WP/20/127

Accounting for Cloud Computing in the National Accounts

by Andrew Baer, Kwangwon Lee, and James Tebrake

IMF Working Papers describe research in progress by the author(s) and are published
to elicit comments and to encourage debate. The views expressed in IMF Working Papers
are those of the author(s) and do not necessarily represent the views of the IMF, its
Executive Board, or IMF management.

©International Monetary Fund. Not for Redistribution

© YYYY International Monetary Fund 2 WP/20/127

IMF Working Paper

Statistics Department

Accounting for Cloud Computing in the National Accounts 1

Prepared by Andrew Baer, Kwangwon Lee, and James Tebrake

Authorized for distribution by Gabriel Quiros-Romero

July 2020

IMF Working Papers describe research in progress by the author(s) and are published to
elicit comments and to encourage debate. The views expressed in IMF Working Papers are
those of the author(s) and do not necessarily represent the views of the IMF, its Executive
Board, or IMF management.

Abstract

Digitalization and the innovative use of digital technologies is changing the way we work,
learn, communicate, buy and sell products. One emerging digital technology of growing
importance is cloud computing. More and more businesses, governments and households are
purchasing hardware and software services from a small number of large cloud computing
providers. This change is having an impact on how macroeconomic data are compiled and
how they are interpreted by users. Specifically, this is changing the information and
communication technology (ICT) investment pattern from one where ICT investment was
diversified across many industries to a more concentrated investment pattern. Additionally,
this is having an impact on cross-border flows of commercial services since the cloud service
provider does not need to be located in the same economic territory as the purchaser of cloud
services. This paper will outline some of the methodological and compilation challenges
facing statisticians and analysts, provide some tools that can be used to overcome these
challenges and highlight some of the implications these changes are having on the way users
of national accounts data look at investment and trade in commercial services.

JEL Classification Numbers: E01, E2, E3, L96

Keywords: cloud computing, information and communication technology, national accounts
Author’s E-Mail Address: [email protected]; [email protected]; [email protected];

1
We would like to thank Marshall Reinsdorf for his detailed comments on an earlier draft of this paper. We are
also grateful to our colleagues Gabriel Quiros, Carlos Sanchez-Munoz, Robert Dippelsman, Jennifer Ribarsky,
and Marcelo Dinenzon for their support and suggestions. Finally, we extend our thanks to David Wasshausen,
Dylan Rassier, Greg Punchak, and Hussein Charara from the U.S. Bureau of Economic Analysis for providing
comments and helpful information based on their experiences.

©International Monetary Fund. Not for Redistribution

 3

Contents

Abstract............................................................................................................................. 2
I. Background .................................................................................................................... 4
II. Cloud Computing Services ............................................................................................ 4
III. Cloud Computing—The Industry .................................................................................. 6
IV. The Challenges Cloud Computing Poses for National Accountants.............................. 10
V. Updating Our Toolbox to Account For The Cloud........................................................ 16
VI. Updating Classification Systems................................................................................. 17
VII. New Data Sources .................................................................................................... 19
VIII. Sample Sizes and Coverage ..................................................................................... 24
IX. Conceptual Challenges............................................................................................... 25
X. What Difference This Makes To Our Users.................................................................. 27
XI. Conclusion ................................................................................................................ 30
XII. References................................................................................................................ 31

Tables
1. Traditional Model .........................................................................................................10
2. Cloud Model.................................................................................................................11
3. Cloud Model with Rationalization .................................................................................12
4. Cloud Model with Rationalization and Foreign Provision ...............................................13
5. Cloud Model with Rationalization and Foreign Provision with Multiple Data Centers .....15
6. Data Needed to Measure Cloud Computing....................................................................20

Figures
1. Global Cloud Computing Market Forecast .......................................................................7
2. Global laaS Cloud Computing Market Share....................................................................7
3. Share of Businesses Purchasing Cloud Computing Services in OECD Countries...............8
4. Businesses Purchasing Cloud Computing Services (2018) ................................................9
5. Share of Businesses Purchasing Cloud Computing Services in OECD Countries...............9
6. Services Producer Price Index for Entrusted Computing Services ...................................23
7. Official U.S. PPI for Cloud-Related Activities and Byrne, Corrado, Sichel Indexes for
Selected AWS Products ....................................................................................................24
8. Relative Size of Computer Services Export in OECD Countries .....................................28
9. Computer Services Export Ratio of Major Exporters ......................................................28
10. Portion of Major Server (HS847150) Importers............................................................29
11. Share of Server Imports in Total Imports......................................................................29

©International Monetary Fund. Not for Redistribution

 4

I. BACKGROUND

Digitalization and the innovative use of digital technologies is changing the way we work,
learn, communicate and buy and sell products. One emerging digital technology of growing
importance is cloud computing. More and more businesses, governments and households are
purchasing hardware and software services from a small number of large cloud computing
providers. This change is having an impact on how macroeconomic data are compiled and
how they are interpreted by users.

Specifically, this has the potential to change the information and communication technology
(ICT) investment pattern from one where ICT investment was diversified across many
industries to a more concentrated ICT investment pattern. Additionally, cloud computing is
having an impact on cross-border flows of commercial services since the cloud service
provider does not need to be in the same economic territory as the purchaser of the cloud
services. This paper outlines some of the methodological and compilation challenges facing
economic statisticians when ‘accounting for the cloud,’ provides some suggested tools that
can be used to overcome these challenges and highlights some of the implications these
changes are having on the way users of economic statistics interpret investment, trade and
productivity.

II. CLOUD COMPUTING SERVICES

Cloud computing services provide customers with on-demand Internet access to ICT
resources, such as computing power, data storage capacity, software services and operating
system functionality. These resources run on servers, storage devices, and networking
equipment housed in data centers operated by the cloud computing provider. The service
provider is responsible for the security, maintenance, and backup of the hardware, software,
and data stored in these facilities.

While ICT equipment rental and outsourcing of data processing functions have long been
sold by service providers, increased access to broadband Internet and the development of
virtualization technologies led to the rapid expansion of cloud computing in the 2000’s.
Virtualization allows the functionality of a single piece of hardware equipment to be
accessed from other computing resources, each with a dedicated operating system. The
related technology of containerization facilitates sharing the functionality of hardware and a
single operating system. Cloud computing relies on these technologies to provide customers
with a specified quantity of ICT functionality when it is needed, while reducing the amount
of ICT equipment that they themselves own and maintain. Amongst the first companies to
provide these services was Amazon, which had developed pooled computing resources for
internal functions that they then offered to external customers beginning in 2006.

Cloud computing services vary based on the type of ICT resources that are accessed.
Infrastructure as a Service (IaaS) is the provision of access to computing, storage, and/or

©International Monetary Fund. Not for Redistribution

 5

networking resources alone. Platform as a Service (PaaS) is geared towards customers that
are developing new software applications where an operating system and development tools
are provided in addition to infrastructure. Finally, Software as a Service (SaaS) provides
customers access to software applications through the Internet where the software runs on the
operating system and computing infrastructure of the cloud services provider.

