Foreign Exchange as a Binding Constraint on Agricultural Modernization: Evidence from Nigeria’s Rice Sector (2000–2023)

Two theoretical frameworks underpin this study. The Growth Diagnostics Framework of Hausmann et al. (2008) argues that growth is constrained by the most binding bottleneck in any production system. Applied to Nigeria’s rice sector, it predicts that interventions targeting credit, extension services, or land development will underperform unless the fundamental constraint on input procurement – foreign exchange access – is simultaneously addressed. The Dual Gap Model of Chenery and Bruno (1962) established that developing countries face both a savings gap and a foreign exchange gap as structural limits on growth. In Nigeria’s context, the foreign exchange gap directly constrains importation of productivity-enhancing inputs, making reserve adequacy a prerequisite for agricultural modernization. Purchasing Power Parity theory, formalized by Cassel (1918), further establishes that naira depreciation systematically raises the domestic cost of internationally traded inputs, compressing farm-level profit margins in the input-intensive modern rice sector.

The empirical literature on FX constraints and agricultural production in Nigeria consistently finds that macroeconomic conditions shape sector performance, though the specific binding constraint mechanism has been underexplored. Studies on rice production and macroeconomic linkages have progressed from early error correction approaches to ARDL cointegration methods. Nkang et al. (2006) first demonstrated long-run cointegrating relationships between macroeconomic variables, rice production, and food security in Nigeria, providing methodological precedent for this study, but their analysis precedes the major foreign exchange deterioration of the post-2014 period. Ammani (2013) documented a significant positive association between foreign exchange availability and rice production dynamics, with estimated elasticities of 0.6–0.8, providing direct antecedent evidence. Ekundayo (2023), applying ARDL methodology, confirmed long-run rice production-macroeconomy relationships but did not isolate reserve adequacy as the operative constraint. These studies converge on macroeconomic drivers of rice sector performance but leave the specific reserve channel unquantified over the full 2000–2023 period.

Research on specific constraints and policy instruments reveals further gaps. Akinsola et al. (2025) found that rice importation affects domestic production through substitution dynamics, but did not examine the foreign exchange reserve mechanism through which import capacity itself is determined. Edwards et al. (2023) identified input supply chains and macroeconomic conditions as critical system drivers through fuzzy cognitive mapping, consistent with the binding constraint hypothesis, but without econometric quantification. The resource curse and Dutch disease literature – notably Sachs and Warner (1995) – provides the mechanism through which oil revenue dependence crowds out the tradeable agricultural sector, explaining the negative GDP-production relationship observed in this study. The existing literature thus confirms that macroeconomic conditions shape Nigerian rice sector performance and that foreign exchange availability is a plausible binding constraint, but leaves unestablished the causal elasticity between reserve adequacy and production over the full post-2000 period. Table 1 summarizes the key empirical studies and the specific gaps this study addresses.

The study area is the Federal Republic of Nigeria. With a population exceeding 220 million and a nominal GDP of approximately $477 billion in 2023, Nigeria is the most populous nation and the largest economy in Africa (World Bank, 2024). Its macroeconomic profile is characterized by dependence on crude oil exports, which contribute approximately 80.6% of total export revenues (Nairametrics, 2024), creating structural vulnerability to commodity price cycles that directly determine foreign exchange availability. Rice is cultivated across diverse agroecological zones – irrigated fadama systems in the North-West and North-Central; rain-fed upland systems in the North-East; and swamp and upland ecologies in the South – with a common trend towards increasing input intensity and therefore greater dependence on imported, foreign-exchange-priced inputs (FAO, 2023).

Secondary annual time-series data were employed, spanning 2000 to 2023 (n = 24 observations). Foreign exchange reserves (USD millions) and the official nominal exchange rate (NGN/USD) were obtained from the World Bank World Development Indicators (World Bank, 2024) and cross-validated against the Central Bank of Nigeria Statistical Bulletin (CBN, 2024). Domestic rice production and import volumes (metric tons) were sourced from the FAO Corporate Statistical Database (FAO, 2024). Real GDP in constant local currency units was sourced from World Bank WDI (World Bank, 2024). All continuous variables were transformed to natural logarithms. All estimations were conducted in EViews 12.

