Food Prices and Social Unrest (1990–2026): The Speed of the Rise, Not the Level Alone
Each of the three documented waves of food-linked unrest since 1990 (the 2008 food riots, the Arab Spring of 2010-2011, the 2022 crises) struck while world food prices were both above a stress level and rising more than 23% a year. In September 2026 prices are above that level again, and rising 15.8% a year: the fastest pace ever recorded in an elevated month outside an unrest window.
This study tests the most-cited claim in the food-security literature, the 2011 “food price threshold” of the New England Complex Systems Institute (NECSI), against the World Bank’s monthly food price index from 1990 to 2026. The level holds as a backdrop: among the months when prices were already elevated, what separates the unrest episodes from the calm years is the speed of the rise. The latest month narrows that gap more than any before it.
This page documents the relationship between the World Bank food price index (Commodity Price Data, “The Pink Sheet”) and documented episodes of food-linked social unrest from 1990 to 2026. It records where the index sat, in nominal and inflation-adjusted terms, during the three major unrest clusters of the period, and the twelve-month rate of change at each. The dataset pairs the monthly food index and its three sub-indices (grains, oils and meals, other food) with a curated, sourced list of documented unrest events, and derives a reproducible classification of the months when prices were both elevated and rising fast. It transposes the NECSI (2011) threshold onto the World Bank index by the same multiple of the same base period, extends the analysis out of sample, and tests the reciprocal case: fast rises that occurred far below the stress level.
Across 1990-2026 the World Bank food price index sat above its stress level (120.1, that is 2.1 times its 2002-2004 average, the multiple behind NECSI’s 2011 threshold) in 59 of 441 months (13.4%), far more often than unrest followed. What distinguishes the three documented unrest clusters is speed: at each flashpoint the index was rising +23.3% to +69.0% year-on-year, while the 19 elevated months outside any unrest window rose at most +15.8%. That ceiling was set by the latest month, September 2026; before 2026, no elevated-but-calm month had risen faster than +8.1%. The current reading is above the stress level and 4.2 points under the 20% surge line, the narrowest margin of any elevated month outside an unrest window. Speed alone is not sufficient either: the fast rises of 1996 and 2004, far below the stress level, were not followed by comparable unrest. This is a descriptive pattern across three clusters, with no causal or forecasting claim; the index measures world benchmark prices, and local retail prices can differ (see Methodology and Limitations).
World Bank food index, nominal (2010 = 100)
12-month change, up from −6.5% a year earlier
stress level (120.1), 4.2 points under the 20% surge line
below the May 2022 record (153.1)
World Bank Pink Sheet, monthly indices updated on 2 October 2026. The dataset and the charts below run through September 2026; they are rebuilt at each monthly update.
- Every documented wave of food-linked unrest since 1990 struck while world food prices were surging, not merely while they were high. At the month of each flashpoint the World Bank food index was rising +69.0% (April 2008), +23.3% (December 2010, the month of Bouazizi’s act) and +31.8% (March 2022). Readings at each episode’s price peak, which differ, are in the table below.
- The price level alone does not discriminate: the index sat above the stress level in 59 of 441 months (13.4%), and 19 of those months fall outside any documented unrest window: March to May 2012, July 2012 to February 2013, June to November 2023, May 2026 and September 2026.
- Those 19 elevated-but-calm months rose by at most +15.8% year-on-year, against at least +23.3% at the three flashpoints: a gap of 7.5 points, and no overlap. Any surge cutoff from 16% to 38% flags the same three periods and no month outside the unrest windows.
- The latest month is the closest call on record. In September 2026 the index read 124.1, above the stress level for the second time this year, and was rising +15.8% a year, up from −6.5% in September 2025. Before 2026 the fastest elevated-but-calm month was October 2012, at +8.1%. Grains (+29.5%) and oils and meals (+21.2%) carry the rise; other food is flat (−0.4%). On the historical pattern this is the elevated-but-calm regime, 4.2 points under the surge line; it is not a forecast of unrest.