Prices

Cloud computing services are primarily priced with subscription and usage fees. Large cloud
providers offer numerous service packages with varying price determining characteristics and
pricing units, making constant quality service pricing very challenging.

IaaS computing capacity can be provided in three categories, either on-demand, in pre-
reserved increments, or at periods when usage is low. Prices are typically charged per hour,
depend on the amount of memory required, with the highest rates for on-demand service and
lowest rates for service provided only at off-peak times. In some cases, hourly rates may be
assessed in intervals of seconds. Over the last number of years some cloud services providers
have moved from per hour charges to per second charges, dramatically changing the price
structure. For each of these categories, prices vary based on the operating system used, and
the amount of central processing units (CPUs), memory, and intermediate file storage
available, among other factors.

IaaS data storage services are usually priced per gigabyte of data stored per month, with
lower rates for each subsequent gigabyte of data stored above pre-determined thresholds.
Monthly discounts are provided for customers who commit to storing data for longer periods
of time. Data transfer services are similarly charged based on the amount of data transferred
in and out of the cloud.

PaaS services are typically charged based on the computing capacity and data storage
requirements used, as with IaaS. SaaS is typically charged as a subscription fee per user per
month, with discounts granted to buyers committing to more users and/or longer service
periods. Usage-based charges per user session are assessed less frequently.

In addition to these service-specific variables, prices are also based on the location of the
infrastructure that is used. Higher prices are charged to access resources in data centers
where demand is higher. While service quality is largely the same regardless of the
infrastructure location, customers may prefer to store sensitive information in certain
countries based on national privacy laws. Customers may also obtain very small
improvements in data latency 2 from using data centers geographically closer to the resource
users.

2
Data latency is the time it takes for data packets to be stored or retrieved.

©International Monetary Fund. Not for Redistribution

 6

While payments for cloud computing services are typically assessed in the provider’s home
country currency by default, some large global providers allow for payments to be made in
the currency of the buyer’s choice. In these cases, charges are determined based on the
seller’s offered exchange rates at the time of the transaction.

III. CLOUD COMPUTING—THE INDUSTRY

The cloud computing market is growing rapidly. According to industry estimates, the
revenue of global public cloud 3 computing services will increase 17.5 percent in 2019 to
US$214.3 billion (Figure 1). Furthermore, the market revenue is projected to reach
$331.2 billion in 2022, which is 54.5 percent higher than the 2019 estimate. All segments of
cloud computing services 4 are expected to show double-digit growth rates up to 2022. SaaS
will remain as the largest segment of the market in the next years, retaining its share at
around 43 percent. On the other hand, it is estimated that the the fastest-expanding segment
will be IaaS, which is forecast to grow an estimated 25.9 percent annually on average,
increasing its portion in the cloud computing market from 16.7 percent in 2018 to
23.1 percent in 2020.

The cloud computing industry is dominated by a small number of very large organizations,
such as Amazon (Amazon Web Services, AWS), Alibaba and Microsoft (Azure). For
example, according to Gartner, as shown in Figure 2, the top five cloud service providers
accounted for nearly 77 percent of the global IaaS market in 2018, up from less than
73 percent in 2017. The providers continue to solidify their position through economies of
scale and aggressive research and development activities. The market forecasts consolidation
will occur as organizations and developers look for standardized, broadly supported
platforms for developing and hosting cloud applications.5

3
Public cloud infrastructure is provisioned for use by multiple customers, while private cloud infrastructure is
restricted for use by a single customer only.
4
Cloud services are usually classified into services related to IT infrastructure (i.e., infrastructure-as-a-service,
or IaaS), platform provision (i.e., platform-as-a-service, or PaaS), access of software (i.e., software-as-a-service,
or SaaS). Gartner Research includes additional categories for business process automation (i.e., business-
process-as-a-service, or BPaaS) and cloud provision of management and security tools.
5
Gartner Research, Press Release 29 July 2019.

©International Monetary Fund. Not for Redistribution

 7

Figure 1. Global Cloud Computing Market Forecast

(Billions of U.S. Dollars)
400
SaaS IaaS BPaaS PaaS Management & Security
331.2

300 289.1
249.8
214.3
200 182.4

100

0
2018 2019 2020 2021 2022
Source: Gartner (April 2019).
Note: Software-as-a-service (SaaS), Infrastructure-as-a-service (IaaS), Business-process-as-a-service
(BPaaS), Platform-as-a-service (PaaS).

Figure 2. Global laaS Cloud Computing Market Share

(Percent)
47.8
Amazon
49.4
15.5
Microsoft
12.7
7.7
Alibaba
5.3
4.0
Google
3.3
1.8
IBM
1.9
23.2
Others 2018 2017
27.4

0 10 20 30 40 50 60
Source: Gartner (July 2019).

©International Monetary Fund. Not for Redistribution

 8

The shift to the cloud can also be observed in the demand-side statistics. In Figure 3, the
OECD reports that, on average, the share of businesses in OECD countries that purchased
cloud computing services increased nearly 6 percentage points between 2016 and 2018.
Businesses classified in the Information and communication sector and the Professional,
Scientific and Technical sector increased their use of the cloud by almost 10 and
8 percentage points respectively within two years. Figures for emerging market and
developing economies indicate the rapid expansion of cloud computing services as well.
According to the China Academy of Information and Communications Technology, a
Chinese government research institute, the public cloud market in China grew 65.2 percent in
2018 and is expected to expand almost threefold by 2022. Also, the public cloud services
revenue in India to reach $2.4 billion in 2019, an increase of 24.3 percent from 2018.6

Figure 3. Share of Businesses Purchasing Cloud Computing Services in

OECD Countries, by Industry1
(Percent)

Sources: OECD, ICT Access, and Usage by Business Database.

1Based on a simple average of the member countries. The latest figure is used is case these there is

not any data for the period. Canada is excluded due to a methodological break in 2017.
2Except of motor vehicles and motorcycles.

The extent of cloud adoption differs across countries, sectors, and firm size. Among OECD
countries, the share of businesses purchasing cloud computing services (Figure 4) in 2018
ranges from a high of 65 percent in Finland to a low of 10 percent in Latvia. The OECD
reports (Figure 3) that over 60 percent of firms in the Information and Communication sector
use cloud computing services but the comparable figure for Accommodation and Food and
Beverage services is only 21 percent. Size also matters as shown in Figure 5, where larger
firms are more likely to purchase cloud services than small firms.
6
Gartner Research, Press Release 18 June 2019.