Unit root tests were conducted using the Augmented Dickey-Fuller (ADF) procedure (Dickey and Fuller, 1979) under intercept-only and intercept-plus-trend specifications, with lag length determined by the Akaike Information Criterion. The ARDL bounds testing approach of Pesaran et al. (2001) was employed to test for long-run cointegration and to estimate long-run and short-run coefficients simultaneously. Three properties justified this choice: consistency in small samples; accommodation of the mixed I(0)/I(1) integration order confirmed by unit root testing; and direct generation of an Error Correction Model (ECM) quantifying adjustment speed. The domestic rice production model is specified as: (1) LNPRODt=β0+β1DEXRATEt+β2LNRESt+β3LNGDPt+β4LNIMPt−1+εt… \matrix{{{\rm{LNPRO}}{{\rm{D}}_{\rm{t}}} = {\beta _0} + {\beta _1}{\rm{DEXRAT}}{{\rm{E}}_{\rm{t}}} + {\beta _2}{\rm{LNRE}}{{\rm{S}}_{\rm{t}}} + } \cr {{\beta _3}{\rm{LNGD}}{{\rm{P}}_{\rm{t}}} + {\beta _4}{\rm{LNIM}}{{\rm{P}}_{{\rm{t}} - 1}} + {\varepsilon _{\rm{t}}} \ldots } \cr } where LNPROD is the natural log of domestic rice production; DEXRATE = Δln(EXRATE_t) approximates the annual percentage rate of naira depreciation and enters as I(0); LNRES is the log of foreign exchange reserves; LNGDP is the log of real GDP; LNIMP_t−1 is the log of lagged rice import volumes; and ε is the error term. The AIC selected an ARDL(2,1,2,2,0) specification. Model adequacy was assessed through the Breusch-Godfrey (BG) LM test for serial correlation and the Breusch-Pagan-Godfrey (BPG) test for heteroskedasticity.

This study employs the asymptotic critical values of Pesaran et al. (2001). Given the annual sample of n = 24 observations, the finite-sample adjusted critical values of Narayan (2005) were additionally applied. For Case II with k = 4 regressors and n ≈ 30, Narayan’s (2005) upper I(1) bound is approximately 5.018 at the 5% significance level. The study’s F-statistic of 5.682 exceeds this adjusted bound, confirming the robustness of the cointegration finding to small-sample correction.

A potential source of endogeneity in this framework is reverse causality between foreign exchange reserves and rice production. This concern is substantially mitigated by two features of the Nigerian context. First, Nigeria’s foreign exchange reserves are overwhelmingly driven by crude oil export revenues (Nairametrics, 2024), which are exogenous to domestic rice sector performance. Second, the ARDL-ECM framework explicitly models dynamic adjustment through lagged dependent and independent variables, which absorb feedback effects. The one-period lag on imports (LNIMP_t₋₁) further reduces contemporaneous simultaneity. These features mitigate, though do not fully eliminate, endogeneity concerns, which are acknowledged as a study limitation.

Table 3 presents descriptive statistics for the study variables in log-transformed form over 2000–2023. Log domestic rice production (L_PRODUCTION) ranges from 14.52 to 16.07, corresponding to growth from approximately 2.0 million metric tons in 2000 to approximately 8.9 million metric tons at the upper bound, with a mean of 15.34. Log foreign exchange reserves (L_RESERVES) has a mean of 24.15 and a standard deviation of 0.42, reflecting oil-cycle-driven variation that provides the primary identifying variation in the empirical analysis. L_EXRATE registered the widest variation (4.62 to 7.33), reflecting accelerating naira depreciation over the study period with direct implications for input procurement costs.