- The reciprocal test. Speed alone is not sufficient either. In April and May 1996 (+21.5% and +22.7%) and from March to May 2004 (up to +27.3%), the index rose faster than 20% a year while sitting far below the stress line, between 66.5 and 73.1. Neither episode was followed by a documented unrest cluster. On this record the elevated level is a necessary backdrop and the fast rise is what separates the episodes within it.
- Robustness disclosure: the stress level transposes NECSI’s multiple (2.1 times the 2002-2004 average, defined on another index) onto the World Bank index; the baskets differ, so it is a transposition, not an equivalence. At 2.2 times the average, December 2010 falls below the line; the sensitivity is detailed in Methodology.
- Reproducible from the World Bank Pink Sheet (CC BY 4.0) and the US CPI-U (BLS, public domain) for the real index, with a sourced event list. Eco3min grants no licence of its own on these data.
441 observations · Monthly · Jan 1990-Sep 2026 · Data: World Bank, CC BY 4.0
Methodology · Cite this dataset
monthly observations (1990-2026)
nominal record, May 2022
of months above the stress level
fastest 12-month surge on record (April 2008)
ceiling of the elevated-but-calm months, set in September 2026
documented unrest clusters (2008, 2010-11, 2021-22)
The index and the unrest record
The chart plots the World Bank food price index monthly from January 1990 to September 2026. The dashed line marks the stress level, 120.1: NECSI (2011) placed its unrest threshold at 2.1 times the 2002-2004 average of the index it used, and the line applies the same multiple to the World Bank index. The index has crossed that line repeatedly, but the red segments highlight only the months when prices were both above the line and rising at least 20% year-on-year. Those red episodes line up with the documented unrest clusters. The months above the line in gray (2012-2013, 2023, May and September 2026) are the periods when prices were high but rising slowly.
The pattern is easier to read as a contrast than as a rule. Read the series alongside its inflation-adjusted counterpart in our companion study on real commodity prices since 1960, and the underlying components in the wheat, rice and palm oil dataset pages.
The consensus: high prices, food riots
The dominant reading is intuitive and well supported. When food becomes unaffordable, populations that spend a large share of income on it are pushed toward the margin of survival, and governments perceived as responsible for food security lose their footing. History offers a long catalogue, from the European bread riots of 1848 onward.
The most-cited modern formulation is quantitative. Researchers at the New England Complex Systems Institute found that the timing of the 2008 food riots and the 2011 Arab Spring coincided with peaks in a global food price index, and identified a specific level, 210 on that index’s 2002-2004 base, above which unrest became statistically likely (Lagi, Bertrand and Bar-Yam, 2011). The finding drew attention partly because the authors had submitted a report to the U.S. government on 13 December 2010 warning of the food-price and instability link, four days before Mohamed Bouazizi’s self-immolation in Tunisia on 17 December. During the 2008 peak, the same work counted more than 60 food riots across 30 countries, ten of which involved multiple deaths.
The econometric literature has since sharpened the claim rather than dropped it. Bellemare (2015), using monthly international data and natural disasters as an instrument, finds that increases in the level of world food prices cause increases in social unrest, while price volatility does not. That is a stronger version of the level story than NECSI’s, and this page is written against it as much as against the threshold.
On a short window, the level story looks decisive: the great price peaks of the period coincided with the great unrest waves. The question this dataset asks is whether the level is what did the work, once the full 1990-2026 record is read on one consistent index.
What the data shows: speed, not level
First, the level threshold cries wolf. The World Bank food index sat at or above the stress level (120.1) in 59 of 441 months, 13.4% of the record. Those months are not confined to the unrest years: 19 of them fall outside any documented unrest window, in three stretches (March to May 2012, July 2012 to February 2013, June to November 2023) and two isolated months in 2026 (May and September). None of the 2012-2013 or 2023 months was followed by an unrest cluster comparable to 2008 or 2011; the two 2026 months are too recent to judge. Being above the level is common; unrest is not.
