Good Intentions Gone Bad? The Dodd-Frank Act and Conflict in Africas - - PowerPoint PPT Presentation

Introduction Empirical Framework Results Discussion Appendix

Good Intentions Gone Bad?

The Dodd-Frank Act and Conflict in Africa’s Great Lakes Region Jeffrey R. Bloem

Ph.D. Candidate Department of Applied Economics University of Minnesota

March 17, 2019

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Minerals and Conflict

◮ ‘Conflict minerals’ find their way into a host of popular consumer products

◮ Cell phones, laptops, jewelry, eyeglasses, cars, airplanes, and medical equipment

◮ Revenues from the extraction of these minerals fuel conflict across Africa

◮ See Berman et al. (2017)

◮ Conflicts are often deadly

◮ Estimates vary between 2 and 6 million people killed due to violent conflict over

the last two decades in the region.

◮ Violent conflict reverses economic development and efforts to alleviate poverty

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Section 1502 of the Dodd-Frank Act

◮ In 2010, US lawmakers passed legislation with the intentions of reducing

conflict in the DRC and surrounding countries

◮ Regulates reporting on supply chain links of tin, tantalum, tungsten, and gold

(3TG) to armed groups

◮ Any company registered with the US SEC must perform due diligence — and file

a report (“Form SD”)

◮ The legislation was—and remains—controversial

◮ US companies claim compliance costs impose an undue burden ◮ Other critics claim the policy is build on faulty assumptions about the causes of

conflict

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix Policy Implementation

(Incomplete) Theory of Change

◮ Theory of change rests on the strength of the link minerals and conflict

◮ Key assumption: Reducing the revenue earned by armed groups from minerals

will reduce conflict

◮ In theory, this tightens the budget constraint of armed groups (e.g. Fearon 2004;

Collier et al. 2009; Dube and Naidu 2015)

◮ This reduces the “feasibility” to perpetuate conflict

◮ In practice, it is not clear this mechanism dominates

◮ For example, consider the “opportunity cost” mechanism (e.g. Becker 1963;

Collier and Hoeffler 1998; Grossman 1991; Dube and Vargas 2013)

◮ A reduction in mineral extraction decreases incomes and the opportunity cost of

joining a rebel group

◮ This could increase conflict Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix Policy Implementation

Background

◮ The Dodd-Frank Act was officially passed by the US Congress in July 2010

◮ Direct consequence: In Sept. 2010 the DRC shut down its entire mineral export

industry (re-opened in 2011)

◮ Real effects: In some areas exports of tin dropped by 90 percent (Seay 2012)

◮ In August 2012 the “final rules” of the legislation are agreed upon by the US

SEC

◮ In July 2013 a lawsuit is in place arguing that the regulation violates US

constitutional rights

◮ Companies required to file first “due diligence” reports in May 2014

◮ In April 2015 US appeals court decides companies must still file annual reports

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix Policy Implementation

Background (continued)

◮ In April 2017, the US SEC suspended enforcement of the legislation

◮ The Financial CHOICE Act of 2017 would have officially abolished the

regulations

◮ Ultimately, dismissed by the US Senate

◮ Many companies still complying with the rules

◮ The law can be enforced again quite quickly ◮ Some companies — such as Apple and Intel — have publicly stated they intend

to follow the rules even if they are abolished

◮ Responding to a “market expectation” for “conflict free” minerals Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Existing Research

◮ Qualitative studies on the effects of the Dodd-Frank Act on livelihoods in the

DRC

◮ See Greenen (2012); Cuvelier et al. (2014); Radley and Vogel (2015); Vogel and

Raeymaekers (2016)

◮ Struggle to quantify the causal relationship

◮ Quantitative studies compare outcomes in geographic areas within the DRC

◮ See Parker et al. (2016); Parker and Vadheim (2017); Stoop et al. (2018) ◮ Important methodological improvement, but still may suffer from endogeneity

issues

◮ The presence of spillovers between geographical regions — a potential SUTVA

violation

◮ Spillovers are relevant in this context (Maystadt et al. 2014) Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Empirical Method

◮ Compare the prevalence of conflict:

◮ Over time (monthly) at the second sub-national administrative level ◮ Across countries covered by the Dodd-Frank Act and other sub-Saharan African

countries

◮ Use a difference-in-differences estimation strategy

◮ Benefits of this approach:

◮ Avoids concerns with spillovers present in within-DRC analysis ◮ Allows impact estimation on the full list of covered countries ◮ Extends core results through 2016 Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Research Questions

◮ Did the Dodd-Frank Act Increase or Decrease Conflict?

