Content uploaded by Thomas Mcgregor
Author content
All content in this area was uploaded by Thomas Mcgregor on Mar 01, 2016
Content may be subject to copyright.
DEPARTMENT OF ECONOMICS
OxCarre
Oxford Centre for the Analysis of Resource Rich Economies
Manor Road Building, Manor Road, Oxford OX1 3UQ
Tel: +44(0)1865 281281 Fax: +44(0)1865 271094
oxcarre@economics.ox.ac.uk www.oxcarre.ox.ac.uk
Direct tel: +44(0) 1865 281281 E-mail: oxcarre@economics.ox.ac.uk
_
OxCarre Research Paper 170
Natural Assets: Surfing a wave of
economic growth
Thomas McGregor
Oxford
&
Samuel Wills
OxCarre
Natural Assets: Surfing a wave of economic growth
Thomas McGregor and Samuel Willsú
February 2016
DRAFT
Abstract
Many natural assets can not be valued at market prices. Non-market valuations
typically focus on the value of an individual asset to an individual user, ignoring
macroeconomic spillovers. We estimate the contribution of a natural asset to ag-
gregate economic activity by exploiting exogenous variation in the quality of surfing
waves around the world, using a global dataset covering over 5,000 locations. Treat-
ing night-time light emissions as a proxy for economic activity we find that high
quality surfing waves boost activity in the local area (<5km), relative to compar-
able locations with low quality waves, by 0.15-0.28 log points from 1992-2013. This
amounts to between US$ 18-22 million (2011 PPP) per wave per year, or $50 billion
globally. The effect is most pronounced in emerging economies. Surfing helps re-
duce extreme rural poverty, by encouraging people to nearby towns. When a wave
is discovered by the international community, economic growth in the area rises by
around 3%.
JEL codes: H41, O13, Q26, Q51, Q56
Key words: Non-market valuation, natural capital, surfing, night-time lights.
úOxford Center for the Analysis of Resource Rich Economies, Department of Economics, University
of Oxford, UK. Samuel Wills would like to thank the Economic and Social Research Council for financial
support [grant number ES/K009303/1]. Corresponding email: thomas.mcgregor@economics.ox.ac.uk
1
Figure 1.1: Wealth breakdown by income group (Jarvis et al., 2011)
1 Introduction
Natural assets are an important part of the world’s capital stock. According to the World
Bank (Jarvis et al., 2011) they account for approximately 5% of global capital, and 30%
of that in developing countries (Figure 1.1). Existing estimates of natural capital focus on
assets that can be valued at market prices, including non-renewable assets like energy and
mineral resources, and renewable assets like cropland, pasture and forests. Many other
forms of natural capital are excluded from estimates because they can not be valued
at market prices. These non-market assets include spectacular mountain ranges, tree-
covered forest paths and, to some eyes, clean six-foot waves peeling seductively down a
point break.
This paper estimates the contribution to aggregate economic activity of a particular non-
market natural asset: surfing waves. We choose surfing waves to exploit a natural exper-
iment: the exogenous distribution of high quality waves around the world; and a unique
dataset: a crowd-sourced online database of wave location, characteristics and quality. In
doing so we estimate the value to individual users as well as macroeconomic spillovers in
the surrounding area.
The quality of surfing waves provides a clean natural experiment. It is entirely predeter-
mined by a specific combination of environmental characteristics that include the shape
of the coastline, the sea-floor, and the direction of prevailing winds and swells. For much
of history this specific combination was of no interest to humans, until surfing became a
popular pastime in the 1960s. Unlike most non-market assets we are able use the exogen-
ous variation in the measured quality of waves to determine the marginal contribution of
these assets to economic activity.1
There are two mechanisms by which non-market assets like waves may contribute to
local economic activity: by stimulating activity within the region, and attracting new
activity to it. Stimulating local activity happens by creating a demand for complementary
goods and services. For surfing this includes manufacturing and retail of surfboards,
wetsuits and other specialised accessories, and services such as board repairs, surfing
lessons and lifeguarding. Attracting new activity happens by drawing new demand to the
1The quality of surfing waves has also been used as a natural experiment to study the emergence of
informal property rights in California (Kaffine, 2009).
2
1992 2000 2013
5km
10km
Figure 1.2: Example of illumination growth in the 5km and 10km surrounding Anchor
Point, a “World Class” (quality 3) wave in southern Morocco.
area to exploit the scarce resource. For surfing this may include transient and seasonal
demand, like tourism; or more permanent demand like retirees and people relocating for
lifestyle reasons. Surfing is particularly well suited to studying these mechanisms because
waves are a common pool resource and, as such, are liable to over-use (Rider, 1998).2
The historical solution has been for surfers to travel and discover new waves, making
exploration a core part of surfing lore.
We are able to determine how good quality waves contribute to economic activity by com-
bining a unique dataset on the characteristics of 5,151 waves around the world, with two
detailed and geographically disaggregated datasets on night-time lights and population.
The wave data is compiled using Python from the website www.wannasurf.com. WannaS-
urf is an online database of surf spots recording their location, quality, difficulty, coastal
geography, best wind, swell and tide conditions and accessibility, amongst other things.
The data on the website is crowd-sourced (like Wikipedia) from a community of 78,000
“WannaSurfers”, on whom data is also available. The second dataset records the amount
of light emitted at night-time around the globe, at a 1km2resolution, which is a useful
proxy for economic activity (Henderson et al., 2011; 2012). The third is from LandScan
and uses a variety of spatial inputs to measure population, also at a 1km2resolution.
We employ a polynomial distributed lag model to determine how waves affect illumination,
and in turn economic activity. Our control group is areas surrounding the lowest quality
waves. This is a relatively high hurdle, as these areas are coastal and of sufficient interest
to surfers to appear in our crowd-sourced database. The model measures the marginal
contribution of higher quality waves to economic activity over the course of our sample,
controlling for both wave and time fixed effects.
We find that high quality waves increase economic activity (proxied by lights) in the
surrounding 5km area by 0.15-0.28 log points, or 16%-32%, over 21 years (1992-2013),
relative to places with low quality waves. This amounts to US$18-22 million (2011 PPP)
2Attempts to allocate property rights and charge entry, as at Cloudbreak in Fiji, have been short-lived
due to public outcry.
3
per wave per year in the surrounding 50km, or US$48 billion globally, which is consistent
with existing survey-based estimates.3The effect is highest for 4-star waves (out of 5)
because the highest quality waves tend to be too difficult for the average surfer. Emerging
economies benefit the most, so long as they have a sufficient level of political stability and
ease of doing business.
Economic activity increases in aggregate, rather than simply being reallocated from other
areas. It is shared amongst nearby towns, and is particularly pronounced in the closest
town and the largest town within 50km. Activity in unlit-rural areas, which are typically
extremely poor, does not increase. However, surfing does reduce extreme poverty by en-
couraging the rural poor to move to more urban areas. Overall the permanent population
around 4-star waves falls, consistent with tourists driving up rents. When new waves
are discovered, surrounding economic growth can rise by up to 3 percent. These results
are robust to other, non-surf related, characteristics of the coastline. Our estimates are
a lower bound for the utility value of indirect natural assets, because many of the rents
from the asset will not accrue to the local area (such as profits to surfwear companies,
and the travel costs of tourists spent to get to the waves).