Second, the unrest episodes share a different signature: pace. At each documented flashpoint the index was climbing steeply: +69.0% year-on-year in April 2008, +23.3% in December 2010 (the month of Bouazizi’s act) and +31.8% in March 2022, the first full month of the war in Ukraine. Across the 19 elevated-but-calm months, the twelve-month change never exceeded +15.8%. The two groups do not overlap in pace. And the surge cutoff is not a tuned parameter: drawn anywhere from 16% to 38%, it flags the same three periods and no month outside the unrest windows. Related framing: our study on agricultural commodities.
Third, the gap is narrowing now. The +15.8% ceiling of the calm group is recent: it is September 2026, the latest month. Until 2026 the fastest elevated-but-calm month was October 2012, at +8.1%, and the gap to the slowest flashpoint was about 15 points. September 2026 cut it to 7.5 points. The index went from −8.6% year-on-year in October 2025 to +15.8% eleven months later, with the index at 124.1, its highest since September 2023. That is the observation this page is built to track: an elevated level whose pace is rising.
Fourth, the mechanism looks like a shock. A rapid surge signals a supply disruption still unfolding (a harvest failure, an export ban, a war) that outruns the ability of households and governments to adjust. A high but stable price has, by definition, already been absorbed: budgets have re-anchored, substitutions have been made, subsidies have been set. Grouping the above-stress months by documented outcome, the ones inside an unrest window ran at a mean of +24.6% year-on-year (40 months), against +1.1% for the elevated-but-calm months (19 months). Speed is the axis on which they separate.
Fifth, and this is the limit of the finding, speed on its own does not do it either. Running the test in reverse, the record contains five months when the index was rising faster than 20% a year while the level sat far below the stress line, outside the run-up to any later crisis: April and May 1996 (+21.5% and +22.7%, index at 72.2 and 73.1) and March to May 2004 (+27.1%, +27.3% and +22.7%, index between 66.5 and 69.8). Neither episode was followed by a documented unrest cluster. The honest formulation is therefore joint rather than exclusive: on this record an elevated level looks like a necessary backdrop, and the speed of the rise is what separates the episodes that erupted from those that did not. What the data contradicts is the level threshold used alone, not the relevance of the level.
The stress level is a transposition. NECSI’s 210 was defined on another global food index, on a 2002-2004 base, which makes it 2.1 times that index’s 2002-2004 average. Applied to the World Bank food index, whose 2002-2004 average is 57.194, the same multiple gives 120.1. The two indices do not weight the same baskets, so the line transposes NECSI’s level rather than reproducing it. The separation found here holds at 2.1 times; it is sensitive to the multiple, as Methodology shows.
What this dataset does not measure. The World Bank index tracks internationally traded food commodities at world benchmark prices, in US dollars. It does not measure the retail prices households actually pay, which are shaped by local currencies, subsidies, tariffs and domestic supply. Record beef prices are one current case of that domestic side, with the number of cattle in the United States at a 75-year low. During the Tunisian revolution, on-the-ground monitoring by the FAO’s GIEWS reported that domestic consumer food prices were broadly stable, and Tunisia’s consumer food price index slipped slightly between November and December 2010, even as international prices surged.
Across 1990-2026, food prices above the stress level coincided with documented unrest only when they were also rising fast. The same levels, reached slowly or held flat, did not, and fast rises from a low base (1996, 2004) did not either: on this record both conditions were present each time. In September 2026 the level is above the line and the pace, at +15.8%, is the fastest ever seen in a calm elevated month.
