## wpiea2020129-print-pdf ## Source details **Canonical URL:** [wpiea2020129-print-pdf](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020129-print-pdf.pdf) ## Other formats - [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020129-print-pdf.pdf.md) - [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020129-print-pdf.pdf.json) --- ### Data sources and sample - Primary source: Dow Jones’ Factiva news aggregator. - Sample restricted to printed articles published by major English-language newspapers and networks in the USA, UK, and Canada: the ABC Network, the BBC, the CBS Network, the Canadian Broadcasting Corp, the NBC Network, the Los Angeles Times, the Financial Times, the Boston Globe, the Globe and Mail, the New York Times, the Telegraph U.K., the Times U.K., the Chicago Tribune, the Telegraph, the Guardian U.K., the Wall Street Journal, the Washington Post, and the Economist. - Motivation for source choice: - Produces a very large sample over a long period: headline measure of relevant articles rising from around 7,000 per month in 1985 to over 20,000 between 2000 and the present, totaling over 7 million articles. - Enables transparent caveats about likely biases because sources are well-known. - Excludes newer media (e.g., Twitter, Facebook) and within-country publications to reduce challenges from changing penetration/composition and overt political agendas, bots, trolls, and nefarious actors. - Robustness checks: Section 4.2 examines impacts for a selection of Middle Eastern countries using different English-language sources, alternate search terms, and French and Arabic language media. ### Article counts and search criteria - Three monthly article counts collected: - xit: Number of articles about social unrest in country i at period t - yit: Number of contemporary articles in country i at period t - zt: Number of contemporary articles in period t - Sample includes the 130 countries countries with at least one million inhabitants and 200 social unrest articles (i.e. ∑t xit ≥200). - Search criteria (summary): - xit must include: Country name AND (“protest*” OR “riot*” OR “revolution” OR ((“civil” or “domestic”) within 10 words of “unrest”)) - yit must include: Country Name AND “today” - zt must include: “today” - Must exclude: Country-specific terms OR “vote of protest” OR “protest vote” OR “protestant*” OR “anniversary” OR “war” OR “memorial” OR “movie” - Location tag: Country i for xit and yit; none specified for zt - Subject tag: Domestic Politics Or Civil Unrest - Word count: 100+ for all three series - Additional normalization filter (for yit and zt): articles include the common and neutral term “today”. - Note: “today” is not included in xit due to constraints in Factiva queries and the low frequency of social unrest articles; see original discussion. ### Article distributions and key statistics - Cross-country distribution is highly skewed: - Median country is featured in an average of 116 articles per month; the average across countries is 350. - Within-country skew is large: the median country attains a peak of almost 900 articles per month. - Median country has an average of three social unrest articles per month, but a maximum of 75. - Time-series growth in contemporary articles: growth from ~7,000 per month in 1985 to over 20,000 per month since 2000 (Figure 1 referenced). ### Constructing the RSUI (indices) - Two indices created from raw article counts: - RSUI_Ait = xit / (1/12 ∑12_{j=1} z_{t−j}) × 100 ̄x_i / ̄z - RSUI_Bit = xit / (1/12 ∑12_{j=1} y_{i,t−j}) - Definitions of averages: - ̄x_i = 1/T ∑_{t=1}^T xit - ̄z = 1/T ∑_{t=1}^T zt - Comparative strengths and weaknesses: - B index: - Naturally interpretable units (share of contemporary articles in a given country about social unrest). - Flaws: - Denominator (1/12 ∑12_{j=1} z_{i,t−j}) often very close to zero for many countries and amplifies noise in xit. - Coverage of a country is endogenous to social unrest (coverage rises following unrest), and the one-period lag understates persistence of social unrest. - A index: - Uses fraction of all articles which are about unrest in country i; robust to the two B-index flaws and selected as the headline measure. - Scaled to have mean 100 within each country to facilitate within-country interpretation. - Lacks useful cross-country information; B index used as a filter for false positives in event coding. ### Event definition and identification - A social unrest event is a period i in country t matching three criteria: 1. Local peak in the A index: RSUI_Ait > max(RSUI_Ai,t−1, RSUI_Ai,t+1). 2. A index satisfies at least one extremum criterion: - (a) RSUI_Ait is in the top 2% of observations for country i; or - (b) RSUI_Ait exceeds the mean for country i by at least four times the standard deviation for country i; or - (c) RSUI_Ait exceeds the rolling 20-year mean for country i by at least four times the rolling 20-year standard deviation for country i. 3. B index threshold: RSUI_Bit > 0.1. - Design choices are acknowledged as somewhat arbitrary; validation pursued in Section 4 against authoritative external sources. - Methodology identifies some [679] monthly events (regional waves summarized below). ### Events identified and regional waves (highlights) - Late 1980s: spike in Western Hemisphere driven by events in South and Central America and the Caribbean (examples: Guatemala democratic transition 1986; Haitian election attacks 1987; Chilean Plebiscite 1988; 1989 Paraguayan Coup d’état; 1990 Nicaraguan elections). - Early 1990s: Europe spike reflecting fall of communism (Bulgaria, Czechoslovakia, Germany, Poland, Romania, Baltic countries). - 2011: Arab Uprisings drive a large wave across almost the whole Middle East, coincident with anti-austerity protests in Europe. - Late 2019–early 2020: protests in Bolivia, Chile, Colombia, Ecuador, and Peru contribute to a spike in unrest events. ### Event screening and mis-identification - Screening inspects only identified events to (1) verify events are not egregiously mis-identified, and (2) generate descriptive labels. - Five main causes of mis-identified events: 1. Events relating to a given country occur elsewhere (e.g., January 2019 protests in Greece labeled as occurring in North Macedonia). 2. Past social unrest events are mentioned as context for current events. 3. Country-specific institutional names match search terms (e.g., Iranian institutions including “Islamic Revolution” matching “revolution”). 4. Alternate usages of search terms (e.g., “Tax protest” parties in Denmark). 5. Diplomatic events/descriptions labeled as “protest.” - Identified some [100] events where articles made no specific mention of contemporaneous civil unrest events. - Two-step strategy: - (1) minimal country-specific search adjustments to reduce mis-identification; - (2) where adjustment infeasible, manually flag and exclude mis-labeled events. - Event screening summary (table text preserved exactly as in source): - Total Removed by search adjustment Removed by hand - Passed human screening581 - Failed human screening996831 - Of which: - Wrong country34268 - Past events231112 - Country-Specific17152 - Incorrect usage1376 - Diplomatic1192 - Possible future event11 ### Search modifications (examples) - Country-specific search exclusions and corrections (examples drawn verbatim): - Bahrain — Wrong country - Mauritanianot ”Dakar” — Riots in Dakar, Senegalese and Mauritanian refugees flee — Wrong country - Mexiconot ”border” — 06/18 picks up protests in US near Mexican border — Wrong country - Montenegronot ”Kremlin” — MNE bids to join NATO — Wrong country - Omannot ”nuclear” — 03/14 Picks up Iran-US nuclear deal talks based in Oman — Diplomatic - Panamanot ”panama papers” — 04/16 picks up Panama papers — Country-Specific - PapuaNew Guineanot ( ”Nauru” OR ”Manus” ) — Riots in Australia’s detention centers in Manus Island — Wrong country - Paraguaynot ”1973” — Articles looked back at protests that happened in Paraguay — Past events - Portugalnot ”Carnation” — Election in Portugal — Past events - Qatarnot ”Syria” — Funds for the Syrian opposition has been allocated and deposited in Qatar — Wrong country - Rwandanot ”StateDepartment” — 10/15 US State Department protest Rwanda’s support for Congolese rebels — Diplomatic - Slovak Republicnot ”velvet revolution” — Parliamentary elections — Country-Specific - Spainnot ( ”UN” near10 ”climate” ) — Global climate protests ahead of U.N. climate talks in Madrid — Wrong country - Taiwan Province of China not ”Vietnam” — Anti-China protest in Vietnam affects Taiwanese businesses — Wrong country - Tajikistannot ”Islamic Revolution” — Throughout: Picks up political party — Country-Specific - Tanzanianot ”Kenya” — Oppositions, US, and Kenya protest over election rerun in Zanzibar — Wrong country - Ukrainenot ”military aid” or impeach* or ”House Intelligence Committee” — UKR mentioned in news articles about Trump impeachment — Wrong country - United Arab Emirates not ”DPW” — 02/06: Picks up controversy over Dubai Port World purchases — Incorrect usage - United Kingdomnot ”Embassy” — 12/11 picks up protests at British Embassy in Tehran — Wrong country - Vietnamnot ”revolution” — Throughout picks up ”revolution” for Party events — Country-Specific - Zambianot ”Zimbabwe” — 08/18 picking up diplomatic incident with Zimbabwe — Diplomatic ### Validation approach (overview) - Two exercises: 1. Show index and event listing agree almost exactly with authoritative narrative sources for four episodes: Arab Uprisings of 2011; Thailand 2006-2014 protests/coups/constitutional crises; Venezuela 2015-2019 protests; color revolutions of the 2000s. 2. Provide further narrative sources and robustness checks (discussed in Section 4 and elsewhere). - Appendix E provides full verbatim descriptions from the primary sources to permit reader verification. ### External-validity highlights (selected concordances) - Arab Uprisings (Worth (2016) timeline): - Tunisia: Bouazizi self-immolation on 17 December 2010 and death on 4 January 2011 reflected in a large RSUI spike and RSUI-implied event in January 2011. - Egypt: first RSUI-implied event dated to February 2011, aligning with external timeline (start of protests last week of January extending through February 2011). - Bahrain, Libya, Yemen: RSUI dates start to February (consistent with Worth’s timing). - Syria: RSUI first picks up sharply in March. - Post-initial outbreak major turning points (Egypt) appear as index spikes and satisfy event definitions, sometimes with a one-month lag due to monthly RSUI frequency. - Thailand 2006-2014: - RSUI records spikes corresponding to February–March 2006 protest peaks, 19/09/06 military coup, November 2008 airport seizures, April 2009 demonstrations, March–May 2010 violent uprising (08/04/10 state of emergency; over 90 deaths), October–November 2013 Amnesty Bill protests (31/10/13 