Coercion Conflict Commodities Jacobus Cilliers University of Oxford - - PowerPoint PPT Presentation

Introduction Baseline Model Full Model Conclusion

Coercion Conflict Commodities

Jacobus Cilliers

University of Oxford

March 17, 2013

Introduction Baseline Model Full Model Conclusion

Introduction

When and why do armed groups coerce?

Introduction Baseline Model Full Model Conclusion

Introduction

When and why do armed groups coerce? Coercion is common. "more died from starvation whilst working on the rice farms than were killed by combatants"

Introduction Baseline Model Full Model Conclusion

Introduction

When and why do armed groups coerce? Coercion is common. "more died from starvation whilst working on the rice farms than were killed by combatants" Strong focus of international community is trade sanctions of "blood minerals" in resource conflicts

Introduction Baseline Model Full Model Conclusion

Introduction

When and why do armed groups coerce? Coercion is common. "more died from starvation whilst working on the rice farms than were killed by combatants" Strong focus of international community is trade sanctions of "blood minerals" in resource conflicts However, need economic theory to predict what impact of change in economic incentives would be on civilian conditions

Introduction Baseline Model Full Model Conclusion

Preview

Model: Two sector trade model, with "armed group" in resource sector

Introduction Baseline Model Full Model Conclusion

Preview

Model: Two sector trade model, with "armed group" in resource sector Results Coercion depends on the (i) type of resource, (ii) factor endowments, and (iii) distribution of military strength. In particular:

Introduction Baseline Model Full Model Conclusion

Preview

Model: Two sector trade model, with "armed group" in resource sector Results Coercion depends on the (i) type of resource, (ii) factor endowments, and (iii) distribution of military strength. In particular:

1

Coercion increases with the price of the resource if there is

• nly one armed group, but decreases if there are many groups.

Introduction Baseline Model Full Model Conclusion

Preview

Model: Two sector trade model, with "armed group" in resource sector Results Coercion depends on the (i) type of resource, (ii) factor endowments, and (iii) distribution of military strength. In particular:

1

Coercion increases with the price of the resource if there is

• nly one armed group, but decreases if there are many groups.

2

Coercion decreases if military strength concentrated with one armed group.

Introduction Baseline Model Full Model Conclusion

Preview

Model: Two sector trade model, with "armed group" in resource sector Results Coercion depends on the (i) type of resource, (ii) factor endowments, and (iii) distribution of military strength. In particular:

1

Coercion increases with the price of the resource if there is

• nly one armed group, but decreases if there are many groups.

2

Coercion decreases if military strength concentrated with one armed group.

Introduction Baseline Model Full Model Conclusion

Preview

Model: Two sector trade model, with "armed group" in resource sector Results Coercion depends on the (i) type of resource, (ii) factor endowments, and (iii) distribution of military strength. In particular:

1

Coercion increases with the price of the resource if there is

• nly one armed group, but decreases if there are many groups.

2

Coercion decreases if military strength concentrated with one armed group. Implication: impact of trade policy depends on both the type of commodity and nature of conflict on the ground.

Introduction Baseline Model Full Model Conclusion

Conflict and Coercion

Ownership over land and labour

Figure: Armed groups fight each other for control over resources and coerce cilivians in the extraction of resources

Introduction Baseline Model Full Model Conclusion

Conflict and Coercion

"with minor exceptions, the objective of military activity is either to secure access to mining sites or ensure a supply of captive labour" UN Panel of Expert Report on ongoing conflict in Kivu provinces, eastern DRC in 2002.

Introduction Baseline Model Full Model Conclusion

Conflict and Coercion

No Monopoly over Violence

Figure: With no monopoly over violence, armed groups can coerce civilians, but are under constant threat of attack from other groups

Introduction Baseline Model Full Model Conclusion

Conflict and Coercion

Trade-off

"since the... informal cease-fire in January 2004, there has been comparatively little fighting...[the armed groups] rather have focused their energies more on

• ppressing the civilian population"

Introduction Baseline Model Full Model Conclusion

Monopoly over violence

Model

Armed groups choose level of employment, LR, and coercion, c, in resource sector: max

c,LR

pRG(TR, LR) − w(LR − LC (c)) − 1 2ψc2 Civilians choose which sector to work in max

LY

pY F(TY , LY ) + w(L − LC (c) − LY )

Introduction Baseline Model Full Model Conclusion

Monopoly over violence

Equilibrium

Figure: Wage rate endogenously determined by the level of employment, L∗

R, where the marginal value products of labour are equalised. Armed

groups coerce until it is cheaper to employ.

