Review on Energy Resilience

profileharsh55
2018-07-05_White_Paper_BigData_EN.pdf

BIG DATA AND PREDICTIVE ANALYTICS IN RENEWABLE ENERGY INVESTMENT

BIG DATA AND PREDICTIVE ANALYTICS IN RENEWABLE ENERGY INVESTMENT

© 2018 AQUILA CAPITAL – FOR PROFESSIONAL INVESTORS ONLY2

Executive summary �� Predictive�Analytics�and�Big�Data�are�currently�two�significant�

disruptive�forces�across�the�global�economy,�a�trend�which�

looks�set�to�continue�for�the�foreseeable�future.�Organisations�

that�are�sophisticated�enough�to�leverage�these�tools�are�devel-

oping�acute�competitive�advantages�based�on�their�nuanced�

insights�and�accurate�forecasts.

�� The�implementation�of�a�strategy�incorporating�Big�Data�or�

Predictive�Analytics�poses�several�technical�challenges�that�

need�to�be�reviewed�in�detail�to�build�out�a�suitable�infrastruc-

ture�to�reliably�grow�and�utilise�the�vast�reserves�of�data�

required�for�actionable�business�insights.

�� Within�Renewable�Energy�Investment,�the�use�of�Predictive�

Analytics�has�not�been�as�widespread�as�in�other�industries.�

One�reason�for�this�could�be�due�to�the�lack�of�large�datasets�

available�to�analyse.�Despite�this,�there�are�currently�applica-

tions�of�predictive�maintenance,�asset�performance�manage-

ment�and�automated�deal�sourcing�all�leveraging�different�

analytical�tools�and�sources�of�data.�Embracing�these�trends�

will�allow�investment�funds�to�refine�their�processes�and�boost�

investor�returns,�providing�a�key�differentiating�factor�between�

competitors.

�� As�data�availability�is�expected�to�improve�in�the�future�it�is�

advisable�for�organisations�to�develop�their�competencies�in�

this�area�early�so�they�can�begin�answering�their�most�crucial�

business�questions�and�locating�their�main�performance�drivers�

to�maintain�their�competitive�positions.

Author:

Saul Butt Alternative�Investment�Associate [email protected]

Lars Meisinger Head�|�Sales�Management�&�Business�Development [email protected]

1. Big Data – Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . p.�3

2. Big Data – Technical Considerations . . . . . . . . . . . . . . . . p.�5

3. Predictive Analytics – Overview . . . . . . . . . . . . . . . . . . . p.�7

4. Predictive Analytics – Technical considerations . . . . . . . p.�10

5. Big Data and Predictive Analytics – current applicability within Renewable Energy . . . . . . . . . . . . . p.�11

6. Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p.�15

BIG DATA AND PREDICTIVE ANALYTICS IN RENEWABLE ENERGY INVESTMENT

© 2018 AQUILA CAPITAL – FOR PROFESSIONAL INVESTORS ONLY 3

The�increasing�democratisation�of�data�and�technology�is�making�

Predictive�Analytics�and�Big�Data�pertinent�themes�for�future�busi-

ness�growth.�They�are�becoming�significant�disruptive�forces�that�

are�changing�the�fortunes�and�strategies�of�organisations�through-

out�the�economy.�Fundamentally,�they�allow�organisations�to�

incorporate�an�adaptive�quantitative�basis�to�their�decision-making,�

which�is�crucial�in�helping�firms�maintain�strong�competitive�posi-

tions�in�an�increasingly�complex�and�dynamic�world.�Conversely,�

failing�to�develop�competencies�in�this�area�could�result�in�an�organ-

isation�operating�at�an�informational�disadvantage�to�competitors.�

In�order�to�appropriately�leverage�these�tools,�it�is�imperative�that�

business�leaders�have�a�thorough�understanding�of�how�Big�Data�

and�Predictive�Analytics�will�impact�their�respective�industries�and�

prepare�accordingly.

The�use�of�Big�Data�and�Predictive�Analytics�is�highly�flexible,�with�

applicability�to�almost�any�business�context.�However,�the�question�

of�Big�Data�and�Predictive�Analytics’�applicability�in�Renewable�Energy�

Investment�is�a�novel�and�challenging�one.�This�brief�paper�will�look�

to�introduce�the�two�concepts�of�Big�Data�and�Predictive�Analytics�

independently�before�attempting�to�describe�their�combined�utility�

within�the�renewable�energy�space.�The�sections�below�will�initially�

offer�a�conceptual�overview�of�Big�Data�and�Predictive�Analytics�

interspersed�with�examples�of�business�applications�before�progress-

ing�to�more�technical�information�concerning�the�challenges�posed�

by�their�implementation.

1. Big Data – Overview

Big�Data�is�a�term�used�to�describe�the�vast�scale�of�new�datasets�

being�assembled�by�organisations.�Formally,�there�remains�no�con-

sensus�regard�ing�what�exactly�is�classified�as�big�data,�as�the�notion�

of�scale�in�relation�to�data�is�continuously�changing.�However,�a�

useful�rule�of�thumb�for�Big�Data�would�be�data�that�is�too�large�to�

store�or�practically�manage�within�an�Excel�spreadsheet.�

Data’s�exponential�growth�results�from�increasing�levels�of�digitali-

sation�and�internet�connectivity.�IBM�estimates�that�90%�of�data�in�

existence�was�created�in�the�past�two�years1.�This�has�coincided�with�

ever�more�powerful�computer�hardware�and�distributed�storage�

technology�(i.e.�The�Cloud),�which�has�followed�Moore’s�Law�of�

doubling�in�power�every�two�years.�This�has�provided�organisations�

with�the�foundation�to�begin�measuring�and�storing�vast�arrays�of�

data�points�on�a�scale�unseen�in�the�past.�

The�intricacy�and�impact�of�insight�derived�from�algorithmic,�statis-

tical�and�visualisation�tools�have�grown�significantly�with�the�arrival�

of�Big�Data.�Due�to�the�value�of�these�insights�many�organisations�

are�attempting�to�adopt�a�more�data-centric�strategy.�However,�the�

process�of�developing�a�Big�Data�initiative�poses�several�challenges,�

which�require�careful�technical�planning�to�overcome.��

Growing volume of digital data

Source: https://www.emc.com/leadership/digital-universe/2014iview/executive-summary.htm

1 https://www.ibm.com/developerworks/community/files/form/anonymous/api/library/054c2ab9-ea33-4c70-b0c6-b5bb2482a098/document/7de665ff-2327-41a8-b7b0-5f0b- ba97356f/media/BIG%20DATA%20%2B%20MAINFRAME.pdf

2013

4.4�ZB

44�ZB

The�digital�universe�is�huge� –�And�growing�exponentially

2020

If�the�Digital�Universe�were�represented�by�the�memory�in�a�stack�of�tablets,�� in�2013�it�would�have�stretched�two�thirds�of�the�way�to�the�Moon�

By�2020,�there�would�be�6.6�stacks�from�the�Earth�to�the�Moon

BIG DATA AND PREDICTIVE ANALYTICS IN RENEWABLE ENERGY INVESTMENT

© 2018 AQUILA CAPITAL – FOR PROFESSIONAL INVESTORS ONLY4

A�further�extension�to�this�framework�similarly�considers�the�value�

associated�with�data.�This�is�a�function�of�both�its�underlying�quality�

and�what�insights�the�data�can�bring.

