (GDP) is, GDP_it, POP_it is SA population, POP_it is SADC population and SADC inflation is INF_ij. All these explanatory variables are chosen to explain the objectives and gap of the study.

The abovementioned equation 2 will be modified in equation 3.3 into natural logarithms except for real effective exchange rates REER_ij and INF_ij as their in-percentage form and written as Equation 3:

The study examines the relationship between exchange rates and agricultural exports across different points in the distribution.

In this study, a balanced panel annual dataset was utilised to model exchange rates and agricultural export flows from SA to selected SADC economies for the period 2010 to 2022. The dataset includes data from 13 SADC economies, namely Angola, Botswana, Democratic Republic of Congo, Lesotho, Madagascar, Malawi, Mauritius, Mozambique, Namibia, Seychelles, Swaziland, Tanzania and Zambia. Data for Zimbabwe and Comoros were excluded because of the unavailability of data for the exchange rate. The description, data sources and selected SADC economies are depicted in Table 1:

The panel QR technique was initially proposed by Koenker and Bassett (1978) to estimate the dependent conditional mean and variance with respect to explanatory parameters (Rehman et al. 2021). This method allows for a more detailed examination of the relationship between agricultural exports and exchange rates. This method will show how this relationship varies across different quantiles of the distribution (Koenker 2004).

Panel QR is considered a more robust alternative to traditional techniques such as OLS, as it can capture heterogeneous effects across various quantiles of the data distribution (Sarkodie & Strezov 2019). In this study, the utilised MMQR with fixed effects proposed by Machado and Silva (2019) is employed to model the relationship between exchange rates and agricultural exports within SA and selected SADC economies. This methodology is chosen in this study for its ability to account for the covariance effects of conditional heterogeneity that may be present in the analysis, providing a more accurate understanding of the dynamics between exchange rates and agricultural exports in the region (Machado & Silva 2019).

The OLS regression has certain limitations, as it primarily focuses on the mean of the dependent variable, assuming a linear relationship between the dependent and independent variables, and, as a result, the MMQR will be employed. This technique was introduced by Koenker and Bassett (1982), Koenker and Zhao (1994) and Zhao (2000), who incorporated the estimates of location and scale functions. This methodology is an extension of OLS regression. The difference between the traditional OLS and QR is that it employs and evaluates the conditional median of various responded quantiles and not following the linear relationship. This approach allows for a more flexible and robust modelling strategy compared to traditional least squares regression. Therefore, panel QR is more vigorous to the prevalence of outliers in estimations and is also useful in explaining the weak association of the conditional means of two variables (Anwar et al. 2021).

Machado and Silva (2019) introduced an augmented panel QR known as the MMQR technique. Considering fixed effects, this model will be adopted and modified to examine the objective of the study. Methods of moments quantile regression is a statistical method that combines moments estimation with QR to provide more efficient and robust estimates of parameters. The regression considers the individual effects on exogenous explanatory variables for the location scale model as Equation 4:

where {σ_i + Z′_itγ > 0 = 1 and (α, β′, δ, γ′)′ represents the parameters of the probabilities and discrete fixed effect is represented by (αi, δi), i = 1, 2, …., n. Z denotes the K-vector of covariates X for location scale, which are various modifications of j. The model can be written as Equation 5:

where Z is a K-vector of the known differentiable with probability 1 that transforms the components of X with an l.

The MMQR by Machado and Silva (2019) is the same as in equation 1, and the model can be written as Equation 6:

Equation 7 is explained in equation 1. Therefore, this equation model is modified as stated in equations 2 and 3:

where ρ_τ (A) = (τ – 1)AI{A ≤ 0} + τAI{A ≥ 0} denotes the check function. The Û = Ȓ_it/(δ^̂_i + Z′_itγ̂) equation represents the standardised residuals and estimates the τ − tℎ sample quantile. Lastly, the study will conduct the quantile slope and symmetric test to access the asymmetric in the analysis.

An application for full ethical approval was made to the Economic and Management Sciences Research Ethics Committee (EMS-REC), and ethics consent was received on 25 November 2020 (reference no: NWU-00924-20-A4).

The descriptive statistics are significant to analyse the summary of data of the variables used in the study. Also, estimating the descriptive analysis is significant in ensuring the quality and reliability of the data before conducting more advanced econometric analyses. The study used 169 observations, which constitutes a large dataset, and the scale variables can be shown in summary in Table 2:

Table 2 outlines the results of the descriptive statistics of the analysis. Descriptive statistics help in providing a concise summary of the data. Measures such as mean, median, mode, standard deviation, skewness and kurtosis offer an overview of the central tendency, dispersion and shape of the data distribution. On average, the real effective exchange rates (REERij) are 98.83, which is the highest varying with a minimum of 65.33 and maximum of 148.47, while inflation (INFij) has an average value of 7.433, ranging from a low of -10 to a high of 30.7. Also, the gross domestic product for SADC economies (LGDPij) has a mean of 23.20 with minimum and maximum values of 20.71 and 25.65, respectively. This is followed by agricultural exports (LGEXij) as the second lowest average of 13.63, which ranges between 10.73 and 15.58. Lastly, the domestic population of SA (LPOPi) is around 19%, and the population of SADC’s (LPOPij) average is 15%; the GDP of SA (LGDPi) is 25.33% for the period of 2010 to 2022. The higher population in SA at 26.76%, compared to the SADC economies, could be attributed to SA being the economy with the largest share of GDP in the region from 2010 to 2022.

Table 3 shows the correlation results as follows:

The correlation matrix will enable the study to estimate whether the dependent and explanatory variables used in the study correlate. This test examined multicollinearity in the study, and all the variables indicated a weaker correlation. The results in Table 3 show that agricultural exports are correlated with the mentioned explanatory variables. A coefficient between 0.10 and 0.30 indicates a weak correlation, whereas a coefficient from 0.30 to 0.60 signifies a moderate correlation. A coefficient of 0.70 to 0.90 denotes a strong correlation. The correlation of agricultural exports with real effective exchange rates (REERij) coefficient is 0.24 is negative and weak, while LGDPi, LGDPij and INFij have weaker negative correlations of 0.03, 0.28 and 0.08 as shown in Table 3. However, LPOPi and LPOPij show a positive weaker correlation with a coefficient 0.03.

Table 4 presents the Pesaran’s test to determine the existence of cross-sectional dependence of the variables in the study.

The Pesaran’s test of cross-sectional independence in Table 4 yielded a test statistic of 1.752 with a p-value of 0.0798. The null hypothesis (H0) for Pesaran’s test states that there is no cross-sectional dependence among the residuals, while the alternative hypothesis posits the existence of cross-sectional dependence (Pesaran’s 2004). The results suggest that there is some evidence to reject the null hypothesis of cross-sectional independence at the 10% significance level.

This study shows the preliminary results of the relationship between exchange rates and agricultural exports across different points in the distribution from SA to SADC. The results include unit roots tests, normality test, slope homogeneity test, cointegration tests, MMQR test, quantile graphs output, slope equality test and symmetric quantile test.