Price Transmission of Tomatoes in South Africa’s Five Major Fresh Produce Markets: Vector Autoregressive Model

Tomatoes are among the most widely consumed vegetables globally, forming an essential component of both fresh diets and processed food industries. Their versatility in culinary applications and growing demand for healthy diets have contributed to their sustained global consumption. The global tomato consumption reached approximately 158 million metric tons (MT) in 2021 and is projected to increase to about 161 million MT by 2026, reflecting a modest but stable growth trend (ReportLinker. 2024). This steady increase suggests that tomatoes remain a staple commodity in global food systems, despite changing dietary patterns. Table 1 shows the top 10 global leading tomato producers. The distribution of tomato production was dominated by China, which contributed approximately 68 241 811 million metric tons, representing about 36.7% of world tomato production. This is followed by India at 20 694 000 million metric tons (11.1%), Turkey at 13 000 000 million metric tons (7.0%), United States at 10 199 753 million metric tons (5.5%), Egypt at 6 275 444 million metric tons (3.4%), Italy at 6 136 380 million metric tons (3.3%), Mexico at 4 207 889 million metric tons (2.3%), Brazil at 3 809 986 million metric tons (2.0%), Nigeria at 3 684 566 million metric tons (2.0%) and Spain at 3 651 940 million metric tons (2.0%).

Asia accounts for 62.62% of global tomato production, followed by America, Europe, Africa, and Oceania at 13.08%, 12.20%, 11.89% and 0.18%, respectively (Table 2). This indicates that global tomato production is heavily skewed towards Asia, reinforcing the region’s central role in both production and consumption dynamics. The regional distribution of production has important implications for global market dynamics. For instance, high production and consumption within the same regions suggest relatively limited reliance on international trade compared to other commodities. However, emerging markets, particularly in Africa, are experiencing gradual increases in consumption driven by population growth, urbanisation, and changing dietary preferences. For example, tomato consumption in Africa reached approximately 22 million tons in recent years and is expected to continue growing steadily in the coming decade (IndexBox, 2025).

Tomato production in South Africa reached approximately 530,843 tons in 2022 (NAMC, 2024). Between the period 2013–2022, the tomato production of South Africa experienced fluctuations; it experienced an increase of 30.93% from 2014 to 2017 and a significant decrease in production of 13.53% from 2018 to 2022 (NAMC, 2024). This decline has been influenced by factors such as high production costs, unfavourable weather conditions, and pests and diseases. South Africa planned to harvest 160,000 tons of tomatoes in 2024 for processing (WPTC, 2024). However, adverse winter conditions, with six days of persistent “black frost” in Northern Limpopo, led to a yield loss, reducing the projected harvest to 140,000 tonnes. These fluctuations underscore the sensitivity of tomato production to environmental conditions and input constraints.

Despite these challenges, the tomato industry remains a key commodity within South Africa’s agricultural economy. The industry employs approximately 25,000 to 28,000 people annually, with employment levels increasing during peak production periods (NAMC, 2024). This highlights the sector’s contribution not only to food supply but also to rural livelihoods and employment creation. In terms of market structure, the majority of tomatoes produced in South Africa are consumed domestically, largely due to their perishable nature. Between 83% and 87% of the annual crop is distributed through national and regional fresh produce markets, with just a smaller portion supplied directly to supermarket chains (PHIP, 2024). This indicates the central role of fresh produce markets in price formation and distribution within the tomato value chain.

A study by Baiyegunhi et al. (2018) examined the Durban-Johannesburg tomato market for the period of 2008–2012. The findings revealed a strong long-term relationship between prices, with price shocks adjusting back to equilibrium within approximately one month, with the Durban market adjusting faster after a shock than Johannesburg. Mandizvidza’s 2018 study on the Limpopo province tomato chain analysed farm-gate-to-wholesale and retail price links. The study found a symmetric price transmission between farm-gate and wholesale, but asymmetry between farm-gate and retail levels, wherein upward price shocks transmit more strongly (Mandizvidza, 2018). This therefore highlights the presence of markup and margin issues.

