Transcript PPT

ADVISORY EXPERT GROUP
BACKCASTING
Statistics Canada
For discussion
 The need for National Account time series
 Considerations and constraints
 Possible approaches
 The need for guidance
2
Statistics Canada • Statistique Canada
Main functions of the National Accounts

The National Accounts allow users to examine the
structure of the economy and the evolution of the
economy.

Users require a time-series of economic data in order to
undertake:
•
•
•
•
•
•
3
monetary and fiscal policy and monitoring;
structural analysis (productivity, input-output or GE modelling);
forecasting and business cycle analysis
administrative uses;
environment-economy linkages; and
research uses.
Statistics Canada • Statistique Canada
How long is long enough?

The National Accounts (NA) record the macroeconomic
history of a nation. The data along, with associated
contextual and analytical information, is a national asset
that needs to be both preserved and invested in.

The greater the coverage of an NA time series, the
greater the benefit, since one of the uses is to compare
recent events to similar economic developments of the
past.

The greater the number of economic developments
covered by the time series the greater the use to policy
makers.
4
Statistics Canada • Statistique Canada
Why it should be done by the NSO

An argument can be made that backcasting should be
left to the users—the NSO need only supply the
metadata required to backcast the data.

This is countered (mainly by users) that:
•
•
•
5
The NSO is in the best position to backcast the data given their
intimate knowledge of the data and accounting rules.
This would avoid multiple ‘official’ measures due to differences
in approaches.
NSO compilation lends credibility to the data.
Statistics Canada • Statistique Canada
Considerations and constraints to
backcasting
 An NSO can take a number of different approaches
to backcasting. These include:





6
Publishing breaks in the time series
Backcasting a portion of the time series with partial detail
Backcasting a portion of the time series with full detail
Backcasting the full time series with partial detail
Backcasting the full time series with full detail
Statistics Canada • Statistique Canada
Considerations and constraints to
backcasting
 There are some important considerations and
constraints national accountants face when backcasting
National Account time series. These include:
•
•
•
•
•
•
Time
Human capacity
Legislative / administrative uses of the data
Coordination (e.g., integrated accounts, decentralized accounts)
Availability of source data and quality of source data
Significance of the changes to be backcasted
 The approach taken in large part is determined by the
above factors.
7
Statistics Canada • Statistique Canada
Considerations and constraints to
backcasting
 In order to properly backcast data, the NA compiler
needs to have an important understanding of the
economic history of the country including:
•
•
•
•
•
•
•
Business cycles
Major economic events
Structural Changes
Changes in government policy
Knowledge of government programs
Demographic trends
Etc...
 A lack of understanding in these areas could result in
more damage than good.
8
Statistics Canada • Statistique Canada
Example – Provincial and Territorial
Accounts of Canada
Yukon Territory: Annual growth ratesReal household final consumption
expenditure
14%
12%
10%
8%
6%
4%
2%
0%
1982
1985
1988
1991
1994
1997
2000
2003
-2%
-4%
-6%
9
Statistics Canada • Statistique Canada
2006
2009
2012
Considerations and constraints to
backcasting
 Revising an integrated set of accounts is an additional
challenge in terms of:
•
•
•
•
Project management
Time constraints
Quality assurance
Scope
 For example, in Canada the benchmark estimates for GDP are
established when compiling the supply-use tables. When conducting a
historical revision it is not realistic to backcast detailed supply-use
tables given the source data and time constraints. At the same time
there is a desire to preserve the integrated nature of the accounts—it
may not always be possible to preserve the same level of integration
when backcasting NA data.
10
Statistics Canada • Statistique Canada
Backcasting techniques
•
11
There are many different ways to backcast NA data.
Some possible approaches include:




Linking
Recompilation
Estimation / modelling
Metadata approach

Most likely backcasting will involve a combination of the
above techniques—the challenge is to understand when a
particular approach is appropriate.
Statistics Canada • Statistique Canada
Possible techniques – linking
• Approach
 Linking datasets requires that the new NA dataset and the old
NA dataset overlap for at least a single period.
 The level associated with the new NA dataset is backcast using
the movements of the vintage NA dataset.
• Implications
 The advantage of this approach is that it is straightforward to
implement and does not require any vintage base statistics.
 There are major drawbacks to this approach in that it assumes
that the impact of the changes in the overlap period are the
same for the entire period being backcast. In addition, it could
result in changing the amplitude of the business-cycle and the
weight of certain series in an aggregate.
12
Statistics Canada • Statistique Canada
Possible techniques – linking
Business investment - R&D
Linked 2001
Business investment - R&D
Linked 1997
250,000
160,000
140,000
200,000
Current Prices
100,000
80,000
60,000
150,000
100,000
40,000
50,000
20,000
-
Business Investment (Recompliation)
-
Mar-81
Apr-82
May-83
Jun-84
Jul-85
Aug-86
Sep-87
Oct-88
Nov-89
Dec-90
Jan-92
Feb-93
Mar-94
Apr-95
May-96
Jun-97
Jul-98
Aug-99
Sep-00
Mar-81
Feb-82
Jan-83
Dec-83
Nov-84
Oct-85
Sep-86
Aug-87
Jul-88
Jun-89
May-90
Apr-91
Mar-92
Feb-93
Jan-94
Dec-94
Nov-95
Oct-96
Current Prices
120,000
Business Investment (Recompliation)
Business Investment (Linked)
Business Investment (Linked)
13
Statistics Canada • Statistique Canada
Possible techniques – recompilation
• Approach
 This requires re-compiling the entire NA database (or parts
of it) from it’s elemental level using source data or proxy
information.
• Implications
 The advantage of this approach is the high quality
estimates it produces.
 The drawback of this approach is the time and cost, as
well as the requirement for detailed source data, which, if
available, may not always exist in a consistent time series.
14
Statistics Canada • Statistique Canada
Possible techniques – recompilation
Growth rates – Real Gross Domestic Product
(before and after the CSNA 1997
comprehensive revision)
3.5%
3.0%
2.5%
% Change
2.0%
1.5%
1.0%
0.5%
-0.5%
Jun-69
Oct-69
Feb-70
Jun-70
Oct-70
Feb-71
Jun-71
Oct-71
Feb-72
Jun-72
Oct-72
Feb-73
Jun-73
Oct-73
Feb-74
Jun-74
Oct-74
Feb-75
Jun-75
Oct-75
Feb-76
Jun-76
Oct-76
Feb-77
Jun-77
Oct-77
Feb-78
Jun-78
Oct-78
Feb-79
Jun-79
Oct-79
0.0%
-1.0%
CSNA 1968
15
CSNA 1993
Statistics Canada • Statistique Canada
Possible techniques – estimation/modelling
• Approach(es)
•
•
•
•
Using auxiliary information: for example, using balance sheet data and assumptions
about rates of return to estimate income flows
Balancing techniques / residual calculations: for example, calculating net operating
surplus of corporations using the expenditure side as the benchmark measure of GDP
Instrumental variable technique: for example, using employment and wage
relationships to back-cast industry production or productivity
Triangulation: using two or more of the methods above in order to do quality
assurance
• Implication
•
•
16
The advantage of this approach is that it is not data intensive or labour intensive and
does not assume the relationship in the link year holds for the backcasted portion of
the time series.
The disadvantage of this approach is that this can usually only be done for selected
variables. Also, the new estimates do not benefit from being constructed within a
framework as it assumes the relationships between the NA variables and instrumental
variables hold for the entire time period.
Statistics Canada • Statistique Canada
Possible techniques – estimation
Modelling (example)
Log index (2000 = 100)
6
5
4
3
2
1
0
-1
Year
1926
1941
GDPI