How has Big Data been used to create value? Examples�of�Big�Data�acting�as�a�source�of�value�creation�can�be�

seen�with�online�platforms�such�as�Facebook,�which�have�the�ability�

to�consolidate�large�datasets�and�develop�acute�insights�into�their�

user-base�to�leverage�and�support�sales�propositions.�More�gener-

ally,�having�large�datasets�has�enabled�analysts�to�better�record�the�

past�and�delve�into�complex�interactions�and�systems,�such�as�those�

between�companies,�humans�and�markets.�

Source: http://www.ibmbigdatahub.com/infographic/four-vs-big-data

Volume The�scale�or�quantity�of�data.�Increasing�the�amount�of�data�allows�for�more�accurate�modelling�and�insight�into� business�questions.�However,�more�data�also�raises�significant�obstacles�to�ordering,�processing�and�storing� information�efficiently.

Velocity The�speed�or�frequency�at�which�data�is�created.�Greater�internet�speeds�and�digital�connectivity�have�allowed� for�real-time�data�streaming.

Variety The�different�types�of�data�being�reviewed.�New�techniques�have�allowed�information�to�be�obtained�from�text,� speech�and�satellite�images.�

Veracity The�accuracy�or�quality�of�the�data.�The�value�of�large�datasets�will�be�severely�undermined�if�the�quality�of�the� data�is�poor.

Some of the characteristics of Big Data are summarised in the four ’Vs’:

Case study – Applications of Big Data Investment Management:�It�was�recently�reported�that�Quan- titative�Hedge�Funds�will�surpass�USD�1tn�AUM2.�Automated�

data-driven�solutions�are�more�scalable�and�accurate�than�tradi-

tional�discretionary�fund�strategies.�The�attainment�of�these�

achievements�is�based�on�the�ability�to�operate�flexibly�within�a�

complex�environment�and�transform�vast�amounts�of�data�into�

accurate�investment�signals.�

Credit Scoring:�A�range�of�data�has�allowed�for�increasingly�reli- able�credit�scoring.�As�opposed�to�assembling�a�credit�score�based�

solely�on�individuals’�past�financial�behaviour,�many�organisations�

are�tapping�into�alternative�Big�Data�sources�such�as�social�media�

accounts.�The�facets�of�one’s�social�media�account�are�often�

indicative�of�personal�behaviour�traits,�which�can�be�highly�inform-

ative�of�future�actions�and�credit-worthiness.�This,�among�other�

applications,�has�made�the�provision�of�social�media�data�a�highly�

profitable�business.�

Healthcare:�The�advances�in�healthcare�analytics�over�the�coming� decades�will�be�phenomenal.�The�quantity�of�data�being�created�

through�healthcare�apps�and�by�consolidating�healthcare�records�

are�providing�unparalleled�insight�into�individuals’�lifestyles�and�

wellbeing.�This�is�being�investigated�in�regards�to�individuals’�

health�and�has�created�the�opportunity�to�forecast�the�probabil-

ity�of�contracting�specific�diseases�and�to�personalise�preventative�

healthcare�treatments�to�best�mitigate�them.

2 https://www.ft.com/content/ff7528bc-ec16-11e7-8713-513b1d7ca85a

BIG DATA AND PREDICTIVE ANALYTICS IN RENEWABLE ENERGY INVESTMENT

© 2018 AQUILA CAPITAL – FOR PROFESSIONAL INVESTORS ONLY 5

2. Big Data – Technical considerations

What are the challenges with managing Big Data? Managing�large�databases�efficiently�is�a�complex�task�and�often�

requires�specialist�knowledge.�Some�of�the�common�issues�associ-

ated�with�managing�database�structures�are:

‘Digital�exhaust’�is�a�term�often�used�to�describe�data�created�by�firms�

as�a�by-product�of�their�operations.�The�challenge�of�data�storage�is�

amplified�for�organisations�that�produce�significant�amounts�of�digital�

exhaust�and�wish�to�store�it�for�future�analysis.�In�this�case,�it�is�required�

to�have�a�detailed�and�forward-looking�data�architecture�such�that�

new�data�sets�can�be�seamlessly�integrated�into�an�existing�system.�

Without�an�appropriate�architecture�it�is�likely�that�the�ability�to�effi-

ciently�store�and�access�data�will�be�constrained�in�the�future.

What are the common approaches to data storage? In�most�circumstances,�relational�databases�are�generally�suitable�

for�data�storage�requirements.�This�structure�allows�information�to�

be�maintained�in�an�efficient�and�easy-to-recover�format.�The�

foundation�of�this�format�is�based�on�data�inputs,�tables�and�rela-

tionships�between�the�tables.�The�core�technology�behind�this�is�

typically�a�free�and�well-established�SQL�structure�that�can�be�easily�

replicated�and�stored�in�the�cloud�to�improve�accessibility.�In�addi-

tion�to�this�there�are�corporate�services�that�provide�customer�support�

and�guarantee�efficient�design3,4,5.

In�the�case�of�Big�Data,�more�novel�storage�methods�are�often�based�

on�NoSQL,�such�as�MongoDB.�This�does�not�rely�on�the�traditional�

relational�structures�between�data�tables�and�allows�for�a�far�quicker�

and�more�dynamic�way�to�store�and�retrieve�data.�This�approach�is�

generally�built�on�flexible�JSON�inputs;�however,�this�data�solution�

is�likely�to�be�excessively�complex�for�most�organisations�without�

vast�data�reserves.