This work contributes to the extensive literature on valuing natural assets. Market-
based techniques are the most straightforward, but are only suited to traded assets and
ecosystem services. This can be extended to non-traded assets through securitisation
(Chichilnisky and Heal, 1998). Non-market techniques are based on either stated or
revealed preferences (Freeman, 1993; Kopp and Smith, 1993). While stated preference
methods like contingent valuation are widely used, bias remains an issue. It can be
improved with appropriate incentives (Carson et al., 2014). Our method relies on revealed
preferences, building on a small literature valuing individual surf breaks using travel costs
(Mavericks, California: Coffman and Burnett, 2009), and hedonic pricing (housing in
Santa Cruz, California: Scorse et al., 2015). We extend these works by studying a panel
of more than 3000 waves in over 130 countries; and by capturing the macroeconomic
spillovers of the asset to surrounding economic activity beyond those captured by the
user or home-owner.
Academic work on valuing non-market natural assets has not yet been adopted consist-
ently by policymakers. Since the Agenda 21 agreement at the 1992 UN Earth Summit in
Brazil there has been an international movement towards integrating environmental and
economic accounts. The most recent iteration is the System of Environmental Economic
Accounting 2012 (SEEA: UN, 2014). This measures natural assets in both physical and
monetary terms, though the scope of the latter is limited to market or near market as-
sets. The SEEA can encompass the value of non-market assets in land values, though
isolating them is difficult. It has prioritised developing “consistent valuation techniques
beyond the System of National Accounts in the absence of market prices”. The World
Bank also estimates natural capital but excludes most non-market assets due to a lack
of data (Jarvis et al., 2011). However, it acknowledges that “missing ecosystem services”
like recreation and aesthetic views may be important, especially in high-income countries.
This work also contributes to the literature on the geographic determinants of economic
activity. Many papers have used night-time lights to this end (Ghosh et al., 2010; Chen
3Lazarow (2009) uses surveys to find that surfing contributes approximately $113-216 million (US$
2011) in direct expenditure to the economy of the Gold Coast, Australia. This covers 8 high quality
waves but excludes macroeconomic spillovers.
4
and Nordhaus, 2011; Henderson et al. 2012; Michalopoulos and Papaioannou, 2013; Smith
and Wills, 2016). Other work uses satellites to study non-marketed ecosystem services
using data on landcover (Sutton and Costanza, 2002; Costanza et al., 2014). Faber
and Gaubert (2015) study the local effects of tourism in Mexican beach towns, using an
instrument for beach quality based on sand colour and offshore islands. In contrast our
quality instrument comes from direct ratings by users, and covers the entire world.
We hope that valuing surf breaks is useful in two ways. The first is development: by
understanding the benefits of surf breaks to local economies, policymakers might be more
willing to invest in infrastructure needed to access them. This is particularly true in
developing countries, where waves remain under-utilised. The second is conservation: by
assessing the value of surf breaks, better cases can be made to conserve them from coastal
erosion, pollution and rising sea levels.
The paper proceeds as follows. Section 2 provides a background to surfing and the geo-
graphical characteristics that give rise to good waves, which underpins our identification
strategy. Section 3 describes our data. Section 4 presents the methodology. Section 5
presents and discusses our findings, and a range of robustness checks. Section 6 concludes.
2 A primer on surfing
Surfing was originally a central part of Polynesian culture. Europeans first observed this
“very dangerous diversion” (King, 1779) in the 1760s, but it was not until the turn of the
20th century that the sport appeared in North America and Australia. It was in these
areas that, after World War II, the global phenomenon of surfing began. Recent reports
estimate continuing growth in the popularity of surfing, with the global population of
surfers rising from 26 million in 2001 to 35 million in 2011 (The Economist, 2012). This is
expected to continue as highly-populated, wave-rich emerging economies like Brazil and
Indonesia increasingly consume leisure.
The waves where surfers practice their craft are created through wind acting on the surface
of the ocean. These waves propagate along the ocean’s surface for up to thousands
of kilometres until they approach shallow water. Resistance from the sea floor slows
movement at the base of the wave, causing the top to spill over, or “break” (Figure 2.1).
For surfers each wave has three key characteristics: size, shape and length, which are
determined by a broad range of factors. The specific combination of factors that creates
good waves underpins our identification strategy.
The size, or amplitude, of a wave is mainly determined by winds that generate swell,
hundreds or thousands of kilometres away from where it is eventually ridden. Important
are the wind’s strength and direction at its source, the area over which it acts, the length
it blows, and how far away the source is. For example, many of the best surfing waves
in Europe occur in the Basque country of northern Spain and southern France. These
waves typically originate in the North Atlantic and are funnelled into the region by the
deep ocean trench of the Bay of Biscay (Figure 2.2).
The shape of a wave describes whether it spills down the face, pitches over, or surges
when it breaks. This is determined locally by the gradient of the sea floor and local wind
5
Figure 2.1: Waves form through wind acting on the surface of the ocean, and break when
they reach shallow water.
i.
ii.
Figure 2.2: Waves formed in the North Atlantic are funnelled by a deep trench into the
Bay of Biscay, creating good surf breaks.
6
i. Spilling wave ii. Pitching wave iii. Surging wave
Figure 2.3: Surfers tend to prefer pitching waves
Figure 2.4: Length describes how long a wave can potenially be ridden.
conditions. A gradual rise in the sea floor causes white-water to spill down the face of the
wave when it breaks. A steep rise in the sea floor - such as moving from deep ocean to a
reef - causes the breaking wave to pitch, creating a “barrel”. A very steeply rising sea-floor
will create a surging wave, as seen at the base of sea cliffs. Local winds also affect this:
“offshore” winds blowing from beach to ocean hold the wave up longer, causing it to pitch
more when it eventually breaks. Onshore winds do the opposite.
The length of a wave describes how long it breaks before reaching the shore. This is
determined locally by the shape of the coastline. Waves break for longer when they reach
the coastline at an angle, causing the whitewater to spill continuously to the left or right.
Long waves therefore typically occur along headlands (“point-breaks”), rivermouths or
coral reefs.
Surfers ride these waves as close to the point of breaking as possible, so the nature of the
breaking process matters. High quality waves will be larger, pitching and longer, all else
being equal. Surfers naturally prefer higher quality waves, but only up to a point dictated
by ability. The direction of the breaking wave is also a consideration as surfers prefer to
face the wave as they ride it (which for most is a wave that breaks to the right). The
shape of a surfboard can be tailored to suit particular types of waves, resulting in local
shaping industries.
High quality waves therefore require a very specific combination of global weather, local
weather and bathymetric conditions, small deviations from which will result in lesser qual-
ity waves. Such a specific combination of characteristics are unlikely to affect economic
activity through any mechanism other than surfing. The range of characteristics also
allows for a lot of natural heterogeneity. Two locations on the same coast may receive
exactly the same swell, but have vastly different quality waves because of the shape of the
sea-floor and coastline. We exploit this heterogeneity in our identification strategy.