spike), January 2014 pre-election protests, and May 2014 martial law and coup (20/05/14, 22/05/14). - RSUI event-coding identified seven events in this period corresponding closely to narrative evidence. - Venezuela 2014-2019: - February 2014 RSUI peak and event. - April–August 2017 sustained RSUI elevation matching Briceño-Ruiz (2019). - January 2019 constitutional crisis and widespread protests reflected as RSUI-implied event; May 2019 failed military uprising identified. - Color Revolutions (2000–2005): - RSUI event coding identifies exactly the month of each of the four color revolutions (Serbia 2000, Georgia 2003, Ukraine 2004, Kyrgyz Republic 2005). ### Sensitivity and robustness (Appendix B and Section 4.2) - Language sensitivity: - French-language searches (Le Figaro, Le Monde, Libération, Agence France Presse, Reuters, The Associated Press and The Canadian Press) generally align with the headline series for Algeria, Lebanon, Syria, and Tunisia; Mauritania differences reflect very small French article counts (on average less than one article per month). - Arabic-language searches face coverage and matching problems: restricting to international Arabic sources yields few matches; expanding to all Arabic hits increases counts but heterogeneity and poor country discrimination (examples: Sudan and Saudi Arabia spikes early 2011 despite limited unrest). - Search-term sensitivity (sub-sample N=19 Middle East and North African countries): - Alternate x_it searches defined: - Expanded (Extension): “attack” OR “conflict*” OR “strike” OR “police*” - Tension (Extension): same as Expanded OR “social” within 10 words of “tension” - Economic (Restriction): “econom*” OR “unemployment” - Youth (Restriction): “youth” - Preferred overlap metric: overlap of the largest observations limited to top 2 percent. - Table 5 summary (Within-country correlations with baseline, N=19): - Overlap of top 2%: - Expanded: 0.92 / 0.69 / 0.87 - Tension: 0.92 / 0.69 / 0.87 - Economic: 0.79 / 0.55 / 0.66 - Youth: 0.45 / 0.23 / 0.33 - Pearson correl: - Expanded: 0.91 / 0.73 / 0.76 - Tension: 0.91 / 0.73 / 0.76 - Economic: 0.84 / 0.57 / 0.71 - Youth: 0.69 / 0.45 / 0.60 - Spearman correl: - Expanded: 0.83 / 0.64 / 0.79 - Tension: 0.83 / 0.64 / 0.79 - Economic: 0.70 / 0.52 / 0.59 - Youth: 0.54 / 0.35 / 0.42 - Interpretation: - Extension searches show very high overlap with baseline (three-quarters of countries have at least 85 percent of observations in common in the top 2 percent). - Economic terms are common in large RSUI observations: in the median country, 79% of the largest headline matches also included an economic term; agreement is over half (55%) in fully 90 percent of countries. - Youth-related terms are less common: median fraction 45%. ### Statistical properties of the index (5.1 Cross-sectional distribution and Appendix D) - Tail model for RSUI_Ait values above cutoff m̄: - P(RSUI_Ait < M) = F(M) for M ≥ m̄ - F(M) = 1 − (a(M − m̄) + 1)^−b - log(1 − F(M)) = −b log a − b log(M − m̄ + 1/a) - Estimation details: - Density estimated by maximum likelihood with m̄ = 200. - Full-sample maximum-likelihood estimation results (N = 6926, m̄ = 200) reported in Table 6: - a = 0.0032 (0.00016) - b = 2.050 (0.0752) - Parameters are sharply identified with small standard errors and fitted and empirical distributions match closely. - Robustness: - Estimates for a and b unaffected materially by choice of m̄. - Within-country estimates of parameters not significantly different from aggregate estimates. - Rule-of-thumb for comparing large tail observations: - For two upper-tail observations M1 and M2 with M1 < M2 and both large relative to ̄m−1/a'130: - If M2 is x% larger than M1 then there usually will be approximately 2x% fewer observations larger than M2 than M1 (using here that b'2). - Rule holds within individual countries and across the sample. ### Correlation, panel regressions and spillovers (Appendix D / Table 7) - Definitions: - s_i,t = Months since last unrest event in country i. - u_i,t = Months since last unrest event in neighbors of country i. - S_k_i,t and U_k_i,t partition these into K intervals. - Time-since averages: - Time since neighboring-country events averages 32 months. - Time since own-country events averages 60 months. - Event regression specification: - event_i,t = α_i + η_t + Σ_{k∈K} β^d_k S_k_i,t + Σ_{k∈K} β^n_k U_k_i,t + e_it - Key regression findings (incremental probabilities; standard errors in parentheses): - Own-country effects (short-term): - Event, last 2-3 months: 0.037 ∗∗∗ (0.007) in (1); 0.027 ∗∗∗ (0.007) in (2); 0.025 ∗∗∗ (0.007) in (3); 0.028 ∗∗∗ (0.008) in (4). - Event, last 4-6 months: 0.029 ∗∗∗ (0.006); 0.020 ∗∗∗ (0.006); 0.018 ∗∗∗ (0.006); 0.021 ∗∗∗ (0.007). - Event, last 7-9 months and beyond: smaller and often statistically insignificant coefficients. - Neighboring-country effects: - Neighbor event, last 2-3 months: 0.018 ∗∗∗ (0.003); 0.017 ∗∗∗ (0.005); 0.016 ∗∗∗ (0.004). - Neighbor event, last 4-6 months: 0.009 ∗∗∗ (0.003); 0.006 (0.004); 0.005 (0.004). - Neighbor event, last 7-12 and 13-24 months: generally positive, smaller, sometimes significant. - Sample and fit: - Full-sample unconditional average unrest-event rate a little over 1 percent of months. - Restricting to observations with at least one prior unrest event raises unconditional average event probability to nearly 2 percent. - Naive R^2 values range from 0.011 to 0.046 across specifications; model fit is modest. - Interpretation: - Own-country unrest associated with an approximate 3 percentage point short-term increase in event probability (approximate quadrupling relative to baseline), but effect statistically insignificant after six months with country fixed effects. - Neighboring-country unrest associated with roughly a doubling of the rate of unrest in the focal country; effects smaller than own-country but somewhat longer-lasting. - Elevated probabilities decay to about 0.5 percent or less within 9 months. ### Predictability (5.3) - Empirical observation: - RSUI spikes dramatically in the month of an event and declines slowly thereafter, but displays a moderate pre-event increase of around 174 index points. - Conditional average around events: moderate pre-event increase ~174 index points; however, such increases have essentially no predictive power because they typically regress sharply to the mean. - Lasso predictive analysis (country-by-country, cross-validated): - Specification: v_{i,t+h} = sum_{j=1}^{M} beta_{j}^{i} v_{i,t-j}, with M = 18, h = 1,...,24 and v_{it} either RSUI or event indicator. - Predictability summary (Table 8, averaged across countries selecting most predictive horizon per country): - Number of non-zero coefficients: Index = 1.03, Events = 0.02 - Horizon of most non-zero coefficients: Index = 1.50, Events = 1.03 - Granger-Newbold predictability: Index = 0.03, Events = 0.00 - Theil U-statistic: Index = 0.92, Events = 1.00 - Diebold-Kilian predictability: Index = 0.11, Events = 0.05 - Findings: - Lasso typically selects at most only one lag of the RSUI at very short lags. - For RSUI events, almost all country-horizon pairs select zero lags (best predictive model is the country average). - Best forecasts explain barely 3 percent of variation in outcomes (Granger-Newbold). - Interpretation and implications: - Correlation (temporal/spatial association) does not imply strong predictive power. - Social unrest is rare and heterogeneously triggered; RSUI useful for timely monitoring of realized unrest but not a reliable early-warning predictor of specific outbreaks. - RSUI and event listings can be used as instruments or sources of exogenous variation for studying causes and consequences of unrest. ### Comparison to other measures (Section C) - ACLED-based index (27 countries, rescaled avg=100 since 2000) and CNTS-based index (117 countries, annual): - RSUI and ACLED-based index show broad agreement: - Both series relatively low in the first decade of the 2000s; substantial and persistent increase starting in 2011. - Both series show a rise in unrest around 2015, a subsequent decline during 2016-2017 and a pick-up since. - At an annual frequency, correlation between RSUI and ACLED-based series is 0.7; at a monthly frequency correlation nearer 0.5. - CNTS-based index shows almost monotonic increase in unrest events from around 2009 onward with little variation prior; pattern hard to reconcile with RSUI and ACLED evidence and may reflect increasing media coverage over time. - Event-level cross-checks: for validated events, RSUI, ACLED (where available), and CNTS show very good agreement; RSUI timing closely aligns with ACLED during times of extreme stress. *Source: Excerpt from the IMF working paper chapter 3.1–3.4, Appendix B, Appendix D, and Sections 4.1–5.3 (wpiea2020129-print-pdf).* ### 3.1 Data ### 3.1 Data ### Data sources and sample - Primary source: Dow Jones’ Factiva news aggregator. - Sample restricted to printed articles published by major English-language newspapers and networks in the USA, UK, and Canada: the ABC Network, the BBC, the CBS Network, the Canadian Broadcasting Corp, the NBC Network, the Los Angeles Times, the Financial Times, the Boston Globe, the Globe and Mail, the New York Times, the Telegraph U.K., the Times U.K., the Chicago Tribune, the Telegraph, the Guardian U.K., the Wall Street Journal, the Washington Post, and the Economist. - Motivation for source choice: - Produces a very large sample over a long period: headline measure of relevant articles rising from around 7,000 per month in 1985 to over 20,000 between 2000 and the present, totaling over 7 million articles. - Enables transparent caveats about likely biases because sources are well-known. - Excludes newer media (e.g., Twitter, Facebook) and within-country publications to reduce challenges from changing penetration/composition and overt political agendas, bots, trolls, and nefarious actors. - Robustness checks: Section 4.2 examines impacts for a selection of Middle Eastern countries using different English-language sources, alternate search terms, and French and Arabic language media. ### Article counts and search criteria - Three monthly article counts collected: - xit: Number of articles about social unrest in country i at period t - yit: Number of contemporary articles in country i at period t - zt: Number of contemporary articles in period t - Sample includes the 130 countries countries with at least one million inhabitants and 200 social unrest articles (i.e. ∑t xit ≥200). - Search criteria vary principally in matching text strings (see Table 1 summary): - xit must include: Country name AND (“protest*” OR “riot*” OR “revolution” OR ((“civil” or “domestic”) within 10 words of “unrest”)) - yit must include: Country Name AND “today” - zt must include: “today” - Must exclude: Country-specific terms OR “vote of protest” OR “protest vote” OR “protestant*” OR “anniversary” OR “war” OR “memorial” OR “movie” - Location tag: Country i for xit and yit; none specified for zt - Subject tag: Domestic Politics Or Civil Unrest - Word count: 100+ for all three series - Additional normalization filter (for yit and zt): articles include the common and neutral term “today”. - Note: “today” is not included in xit due to constraints in Factiva queries and the low frequency of social unrest articles; see original discussion. ### Article distributions and key statistics - Cross-country distribution is highly skewed: - Median country is featured in an average of 116 articles per month; the average across countries is 350. - Within-country skew is large: the median country attains a peak of almost 900 articles per month. - Median country has an average of three social unrest articles per month, but a maximum of 75. - Figures referenced: - Figure 1: Number of contemporary articles per month — shows growth from ~7,000 per month in 1985 to over 20,000 per month since 2000. - Figure 2: Cross-country average monthly article counts — panels for (a) Social unrest articles, xit, and (b) Contemporary country articles, yit. ### Constructing the RSUI (indices) - Two indices created from raw article counts: - RSUI_Ait = xit / (1/12 ∑12_{j=1} z_{t−j}) × 100 ̄x_i / ̄z - RSUI_Bit = xit / (1/12 ∑12_{j=1} y_{i,t−j}) - Definitions of averages: - ̄x_i = 1/T ∑_{t=1}^T xit - ̄z = 1/T ∑_{t=1}^T zt - Comparative strengths and weaknesses: - B index: naturally interpretable units (share of contemporary articles in a given country about social unrest). - Flaws: - Many countries have very little coverage outside of social unrest events, making the denominator (1/12 ∑12_{j=1} z_{i,t−j}) often very close to zero and amplifying noise in xit. - Coverage of a country is endogenous to social unrest (coverage rises following unrest), and the one-period lag is insufficient for events with persistence over several months; the B index thus understates persistence of social unrest. - A index: uses fraction of all articles which are about unrest in country i; robust to the two B-index flaws and selected as the headline measure. - Scaled to have mean 100 within each country to facilitate within-country interpretation. - Lacks useful cross-country information; B index is used as a filter for false positives in event coding (Section 3.3). - Empirical illustration: - Figure 3 (six Middle Eastern countries, June 2010 - December 2014) shows large intermittent spikes and highlights B index endogeneity: major spikes in A (e.g., Tahrir Square protests, Tunisian Parliamentary elections of 2011) are muted or invisible in B because the denominator rises after unrest. ### Event definition - A social unrest event is a period i in country t matching three criteria: 1. Local peak in the A index: RSUI_Ait > max(RSUI_Ai,t−1, RSUI_Ai,t+1). 2. A index satisfies at least one extremum criterion: - (a) RSUI_Ait is in the top 2% of observations for country i; or - (b) RSUI_Ait exceeds the mean for country i by at least four times the standard deviation for country i; or - (c) RSUI_Ait exceeds the rolling 20-year mean for country i by at least four times the rolling 20-year standard deviation for country i. - Rationale: percentile and standard-deviation thresholds ensure robustness to cross-country variation; part (c) addresses changing media focus over time. 3. B index threshold: RSUI_Bit > 0.1 — a weak cross-country information filter to exclude false positives in countries with very rare unrest. - Design choices are acknowledged as somewhat arbitrary; validation is pursued in Section 4 against authoritative external sources. ### Events identified and regional waves - Methodology identifies some [679] monthly events (figure shows fraction of countries with social unrest events by region). - Regional patterns highlighted: - Late 1980s: spike in Western Hemisphere driven by events in South and Central America and the Caribbean (e.g., Guatemala democratic transition 1986; Haitian election attacks 1987; Chilean Plebiscite 1988; 1989 Paraguayan Coup d’état; 1990 Nicaraguan elections). - Early 1990s: Europe spike reflecting fall of communism (events in Bulgaria, Czechoslovakia, Germany, Poland, Romania, Baltic countries). - 2011: Arab Uprisings drive a large wave across almost the whole Middle East, coincident with anti-austerity protests in Europe. - Late 2019–early 2020: protests in Bolivia, Chile, Colombia, Ecuador, and Peru contribute to a spike in unrest events. ### Event screening and mis-identification - Screening approach: check only identified events (not every month/country) by inspecting matching Factiva articles to (1) verify events are not egregiously mis-identified, and (2) generate descriptive labels. - Five main causes of mis-identified events: 1. Events relating to a given country occur elsewhere (e.g., January 2019 protests in Greece labeled as occurring in North Macedonia). 2. Past social unrest events are mentioned as context for current events. 3. Country-specific institutional names match search terms (e.g., Iranian institutions including “Islamic Revolution” matching “revolution”). 4. Alternate usages of search terms (e.g., “Tax protest” parties in Denmark). 5. Diplomatic events/descriptions labeled as “protest.” - Identified some [100] events where articles made no specific mention of contemporaneous civil unrest events. - Two-step strategy to address mis-identified events: (1) minimal country-specific search adjustments to reduce mis-identification; (2) where adjustment is infeasible, manually flag and exclude mis-labeled events. - Event screening summary (table text preserved exactly as in source): - Total Removed by search adjustment Removed by hand - Passed human screening581 - Failed human screening996831 - Of which: - Wrong country34268 - Past events231112 - Country-Specific17152 - Incorrect usage1376 - Diplomatic1192 - Possible future event11 ### Validation approach (overview) - Two exercises to justify methodological choices: 1. Show index and event listing agree almost exactly with authoritative narrative sources for four diverse episodes: Arab Uprisings of 2011; protests/coups/constitutional crises in Thailand during 2006-2014; protests in Venezuela in 2015-2019; and the color revolutions of the 2000s. 2. Provide further narrative sources and robustness checks (discussed in Section 4 and elsewhere). *Source: Excerpt from the IMF working paper chapter 3.1–3.4 (data, index construction, events, and screening).* ### Appendix B. ### Appendix B. ### Sensitivity and robustness of the RSUI measure - The authors vary language, sources, and search terms and –where applicable – get very similar results for a subset of Middle Eastern Countries since 2011. - Interpretation: using media reports to identify social unrest is relatively forgiving; major events are sufficiently stark and obvious that they are revealed no matter the precise details of the methodology used to identify them. - The authors do not place much weight on comparisons to alternate measures such as the Cross-National Time-Series Database (CNTSD, Banks and Wilson (2020)) or The Armed Conflict Location & Event Data Project (ACLED) because: - Comparison to these alternatives is a relative test and cannot distinguish between shortcomings of the authors’ measures and those of the alternatives themselves. - Example doubts: do differences with the CNTSD simply reflect the small number of sources used in that measure? Are differences with ACLED merely a function of that source’s emphasis on conflict? - Preferred tests are absolute ones: the authors’ measure matches authoritative, external reports of social unrest and is not unduly sensitive to small changes. - Note: comparisons to the relevant CNTS and ACLED series are still reported in Appendix C. ### External validity (Section 4.1) - Approach: compare the index and event coding to four major waves of protest where alternative authoritative narrative sources exist. - Appendix E provides full verbatim descriptions from the primary sources to permit reader verification. - The four episodes vary widely by location, time period, government type, and political outcomes; yet across all, the index and implied events match closely with alternate narrative sources. ### 4.1.1 The Arab Uprisings of 2011 - Primary external source: Worth (2016) — a comprehensive timeline of events between June 2010 and December 2014 covering Bahrain, Egypt, Libya, Syria, Tunisia, and Yemen. - The Worth timeline events are shown as vertical lines in Figure 5 (same period and countries as Figure 3, which includes RSUI event codings). - Key concordances: - Cross-country timing of the initial round of protest almost exactly coincides with initial spikes in the RSUI early 2011. - Tunisia: Bouazizi self-immolation on 17 December 2010 and subsequent death on 4 January 2011 is reflected in a large spike in the RSUI and a RSUI-implied event in January 2011; prior to this there are no RSUI-implied events in any of these six countries. - Egypt: first RSUI-implied event dated to February 2011, agreeing almost identically with the external timeline that dates start of protests to the last week of January and extending through February 2011. - Bahrain, Libya, Yemen: RSUI dates the start to February (consistent with Worth’s timing of February or very late January). - Syria: RSUI first picks up sharply in March, consistent with Worth. - The index and event coding match the cross-country order of the start of the Arab Uprisings exactly, and the timing very closely. - Post-initial outbreak: - Major turning points of the Egyptian revolution show as index spikes and satisfy event definitions, sometimes with a one-month lag when events occur in the second half of a month (monthly RSUI frequency limits precision). - Examples matched: running battles for control of Tahrir Square in November 2011; election of Muhammed Morsi and unrest in June 2012; collapse of the constituent assembly in November 2012; Morsi’s overthrow in July 2013. - Similar identification of major post-revolutionary events in Libya, Tunisia, and Yemen. - Differences between RSUI and Worth: - RSUI distinguishes separately the deaths of Tunisian politicians Chokri Belaid and Mohamed Brahmi, which Worth mentions together only after the latter’s death. - RSUI includes Bahrain events in April 2012 (a bomb blast injuring several policemen and domestic protest affecting the Bahrain Grand Prix); these are arguably major unrest events and thus reasonably included. - Figure 5 annotations (examples shown in source): - 14/02/11: Protests start. - 14/03/11: Saudi troops arrive - 15/02/11: Large−scale Benghazi protests - 27/02/11: Transitional Council is formed - 17/03/11: UN authorizes air strikes - 20/08/11: Qaddafi killed - 11/09/12: US Ambassador killed in Benghazi - 01/11/12: Clashes in Benghazi - 25/06/14: Elections; Islamist coup in Tripoli - 17/12/10: Bouazizi self−immolation - 14/01/11: Ben Ali resigns - 23/10/11: Parliamentary elections - 25/07/13: Mohamed Brahmi assassinated - 28/09/13: Ennahda movement agrees to cede power - 25/01/11: Protests start - 28/01/11: Protestors take Tahrir Square - 11/02/11: Mubarak resigns - 19/03/11: Constitutional referendum - 01/06/12: Morsi elected president - 30/06/12: Anti−Morsi protests - 14/08/12: Rabaa square cleared - 01/11/12: Morsi decree grants him broad powers - 03/07/13: Morsi deposed - 30/03/14: Sisi elected president - 15/03/11: Demonstrations start in Syria - 27/01/11: First large−scale protests - 18/03/11: Gunman fires on protestors - 03/06/11: President injured in bombing - 23/11/11: President signs transition document - 10/09/14: Houthi rebels capture Sanaa - Methodological note: RSUI is monthly; events in the second half of a month can appear as spikes in the following month. ### 4.1.2 Thailand 2006-2014 - Narrative sources: Prasirtsuk (2009, 2010, 2015), Dalpino (2011), Baker (2016), Keyes (2006); verbatim extracts in Appendix E. - Timeline and RSUI concordance: - January 2006: anti-government protests erupted following revelations about the Prime Minister’s family; RSUI shows a sharp spike in February and March 2006 corresponding to protest peaks. - September 2006: military coup reflected by a second RSUI spike in September. - 2008-2010: return of the ex-Prime Minister in February prompted political disputes, violent clashes in Bangkok in September, and seizure of two major airports in November — all reflected as RSUI peaks in corresponding months. - April 2009: major anti-government demonstrations in Bangkok coinciding with an ASEAN summit in Pattaya; local RSUI spike. - March–May 2010: after a Supreme Court ruling, massive anti-government protests in March 2010; April–May 2010 turned violent with state of emergency (08/04/10), widespread fires including the Stock Exchange of Thailand and a major department store, and over 90 deaths — RSUI records very high values in April and May. - 2011 transfer of power prompted relatively little unrest by preceding standards. - October–November 2013: Amnesty bill in October 2013 leads to mass anti-government demonstrations in November (31/10/13 RSUI spike). - January 2014: pre-election protests (01/14 RSUI spike). - May 2014: military declares martial law (20/05/14) and coup (22/05/14) — RSUI spike. - RSUI event-coding identified seven events in this period corresponding closely to the narrative evidence. The 2006 events do not register as RSUI-identified events (later events received more coverage due to increased severity and violence). - Figure 6 and Figure 7 annotations (examples shown in source): - 23/01/06: Anti tax avoidance protests - 01/06: Protests call for PM's resignation - 08/03/06: Buddhist monks back PAD protest movement - 19/09/06: Military coup - 03/08: Yellow shirts protest immunity - 26/08/08: "Whistle Day" protests - 09/08: Red v. yellow shirt clashes - 07/10/08: Yellow shirts clash with police - 15/10/08: Preah Vihear temple dispute - 26/11/08: PAD protestors seize two major airports - 04/09: Red shirts take over ASEAN meeting - 03/10: Red shirt protests against government - 04/10: Red shirts seize sections of Bangkok - 08/04/10: Abhisit declares state of emergency - 05/10: Red shirt advisor killed by sniper - 05/10: Widespread fires, over 90 die - 07/11: Yingluck Sinawatra elected - 31/10/13: Amnesty Bill sparks protest - 24/01/14: Professional groups call for government ouster - 20/05/14: Military declares martial law - 22/05/14: Military coup ### 4.1.3 Venezuela 2014-2019 - Narrative sources: Gutíerrez (2017) and Briceño-Ruiz (2019) — peer-reviewed Latin American political science journal articles documenting 2013-2019 events. - RSUI concordance: - February 2014: first major wave of anti-government protests corresponds to the first RSUI peak and is coded as an RSUI event. - 2015–2016: RSUI remains at lower levels with no implied events and a small peak around the opposition’s parliamentary victories in late 2015 — aligns with narrative sources that record important political events but no major protests or civil disruptions between mid-2014 and mid-2016. - April–August 2017: RSUI picks up sharply and remains elevated until August, agreeing almost exactly with Briceño-Ruiz (2019) reporting “a new wave of protests from April to August 2017”. - January 2019: constitutional crisis, power struggle, widespread protests, and a humanitarian crisis are reflected in both narrative descriptions and an RSUI-implied event for January 2019. - May 2019: RSUI identifies a further event corresponding to a failed military uprising. - Overall: events in Venezuela line up very closely with the two peer-reviewed narrative sources. ### 4.1.4 Color Revolutions (2000–2005) - External source: Tucker (2007) — peer-reviewed APSA publication focusing on four prominent color revolutions: - Yugoslavia’s “Bulldozer revolution” in October 2000; - Georgia’s 2003 “Rose revolution”; - 2004 “Orange revolution” in Ukraine; - 2005 “Tulip revolution” in Kyrgyz Republic. - Findings: - Figures 10 and 11 present RSUI and event coding for these countries during 2000-2005. - In all four cases, the RSUI event coding identifies exactly the month of each of the four color revolutions. - Secondary events following main revolutionary episodes are associated with RSUI spikes in Georgia and Kyrgyz Republic (delayed elections). - The event coding is somewhat conservative: identifying more minor follow-up events would increase false positives elsewhere — reflecting a deliberate trade-off between false positives and false negatives. *Source: Appendix B, wpiea2020129-print-pdf - Appendix B.* ### 4.2 Internal Validity ### 4.2 Internal Validity ### Overview - The section reports results from alternative measures of the RSUI where parameters defining the RSUI are varied, including language, sources, and search terms. - The methodology treats current IMF member country descriptions as fixed for historical time series; events in the then-Federal Republic of Yugoslavia are proxied using the time series for Serbia. ### Key graphical evidence (described) - Figures illustrate RSUI event coding and alignment with known major events (Venezuela January 2013-June 2019; Color Revolutions August 2000-December 2005). - Language-coverage plot (Figure 12) shows the number of contemporary articles per month, z_t, by language with: - French coverage limited throughout. - Arabic content growing rapidly following 2011. --- ### 4.2.1 Sensitivity to language Findings: - French-language searches (sources: Le Figaro, Le Monde, Libération, Agence France Presse, Reuters, The Associated Press and The Canadian Press) generally align with the headline series: - Major spikes occur in exactly the same months for Algeria, Lebanon, Syria, and Tunisia when comparing English and French. - Morocco shows some disagreement during relatively quiet periods, but major events (Arab Spring, Hirak Rif protests of late 2016 and early 2017) are clearly identified in both languages. - Mauritania differences likely reflect very small French article counts (on average less than one article per month). - Arabic-language searches faced coverage and matching problems: - Restricting Arabic sources to well-known international news sources yielded only a handful of articles and a very low match rate to social unrest terms. - Expanding Arabic sources to allow all search hits (including news websites, blogs, and less well-regarded outlets) increased article counts but introduced heterogeneity in source quality. - Arabic-language indices did a poor job of distinguishing country-specific unrest: - Example: Sudan and Saudi Arabia show large spikes in early 2011 despite very limited unrest at those times. - The Arabic peak in November 2010 was driven by only 18 matching social unrest articles versus a February 2011 English-language peak of some 1600 articles. - Possible reasons: pan-Arabic reporting tendency and text-matching methods that work in English not translating well to Arabic. Conclusion: - Comparison across languages suggests either strong agreement with French or fundamental limits imposed by availability and matching quality of Arabic sources. --- ### 4.2.2 Sensitivity to search terms Design: - Four extra searches, all derivatives of the headline search for x_it (see Table 4), were conducted for a sub-sample of 19 Middle East and North African countries. - Two extensions of the headline search: - Expanded (Extension): adds broader, potentially more violent terms. - Tension (Extension): includes the Expanded terms plus a proximity condition for “social” within 10 words of “tension”. - Two restrictions of the headline search: - Economic (Restriction): restricts matches to those also including economic-related terms. - Youth (Restriction): restricts matches to those also including “youth”. Table 4: Alternate search terms - Name / Type / Terms - Expanded / Extension / “attack” OR “conflict*” OR “strike” OR “police*” - Tension / Extension / “attack” OR “conflict*” OR “strike” OR “police*” OR “social” within 10 words of “tension” - Economic / Restriction / “econom*” OR “unemployment” - Youth / Restriction / “youth” - Note: Alternate searches for x_it. All searches are derivatives of the headline search. Those labeled “Extension” include OR operators for the extra search terms, so return extra matches which may include terms listed. Those labeled “Restriction” append the listed terms using an AND operator, so return only those headline matches which also include the terms listed. Findings (overview and Table 5 results): - Preferred overlap metric: overlap of the largest observations limited to top 2 percent. - Extension searches (Expanded, Tension) show very high overlap with baseline: - Three-quarters of countries have at least 85 percent of observations in common in the top 2 percent, indicating the headline index is not driven by spurious matches tied to particular search terms. - Restrictions inform potential drivers: - Economic terms are common in large RSUI observations: - In the median country, 79% of the largest headline matches also included an economic term. - Agreement is over half (55%) in fully 90 percent of countries. - Youth-related terms are less common: - Median country common fraction of top matches including youth terms is 45%. Table 5: Within-country correlations with baseline, varying search terms (N=19) - Overlap of top 2% / Expanded / 19 / 0.92 / 0.69 / 0.87 - Overlap of top 2% / Tension / 19 / 0.92 / 0.69 / 0.87 - Overlap of top 2% / Economic / 19 / 0.79 / 0.55 / 0.66 - Overlap of top 2% / Youth / 19 / 0.45 / 0.23 / 0.33 - Pearson correl / Expanded / 19 / 0.91 / 0.73 / 0.76 - Pearson correl / Tension / 19 / 0.91 / 0.73 / 0.76 - Pearson correl / Economic / 19 / 0.84 / 0.57 / 0.71 - Pearson