Introduction Baseline Model Full Model Conclusion

Monopoly over violence

Results

Coercion increases with w, which increases when: Labour is scarce

e.g. Domar (1970), Fenske (2012), Austin(2005)

Introduction Baseline Model Full Model Conclusion

Monopoly over violence

Results

Coercion increases with w, which increases when: Labour is scarce

e.g. Domar (1970), Fenske (2012), Austin(2005)

Price of either resource of yeoman good increases.

E.g. Coercion increased with price of rubber in the Congo Free State

Introduction Baseline Model Full Model Conclusion

Monopoly over violence

Results

Coercion increases with w, which increases when: Labour is scarce

e.g. Domar (1970), Fenske (2012), Austin(2005)

Price of either resource of yeoman good increases.

E.g. Coercion increased with price of rubber in the Congo Free State

Production is labour intensive

Introduction Baseline Model Full Model Conclusion

Many Armed Groups

Many armed groups

n armed groups, decide between conflict and coercion max

Li,ci,fi pRG(π (fi, f−i) · TR, Li R) − w(Li R − LC (ci))

Contest function determines share of land that each armed group receives: π (fi, f−i) =

• fi

∑ fj

Introduction Baseline Model Full Model Conclusion

Many armed groups

Equilibrium

Cost of coercion is opportunity cost of not fighting. w − κ (c∗

i ) · r = 0

(1) In equilibrium C ∗ = s − n − 1 n r w

• TR

(2) Coercion depends both on ratio of factor prices and number of armed groups.

Introduction Baseline Model Full Model Conclusion

Many armed groups

Results

Coercion, surprisingly, now decreases with the resource price if there are many armed groups. ↑ pR = ⇒ ↑ L∗

R

= ⇒ ↑ MRPT and ↓ MRPL = ⇒ ↑ r w

• =

⇒ ↓ C ∗

Introduction Baseline Model Full Model Conclusion

Many armed groups

Results

Higher levels of coercion, if fewer armed groups, or military power centralised in one armed group. Intuition: oligopolist producers of violence: Groups would prefer that no-one fights, since can allocate all resources to coercion.

Introduction Baseline Model Full Model Conclusion

Many armed groups

Results

Higher levels of coercion, if fewer armed groups, or military power centralised in one armed group. Intuition: oligopolist producers of violence: Groups would prefer that no-one fights, since can allocate all resources to coercion. If one group fights, then fighting is less effective for other groups: need to fight more to gain the same share of land.

Introduction Baseline Model Full Model Conclusion

Many armed groups

Results

Higher levels of coercion, if fewer armed groups, or military power centralised in one armed group. Intuition: oligopolist producers of violence: Groups would prefer that no-one fights, since can allocate all resources to coercion. If one group fights, then fighting is less effective for other groups: need to fight more to gain the same share of land. Fighting of one group thus creates "negative externality" of fighting to other groups.

Introduction Baseline Model Full Model Conclusion

Many armed groups

Results

Higher levels of coercion, if fewer armed groups, or military power centralised in one armed group. Intuition: oligopolist producers of violence: Groups would prefer that no-one fights, since can allocate all resources to coercion. If one group fights, then fighting is less effective for other groups: need to fight more to gain the same share of land. Fighting of one group thus creates "negative externality" of fighting to other groups. Larger armed group has more "market power" in use of violence: larger impact on total level of fighting, therefore fights less.

Introduction Baseline Model Full Model Conclusion 1

Sierra Leone: no accounts of coercion in diamond areas more conflict diamond-rich areas, but lower levels of civilian victimisation (Bellows and Miguel, 2006) Diamonds have high price, therefore land was valuable factor of production, not labour. Military resources thus allocated to conflict, not coercion.

Introduction Baseline Model Full Model Conclusion 1

Sierra Leone: no accounts of coercion in diamond areas more conflict diamond-rich areas, but lower levels of civilian victimisation (Bellows and Miguel, 2006) Diamonds have high price, therefore land was valuable factor of production, not labour. Military resources thus allocated to conflict, not coercion.

2

Kivu Provinces: all coercion in 2002 committed by Rwandan Army, by far the strongest group. When the Rwandan Army departed, it left a vacuum of power, leading to a proliferation of smaller groups Many small groups lead "coordination problem". Each group fights more in anticipation that the other will do the same. No group strong enough to decrease fighting. As a result, less military resources available for coercion.

Introduction Baseline Model Full Model Conclusion

Summary

1

Violence aimed at ownership over factors of production.

2

Coercion also depends on strategic interaction in fighting between armed groups

3

Impact of change in resource price on level of coercion ambiguous and depends on nature of conflict on the ground.

Introduction Baseline Model Full Model Conclusion

Implications

1

Impact of trade policy depends on number of armed groups, the resource that armed groups operate and factor endowments.

2

Trade sanctions could have negative unintended consequences.

3

Cease-fire agreement interpreted as collusion.

