The Macroeconomic

Effects of Social Unrest

NOVEMBER 23, 2020

Metodij Hadzi-Vaskov, Samuel Pienknagura, and Luca Antonio Ricci

INTERNATIONAL MONETARY FUND 1

 Motivation and Literature

INTERNATIONAL MONETARY FUND 2

 Motivation: the issue
• Social unrest has been increasing in recent years:
• Frequency of severe social protests has increased
recently, affecting diverse places from France and
Hong Kong in AEs to Chile and Lebanon in EMs,
among others in 2019
• The latest Global Peace Index (2020) suggests that
the number of riots, general strikes and anti-
government demonstrations around the world

1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
increased by 244 per cent over the last decade
(2011 to 2019)
Source: Authors' calculations based on Barrett, Appendino, Nguyen and de Leon Miranda (2020).
• Social unrest is likely to pick up again after the pandemic due to the adverse impact on inequality and poverty
(Furceri, Loungani, Ostry, and Pizzuto, 2020; Ahmed, Ahmed, Pissarides, and Stiglitz, 2020)

• Studies on macroeconomic effects of social unrest are limited (more on large conflicts)
• Conflicts are a key hurdle to economic growth (Rodrik, 1999), produce persistent negative effects on output (Cerra and
Saxena, 2008; IMF, 2019; Rother et al. 2016), and result in large overall macroeconomic costs (Novta and Pugacheva, 2020)
• Unrest usually an element of broader political instability that negatively affects growth (Alesina et al. 1996; Aisen and Veiga,
2013; Jong-A-Pin, 2009)
• Recent IMF work: Saadi Sedik and Xu (2020); Hlatshwayo and Redl (2020); Barrett et al. (2020)
INTERNATIONAL MONETARY FUND 3
 Motivation: what we do
• We leverage on a novel index.
• Barrett, Appendino, Nguyen and de Leon Miranda (2020).
• Timely, transparent, high-frequency indicator with broad and consistent cross-country
coverage based on counts in relevant media reports.
• Compared to existing indicators, it provides higher frequency (than CNTSD, annual)
and broader coverage (than ACLED, mainly SSA) with more objectivity and replicability
(than ICRG, subjective assessment).

• We thus investigate the macroeconomic impact considering:

• a broader pool of social events, including less severe than civil conflicts (compared to
the literature looking at the economic costs of conflicts);
• impact also on higher frequency indicators;
• demand & supply components;

INTERNATIONAL MONETARY FUND 4

 Preview
• Economic activity declines following spikes in the unrest index, with GDP remaining
on average ¼ pp below the pre-shock baseline after 6 quarters.

• Adverse effect on GDP is driven by sharp contractions of services and

manufacturing (sectoral decomposition) and consumption (demand decomposition).

• Social unrest lowers confidence, raises uncertainty, albeit its adverse economic
effects can be dampened by strong institutions and policy space.

• Economic impact differs by type of unrest episodes:

mainly due to politics/elections < socio-economic reasons < both

• Results are robust to controlling for previous adverse growth events

INTERNATIONAL MONETARY FUND 5
 Data and methodology

INTERNATIONAL MONETARY FUND 6

 Data on unrest

• Social unrest data comes from Barrett, Appendino, Nguyen and de Leon Miranda
(2020).

• They propose a monthly news-based index starting in 1985m1 that quantifies social unrest for
a large set of countries—the reported social unrest index (RSUI).
• The primary source is Dow Jones' Factiva news aggregator. They restrict their sample to
printed articles published in major English-language newspapers and networks (USA, CAN,
UK).
• The index takes the following form:

• Where 𝑥𝑖𝑡 is the article count related to unrest events in country i in month t; 𝑧𝑡 is the article
count in period t; 𝑥ഥ𝑖 and 𝑧ҧ are the averages over all time periods.
INTERNATIONAL MONETARY FUND 7
 Data on unrest (II)

• To obtain the article counts, the authors apply the following search criteria:

INTERNATIONAL MONETARY FUND 8

 RSUI

Source: Barrett et al. (2020). Source: Barrett et al. (2020).

INTERNATIONAL MONETARY FUND 9

 Data on unrest (III)
• In addition to constructing the index, the authors propose an algorithm to identify RSUI-implied
events.