correl / Youth / 19 / 0.69 / 0.45 / 0.60 - Spearman correl / Expanded / 19 / 0.83 / 0.64 / 0.79 - Spearman correl / Tension / 19 / 0.83 / 0.64 / 0.79 - Spearman correl / Economic / 19 / 0.70 / 0.52 / 0.59 - Spearman correl / Youth / 19 / 0.54 / 0.35 / 0.42 Illustration: - Figure 15 shows RSUI June 2010-present for the largest six Arabic-speaking countries by search term (Baseline, Expanded, Tension, Economic, Youth), illustrating similarities and differences across searches. --- ### 5 Statistical Properties of the Index (beginning) — 5.1 Cross-sectional distribution Model proposal: - For values in excess of some cutoff m̄, the full-sample RSUI_Ait is modeled with: - P(RSUI_Ait < M) = F(M) for M ≥ m̄ - F(M) = 1 − (a(M − m̄) + 1)^−b - Interpretation: m̄ is a filter for noise; for M above m̄ the fraction of observations above M is proportionate to M − m̄ + 1/a. Equivalently: - log(1 − F(M)) = −b log a − b log(M − m̄ + 1/a) - Estimation details: - Density estimated by maximum likelihood with m̄ = 200. - Full-sample maximum-likelihood estimation results (N = 6926, m̄ = 200) reported in Table 6. - Parameters are sharply identified with small standard errors and fitted and empirical distributions match closely (Figure 16). Table 6: Full-sample maximum-likelihood estimation of cross sectional density, N=6926, m̄=200. Standard errors in parentheses. - a = 0.0032 (0.00016) - b = 2.050 (0.0752) *Source: wpiea2020129-print-pdf - 4.2 Internal Validity (IMF) ### Appendix D provides further robustness checks, showing that 1) estimates for the parameters ### Appendix D provides further robustness checks, showing that 1) estimates for the parameters ### Robustness of cross-sectional and within-country parameter estimates - Estimates for the parameters of the cross-sectional distribution are unaffected by the choice of the cutoff ̄m. - Within-country estimates of the parameters are not significantly different from the aggregate estimates. - These robustness results provide a useful interpretation of large observations of the headline index. ### Rule-of-thumb for comparing large tail observations - For two observations in the upper tail M1 and M2 with M1 < M2 and both large relative to ̄m−1/a'130: - If M2 is x% larger than M1 then there usually will be approximately 2x% fewer observations larger than M2 than M1 (using here that b'2). - This rule of thumb holds both within individual countries and across the sample as a whole. ### Correlation of unrest events in time and space — definitions and data patterns - Definitions: - s_i,t = Months since last unrest event in country i. - u_i,t = Months since last unrest event in neighbors of country i (neighbors defined as countries which share borders). - Indicators S_k_i,t and U_k_i,t partition s_i,t and u_i,t into K distinct intervals (t_0,t_1], (t_1,t_2], ..., (t_{K−1},t_K]. - Empirical patterns: - Time since neighboring-country events averages 32 months. - Time since own-country events averages 60 months. - Figure 17 (distribution of time since last event) shows own-country and neighbor-country densities over months since last event (0–300 months plotted). ### Panel regression framework and interpretation - Event equation estimated: - event_i,t = α_i + η_t + Σ_{k∈K} β^d_k S_k_i,t + Σ_{k∈K} β^n_k U_k_i,t + e_it - event_i,t is an indicator for an unrest event; α_i and η_t are country and time fixed effects. - β^d and β^n are interpretable as the incremental probability of an unrest event conditional on a domestic or neighboring event during the last (t_{k−1}, t_k] months. - Sample notes: - The full-sample unconditional average unrest-event rate is a little over one percent of months. - Restricting to observations with at least one prior unrest event raises the unconditional average event probability to nearly two percent. ### Key regression findings (from Table 7) - Own-country (Event) effects (incremental probability, with standard errors in parentheses): - Event, last 2-3 months: 0.037 ∗∗∗ (0.007) in (1); 0.027 ∗∗∗ (0.007) in (2); 0.025 ∗∗∗ (0.007) in (3); 0.028 ∗∗∗ (0.008) in (4). - Event, last 4-6 months: 0.029 ∗∗∗ (0.006); 0.020 ∗∗∗ (0.006); 0.018 ∗∗∗ (0.006); 0.021 ∗∗∗ (0.007). - Event, last 7-9 months: 0.012 ∗∗ (0.005); 0.004 (0.006); 0.002 (0.006); 0.002 (0.006). - Event, last 10-12 months: 0.011 ∗∗ (0.005); 0.003 (0.005); 0.002 (0.006); 0.003 (0.006). - Event, last 13-24 months: 0.004 ∗ (0.002); −0.003 (0.002); −0.004 (0.003); −0.003 (0.003). - Event, last 25-48 months: 0.001 (0.002); −0.004 ∗∗ (0.002); −0.004 (0.002); −0.005 ∗∗ (0.003). - Event, last 49-60 months: 0.004 (0.003); 0.001 (0.003); 0.001 (0.003); 0.001 (0.004). - Neighboring-country (Neighbor event) effects: - Neighbor event, last 2-3 months: 0.018 ∗∗∗ (0.003); 0.017 ∗∗∗ (0.005); 0.016 ∗∗∗ (0.004). - Neighbor event, last 4-6 months: 0.009 ∗∗∗ (0.003); 0.006 (0.004); 0.005 (0.004). - Neighbor event, last 7-9 months: 0.008 ∗∗∗ (0.003); 0.005 (0.004); 0.005 (0.004). - Neighbor event, last 10-12 months: 0.008 ∗∗ (0.003); 0.006 (0.005); 0.006 (0.005). - Neighbor event, last 13-24 months: 0.008 ∗∗∗ (0.002); 0.007 ∗ (0.004); 0.007 ∗∗ (0.003). - Neighbor event, last 25-48 months: 0.002 (0.002); −0.001 (0.003); −0.0003 (0.003). - Neighbor event, last 49-60 months: 0.002 (0.003); 0.002 (0.005); 0.003 (0.005). - Constant terms reported: - Constant: 0.011 ∗∗∗ (0.001) in (1); 0.017 ∗∗∗ (0.001) in (2); 0.012 ∗∗∗ (0.001) in (3). - Fixed effects vary by specification: None; Country; Two-ways; Country (as listed for columns (1)–(8)). - Model fit and sample sizes: - Naive R^2 values: 0.011, 0.017, 0.022, 0.028, 0.046, 0.020, 0.028, 0.032 (columns (1)–(8)). - Observations: 52,736; 26,126; 26,126; 26,126; 26,126; 30,828; 18,768; 18,768 (columns (1)–(8)). - Normalized AIC values shown per column (e.g., -1.706, -1.247, ... -1.177). - R^2 after fixed effects reported as small (e.g., 0.005, 0.003, 0.003, 0.002, 0.001, 0.005) and adjusted R^2 values include negative entries in some specs. - Statistical significance notation: - ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01. - Double-clustered standard errors are shown in parentheses. ### Interpretation and summary implications - Own-country unrest: - Within-country unrest is associated with an approximate 3 percentage point increase in the short-term probability of unrest (an approximate quadrupling relative to baseline), but the effect is statistically insignificant after six months when country fixed effects are included. - Specification (3) shows a rise of nearly 4 percent immediately following an unrest event (partly reflecting selection); correcting for country means (specification (4)) reduces the immediate increase to around 3.5 percent and shortens persistence. - Neighboring-country unrest: - Unrest in neighboring countries is associated with roughly a doubling of the rate of unrest in the focal country; effects are smaller than own-country effects (a little over half the magnitude) but somewhat longer-lasting. - Magnitude and persistence: - Social unrest is rare: usually around 1.3 percent of countries per month experience unrest. - Following an event in the same country, the rate typically triples; following an event in a neighboring country, the rate typically doubles. - These elevated probabilities decay to about 0.5 percent or less within 9 months. - Model fit and diagnostics: - The regressions have fairly poor fit (naive R^2 no better than five percent), reflecting substantial residual variation beyond averages conditional on time since last event. - The Akaike information criterion suggests richer models improve fit (decline in AIC for richer specifications). *Italic: Source: wpiea2020129-print-pdf - Appendix D provides further robustness checks, showing that 1) estimates for the parameters* ### 5.3 Predictability ### 5.3 Predictability ### Overview - Question addressed: can the RSUI itself predict social unrest outbreaks before they occur? - Key empirical observation: the RSUI spikes dramatically in the month of an event and declines slowly thereafter, but displays a moderate increase (around 174 index points) in the period prior to the event. ### Evidence from conditional averages and timing - Left-hand panel (Figure 18): conditional average of the RSUI around an event — sharp spike at event month, slow decline afterward; moderate pre-event increase of around 174 index points. - Right-hand panel (Figure 18): mean RSUI conditioned on observing increases of at least 174 index points — such increases have essentially no predictive power; the RSUI declines sharply back towards its long-run average in subsequent months. - Empirical implication: almost all social unrest events are preceded by moderate increases in the RSUI, but moderate increases in the RSUI are usually followed by a swift regression to the mean. ### Lasso predictive analysis (method and specification) - Model estimated separately for each country using a cross-validated lasso: - Specification: v_{i,t+h} = sum_{j=1}^{M} beta_{j}^{i} v_{i,t-j} - Where M = 18, h = 1,...,24 and v_{it} is either the RSUI or event indicator for country i in period t. - Purpose: calculate the best country- and horizon-specific predictive model for the RSUI using up to a year-and-a-half of prior outcomes data; cross-validation aims to minimize out-of-sample forecast error. ### Measures of predictability computed - Three predictability measures computed for each country-horizon model: - Granger-Newbold measure: fraction of variance of the outcome explained by the forecast (analogous to R^2). - Theil uncertainty statistic: less than one when the forecast is better than the long-run average. - Diebold and Kilian (2001) statistic: measures predictability relative to the long-run error variance. ### Summary statistics of predictive performance (Table 8) - Results compiled by selecting the most predictive horizon for each country and then averaging across countries. Despite selecting the best horizon per country, predictability is low: - Number of non-zero coefficients: Index = 1.03, Events = 0.02 - Horizon of most non-zero coefficients: Index = 1.50, Events = 1.03 - Granger-Newbold predictability: Index = 0.03, Events = 0.00 - Theil U-statistic: Index = 0.92, Events = 1.00 - Diebold-Kilian predictability: Index = 0.11, Events = 0.05 - Additional empirical notes: - The lasso typically selects at most only one lag of the RSUI when predicting future values of the index, and typically at very short lags. - For RSUI events, almost all country-horizon pairs select zero lags, i.e. the best predictive model in almost all cases is identically the country average. - The Granger-Newbold statistic shows that the best forecast explains barely three percent of the variation in outcomes. ### Interpretation - Correlation versus prediction: correlation (or temporal/spatial association) does not imply strong predictive power. Social unrest indicates a higher