7
Figure 3.1: Overview of WannaSurf wave locations
3Data
3.1 Waves
Wannasurf (www.wannasurf.com) is an online “world surf spot atlas, made by surfers for
surfers”. It records the location, quality, type, accessibility, coastal and oceanic character-
istics of 5,288 surf spots (waves) around the world (Figure 3.1). Of these we drop 137 for
which the data on quality is either missing or rated 0 stars (“choss”), leaving 5,151 surf
spots in our dataset. The data is crowd-sourced from a community of 78,000 website users
who create and edit the information on each surf spot, in a similar way to Wikipedia.4
The geographic coordinates of each wave are given precisely. The waves are distributed
amongst 146 countries, though they are particularly concentrated in Australia (888 waves)
and the US (878), as shown in Figure 3.2. This is to be expected because of the long
coastline and large surfing population in these countries, from whom the data is crowd-
sourced.5
Each wave is assigned one of five quality ratings, ranging from “sloppy” (1-star) to “totally
epic” (5-star).6Most waves fall into the 2 and 3-star ratings, as shown in Figure 3.1. Wave
quality is not evenly distributed across countries, with Namibia, Western Sahara and the
Maldives having the highest average quality, and Ukraine, Qatar and Kuwait the lowest.
There is a large exogenous element to this, because coastal structure and exposure to wind
and swell are important components of quality. There is also some degree of selection.
Wannasurf contributors are more likely to record unremarkable waves in countries with
4Each user has a publicly available profile including information on their age, location and surfing
preferences, which we do not use.
5The website was created in 2004, and mainly codified existing knowledge that was previously available
offline. As such we do not have the “discovery date” of each wave.
6This rating is crowd-sourced. There is also a user poll for wave quality rating, though we don’t use
this as it typically has less than 100 respondants for most locations.
8
Figure 3.2: Our data on waves is distributed around the world, though is particularly
concentrated in Australia and the US.
Star Rating Description Frequency Share
1 Sloppy 384 7.5%
2 Normal 2,027 39.4%
3 Regional Classic 2,129 41.3%
4 World Class 450 8.7%
5 Totally Epic 161 3.1%
Total 5,151 100%
Table 3.1: Breakdown of waves by quality.
large surfing populations, either local or tourist. On the other hand, only good waves
tend to be recorded away from the beaten path. Western Sahara is an example. Of the
four waves recorded, three are 4-star, and one is 5-star.
The characteristics of each wave are also recorded. These include variables on accessibility
(“Distance”, “Easy to find?”, “Public access?”, “Crowd”), difficulty (“Experience”), the
type of wave (“Frequency”, “Type”, “Direction”, “Length”, “Bottom”, “Power”) and
oceanic conditions (“Good swell direction”, “Good wind direction”, “Swell size”, “Best
tide”). Of these we make particular use of the “Type” variable, which indicates whether
the shoreline is a beach, a reef, a rivermouth, a headland (point-break) or a breakwater.
Wave quality also varies by Type, as shown in Figure 3.3.
Finally, we also conduct a small event study around the date that waves were discovered.
The date of discovery is taken from two sources. The first is the date of the “Rip Curl
Pro Search” competition, which was an event on the surfing world tour organised by
the Association of Surfing Professionals annually from 2005-2010. It took place in a
different location each year, which was previously relatively unknown to the global surfing
community. The second is the “Google Earth Challenge”, which was a competition run
by Surfing Magazine in 2007 to discover a previously unknown wave using Google Earth.
Surfing Magazine is a leading industry publication read by millions worldwide. Table 3.2
shows the seven wave discoveries in our sample.
9
Figure 3.3: Breakdown of wave quality by type.
Wave Country Date of Discovery Quality Source
St Leu Reunion 2005 World Class Rip Curl
La Jolla Mexico 2006 World Class Rip Curl
El Gringo Chile 2007 World Class Rip Curl
Skeleton Bay Namibia 2007 Totally Epic Surfer Mag
Uluwatu Indonesia 2008 Totally Epic Rip Curl
Supertubos Portugal 2009 Totally Epic Rip Curl
Middles Puerto Rico 2010 Regional Classic Rip Curl
Table 3.2: Wave Discoveries
10
Figure 3.4: Night-time light and population data
3.2 Night-time lights
The Defence Meterological Satellite Program’s Operational Linescan System (DMSP-
OLS) uses satellites to record the average annual night-time light intensity around the
world, from 1992-2013 (Figure 3.4). The data is provided at a resolution of 30x30 arc-
seconds (approximately 1 square kilometre near the equator), and ranges from 0 to 63.
The data is constructed by overlaying all daily images over the course of a year, discarding
those that are obfuscated by cloud cover, lightning, aurora, etc. for a given pixel.
There is a strong link at the national level between the growth of GDP and mean light
intensity (Doll et al., 2006; Henderson et al., 2012; Michalopoulos and Papaioannou, 2014).
This is illustrated in Figure 3.5, which plots the log of the sum of light readings by country
against two measures of log PPP-adjusted GDP: based on expenditure and production
(Penn World Tables 8.1). The associated regressions yield an adjusted r-squared of .82
and .80 respectively. We make use of the high spatial resolution of the data to study
economic activity at a sub-national level, as has been done in a number of other studies
(Chen and Nordhaus, 2011; Michalopoulos and Papaioannou, 2013; Hodler and Raschky,
2014; Jedwab and Moradi, 2015; Jedwab et al., 2015).7
7This data is subject to “top-coding”, where economic activity beyond the maximum luminosity rating
11
6 8 10 12 14 16
ln(GDP), Expenditure, 2003
0 5 10 15 20
Ln(Total Lights, 2003)
Figure 3.5: PPP-adjusted GDP vs Night-time lights (in logs), 2003 (see Smith and Wills,
2016).
3.3 Population
The LandScan data set is produced by the Oak Ridge National Laboratory and provides
annual mid-year spatial population counts at a 30x30 arcsecond resolution from 2000-2013
(Figure 3.4).8It reports “ambient” population, which is the average over a 24 hour period,
rather than simply where people sleep. However, it excludes intermittent populations such
as tourists or temporary relief workers, and may not reflect things like seasonal migrations
or refugee movements. The dataset is constructed by distributing known national and sub-
national population counts across the grid according to a likelihood model that uses inputs
including land cover data, roads data, and high resolution satellite imagery, among other
sources.9This data is similar to that from NASA’s Socioeconomic Data and Applications
Center (SEDAC) which also measures population at a 30x30 arcsecond resolution, but is
only available for the years 1990, 1995 and 2000 (see Dell, 2010 and Alesina et al., 2015
amongst others).
The LandScan data pays special attention to coastal features. To account for the dynamics
of coastal change the LandScan model extends all coastal boundaries several kilometres
seaward. This ensures that all shore and small island features are included within an
administrative unit boundary.10
3.4 Urban and rural classification
SEDAC also provides an “Urban Extents Grid”, which uses 1995 population count estim-
ates to classify each square of a 30x30 arcsecond global grid as either urban or non-urban.
of 63 cannot be distinguished. This is mostly an issue in the centre of dense, economically active areas
in developed countries (Michalopoulos and Papaioannou, 2014), and is not a particular concern in our
study of coastal areas. The data is also subject to “overglow”, where lights appear larger over water and
snow. We address this by clipping our dataset to the shoreline.