Where the sceptics are right
The strongest objection: a published, instrumented result points the other way. Bellemare (2015), in the American Journal of Agricultural Economics, uses natural disasters to instrument for food prices and finds that increases in the world food price level cause increases in social unrest, while price volatility does not, a result robust to real versus nominal prices, to commodity-specific indices and to alternative definitions of volatility. That is a properly identified causal estimate on a continuous outcome; this page is a descriptive overlay on three curated clusters, and where they conflict the econometric estimate carries more weight. Two things keep the present finding standing. Volatility is not speed: Bellemare’s measure is the variance of the price distribution, whereas the variable here is the signed twelve-month rate of change, and a steady fast climb has high speed and low volatility. And his level result is about increases in the level, which differs from the threshold claim tested here. The fair reading is that the two are compatible: the level matters, and among elevated months the pace of the rise is what discriminates.
The strongest mechanical objection: the index may be a symptom, not a cause. Todd Smith’s study of urban Africa argues that domestic consumer prices, not the international index, are the variable that actually moves people, and that rainfall scarcity works better as an instrument, implying that a common driver (weather, supply) can produce both the price surge and the unrest. On this view the world index is a real-time barometer of stress that happens to move first. The speed finding is consistent with that: a fast surge is exactly what a fresh supply shock looks like in the index.
The separation is fragile at the margins. It holds at the multiple NECSI used, 2.1 times the base average, and nowhere else without qualification. At 2.2 times (125.8), December 2010, the month of Bouazizi’s act, falls below the line. At 2.0 times (114.4), months just before and after the unrest windows enter the calm group, some rising as fast as +40% a year; excluding the twelve months around each window, the calm ceiling is again September 2026, at +15.8%. And the gap between calm and flashpoint months is now 7.5 points, set by a single month. A fourth surge that passes without unrest, or a slow plateau that erupts, would revise the picture.
The sample is small, and the outcome variable is curated. Three documented clusters in thirty-six years is a pattern, not a statistically validated law, and the unrest windows here are compiled from Lagi et al. (2011) and the cited sources rather than drawn from a comprehensive event database; comprehensive event data (ACLED) is licensed and not redistributed here, which is a real limitation, not a technicality. Because the windows are curated around known episodes, the fact that 40 of the 44 window months also sit above the stress level is partly a property of how the windows were drawn, not an independent result. This dataset offers a descriptive overlay of a sourced event list against a price series, without confidence intervals.
The 2022 case is the weakest of the three. The most prominent 2022 collapse, in Sri Lanka, where mass protests drove President Gotabaya Rajapaksa to flee on 13 July 2022, was largely domestic in origin: a 2021 ban on chemical fertiliser that gutted local harvests, a debt and foreign-exchange crisis, and the loss of tourism. The global surge amplified an already collapsing domestic food situation rather than causing it, which is precisely the “contributing amplifier, not sole trigger” framing this study defends.
And the level story’s own forward call did not hold, instructively. NECSI extrapolated that the rising trend would cross into “high impact even without price peaks” in 2012-2013. The index did sit above the stress level from March 2012 to February 2013, apart from June 2012, but prices were flat to falling (−0.9% year-on-year in March 2012), and no comparable wave followed. That is the level thesis’s sharpest miss, and the speed finding explains it: 2012 was a plateau, not a surge.
Unrest frequency by regime
The natural measure here is the frequency of documented unrest conditional on the price regime. With only three clusters, this is descriptive and heavily caveated: the counts are small and what constitutes a “cluster” is a judgement disclosed in Methodology.
“elevated AND surging” periods coinciding with documented unrest
elevated-but-calm months followed by a comparable wave (the two 2026 months are too recent to judge)
fast rises below the stress level (1996, 2004) followed by a comparable wave
At the 20% cutoff the “elevated AND surging” flag produces four contiguous runs (March to July 2008, December 2010 to August 2011, May to October 2021 and March to April 2022), which map onto three unrest clusters, since the 2021 and 2022 runs belong to one episode split by a dip below the cutoff. The three periods flagged are the same for any cutoff from 16% to 38%; only how they are cut into runs changes. At 15%, September 2026 itself would enter the flag.