probability of future unrest, but because social unrest is sufficiently rare and driven by many other factors, using past unrest to predict future unrest yields only a very small improvement in forecasting power. - Rarity and heterogeneous triggers: social unrest events are rare and can be triggered by varied and sometimes minor events (example: increased subway fares). This heterogeneity and rareness make individual events hard to predict. - Analogy used in text: the probability of winning a lottery rises conditional on buying a ticket, but lottery wins are so rare that holding a lottery does not reliably predict winning. ### Implications for monitoring and research - Monitoring: the RSUI is useful for timely, high-frequency monitoring of realized social unrest (spikes and events), but it is not a reliable early-warning predictor of specific outbreaks. - Research uses: because exact timing of social unrest is near-random and regional waves of unrest exist, the RSUI and event listings can be used as instruments or sources of exogenous variation for studying causes and consequences of unrest (for example, impacts on financial markets, investment, trade disruption, or effects of corruption/governance/poverty). - Practical consequence: average effects (as identified in other analyses) can be estimated and interpreted, but predicting individual events remains largely infeasible with the RSUI alone. *Source: wpiea2020129-print-pdf - 5.3 Predictability* ### References to riots in Cairo and ### wpiea2020129-print-pdf - References to riots in Cairo and ### Table 10: Search modifications and reasons - Lists country-specific search exclusions and corrections (examples drawn verbatim): - Bahrain — Wrong country - Mauritanianot ”Dakar” — Riots in Dakar, Senegalese and Mauritanian refugees flee — Wrong country - Mexiconot ”border” — 06/18 picks up protests in US near Mexican border — Wrong country - Montenegronot ”Kremlin” — MNE bids to join NATO — Wrong country - Omannot ”nuclear” — 03/14 Picks up Iran-US nuclear deal talks based in Oman — Diplomatic - Panamanot ”panama papers” — 04/16 picks up Panama papers — Country-Specific - PapuaNew Guineanot ( ”Nauru” OR ”Manus” ) — Riots in Australia’s detention centers in Manus Island — Wrong country - Paraguaynot ”1973” — Articles looked back at protests that happened in Paraguay — Past events - Portugalnot ”Carnation” — Election in Portugal — Past events - Qatarnot ”Syria” — Funds for the Syrian opposition has been allocated and deposited in Qatar — Wrong country - Rwandanot ”StateDepartment” — 10/15 US State Department protest Rwanda’s support for Congolese rebels — Diplomatic - Slovak Republicnot ”velvet revolution” — Parliamentary elections — Country-Specific - Spainnot ( ”UN” near10 ”climate” ) — Global climate protests ahead of U.N. climate talks in Madrid — Wrong country - Taiwan Province of China not ”Vietnam” — Anti-China protest in Vietnam affects Taiwanese businesses — Wrong country - Tajikistannot ”Islamic Revolution” — Throughout: Picks up political party — Country-Specific - Tanzanianot ”Kenya” — Oppositions, US, and Kenya protest over election rerun in Zanzibar — Wrong country - Ukrainenot ”military aid” or impeach* or ”House Intelligence Committee” — UKR mentioned in news articles about Trump impeachment — Wrong country - United Arab Emirates not ”DPW” — 02/06: Picks up controversy over Dubai Port World purchases — Incorrect usage - United Kingdomnot ”Embassy” — 12/11 picks up protests at British Embassy in Tehran — Wrong country - Vietnamnot ”revolution” — Throughout picks up ”revolution” for Party events — Country-Specific - Zambianot ”Zimbabwe” — 08/18 picking up diplomatic incident with Zimbabwe — Diplomatic - Purpose: documents heuristic search modifications and reasons (Wrong country, Diplomatic, Country-Specific, Past events, Incorrect usage). ### Appendix B — Other narrative approaches: comparison of RSUI to narrative sources - Scope: five episodes compared to RSUI: Philippines 1986 (EDSA), Korea “June struggle” 1987, Korea 2016 impeachment protests, end of Apartheid in South Africa, and unrest in five Middle-Eastern countries (1999–2019). - Key methodological point: - RSUI can identify fast-moving protest episodes to the month and is conservative in event coding to reduce false positives. ### B.1 Philippines, 1986 (EDSA / People Power) - Narrative summary (from Schock (1999)): - 1985: “rising protests” following assassination of Benigno Aquino. - 07/02/86: Elections in response to rising protest - 16/02/86: Corazon Aquino rejects results - 22/02/86: Soldiers mutiny for Aquino - 23/02/86: Manila Archbishop condemns elections - 25/02/86: Troops retreat causes major defections - 26/02/86: Marcos desposed, Aquino sworn in - RSUI observations: - Singular peak in February 1986; RSUI identifies February 1986 as an RSUI event. - Figures referenced: Figure 21 and Figure 22 showing month-level RSUI peaks. ### B.2 Korea, 1986–1987 (June Struggle) - Narrative summary (from Choe & Kim (2012)): - 10/86: Geonguk University protests repressed - 14/01/87: Student Death Sparks Protest - 10/06/87: Nationwide Democratic Protests Occur - 26/06/87: Height of the "June Struggle" - 29/06/87: Chun finally agrees to direct presidential elections - RSUI observations: - Pronounced RSUI peak in June 1987; RSUI-implied event dated to June 1987 (Figure 23 and Figure 24). - Demonstrates ability to capture rapid protest escalation. ### B.2.1 The End of Apartheid (South Africa) - Narrative timelines (Waldmeir (1997); Klopp & Zuern (2007)) — selected dated events (asterisks denote unrest events as per source): - 02/90: **ANC, SACP legalized; Mandela released** - 26/03/90: **Police kill 17 protestors at Sebokeng** - 14/07/90: Inkatha Freedom Party (IPF) formed - 22/07/90: IPF−ANC fighting spreads to J'burg area - 08/90: ANC ends armed struggle; violence escalates. - 12/90: Start of multi−party talks - 25/07/91: News that IPF funded by security police - 03/92: Whites−only referendum endorses reform. - 15/06/92: **Negotiations stall, ANC promises demonstrations** - 16/06/92: **ANC mass action begin, IFP attacks residents at Boipatong** - 07/09/92: **Bisho Massacre** - 26/09/92: ANC & National Party sign record of understanding - 04/93: **Head of ANC armed wing assassinated** - 07/93: Agreement on election date; IFP walks out of talks - 11/93: Agree on interim constitution - 12/93: Multi−party interim government begins - 03/94: Overthrow of Bophuthatswana; rout of white right - 27/04/94: **Nonracial, democratic elections; ANC wins** - 05/94: Mandela inaugurated - RSUI observations: - Highest two RSUI peaks match Mandela release (February 1990) and first non-racial elections (late March 1994, peak in April due to monthly frequency). - Intermediate unrest events (Sebokeng deaths, CODESA breakdown, Bisho Massacre, assassination of Chris Hani) correspond to RSUI peaks. - RSUI is conservative: many political but non-unrest events register weakly or not at all on the RSUI. - Figures: Figure 25 (external events) and Figure 26 (RSUI-implied events). ### B.2.2 Korea, 2016–2017 (Candlelight Revolution / Impeachment of President Park) - Narrative timeline (Kim (2017)): - 10/16: Choi Soon−Sil scandal exposed - 29/10/16: Candlelight Revolution begins - 11/11/16: Candlelight Protests continue - 03/12/16: Height of Protests - 09/12/16: National Assembly votes to impeach - 07/01/17: Rally to commemorate Sewol Ferry and 11th Candlelight Protest - 09/03/17: Supreme court upholds impeachment - 09/05/17: Moon wins election - RSUI observations: - RSUI picks up sharp increase in November 2016; November coded as an RSUI event (Figure 28). - March 2017 verdict corresponded with further large protests and three deaths; coded as an RSUI event. - May 2017 election produced a local RSUI spike but was not coded as an RSUI event due to event screening criteria (no contemporaneous reports of unrest in that month). ### B.3 Middle East, 1999–2019 (USIP timelines vs RSUI) - Data sources: USIP timelines for five countries: Egypt, Iran, Iraq, Libya, Tunisia (January 1999–June 2019). - RSUI alignment highlights: - Initial spikes in early 2011 across Tunisia, Egypt, Iraq, Libya, Syria coincide with Arab uprisings as per USIP. - Tunisia: RSUI-implied event in January 2011 corresponding to post-Bouazizi unrest. - Egypt: RSUI dates first Arab Spring RSUI-implied event to February 2011 (first wave started January 25th). - Iraq and Libya: RSUI dates start of Arab Spring to February 2011; Syria to April 2011 — coincides with external timeline. - RSUI captures secondary protests and major turning points (e.g., Egypt: post-election unrest June 2012; protests after Morsi’s decree; Morsi’s overthrow July 2013). - Iran: RSUI captures widespread unrest in January 2018 and June 2009 Green Movement. - Differences noted: - RSUI includes parliamentary/presidential elections in Tunisia (2014) and Iran (June 2013) as events; USIP omits these. - USIP identifies more within-month separate events; RSUI reflects smaller episodes as elevated index levels but may not code them as events. - Three RSUI-implied events not in USIP: August 2011 and November 2011 in Libya (onset of civil war) and July 2018 in Iraq (linked to electricity cuts on July 8th; USIP dates Basra protests to September 2018). - Figures: Figure 29 (USIP events) and Figure 30 (RSUI event coding). ### Section C: Comparison to other measures - Comparative measure mentioned: - ACLED (Armed Conflict Location and Event Database) — includes counts of demonstrations and riots at country level. - Note: ACLED coverage has expanded dramatically since 2015; primary focus historically on African countries with a sample of 27 countries having data starting in [text truncated]. - Purpose: situates RSUI relative to event-level datasets like ACLED; emphasizes differences in coverage, unit of observation, and time-series patterns. *Italic: Content derived from wpiea2020129-print-pdf - References to riots in Cairo and (source PDF). * ### 1998. We therefore create an index for each of these countries analogous to the RSUI, summing ### RSUI validation and event documentation (excerpt) ### Comparison of RSUI with alternative data sources - Construction details: - ACLED-based index: number of events in each month for each country, rescaled so that the index averages to 100 since 2000, combined by taking a simple average across countries in each month for 27 countries. - CNTS-based index: reports number of riots and anti-government demonstrations (annual frequency), summed into country indices and averaged to give an aggregate annual index for 117 countries. - Broad findings: - Despite different phenomena and methodologies, RSUI and ACLED-based index show broad agreement: - Both series relatively low in the first decade of the 2000s and then exhibit a substantial and persistent increase starting in 2011. - Both series show