8This is the same resolution as the lights data, although the pixels are not aligned. We use grid cells
aligned with the lights rasters but not the population rasters. To address this the Zonal Statistics tool
in ArcGIS internally resamples the raster files so that they are aligned.
9For further detail http://web.ornl.gov/sci/landscan/landscan_documentation.shtml)
10For more information see the LandScan documentation http://web.ornl.gov/sci/landscan/landscan_documentation.shtml#01.
12
The classification is based on contiguous lighted squares (as of 1995) and squares known
to hold at least 5000 people.
3.5 Political stability and ease of doing business
The World Bank provides data on political stability in its Worldwide Governance Index,
and on the ease of doing business in its Doing Business Survey. We collect countries into
four groups with similar numbers of waves based on their 2014 scores on each. Table B.2
groups countries by political stability, and table B.1 groups them by the quality of their
business environment.
4 Methodology
4.1 Measuring economic activity
In our analysis we use two measures of economic activity as the dependent variable:
illumination in the immediate vicinity and illumination in nearby towns.
Illumination in the immediate vicinity is measured using luminosity in surrounding circles
of various radii. We draw these circles at 1km, 5km, 10km and 50km around each wave
and take the sum of illumination within each circle for each year. Because waves are
located on the coastline, and there is no economic activity generated out to sea, we clip
these circles so as only to include area covered by land.
We also separate each circle into urban, lit-rural and unlit-rural areas. The distinction
is based on the 1990 SEDAC Urban Extents Grid dataset, which demarcates urban and
rural areas using a combination of population counts (persons), settlement points, and the
presence of night-time lights. Urban areas are those with significant lit cells, a buffered
settlement point, or a total population greater than 5,000 persons. Lit-rural areas are
non-urban cells that were lit in 1992. Unlit-rural areas are non-urban cells that were
not lit in 1992 but have a positive population. Studying unlit-rural areas allows us to
determine the impact of natural assets on rural poverty (see Smith and Wills, 2016).
Illumination in nearby towns is measured by endogenously locating towns by their popu-
lation density. A town is defined by a perimeter enclosing cells with a population density
of 300 persons per square kilometre or more.11 Each wave is linked to two towns: the
closest town and the largest town within 50km radius (based on total population).
4.2 Identification
Estimating the economic return to a non-market natural asset presents challenges of
endogeneity and attributability. We address this using a natural experiment that exploits
the exogenous variation in the quality of surfing waves.
11We also use a cut-offof 600 persons per square kilometer as a robustness check.
13
Endogeneity can arise for a variety of reasons. The first, and most obvious is due to
omitted variables: observable and unobservable characteristics that are correlated with
both the location of waves and local economic activity. If these omitted variables are
time-varying then controlling for time fixed-effects will not help. Examples of omitted
variables for areas near waves could include geographic characteristics, political stability
and institutional quality.
Endogeneity can also arise due to reverse causality. The exploitation of surfing waves may
depend largely on the level of economic activity, and associated infrastructure, already
established in the area. We believe this to be less of an issue for surfing, given the strong
history of intrepid exploration by surfers, the rarity of top quality waves and the sheer
isolation of some locations in our data. For example, to access Red Bluffin Western
Australia (a “totally epic” quality 4 wave) one must drive for 4.5 hours along a dirt road
from Carnarvon, an isolated town with population less than 5000. Reverse causality
remains a concern.
Attributing changes in economic activity to a specific natural asset requires strong iden-
tification of the asset itself. There may be many other factors, natural or otherwise, that
attract economic activity to the area surrounding a wave specific area. For surfing waves
this may include trade, boating, fishing and the benefits of nearby beaches.
We address the identification challenges of endogeneity and attributability by exploiting
the exogenous variation in wave quality, rather than the existence of waves per se. Whilst
the location of a wave may well be endogenous to local economic activity for the afore-
mentioned reasons, the variation in the quality of a wave’s surfing potential is exogenous.
In other words, the quality of a surfing wave can be treated as a natural experiment.
We also conduct robustness tests to verify the success of our identification strategy. This
includes varying the baseline quality to control for selection of low quality waves into our
database, and testing whether particular coastal features (reefs, rivermouths, etc) drive
our results.
4.3 Estimating Equations
The effect of surfing wave quality on spatial outcomes are estimated using the following
polynomial distributed lag model:
Yi,t =–+—(t)Qi+“(t)+Wi+Zt+‘i,t (4.1)
where
—(t)=—1t+—2t2+—3t3+—4t4
“(t)=“1t+“2t2+“3t3+“4t4
where Yi,t is the outcome of interest for wave iat time t=[0,...,21],Qiis an indicator
equal to zero if the wave is of poor quality (1-star) and one if the wave is of some higher
quality (2-5 stars), Ztis time fixed effects and Wiis wave fixed effects. The polynomial
14
structure is imposed on —(t)and “(t)to reduce the effects of collinearity in the data. The
standard errors are clustered at the wave level to address the spatial correlation between
observations (surf breaks).
An alternative linear specification is used test the relative significance of wave qualities,12
Yi,t =–Õ+
t
ÿ
s=0
—Õ(Zs◊Qi)+Zs+Wi+‘i,t.(4.2)
The dependent variable in our estimating equation, Yi,t, varies between the log of lights
and the log of population, either in the immediate vicinity of waves or in nearby towns,
depending on the outcome of interest.
We drop observations for which there is no GPS data for the location of the wave which
leaves us with 5,288 waves globally. We then drop those observations for which the wave
quality is either missing or rated 0 stars. In total these make up 2.6% of all waves, leaving
5,151 observations in total.
5 Results
This section shows that good quality waves boost economic activity in the surrounding
areas, relative to areas with low quality waves. The effect is most pronounced in emerging
economies. Activity (proxied by night-time illumination) increases overall, rather than
simply being reallocated from nearby areas. However we do find that the permanent
population falls around good waves, which we attribute to tourism. The increase in
activity is broad-based: it occurs in the immediate vicinity and in nearby towns. It
also reduces rural poverty by encouraging the poor to move to areas of higher activity.
These results are robust to a variety of controls, including non-surfing related coastal
characteristics and alternative specifications.
5.1 Good surfing waves boost nearby economic activity
Areas with high quality surfing waves have higher economic activity than those with low
quality waves, peaking with 4-star waves. This is a relatively high hurdle. Our control
group is the area surrounding the lowest quality waves, and so is already on the coast and
sufficiently known by surfers to appear in WannaSurf. We also control for time and wave
fixed effects, to isolate the marginal effect of good surfing conditions.
Figure 5.1 uses the model in equation 4.1 to show how waves affect economic activity
within a 5km radius. Higher quality waves increased economic activity, with 4-star waves
increasing activity by 0.28 log points relative to 1-star waves over our sample. 5-star
increased relative activity by 0.19 log points, 3-star waves by 0.15 log points, and 2-star
waves by 0.03 log points (which was not significant). The relative effect on activity at
12Results available online.
15
Figure 5.1: Effect of waves of various qualities on economic activity within 5km.