©International Monetary Fund. Not for Redistribution

 9

Figure 4. Businesses Purchasing Cloud Computing Services (2018)1

(Percent)
70

60

50

40
32.4
30

20

10

OECD
Czech Republic
Australia
Iceland

Austria
Brazil2/
Netherlands
Norway

France
Luxembourg

Slovak Republic
Italy

Korea
Latvia
Ireland

Belgium
Japan2/

Estonia

Switzerland2/
Germany

Hungary
Finland
Sweden

Canada2/

United Kingdom

Lithuania
Portugal

Spain
Denmark

Slovenia
Sources: OECD, “ICT Accessd and Usage b by Business.”
1
The latest figure is used in case there is not any data for 2018.
2Difference in methodology.

Figure 5. Share of Businesses Purchasing Cloud Computing Services

in OECD Countries, by Business Size1
(Percent)
60
2016
50 2018

40

30

20

10

0
All businesses 250+ 50-249 10-49
Sources: OECD, “ICT Access and Usage by Business.”
1Based on a simple average of the member countries. The latest figure is used in case there

is no data for the period. Canada is excluded due to a methodological break in 2017.

©International Monetary Fund. Not for Redistribution

 10

IV. THE CHALLENGES CLOUD COMPUTING POSES FOR NATIONAL ACCOUNTANTS

Cloud computing is posing several challenges for national accountants. The transactions
associated with the sale and purchase of cloud-based IT services are quite different from the
transactions associated with the more traditional model where firms would self-provision
their IT services. One way to illustrate the far-reaching nature of these changes is by way of
an example that compares how national accountants record ‘traditional’ computing services
such as infrastructure, storage and networking with how they record ‘cloud computing’
services in the various economic statements that they compile.

Consider four firms:

• a firm that manufactures and sells IT equipment (e.g., servers);
• an accounting firm that requires a server to run their accounting practice;
• a clothing wholesaler that requires a server to deliver wholesaling services; and
• a computer processing firm that provides cloud computing services such as
processing services and storage services.

In the traditional model, where firms internally provision their IT services, the accounting
firm and wholesaler would purchase servers from the IT equipment manufacturer/wholesaler,
hire staff to maintain their servers and use this capital and labor to supply IT services to the
rest of the firm. Table 1 presents a stylized production and generation of income account
under this traditional model.

Table 1. Traditional Model

Transaction IT Equipment Accounting Wholesaler

Manuf acturer Firm
Goods and Services Account
Gross Fixed Capital Formation – IT Equipment. 50,000 50,000

Output
Servers 100,000
Accounting Services 25,000
Wholesaling Services 40,000

Intermediate Consumption
Inputs related to the provision of internal IT services 5,000 10,000
Other inputs 50,000 10,000 15,000

Gross Value Added 50,000 10,000 15,000

Compensation of Employees (IT employees) 3,000 5,000
Compensation of Employees (non-IT employees) 25,000 5,000 7,000
Operating Surplus, Gross 25,000 2,000 3,000

In the cloud services model we observe a different set of transactions. In this model the IT
services firm purchases servers from the IT equipment manufacturer. The accounting firm
and wholesaler no longer record a capital outlay, nor do they need to hire labor to manage

©International Monetary Fund. Not for Redistribution

 11

their servers. Instead, they have a new cloud services expense which records the purchase of
cloud services from the IT services firm. The structure of the production and generation of
income accounts under the cloud services model is illustrated in table 2.

Table 2. Cloud Model

Transaction IT Equipment Accounting Wholesaler IT Services

Manuf acturer Firm Firm
Goods and Services Account
Gross Fixed Capital Formation – IT 100,000
Equipment

Output
Servers 100,000
Accounting Services 25,000
Wholesaling Services 40,000
Cloud Computing Services 22,000

Intermediate Consumption
Inputs related to the provision of internal IT
services
Cloud Computing Purchased Services 8,000 14,000
(IaaS, PaaS, SaaS)
Other 50,000 10,000 15,000 15,000

Gross Value Added 50,000 7,000 11,000 7,000

Compensation of Employees (IT) 0 0 6,000
Compensation of Employees (Non-IT) 25,000 5,000 7,000
Operating Surplus, Gross 25,000 2,000 4,000 1,000

The structure of the production and generation of income accounts have changed with the
introduction of the IT services firm and subsequent sale of cloud services. In the traditional
model, the accounting firm and wholesaler both established internal IT departments and these
departments provided IT services to the rest of the organization. These IT services were not
recorded as output by the firm since they represent output produced and consumed by the
same firm. 7 With the introduction of the IT services firm, the IT services are produced
outside the firm and sold on the market. Given these services are now sold on the market the
national accountant must record the output of IT services in the IT services industry and then
show the purchase of these services by the consuming industries.

The result is a new product has been added to both the output and input matrixes and
economy wide output and input has increased – i.e., there has been a grossing up of output
and inputs by the value of the IT services. 8 Economy-wide gross value added under these
7
The System of National Accounts 2008 (2008 SNA) defines output as “goods and services produced by an
establishment…excluding the value of goods and services consumed by the same establishment, except for
goods and services used for capital formation (fixed capital or changes in inventories) or own final
consumption” (paragraph 6.89).
8
This is not a new phenomenon. The fragmentation of production has led to an expansion of output and inputs
in other industries as well.

©International Monetary Fund. Not for Redistribution

 12

assumptions remains the same (this is by design at this point – mainly for illustrative
purposes). Economy-wide gross operating surplus increases due to efficiency gains.

In addition to changes in total output and intermediate consumption the economy wide
investment pattern has shifted. In the traditional model both the accounting firm and
wholesaler recorded IT investment. In the cloud services model IT investment is
concentrated in the IT services firm. Finally, IT labor services have also shifted from the
accounting firm and government department to the IT services firm.

The above example, while illustrative, is simplified and does not fully capture all the possible
changes that could occur. As noted earlier, cloud computing has arisen because it is more
efficient for firms to purchase computing services then to purchase, operate and maintain IT
equipment. In the past, firms would invest heavily in IT equipment and IT capacity, often
having to purchase up to their maximum computing capacity needs even though their
maximum capacity was only required periodically. Cloud computing allows firms to
purchase what they need when they need it, allowing them to scale their operations quickly if
demand spikes. In theory, cloud computing should result in an economy wide rationalization
of IT services.

This rationalization can be overlaid on the previous example. Assume that the IT services
firm only needs to purchase a single server because it has determined that it can meet the
needs of both the accounting firm and wholesaler by shifting capacity as required. In
addition, it has determined that it only needs to hire half as many employees to maintain the
server rather than the full complement of employees required by the accounting firm and
wholesaler. These changes are shown in table 3.