The Welfare Cost of Lawlessness: Evidence from Somali Piracy

Tim Besley, Thiemo Fetzer & Hannes-Felix Mueller

Presentation at CSAE Conference 2013

March 16, 2013

Thiemo Fetzer (LSE) The Welfare Cost of Lawlessness March 16, 2013 1 / 1

Motivation

What are the welfare costs of predation? Bringing goods to market is subject to predation/ theft Piracy is a specific form of such predation. Little is known about its welfare costs, but it became salient because of upsurge in piracy attacks off the coast of Somalia Piracy became endemic as the Somalian coast provides safe havens for pirates. Hence, piracy is a symptom of the inability to establish law and order. Broader debates Establishing rule of law and order, functioning institutions with a monopoly of power is is important for development (Acemoglu et al, 2001) Taxation is a cheap from of extraction: high tax states tend to be rich, high predation states are poor (Besley and Persson, 2010)

Thiemo Fetzer (LSE) The Welfare Cost of Lawlessness March 16, 2013 2 / 1

Preview of Results

Estimate the Somalian piracy increases cost of shipping by around 8-13% The same net revenues that piracy generates, could be obtained with a tax of just 0.8%. Overall estimate suggest that cost lie in a range between 0.9 - 3.3 billion USD. Hence, taxation is a lot cheaper from of making transfers (see e.g. Olken and Barron, 2010).

Thiemo Fetzer (LSE) The Welfare Cost of Lawlessness March 16, 2013 3 / 1

Piracy off the Somalian Coast

Figure 1: Attacks and Treatment Areas

Somalia Gulf of Aden

Thiemo Fetzer (LSE) The Welfare Cost of Lawlessness March 16, 2013 4 / 1

Time-Series of Attacks

10 20 30 40 num ber of a ttacks

2003m 1 2 004m 7 2006m1 2007m7 2009m1 2010m7

Figure 2: Time Series of Attacks in Somalia

Thiemo Fetzer (LSE) The Welfare Cost of Lawlessness March 16, 2013 5 / 1

Visualization of Maritime Traffic

Figure 3: Calculated Shipping Lanes

Number of Observations 7884 2909 1073 396 146 54 20 7 3

Thiemo Fetzer (LSE) The Welfare Cost of Lawlessness March 16, 2013 6 / 1

Chartering Who pays the costs?

War risk insurance is made directly dependant on the area of sea that a ship crosses. May 2008: Joint War Committee London declared Gulf of Aden as high risk area, expanded over time covering the whole area we refer to as Somalia ”Charterers shall indemnify Owners against all liabilities costs and expenses arising out of actual or threatened acts of piracy or any preventive or other measures taken by Owners [...], including but not limited to additional insurance premiums, additional crew costs and costs of security personnel or equipment.” Intertanko Model Chartering Contract Clause

Thiemo Fetzer (LSE) The Welfare Cost of Lawlessness March 16, 2013 7 / 1

Empirical Strategy

Obtain data on 24,000 ship chartering contracts in the dry bulk sector (1/3 of global tonnage).

Who pays?

Obtain piracy incidence data that is geo-coded Match two datasets based on whether the shortest maritime route would pass through piracy areas. Difference-in-Difference type estimation zisdt = log C(s, d, t, Adt, xidt) = αs + αd + αt + γAdt + βxidt + ǫisdt (1) αs ship type FE, αd dyad FE, αt time FE Adt = δdr × E[art+1] δdr = 1 if route d crosses piracy area. Note this is an intention to treat. Exploit arguably exogenous variation in Adt due to weather induced seasonality.

Seasonality Thiemo Fetzer (LSE) The Welfare Cost of Lawlessness March 16, 2013 8 / 1

Main Results

(1) (2) (3) (4) VARIABLES daily charter rate per DWT daily charter rate per DWT daily charter rate per DWT daily charter rate per DWT number of attacks (Somalia) 0.00572*** 0.00589*** 0.00259*** (0.00128) (0.00130) (0.000898) attacks * handysize (Somalia) 0.00336 (0.00283) attacks * handymax (Somalia) 0.00665*** (0.00199) attacks * panamax (Somalia) 0.00597*** (0.00137) attacks * small capesize (Somalia) 0.00637*** (0.00189) attacks * capesize (Somalia) 0.00206 (0.00222) ballast bonus per DWT

• 8.04e-06

(5.13e-06) ship age

• 0.00614***

(0.000791) handysize 0.623*** 0.637*** 0.607*** 0.621*** (0.0224) (0.0203) (0.0369) (0.0255) handymax 0.401*** 0.403*** 0.354*** 0.394*** (0.0218) (0.0203) (0.0311) (0.0258) panamax 0.149*** 0.150*** 0.151*** 0.153*** (0.0142) (0.00912) (0.0221) (0.0154) capesize