Figure 5.2: Distribution of experience required by wave quality.
1km, 10km and 50km radii display a similar pattern, where 4-star waves have the largest
effect on local economic activity (though slightly less pronounced, see Appendix A).
The economic impact of good quality waves is significantly higher than bad quality waves
after 21 years. We test the significance of wave quality against 1-star waves in the main
specification, and against other quality waves in the linear specification (equation 4.2).
Both 3-star and 4-star waves increase illumination significantly more than 2-star waves
at the 1% or 5% level for most distances (1km - 50km). 4-star waves only significantly
increase illumination more than 3-star waves at the 50km radius.
Wave quality has an inverse U-shaped effect on economic activity. This is because the
highest quality waves require a lot of experience to ride, being disproportionately rated
for “Pros or Kamikazes only” (experience level 3) as illustrated in Figure 5.2. This limits
their appeal.
16
5.2 Emerging economies benefit from surfing the most
High quality waves increase economic activity on average, though the effect is concentrated
in emerging economies. Waves largely generate economic activity through tourism. The
strength of this channel will depend on both supply and demand. The demand for tourism
will depend on the institutional and political characteristics of the country. The supply
of tourism services will depend on the ease with which new businesses can respond to an
inflow of prospective surfers. Using World Bank data on political stability and ease of
doing business we find that waves have the most pronounced effect in countries that score
“low” on both, as they have significant scope to grow.
Figure 5.3 shows how 4-star waves affect economic activity in the surrounding 5km, based
on their country’s political stability and ease of doing business. We omit the USA and
Australia from the analysis as they are large, developed and stable countries who over-
whelmingly dominate our sample. Waves have the largest effect on countries with interme-
diate political and business environments. Countries with “low” “or moderate” political
stability are sufficiently stable to attract surfers, unlike those countries scoring “very
low”. However, they are unstable enough that their tourism industry still has scope to
grow. Similarly, countries with a “moderate” business environment are able to facilitate
economic activity,allowing tourism to expand to meet demand from surfers, unlike those
scoring “low” or “very low”. However, they will not have such well established tourism
infrastructure that surfers will not add to economic activity.13
5.3 Surfing increases activity overall, rather that redistributing
it from surrounding regions
Economic activity near good waves increases in aggregate, rather than just being drawn
away from other areas. Figure 5.4 shows how waves of different quality affect economic
activity in surrounding concentric rings out to a 50km radius, relative to 1-star waves. If
the increase in activity described in Section 5.1 simply reallocated activity from surround-
ing areas, then we would expect higher activity in the closest rings, and lower activity
further out. Instead we find that activity is higher in all rings, and falls the further from
the wave one travels. If anything this suggests that surfing generates positive spillovers
for the surrounding areas.
We find that 4-star waves drive the largest increase in surrounding activity at all distances
over 21 years. Within 5km the effect was 0.28 log points, falling to 0.26 log points in
the 5-10km, and 10-50km rings. The effect of 2-star waves remained insignificant at all
distances, 3-star waves fell from 0.15 to 0.10 log points, and 5-star waves rose from 0.12
to 0.20 log points. This is again consistent with an inverse-U relationship between wave
quality and economic activity.
While surfing does not redistribute economic activity it does redistribute the permanent
population. The permanent population falls in the 5km surrounding 2-5 star waves, rel-
ative to 1-star waves (see Figure 5.5, panel i.). This is most pronounced for 4-star waves,
13To confirm that our results are not being driven by large, developed, wave-rich countries we re-run
the analysis specifically for Australia and the USA, as shown in Figure B.1.
17
i.
ii.
Figure 5.3: Effect of wave quality on local illumination by i. business environment and ii.
political stability
18
i.
ii.
Figure 5.4: The effect of waves of various quality on economic activity in expanding
concentric rings of i. 5-10km and ii. 10-50km.
19
which increase economic activity the most and reduce the local permanent population
by -0.35 log points over our sample. Panel iii. of Figure 5.5, shows that the population
increases at further distances, particularly at 10km-50km. This is consistent with the per-
manent population moving away from the waves because tourists, which are not included
in the LandScan data, drive up property values.
5.4 Surfing benefits areas of existing economic activity, includ-
ing nearby towns
We have seen that surfing waves increase nearby economic activity. Now we investigate
how this happens. First, we break the 5km circle surrounding each wave into urban and
lit-rural areas. Second, we construct our own measure of the closest town, and the largest
town within 50km. We find that good waves increase activity in each. In the next section
we turn our attention to unlit-rural areas, and the role of natural assets in economic
development.
Figure 5.6 shows how the 5km surrounding each type of wave is divided between urban,
lit-rural and unlit-rural areas. Overall the largest proportion of waves are in urban areas
(48%), followed by lit-rural (43%) and unlit-rural areas (9%). As wave quality increases,
so too does the share of waves in rural areas. This is consistent with selection in the
WannaSurf database. While a 1-star wave in an urban area might be surfed sufficiently
often to warrant entry in the WannaSurf database, the same wave in a rural area might
not. In contrast, surfers might be willing to travel to rural areas to surf a totally epic
(5-star) wave. Selection may affect our results, because our interpretation of faster light
growth near high quality waves might actually be a story of faster light growth in rural
areas. As we will show next, the second interpretation can be discarded because our
results also hold when considering urban, and rural areas individually.
Figure 5.7 investigates how aggregate light growth in the 5km surrounding each wave is
allocated between urban and lit-rural areas. The main results in Section 5.1 is confirmed:
better quality waves increase illumination in the surrounding areas, peaking with 4-star
waves. The effect is larger in lit-rural than urban areas because their initial level of
illumination is lower, allowing for a larger percentage change. As well as understanding the
nature of growth near natural assets, this also provides some robustness for the selection
effects mentioned above.
As well as increasing economic activity in their immediate surroundings, waves also in-
crease economic activity in nearby towns. Figure 5.8 shows how waves affect illumination
in the closest town (defined as having a population density over 300 people per km2),
and in the largest town within 50km. Illumination in the closest town increases by 0.14
log points for 3 and 4-star waves, while there is a small an insignificant effect for 2 and
5-star waves. In contrast illumination in the largest nearby town increases for all wave
qualities, with the largest effect of 0.15 log points for 4 and 5-star waves. The effect on
the largest nearby town is larger than on the closest town for each wave quality. This
suggests that the economic benefits of natural assets like surfing waves tend to accrue to
areas of existing economic activity, where there is the infrastructure needed to support
tourism and other non-recreation activities. The effect is most pronounced for the highest
20
i.
ii.
iii.
Figure 5.5: Effect of waves on population within surrounding i. 5km, ii. 5-10km and iii.
10-50km concentric rings.
21
Figure 5.6: Breakdown of the 5km surrounding waves of each quality
i.
ii.
Figure 5.7: Effect of waves on lights in i. urban and ii. lit-rural areas within a 5km radius.
22
i.
ii.
Figure 5.8: Effect of wave quality on the i. closest town (>300 people per km2) and ii.
largest town within 50km, relative to lowest quality waves
quality waves, which can be attributed to their proportionally greater incidence in rural
areas, and the greater requirements needed to service surfers riding those waves (such as
board repairs, healthcare, etc).