a rise in unrest around 2015, a subsequent decline during 2016-2017 and a pick-up since. - At an annual frequency, the correlation between RSUI and the ACLED-based series is 0.7. - At a monthly frequency, the correlation between the two indices is nearer to one half. - CNTS-based index behavior: - CNTS shows an almost monotonic increase in unrest events from around 2009 onward with very little variation prior. - This pattern is hard to reconcile with RSUI and ACLED evidence. - Possible explanation: CNTS may use a raw count of reports of unrest; increasing media coverage over time could mechanically drive an upward trend. - Comparison to the average rate of RSUI-implied events does not explain CNTS behavior better; increases in the early 2000s and 2010s in RSUI-implied events appear uncorrelated with the CNTS index. - Event-level cross-checks: - For specific validated events (Figures 33–36), RSUI, ACLED (where available), and CNTS show very good agreement. - Notably, timing of spikes in RSUI closely aligns with ACLED index during times of extreme stress. - Figures referenced: - Figure 31: RSUI vs. ACLED, 27 countries, January 2000 - March 2020, simple average (rescaled avg=100). - Figure 32: RSUI vs. CNTS, 117 countries, January 1995 - March 2020, simple average (rescaled avg=100). ### Robustness of RSUI statistical properties - Cross-sectional parameter stability: - Country-specific maximum likelihood estimates (MLE) of parameters a and b were compared to full-sample estimates; error bars show two-standard-error ranges percent confidence intervals. - Full-sample estimates fall within country-specific estimates in over 80 and 90 percent of countries for the estimates of a and b respectively. - Some countries omitted due to MLE convergence failures. - Threshold sensitivity: - Estimates evaluated as the arbitrary cutoff m̄ changes. - Changing m̄ has a mechanical impact on a, proportionate to 1/a; panel (a) shows estimates transformed into the 1/a space (standard errors via delta method). - Parameter a is reasonably stable, tracking the full-sample-implied value for 1/a reasonably closely. - Parameter b is very stable as m̄ rises; the standard error for b actually decreases despite reduction in sample size as m̄ rises. - Interpretation: RSUI is particularly informative around major events; noise is larger when activity is low, but signal becomes increasingly precise for large events. The upper tail of the RSUI distribution is particularly easy to interpret. - Figures referenced: - Figure 37: Country-specific parameter estimates (MLE estimates of a and b; full-sample estimates shown). - Figure 38: Parameter estimates with changing threshold cutoffs (panels showing 1/a and b against RSUI cutoff). ### Verbatim external event descriptions (selected timelines) - Arab Uprisings (verbatim event entries): - Bahrain: 2011-02-14 Protests start in Bahrain. - Bahrain: 2011-03-14 Saudi Arabia sends armored columns into Bahrain. Security forces crush the uprising, carrying out mass arrests within a few days. Bahrain’s king declares martial law. The monument in Pearl Square, the uprising’s heart, is razed. - Egypt: 2011-01-25 Egyptian protests start. - Egypt: 2011-01-28 Egypt’s Friday of Anger. Protestors battle police and win control of Tahrir Square in Cairo. Similar battles in other cities across Egypt. - Egypt: 2011-02-11 Egypt’s President, Hosni Mubarak, steps down. The Supreme Council of the Armed Forces takes power, promising a transition to civilian authority. - Egypt: 2011-03-19 Egypt’s first post revolutionary vote, on a constitutional referendum that could lead to early elections, ends in triumph for the Muslim Brotherhood and hints at rising polarization between Islamist and secularist camps. - Egypt: 2011-11-01 There are more protests and violence in Egypt after the ruling military council moves to ensure it will play a dominant role even after civilian government is elected. - Egypt: 2012-06-01 Muhammed Morsi of the Muslim Brotherhood is elected president of Egypt, narrowly defeating an ally of former president Hosni Mubarak. - Egypt: 2012-06-30 There are huge protests across Egypt against President Muhammed Morsi. A pro-Islamist sit-in begins in Rabaa Square in Cairo. - Egypt: 2012-08-14 Egyptian security forces clear the two main Islamist protest encampments, killing about a thousand people and arresting thousands more. Within weeks, most of the Muslim Brotherhood is jailed. - Egypt: 2012-11-01 The Egyptian president Muhammed Morsi issues a decree temporarily granting himself broad powers above any court as the guardian of the revolution. Large-scale protests begin. - Egypt: 2013-07-03 The Egyptian military deposes President Muhammed Morsi and puts him under arrest in an unknown location. Islamist protests expand and confrontation with police begin, with dozens shot and killed. - Egypt: 2014-03-30 Abdel Fattah al Sisi is elected Egypt’s president. - Libya: 2011-02-15 Libya’s first large-scale protests occur in the eastern city of Benghazi. A police crackdown leads to violent confrontations across Libya. - Libya: 2011-02-27 Libya’s National Transitional Council is formed. Two weeks later, France recognizes it as Libya’s legitimate government; other countries follow. - Libya: 2011-03-17 The United Nations Security Council authorizes military action in Libya. Air strikes begin immediately, with NATO in command. - Libya: 2011-08-20 Muammar al Qaddafi, Libya’s ruler for forty-two years, is captured and killed by rebels while trying to flee from the coastal city of Sirte. - Libya: 2012-09-11 U.S. Ambassador to Libya, J. Christopher Stevens is killed in an attack on American compound in Benghazi following violent protests. - Libya: 2012-11-01 Amid rising lawlessness in Libya, nine people are killed in clashes between the military and Islamist militia fighters in Benghazi. - Libya: 2014-06-25 Libyan elections are marked by low turnout, and are followed by clashes between followers of outgoing and incoming parliaments. Islamists carry out a coup in Tripoli. - Syria: 2011-03-15 Demonstrations start in Syria, triggered by police mistreatment of teenagers arrested in the southern town of Daraa for writing anti-government graffiti. Protests spread quickly and are met with mass arrests and shootings. - Tunisia: 2010-12-17 Mohamed Bouazizi sets himself on fire in Sidi Bouzid, Tunisia. Protests start and spread across Tunisia. - Tunisia: 2011-01-14 Tunisia’s president, Zine El Abidine Ben Ali, capitulates in the face of rising protests and flees the country with his inner circle. - Tunisia: 2011-10-23 Tunisia holds parliamentary elections, the first to take place since the uprising. Ennanda, the Islamist party led by Rached Ghannouchi, wins a plurality and subsequently forms a coalition government. - Tunisia: 2013-07-25 In Tunis, Mohamed Brahmi is assassinated, making him the second opposition leader to be killed in five months. Protests against the Islamist-led government intensify. - Tunisia: 2013-09-28 The Tunisian crisis eases. An independent caretaker government takes power and dialogue on a new constitution begins. - Yemen: 2011-01-27 The first large-scale protests in Yemen take place. - Yemen: 2011-03-18 Gunman aligned with Yemen’s government open fire on protestors in the capital, killing at least fifty-two and prompting top military officers to defect. President Ali Abdullah Saleh signals soon afterward that he may step down before the year’s end. Much of the country falls out of the government’s control. - Yemen: 2011-06-03 The Yemeni President Ali Abdullah Saleh is seriously injured by a bombing in the mosque of his compound. He is flown to Saudi Arabia for treatment, leaving the country rudderless. - Yemen: 2011-11-23 After months of evasive maneuvering, the Yemeni president Ali Abd- uleh Saleh flies to Riyadh, Saudi Arabia, to sign a transition document brokered by Arab and American diplomats. - Yemen: 2014-09-10 In Yemen, the Houthi rebels, now in league with the former Yemeni president Ali Abdullah Saleh, capture Sanaa, becoming the country’s de facto rulers. - (Table 11: Verbatim descriptions of Arab Uprising events from Worth (2016)) - Venezuela (selected verbatim entries, 2014–2019): - Feb 2014 Briceno-Ruiz Between February and April, street protests occur supporting the Exit, a proposal pushed by Leopoldo Lopez, Antonio Ledezma, and Maria Corina Mochado to oust the regime. Protests are brutally suppressed. - Dec 2014 Briceno-Ruiz Maduro calls extraordinary session of the National Assembly during recess to appoint his own judges, which would have been appointed by the National Assembly in 2016. - Dec 2015 Gutierrez Maduro and Chavismo suffer significant electoral defeat when the opposition obtains the necessary majority in the National Assembly. - Dec 2015 Briceno-Ruiz Vacation is suspended in the Supreme Court to hear a case against the results of the National Assembly election, which holds Opposition in contempt. - 2016-01-15 Gutierrez Maduro begins governing using an illegal decree of economic emergency, which has not been approved by the National Assembly. - Apr 2017 Briceno-Ruiz A new wave of protests occurs from April to August 2017, and which are repressed violently by government. Torture, arbitrary arrests and human rights violations of young leaders occurs. - 2019-01-05 Briceno-Ruiz Condeming Maduro for usurping controversial May 2018 elections, the National Assembly declares Juan Guaido National Assembly president for the 2019 period. - 2019-01-13 Briceno-Ruiz Guaido is arrested and then released by members of the Bolivarian Intelligence Service (SEBIN), in a police raid, after which he is freed. A muddled explanation from the Minister of Information says it is carried out by rogue elements. - 2019-01-21 Briceno-Ruiz The Supreme Court, controlled by Maduro, declares Guaido’s appointment null, and all acts that arrive from it void. - 2019-01-23 Briceno-Ruiz Guaido calls for a huge protest on January 23, a symbolic day for Venezuelans marking when former dictator Marcos Peres Jimenez was overthrown in 1958. - 2019-01-24 Briceno-Ruiz The high military commander, controlled by Chavismo, gives his unconditional support to Maduro. - 2019-01-25 Briceno-Ruiz In a public appearance in front of 1,000 Venezuelans in Caracas, Guaido declares himself interim president of Venezuela. Mass protests begin. - (Table 12: Verbatim descriptions of Venezuelan events 2013-2019 from Gutierrez (2017) and Briceño-Ruiz (2019)) - Thailand (selected verbatim entries, 2006–2014): - 2006-01-23 Keyes Protests intensify when Thaksin and his family sell their holdings worth nearly two billion bhat in the telecommunications Shin Corporation to a Singapore holding company (Temasek) without incurring any taxes. Tax avoidance seen as corruption. - 2006-01-01 Keyes By January or February, a large but loosely-connected movement begins to mobilize