Falling agricultural prices have been found to generally increase armed civil conflict in producing regions, but not the risk of the most lethal conflicts – civil wars. This column shows that the evidence for civil-war onset is consistent with the broader evidence on armed civil conflict when agricultural price indices isolate changes in world prices from changes in countries’ exports. Across 118 countries over 1957–2007, a one standard deviation decline in world agricultural prices raised the risk of civil-war onset by around 30%, with larger effects in sub-Saharan Africa. The evidence suggests that the mechanism operates through producer incomes rather than consumer prices.
Civil wars have cost millions of lives through combat, famine, and disease (Fearon and Laitin 2003, Ghobarah et al. 2003, Eck and Hultman 2007). A widely debated global driver of civil-war risk is changes in world agricultural prices. Falling agricultural prices reduce rural incomes and lower the opportunity cost of joining an insurgency (Dal Bó and Dal Bó 2011). This view is supported by case studies of the civil wars in Rwanda and Burundi in the 1990s, which started following the collapse in the world price of coffee (Verwimp 2003, Nkurunziza and Ngaruko 2005). But Bazzi and Blattman’s (2014) influential cross-country analysis finds no effect of falling world agricultural prices on the risk of civil-war onset. Their null result for civil-war onset is, in turn, an exception in Blair et al.’s (2021) meta-analysis, which concludes that falling agricultural prices generally increase armed civil conflict in producing regions.
In Ciccone (2026), I find that the evidence linking falling agricultural prices to armed civil conflict extends to civil war when the price shocks countries face are measured to isolate changes in world prices. In a global sample of 118 low- and middle-income countries over 1957–2007, a one standard deviation decline in world agricultural prices raises the risk of civil-war onset by around 30% relative to an average annual onset rate of around 2%. In sub-Saharan Africa, the increase is around 45%.
From price indices to price shocks
The civil-war exception in Blair et al.’s (2021) meta-analysis turns on how country-specific commodity price indices are constructed. Countries typically produce and export several agricultural commodities that sell at different world prices. To obtain a comprehensive measure of the world agricultural prices a country faces, individual commodity prices must be aggregated into a price index. This requires specifying commodity weights. If the chosen weights do not change over time, then changes in the country-specific price index – the world-price shock a country faces – reflect only changes in commodity prices. This is the approach introduced by Deaton and Miller (1995). Bazzi and Blattman (2014) use a different approach and aggregate individual commodity prices into a price index using commodity weights that vary over time. Year-on-year changes in their price index therefore reflect both changes in commodity prices and changes in commodity weights. The resulting price shocks no longer isolate price changes.
I examine whether the broader evidence on armed civil conflict extends to civil-war onset using four different agricultural price indices. All four use countries’ commodity exports to obtain the country-specific commodity weights. The key difference is whether these weights are time-invariant or time-varying. Three price indices use time-invariant commodity weights to isolate changes in world prices. The first assigns a weight of one to each country’s principal agricultural export over the 1957–2007 sample period and zero to all other agricultural exports, following Caselli and Tesei (2016). The second uses export weights fixed around the middle of the sample period (1980), following Deaton and Miller (1995). The third uses average export weights over the sample period, following Ciccone (2026). The fourth price index uses time-varying export weights, following Bazzi and Blattman (2014).
The empirical analysis shows that, using any of the three fixed-weight price indices, the broader evidence on agricultural prices and armed civil conflict extends to civil-war onset. By contrast, using the price index with time-varying weights, there is no effect of agricultural price shocks on civil-war onset.
The basic correlations underlying these results are illustrated in Figure 1. The figure compares average agricultural price growth over the three years preceding civil-war onset and non-onset years. For all three indices that isolate changes in world prices, agricultural prices grew less before civil-war onset years than before non-onset years. The gap is 9.8 percentage points using the principal agricultural export (p=0.02), 5.6 points using export weights fixed in 1980 (p=0.03), and 4.3 points using average export weights (p=0.06). This pattern is reversed when price growth is calculated using the index with time-varying weights, although the difference is not statistically significant.
One concern in interpreting Figure 1 is that civil-war risk in large agricultural exporters could itself affect world prices. If the specific concern is that rising civil-war risk reduces expected world supply, this would put upward pressure on world prices before civil-war onset and therefore work against the pattern for the three price indices that isolate changes in world prices (by raising price growth before civil-war onset relative to non-onset years).
Figure 1 World agricultural prices preceding civil-war onset versus non-onset years, across four export-weighting schemes
Note: Each pair of bars compares the 3-year agricultural export price growth (%, t-3 to t) preceding years t with a civil-war onset versus years t without one, for countries with at least one onset over 1957-2007. The four pairs differ in how commodities are weighted: principal export commodity (following Caselli and Tesei 2016), fixed 1980 export weights (following Deaton and Miller 1995), fixed average export weights (Ciccone 2026), and time-varying export weights (as in Bazzi an Blattman 2014). Whiskers ±1 s.e.
Why do the results differ when agricultural price indices use time-varying commodity weights? Because changes in these indices also pick up changes in countries’ export composition. I show this by decomposing the change in the time-varying-weight index into two components: one that reflects changes in world commodity prices and one that reflects changes in countries’ export composition. The price component shows that falling world prices significantly increase the risk of civil-war onset, but this effect is offset by the component that reflects changes in export weights. An agricultural price index with time-varying commodity weights combines the two components and therefore obscures the effect of world prices on the risk of civil-war onset.
There is another measurement issue. Export data for low- and middle-income countries are often missing and partly interpolated, extrapolated, or backcast. Measurement error in year-by-year export shares feeds directly into indices with time-varying weights. I show that averaging export shares over a long period reduces measurement error in the commodity weights and, under a restriction on the persistence of measurement error, can recover the causal price effect even though the resulting time-invariant weights do not track export composition year by year.
Producers rather than consumers
World agricultural prices affect not only producers but also consumers through food prices. To isolate the producer channel, I control for a country-specific consumer food-price shock following McGuirk and Burke (2020). This leaves the estimated effect of world agricultural prices essentially unchanged. I also find that the effects of world agricultural prices are substantially larger in countries whose agricultural exports are mainly cash crops such as coffee, cocoa, and cotton. Other patterns point in the same direction. The effects of world agricultural prices on the risk of civil-war onset are stronger among net food exporters and lower-income countries. Together, these findings suggest that falling world prices increased the risk of civil-war onset by depressing rural incomes.
No civil-war exception
Whether agricultural price indices use fixed or time-varying commodity weights changes the conclusion about world agricultural prices and civil-war onset. Once changes in world agricultural prices are isolated from changes in countries’ exports, the civil-war exception in Blair et al.’s (2021) meta-analysis disappears: falling agricultural prices increase the risk of armed civil conflict up to and including full-scale civil war.
The price and export data used in the analysis end in 2007, but large swings in world agricultural prices have not stopped. Two forces pull in opposite directions. I find that world prices passed through to the prices farmers receive more strongly after 1980 than before, in line with the agricultural policy reforms of the 1980s and 1990s. And stronger pass-through would make the effect of falling world prices on civil war onset larger than over 1957–2007. At the same time, incomes in Sub-Saharan Africa have risen since then, and I find this effect to be concentrated in lower-income countries. Whether falling agricultural prices have continued to raise the likelihood of civil war onset since 2007 is the natural next question.
Source : VOXeu