17
1956
1971
GDPI 1961 1986
Statistics Canada • Statistique Canada
1986
2001
GDPI Projector
Possible techniques – metadata approach
• Approach
 The NSO does not attempt to link the NA databases, but
rather provides metadata explaining the differences between
the databases.
• Implications
 The advantages of this approach are its limited use of
resources and time
 The disadvantage of this approach includes difficulty for
users in compiling NA time series and the potential for
multiple versions of NA time series.
18
Statistics Canada • Statistique Canada
Mar-47
Dec-48
Sep-50
Jun-52
Mar-54
Dec-55
Sep-57
Jun-59
Mar-61
Dec-62
Sep-64
Jun-66
Mar-68
Dec-69
Sep-71
Jun-73
Mar-75
Dec-76
Sep-78
Jun-80
Mar-82
Dec-83
Sep-85
Jun-87
Mar-89
Dec-90
Sep-92
Jun-94
Mar-96
Dec-97
Sep-99
Jun-01
Mar-03
Dec-04
Sep-06
Jun-08
Mar-10
Dec-11
Sep-13
millions of dollars
Possible techniques – metadata approach
Nominal GDP - CSNA Vintages
2,500,000
2,000,000
19
1,500,000
1,000,000
500,000
-
CSNA 1968
CSNA 2008
Statistics Canada • Statistique Canada
CSNA 1993
-2%
20
Mar-59
Jul-60
Nov-61
Mar-63
Jul-64
Nov-65
Mar-67
Jul-68
Nov-69
Mar-71
Jul-72
Nov-73
Mar-75
Jul-76
Nov-77
Mar-79
Jul-80
Nov-81
Mar-83
Jul-84
Nov-85
Mar-87
Jul-88
Nov-89
Mar-91
Jul-92
Nov-93
Mar-95
Jul-96
Nov-97
Mar-99
Jul-00
Nov-01
Mar-03
Jul-04
Nov-05
Mar-07
Jul-08
Nov-09
Mar-11
Jul-12
Nov-13
% Change
Possible techniques – metadata approach
Nominal GDP Growth Rates - CSNA Vintages
8%
6%
4%
2%
0%
-4%
-6%
CSNA 1968
CSNA 1993
CSNA 2008
Statistics Canada • Statistique Canada
Possible techniques – metadata approach
(example)
Time series
Frequency
In use until
International
CSNA Manual or
manual or guide
Guide
1926-1996
Annual
1997
Orange Book
1947q1-1997q2
Quarterly
1997
SNA 1968
Orange Book
1961q1-2012q2
Quarterly
2012
SNA 1993
Blue Book
1981q1-present
Quarterly
On-going
SNA 2008
Red Book
Changes between CSNA 2008 and CSNA 1993
Classifications systems
Household Final Consumption Expenditure
North American Product Classification System
North American Industrial Classification System
Asset and Liabilities
Sectors
-Persons and unincorporated businesses, -corporations +household, +NPISH,
+financial corporations + non-financial corporations, + Aboriginal general
government
Asset boundary
Research and development, weapons systems
Major presentational changes
21
Statistics Canada • Statistique Canada
Lessons learned at Statistics Canada
 You need to be an economic historian
 Information management is important
 Saving for later is a bad idea:
•
•
•
•
It is difficult to “get back to it’
Fatigue prevents it
Key staff departures
Memory fades...
 The classification system of today is not the right
classification system for yesterday and will not be the
right classification system for tomorrow.
 Scale the project to avaliable Communicate,
Communicate, Communicate...
22
Statistics Canada • Statistique Canada
Lessons learned at Statistics Canada
“I made the following chart to
illustrate one of the great
frustrations that journalists,
economists and academics have
with StatsCan. One minute, the
agency, tasked with measuring the
tick tock of the economy and
society, tracks seemingly vital data
(such as detailed breakdowns of
public sector employment and
wages by all levels of government,
or the total value of government
transfer payments to persons by
province and type of transfer), the
next, *poof*, they’re terminated.
Good thing StatsCan still tracks
the square footage of fungi
production.”
By Jason Kirby
http://www.macleans.ca/economy/eco
nomicanalysis/the-case-of-thedisappearing-statistics-canada-data/
August 13, 2014
23
Statistics Canada • Statistique Canada
Questions for the AEG to consider
 Does the NA community need to work together to
establish best practices and general approaches to
backcasting?
 If so what should be included in such a guide?
24
Statistics Canada • Statistique Canada