Early stage data strategies When�developing�a�strategy�transitioning�towards�more�empirical�

decision-making,�organisations�often�already�possess�large�quanti-

ties�of�highly�valuable�information�internally.�Usually,�this�information�

has�not�been�appropriately�stored�or�formatted�in�a�useable�form.�

In�order�to�make�this�information�accessible�for�analysis,�collating�

the�data�within�a�centralised�data�warehouse�is�a�key�step�in�begin-

ning�an�initiative�of�data-driven�decisions�and�automation.�

The�standardisation�of�information�is�fundamental.�It�is�estimated�

that�data�analysts�spend�around�80%�of�their�time�cleaning,�struc-

turing�and�reviewing�data,�with�the�time�demands�of�applying�

value-adding�analytics�being�relatively�trivial6.�Having�data�stored�in�

a�standardised�format�would�greatly�improve�the�ability�to�provide-

broad-based�and�detailed�analytics.�Furthermore,�with�data�easy�to�

access,�the�automation�of�tasks�such�as�reporting�and�performance�

indexing�becomes�a�far�more�feasible�proposition.�

In�addition�to�this,�information�concerning�an�organisation’s�past�per-

formance�will�often�prove�to�be�highly�valuable�in�developing�an�

understanding�of�its�core�performance�drivers.�Therefore,�prior�to�

delving�into�acquiring�exotic�data�sets,�it�is�cost-effective�to�collect�the�

low-hanging�fruit�provided�by�collating�internally-stored�data.�

Advanced stage data strategies Many�organisations�do�not�possess�‘Big�Data’,�which�can�put�them�at�

a�competitive�disadvantage.�If�an�organisation�already�owns�a�devel-

oped�data�architecture,�one�option�to�overcome�this�would�be�to�

acquire�or�lease�external�datasets�to�derive�further�insights.�A�compli-

cation�with�this�approach�is�that�the�valuation�of�datasets�is�a�novel�

and�developing�field7,8.�It�is�often�difficult�to�estimate�where�and�how�

much�value�will�be�created�by�a�specific�dataset�prior�to�its�acquisition,�

which�makes�the�market�pricing�of�data�highly�subjective.9

Data� Formatting

The�correct�formatting�of�data�is�required�to� avoid�data�duplication,�which�can�cause�infor- mation�to�grow�unnecessarily�large.�Further,�the� speed�at�which�data�can�be�accessed�and� processed�will�be�depends�on�its�format.�

Storage�space Organisations�need�to�decide�how�to�store�data,� whether�through�acquiring�servers�in-house�or� leasing�server�space�from�external�organisations.� Both�of�these�options�have�different�associated� cost�structures�and�benefits.

Processing� power

To�access�or�process�Big�Data,�a�substantial� amount�of�computational�power�is�required.� This�can�either�come�from�more�powerful� computers,�which�will�suffice�up�to�a�point.� However,�distributed�systems�are�far�more� scalable�and�increasingly�affordable.

Data�quality The�volume�of�Big�Data�is�presently�a�significant� point�of�discussion�in�strategy�development.� However,�if�the�quality�of�the�assembled�data�is� poor,�then�its�value�will�be�substantially� diminished.�

Data�security There�is�a�trade-off�in�providing�employees�with� significant�accessibility�of�corporate�data�in�a� centralised�format�and�the�potential�severity�of�a� data�breach.�It�is�imperative�that�data�is�stored� using�secure�servers,�strong�passwords�and� advanced�level�encryption�to�avoid�the�loss�of� customer�and�business�data.

Source: http://pwc.blogs.com/analytics_means_business/2017/09/big-data-big-deal.html

3 http://www.oracle.com/technetwork/topics/entarch/oracle-wp-big-data-refarch-2019930.pdf 4 https://tech.winton.com/2017/03/data-technologies-at-winton/ 5 https://tech.winton.com/2017/09/creating-a-scalable-data-ingestion-process/ 6 www.forbes.com/sites/gilpress/2016/03/23/data-preparation-most-time-consuming-least-enjoyable-data-science-task-survey-says 7 https://hbr.org/2016/09/do-you-know-what-your-companys-data-is-worth 8 https://svds.com/valuing-data-is-hard/ 9 https://sloanreview.mit.edu/article/whats-your-data-worth/

BIG DATA AND PREDICTIVE ANALYTICS IN RENEWABLE ENERGY INVESTMENT

© 2018 AQUILA CAPITAL – FOR PROFESSIONAL INVESTORS ONLY6

Opportunities to develop datasets It�is�possible�to�grow�an�organisation’s�data�resources�without�acquir-

ing�datasets�from�third-party�providers�by�scraping�the�internet�for�

information.�This�approach�allows�firms�to�grow�their�datasets�by�

focusing�on�the�key�data�they�require�for�their�decision-making.�

Some�tools�to�simplify�the�web-scraping�process�are�import.io�and�

mozenda.com.�

Combining datasets and looking externally Novel�insights�have�often�been�found�by�combining�multiple�data-

sets�and�uncovering�new�correlations.�Considering�Renewable�Energy�

investments,�there�are�several�useful�external�datasets�that�provide�

market�data�that�could�be�utilised.�Some�examples�include:�

Take away Big�Data�is�changing�organisations�and�is�a�key�source�of�com-

petitive�advantage.�It�is�providing�the�foundations�for�more�

empirically-based�decision-making.

The�storage�and�management�of�growing�amounts�of�data�will�

often�require�significant�planning�and�investment�to�develop�and�

maintain�an�effective�data�infrastructure.

For�organisations�beginning�to�transition�towards�empirically-as-

sisted�decision-making,�a�robust�centralised�database�and�processing�

historic�operational�data�will�often�provide�significant�value.

0� 5� 10� 15� 20� 25� 30� 35� 40

Obstacles to use of alternative data by asset managers and hedge funds

Prohibitively�high�fees

Internal�procurement�processes�are�too�cumbersome/slow

Lack�of�time�needed�to�evaluate�data

Management�not�convinced�of�data’s�value

Difficulty�with�uncustomised�datasets

Data�not�compatible�with�analysis�systems

Human�capital�needed�for�integration�not�available

There�are�currently�no�real�obstacles

per cent

Renewables. ninja (weather�data)

Provides�hourly�forecasts�of�weather�conditions� around�the�globe.

OPEN.ei� (Energy� Production)

A�range�of�free�datasets�concerning�energy�and� renewables.�The�information�is�peer-verified�to� establish�the�quality.

Energy�Demo� (Solar�Energy� Production)

Open�source�data�providing�the�location�of� numerous�solar�producing�assets.

Quandl�(Alter- native�Data)

A�combination�of�free�and�paid�datasets,�this�is� perhaps�the�most�established�and�comprehen- sive�source�of�alternative�investment� information.