Table 4 presents ADF unit root test results. DEXRATE is stationary at levels (I(0)) by construction. L_PRODUCTION is marginally stationary at levels (I(0)*); all remaining variables are non-stationary in levels but stationary upon first differencing (I(1)). No variable is integrated of order two, satisfying the ARDL prerequisite. The mixed I(0)/I(1) integration order validates the ARDL bounds testing approach over classical Engle-Granger or Johansen frameworks.

The AIC selected an ARDL(2,1,2,2,0) specification with k = 4 regressors. Table 5 and Figure 1 present the bounds test results. The F-statistic of 5.682 exceeds the Pesaran et al. (2001) upper I(1) critical values at all significance levels (10%: 3.09; 5%: 3.49; 1%: 4.37) and also exceeds Narayan’s (2005) small-sample upper bound of approximately 5.018 at the 5% level, confirming cointegration under both asymptotic and finite-sample critical values.

Fig. 1.

ARDL bounds test: F-statistic (5.682) vs Pesaran et al. (2001) critical values at 10%, 5%, and 1% significance levels (k = 4 regressors)

The F-statistic exceeds all I(1) upper bounds.

Source: Authors’ computation from EViews 12 (2025).

Table 6 and Figure 2 present the long-run ARDL coefficient estimates. L_RESERVES is positive and highly significant (1.684, SE = 0.326, p < 0.001). L_GDP is negative and highly significant (−0.340, SE = 0.075, p = 0.001). DEXRATE (−0.521, p = 0.178) and L_IMPORTS(−1) (0.011, p = 0.470) are not statistically significant at conventional levels.

Fig. 2.

Long-run ARDL coefficient estimates with 95% confidence intervals. Filled squares: significant at p < 0.05; diamonds: not significant. Dashed vertical line at zero for reference.

Source: authors’ computation from EViews 12, 2025.

Table 7 presents the ECM results. R² = 0.861; Adjusted R² = 0.806. The error correction term CointEq(−1) = −0.790 (t = −7.151, p < 0.001) is negative, significant, and less than one in absolute value, confirming a valid and stable adjustment process. Among short-run terms, D(L_GDP) = −0.152 (p = 0.013) and D(L_GDP(−1)) = +0.172 (p = 0.009) are statistically significant; all other short-run regressors are not. Both diagnostic tests confirm model validity: BG serial correlation F = 0.526 (p = 0.604); BPG heteroskedasticity F = 0.545 (p = 0.787).

The Central Bank of Nigeria should establish a dedicated, concessional foreign exchange allocation window for importers of strategic agricultural inputs – principally fertilizers, improved seed varieties, and farm mechanization equipment. The elasticity of 1.684 implies that a sustained 10% improvement in effective foreign exchange access for the sector would generate approximately a 16.8% increase in domestic rice output in long-run equilibrium. Such a window would insulate input procurement from acute scarcity episodes, stabilize input costs, and ensure that existing agricultural credit programmes are realized in actual input volumes rather than being eroded by depreciation. Complementarily, accelerating export diversification to reduce structural oil revenue dependence would address the root cause of foreign exchange volatility and provide the macroeconomic stability the sector requires.

Three limitations warrant acknowledgement. First, the annual sample of n = 24 observations is small by time-series standards. The ARDL framework is designed for small samples and Narayan’s (2005) small-sample bounds correction was applied; nevertheless, results should be interpreted with appropriate caution. Second, the national-level analysis masks subnational heterogeneity: the foreign exchange constraint may bind with differential intensity across irrigated and rain-fed production systems and across geopolitical zones with varying input market integration. State-level or farm-survey panel analysis would be valuable in future research. Third, while endogeneity concerns have been substantially mitigated by the oil revenue exogeneity of Nigeria’s reserve position and the ARDL-ECM dynamic structure, they are not fully eliminated; future research employing instrumental variable approaches would provide a more formal treatment.