• 0.0385
• 0.0513*
• 0.0899
• 0.0245

(0.0398) (0.0305) (0.0885) (0.0486) route fixed effect yes yes yes yes month fixed effect yes yes yes yes Observations 24,363 24,332 10,058 24,363 R-squared 0.873 0.877 0.861 0.874

Notes: Standard errors in parentheses. Standard errors are clustered at the route level. *** p<0.01, ** p<0.05, * p<0.1. "DWT" is deadweight tonnage. "Daily charter rate per DWT" is the log of the time charter rate per day per deadweight

• tonnage. All attack variables are interactions between a dummy that indicates whether a ship will cross the pirate territory

and the number of attacks. "Handysize" is a dummy that indicates ships with DWT<35000. "Handymax" are ships with 35000<DWT<55000. "Panamax" are ships with 55000<DWT<80000. "Small capesize" are ships with 80000<DWT<150000 (omitted). "Capesize" are ships with DWT>150000. "Ballast bonus" is a payment that compensates the ship owner for travelling without cargo on return. Column (3) only uses data after the surge in piracy in the Somalia region May 2008. Column (4) controls for interactions between ship categories and a Somalia dummy.

Thiemo Fetzer (LSE) The Welfare Cost of Lawlessness March 16, 2013 9 / 1

Observations

Both the surge in piracy and the seasonal variation in intensity are used to identify the impact on chartering rates. Point estimate suggests that an increase in 1 attack per month increases the chartering rate per DWT by 0.572 %-age points. So increase in piracy by 14.3 attacks (before/ after May 2008) implies cost increase of 8.2% age points. But strong seasonal variation. Need to check robustness of results Robustness:

1 Robustness to treatment assignment 2 Robustness to model of expectations 3 Is there a quantity reaction? Lehman 4 Substitution to different types of ships Thiemo Fetzer (LSE) The Welfare Cost of Lawlessness March 16, 2013 10 / 1

Robustness

(1) (2) (3) (4) (5) (6) (7) (8) VARIABLES daily charter rate per DWT daily charter rate per DWT daily charter rate per DWT daily charter rate per DWT daily charter rate per DWT daily charter rate per DWT daily charter rate per DWT daily charter rate per DWT forecast number of attacks (Somalia) 0.00785*** 0.0102*** (0.00175) (0.00274) number of attacks (Somalia) 0.00590*** 0.00289*** 0.00673*** 0.00580*** (0.00129) (0.000896) (0.00160) (0.00151) annual GDP growth start region 0.02368* (0.01159) annual GDP growth end region

• 0.0047

(0.01375) number of attacks (Gulf of Aden) 0.0100*** (0.00244) Somalia number of attacks (route through Gulf of Aden) 0.00660*** (0.00146) Somalia number of attacks (route not through Gulf of Aden) 0.00226** (0.000894) ship size controls yes yes yes yes yes yes yes yes route fixed effect yes yes yes yes yes yes yes yes month fixed effect yes yes yes no yes yes yes yes region-specific month fixed effects no no no yes no no no no Observations 24,363 24,363 24,332 24,363 24,363 24,363 24,363 24,363 R-squared 0.873 0.873 0.877 0.900 0.873 0.874 0.873 0.873

Notes: Standard errors in parentheses. Standard errors are clustered at the route level. *** p<0.01, ** p<0.05, * p<0.1. "DWT" is deadweight tonnage. "Daily charter rate per DWT" is the ln of the time charter rate per day per deadweight tonnage. All piracy variables are interactions between a dummy that indicates whether a ship will cross the pirate territory and the number of attacks. "Forecast number of attacks" is the forecasted number of attacks next month calculated using an AR(2) model in column (1) and a Markov chain model for column (2). Column (4) controls for a seperate set of time fixed effects for each of 24 start regions. Column (5) takes a smaller treatment area and uses piracy estimates from attacks in this area. Column (6) uses the piracy estimates from our main specification and applies them to shipments though different areas. Columns (7) and (8) use the following treatment assignment: alternative routes not using the Suez canal were used if the alternative route was at most 10% and 20% longer than the Suez route, respectively.

Thiemo Fetzer (LSE) The Welfare Cost of Lawlessness March 16, 2013 11 / 1

Exploiting Seasonal Variation

Seasonality Means Figure 4: Shipping Cost Prediction of Pirate Activity and Wind Speed .1 .2 .3 .4 2007m1 2008m1 2009m1 2010m1 Estimated Cost Increase (in %) 90% confidence interval Pirates become active Monsoon season

Figure:

Thiemo Fetzer (LSE) The Welfare Cost of Lawlessness March 16, 2013 12 / 1

Extended Results

(1) (2) (3) (4) (5) VARIABLES

Suez Canal cargo traffic Suez Canal cargo traffic average ship size daily charter rate per DWT daily charter rate per DWT

number of attacks (Somalia)