• An RSUI-implied event has to satisfy the following three criteria:

1. It must be a local peak: 𝑅𝑆𝑈𝐼𝑖𝑡 = 𝑚𝑎𝑥 𝑅𝑆𝑈𝐼𝑖𝑡+1 , 𝑅𝑆𝑈𝐼𝑖𝑡−1
2. The index satisfies one of the following:
▪ 𝑅𝑆𝑈𝐼𝑖𝑡 > 𝑅𝑆𝑈𝐼𝑖𝑡 + (4 ∗ 𝑠𝑑(𝑅𝑆𝑈𝐼𝑖𝑡 )) or
▪ 𝑅𝑆𝑈𝐼𝑖𝑡 is in the top 2% or
▪ 𝑅𝑆𝑈𝐼𝑖𝑡 exceed the 20-year moving average by 4 times the 20-year st. dev.
3. The unrest article count for the month is at least 10 percent the average monthly count for the country
over the past 12 months.

• We label an event satisfying all five criteria as a major event.

INTERNATIONAL MONETARY FUND 10

 Data on unrest (IV)

• In the empirical analysis we use the dataset constructed by Barrett, Appendino, Nguyen and de
Leon Miranda (2020) as follows:

o For each country, we aggregate the social unrest index (RSUI) at the quarterly level by
taking the maximum level over the quarter.

o We also aggregate the RSUI-implied event dummy constructed by the authors and the
major event dummy at the quarterly level (taking the maximum over the quarter) and focus
on new events—events that are at least 8 quarters apart from each other.

o We will use both the index and the event, in separate exercises

o Based on the event’s main underlying cause/trigger, we distinguish between 3 types:

(1) Political/elections (2) Socio-economic (3) Mixed

INTERNATIONAL MONETARY FUND 11

 Volatility of RSUI varies substantially across countries

1200

1000

800
Based on quarterly data ,where the quarter is defined by the max of the 3 months

600

400

200

St. Dev. RSUI Mean

INTERNATIONAL MONETARY FUND 12

 Macroeconomic data and sample

• We include all 96 countries with both RSUI and quarterly GDP data over the
period 1990-2018, excluding fragile states.
• Quarterly GDP and other NA accounts data comes from national sources (SA by
authorities if available, otherwise we use Haver for SA).
• Debt/GDP comes from the Fiscal Monitor database.
• Rule of law estimates come from the World Governance Indicators (Kaufmann,
Kraay, and Mastruzzi, 2010)
• Data on exchange rate regimes comes from Ilzetzki, Reinhart and Rogoff (2019).
• Data for confidence indicators, monthly economic activity and industrial production
are from Haver.
• Data on uncertainty comes from the latest vintage of the World Uncertainty Index
proposed by Ahir, Bloom and Furceri (2018).
• Excluded fragile states as defined by the World Bank in at least one year since 2006.

INTERNATIONAL MONETARY FUND 13

 Macroeconomic data and sample

Advanced economies (33): Emerging Mkts and Low-income Countries (63):

Australia, Austria, Belgium, Canada, Cyprus, Albania, Algeria, Argentina, Armenia, Azerbaijan,
Czech Republic, Denmark, Estonia, Finland, Bahrain, Bangladesh, Bolivia, Brazil, Bulgaria,
France, Germany, Greece, Hong Kong, Burkina Faso, Chile, China, Colombia, Croatia,
Ireland, Israel, Italy, Japan, Korea, Latvia, Ecuador, Egypt, Ethiopia, Ghana, Guatemala,
Lithuania, Netherlands, New Zealand, Honduras, Hungary, India, Indonesia, Iran,
Norway, Portugal, Singapore, Slovakia, Jordan, Kazakhstan, Kenya, Kuwait, Kyrgyz
Slovenia, Spain, Sweden, Switzerland, Republic, Macedonia, Malaysia, Mexico,
Taiwan PoC, UK, United States Moldova, Montenegro, Morocco, Nicaragua,
Niger, Pakistan, Panama, Paraguay, Peru,
Philippines, Poland, Qatar, Romania, Russia,
Rwanda, Saudi Arabia, Senegal, Serbia, South
Africa, Sri Lanka, Tanzania, Thailand, Tunisia,
Turkey, Uganda, Ukraine, United Arab Emirates,
Venezuela, Vietnam, Zambia

INTERNATIONAL MONETARY FUND 14

 Econometric approach

• We assess the macroeconomic impact of episodes of social unrest using the local projection
method proposed by Jordà (2005) and Teulings and Zubanov (2014).