Source: https://www.ft.com/content/d86ad460-8802-11e7-bf50-e1c239b45787

BIG DATA AND PREDICTIVE ANALYTICS IN RENEWABLE ENERGY INVESTMENT

© 2018 AQUILA CAPITAL – FOR PROFESSIONAL INVESTORS ONLY 7

3. Predictive Analytics – Overview

Analytics�is�broadly�defined�as�the�systematic�processing�of�data�to�

generate�insight.�The�forms�of�analytics�can�be�categorised�into�the�

following�branches:

Forms of analytics Descriptive analytics:�the�most�commonly�applied�form�of�ana- lytics.�It�is�used�to�provide�insight�into�what�has�happened�in�the�

past.

Diagnostic analytics:�seeks�to�understand�why�particular�events� occurred.�This�is�often�based�on�testing�correlations�between�

various�relevant�variables.

Predictive analytics:�attempts�to�develop�forecasts.�Probability� and�statistics�are�employed�to�understand�the�likelihood�of�an�

outcome�occurring�based�on�past�information.�This�sees�the�com-

plexity�of�predictive�analytics�rising�significantly�relative�to�earlier�

stages�of�analytics.

Prescriptive analytics:�attempts�to�estimate�the�best�responses� to�forecasted�events.�When�operating�within�a�stochastic�context�

this�requires�complex�computational�capabilities�and�will�be�based�

on�expected�value�calculations.�The�benefit�of�accurately�model-

ling�best�response�decisions�can�be�significant.�This�form�of�

analytics�is�primarily�dominated�by�Machine�Learning�and�Artifi-

cial�Intelligence.�

Predictive�Analytics�are�changing�the�way�that�organisations�approach�

decision-making.�They�allow�firms�to�not�just�look�at�the�past�and�

evaluate�their�performances�but�to�forecast�what�will�likely�happen.�

The�tools�available�to�apply�predictive�analytics�are�extremely�broad�

and�flexible,�which�makes�defining�a�clear�business�problem�crucial�

to�identifying�the�appropriate�solution.�When�defining�a�business�

problem,�it�is�often�useful�to�go�through�each�stage�in�the�above�

graph.�Additionally,�incorporating�visualisation�tools�into�the�analyt-

ics�process�can�help�to�quickly�communicate�quickly�the�narrative�

behind�a�problem�and�develop�actionable�insights.�

V al

ue

Stages of analytical analysis

Complexity

Hindsight Insight Foresight

Information Optimisation

Descriptive

What�has�� happened?

Diagnostic

Why�has�this� happened?

Predictive

What’s�likely�to� happen?

Prescriptive

How�best�to�act�in� the�future?��

BIG DATA AND PREDICTIVE ANALYTICS IN RENEWABLE ENERGY INVESTMENT

© 2018 AQUILA CAPITAL – FOR PROFESSIONAL INVESTORS ONLY8

What research questions can Predictive Analytics be applied to? 1) Categorisation�–�Programs�look�to�decipher�whether�an�obser- vation�belongs�in�one�category�or�another.�An�example�of�this�would�

be�whether�to�categorise�a�stock�as�a�buy�or�a�sell,�based�on�its�

fundamentals.

2) Numerical estimation –�This�type�of�estimation�is�generally� slightly�more�complex,�where�a�program�will�be�used�to�quantify�an�

outcome.�An�example�of�this�would�be:�based�on�a�stock’s�changing�

value,�what�is�the�probability�that�it�will�continue�to�rise�or�fall?

3) Correlations�–�Models�looking�to�uncover�correlations�are�often� very�useful�in�understanding�the�interrelatedness�of�variables�within�

a�complex�environment.�

What is the interaction between Big Data and Predictive Analytics? Big�Data�has�been�integral�in�developing�the�accuracy�and�efficacy�

of�predictive�analytics.�This�has�been�achieved�through�the�breadth�

and�depth�of�data�now�available.�The�increasing�digitalisation�of�

society�has�led�to�a�plethora�of�new�data�points�by�which�to�refine�

the�modelling�of�complex�relationships.�This�much�broader�array�of�

information�that�can�be�incorporated�into�models�has�shed�new�light�

on�desired�research�topics.�Further,�it�is�clear�that�having�more�com-

plete�population�data�will�allow�a�model�to�more�accurately�replicate�

reality.�Therefore,�as�the�depth�and�coverage�of�Big�Data�increases,�

the�reliability�of�models�will�likely�improve.

Common techniques applied in Predictive Analytics

Regressions Regressions�provide�insights�into�how�highly�correlated�variables�are�to�a�particular�outcome�or�event.�Once�estab- lished,�the�model�inputs�can�be�adjusted�to�understand�the�expected�outcome�in�different�scenarios.

Time�Series�Models This�is�an�extension�to�a�simple�regression�model,�with�the�variable�of�time�considered�in�order�to�understand�how� trends�will�develop�and�progress.�This�is�key�to�creating�insightful�forecasts.

Geospatial�Models This�is�an�extension�to�a�simple�regression�model,�with�the�variable�of�space�considered�to�show�how�trends� develop�in�areas.�This�is�a�particularly�powerful�tool�when�the�element�of�time�is�additionally�considered.

Machine�Learning Utilises�statistical�methods�along�with�computer�science�principles�to�allow�programs�to�learn�independently.�The� algorithms�developed�for�machine�learning�are�useful�in�uncovering�patterns�within�data.�In�the�machine�learning� space�there�is�a�plethora�of�tools�available�of�differing�complexity.�The�most�basic�models�tend�to�be�linear,�while� more�advanced�methods�can�model�non-linear�relationships.

Artificial�Intelligence These�are�programs�that�are�often�more�advanced�than�machine�learning�and�look�to�learn�independently�and� respond�to�environmental�stimuli.�These�programs�can�be�extraordinarily�useful�for�prescriptive�analytics�when�run� in�simulations.�Common�tools�are�reinforcement�learning�and�artificial�neural�networks.