• 359.3***
• 32.89
• 20.48

(87.17) (36.90) (56.68) cargo traffic break

• 20801***

(867.8) war risk area (Somalia) 0.121*** 0.360*** (0.0319) (0.0975) war risk area * windspeed (Somalia)

• 0.0373***

(0.0125) windspeed (Somalia) 0.00470 (0.00423) ship size controls

• yes

yes route fixed effect

• yes

yes yes month fixed effect * * yes yes yes Observations 108 108 12,753 24,363 24,363 R-squared 0.611 0.94 0.482 0.873 0.874

Notes: Standard errors in parentheses. Standard errors are clustered at the route level in columns (4) and (5). *** p<0.01, ** p<0.05, * p<0.1. Columns (1) and (2) use monlty time-series data from the Suez canal. Column (3) uses all data but aggregates by dyad. "Suez canal cargo traffic" is measured in DWT. "Average ship size" is the dyad average for that month in deadweight tons. "DWT" is deadweight tonnage. "Daily charter rate per DWT" is the ln of the time charter rate per day per deadweight tonnage. "Cargo traffic break" is a dummy variable that takes a value of 1 after the volume in trade through the Suez canal collapses in November 2008. All piracy variables are interactions between a dummy that indicates whether a ship will cross the pirate territory and the number of (expected) attacks. "War risk area" is a dummy that indicates whether the area was defined as a war risk area by the Joint War Comittee. "Windspeed" is the (predicted) monthly windspeed in the piracy area in the same month. (*) Columns (1) and (2) control for a linear time trend.

Lehman Seasonality Thiemo Fetzer (LSE) The Welfare Cost of Lawlessness March 16, 2013 13 / 1

Studying Suez Canal Traffic

20000 30000 40000 50000 60000 70000 10 20 30 40 50 2002m1 2004m1 2006m1 2008m1 2010m1 ... Confidence Band Cargo Confidence Band Piracy Somalian Piracy (attacks) Cargo through Suez (tons) Go back robustness. Go back extended results. Thiemo Fetzer (LSE) The Welfare Cost of Lawlessness March 16, 2013 14 / 1

Welfare Loss Estimates

Very simple framework with one composite traded good Try to compare the (distortionary) piracy tax to a general tax on shipping through the area Present three sets of estimates based on point estimates.

Table 6: The Welfare Cost of Piracy in 2010 Panel A: Gulf of Aden (1) (2) (3) L1 (in million USD) L2 (in million USD) L3 (in million USD) low estimate 638 649 1495 high estimate 999 1026 2264 Panel B: Somalia (1) (2) (3) L1 (in million USD) L2 (in million USD) L3 (in million USD) low estimate 935 952 2127 high estimate 1462 1503 3250

Calulcalations are discussed in section 4 and the appendix D. Column (2) adjusts the welfare loss by taking into account the change in trade. Column (3) adjusts the cost to take into account the share of costs borne by charterers. Panel B uses data on trade to and from the Middle East to calculate the costs for the area including the Indian Ocean.

Thiemo Fetzer (LSE) The Welfare Cost of Lawlessness March 16, 2013 15 / 1

Conclusion

Find a robust effect of piracy (law and order) on transport costs. An equivalent tax rate of 0.8% on all shipping raises the same amount of net revenues as are generated by pirates, contrasts to increases in shipping costs between 8% - 13%. Piracy is a costly way of making transfers. Outlook: piracy is an externality. It requires coordination of various

• stakeholders. This is not in sight. Historically, very similar situations.

The London and China Telegraph from 4th February 1867 noted that “Besides we are not the only Power with large interests at stake. French, Americans, and Germans carry on an extensive trade [...] Why should we then incur singly the expense of suppressing piracy if each provided a couple of gunboats the force would suffice for the safety foreign shipping which is all that devolves upon [..] why should the English tax payer alone bear the expense?”

Thiemo Fetzer (LSE) The Welfare Cost of Lawlessness March 16, 2013 16 / 1

Seasonality induced variation in Piracy

Go back.

Figure A2: Wind Speed and Attacks in the Somalia Area 4 5 6 7 8 9 windspeed (m/s) 10 20 30 40

January February March April May June July August September October NovemberDecember average attacks post May 2008 average windspeed (lagged on month)

Thiemo Fetzer (LSE) The Welfare Cost of Lawlessness March 16, 2013 17 / 1

Seasonal Variation

Table 1: Seasonality in Attacks in Somalia Region month before May 2008 after May 2008 January 1.5 12.5 Februar 2.7 7.5 March 2.9 31.5 April 5.2 34.1 May 3.7 21.0 June 1.8 10.5 July 3.5 4.6 August 2.1 9.4 September 1.3 14.6 October 3.8 18.7 November 2.2 28.0 December 2.4 12.6 average 2.8 17.1 difference (after-before) 14.3