• The procedure does not constrain the shape of the impulse response functions and is less
sensitive to misspecification than estimates of VAR models.

• The benchmark specification at a quarterly frequency is as follows:

𝑦𝑖,𝑡+ℎ − 𝑦𝑖,𝑡−1 = 𝛼𝑖ℎ + 𝛾𝑡ℎ + 𝛽 ℎ unrest 𝑖,𝑡 +𝛿𝑋𝑖,𝑡 + 𝜀𝑖,𝑡+ℎ

• 𝑦 is the variable of interest (GDP, sectoral value added, demand components, confidence);
unrest is either the index proposed by Barrett et al. 2020 (RSUI) or the event dummy;
and X are a set of controls that includes past values of the dependent variable and past and
future values of the shock (in the case of regressions using the RSUI).

• One of the main advantages of the LP method in estimating the effects of shocks is its
flexibility in dealing with non-linearities and state dependency (Ramey and Zubairy, 2018).

INTERNATIONAL MONETARY FUND 15

 Results: Impact of RSUI

INTERNATIONAL MONETARY FUND 16

 GDP declines following spikes in the unrest index

GDP • GDP experiences a steady decline following a shock

Response to a one st.dev. to the unrest index of one standard deviation in the unrest index.

• GDP declines by -0.1 percentage points qoq on

impact.

• After 6 quarters, quarterly GDP remains .25

percentage points below its pre-shock level.

• A shock of one standard deviation is equivalent to

the protests following the Peña Nieto election in
2012 or Chile’s presidential election protests in
2013.

• For comparison, the protests of July 2019 in HKG

and the yellow vest protests of 2018 in France
resulted in an increase of 4 st.dev. in RSUI, while
the events of Oct/Nov 2019 in Chile resulted in a 10
st.dev. increase in RSUI.
Standard errors are clustered at the country level.

INTERNATIONAL MONETARY FUND 17

 The adverse effects on GDP appear to be driven by a
sharp contraction in manufacturing and services
Agriculture Manufacturing Services
Response to a one st.dev. to the unrest index Response to a one st.dev. to the unrest index Response to a one st.dev. to the unrest index

Standard errors are clustered at the country level.

Standard errors are clustered at the country level. Standard errors are clustered at the country level.

INTERNATIONAL MONETARY FUND 18

 On the demand side, the effect of unrest is more evident
on consumption
Consumption Investment Trade balance
Response to a one st.dev. to the unrest index Response to a one st.dev. to the unrest index Response to a one st.dev. to the unrest index

Standard errors are clustered at the country level.

Standard errors are clustered at the country level. Standard errors are clustered at the country level.

INTERNATIONAL MONETARY FUND 19

 Unrest increases uncertainty and somewhat dampens
consumer confidence

World Uncertainty Index for the country Consumer confidence Business Confidence
Response to a one st.dev. to the unrest index Response to a one st.dev. to the unrest index Response to a one st.dev. to the unrest index

Standard errors are clustered at the country level. Standard errors are clustered at the country level. Standard errors are clustered at the country level.

INTERNATIONAL MONETARY FUND 20

The adverse effects of unrest are evident in all countries regardless
of income levels, but impact is (about 3 times) larger in EMs

GDP, advanced economies GDP, EMs

Response to a one st.dev. to the unrest index Response to a one st.dev. to the unrest index

Standard errors are clustered at the country level. Standard errors are clustered at the country level.

INTERNATIONAL MONETARY FUND 21

 Results: role of institutions and policy space

INTERNATIONAL MONETARY FUND 22

 Econometric approach (state-dependent LP)
• In addition to the benchmark regression presented previously, we explore specifications that
condition the response of activity to social unrest shocks to particular states S.

• The typical specification takes the following form (unrest=RSUI)

𝑦𝑖,𝑡+ℎ − 𝑦𝑖,𝑡−1 =
ℎ ℎ ℎ
𝐹(𝑧𝑖,𝑡−1 ) 𝛼ℎ𝑖𝑔ℎ,𝑖 + 𝛾ℎ𝑖𝑔ℎ,𝑡 + 𝛽ℎ𝑖𝑔ℎ 𝑢𝑛𝑟𝑒𝑠𝑡𝑖,𝑡 + 𝛿ℎ𝑖𝑔ℎ 𝑋𝑖,𝑡 +
ℎ ℎ ℎ
(1 − 𝐹(𝑧𝑖,𝑡−1 )) 𝛼𝑙𝑜𝑤,𝑖 + 𝛾𝑙𝑜𝑤,𝑡 + 𝛽𝑙𝑜𝑤 unrest 𝑖,𝑡 + 𝛿𝑙𝑜𝑤 𝑋𝑖,𝑡 + 𝜀𝑖,𝑡+ℎ .