BIG DATA AND PREDICTIVE ANALYTICS IN RENEWABLE ENERGY INVESTMENT

© 2018 AQUILA CAPITAL – FOR PROFESSIONAL INVESTORS ONLY 9

Case Study – Innovative Uses of Predictive Analytics Asset�maintenance:�Rolls�Royce�operates�within�the�aviation�main-

tenance�business,�through�leasing�jet�engines.�As�opposed�to�

simply�leasing�the�assets,�they�have�augmented�their�service�by�

providing�advanced�predictive�analytics�based�on�data�collected�

across�all�their�jet�engines.�This�allows�Rolls�Royce�to�forecast�

when�their�customers’�engines�require�maintenance,�which�lowers�

repair�costs�and�downtime.�This�new�business�model�structure�

would�not�be�possible�without�the�developments�in�storing�and�

forecasting�using�large�datasets.�

Pricing�Forecasts:�Kayak,�an�online�travel�agent,�provides�fore-

casts�for�customers�to�identify�the�ideal�time�to�purchase�flight�

tickets�at�the�lowest�prices.�This�has�been�achieved�through�col-

lating�a�vast�database�from�airlines’�daily�ticket�pricing�and�

uncovering�pricing�patterns.

Risks associated with Predictive Analytics There�are�a�number�of�risks�associated�with�implementing�predictive�

models,�with�a�lack�of�training�data�and�transparency�in�black�box�

models�being�two�key�issues.�Poor�or�inadequate�data�being�used�

to�fit�a�model�will�often�result�in�forecasts�that�misrepresent�reality.�

A�means�to�overcome�this�is�to�include�more�predictive�variables.�

However,�this�can�lead�to�overfitting�and�increasing�model�complex-

ity,�making�it�extremely�difficult�to�interpret�the�fundamental�reasoning�

behind�a�model’s�output.�

Additionally,�analytical�models�are�particularly�accurate�in�stable�

environments.�However,�when�there�are�fundamental�changes�in�

what�a�model�is�measuring,�there�is�no�way�to�accurately�account�

for�this.�This�makes�it�important�to�maintain�models�with�current�

data�and�understand�when�to�disregard�their�output�to�avoid�falla-

cious�assumptions.�

A u

to m

o ti

ve

M an

u fa

ct u

ri n

g

C o

n su

m er

Fi n an

ce

A g

ri cu

lt u

re

En er

g y

H ea

lt h

ca re

Ph ar

m ac

eu ti

- ca

ls

Pu b

lic /s

o ci

al

M ed

ia

Te le

co m

Tr an

sp o

rt �a

n d �

lo g

is ti

cs

Real�time�optimisation

Strategic�optimisation

Predictive�Analytics

Predictive�maintenance

Radical�personalisation

Discover�new�trends/anomalies

Forecasting

Process�unstructured�data

Machine learning has the potential to be applied across many industries

Impact potential Low High

Source: https://www.mckinsey.com/~/media/McKinsey/Business%20Functions/McKinsey%20Analytics/Our%20Insights/The%20age%20 of%20analytics%20Competing%20in%20a%20data%20driven%20world/MGI-The-Age-of-Analytics-Full-report.ashx

BIG DATA AND PREDICTIVE ANALYTICS IN RENEWABLE ENERGY INVESTMENT

© 2018 AQUILA CAPITAL – FOR PROFESSIONAL INVESTORS ONLY10

4. Predictive Analytics – Technical considerations

Challenges of establishing Predictive Analytics For�analytics�to�be�valuable��there�must�be�a�suitable�quantity�and�

variety�of�data�to�process.�Without�such�data�it�is�impossible�to�gen-

erate�statistical�analyses�that�have�the�required�accuracy�or�reliability�

upon�which�to�base�business�decisions.

It�is�also�necessary�to�invest�in�hiring��appropriate�talent:�an�analyst�

must�have�the�right�technical�capabilities,�along�with�a�strong�math-

ematical�and�computer�science�background.�Sector�knowledge�is�

crucial�to�identify�the�key�business�issues�and�to�structure�research�

accordingly.

Large�technology�companies�such�as�Google�and�Facebook�are�con-

tinuously�publishing�open-source�software�libraries�that�enable�

analysts�to�conduct�ever�more�advanced�computations.�Additionally,�

the�existing�software�is�being�constantly�developed�to�make�them�

more�user-friendly.�These�two�trends�mean�it�is�easier�for�organisa-

tions�to�undertake�valuable�analytics.

What are the Common Tools Utilised in Predictive Analytics?

Python Python�is�perhaps�the�programming�language�that�is�most�synonymous�with�Data�Science.�It�includes�the�most� advanced�packages�available�for�Machine�Learning�and�has�strong�data�management�tools.�The�flexibility�of�the� language�allows�for�efficient�algorithms�to�be�constructed.

R Similarly�to�Python,�R�is�a�high-level�programming�language.�It�has�extremely�advanced�statistics�and�visualisation� libraries.

Tableau Allows�users�to�quickly�create�elegant�visualisations�and�conduct�data�analysis�to�understand�the�stories�within� large�datasets.

SQL As�previously�discussed,�SQL�is�a�database�software.�The�language�is�declarative,�meaning�that�it�will�select�the� optimal�way�to�execute�a�task,�making�it�an�efficient�data-querying�language.

Jupyter�Notebooks A�reporting�technology�that�allows�analysts�to�incorporate�code,�text�and�graphs�together.�This�allows�for�trans- parency�in�methodology�and�encourages�collaboration.

GitHub An�online�code�repository.�This�allows�individuals�to�share�code�and�approaches�to�solving�problems.�This�is�the core�platform�to�collaborate�on�software�development�and�can�host�a�great�deal�of�open-source�software.

Azure� The�distributed�computational�platform�from�Microsoft.�It�is�a�user-friendly�means�by�which�to�parallelise�big� computational�tasks�over�large�datasets.�

LaTeX A�report�processing�language�that�allows�the�user�to�design�reports�with�significant�additional�flexibility�in relation�to�traditional�text�applications�such�as�Word.

Take away Predictive�Analytics�can�provide�potential�solutions�to�a�range�of�

business�problems.

Simple�diagnostic�analytics�using�visualisation�tools�can�Often�

create�substantial�value�in�understanding�business�problems.

The�accuracy�of�a�model’s�output�is�highly�correlated�to�the�quan-

tity�of�data�available�to�train�it.