protests calling for Thaksin’s resignation. The movement, which takes the name People’s Alliance for Democracy (PAD), begins with attacks on Thaksin’s ethical conduct. - 2006-09-19 Keyes The army, led by General Soonthi Boonyaratglin, stages a coup. Thaksin’s supporters put up no resistance and no blood is shed. - 2008-11-26 Prasirtsuk (2009) The PAD escalate their pressure by seizing both major airports in Bangkok (Suvanaphumi and Donmuang) on November 26 in a blackmail strategy to paralyze the government. - 2010-05-01 Dalpino Red shirt leaders surrender or go underground, and protestors set 39 buildings on fire, including the stock exchange and Central World mall. Estimates count 91 dead. - 2014-05-20 Prasirtuk (2015) Three months after post-election stalemate, on May 20, the military declares martial law to restore peace and stability. - 2014-05-22 Prasirtuk (2015) Army Commander General Prayuth Chan-ocha calls leaders of both sides to a negotiation table at the army club, acting as mediator. On May 22, after talks fail, Prayuth abruptly declares the coup. - (Table 13: Verbatim descriptions) - Color Revolutions (selected verbatim entries): - Serbia: 2000-10-05 Opposition leaders call for a total blockade of all institutions in Serbia, and over 200,000 people take to the streets of Belgrade....Over half a million people march on Belgrade and quickly seized control of major government institutions. - Serbia: 2000-10-06 The following day, Milosevic resigns. - Georgia: 2003-11-02 Following the November 2, 2003, Georgian parliamentary elections, small-scale protests begin in Tbilisi as various reports of fraud begin to accumulate. - Georgia: 2003-11-23 After originally claiming that he would not step down, Shevardnaze resigns the following day. - Georgia: 2004-01-04 New presidential elections are set for January 4, 2004, which Saakashvili wins with 96 percent of the vote in an essentially uncontested election. - Ukraine: 2004-10-31 On October 31, 2004, Viktor Yanukovich and Viktor Yushchenko receive 39.3 percent and 39.9 percent of the vote, respectively. - Ukraine: 2004-11-21 After reports of voter fraud, the Central Election Commission declares Yanukovich the winner by a 49.5 percent to 46.6 percent margin. - Ukraine: 2004-12-03 Ukraine’s Supreme Court declares the second round election results to be invalid and orders that the round be run again on December 26. - Ukraine: 2004-12-26 The second round features over 12,000 international observers, and Yushchenko wins with 52.0 percent to 44.2 percent. - Kyrgyz Rep.: 2005-02-27 and 2005-03-15 Protests begin and culminate with 30,000 protesters converging in the main square of Bishkek; Akayev eventually agrees to resign. - (Table 14: Verbatim descriptions from Tucker (2007)) - Philippines 1986 (selected verbatim entries): - 1986-02-07 Schock As a response to rising protests, Marcos calls for snap elections to be held on February 7, 1983. - 1986-02-16 Schock Widow of Benigno, Corazon Aquino, responding to the fraudulent declaration of electoral victory for Marcos by the Batasang Pambansa (national assembly), led a rally of over two million people, proclaiming victory for herself and the people against Marcos. - 1986-02-22 Schock Defense Minister Jaun Ponce Enrile leads two battalions of soldiers in a mutiny, barricading two military camps in Quezon City. He is joined by General Fidel Ramos, and they announce their withdrawel of support from Marcos for Aquino. - 1986-02-25 Schock Marcos orders two battalions to the military camps to put down the mutiny, but as the tanks approach civilians form human barricades. The loyalist troops retreat in the face of mass unarmed civilians led by priests and nuns, which sparks nationwide defections. - 1986-02-26 Schock Corazon Aquino is sworn in as president and Marcos is escorted to Honalulu by the US military. - (Table 15: Verbatim descriptions) - Korea 1987 (selected verbatim entries): - Oct 1986 Choe After protests at Geonguk University in October 1986, the Chun government harshly surppreses the protesting students and claims they are supporters of the communist North Korean regime. - 1987-01-14 Choe Many Korean citizens participate in a protest condemning the Chun regime for killing Park Jong Chul, a student activist, under torture. - 1987-04-13 Choe The Chun government announces Protection of the Constitution, revoking promise and prohibiting any discussion of constitutional revision, a peaceful government transfer under the provisions of the present constitution, or the enforcement of the presidential election by the end of 1987. - 1987-06-10 Choe Nationwide democratic movement known as June Struggle begins, and more than 400,000 participate despite the Chun government mobilizing 60,000 police to supress the rallies. - 1987-06-26 Choe More than 1,400,000 people in all parts of country join the democratic movements under the slogan of ’Cancellation of the April 13 Announcement of the Protection of the Constitution and Constitutional Revision of Direct Presidential Elections’. - 1987-06-29 Choe The Chun government makes a democratic declaration and agrees to constitutional revision for direct presidential elections. - (Table 16: Verbatim descriptions) - End of Apartheid (selected verbatim entries, 1990–1994): - Feb 1990 Waldmeir de Klerk legalizes the ANC and SACP and releases Mandela. - 1990-03-26 Klopp & Zuern In a protest in Sebokeng, the police fire upon a crowd, killing seventeen unarmed protestors and injuring over 400. The ANC responds by suspending prenegotiation talks. - 1990-07-14 Klopp & Zuern Inkatha’s leader Mangosuthu Bethuelezi announces that Inkatha will become a national political party open to all races: the Inkatha Freedom Party (IPF). - 1990-07-22 Klopp & Zuern Fighting between Inkatha and the ANC, which has become endemic in KwaZulu-Natal, spreads to the Johannesburg area. - Aug 1990 Waldmeir ANC announces it is suspending armed struggle; violence escalates. - Dec 1990 Waldmeir Start of formal multi-party talks, the Convention for a Democratic South Africa (CODESA). - 1991-07-25 Klopp & Zuern The Weekly Mail breaks Inkathagate, revealing a state-supported security police operation that funded Inkatha in its violent campaign against the ANC. - Mar 1992 Waldmeir Whites-only referendum endorses reform. - 1992-06-15 Klopp & Zuern When multiparty negotiations at CODESA reach a deadlock, the ANC responds with a promise to demonstrate. - 1992-06-16 Klopp & Zuern After ANC mass action begins, IFP supporters attack residents at Boipatong. - 1992-09-26 Klopp & Zuern The ANC and the National Party sign the crucial Record of Understanding on September 26, 1992, effectively marginalizing the IFP. - Apr 1993 Waldmeir Chris Hani (head of ANC armed-wing) assassinated. - Jul 1993 Waldmeir Agreement reached on election date, but IFP walks out of talks. - Nov 1993 Waldmeir Agreement on an interim constitution. - Dec 1993 Waldmeir Transitional Executive Council, the multi-party interim government, begins to operate. - Mar 1994 Waldmeir Overthrow of Bophuthatswana (a Bhantuland), rout of the white right. - 1994-04-27 Klopp & Zuern South Africa holds its first nonracial, democratic elections, with ANC the victor. - May 1994 Waldmeir Mandela is inaugurated. Government of National Unity formed. - (Table 17: Verbatim descriptions of South African events 1990-1994 from Klopp & Zuern (2007) (daily) and Waldmeir (1997) (monthly)) - Korea 2016 (selected verbatim entries): - Oct 2016 Kim A scandal exposes that Choi, personal confidante of Park, has been conspiring with the president to amass a fortune, shape national policy and secure positions for her friends and relatives. - Oct 2016 Kim Blacklist is discovered and published by Hankook Ilbo showing that the Park Administration targeted creative professionals who were seen as anti-government. The blacklist includes over 9,000 visual artists, writers, film directors, and others critical of the Sewol Ferry sinking. - 2016-10-29 Kim Beginning on October 29th, weekly anti-Park protests are held on an enormous scale for 17 consecutive weeks, - 2016-12-03 Kim At protest height on December 3rd, the total estimate of people who join the protests on a single day exceeds 2.3 million people across the country - 1.6 milion in Seoul alone. - 2016-12-09 Kim After weeks of debate, on December 9th the South Korean National Assembly votes to pass an impeachment bill, holding that the President has violated her oath of office as well as core tenets of the Constitution and laws governing the operation of the presidency. - 2017-03-09 Kim Constitutional Court justices appear on national television, ruling 8-0 to uphold the National Assembly vote on impeachment. - 2017-05-09 Kim In snap election held on May 9th, liberal Moon Jae-in wins by a landslide, taking 41 percent of the vote, and begins his term. - (Table 18: Verbatim descriptions) - USIP Middle East timelines (selected verbatim entries): - Iran: 1999-07-08 Students demonstrated at Tehran University after reformist Salaam newspaper was closed by the judiciary. Protests continued for six days. More than 1,000 students were arrested. - Iran: 2009-06-13 The opposition Green Movement launched the most serious challenge to the theocracy since the revolution. Millions turned out on the streets of several Iranian cities for the next two weeks with banners declaring Where is my vote? - Tunisia: 2010-12-17 Mohammed Bouazizi, a street vendor, set himself on fire to protest government corruption, igniting growing protests throughout the month. - Tunisia: 2012-06-12 In Tunis, Salafis rioted, protesting an art exhibit that they found offensive. One man was killed, 62 policemen were injured, and 160 protesters were arrested. A curfew was imposed. - Tunisia: 2018-11-22 Some 650,000 civil servants went on strike to protest mounting inflation and the governments refusal to raise wages. Thousands demonstrated nationwide. The prime minister had been under pressure from international lenders who threatened to stop financing Tunisia's economy if the government did not tackle its large budget deficit. In November, nearly 750 protests, mainly over the economic situation, were held across Tunisia. - West Bank & Gaza: 2018-03-30 Tens of thousands of Gazans began six weeks of Friday protests called the Great March of Return at the border fence with Israel. Initially led by independent activists, the demonstrations were co-opted by Hamas. Demonstrators threw Molotov cocktails, launched stones with slings, burned tires, or sought to breach the fence. Israel responded with live fire, killing more than 100 Palestinians and wounding thousands. - (Table 19: USIP timeline of events in the Middle East) *Source: Excerpt from working paper (RSUI validation, robustness checks, and verbatim external event descriptions).* --- _Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020129-print-pdf.pdf_