Thiemo Fetzer (LSE) The Welfare Cost of Lawlessness March 16, 2013 18 / 1

Introduction Specification and Data Results Interpretation Conclusion

Climate, ecosystem resilience and the slave trade

James Fenske & Namrata Kala

University of Oxford and Yale University

1 / 23

Introduction Specification and Data Results Interpretation Conclusion Questions and motivation Overview

Introduction

Question

1

Did (environmental) supply factors affect the dynamics of the slave trade? Motivation The slave trade matters today: Parts of Africa that exported more slaves in the past are poorer (Nunn, 2008), less trusting (Nunn and Wantchekon, 2010), and more ethnically fractioned (Whatley and Gillezeau, 2010). Explaining the slave trade helps explain Africa’s poverty. History provides a window into the effects of environmental change: Abrupt and persistent climate change contributes to social collapse (e.g. Weiss, 1994; Haug et al., 2003). We find that even smaller, short-run anomalies caused large shifts in African economies’ coping mechanisms.

2 / 23

Introduction Specification and Data Results Interpretation Conclusion Questions and motivation Overview

What we do

Match gridded annual reconstructions of temperature anomalies with port-level annual data on slave exports (1730-1866). Show that

1

ports exported fewer slaves in warmer years (years of lower agricultural productivity (Tan and Shibasaki, 2003; Lobell and Field, 2007)).

2

the effect of temperature was greatest where agriculture is most sen- sitive to temperature.

3

both climate trends and shocks around them can explain slave exports.

4

historical temperature anomalies, particularly those at the peak of the slave trade, predict contemporary luminosity around ports.

Provide a simple model in which agricultural TFP shocks reduce slave exports: we argue that higher temperatures raised the costs of harvest- ing slaves, due to greater costs of taxation, disorder, higher mortality, and lower productivity in supporting sectors. Validate this model using case studies of three influential ports: Benguela, Whydah, and Mozambique.

3 / 23

Introduction Specification and Data Results Interpretation Conclusion Questions and motivation Overview

Contribution

Environment and development. Geography acts over the long run (Bleakley and Lin, 2012; Easterly and Levine, 2003; Ashraf and Michalopoulos, 2011). Un-

• bservables and mechanisms are challenges in this literature (Dell et al., 2009;

Bluedorn et al., 2011; Jia, 2011, Fenske 2012).

We provide a new mechanism connecting past shocks to present out- comes. We find large impacts of (relatively) small environmental shocks (Horn- beck 2012; Weiss and Bradley, 2001).

Economics of conflict. We show that the response of endemic, predatory violence to economic shocks differs from the dynamics of recent conflicts (e.g. Besley and Persson, 2011; Ciccone, 2011; Collier and Hoeffler, 2004; Miguel et al. 2004). Economics of the slave trade. The only supply shock shown to be significant is guns-for-slaves cycle (Whatley, 2008). We show that African shocks mattered, and in which direction.

4 / 23

Introduction Specification and Data Results Interpretation Conclusion Questions and motivation Overview

1

Introduction

2

Specification and Data

3

Results

4

Interpretation

5

Conclusion

5 / 23

Introduction Specification and Data Results Interpretation Conclusion Specification Data

1

Introduction

2

Specification and Data

3

Results

4

Interpretation

5

Conclusion

6 / 23

Introduction Specification and Data Results Interpretation Conclusion Specification Data

Specification

We estimate: slavesi,t = βtemperaturei,t + δi + ηt + ǫi,t. slavesi,t is the number of slaves exported in year t from port i. temperaturei,t is the temperature for port i in year t. δi and ηt are port and year fixed effects. This is estimated using a tobit. Standard errors are clustered by nearest climate point X year, since there are fewer climate points than ports. We will find β < 0.

7 / 23

Introduction Specification and Data Results Interpretation Conclusion Specification Data

Ports and temperature points

8 / 23

Introduction Specification and Data Results Interpretation Conclusion Specification Data

Data

Slave trade The Transatlantic Slave Trade Database: voyage-level data on > 34,000 voyages - year of departure and principal port of slave pur-

• chase. We construct an annual panel of port-level slave exports from

134 ports over 137 years (1730-1866). Temperature Global surface temperature anomalies (annual), on a 5 degree grid, 1730 onwards, from Mann, Bradley, and Hughes (1998, 2004). An “anomaly” is the deviation from the mean temperature from the baseline 1902-1980. We convert these into temperatures. Multi-proxy data: constructed using historical data from coral, ice cores, tree rings, and other long instrumental records. In addition to temperature anomalies relative to 1902-1980, also com- pute climate anomalies by removing the 30-year local running mean from each observation.