◮ In the DRC? ◮ In all covered countries (DRC + surrounding countries)?

◮ What are the underlying mechanisms of the effects?

◮ Specifically: “feasibility” vs. “opportunity cost”

◮ What is the effect of the recent enforcement suspension by the US SEC?

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Data

◮ Armed Conflict Location and Event Data (ACLED) project

◮ Subset includes data from 39 sub-Saharan African countries from 2004 through

2016

◮ Enforcement suspension analysis: May 2014 - October 2018 ◮ Construct a monthly panel dataset: 156 time periods and 3,764 administrative

regions

◮ Outcome variables:

◮ (A) All conflict ◮ (B) Violence against civilians ◮ (C) Rebel group battles ◮ (D) Riots and protests ◮ (E) Deadly conflict Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Conflict Trends by Type

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Estimation Specification (1)

◮ Linear regression model: yrct = αrc + γt + β · 1{rc = DRC} · 1{t ≥ July 2010} + ǫrct (1)

◮ yrct type of conflict in administrative area r in country c in month t ◮ αrc and γt are geographic and month fixed effects ◮ β is the coefficient of interest and is the DID estimate of the effect of the

Dodd-Frank Act

◮ ǫrct is an error term

◮ Implement a variant of Fisher’s permutation test (Fisher 1935) for robustness

check on inference

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Estimation Specification (2)

◮ Linear regression model: yrct = ηrc + λt + δt · 1{rc = DRC} · 1{t = 2005, 2006, 2007, ..., 2016} + ξrct (2)

◮ yrct type of conflict in administrative area r in country c in month t ◮ ηrc and λt are geographic and month fixed effects ◮ δt is a vector of coefficients and is the year-specific DID estimate of the effect of

the Dodd-Frank Act

◮ ξrct is an error term

◮ Tests the assumption that conflict would not have evolved differently in the

absence of the Dodd-Frank Act

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix Core Results

Effect of the Dodd-Frank Act on Conflict

Conflict, All Violence Against Rebel Group Riots and Protests Deadly Conflict Types Civilians Battles (1) (2) (3) (4) (5) Panel A: DRC Only Effect of Dodd-Frank 0.143*** 0.076*** 0.063*** 0.113*** 0.068*** (0.007) (0.004) (0.002) (0.005) (0.005) Observations 433,992 433,992 433,992 433,992 433,992 Baseline DRC mean 0.140 0.084 0.082 0.050 0.072 Geographic and time FEs Yes Yes Yes Yes Yes R-squared 0.141 0.097 0.084 0.125 0.074 Panel B: All Covered Countries Effect of Dodd-Frank 0.001 0.008

• 0.001

0.003

• 0.004

(0.016) (0.010) (0.007) (0.012) (0.010) Observations 574,236 574,236 574,236 574,236 574,236 Baseline Covered mean 0.030 0.015 0.013 0.010 0.015 Geographic and time FEs Yes Yes Yes Yes Yes R-squared 0.129 0.087 0.076 0.116 0.067 Notes: The dependent variable is a binary variable indicating the existence of a conflict event at the second sub-national administrative area within a given month. Standard errors clustered at the country level are in parentheses. Bonferroni adjusted p-values are noted as follows *** p<0.01, ** p<0.05, * p<0.1.