Table 3. Cloud Model with Rationalization

Transaction IT Equipment Accounting Wholesaler IT Services

Manuf acturer Firm Firm
Goods and Services Account
Gross Fixed Capital Formation – IT 50,000
Equipment

Output
Servers 50,000
Accounting Services 25,000
Wholesaling Services 40,000
Cloud Computing Services 20,000

Intermediate Consumption
Inputs related to the provision of internal
IT services
Cloud Computing Purchased Services 7,000 13,000
Other 25,000 10,000 15,000 7,000

Gross Value Added 25,000 8,000 12,000 13,000

©International Monetary Fund. Not for Redistribution

 13

Compensation of Employees (IT) 0 0 3,000

Compensation of Employees (Non-IT) 12,500 5,000 7,000
Operating Surplus, Gross 12,500 3,000 5,000 10,000

In this case, the economy-wide output will decrease since both the output of the
manufacturing firm and cloud services firm declined (total output=135,000). Additionally,
total economy-wide value added also declines, due mainly to the decline in the production of
servers.

The rationalization propagated by the move to the cloud results in a reduction in the output of
servers, labor and cloud computing services. The decline in servers and labor is relatively
straightforward but the reduction in cloud computing services requires a little more
explanation.

In the traditional model both the accounting firm and wholesaler were required to purchase
more IT capacity than they required since they had to purchase up to their maximum
requirements rather than their average requirements. Because the IT service provider can
leverage capacity across firms they only need to purchase the average capacity of each firm.
This results in a lower measured output of overall computing services when moving to the
cloud model because the excess capacity is no longer required. The accounting firms and
wholesaler are now only purchasing their average requirements and their gross value added
should increase due to this rationalization.

Productivity in the accounting firm, wholesaler and IT services firm should increase as
computing resources are centralized. Over the longer run, increases should be seen in both
labor productivity and multi-factor productivity since the same output will be produced with
less labor (IT staff) and less capital (excess capacity is no longer required). These
productivity gains should be most notable in those industries where there is considerable
movement from traditional ICT investment to the use of cloud services.

Another possible and important change involves the geographic provision of the cloud
services. In the traditional model, the server would be located at the same physical premises
as the other factors of production. Under the cloud model it is quite possible for the cloud
provider to be in one economic territory and the cloud purchaser to be in another economic
territory. This means that the purchase of the cloud services could represent an import rather
than a domestic purchase.

The fact that these services can be provided by firms located in another economic territory
further changes the picture. Incorporating the notion that the IT service provider is outside
the economic territory means additional import and export transactions need to be recorded.
These changes are illustrated in Table 4.

Table 4. Cloud Model with Rationalization and Foreign Provision

©International Monetary Fund. Not for Redistribution

 14

Economic Territory A Economic Territory B

Transaction IT
Accounting
Equipment Wholesaler IT Services Firm
Firm
Manuf acturer
Goods and Services Account
Exports of Servers 50.000
Exports of Cloud Services 20,000
Imports of Cloud Services 7,000 13,000
Gross Fixed Capital Formation 50,000
– IT Equipment / Input of IT
Equipment

Output
Servers 50,000
Accounting Services 25,000
Wholesaling Services 40,000
Cloud Computing Services 20,000

Intermediate Consumption
Inputs related to the provision of
internal IT services
Cloud Computing Purchased 7,000 13,000
Services
Other 25,000 10,000 15,000 7,000

Gross Value Added 25,000 8,000 12,000 13,000

Compensation of Employees 0 0 3,000
(IT)
Compensation of Employees 12,500 5,000 7,000
(Non-IT)
Operating Surplus, Gross 12,500 3,000 5,000 10,000

Under this scenario economy-wide output declines in territory A since the cloud services
(and the associated output) are provided by a foreign firm in territory B. Intermediate
consumption remains the same since the accounting firm and wholesaler continue to
purchase cloud computing services. The only difference is these services are now provided
by a foreign entity and recorded as an import.

As noted in the first section of this paper, the provision of cloud services is concentrated in a
small number of large firms. For these firms to be able to deliver service to their clients they
have established business operations and data centers in various locations across the globe.

Expanding the example from earlier a third territory (C) is added. Territory C contains a data
center controlled by the cloud computing firm headquartered in territory B. The cloud
computing firm also maintains a data center in territory B. The accounting firm and
wholesaler located in territory A continue to purchase cloud computing services from the
firm in territory B. The firm in territory B is free to move production form B to C to deliver
the cloud computing services to the firm in territory A.

©International Monetary Fund. Not for Redistribution

 15

Table 5. Cloud Model with Rationalization and Foreign Provision with Multiple
Data Centers

Economic Territory A B C
Transaction IT Equipment Accounting Wholesaler IT Services IT
Manuf acturer Firm Firm Services
Firm
Goods and Services
Account
Exports of Servers 50.000
Exports of Cloud Services 10,000 10,000
Imports of Cloud Services 7,000 13,000
Gross Fixed Capital 25,000 25,000
Formation – IT Equipment
/ Imports of IT Equipment

Output
Servers 50,000
Accounting Services 25,000
Wholesaling Services 40,000
Cloud Computing Services 10,000 10,000

Intermediate Consumption
Inputs related to the
provision of internal IT
services
Cloud Computing 7,000 13,000
Purchased Services
Other 25,000 10,000 15,000 5,000 2,000

Gross Value Added 25,000 8,000 12,000 5,000 8,000

Compensation of 0 0 1,000 2,000
Employees (IT)
Compensation of 12,500 5,000 7,000
Employees (Non-IT)
Operating Surplus, Gross 12,500 3,000 5,000 4,000 6,000

The question facing the national accountant is whether the accounting firm and wholesaler
are importing from B or B and C. If their direct payment is going to B, then the import is
from B even if B subsequently imports from C. Once this is determined the compiler is still
left with the question ‘how much’ since during any given accounting period A could receive
all their services from B or all their services from C or a combination thereof.

The Sixth Edition of the IMF's Balance of Payments and International Investment Position
Manual (BPM6) details measurement of trade on an ownership basis, rather than a physical
basis. If the consumer of cloud services located in country A is billed by an establishment in
country B that manages the delivery of the services, the source of the imports of services is
country B. That establishment may have a contract with a data center in country C where the
actual computing takes places, and payments under that contract would be an import by

©International Monetary Fund. Not for Redistribution

 16

country B from country C. Compiling these service flows in the proper economic territorites
is another challenge facing national accountants and balance-of-payments statisticians.