Read directionally: when the index has been both above the stress level and rising 20% or more a year, documented unrest has coincided in every case on record; when it has been above the level but rising slowly, or rising fast from a level far below the line, it has not.
Past patterns are not predictive of future outcomes. Regime-conditional frequencies describe a small historical record across three clusters, not expected probabilities.
Levels to Watch
These reference points describe where the current reading sits against the historical record. They are descriptive markers for reading future World Bank releases, not signals or targets.
3.3% above the 120.1 stress level, the highest reading since September 2023, and 18.9% below the May 2022 record of 153.1. Being above the level has, on its own, not been associated with unrest in the historical record.
Up from −6.5% in September 2025 and +12.0% in August 2026, and 4.2 points under the 20% surge line. In the historical record, the three unrest flashpoints coincided with readings of +23.3% or more.
The 19 elevated months outside any unrest window never exceeded +15.8% year-on-year, and that maximum is the current month; before 2026 the maximum was +8.1% (October 2012). A move above +20% a year while the index stays above 120.1 would put the reading in the regime that, on this record, has coincided with each documented cluster.
Grains rose fastest at all three price peaks, with oils and meals close behind; in September 2026, grains (+29.5%) and oils and meals (+21.2%) lead again while other food is flat (−0.4%). Which component rises fastest identifies the nature of the supply shock (see the attribution chart below).
The three clusters, side by side
Each row records the World Bank food index at the episode’s price peak, in nominal and inflation-adjusted terms, its twelve-month rate of change, and the sub-index that rose fastest. The twelve-month change at the price peak differs from the flashpoint month quoted in the summary above, which is why both columns are shown.
| Episode | Peak (nominal) | Real at peak (US CPI) | 12-month change at peak | At flashpoint | Fastest sub-index at peak | Documented unrest |
|---|---|---|---|---|---|---|
| 2008 (peak Jun 2008) | 131.2 | 130.7 | +61.0% | +69.0% (Apr) | Grains (+80.4%) | 60+ food riots, 30 countries |
| 2010-11 (peak Feb 2011) | 129.3 | 127.4 | +35.5% | +23.3% (Dec 2010) | Grains (+50.1%) | Arab Spring; regional food riots |
| 2021-22 (peak May 2022) | 153.1 | 114.2 | +20.0% | +31.8% (Mar 2022) | Grains (+26.4%) | Sri Lanka collapse; scattered protests |
| September 2026 (latest month) | 124.1 | 80.3 | +15.8% | n/a | Grains (+29.5%) | None comparable to date |
Peak values are the highest nominal reading within each episode (2007-2008, mid-2010 to 2011, 2021-2022). The real index is the nominal index deflated by the US CPI-U and rebased to 2010 = 100. Fastest sub-index: the World Bank component with the highest year-on-year change at the peak month. At the December 2010 and March 2022 flashpoints, oils and meals rose faster than grains (+29.4% against +27.5%, then +41.1% against +32.9%). May 2022 rounds to +20.0% but is just under the cutoff (+19.98%), hence outside the flagged run.
What drove each surge
The 2008 spike was a grains event first, with oils and meals not far behind; in 2010-2011 and 2021-2022 the same two components led, and at the December 2010 and March 2022 flashpoints oils and meals were ahead of grains, in 2022 amid the disruption of Ukrainian and Russian exports. The common thread is the pace of the aggregate move. The individual components are tracked in the corn and sugar dataset pages, and the broader commodity picture in the commodity price hub.
Same level, opposite speed
Each dot is one month. To the right of the stress line, the same range of high price levels, the months of the red episodes sit high on the vertical axis, while the elevated-but-calm months sit lower; the highest of them is September 2026. The horizontal position (how high) does not separate the two groups; the vertical position (how fast) does. Note also the left of the plot: the months of 1996 and 2004 sit high on speed but far to the left of the stress line, and neither episode was followed by a documented cluster, which is why the reading here is joint rather than “speed alone”.