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix Core Results

Placebo Estimates from Permutation Tests

Effect of Dodd-Frank on Conflict

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix Core Results

Year-Specific Effects, DRC Only

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix Supplemental Analysis A

Mechanisms within the DRC

◮ Test for mechanisms: “opportunity cost” vs. “feasibility”

◮ Use data from the International Peace Information Service (IPIS) – from 2009

through 2015

◮ Use information from mine visits on: (i) number of workers and (ii) the presence

• f an armed group

◮ Identification strategy

◮ Compare outcomes over time between 3T mining sites and all other types of

mining site, including gold

◮ Follows strategy used by Parker and Vadheim (2017) and Stoop et al. (2018) Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix Supplemental Analysis A

“Opportunity Cost” > “Feasibility”

◮ Find evidence of roughly a 50% reduction in the number of workers at 3T

mineral mines

◮ Evidence of a strong effect of the “opportunity cost” mechanism

◮ Fine weak evidence of some reduction in the presence of an armed rebel group

◮ Evidence of a weak effect of the “feasibility” mechanism Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix Supplemental Analysis B

Did Enforcement Suspension Help?

◮ Repeat core estimation strategy

◮ From May 2014 (beginning of enforcement) through October 2018 ◮ Estimate the effect of enforcement suspension in April 2017

◮ Do not find robust effects both in the DRC and in all covered countries

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Conclusion

◮ Find evidence of unintended consequences of the conflict minerals legislation

in the DRC

◮ May be more dramatic than previously reported ◮ Roughly a 100 percent increase in the probability of conflict ◮ Compare to between 30 to 50 percent increase of various types of conflict (Stoop

et al. 2018)

◮ Precise null effect on the probability of conflict within all covered countries

pooled together

◮ Hides important heterogeneity

◮ Potential mechanisms (within the DRC)

◮ The “opportunity cost” mechanism may outweigh the “feasibility” mechanism

◮ The recent suspension of enforcement by the US SEC

◮ Find little effect both in the DRC and in all covered countries pooled together Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Mechanisms Results

(1) (2) (3) (4) (5) (6) IHS IHS IHS Binary Binary Binary number number number presence of presence of presence of workers workers workers armed group armed group armed group Truncated IPIS Data (2009 - 2015) Effect of Dodd-Frank

• 0.978***
• 1.001***
• 0.765**
• 0.308*
• 0.287*
• 0.268*

(0.264) (0.285) (0.255) (0.130) (0.136) (0.119) 3T Mineral Mine 0.760** 0.742** 0.557*

• 0.130
• 0.131

0.0785 (0.252) (0.261) (0.254) (0.113) (0.117) (0.0959) Post July 2010 0.209 0.119

• 0.154

0.161 0.0895 0.150 (0.245) (0.326) (0.404) (0.0917) (0.120) (0.102) Observations 2,371 2,371 2,371 2,621 2,621 2,621 Baseline 3T mean

• 0.327

0.327 0.327 R-squared 0.021 0.053 0.185 0.149 0.166 0.426 Year FEs No Yes Yes No Yes Yes Territory FEs No No Yes No No Yes Notes: Standard errors clustered at the territory level are shown in parentheses. Bonferroni adjusted p-values are noted as follows *** p<0.01, ** p<0.05, * p<0.1.

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Supported by Anecdotal Evidence

“When his father could no longer make enough money from the tin mine, when he could no longer pay for school, Bienfait Kabesha ran off and joined a militia. It

• ffered the promise of loot and food, and soon he was firing an old rifle on the front

lines of Africa’s deadliest conflict. He was 14.”

• Raghavan, S. (2014) The Washington Post

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Effect of Enforcement Suspension on Conflict

Conflict, All Violence Against Rebel Group Riots and Protests Deadly Conflict Types Civilians Battles (1) (2) (3) (4) (5) Panel A: DRC Only Effect of Enforcement Suspension 0.007 0.027*** 0.010***

• 0.012

0.014*** (0.007) (0.004) (0.003) (0.005) (0.003) Observations 147,976 147,976 147,976 147,976 147,976 Basline DRC mean 0.357 0.179 0.156 0.247 0.184 Geographic and time FEs Yes Yes Yes Yes Yes R-squared 0.181 0.116 0.135 0.164 0.131 Panel B: All Covered Countries Effect of Enforcement Suspension

• 0.002

0.005

• 0.006
• 0.014
• 0.006

(0.0111) (0.010) (0.005) (0.008) (0.004) Observations 195,676 195,676 195,676 195,676 195,676 Basline Covered mean 0.092 0.052 0.022 0.051 0.037 Geographic and time FEs Yes Yes Yes Yes Yes R-squared 0.177 0.129 0.125 0.153 0.122 Notes: The dependent variable is a binary variable indicating the existence of a conflict event at the second sub-national administrative area within a given month. Standard errors clustered at the country level are in parentheses. Bonferroni adjusted p-values are noted as follows *** p<0.01, ** p<0.05, * p<0.1.