The final measurement challenge facing national accountants concerns their ability to
develop volume estimates to have an accurate measure of real growth. Cloud computing
services are transacted between firms, and therefore cloud computing introduces the need for
new service price indexes. In the past, when firms purchased their own equipment these
investment goods would be deflated using ‘goods’ prices which tended to be readily
available. While ICT goods prices are a challenge, they are less of a challenge than the
production of reliable service price indexes. In addition to the difficulty in collecting and
compiling service prices, price statisticians and national accountants also face challenges
associated with the fact that these services are often bundled and sold on contract with a wide
range of discounts and incentives. Papers presented at the 2018 meeting of the Voorburg
Group on Services Statistics show the approaches that national statistical offices in Finland,
Poland, Israel, and the United States have taken in developing producer price indexes that
include the provision of cloud computing services.9

The above example illustrates the significant structural changes a large-scale shift by firms
towards the adoption of a cloud strategy for the provision of IT services has on an economy.
All things equal, if a large share of firms begin to purchase a large share of their IT services
users of national account data can expect (all things equal):
• a softening in output and investment in IT equipment;
• investment in IT equipment to become increasingly industrially and geographically
concentrated;
• increases in the cross-border flow of commercial services;
• increases in the relative size of IT services industries;
• increased concentration of IT related labor in those industries providing cloud
services;
• potential gains in productivity due to the rationalization associated with cloud
computing;
• appearances of new products and processes made possible by cloud computing; and
• change in the geographic concentration of imports of ICT equipment.

V. UPDATING OUR TOOLBOX TO ACCOUNT FOR THE CLOUD

It is apparent that the wide-scale move to cloud computing will have a significant impact on
the interpretation of macroeconomic statistics. As can be seen from the previous examples,
the current macroeconomic frameworks can properly capture and record this activity and
present it to users in a meaningful way. While the framework appears to be sufficient, the
more significant challenges facing national accountants relates to the statistical infrastructure

9
Links to the papers by Tag, et al (Finland), Cebula, et al (Poland), Tal (Israel), and Baer, et al. (U.S.) are
provided in the references.

©International Monetary Fund. Not for Redistribution

 17

in place at most statistical offices to properly capture this activity. In most NSOs, the existing
statistical infrastructure is more suited for the traditional IT service delivery model described
above. It is less suited for the emerging cloud model use to deliver IT services. To properly
account for cloud computing, there are three areas where NSOs may want to invest to ensure
their national accounts are “cloud” ready. First, NSOs and International Organizations (IOs)
will need to invest in updating classification systems to ensure cloud computing services are
property identified in national accounts. Second, NSOs and IOs will need to develop new
data sources to ensure the activity is properly captured. Third, NSOs and IOs may need to
consider conceptual changes to their frameworks, specifically as it relates to cross border
flows and the recording of investment in software.

VI. UPDATING CLASSIFICATION SYSTEMS

Classification systems are critical building blocks that support the reporting and presentation
of economic data. There are many classification systems that support statistical systems such
as geographic classification systems, institutional sector classification systems, occupational
classification systems, product classification systems and industry classification system. The
classification systems most impacted by the advent of cloud computing are the product
classification system and the industrial classification system.

A product classification system classifies goods and services that are the result of production
in the economy. The classification system groups the millions of goods and services in
homogenous groups or classes to help data users better interpret the goods and services that
are being produced, used and consumed in an economy.

An industrial classification system classifies the producers of goods and services that are the
result of production in the economy into distinguishable groups or classes. The producers are
assigned to groups or classes based on the similarity in their production function and the
goods and services they produce.

The phrases “cloud computing,” “IaaS,” “PaaS,” and “SaaS” do not appear in either Central
Product Classification (CPC) Version 2.1 or International Standard Industrial Classification
of All Economic Activities (ISIC) Version 4. As a result, it is not clear that statistical agencies
are classifying these services consistently in production statistics and associated deflators.
With the large and growing sales of cloud computing, future updates to product
classifications should provide clear guidance on these services individually. Strong interest in
data about cloud computing services calls for new product and industrial categories dedicated
to these services alone.

Presentations at the 2018 meeting of the Voorburg Group on Services Statistics highlighted
some inconsistencies in the classification of cloud computing products. While there was
general agreement that IaaS is classified in CPC subclass 83159, Other hosting and IT
infrastructure provisioning services, the treatment of PaaS and SaaS varied.

©International Monetary Fund. Not for Redistribution

 18

The 2018 report of the Eurostat Task Force on Price and Volume Measures for Services
Activities report stated that PaaS “is most likely” classified in CPC class 8314, Information
technology (IT) design and development. For the U.S., however, this service is classified in a
category that concords with CPC subclass 83159, Other hosting and IT infrastructure
provisioning services.

The Eurostat Task Force report states that SaaS should be classified with software
publishing, which would place it in CPC subclass 84392, Online software. In the U.S.,
however, a distinction is made based on whether the provider holds copyright to the software
that is provisioned. If they do own the rights, then providing the software through the Internet
is classified with software publishing. If they do not own the rights then the service is
classified as CPC subclass 83152, Application service provisioning (ASP), defined as the
provision of leased software from a centralized, hosted, and managed computing
environment.

As cloud computing services have grown, large providers are increasingly offering service
bundles. These bundles allow clients to use IaaS for applications that they prefer to manage
themselves, PaaS for application that they prefer to design themselves, and SaaS for third-
party applications that they prefer to access only as needed. While Microsoft has always
bundled hosted infrastructure with their original software products, infrastructure-focused
service providers have also recently developed software applications that they are offering to
customers through a SaaS model. This includes business intelligence tools, such as Amazon
QuickSight and Google Cloud DataLab. Within the SaaS category, large service providers
are similarly offering a mix of their own developed software solutions and those developed
by other publishers. These bundles create additional challenges in consistent product
classification.

While most cloud computing services are sold to business and institutional buyers, cloud data
storage services and some SaaS products are also sold for personal consumption expenditure.
Cloud storage services are included in the Classification of Individual Consumption
According to Purpose (COICOP) class 08.3.3, Internet access provision and net storage
services. SaaS is classified in COICOP subclass 08.3.9.2, Subscription to audio-visual
content, streaming services and rentals of audio-visual content.

As can be seen, cloud computing is combined with other types of products making it difficult
for users to understand the transformative shift that may be occurring.

For industrial classification, cloud computing providers are currently recorded in ISIC
industry 6311, Data processing, hosting, and related services, along with a variety of services
such as data entry and time-share provision of mainframe computers.

©International Monetary Fund. Not for Redistribution

 19

Again, because the classification systems were developed prior to the widescale adoption of
cloud computing it is not possible for most NSOs to answer questions such as the size and
growth of the cloud computing industry. In order to address these issues it is important for
the international community to develop an agreed upon set of product and industrial classes
that will facilitate both the collection, presentation and analysis of the data.

Since cloud computing activities have effectively replaced most of the data entry and
mainframe-related services, ISIC 6311 could be re-named to Cloud Computing Services.
Products within the activity could include IaaS, PaaS, and SaaS, except where the provider
holds copyright. Cloud provision of software to which the provider holds copyright could
remain classified in the software publishing industry.