Shrestha et al. (2014) noted that poor marketing infrastructure and lack of market integration hinder cointegration in fresh produce chains, reinforcing the importance of infrastructure improvements in facilitating competitive pricing. Alemu and Ogundeji (2010) analysed price transmission in South African food markets. They found asymmetric transmission between producer and retailer prices, where retailers adjust quickly to margin increases but slowly to decreases, indicating potential market power and anticompetitive behaviour (Alemu and Ogundeji, 2010). These two similar studies noted a price transmission on data ranging from 2000–2012, making this study relevant, as it will give an up-to-date analysis from 2019–2024. Previous studies examining tomato price transmission in South Africa largely focused on one or two regional markets. No study has attempted to investigate the dynamics across all five major national fresh produce markets. This constitutes a knowledge gap that the present study aims to fill.

The monthly time series data ranging from January 2019 to December 2024 was used in this study. The data was sourced from the Directorate: Economic and Statistics Services at the Department of Agriculture. The focus was on five major fresh produce markets in South Africa, including Johannesburg, Tshwane, Bloemfontein, Cape Town, and Durban. The study uses the average monthly tomato prices per ton from the five fresh produce markets to analyse price transmission. Stata SE 15 statistical tool was used to analyse the empirical results for this study.

To analyse the price transmission of tomato on all five fresh produce markets in South Africa, the study employed a Vector Autoregressive Model. The VAR Model’s ability to capture dynamic interrelationships among multiple time series variables simultaneously made it especially suitable for this study. It allows each market to be modelled as a function of its own past values and the past values of all other markets, which therefore facilitates a comprehensive analysis of price interactions and transmission between markets. Three separate procedures were performed: (i) evaluating the stationarity of tomato prices in all five markets, (ii) examining how past prices in one market influence its own current prices as well as those in other markets, (iii) assessing the impact of price shocks in one market on tomato prices in the other markets over time.

The Augmented Dickey-Fulker (ADF) test, originally developed by Dickey and Fuller (1981), is used to determine whether a time series contains unit roots, which would indicate non-stationarity. In this study, the ADF was used to assess the stationarity of the data series. The null hypothesis of a unit root (i.e., time series data is non-stationary) was tested against the alternative hypothesis that data series are stationary. The ADF test builds on the Dickey-Fuller (DF) test, which checks for the null hypothesis that δ > 0 against the alternative hypothesis that δ > 0 in the following equation: (1) ΔYt=δYt−1+εt \Delta {Y_t} = \delta {Y_{t - 1}} + {\varepsilon_t} where:

• Δ – is the first difference operator

• Y_t – is time series data and

• ε_t – is a random error term.

If δ is found to be zero, the conclusion is that the time series Y_t is non-stationary. If δ is negative, Y_t is stationary (Dickey and Fuller, 1979).

The DF test assumes that the error terms are independently and identically distributed. However, this is an assumption that is not frequently satisfied in economic time series data. Therefore, it is a limited or low power test (Gujarati and Porter, 2009). The ADF test adjusts the DF test to account for possible autocorrelation in the error terms by adding the lagged difference terms of the regression, as shown in equation (2) (2) ΔYt=α+δYt−1+γt+∑λtΔYt−1+εt \Delta {Y_t} = \alpha + \delta {Y_{t - 1}} + \gamma t + \sum {\lambda_t}\Delta {Y_{t - 1}} + {\varepsilon_t} where:

• Δ – is the first difference operator

• Y_t – is time series data

• α – is the intercept

• ΔY_t–1 – are the lagged difference terms of the time series data

• ε_t – is a random error term.

Table 3 presents the descriptive statistics of the price per ton of tomatoes from five major fresh produce markets in South Africa.