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Placebo Estimates from Permutation Tests

Effect of Enforcement Suspension on Conflict

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Year-Specific Effects, All Covered Countries

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Country-Specific Effects

Conflict, All Violence Against Rebel Group Riots and Protests Deadly Conflict Types Civilians Battles (1) (2) (3) (4) (5) Panel A: Democratic Republic of Congo Effect of Dodd-Frank 0.143*** 0.0756*** 0.0627*** 0.113*** 0.068*** (0.023) (0.016) (0.016) (0.021) (0.018) Observations 432,432 432,432 432,432 432,432 432,432 R-squared 0.141 0.098 0.084 0.125 0.074 Panel B: Angola Effect of Dodd-Frank

• 0.0308***
• 0.0108***
• 0.00535***
• 0.0229***
• 0.0141***

(0.00307) (0.00143) (0.000759) (0.00247) (0.00110) Observations 450,060 450,060 450,060 450,060 450,060 R-squared 0.115 0.071 0.042 0.111 0.047 Notes: The dependent variable is a binary variable indicating the existence of a conflict event at the 2nd subnational admin- istrative area within a given month. Standard errors clustered by the 2nd subnational administrative area in parentheses *** p<0.01, ** p<0.05, * p<0.1.

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Country-Specific Effects (continued)

Conflict, All Violence Against Rebel Group Riots and Protests Deadly Conflict Types Civilians Battles (1) (2) (3) (4) (5) Panel C: Burundi Effect of Dodd-Frank 0.0339*** 0.0325*** 0.000903 0.0363*** 0.00503 (0.00659) (0.00487) (0.00262) (0.00624) (0.00342) Observations 448,812 448,812 448,812 448,812 448,812 R-squared 0.112 0.069 0.040 0.109 0.046 Panel D: Central African Republic Effect of Dodd-Frank 0.0715*** 0.0601*** 0.0297*** 0.0223** 0.0544*** (0.0143) (0.0107) (0.00894) (0.00958) (0.0109) Observations 436,020 436,020 436,020 436,020 436,020 R-squared 0.116 0.074 0.045 0.112 0.051 Notes: The dependent variable is a binary variable indicating the existence of a conflict event at the 2nd subnational admin- istrative area within a given month. Standard errors clustered by the 2nd subnational administrative area in parentheses *** p<0.01, ** p<0.05, * p<0.1.

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Country-Specific Effects (continued)

Conflict, All Violence Against Rebel Group Riots and Protests Deadly Conflict Types Civilians Battles (1) (2) (3) (4) (5) Panel E: Republic of Congo Effect of Dodd-Frank

• 0.0272***
• 0.0112***
• 0.00459***
• 0.0178***
• 0.0133***

(0.00504) (0.00165) (0.00121) (0.00585) (0.00132) Observations 432,276 432,276 432,276 432,276 432,276 R-squared 0.115 0.071 0.042 0.112 0.047 Panel F: Rwanda Effect of Dodd-Frank

• 0.00351

0.00452

• 0.00395**
• 0.0119**
• 0.0156***

(0.0144) (0.0118) (0.00179) (0.00513) (0.00417) Observations 429,468 429,468 429,468 429,468 429,468 R-squared 0.113 0.071 0.041 0.111 0.047 Notes: The dependent variable is a binary variable indicating the existence of a conflict event at the 2nd subnational admin- istrative area within a given month. Standard errors clustered by the 2nd subnational administrative area in parentheses *** p<0.01, ** p<0.05, * p<0.1.