Historical turning points
2007-2008: the grains spike
The index rose at its fastest pace on record, +63.9% year-on-year in March 2008 and +69.0% in April, and peaked at 131.2 in June 2008, led by grains (+80.4% year-on-year). More than 60 food riots followed across 30 countries (Lagi et al., 2011). Prices then fell sharply as the global financial crisis took hold, dropping back below the stress level from August 2008.
2010-2011: the Arab Spring window
The index climbed back above the stress level in December 2010, rising +23.3% year-on-year that month, and peaked at 129.3 in February 2011 (+35.5%), led again by grains (+50.1%). The Tunisian and Egyptian uprisings unfolded through this window, though, as the domestic-price evidence shows, international and local retail prices diverged. The level then eased without a comparable wave, while staying above the stress line through most of 2012 and until February 2013.
2021-2022: the wartime record
Post-pandemic supply strains and then the Russian invasion of Ukraine drove the index to an all-time nominal high of 153.1 in May 2022 (+20.0% year-on-year); at the March 2022 flashpoint it was rising +31.8%, with oils and meals up 41.1%. Documented unrest was more concentrated than in 2008 or 2011, most prominently Sri Lanka’s collapse, and in that case largely domestic in origin. Prices then receded: the index stood 15.7% lower year-on-year in June 2023, while still above the stress level until November 2023.
The quiet surges: 1996 and 2004
Twice in the record the index rose faster than 20% a year without the level ever approaching the stress line: April and May 1996, at 72.2 and 73.1 on rises of +21.5% and +22.7%, and March to May 2004, between 66.5 and 69.8 on rises of up to +27.3%. Neither episode is associated with a documented unrest cluster. They are the closest thing this record offers to a falsification test of speed taken on its own, and they are the reason the finding here is stated as a joint condition.
2026: current observation
The index crossed the stress level in May 2026 (121.1, +11.8% year-on-year), slipped back below it in the summer, and stood at 124.1 in September 2026, its highest since September 2023 and 18.9% below the 2022 record. The twelve-month change has climbed from −8.6% in October 2025 to +15.8%, led by grains (+29.5%) and oils and meals (+21.2%). In real terms, deflated by the US CPI, the index is at 80.3, far below its 2008 record of 130.7, and up 11.6% in a year. On the historical pattern this is the elevated-but-calm regime at its fastest recorded pace: a high level, an accelerating pace, under the surge signature as of September 2026. This describes the record; it is not a forecast. Whether the pace goes on rising depends on the physical side (harvests, energy inputs, fertiliser), the terrain of the reading of commodities as regimes.
Methodology
The series is the World Bank food price index and its three sub-indices (grains, oils and meals, other food), monthly from January 1990 to September 2026, from the “Monthly Indices” sheet of the Pink Sheet historical file updated on 2 October 2026, in nominal US dollars with 2010 = 100. The real index divides the nominal index by the US CPI-U (all items, not seasonally adjusted, BLS series CUUR0000SA0) and rebases the result to 2010 = 100: the Pink Sheet publishes its monthly indices in nominal terms only. The BLS published no CPI for October 2025, so the real index is blank for that month, and only that month; nothing is interpolated. Year-on-year change is the twelve-month percentage change of the nominal index, computed on the World Bank’s values, which are rounded to one decimal.