BIG DATA AND PREDICTIVE ANALYTICS IN RENEWABLE ENERGY INVESTMENT

© 2018 AQUILA CAPITAL – FOR PROFESSIONAL INVESTORS ONLY 11

5. Big Data and Predictive Analytics – current applicability within renewable energy

The�use�of�Big�Data�in�the�management�of�financial�asset�classes�has�

produced�excellent�results�for�some�organisations.�These�successes�

has�been�based�on�leveraging�the�three�trends�of:�increasing�computer�

power,�data�proliferation�and�improving�analytical�methodologies10.�

Within�Renewable�Energy�investment,�the�use�of�analytics�has�been�

slightly�more�subdued�due�to�the�low�availability�of�data.�This�is�largely�

due�to�the�nature�of�the�asset�class�being�privately�held�with�a�high-

ly-fragmented�ownership�structure,�meaning�the�opportunities�to�

assemble�consolidated�market�data�is�extremely�limited.�Despite�the�

sparse�data�environment�within�the�Renewable�Energy�investment�

industry,�there�exists�opportunities�to�undertake�valuable�analytics.�

Asset Performance Management�-�APM�entails�collating�past�per- formance�information�from�a�portfolio�of�assets�and�uncovering�

correlations�between�certain�variables�and�a�target�metric,�such�as�

maintenance�spending�relative�to�profit.�Based�on�this,�a�forecast�

will�be�made�to�calculate�how�performance�can�be�adjusted�to�opti-

mise�on�the�target�metric.�This�is�often�a�balancing�act�in�looking�to�

manage�the�asset�conservatively,�while�also�attempting�to�maximise�

energy�output.�

This�is�often�an�ideal�research�project�for�organisations�with�a�large�

backlog�of�historic�data.�After�establishing�past�relationships,�pre-

dictive�analytics�methods�such�as�regression�analysis�can�be�applied�

to�find�the�optimal�conditions�to�run�a�project.�

Data�related�to�Asset�Performance�is�growing�significantly.�One�of�

the�main�drivers�of�this�has�been�the�incorporation�of�sensors�used�

to�measure�the�operating�efficiency�of�an�asset.�Once�data�has�been�

recorded�for�a�great�number�of�assets�it�can�be�gen�eralised�into�a�

model�and�used�for�forecasts.

APM�becomes�a�more�compelling�proposition�when�combining�past�

operational�data�with�alternative�datasets�to�hold�external�factors�

constant�and�understand�how�much�of�a�project’s�success�is�depend-

ent�on�the�asset�management’s�decision-making.�One�growing�form�

of�large�datasets,�which�could�be�significant�in�this�case,�is�weather�

data.�This�is�being�assembled�using�asset�sensors,�satellite�imaging�

and�independent�weather�stations.�It�provides�time�series�data�on�

progressing�weather�patterns,�which�can�then�be�used�to�measure�

the�impact�of�weather�on�asset�performance.�

10 https://www.man.com/the-rise-of-machine-learning 11 https://www.cio.com/article/3006300/big-data/weather-company-forecasts-more-big-data-for-ibm-watson-analytics.html 12 https://www.renewableenergyworld.com/articles/2014/02/the-interconnection-nightmare-in-hawaii-and-why-it-matters-to-the-u-s-residential-pv-industry.html

Case Study – General Electric GE�has�become�a�market�leader�in�asset�performance�manage-

ment�for�renewable�energy�production.�Through�using�the�

principles�of�the�Internet�of�Things�and�installing�sensors�with�

internet�connection�to�small�SQL�databases,�GE�can�assist�in�

assembling�personalised�information�for�energy�producers.�

On�top�of�GE’s�ability�to�assist�asset�owners�in�collecting�perfor-

mance�data,�they�have�similarly�used�their�scale�of�insight�in�

renewable�assets�to�provide�performance�and�optimisation�rec-

ommendations�to�increase�profitability�and�reduce�downtime.�

Case Study – Energy storage and market pricing Using�predictive�analytics�to�model�the�conditions�for�an�asset’s�

optimal�operating�efficiency�and�energy�output�is�a�valuable�tool�

in�improving�investor�returns.�However,�the�requirement�of�this�

service�will�become�far�greater�with�the�implementation�of�smart�

grids�and�storage�facilities.�In�order�to�avoid�oversupply�to�the�

grid�because�renewable�energy�sources�produce�simultaneously12,�

it�is�likely�that�batteries�will�be�used�increasingly�to�regulate�the�

supply�of�electricity.�In�order�to�achieve�this,�optimisation�models�

forecasting�the�appropriate�times�to�release�energy�into�the�grid�

will�become�a�significant�differentiator�in�the�profitability�of�

renewable�energy�assets.

Case study – IBM Weather�data�and�modelling�has�advanced�significantly�since�IBM’s�

recent�acquisition�of�The�Weather�Company,�which�provides�infor-

mation�to�Watsons�Artificial�Intelligence�platform.�The�system�is�

currently�collecting�“four�gigabytes�of�data�per�second�from�40�

million�mobile�phones,�147,000�sensors�in�weather�observing�sta-

tions,�as�well�as�50,000-plus�daily�flights�that�collect�data�on�

temperature,�turbulence�and�barometric�pressure.”�The�insights�

this�will�generate�in�collaboration�with�IBM’s�Artificial�Intelligence�

capabilities�will�allow�renewable�energy�producers�to�better�fore-

cast�energy�production�and�plan�maintenance�works11.�

BIG DATA AND PREDICTIVE ANALYTICS IN RENEWABLE ENERGY INVESTMENT

© 2018 AQUILA CAPITAL – FOR PROFESSIONAL INVESTORS ONLY12

Reliability-centred management & predictive maintenance – The�aviation�industry�initiated�the�preliminary�study�of�Reliability�

Centred�Management�(RCM)�in�the�1960s�to�combat�their�assets’�

escalating�O&M�costs.�This�resulted�in�a�broad�and�collaborative�

research�initiative�concerning�the�maintenance�and�durability�of�

mechanical�components�to�reduce�costs�associated�with�component�

defects,�replacement�and�asset�downtime.�The�resultant�adaptation�

to�maintenance�spending�was�to�allocate�expenditure�based�on�a�

component’s�relative�importance�to�the�asset�and�replacement�cost,�

which�was�developed�into�a�more�effective�weighted�maintenance�

spending�model.