Bloemfontein has the highest average price per ton (R10178.21), with the second highest average price per ton being Johannesburg at R9226.68. Bloemfontein also has the widest spread (min–max) at R18 590.85, therefore showing notable volatility. This may stem from smaller scale and lower liquidity, which can increase price shocks. These results are in line with those of Chikazunga et al. (2008), who highlighted how smaller markets often face greater volatility due to few agents/buyers. Figure 1 illustrates how the tomato prices of all the five fresh produce markets evolve over time.

Fig. 1.

Tomato prices of five fresh produce markets in 2019–2024

Source: own elaboration.

Table 4 represents the results of the unit root test, indicating whether the variables are stationary or non-stationary.

The results show that all market price variables (Johannesburg, Tshwane, Bloemfontein, Cape Town, and Durban) are statistically significant at the 1% level, both in terms of levels and first differences. This implies that the null hypothesis could not be accepted, as non-stationarity occurs at >5%. The strong rejection of the unit root hypothesis across all markets highlights the stability of price dynamics within the selected fresh produce markets. This finding implies that price shocks in these markets are not persistent over time, but rather tend to revert to equilibrium levels.

Table 5 represents the results of the suitable VAR lag that was selected.

The results in Table 5 reveal that lag 1 was significant, with the lowest AIC (83.128), HQIC (83.516), and FPE (8.7e+29). Lag 0 was not significant but has the lowest SBIC (83.832) with lag 4 being significant with a high AIC (83.147), HQIC (84.505), FPE(9.8e+29), and SBIC (86.574). Liew (2004) emphasizes that AIC and FRE perform best for identifying the true lag order, minimizing underestimation while maintaining consistency. Therefore, lag 1 was considered. The selection of lag 1 has meaningful economic implications. It suggests that price adjustments across markets occur relatively quickly, with current prices being influenced primarily by prices from the immediately preceding period. This reflects short-term transmission mechanisms and supports the notion of rapid information diffusion across South Africa’s fresh produce markets. Appropriate lag selection is crucial because the Johansen cointegration test is sensitive to lag specification; incorrect lag choice may lead to biased or misleading inference (Baek and Koo, 2008).

Table 6 presents the results of the cointegration test to check cointegration among variables.

The trace test indicates five cointegrating relationships at the 5% level. The maximum eigenvalue test also indicates five cointegrating relationships at the 5% level. Both the trace and max-eigenvalue test reject the null hypothesis that there is no cointegration amongst the variables. The existence of multiple cointegrating relationships implies that the fresh produce markets are highly integrated and share common long-run equilibrium price dynamics. This suggests that prices in one market are not determined in isolation but are linked to prices in other markets over time. Such integration is indicative of efficient spatial arbitrage, where price differences across markets are corrected through trade and market interactions.

Table 7 presents the results of the vector autoregression model, showing how the previous tomato price of a market influences its own current price and that of other markets.

The VAR results revealed that the Johannesburg market’s average tomato price per ton from the previous month positively and significantly influences the current month’s average tomato price in its own market (Coef = 1.868, p = 0.012), as well as in Tshwane (Coef = 1.540, p = 0.024), Bloemfontein (Coef = 2.200, p = 0.006), Cape Town (Coef = 1.329, p = 0.051), and Durban (Coef = 1.549, p = 0.034) market. This means that a 1-unit increase in Johannesburg’s tomato price in the previous month led to increases across all five markets, namely, Johannesburg, Tshwane, Bloemfontein, Cape Town and Durban market at 1.868, 1.540, 2.200, 1.329 and 1.549, respectively. By revealing Johannesburg market to be the price leader in the fresh produce market, these results align with the Competition Commission’s (2025) report, showing that Johannesburg market still maintains a dominant share of trade volumes and serves as a key hub in the country’s fresh produce market. These consistent cross-market effects signal Johannesburg market’s leadership role in price formation, likely due to its total size, buyer diversity, and a historical role as the country’s largest NFPM (Johannesburg fresh produce market). The findings presented by Baiyengunhi et al. (2018) further revealed that Johannesburg market is the leading terminal in tomato price transmission to Durban.