The stress level and its caveat. NECSI (2011) placed its unrest threshold at 210 on a global food index with a 2002-2004 base of 100, that is 2.1 times the base-period average. This page applies the same multiple to the same period of the World Bank food index. The baskets and weights differ, so the line is a transposition, not an equivalence, and the separation is sensitive to it:
| Multiple | Stress level | Months above | Calm months | Reading |
|---|---|---|---|---|
| 2.0× | 114.4 | 92 | 49 | months next to the unrest windows enter the calm group, up to +40.0% a year; excluding the 12 months around each window, 21 calm months, ceiling +15.8% (September 2026) |
| 2.1× (used) | 120.1 | 59 | 19 | all three flashpoints above the line; calm ceiling +15.8% |
| 2.2× | 125.8 | 27 | 4 | December 2010 falls below the line; calm ceiling +8.1% |
Regime and episode definitions
The 20% surge cutoff is not tuned. Set anywhere from 16% to 38%, the “elevated AND surging” flag identifies the same three periods (2008, 2010-11 and 2021-22) and no month outside the unrest windows; only the number of contiguous runs changes. At 15% and below, September 2026 enters the flag; above 38%, the 2010-11 or 2021-22 period drops out. The 19 elevated months outside any unrest window never exceeded +15.8% year-on-year, while the three flashpoints were at +23.3% or more. Documented unrest windows are curated from the sourced event list, not derived from prices: 2008 (February to July), 2010-11 (November 2010 to November 2011) and 2021-23 (May 2021 to May 2023; the cluster is named 2021-22 after its price surge). Four window months sit below the stress level (February 2008, November 2010, October and November 2011); the alignment of the other 40 with above-stress months is partly a property of the curation, disclosed here rather than presented as a result.
Filter Definitions
Dataset Design
| Variable | Type | Unit | Source | Calculation |
|---|---|---|---|---|
| food | float | index (2010 = 100) | World Bank | direct |
| oils_meals / grains / other_food | float | index (2010 = 100) | World Bank | direct |
| food_real | float | index (2010 = 100) | World Bank, BLS | deflated by US CPI-U |
| food_yoy / food_chg6 / food_real_yoy | float | % | Eco3min | 12-month and 6-month % change |
| oils_meals_yoy / grains_yoy / other_food_yoy | float | % | Eco3min | 12-month % change |
| above_stress / surge / danger | int (0/1) | flag | Eco3min | see definitions above |
| unrest_window | int (0/1) | flag | Eco3min | curated, sourced |
Python Reproduction Code
# Reproduce the flags from the published CSV (World Bank data, CC BY 4.0) import pandas as pd url = "https://eco3min.fr/wp-content/uploads/2026/10/food-price-speed-not-level-v2.csv" df = pd.read_csv(url, parse_dates=["date"]).set_index("date") STRESS = round(2.1 * df.loc["2002":"2004", "food"].mean(), 1) # 120.1 SURGE = 20 df["danger"] = (df.food >= STRESS) & (df.food_yoy >= SURGE) # reciprocal test: fast rises that happened below the stress level df["fast_but_low"] = (df.food < STRESS) & (df.food_yoy >= SURGE)
Download the Dataset
The CSV contains the monthly World Bank food index, its three sub-indices, the real index deflated by the US CPI-U, the twelve-month and six-month changes, and the regime flags; the XLSX adds the sourced unrest-event list, a data dictionary and the source terms. Data terms: World Bank, Commodity Price Data (The Pink Sheet), licensed by the World Bank under CC BY 4.0 and its Terms of Use for Datasets; US CPI-U, U.S. Bureau of Labor Statistics, public domain; event list compiled from Lagi et al. (2011) and the cited sources. Eco3min grants no licence of its own on these data, and no endorsement by the World Bank is implied. Eco3min text and analysis: CC BY 4.0, with attribution to Eco3min Research.
This study sits within Eco3min’s commodity coverage. Explore the underlying series in the commodity price data hub.
Data Sources & References
- Primary World Bank, Commodity Price Data (The Pink Sheet), historical monthly indices (Food, Grains, Oils & Meals, Other Food), nominal US dollars, 2010 = 100, file updated on 2 October 2026, covering January 1990 to September 2026. Licence: CC BY 4.0.
- Primary U.S. Bureau of Labor Statistics, Consumer Price Index for All Urban Consumers (CPI-U), all items, not seasonally adjusted, series CUUR0000SA0, retrieved 20 October 2026. Public domain.