The Extended Balance of Payments Services (EBOPS) classification system is used to record
and present international imports and exports of trade in services. Imports and exports of
cloud services would be considered part of the telecommunication, computer and
information services category. This category is broken down into 7 classes and sub-classes
(such as telecommunication services, software services, new agency services. It is not clear
whether cloud services would be recorded in the other information services category or the
other computer services category. To increase the usability of the classification system and
quality of the resulting data it would be useful to update the system to include classes related
to the different type of cloud products.

Another important piece of statistical infrastructure that an NSO will need to update is their
business register. As new cloud computing firms emerge, these enterprises will need to be
added to the business register and coded to the appropriate industries so that they form part of
the overall universe of enterprise and institutional units in a given country.

The difficulty NSOs will face when updating their business register concerns how to properly
classify the different units within the enterprise. For example, in many cases the service
provider is a very large multinational organization with affiliates operating in many different
countries. The NSOs will need to determine the exact nature of the operations of the affiliate.
Is the affiliate simply a sales office or is the affiliate delivering the cloud service to local and
foreign customers? Is the affiliate engaged in research and development activities, producing
important intellectual property that benefits the entire multinational organization? The
determination of these questions is important because it helps define the types of transaction
occurring in the economy.

VII. NEW DATA SOURCES

Probably the largest cloud computing measurement challenge NSOs are facing is the need to
update its source data collection systems to capture cloud computing activity. Most NSOs use
the classification systems noted above as the starting point to design their collection
instruments. As these new product classes and industries are developed the questionnaires,

©International Monetary Fund. Not for Redistribution

 20

administrative data processing systems, methods, computational systems need to be updated

to reflect cloud computing activity.

One way to present these changes is by identifying the type of information that needs to be
collected or acquired, identify possible collection vehicles that may need to be updated or
acquired and then make note of any changes that need to be made to these traditional vehicles
to ensure the NSO can capture the activity. This is summarized in the following table.

Table 6. Data Needed to Measure Cloud Computing

Data to be Possible Collection Required Update

Acquired Vehicles
Purchases of IT Business Surveys In most cases, countries are only collecting information
processing and Administrative related to the the purchase of IT consulting firms. Fee
services Tax Forms countries are explicitly collecting purchased service
expenditures related to the provision of cloud
computing services such as software subscriptions,
processing time and storage space.
Purchases of IT Business Surveys, There is no change needed with respect to the
processing Administrative Tax collection of data related to the purchase of IT
equipment Forms and Customs equipment. It is possible that NSOs will start to see a
Forms greater consolidation of activity amount a smaller
number of firms.
Purchases of Business Surveys, A decision needs to be made regarding the difference
Sof tware Administrative Tax between the purchase of software (which is capitalized)
Forms and Customs and a subscription to software services. If it is deemed
Forms, Own-account that a software subscription is a purchased service that
estimation methods is not capitalized it will have a significant impact on
how software is recorded in macroeconomic accounts.
Prices of cloud Producer Price New prices indexes will need to be developed to record
services Reports, Consumer the provision of cloud services to both producers and
Price Reports consumers. Given the nature of the services and the
rapid technological change, some form of hedonic
measurement is probably required to ensure only pure
price change is recorded.
Trade in cloud International Trade in Surveys should be expanded to include detailed
services Services Surveys questions on imports and exports of cloud services.
These data would help clarify residence of service
providers in complex global cloud service supply
chains where intellectual property assets, IT
processing equipment, and customers are all in
multiple locations.

©International Monetary Fund. Not for Redistribution

 21

To better illustrate the type of changes required, consider the latest annual manufacturing
business survey issued by the U.S. Census Bureau, Statistics Canada and the Office of
National Statistics in the U.K. as presented in figures 1, 2, and 3.

In the case of Canada, cloud computing services could get classified into one of three
expense categories: Professional and business fees, office and computer related expenses and
telephone, internet and other telecommunication expenses. The questions posed to
respondents are shown below.

In reviewing the reporting guide for the questionnaire there is no specific mention of cloud
services. Respondents are instructed to include data processing services fees under
professional and business fees and data processing expense (equipment, software and
software licenses) under office and computer related expenses. Whereas under the
telecommunication services question respondents are instructed to include such items as
internet service charges, online information provision services, and online access services.
Given these instructions it is not clear where to record cloud related storage, processing and
software services. Even if the expenses were properly classified, it would be difficult to
isolate the economic impact associated with a large scale move to cloud computing.

©International Monetary Fund. Not for Redistribution

 22

Canada is not alone in its challenge to better articulate and isolate the impact of cloud
computing on the economy. A similar situation exists in the U.S., where businesses are asked
to report expenses for a broad category labeled “Data processing and other purchased
computer services.” This includes all “computer-related advice and services.” Respondents
are not instructed on how to classify cloud computing expenses.

Since distinct classification categories for cloud computing services do not currently exist, it
is difficult to find official price statistics that show recent price developments. While NSOs
have made progress in collecting prices for emerging cloud computing services, this data is
frequently hidden within product categories that include other tangentially related services
and are not labeled as cloud computing. New official price indexes dedicated to measuring
cloud computing prices are needed to identify these price developments.

©International Monetary Fund. Not for Redistribution

 23

The Bank of Japan produces a price index for Entrusted computing services (cloud
applications), which has steadily declined since 2010. This product category closely aligns
with SaaS.

Figure 6. Services Producer Price Index for Entrusted Computing Services

(Cloud Applications)
Bank of Japan, 2015=100

While official price statistics for IaaS services alone are not available, recent research studies
have produced interesting results. Byrne, Corrado, and Sichel (2018) constructed price
indexes for some of the most important Amazon Web Services (AWS) IaaS products in the
U.S. and recorded quality-adjusted price declines of approximately 50 percent between 2009
and 2016. Coyle and Nguyen (2019) evaluated similar products from AWS offered in the
U.K and found even larger price decreases over this period. However, the U.S. Producer
Price Index (PPI) for the service category that most closely aligns with these services
(Hosting, ASP, and other IT provisioning services) was little changed over the same period.

©International Monetary Fund. Not for Redistribution

 24

Figure 7. Official U.S. PPI for Cloud-Related Activities and Byrne, Corrado,
Sichel Indexes for Selected AWS Products

110.0
100.0
90.0
80.0
70.0
60.0
50.0
40.0
30.0

U.S. Hosting, ASP and other IT infrastructure provisioning services PPI

Amazon EC2 Compute (Byrne, Corrado, Sichel)
Amazon RDS Database (Byrne, Corrado, Sichel)

VIII. SAMPLE SIZES AND COVERAGE

NSOs will additionally need to look at how they design their samples in the face of a wide-
scale adoption of cloud computing services. For example, when statisticians design imports
and exports of commercial services questionnaires they generally stratify their surveys to
include those firms likely to import or export services such as accounting services, legal
services, engineering services etc. The number of firms in an economy importing or
exporting these types of services has traditionally been limited to larger, more mature firms.
In a world where cloud computing is the primary option for the purchase of IT services the
range of firms that should now be considered an importer is increasing and includes smaller
firms in less traditional importing industries. Without an expansion of a country’s sample of
importers, there could be a drift in import statistics due to a growing under coverage in
imports of cloud services.