The previous month’s price in Bloemfontein market was found to negatively and significantly influence not only its own market (Coef = −0.640, p = 0.089) but also the current prices in Johannesburg (Coef = −0.760, p = 0.028), Tshwane (Coef = −0.791, p = 0.013), and Durban (Coef = −0.668, p = 0.044) market. A 1-unit increase in Bloemfontein market’s previous month price causes a decrease in the price of its own market, Johannesburg, Tshwane, and Durban market at 0.640, 0.760, 0.791, and 0.688, respectively. This negative influence may reflect local supply surges or institutional pricing practices distinct from those of the larger metropolitan markets.

Cape Town market’s previous month prices influence its own current tomato price (Coef = 0.351, p = 0.097). A 1-unit increase in its own previous month price increases its current price by 0.351. The market in Cape Town seemed to be independent, with only marginal evidence of price influence from Johannesburg market only. These findings align with the study by Nyopa and Khumalo (2022), who highlighted that markets with a high own variance share but low cross-market spillovers show relative stability and insulation.

In Bloemfontein market, the average tomato price from two months ago negatively and significantly influences the current average tomato price in its own market (Coef = −0.806, p = 0.71), as well as in Johannesburg (Coef = −0.749, p = 0.067) and Tshwane (Coef = 0.658, p = 0.053) market.

Table 8 to Table 10 present the results of the diagnostic test that was done using VAR’s Lagrange-Multiplier test, Jarque-Bera test and Eigenvalue stability condition.

Table 8 reports the Lagrange-Multiplier test results for serial correlation. The p-values for both lag 1 (0.416) and lag 2 (0.603) exceed the 5% significance level, indicating that the null hypothesis of no autocorrelation cannot be rejected. This implies that the residuals are serially uncorrelated and behave as white noise. The absence of autocorrelation is a critical requirement for a well-specified VAR model, as serially correlated residuals would indicate omitted dynamics or model misspecification. These findings suggest that the selected lag structure is appropriate and sufficiently captures the dynamic interactions among the variables. Similar studies confirm that p-values greater than 0.05 in LM tests indicate no residual autocorrelation and validate the model specification (Saputri et al., 2026). The Jarque-Bera test results presented in Table 9 reveal that p-values for all markets–Johannesburg (0.942), Tshwane (0.768), Bloemfontein (0.733), Cape Town (0.865), and Durban (0.270)–were higher than 5%, implying that the residuals are normally distributed. The normality of residuals enhances the reliability of statistical inference, including hypothesis testing and confidence intervals. This suggests that the estimated VAR coefficients are statistically valid. The literature similarly emphasises that normally distributed residuals strengthen the credibility of econometric results and inference (Saputri et al., 2026).

The VAR’s stability condition requires that the moduli (absolute values) of all eigenvalues from the model’s companion matrix must be less than 1. As depicted in Table 10, all five eigenvalues lie inside the unit circle, with modulus 0.588454, 0.588454, 0.259789, 0.205821, and 0.050944. Since all moduli were <1, this implies the VAR is stable. A stable VAR model ensures that shocks to the system dissipate over time rather than exploding, thereby guaranteeing meaningful dynamic analysis (Mapila, 2025).

Figure 2 presents the results of the Impulse Response Function to identify how one price shock in one market influences the price in another market.

Fig. 2.

Results of the Impulse Response Function

Source: own elaboration.

Johannesburg market shocks lead to an average increase of R 3 000 in Bloemfontein and Durban markets within one month. By contrast, Tshwane and Cape Town markets responded more moderately. The peak impact is immediate, and the effect fades to baseline by the fifth or sixth month, indicating efficient short-run reversion in a stable system. The 95% confidence bands exclude zero during these initial months, confirming that the shock effects are statistically significant. In contrast, a shock in Bloemfontein market mainly affects the same market, with negligible, non-significant spillovers elsewhere. These findings therefore support Johannesburg’s role as the price leader and highlight asymmetric inter-market transmission, which is consistent with patterns observed in stationary VAR applications.