9 / 23

Introduction Specification and Data Results Interpretation Conclusion Specification Data

Summary statistics

Mean s.d. Min Max N Slaves exported 444 1,813 34,927 18,358 Slaves (non-zero) 2,543 3,673 1.23 34,927 3,206

• Temp. (interpolated)

25.2 2.33 13.3 27.5 18,358

• Temp. (closest point)

25.2 2.34 13.3 27.4 18,358 Climate 25.2 2.32 13.4 27.3 18,224 Deviation from climate

• 0.00043

0.13

• 0.86

0.62 18,224 Year 1,798 39.5 1,730 1,866 18,358 AEZ: Desert 0.030 0.17 1 18,358 AEZ: Subhumid 0.28 0.45 1 18,358 AEZ: Forest 0.43 0.50 1 18,358 AEZ: Dry Savannah 0.15 0.36 1 18,358 AEZ: Moist Savannah 0.11 0.32 1 18,358

10 / 23

Introduction Specification and Data Results Interpretation Conclusion Main results Mechanisms Robustness

1

Introduction

2

Specification and Data

3

Results

4

Interpretation

5

Conclusion

11 / 23

Introduction Specification and Data Results Interpretation Conclusion Main results Mechanisms Robustness

Higher temperatures reduce slave exports

(1) Temperature

• 3,052.036***

(589.096) Year F.E. Y Port F.E. Y Observations 18,358 Standard errors clustered by Year (786.560) Artificial square (633.235)

Notes: ***Significant at 1%, **Significant at 5%, *Significant at 10%. Standard errors clus- tered by closest climate point X year in parentheses. The dependant variable is slave exports. All regressions are tobit.

12 / 23

Introduction Specification and Data Results Interpretation Conclusion Main results Mechanisms Robustness

The effects are strongest in drier regions

(1) Temperature X Desert

• 3,862.479**

(1,888.139) Temperature X Dry Savannah

• 3,924.739***

(706.096) Temperature X Sub-humid

• 2,643.011***

(863.810) Temperature X Moist Savannah

• 1,570.826*

(801.907) Temperature X Humid forest 239.193 (946.407) Year F.E. Y Port F.E. Y Obs. 18,358

Notes: ***Significant at 1%, **Significant at 5%, *Significant at 10%. Standard errors clustered by closest climate point X year in parentheses. The dependent variable is slave exports. All regressions are tobit.

13 / 23

Introduction Specification and Data Results Interpretation Conclusion Main results Mechanisms Robustness

Climate trends and shocks around them matter

(1) (2) (3)

• Dev. from normal
• 1,244.226**
• 2,640.017***

(573.612) (542.358) Normal

• 18,583.692***
• 20,727.432***

(1,726.335) (1,760.309) Year F.E. Y Y Y Port F.E. Y Y Y Obs. 18,224 18,224 18,224

Notes: ***Significant at 1%, **Significant at 5%, *Significant at 10%. Standard errors clustered by closest climate point X year in parentheses. The dependent variable is slave exports. All regressions are tobit.

14 / 23

Introduction Specification and Data Results Interpretation Conclusion Main results Mechanisms Robustness

Old shocks matter today

(1) (2) (3) (4) (5) (6) (7) (8) Anomaly 0.004** 0.004** 0.065*** 0.050*** 0.139*** 0.059** 0.166*** 0.050*** (0.002) (0.002) (0.022) (0.019) (0.050) (0.027) (0.045) (0.017) Controls N Y Y Y Y Y Y Y Time Period All All 1730s 1740s 1750s 1760s 1770s 1780s Obs. 134 134 134 134 134 134 134 134 R2 0.061 0.466 0.482 0.478 0.484 0.464 0.500 0.480 (9) (10) (11) (12) (13) (14) (15) (16) Anomaly 0.037** 0.050** 0.019 0.036 0.044* 0.031* 0.035* 0.047* (0.018) (0.023) (0.026) (0.033) (0.026) (0.018) (0.020) (0.028) Controls N Y Y Y Y Y Y Y Time Period 1790s 1800s 1810s 1820s 1830s 1840s 1850s 1860s Obs. 134 134 134 134 134 134 134 134 Removed 0.465 0.474 0.444 0.449 0.458 0.456 0.461 0.459 Notes: ***Significant at 1%, **Significant at 5%, *Significant at 10%. Robust standard errors in parentheses. The dependent variable is luminosity. All regressions are OLS. Controls are absolute latitude, longitude, number of luminosity points, AEZ dummies, distance to the nearest port of slave demand, and average temperature over the period 1902-1980. 15 / 23

Introduction Specification and Data Results Interpretation Conclusion Main results Mechanisms Robustness

Robustness 1

Heterogeneity: Port linear trends. Re- gion quadratic trends. Include anoma- lies at the nearest/modal New World port. Include anomalies in shipping