Design strategies and samples sizes should therefore be examined across a host of surveys
and statistical programs including:
• Imports of Commercial Services Surveys
• Exports of Commercial Services Surveys
• Surveys related to computer design services
• Price report surveys
• Annual and quarterly business statistics surveys for cloud activities
• Business expense surveys

©International Monetary Fund. Not for Redistribution

 25

IX. CONCEPTUAL CHALLENGES

While most of the challenges related to measuring cloud computing can be addressed by the
current suite of measurement frameworks there are a few conceptual challenges that have
arisen related to the subscription of software and the location of data centers.

When a firm purchases software such as Microsoft Office, SAP, SAS – whether it is a license
or outright purchase of a copy, the SNA recommends that the software be capitalized as a
fixed asset and depreciated over time provided it is used continuously in production for more
than one year. The SNA goes on to say that if an institutional unit acquires a license to use
software and the license is “purchased with a single payment for use over a multiyear
period,” 10 it should be recorded as an asset. If, on the other hand the acquisition of the copy
with a license to use is paid for using a regular (monthly, quarterly) stream of payments it
should be recorded as payments for service since there is no indication that the firm requires
the service for more than one year.

The pay as you go nature of cloud computing poses some guidance questions that need to be
addressed. Consider the following scenarios:
a. A firm sets up an account with a cloud provider by which its employees can
use a specific software on an as needed basis for the next two years. The firm
is billed according to its usage.
b. A firm subscribes to a cloud provisioned software service on a month to
month basis. It has been doing this for the last three years.
c. A firm subscribes to a cloud provisioned software service for a baseline of
20 users and enters into a three-year contract. The contract states that
additional users can use the software and that the firm will be billed monthly
for this additional usage.

In each of the above scenarios it is unclear how these transactions should be classified. In the
first two scenarios it is unclear whether the purchase should be recorded as an asset or
intermediate expense. In the third case it is unclear as to the value of the asset. Given the
changing and dynamic pricing models associated with cloud computing there are many more
scenarios that could be illustrated. It is not realistic to think that treatment should be
determined on a case by case basis. Guidance and recommendations are required to identify
when the cloud services should be treated as an asset or as intermediate consumption.

A second conceptual challenge facing statisticians concerns the geographic location of the
‘production’ of cloud services. From a statistical compiler’s perspective when all the family
photos are stored on the family laptop the calculations are straight forward. When it is
decided to upload all the family photos to the cloud and pay a monthly storage fee it gets a
little more complicated. Microsoft has a nice feature on its website where it informs its

10
SNA 10.100.

©International Monetary Fund. Not for Redistribution

 26

clients where it stores their data. The figure below lists where customer data for Asia
countries is stored. As you can see, the data are stored in a variety of locations and the
location differs depending on the platform.

This poses a significant issue for national accounts and balance of payments compilers. If the
client is in Asia and is purchasing storage services from a Microsoft data center – which
country is the client transacting with? In this case the cloud truly is multi-jurisdictional but
our macroeconomic standards dictate that we need to assign production to an economic
territory. The above situation does not so much demand a conceptual solution as it demands
some form of pragmatic solution. The reason for this pragmatic approach is because it is not
realistic (or possible?) to expect Microsoft to provide statistics compilers the value of storage
service trade between Hong Kong and Singapore. Especially because Microsoft is probably
able to move storage between Hong Kong and Singapore on a frequent basis (daily, weekly,
monthly).

What is probably needed in this case is some sort of recommendation that states the
production of the services will be attributed to the location where the activity predominately
(great than 50 percent) takes place.

It is clear that the wide-scale adoption of cloud computing services is not only pushing the
statistical bounds, it is also pushing the conceptual boundaries associated with

©International Monetary Fund. Not for Redistribution

 27

macroeconomic frameworks such as the Balance of Payments and System of National

Accounts.
X. WHAT DIFFERENCE THIS MAKES TO OUR USERS

The shift to cloud computing will affect the pattern of investment. As described in Figure 3
above, the share of businesses purchasing cloud computing services has increased across all
industries among OECD countries. At the same time, the statistics show that the investment
in cloud computing is being concentrated industrially. Differences in the rate of adoption can
also be found geographically. In figure 4 above we have found that, even among OECD
countries, the reported share of businesses purchasing cloud computing services varies by as
much as 55 percentage points in 2018.

The expansion of cloud computing industry can also be seen in the international trade
statistics. According to the ICT services complementary grouping recommended by the
Partnership on Measuring ICT for Development, an international initiative, the Balance of
Payments Services classification (EBOPS) 2010 commodity group 9.2 Computer services
includes the CPC codes for cloud computing products.11 Across OECD countries, computer
services exports have increased rapidly since 2010, suggesting a possible surge in trade of
cloud computing services. From 2010 to 2017, the computer services exports of United
States and European countries (28 countries) registered average annual growth of
14.3 percent and 7.5 percent respectively.

As described in Figure 8, it is reported that Ireland is the biggest exporter for computing
services among OECD countries. The country’s exports accounted for 29 percent of the total
computing services exports of OECD countries in 2016. Also, the exports represent
43 percent of Ireland’s total exports of services for the year (Figure 9). This share is notably
high, compared to other major computer services exporters such as Germany (10 percent)
and the United States (3 percent). This distinctiveness of Ireland’s figure could partly be
explained by the fact that multinational enterprises credit a large proportion of global
computing sales to business units resident in Ireland.

As shown above, current trade statistics provide a rough picture of cloud computing services
trade. However, the importance of accurately capturing the trade will grow, as the cloud
computing industry expands. Furthermore, the scalability of the service and the mobility
between data centers require well timed measurement of the trade. Besides a modified
EBOPS classification for the cloud computing services, national accountants would
especially need more information on the residency and volume measures for GDP estimates.
In this regard, information regarding the import and export prices of cloud computer services
are required.

11
UNCTAD. “International Trade in ICT Services and ICT-enabled Services: Proposed Indicators from the
Partnership on Measuring ICT for Development.”

©International Monetary Fund. Not for Redistribution

 28

Figure 8. Relative Size of Computer Services Export in OECD Countries

(2016, in U.S. Dollar)

23.7%
Ireland 29.0%
Germany
United States
Netherlands
Sweden 4.0%
France
4.4%
Israel
Spain 4.8%
13.3%
Others
5.5%
6.4% 8.9%
Source: OECD International Trade in Services Statistics.

Figure 9. Computer Services Export Ratio of Major Exporters

(Percent)
50
Ireland
45 Germany
40 United States
Netherlands
35 Sweden
30
25
20
15
10
5
0
2010 2011 2012 2013 2014 2015 2016 2017
Source: OECD International Trade in Services Statistics.