5.5 Surfing also reduces rural poverty
Natural assets account for a large share of the capital stock in developing countries.
There is 365,000km of coastline in the world - nine times the circumference of the earth. A
significant proportion of this lies in developing countries, with Africa’s coastline stretching
for 26,000km, and Asia’s 62,800km. The potential for surfing assets to exist along these
coastlines is huge. This section shows that waves have significant potential for reducing
poverty in their local areas.
To understand the potential for waves to reduce poverty we turn our attention to unlit
rural areas. These are areas within 5km of a wave that were unlit but had a positive
23
Figure 5.9: Population living in unlit rural areas within 5km of surfing waves.
population in 2000 (when our data begins). Smith and Wills (2016) show that the share
of population living in unlit rural areas - the rural poor - is a good proxy for extreme
poverty. When households cross the extreme poverty line they quickly illuminate, due to
the high returns from longer working hours. Figure 5.9 shows that the population living
in unlit rural areas falls in areas near good waves by up to 0.7 log points, relative to 1-star
waves. This suggests that there is significant potential for natural assets like surfing waves
to be harnessed for reducing extreme poverty.
There are two ways that the population living in unlit rural areas can fall: by unlit
areas becoming lit, or by people moving away from unlit areas. Figure 5.10 shows the
proportion of areas that were unlit in 2000 that became lit in subsequent years. Areas near
good waves illuminate slower than areas near bad waves. It suggests that waves do not
cause unlit rural areas to become illuminated. This implies that people move away from
unlit areas. Testing this implication directly produces a statistically insignificant result.
However, we do find evidence that the population in nearby towns increases (Figure 5.11).
In large towns this amounts to a population rise of up to 0.35 log points for 5-star waves
over our sample. Figure 5.11, panel ii.). This is consistent with surfing waves reducing
extreme rural poverty by attracting people from rural areas to nearby towns.
5.6 Surfing waves contribute around US$50 billion to global eco-
nomic activity each year
We find that surfing waves contribute US$51.2 billion (2011 PPP) globally each year to
economic activity in their surrounding 50km. To arrive at this figure we allocate all pixels
of luminosity within 50km of a wave to a particular wave quality (proportionately if within
50km of more than one wave). This accounts for 9.4% of global illumination. We then
aggregate total illumination and total GDP (US$ 94.1 trillion in 2011 PPP , World Bank
WDI) to find an average value for each pixel of luminosity. Using the parameter estimates
from equation 4.1 we deduce the marginal contribution of surfing waves to activity for
24
Figure 5.10: Share of unlit areas in 2000 that became lit in subsequent years, within 5k
of a wave.
i.
ii.
Figure 5.11: Effect of waves on population in nearby towns (quadratic polynomial model)
25
Star rating Annual Contribution per Wave Frequency Annual Contribution Total
3 17.9 2,129 38,109
4 22.3 450 10,035
5 19.0 161 3,059
Total 2,740 51,203
Table 5.1: Annual global contribution of waves to economic activity in surrounding 50km
(US$ million 2011 PPP)
each wave quality, averaged over 22 years. We exclude 2-star waves because their effect
is not statistically significant from zero. The results are reported in Table 5.1.
Each 4-star wave contributes US$22.3 million on average to surrounding economic activity
each year. 5-star waves contribute US$19.0 million on average, and 3-star waves US$17.9
million on average. The largest aggregate contribution comes from 3-star waves which,
due to their prevalence, account for 75% of surfing’s contribution to global GDP.
5.7 Discovering a new wave significantly raises activity in the
local area
To provide further evidence that good quality waves improve local economic activity we
conduct a small event study, using the date that waves were discovered. This is different to
the previous analysis because we are exploiting exogenous variation over time, rather than
over space. Using discovery dates from two sources, the “Rip Curl Pro Search” competition
and the “Google Earth Challenge” (see Section 3.1), we find that illumination near waves
grows 4% faster after they are discovered by the international surfing community.
There are two challenges in conducting this type of event study. First, we require a
meaningful definition of a wave being “discovered”. If a wave is only known to a handful
of locals, it is unlikely to generate much local economic activity. What we really mean by
“discovery” is that the global surfing community becomes aware, for the first time, about
a new high-quality wave. Second, we need to define when a discovery takes place. There
is no official surfing body or archive that stores and maintains this kind of information.
Discoveries must also take place within our night-time lights sample, from 1992-2013.
To estimate the impact of the discovery of a surfing wave on local economic activity we
fit the following linear model on our sample of seven wave discoveries:
Yi,t =–+—i,tWiúTt+”i,t DiúTt+Zi+Wi+‘i,t (5.1)
where Yi,t is the log of lights within 50km of each wave i=[0,...,5] at time t=[0,...,21],
Ttis a continuous time variable cantered on zero at the year of discovery, Wiis a wave
fixed effect, Ztis time fixed effects, Diis our discovery indicator equal to 0 before the
year of discovery and 1 after. The coefficient —i,t measures the linear growth rate of lights
before the wave was discovered, and ”i,t measures the change in the rate of growth of
lights after the wave was discovered. Eq. 5.1 compares light growth over time pre and
26
post discovery, controlling for changes in light intensity over time that are common to all
waves and differences in light intensity between each wave.
Our main coefficient of interest, ”i,t, estimates the change in trend light growth after the
wave was discovered: our treatment effect. We find ˜
”i,t =0.04 which is significant at
better that the 1% level (p“>|t|=0.005).
This result implies that discovering a wave leads to an increase in annual light growth
of around 4% per year, on average across our seven discoveries. Translating our headline
results to annualised growth rates yields increases in annual growth rates of 0.7% for 3-
star waves, 1.2% for 4-star waves and 0.8% for 5-star waves, all relative to 1-star (normal)
waves. This is substantially larger than our headline results for 3-star, 4-star and 5-star
waves.
The following figures plot the results from Eq. 5.1 for each of our seven wave discoveries.
They compare the predicted light growth pre and post discovery with the actual light
growth over time, controlling for changes in light intensity over time that are common to
all waves. The red line in each of the figures denotes the date of discovery for each wave,
centred at zero, whilst the solid and dashed black lines denote the fitted linear model pre
and post discovery respectively for each wave. The results are clear. Discovering a wave
increases trend growth rates in the local area.
5.8 Robustness
To check the robustness of our main results we investigate two alternative explanations.
The first adds to the wave event study of the previous section by investigating whether
our results may be driven by other coastal characteristics unrelated to surfing. The second
tests whether the selection into our database of low quality waves near towns is driving
our results; complementing the town-based analysis in Section 5.4. Our results are robust
to both tests.
Non-surfing coastal characteristics To test whether our identification strategy is
valid we investigate whether other, non-surfing related, coastal characteristics might be
driving our results. If the particular meteorological and bathymetric conditions that
create good surfing waves also give rise to other economic activity, like swimming on
sandy beaches, diving on coral reefs, fishing from rivermouths, or trading from harbours,
then our results may be biased. To test this we exploit data on the type of each wave,
outlined in Table 5.2.