• countries. Include prices in Africa and

New World. Estimate pre-and-post

• 1807. Drop inactive ports. Effects are

concentrated in the periods 1742-58, 1780-1790, and 1827-1845. No differ- ence in El Nino years. Weakly signifi- cant stronger effect for larger ports. Measurement: Use “known” slaves. Use closest temperature point. Use slaves arriving at destination ports. Use slaves normalized by population density in 1700. Most slaves came from near the coast. “Ethnic”-level shocks. Level of observation: Collapse data into 1-degree by 1-degree (and 5- degree by 5-degree) boxes. Collapse to modern countries. Collapse to re- gions. Outliers: Remove influential observa- tions. Remove the bottom 50% of

• ports. Remove bottom 50% of years

by port. Drop each region. Estimation: Use OLS + Conley’s

• OLS. Use whether port exported any

slaves as outcome. First Differences. Port mean anomaly in place of fixed effects. Lag temperature as control. Include running means/variances of slave exports/temperature. Replace year F.E. with trends.

16 / 23

Introduction Specification and Data Results Interpretation Conclusion Main results Mechanisms Robustness

Robustness 2

Alternative mechanisms: Ships sink more in colder years, not less. Wind speed falls by very little. Modern tem- perature shocks are quite correlated

• ver space.

Shocks to interior eth- nic groups give similar results. There are no heterogeneous effects for cattle- keepers or in tsetse-suitable areas. Lag slave exports: Include lag slaves. Instrument for lag slaves with lag dif- ference. Port mean anomaly, year F.E., lag slave exports, and ini- tial slave exports. OLS with Lag. Arellano-Bond.

17 / 23

Introduction Specification and Data Results Interpretation Conclusion Argument Case studies

1

Introduction

2

Specification and Data

3

Results

4

Interpretation

5

Conclusion

18 / 23

Introduction Specification and Data Results Interpretation Conclusion Argument Case studies

Anomalies as a cost shock

A coastal African ruler maximizes profits from selling slaves. This can be thought of as his utility from consuming imported goods (as in Fenoaltea (1999)). Slave traders will pay p per slave; the ruler is a price taker. This also works with downward-sloping demand, as long as p(S) is not too convex. The ruler “produces” S slaves using an army he controls. The cost of harvesting S slaves is C(S, T), where T is temperature. CS > 0, CSS > 0, CT > 0, and CST > 0. That is, there are convex costs of slave harvesting, and higher temperature raises these. So, the ruler’s profit is pS − C(S, T), and anomalies lower slave ex- ports: dS dT = −CST CSS < 0

19 / 23

Introduction Specification and Data Results Interpretation Conclusion Argument Case studies

The costs of harvesting slaves

Why do we think CST > 0?

The costs of taxing the peasantry to feed a slave-harvesting army rise during bad harvests. The mortality of slaves and porters rises when temperature rises. Areas of slave supply become more disordered. Complementary economic activities suffer.

The temperature-agriculture and temperature-mortality links are well established in the literature (Alsop, 2007; Burgess et al., 2011; Kala, Kurukulasuriya and Mendelsohn, 2011; Kurukulasuriya and Mendel- sohn, 2006).

20 / 23

Introduction Specification and Data Results Interpretation Conclusion Argument Case studies

Case studies

Bad harvests raised costs slave supply:

Droughts lead to disorder, dispersal, confrontation with local states, mortality of slave caravans in Benguela. Tribute was often paid as slaves. The Dahomean army was the major captor of slaves in Whydah, and competed with Oyo. Mahlatule droughts of the 1790s and 1820s-1830s in Mozambique dis- rupted settlement patterns, trading networks, and local states; pushed Nguni states north; checked Portuguese and Afro-Portuguese expan- sion, and; made rivers impassable.

Complementary sectors supported the slave trade:

Soldiers, officials, porters, and traders depended on local markets for supplies and to supplement their incomes in Benguela. Stability in interior markets aided traders in Whydah. The slave trade drew on the local retail trade, agriculture, fishing and salt-making. The port imported natron and kola from the interior. Mozambique Island was chronically short of locally-sourced provisions.

21 / 23

Introduction Specification and Data Results Interpretation Conclusion

1

Introduction

2

Specification and Data

3

Results

4

Interpretation

5

Conclusion

22 / 23

Introduction Specification and Data Results Interpretation Conclusion

Conclusion

Supply shocks within Africa influenced the dynamics of the slave trade. These are large in magnitude: a one degree anomaly reduced exports by ∼3,000 slaves. We interpret these as shocks to the cost of slave supply, operating through mortality and the productivity of complementary sectors. Understanding the impacts of smaller weather and climate shocks crucial for climate change adaptation. They can tax societies’ cop- ing mechanisms, which in turn can have large, long-term impacts on development.

23 / 23