13 https://www.coresystems.net/blog/the-difference-between-predictive-maintenance-and-preventive-maintenance

Traditionally,�manufacturers�would�assist�asset�operator�RCM�by�pro-

viding�a�recommended�preventative�maintenance�program�for�its�

components.�Specifically,�they�would�provide�suggestions�for�when�

to�undertake�maintenance�that�would�typically�be�based�on�time,�

events�or�meter�readings.�An�example�of�this�would�be�having�a�

wind�turbine�gearbox�serviced�every�18�months.�These�measures�

have�proven�highly�effective�in�reducing�maintenance�costs.�However,�

these�preventative�measures�are�generally�not�specific�to�the�context�

in�which�an�asset�is�used.�For�example,�a�wind�farm�in�Portugal�

exposed�to�extremely�hot�and�dry�weather�will�likely�face�different�

maintenance�issues�from�an�identical�asset�based�in�Sweden�with�

freezing�temperatures�and�high�precipitation.�

A�recent�development�in�RCM�that�can�account�for�the�variance�in�

assets’�maintenance�requirements�is�predictive�maintenance.�This�

approach�utilises�various�data�sources�to�predict�when�maintenance�

expenditure�is�required,�indicate�what�actions�to�undertake�and�fore-

cast�the�benefit�of�the�expenditure.�When�completing�predictive�

maintenance,�traditional�factors�such�as�the�age�of�components�can�

be�used.�However,�novel�factors�such�as�thermal�imaging�and�vibra-

tion�monitors�along�with�external�factors�such�as�weather�are�

becoming�increasingly�popular�to�develop�a�clear�picture�of�what�

issues�an�asset�is�likely�to�incur.�This�approach,�utilising�regression�

analysis�and�optimisation�models,�allows�for�specific�maintenance�

programs�to�be�created�for�each�component�and�will�flag�issues�prior�

to�a�major�defect�occurring,�resulting�in�better�allocation�of�O&M�

spending.�This�novel�method�can�combine�both�predictive�and�pre-

scriptive�analytics�with�the�benefits�of�this�approach�estimated�to�

be�over�30%�cheaper�than�traditional�preventative�measures13.

1990 1995 2000 2005 2010 2015 2020

140

120

100

80

60

40

20

0

N um

b er

o f

p ro

vi d

er s

Alternative data provision is a fast-growing niche

Source: https://www.bcg.com/publications/2017/principal-investors-private-equi- ty-fund-strategy-operations-digital-deal-sourcing-private-equity.aspx

BIG DATA AND PREDICTIVE ANALYTICS IN RENEWABLE ENERGY INVESTMENT

© 2018 AQUILA CAPITAL – FOR PROFESSIONAL INVESTORS ONLY 13

Maintenance strategies in an industrial context

Source: https://www2.deloitte.com/content/dam/Deloitte/de/Documents/deloitte-analytics/Deloitte_ Predictive-Maintenance_ PositionPaper.pdf

Reactive maintenance Assets�fail�before�being�maintained.

Preventive maintenance Systems�are�maintained�at�fixed�intervals�

to�ensure�continuous�availability.

Condition-based maintenance Systems�are�maintained�based�on�simple�

rules�using�equipment�information.

Predictive maintenance Systems�are�maintained�before�failure,�but�

run�as�long�as�possible�without�interruption.

�� no�analytics�or�sensors�needed

�� spare�parts�are�fully�used�up

�� no�analytics�or�sensors�needed

�� spare�parts�are�fully�used�up

+ -

�� limited�failure�forecasting�

accuracy

�� medium�risk�of�failures�remains

�� good�usage�of�equipment�life�

cycle

�� limits�unplanned�downtimes +

-

�� optimal�usage�of�equipment�

lifetime

�� minimises�unplanned�downtimes

�� upfront�investment�necessary

�� expert�knowledge�required

+

-

�� no�in-process�sensors�needed

�� reduces�unplanned�downtimes

�� waste�in�life�cycles�of�spare�parts

�� excessive�maintenance�and�

downtime

+

-

production downtime warning maintenance asset failure avoided asset failure

BIG DATA AND PREDICTIVE ANALYTICS IN RENEWABLE ENERGY INVESTMENT

© 2018 AQUILA CAPITAL – FOR PROFESSIONAL INVESTORS ONLY14

In�order�to�improve�the�validity�of�predictive�maintenance,�a�range�

of�historic�data�must�be�available�to�refine�forecasts�of�component�

failure.�One�means�to�achieve�this�has�been�to�encourage�industries�

to�centralise�their�data�such�that�each�market�participant�will�have�

a�wider�and�potentially�longer�backlog�of�historic�data.�T-Systems�

recently�indicated�that�even�small�degrees�of�corporate�collaboration�

through�sharing�data�on�a�single�component�can�yield�significant�

results14.�A�good�example�of�data�sharing�can�be�found�in�Norway,�

whereby�due�to�the�extensive�cross-ownership�of�hydroelectric�pro-

ducing�assets,�Energy�Norway�has�been�able�to�centralise�a�database�

of�between�90�–�140�TWh�of�Hydropower�production�data�a�year.�

With�this�vast�reserve�of�data,�a�research�project�called�MonitorX has�begun,�which�looks�to�use�machine�learning�and�artificial�intel-

ligence�in�collaboration�with�Big�Data�to�model�component�lifetime�

optimisation�based�on�a�variety�of�surrounding�conditions�and�uncer-

tainties15.�This�collaboration�includes�11�Norwegian�and�Swedish�

power�producers�and�two�suppliers�that�are�working�together�in�

testing�new�analytics�methodologies�to�better�evaluate�their�main-

tenance�expenditure�and�processes16.

Deal sourcing & due diligence�–�Combining�web-scraping�with� quantitative�analysis�methods�such�as�natural�language�processing��

has�made�it�possible�for�managers�to�translate�substantial�reserves�

of�information�online�into�insights�on�potential�acquisition�targets.�

With�web-scraping�programs�it�is�possible�to�trawl�the�internet�to�

find�specific�assets�that�have�been�mentioned�in�the�news�or�on�

social�media.�Once�an�asset�has�been�located,�relevant�information�

can�be�found�and�compared�against�a�set�of�criteria�specified�by�the�

investment�manager.�If�the�asset�meets�a�specific�set�of�requirements�

it�can�then�be�flagged�as�worthy�of�greater�research.�In�a�recent�BCG�

report,�the�importance�of�leveraging�proprietary�information�from�

web-scraping�and�data�analytics�to�guide�both�deal�sourcing�and�

due�diligence�were�highlighted�as�being�vital�to�avoid�being�at�a�per-

manent�information�disadvantage�to�competing�funds17.