Figure 3.3 shows that the distribution of wave quality varies by type, with rivermouths,
reefs and point-breaks all being better than average. Rivermouths make up only 2.6% of
our observations (2.7% of 4-star and 1.2% of 5-star waves) and so there are not enough
to systematically bias our results. Reefs and point-breaks comprise a greater share, so
we re-run our analysis excluding these observations to see if our results still hold (see
Figure 5.13). Excluding reefs, point-breaks and both produces a similar outcome to the
main results in Section 5.1. 4-star waves increase illumination in the surrounding 5km by
0.20-0.25 log points over our sample.We also get broadly similar results when we exclude
27
i. ii.
iii. iv.
v. vi.
vii.
Figure 5.12: Event study results: five wave discoveries
28
Wave Type Description Frequency Percent
0 Beach-break 2,013 41.6
1Breakwater/jetty 124 2.6
2 Don’t know 12 0.3
3 Point-break 643 13.3
4 Reef-artificial 23 0.5
5Reef-coral 261 7.5
6Reef-rocky 998 20.6
7 Rivermouth 118 2.4
8 Sand-bar 549 11.3
N/A Missing 85 N/A
Total 4,926
Table 5.2: Breakdown of waves by type
reefs and point-breaks from our analysis of closest and largest nearby towns in Section
5.4, as described in Appendix C. For both the closest and the largest nearby towns the
effect is of a similar magnitude to our main results, peaking with 4-star waves for the
closest town and 5-star waves for the largest town.
Selection of low-quality urban waves The WannaSurf database exhibits some se-
lection bias, where the lowest quality waves are more likely to appear in the data when
they are close to towns and cities (54%). This might bias our results if lights in towns
grow slower than in rural areas. To test this we re-run the analysis using 2-star waves as
the baseline, rather than 1-star. A similar share of 2-star waves are located in towns or
cities (49%), as are the 4-star waves that are our focus (48%). Figure 5.14 shows that our
results are broadly the same as with the 1-star baseline, with slightly smaller coefficients
(also see Appendix D).
29
i.
ii.
iii.
Figure 5.13: Robustness test excluding i. reefs, ii. points and iii. both.
30
Figure 5.14: Effect of waves on illumination in the surrounding 5km, with quality 1 as
baseline.
6 Conclusion
This paper offers a global panel study on the effect of a non-market natural asset on
economic activity. We combine three high-resolution spatial datasets, on the location of
surfing waves, night-time light emissions and population, to answer the question: how
much do surfing waves contribute to the surrounding economy?
We find that surfing waves contribute approximately US$50 billion to global economic
activity each year. This amounts to an average contribution of US$18-25 million per
wave per year.The effect on activity increases with wave quality, except for the highest
quality waves which require a lot of skill to ride. Emerging economies benefit the most
from surfing, as long as they have a sufficient level of political stability and ease of doing
business. Furthermore, the increase in activity does not just represent a reallocation
away from surrounding areas. Waves do, however, cause the permanent population to
move further away - which is consistent with tourists driving up property prices. Surfing
also appears to play a role in reducing extreme poverty, again by encouraging people to
move away from unlit rural areas into nearby towns. These results capture the value of
macroeconomic spillovers from the natural asset, in contrast to most existing methods of
non-market valuation. The results are also robust to a range of tests for endogeneity and
attributability.
While this began as a personal interest project for a couple of sandy-footed economists,
it also has several policy implications. The first is to provide policymakers with an
understanding of the potential benefits of waves for economic development, especially
in developing countries. This is true for both naturally occurring waves and artificially
constructed waves - be they offshore artificial reefs or onshore wave pools. The second is
to highlight the importance of conserving the quality of waves. This involves limiting both
coastal pollution and changes to the characteristics of waves through dredging, coastal
manipulation or rising sea levels.14
14It may also involve protection from sharks, which has been at the forefront of the surfing community’s
31
The paper also suggests a range of extensions. By providing a methodology for valuing
the economic spillovers from non-market natural assets we capture externalities that may
not be included in other methods of non-market valuation. This methodology is relevant
for any natural asset that exogenously varies in quality around the world, including rock-
climbing cliffs and natural reserves (including UNESCO natural heritage sites).
References
[1] Carson, R. T., T. Groves, and J. A. List (2014). Consequentiality: A theoretical
and experimental exploration of a single binary choice. Journal of the Association of
Environmental and Resource Economists 1 (1), 171–207.
[2] Chen, X. and W. D. Nordhaus (2011). Using luminosity data as a proxy for economic
statistics. Proceedings of the National Academy of Sciences 108(21), 8589–8594.
[3] Chichilnisky, G. and G. Heal (1998). Economic returns from the biosphere. Nature 629.
[4] Coffman, M. and K. Burnett (2009). The value of a wave: An analysis of the mavericks
region, half moon bay, california. Study commission for Save The Waves Coalition.
[5] Costanza, R., R. de Groot, P. Sutton, S. van der Ploeg, S. J. Anderson, I. Kubiszewski,
S. Farber, and R. K. Turner (2014). Changes in the global value of ecosystem services.
Global Environmental Change 26, 152–158.
[6] Farber, B. and C. Gaubert (2015). Tourism and economic development: Evidence
from mexico’s coastline. UC Berkeley Economics Department working paper .
[7] Freeman, A. M. (1993). Valuing natural assets: the economics of natural resource
damage assessment. Washington DC: Resources for the Future.
[8] Ghosh, T., R. L Powell, C. D Elvidge, K. E Baugh, P. C Sutton, and S. Anderson
(2010). Shedding light on the global distribution of economic activity. The Open
Geography Journal 3 (1).
[9] Henderson, J. V., A. Storeygard, and D. N. Weil (2011). A bright idea for measuring
economic growth. The American economic review 101 (3), 194.
[10] Henderson, J. V., A. Storeygard, and D. N. Weil (2012). Measuring economic growth
from outer space. The American Economic Review, 994–1028.
[11] Jarvis, M., G.-M. Lange, K. Hamilton, D. Desai, B. Fraumeni, B. Edens, S. Ferreira,
H. Li, L. Chakraborti, W. Kingsmill, et al. (2011). The changing wealth of nations:
measuring sustainable development in the new millennium. The World Bank.
[12] Kaffine, D. T. (2009). Quality and the commons: The surf gangs of california. Journal
of law and economics 52 (4), 727–743.
[13] King, J. (1779). Diary, kealakekua bay, hawaii.
mind since former World Champion Mick Fanning was attacked by a shark in the 2015 final of a World
Tour event in Jeffrey’s Bay, South Africa
32
[14] Kopp, R. J. and V. K. Smith (1993). Valuing Natural Assets: The Economics of
Natural Resource Damage Assessment. Resources for the Future, Washington, DC.
[15] Michalopoulos, S. and E. Papaioannou (2013). Pre-colonial ethnic institutions and
contemporary african development. Econometrica 81 (1), 113–152.
[16] Polyakov, M., D. J. Pannell, R. Pandit, S. Tapsuwan, and G. Park (2015). Capitalized
amenity value of native vegetation in a multifunctional rural landscape. American
Journal of Agricultural Economics 97 (1), 299–314.
[17] Rider, R. (1998). Hangin’ten: The common-pool resource problem of surfing. Public
Choice 97 (1-2), 49–64.