- Research Lagi, M., Bertrand, K. Z., & Bar-Yam, Y. (2011). The Food Crises and Political Instability in North Africa and the Middle East. arXiv:1108.2455.
- Research Bellemare, M. F. (2015). Rising Food Prices, Food Price Volatility, and Social Unrest. American Journal of Agricultural Economics, 97(1), 1-21. doi:10.1093/ajae/aau038.
- Research Smith, T. G. (2014). Feeding Unrest: Disentangling the Causal Relationship between Food Price Shocks and Sociopolitical Conflict in Urban Africa. Journal of Peace Research, 51(6), 679-695.
- Research Raleigh, C., Choi, H. J., & Kniveton, D. (2015). The Devil is in the Details. Global Environmental Change, 32, 187-199.
- Context FAO GIEWS country brief, Tunisia (31 January 2011); New Security Beat / Stimson Center (2014) on international versus local prices.
- Context USIP and UK House of Commons Library (2022) on the Sri Lankan economic crisis and the Aragalaya protests.
Limitations
- Global, not local. The World Bank index is a world benchmark price in US dollars; it does not capture the retail prices households pay, which reflect currencies, subsidies and domestic supply.
- Transposed threshold. The stress level applies NECSI’s multiple (2.1 times the 2002-2004 average) to a different index; at 2.2 times, December 2010 falls below the line, and at 2.0 times the calm group includes fast months next to the unrest windows.
- Small sample, narrow margin. Three documented clusters in thirty-six years support a descriptive pattern, not a statistically validated causal relationship; the gap between the fastest calm month and the slowest flashpoint is 7.5 points, set by September 2026.
- Curated event list, and what follows from it. The unrest overlay is a sourced but non-exhaustive list of documented events, not a comprehensive unrest dataset; comprehensive event data (e.g. ACLED) is licensed and not redistributed here. Because the windows were drawn around known episodes, the alignment between window months and above-stress months is partly built in.
- Speed is not the whole story. The fast rises of 1996 and 2004, far below the stress level, were not followed by unrest, so the level is not dispensable; the finding is a joint condition, not a replacement of level by speed.
- Correlation, not causation. A common driver (weather, supply shocks) may produce both the price surge and the unrest; the index may be a coincident barometer rather than a cause. A properly instrumented study (Bellemare, 2015) finds a causal effect running from price levels to unrest.
- Real index. The real series uses the US CPI-U as deflator, a choice suited to a dollar-denominated index but not the only one; it is missing for October 2025, when the BLS published no CPI.
- Backward-looking. The record documents what happened; it is not a forecast.
Frequently Asked Questions
What is the food price “threshold” for unrest?
How high are global food prices in 2026?
Doesn’t published research find that price levels, not volatility, drive unrest?
Isn’t this just correlation? Couldn’t drought or war drive both prices and unrest?
Does this measure what people actually pay for food locally?
Can I reuse the dataset?
Eco3min Research (2026). Food Prices and Social Unrest (1990-2026): The Speed of Surges, Not the Price Level. Eco3min. https://eco3min.fr/en/food-price-speed-not-level/. Underlying data: World Bank, Commodity Price Data (The Pink Sheet), CC BY 4.0; US CPI-U, BLS; event list per Lagi et al. (2011) and cited sources. Eco3min text and analysis: CC BY 4.0.
Related research
Last updated — 3 November 2026
Disclaimer – Financial and Legal Information: The analyses, commentary, and content published on eco3min.fr are provided for informational and educational purposes only. They do not constitute investment advice or a solicitation to buy or sell financial instruments. Past performance is not indicative of future results. All investment decisions involve risk and are the sole responsibility of the reader. Data, charts and calculated series are provided as is and may contain errors or lag behind their source: check any figure against the primary source cited before relying on it. Content describing laws, regulations or contracts reflects the law as of its publication date and does not constitute legal or tax advice; applying it to a specific situation requires a qualified legal or tax professional.