Further applicable datasets Macroeconomic data�-�Gaining�insight�into�detailed�macroeco- nomic�data�can�be�a�key�driver�in�strategic�and�corporate�level�

decision-making.�Understanding�the�exact�point�at�which�one�stands�

in�the�business-cycle�can�be�highly�informative�as�to�the�impending�

economic�forces�to�which�a�business�and�its�clients�will�be�exposed.�

In�relation�to�this,�the�investment�manager�can�begin�preparing�the�

appropriate�investment�products�that�clients�will�require�over�sub-

sequent�months�and�years.�This�allows�the�fund�managers�to�make�

forward-looking�decisions�as�opposed�to�reacting�to�current�market�

trends.�Recently�there�have�been�a�number�of�innovations�in�mac-

ro-economic�forecasting�with�inflation�forecasts�being�run�in�real�

time�using�online�data�such�as�the�Billion�Price�Project��or�satellite�

imaging�to�infer�economic�activity�and�GDP�growth.�Leveraging�these�

tools�could�allow�managers�to�develop�more�nuanced�and�timely�

economic�indicators�to�support�their�decision-making.�

Case study – Private equity due diligence developments The�application�of�automated�due�diligence�through�assembling�

data�with�web-scraping�is�providing�a�significant�advantage�to�

tech-savvy�private�equity�funds.�According�to�a�recent�article18,�

the�information�found�through�trawling�the�internet�is�provid-

ing�new�insights�that�support�private�equity�funds�in�their�

increasingly�competitive�bidding�processes.�Additionally,�the�

automation�of�the�research�is�reducing�the�time�and�costs�asso-

ciated�with�typical�due�diligence�processes,�creating�a�leaner�

and�more�detailed�analysis.�

14 https://www.t-systems.com/de/en/industries/automotive/sales-aftersales-solutions/remote-support/predictive-maintenance-379704 15 https://www.energinorge.no/energiforskning/fornybar-energiproduksjon/vannkraft/maskin-elektro/pagaende-maskin-elektro/monitorx/ 16 https://www.energinorge.no/energiforskning/nyheter/2016/verktoy-for-a-unnga-havari/ 17 https://www.bcg.com/publications/2017/principal-investors-private-equity-fund-strategy-operations-digital-deal-sourcing-private-equity.aspx 18 https://www.forbes.com/sites/baininsights/2017/04/07/data-mining-your-way-to-better-due-diligence-in-private-equity/#6cda573c243e

BIG DATA AND PREDICTIVE ANALYTICS IN RENEWABLE ENERGY INVESTMENT

© 2018 AQUILA CAPITAL – FOR PROFESSIONAL INVESTORS ONLY 15

6. Summary

Big�Data�and�Predictive�Analytics�in�combination�are�providing�a�

com�petitive�edge�in�many�industries.�The�cost�of�accessing�data�is�

falling,�technology�is�ever�more�powerful�and�analytical�methods�

are�increas�ingly�sophisticated.�However,�the�scarcity�of�relevant�data�

in�the�renewable�energy�investment�industry�remains�the�limiting�

factor�in�the�application�of�further�predictive�analytics.

At�present,�the�renewable�energy�industry�can�develop�detailed�

descriptive�analytics�of�its�past�performance�by�combining�assets’�

operating�information�with�external�datasets�such�as�weather�con-

ditions.�However,�with�the�persistent�proliferation�of�available�

information�through�data�collection�technolo�gies�and�large�corpo-

rate�data�providers�it�is�expected�that�data�will�become�sufficiently�

abundant�in�the�near�future�to�support�more�elaborate�predictive�

analytics�methods.�This�seemingly�secular�trend�in�data�growth�will�

support�the�development�of�intriguing�new�insights�in�Renewable�

Energy�Investments�along�with�accurate�forecasts�of�asset�returns�

and�performance.�It�is�therefore�crucial�to�begin�devel�oping�the�req-

uisite�capabilities�to�manage�and�process�data,�such�that�Renewable�

Energy�Investment�funds�can�gain�early�insight�into�their�performance�

drivers�and�business�problems.�Once�these�key�variables�and�issues�

have�been�isolated,�organisations�can�undertake�deep�analyses�into�

these�topics�when�suitable�datasets�become�available.�This�will�

provide�a�platform�to�uncover�best�practices�and�maintain�superior�

decision-making�ahead�of�competitors.

BIG DATA AND PREDICTIVE ANALYTICS IN RENEWABLE ENERGY INVESTMENT

© 2018 AQUILA CAPITAL – FOR PROFESSIONAL INVESTORS ONLY16

For more information, please contact:

Aquila Group Valentinskamp�70 20355�Hamburg Germany Tel.:�� +49�(0)40�87�50�50-199 E-mail:�[email protected] Web:� www.aquila-capital.com

Follow us on

Hamburg�·�Frankfurt�·�London�·�Luxembourg�·�Madrid�·�Oslo�·�Zurich�·�Invercargill�·�Singapore�·�Tokyo�

Important notice:�Please�note�that�all�information�has�been�collected�and�examined�carefully�and�to�the�best�of�our�knowledge;�however,�the�information�is�provided� without�any�guarantee.�All�information�is�believed�to�be�reliable�but�we�are�not�able�to�warrant�its�completeness�or�accuracy.�This�document�does�not�constitute�an� offer�or�solicitation�for�the�purchase�or�sale�of�any�financial�instrument,�is�launched�for�informational�purposes�only�and�is�not�intended�to�provide�and�should�not�be� relied�on�for�accounting,�legal�or�tax�advice,�or�investment�recommendations.�Investors�are�alerted�that�future�performance�can�vary�greatly�from�past�results.�The� value�of�investments�and�income�from�them�may�rise�as�well�as�fall.�You�may�not�get�back�the�amount�initially�invested.�All�investments�involve�risks�including�the�risk� of�a�possible�loss�of�principal.�Emerging�markets�may�be�subject�to�increased�risks,�including�less�developed�custody�and�settlement�practices,�higher�volatility�and� lower�liquidity�than�non-emerging�market�securities.�Some�information�quoted�was�obtained�from�external�sources�we�consider�to�be�reliable.�No�responsibility�can� be�accepted�for�errors�of�fact�obtained�from�third�parties,�and�this�data�may�change�with�market�conditions.�This�does�not�exclude�any�duty�or�liability�that�Aquila� Capital�has�to�its�customers�under�any�regulatory�system.�Investment�regions,�sectors�and�strategies�illustrated�in�the�document�are�for�illustrative�purposes�only�and� should�not�be�viewed�as�a�recommendation�to�buy�or�sell.�The�terms�Aquila�and�Aquila�Capital�comprise�companies�for�alternative�and�real�asset�investments�as�well� as�sales,�fund�management�and�service�companies�of�the�Aquila�Group.�The�respective�responsible�legal�entities�of�the�Aquila�Group�that�offer�products�or�services� are�named�in�the�corresponding�agreements,�sales�documents�or�other�product�information.�

A�publication�of�Aquila�Capital�Investmentgesellschaft�mbH.�As�of�July�2018.�Author:�Saul�Butt,�Lars�Meisinger