[18] Scorse, J., I. Reynolds, A. Sackett, et al. (2015). Impact of surf breaks on home
prices in santa cruz, ca. Tourism Economics 21 (2), 409–418.
[19] Smith, B. and S. Wills (2016). Left in the dark? oil and rural poverty. Techni-
cal report, Oxford Centre for the Analysis of Resource Rich Economies, University of
Oxford.
[20] Sutton, P. C. and R. Costanza (2002). Global estimates of market and non-market
values derived from nighttime satellite imagery, land cover, and ecosystem service val-
uation. Ecological Economics 41 (3), 509–527.
[21] The Economist (2012). Surf economics: Beach rush. The Economist March 17.
[22] Trends, L. (2013). 2013 sima retail distribution study. Technical report, Surf Industry
Manufacturers Association.
[23] United Nations (2014). System of Environmental-Economic Accounting 2012 - Cen-
tral Framework. "The United Nations", New York.
33
Appendix
A Aggregate economic activity
Figure A.1: Effect of waves of various qualities on economic activity within 1km, 10km
and 50km.
34
B Ease of doing business and political stability cat-
egories
Figure B.1: Headline results are not being driven by the USA or Australia
35
WB DB - very low WB DB - low WB DB - moderate WB DB - high
Country Waves Country Waves Country Waves Country Waves
Brazil 291 South Africa 207 Spain 182 France 290
Indonesia 136 Puerto Rico 53 Japan 118 Portugal 162
Ecuador 47 Chile 50 Italy 113 United Kingdom 149
Argentina 34 Costa Rica 50 Mexico 92 New Zealand 114
Venezuela 24 Greece 50 Peru 76 Ireland 58
Nicaragua 20 Morocco 43 Mauritius 14 Canada 38
Senegal 20 Philippines 29 Belgium 12 Netherlands 32
Barbados 19 Panama 27 United Arab Emirates 9 Germany 22
Sri Lanka 18 Israel 25 Bulgaria 8 Taiwan 18
India 13 Uruguay 24 Poland 4 Denmark 15
Micronesia 11 Turkey 18 Croatia 3 Iceland 14
Bahamas 10 Namibia 17 Malaysia 14
Papua New Guinea 10 Dominican Republic 16 Sweden 11
Angola 9 Seychelles 15 South Korea 7
Maldives 9 Russia 14 Hong Kong 5
Verde 9 Thailand 14 Switzerland 5
Ghana 8 El Salvador 12 Lithuania 3
Madagascar 8 China 11 Estonia 2
Egypt 7 Colombia 11 Finland 2
Mozambique 7 Tunisia 11 Latvia 2
Lebanon 6 Samoa 10 Austria 1
Guinea 5 Vietnam 10
Algeria 4 Cyprus 8
Liberia 4 Guatemala 7
Sao Tome And Principe 4 Dominica 6
Tog o 4 Fi j i 6
Cameroon 3 Malta 5
Cote d’Ivoire 3 Oman 4
Gambia 3 Brunei Darussalam 3
Kenya 3 Saint Lucia 3
Myanmar 3 Albania 2
Nigeria 3 Tonga 2
Sierra Leone 3 Vanuatu 2
Tanzania 3 Jamaica 1
Benin 2 Kuwait 1
Gabon 2 Qatar 1
Grenada 2 Trinidad And Tobago 1
Haiti 2 Ukraine 1
Kiribati 2
Palau 2
Rep Congo 2
Bangladesh 1
Belize 1
Cambodia 1
Honduras 1
Iran 1
Saint Kitts And Nevis 1
Saint Vincent And T.. 1
Solomon Islands 1
Timor-Leste 1
Zimbabwe 1
Tot a l 78 5 77 0 63 1 96 4
Table B.1: Countries and wave count by World Bank Doing Business categories.
36
WB WGI Pol Stab - very low WB WGI Pol Stab - low WB WGI Pol Stab - moderate WB WGI Pol Stab - high
Country Waves Country Waves Country Waves Country Waves
South Africa 207 Brazil 291 France 290 Portugal 162
Indonesia 136 Spain 182 United Kingdom 149 Japan 118
Mexico 92 Greece 50 Italy 113 New Zealand 114
Peru 7 6 Ecua dor 4 7 Pu erto R ico 53 Irela nd 58
Morocco 43 Argentina 34 Chile 50 Canada 38
Philippines 29 Panama 27 Costa Rica 50 Netherlands 32
Israel 25 Reunion 21 Namibia 17 Uruguay 24
Venezuela 24 Nicaragua 20 Seychelles 15 Germany 22
Senegal 20 Dominican Republic 16 Malaysia 14 Barbados 19
Sri Lanka 18 Vietnam 10 Belgium 12 Taiwan 18
Turkey 18 Bulgaria 8 Verde 9 Denmark 15
Russia 14 South Korea 7 Cyprus 8 Saint Martin 15
Thailand 14 Sao Tome And Principe 4 Fiji 6 Iceland 14
India 13 Gabon 2 Oman 4 Mauritius 14
El Salvador 12 Benin 2 Croatia 3 Micronesia 11
China 11 Belize 1 Albania 2 Sweden 11
Colombia 11 Trinidad And Tobago 1 French Guiana 2 Bahamas 10
Tunisia 11 Kuwait 1 Kiribati 2 Samoa 10
Papua New Guinea 10 Jamaica 1 Latvia 2 Maldives 9
Angola 9 Cambodia 1 Vanuatu 2 United Arab Emirates 9
Ghana 8 Saint Kitts And Nevis 1 Aruba 7
Madagascar 8 Solomon Islands 1 Dominica 6
Egypt 7 Hong Kong 5
Guatemala 7 Malta 5
Mozambique 7 Switzerland 5
Lebanon 6 Polan d 4
Guinea 5 Virgin Islands, U.S. 4
Algeria 4 Brunei Darussalam 3
Liberia 4 Lithuania 3
Tog o 4 Saint Lucia 3
Cameroon 3 Anguilla 2
Cote d’Ivoire 3 Bermuda 2
Gambia 3 Estonia 2
Kenya 3 Finland 2
Myanmar 3 Grenada 2
Nigeria 3 Tong a 2
Sierra Leone 3 Aust ria 1
Tanzania 3 Qatar 1
Haiti 2 Saint Vincent And T.. 1
Rep Congo 2
Bangladesh 1
Honduras 1
Iran 1
Somalia 1
Timor-Leste 1
Ukraine 1
Zimbabwe 1
Total Waves 888 726 805 783
Table B.2: Countries and wave count by WB Worldwide Governance Indicators, Political Sta-
bility categories
37
C Excluding certain wave types
i.
ii.
iii.
Figure C.1: Robustness test: excluding i. reefs, ii. points and iii. both from our analysis
of illumination in the closest town.
38
i.
ii.
iii.
Figure C.2: Robustness test: excluding i. reefs, ii. points and iii. both from our analysis
of illumination in the largest nearby town.
39
D Alternative baseline
Figure D.1: Effect of waves on illumination within surrounding 1km, 10km and 50km,
with quality 1 as baseline.
40
Figure D.2: Effect of waves on illumination at various distance buckets, quality 1 as
baseline.
41