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Country-Specific Effects (continued)

Conflict, All Violence Against Rebel Group Riots and Protests Deadly Conflict Types Civilians Battles (1) (2) (3) (4) (5) Panel G: Tanzania Effect of Dodd-Frank

• 0.0216***
• 0.00762***
• 0.00362***
• 0.0182***
• 0.0101***

(0.00268) (0.00132) (0.000880) (0.00219) (0.00132) Observations 453,336 453,336 453,336 453,336 453,336 R-squared 0.113 0.070 0.041 0.110 0.046 Panel H: Uganda Effect of Dodd-Frank

• 0.0353***
• 0.0163***
• 0.0275***
• 0.00668
• 0.0342***

(0.00722) (0.00350) (0.00478) (0.00406) (0.00456) Observations 450,996 450,996 450,996 450,996 450,996 R-squared 0.114 0.071 0.045 0.114 0.049 Notes: The dependent variable is a binary variable indicating the existence of a conflict event at the 2nd subnational admin- istrative area within a given month. Standard errors clustered by the 2nd subnational administrative area in parentheses *** p<0.01, ** p<0.05, * p<0.1.

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Country-Specific Effects (continued)

Conflict, All Violence Against Rebel Group Riots and Protests Deadly Conflict Types Civilians Battles (1) (2) (3) (4) (5) Panel I: Zambia Effect of Dodd-Frank

• 0.00539

0.00332

• 0.00325**
• 0.00621
• 0.0113***

(0.00553) (0.00322) (0.00152) (0.00514) (0.00141) Observations 436,332 436,332 436,332 436,332 436,332 R-squared 0.112 0.070 0.041 0.109 0.047 Placebo tests (other countries) 5th percentile

• 0.042
• 0.029
• 0.010
• 0.029
• 0.020

95th percentile 0.079 0.026 0.015 0.041 0.051 Geographic and time FEs Yes Yes Yes Yes Yes Notes: The dependent variable is a binary variable indicating the existence of a conflict event at the 2nd subnational admin- istrative area within a given month. Standard errors clustered by the 2nd subnational administrative area in parentheses *** p<0.01, ** p<0.05, * p<0.1.

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Synthetic Control Estimation

All Conflict

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Synthetic Control Estimation

Violence Against Civilians

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Synthetic Control Estimation

Rebel Group Battles

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Synthetic Control Estimation

Riots and Protests

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Synthetic Control Estimation

Deadly Conflict

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Year-Specific Effects, All Covered Countries

Effect of Enforcement Suspension on Conflict

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Year-Specific Effects, All Covered Countries

Effect of Enforcement Suspension on Conflict

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?

Introduction Empirical Framework Results Discussion Appendix

Alternative Dependent Variable Definitions

Conflict, All Violence Against Rebel Group Riots and Protests Deadly Conflict Types Civilians Battles (1) (2) (3) (4) (5) Panel A: DV = 1 if > 5 Conflict Events Effect of Dodd-Frank 0.039*** 0.017*** 0.009*** 0.013*** 0.019*** (0.002) (0.000) (0.000) (0.001) (0.003) Observations 433,992 433,992 433,992 433,992 433,992 Baseline DRC mean 0.030 0.009 0.015 0.001 0.049 Geographic and time FEs Yes Yes Yes Yes Yes R-squared 0.132 0.117 0.070 0.067 0.059 Panel B: DV = 1 if > 10 Conflict Events Effect of Dodd-Frank 0.019*** 0.005*** 0.003*** 0.002*** 0.013*** (0.001) (0.000) (0.000) (0.001) (0.002) Observations 433,992 433,992 433,992 433,992 433,992 Baseline DRC mean 0.014 0.003 0.007 0.000 0.035 Geographic and time FEs Yes Yes Yes Yes Yes R-squared 0.085 0.048 0.047 0.040 0.049 Notes: The dependent variable is a binary variable indicating the existence of either more than 5 or ten conflict events at the second sub-national administrative area within a given month. Standard errors clustered at the country level are in

• parentheses. Bonferroni adjusted p-values are noted as follows *** p<0.01, ** p<0.05, * p<0.1.

Jeffrey R. Bloem University of Minnesota Good Intentions Gone Bad?