• 𝐹(𝑧𝑖,𝑡−1 ) is a smooth function of the state variable taking a value between 0 and 1; z is a
1
normalized version of the state variable such that 𝐹 0 = 2 .

• The state variables account for:

i) Rule of law; ii) Debt level; iii) Exchange rate flexibility; iv) Product market competition, and v)
Labor market flexibility

• As is common in the literature, we assume 𝐹(𝑧𝑖,𝑡−1 ) = exp −𝜆0 𝑧𝑖,𝑡−1 / 1 + exp −𝜆0 𝑧𝑖,𝑡−1

INTERNATIONAL MONETARY FUND 23

 Strong institutions and available policy space can dampen
the adverse effects of unrest
Black (red) lines corresponding to weak (strong) institutions/policy space

GDP’s response to social unrest, by level of GDP’s response to social unrest, by debt level GDP’s response to social unrest, by XR-
rule of law flexibility

Standard errors are clustered at the country level. Standard errors are clustered at the country level. Standard errors are clustered at the country level.

INTERNATIONAL MONETARY FUND 24

 Some structural factors affect performance following
episodes of unrest, others do not
Black (red) lines corresponding to low (high) competition/flexibility

GDP’s response to social unrest, by level of GDP’s response to social unrest, by level of
Product Market Competition Labor Market Flexibility • Prior was for structural factors to
affect medium-term performance,
not necessarily the immediate
response to unrest events.

• Open question: what could be

plausible mechanisms through which
low PMC amplifies the adverse
effects of social unrest?

Standard errors are clustered at the country level. Standard errors are clustered at the country level. Standard errors are clustered at the country level.

INTERNATIONAL MONETARY FUND 25

 Results: large events and different types of episodes

INTERNATIONAL MONETARY FUND 26

 Not surprisingly, large unrest events lead to larger GDP
contractions (twice as much)
GDP GDP
Response after a new event (identified by algorithm) Response after a new major event (identified by algorithm)

Standard errors are clustered at the country level. Standard errors are clustered at the country level.

INTERNATIONAL MONETARY FUND 27

 Unrest episodes motivated by socio-economic issues lead to sharper
GDP contractions than episodes related to politics/elections…

GDP GDP
Response to events triggered by politics Response to events associated with social issues

Standard errors are clustered at the country level. Standard errors are clustered at the country level.

INTERNATIONAL MONETARY FUND 28

 …while episodes triggered by intertwined socio-economic and
political issues are associated with largest GDP contractions
GDP GDP
Response to events associated with both Response to events that cannot be labeled

Standard errors are clustered at the country level. Standard errors are clustered at the country level.

INTERNATIONAL MONETARY FUND 29

 Addressing endogeneity concerns

INTERNATIONAL MONETARY FUND 30

 No clear growth decline pre-unrest, apart from in t-3
average growth deviation below zero for 4q after unrest
Growth distribution around new events Distribution of growth deviations around new Distribution of growth deviations around new
events: simple average events: moving average

INTERNATIONAL MONETARY FUND 31

 Methodology (I)
We refine our regressions to control for potential endogeneity both when using RSUI and events. When using
“events” we run the following regression:

𝑦𝑖,𝑡+ℎ − 𝑦𝑖,𝑡−1 = 𝛼𝑖ℎ + 𝛾𝑡ℎ + 𝛽𝑠𝑔

ℎ
sg event 𝑖,𝑡 + 𝛽𝑠ℎ sevent 𝑖,𝑡 + 𝛽𝑔ℎ 𝑔event 𝑖,𝑡 +𝛿𝑋𝑖,𝑡 + 𝜀𝑖,𝑡+ℎ

• sg event 𝑖,𝑡 : a dummy taking value one if country i experienced a social unrest event in period t and
experienced two or three quarters of below average growth in the three quarters preceding the unrest
event;

• sevent 𝑖,𝑡 : takes value one if country i experienced a social unrest event and did not experienced two or
three quarters of below average growth in the three quarters preceding the unrest event;

• g event 𝑖,𝑡 : takes value one if country i experienced two or three quarters of below average growth prior to t
and did not experience a social unrest event. The excluded group are countries that neither experienced
unrest nor low growth.