The shift to cloud computing is also having an impact on goods trade. According to the
UNCTAD 2013 World Information Report, servers can be imported under three sub-
categories of Harmonized System Codes (HS) 8471. Automatic data processing machines
and units: 847141, 847149, and 847150. Among them, the largest and most relevant category
is HS 847150, covering computer servers without keyboard and monitors.12 Figure 10 shows

12
Coyle and Nguyen (2019).

©International Monetary Fund. Not for Redistribution

 29

the world share of imports of servers for selected countries. The United States is the
dominant importer and its portion has continuously increased since 2014. The import value
itself increased more than threefold between 2009 and 2018. Ireland, a location of a number
of data centers, has reported significant growth in imports of servers (Figure 11).

Figure 10. Portion of Major Server (HS847150) Importers

(Percent)
45
United States China1/ Japan Germany Canada
40

35

30

25

20

15

10

0
2009 2010 2011 2012 2013 2014 2015 2016 2017 2018
Sources: UN Comtrade.
1Include Hong Kong SAR.

Figure 11. Share of Server Imports in Total Imports

(Top five countries in 2017, percent)
1.6
Ireland USA Israel
1.4
Finland Japan
1.2

1.0

0.8

0.6

0.4

0.2

0.0
2009 2010 2011 2012 2013 2014 2015 2016 2017 2018
Source: UN Comtrade.

©International Monetary Fund. Not for Redistribution

 30

Investment patterns, trade flows, economy-wide output and value added are some of the key
macroeconomic indicators that will be impacted by a wide scale move by businesses and
households towards cloud computing. Given the potential size of the shift it is important that
national accountants and balance of payments compilers provide granular timely data to users
so that they can better understand the significance of the transformation.

XI. CONCLUSION

This paper was a first attempt to outline the challenges associated with ‘accounting’ for a
wide-scale sales and purchases of cloud computing services. The paper argued that the
delivery of the service is concentrated and global. The potential number of buyers is very
large and given the scale and global nature of the activity the implications on macroeconomic
indicators and measurement could be large – impacting output, trade, investment and real
value added. The paper then noted that most NSOs and IOs are currently not well-placed to
measure this transformative change since its measurement requires a new set of tools and
instruments. Finally, the paper noted that while most of the challenge is on the measurement
side of the border, there are a few conceptual lines that may need to be crossed or moved in
order to ensure users are able to interpret and have an appreciation for the changes that are
occurring.

©International Monetary Fund. Not for Redistribution

 31

XII. REFERENCES

“Gartner Forecasts Worldwide Public Cloud Revenue to Grow 17.5 Percent in 2019.” Last
modified April 2, 2019. https://www.gartner.com/en/newsroom/press-releases/2019-
04-02-gartner-forecasts-worldwide-public-cloud-revenue-to-g.

“Gartner Says Worldwide IaaS Public Cloud Services Market Grew 31.3% in 2018.” Last
modified July 29, 2019. https://www.gartner.com/en/newsroom/press-releases/2019-
07-29-gartner-says-worldwide-iaas-public-cloud-services-market-grew-31point3-
percent-in-2018.

Baer, Andrew, Thomas Hill, Thomas Ponikowski, Bonnie Murphy, Melanie Santiago, and
Jason Carnival. 2018. “ISIC 6311 Data processing, hosting, and related services.”
https://voorburggroup.org/Documents/2018%20Rome/Papers/1014.pdf.

Byrne, David, Carol Corrado, and Daniel E. Sichel. 2018. “The Rise of Cloud Computing:
Minding Your P’s, Q’s and K’s.” NBER Working Paper Series.
http://www.nber.org/papers/w25188.

Cebula, Beata and Agnieszka Matulska-Bachura. 2018. “ISIC 6311 Data processing, hosting,
and related services.”
https://voorburggroup.org/Documents/2018%20Rome/Papers/1018.pdf

“Coyle, Diane, David Nguyen. 2018. “Cloud Computing and National Accounting.” ESCoE
Discussion Paper 2018-19. https://www.escoe.ac.uk/research/discussion-papers-3/.

Coyle, Diane, David Nguyen. 2019. “Cloud Computing, Cross-Border Data Flows and New
Challenges for Measurement in Economics.” National Institute Economic Review
249 (August). https://journals.sagepub.com/doi/full/10.1177/002795011924900112.

Dignan, Larry. “Top cloud providers 2019: AWS, Microsoft Azure, Google Cloud; IBM
makes hybrid move; Salesforce dominates SaaS.” ZDNet. Last modified
August 15, 2019. https://www.zdnet.com/article/top-cloud-providers-2019-aws-
microsoft-azure-google-cloud-ibm-makes-hybrid-move-salesforce-dominates-saas/.

Eurostat. 2018. “Final report of the Task Force ‘Price and Volume Measures for Service
Activities.’”https://unstats.un.org/unsd/nationalaccount/aeg/2018/M12_3b_Price_Vol
ume_Services_Activities.pdf.

Mathias, Craig. “What is virtualization? Far more than just virtual machines.” Network
World. Last modified October 26, 2017.
https://www.networkworld.com/article/3234795/what-is-virtualization-definition-
virtual-machine-hypervisor.html.

Miller, Ron. “How AWS came to be.” TechCrunch. Last modified July 2, 2016.
https://techcrunch.com/2016/07/02/andy-jassys-brief-history-of-the-genesis-of-aws/

©International Monetary Fund. Not for Redistribution

 32

Rubens, Paul. “What are containers and why do you need them?” CIO. Last modified June
27, 2017. https://www.cio.com/article/2924995/what-are-containers-and-why-do-you-
need-them.html.

Tag, Susanna, Anna-Riikka Pitkanen. 2018. “Mini-presentation on Data processing, hosting,

and related activities (ISIC 6311) with emphasis on cloud computing.”
https://voorburggroup.org/Documents/2018%20Rome/Papers/1016.pdf.

Tal, Itsik, 2018. “ISIC 6311 Data processing, hosting, and related services.”
https://voorburggroup.org/Documents/2018%20Rome/Papers/1020.pdf

“UNCTAD. 2015. “International Trade in ICT Services and ICT-enabled Services: Proposed
Indicators from the Partnership on Measuring ICT for Development”, UNCTAD
Technical Notes on ICT for Development.
https://unctad.org/en/PublicationsLibrary/tn_unctad_ict4d03_en.pdf

Weins, Kim. “Cloud pricing comparison: AWS vs. Microsoft Azure vs. Google Cloud vs.
IBM Cloud.” InfoWorld. Last modified November 22, 2017.
https://www.infoworld.com/article/3237566/cloud-pricing-comparison-aws-vs-azure-
vs-google-vs-ibm.html.

©International Monetary Fund. Not for Redistribution