• Growth events based on negative deviations from country averages (at least 2 out of the last 3 quarters).
Robust to alternative definitions based on: similar deviations from MA, or from zero growth, as well as
negative growth in t-1
INTERNATIONAL MONETARY FUND 32
 Methodology (II)
We follow a slightly modified approach when using RSUI as our shock. In this case we run the following LP
model:

𝑦𝑖,𝑡+ℎ − 𝑦𝑖,𝑡−1
= 𝛼𝑖ℎ + 𝛾𝑡ℎ + 𝛽𝑙𝑜𝑤
ℎ
∗ 𝑙𝑜𝑤𝑖,𝑡 ∗ 𝑅𝑆𝑈𝐼
ℎ ℎ ℎ
+ 𝛽𝑛𝑜 𝑙𝑜𝑤 ∗ 1 − 𝑙𝑜𝑤𝑖,𝑡 ∗ 𝑅𝑆𝑈𝐼𝑖,𝑡 +𝛿𝑙𝑜𝑤 ∗ 𝑙𝑜𝑤𝑖,𝑡 +𝛿𝑙𝑜𝑤 ∗ (1 − 𝑙𝑜𝑤𝑖,𝑡 ) +𝛿𝑋𝑖,𝑡 + 𝜀𝑖,𝑡+ℎ

Where 𝑙𝑜𝑤𝑖,𝑡 is a dummy variable that takes value one if country i experienced two or three quarters of below
average growth prior.

• Thus, 𝛽𝑙𝑜𝑤
ℎ
captures the effect of an increase in the social unrest index (RSUI) in countries with prior low
ℎ
growth, and 𝛽𝑛𝑜 𝑙𝑜𝑤 captures the effect of an increase in the social unrest index (RSUI) in countries with no
prior low growth.

INTERNATIONAL MONETARY FUND 33

 Social unrest affects growth even when controlling for
prior adverse growth events (I)
Panel 1. Impulse response of events of both Panel 2. Impulse response of events of low
social unrest and low growth growth and no social unrest
• Panel 3 shows that social unrest affects
growth even if not preceded by adverse
growth events

• Panels 1 and 2 show that adverse growth

events are persistent, but when followed by
social unrest then the future growth impact Panel 3. Impulse response of social unrest and Panel 4. Difference in impulse response
no low growth between events of low growth and social
is larger unrest and low growth with no unrest

• Indeed, Panel 4 shows that social unrest

significantly worsens the growth impact of
adverse growth events

INTERNATIONAL MONETARY FUND 34

 Social unrest affects growth even when controlling for
prior adverse growth events (II)

• Similar results when using RSUI.

• A one st. dev. spike in RSUI lowers GDP in

countries that experienced prior low growth.

• Similarly, spikes in RSUI diminishes GDP in

countries that were not experiencing low
growth, albeit effects are smaller and last
fewer quarters.

INTERNATIONAL MONETARY FUND 35

 Conclusions

INTERNATIONAL MONETARY FUND 36

 Conclusions (I)

• Economic activity declines following a one standard deviation increase in the unrest index, with GDP
remaining on average ¼ pp below the pre-shock level after 6 quarters

• Adverse effect on GDP is driven by sharp contractions of:

• Manufacturing and Services (from sectoral perspective); and
• Consumption (from demand side)

• Unrest is found to:

• dampen (consumer) confidence and raise uncertainty
• adversely affect both AE and EM, though effect seems larger for EMs

• Strong institutions and available policy space can dampen the adverse effects of unrest

INTERNATIONAL MONETARY FUND 37

 Conclusions (II)

• Effects are larger when considering events of social unrest—which are increases of at least 4
standard deviations in the RSUI.

o On average, GDP remains 2 percentage points below baseline after 6 quarters of a social unrest
event.

o For comparison, the impact of social unrest events is larger than the impact on GDP of a negative
shock (2 st.dev.) to commodity terms-of-trade growth.

• Unrest episodes motivated by socio-economic reasons result in sharper GDP contractions compared
to those associated mainly with politics/elections; episodes triggered by a combination of socio-
economic and political issues see sharpest contractions

• Results are robust to controlling for previous adverse growth events

• Results based on monthly data somewhat weaker given the relatively high noise at such frequency

INTERNATIONAL MONETARY FUND 38