BOP OF INDIA

BOPCOM-04/03
________________________________________________________________________
Seventeenth Meeting of the
IMF Committee on Balance of Payments Statistics
Pretoria, October 26–29, 2004
Towards A Revisions Policy for India’s
Balance of Payments Statistics
Prepared by Michael Debabrata Patra and Bhupal Singh
Reserve Bank of India- 2 -
TOWARDS A REVISIONS POLICY FOR
INDIA’S BALANCE OF PAYMENTS STATISTICS
Executive Summary
· Revisions policies, practices and studies form a key element of the Fund’s Data
Quality Assessment Framework (DQAF).
· Country practices reveal a wide diversity surrounding the key elements: regular,
well-established and transparent schedule of revisions; clear identification of
preliminary data; and public availability of revisions studies.
· A key driver of the formulation of revisions policy is user needs.
· Revisions policies for BoP statistics face a dilemma – incorporating more
accurate/complete information versus preserving data stability
· Extraordinary revisions overriding the established revisions cycle are common,
driven by detection of errors, significant revisions and methodological change.
· Historical revisions are difficult to identify and are typically associated with
countries which have stabilised their revision cycles.
· In India, several elements of a stable and consistent revisions policy were already
in place: consultations with key users; strong similarity with the country
experience; timeliness in preliminary data releases; clear identification of
preliminary and revised data; systematic dissemination of revised data.
· The lag structure of data reported for BoP compilation in India shows that the
provisional data are received within a time period ranging from 7 to 80 days from
the reference date. The maximum period up to which revisions occur is 14
months. The maximum lag of 24 months is reported for reinvested earnings of
FDI enterprises.
· The Mean Absolute Percentage Error (MAPE) for India’s gross current receipts
and payments in low, attesting to co-movement between revised and preliminary
data. Both the MAPE and the MARE attest to the stability and reliability of the
revised CAB. For CAB, the RMSRE and MRE coincide.
· The two measures for errors and omissions i.e., average absolute error (AAE) and
root mean square error (RMSE), indicate a high level of internal consistency in
India’s BoP data.
· Quantitative indicators of data quality need to be back-tested more rigorously.
They can only reinforce qualitative assessment of the data, not substitute for it.
· A clear understanding of the lag structures, institution of an electronic reporting
system for banking data (FET-ERS) and benchmarking national requirements
against the cross-country experience were key factors leading to announcement of
a revisions policy for India’s balance of payments data. - 3 -
TOWARDS A REVISIONS POLICY FOR
INDIA’S BALANCE OF PAYMENTS STATISTICS
Michael Debabrata Patra and Bhupal Singh
·
I. Introduction
This paper is intended to be a tribute to the seminal work undertaken in the IMF
to coagulate good practices in revisions of macro economic statistics and encourage their
wider adoption among national statistical authorities. Revisions policies, practices and
studies form a key element of the cascading structure of the Fund’s Data Quality
Assessment Framework (DQAF). More crucially, they form aspects of good governance
(Carson, Khawaja and Morrison, 2003). This, and other surveys of country practices
(IMF, 2002) reveal a wide diversity surrounding the key elements of a good revisions
policy contained in the DQAF: (i) regular, well-established and transparent schedule of
revisions; (ii) clear identification of preliminary data; and (iii) public availability of
revisions studies.
With the foregoing as its leit motif, this paper elucidates how processes of
governance in balance of payments (BoP) statistics in India reached critical mass, leading
to the announcement of a revisions policy on September 30, 2004 for the first time in
India (www.rbi.org.in). The next section presents the main findings of a cross-country
survey on revisions policy undertaken to learn and adapt. Section III sets out the specific
features of the data reporting system in India – timeliness, accuracy, length of revision
cycles, public awareness. Section IV attempts some quantitative assessment of the results
of recent revisions as a prototype of a revisions study. In conclusion, the paper sets out
the revisions policy for India’s BoP statistics which essentially forms out of this
accumulated analysis.

·
Michael Debabrata Patra is Adviser and Bhupal Singh is Assistant Adviser in the Reserve Bank of India.
The views expressed in this paper are those of the authors and not of the institution to which they belong.
The usual disclaimers apply. - 4 -
II. A Summary of the Country Experience
As Carson et al underscore, a key driver of the formulation of revisions policy is
user needs. Users – policy authorities, investors and financial market
participants/analysts, international organisations, researchers and the media – ask the
following questions
- How accurate are first releases?
- What is the likelihood of further revision, which way and by how much?
- What are the causes of revision?
- Is there a well-defined policy?
For most national authorities, the undertaking of revisions has two hovering
spectres - additional work and the discovery of mistakes in a world rapidly shifting under
their feet and necessitating generation of new data as well as conceptual and
methodological revisions. These fears emerge from the country studies as the major
impediments to publicly announcing revisions policies. Formally announced updating of
weights/base year, bringing in authentic/new data that is well known, statistical
refinement (seasonality, detrending, etc.,) changes in concepts/definitions/presentations
recommended by supra-national standard setting bodies constitute the virtuous segment
of the revisions policy motivation cycle. Here, resource constraints are clearly an overall
constraining factor.
Countries which have announced revisions policies for BoP statistics have
typically grappled with the horns of a dilemma – the desire to incorporate more
accurate/complete information into official data series versus the practical necessity of
preserving the stability of the data series since users interchange stability with accuracy
(Australian Bureau of Statistics 2002; Central Bank of Chile, 2003, Statistics New
Zealand, 2002). Accordingly, partial revisions to data for the current year are carried out
during the year with every monthly/quarterly release (Australia, Chile, Estonia, Italy,
Norway, New Zealand).
Final revisions are generally undertaken with long intervals, the earliest being
semi-annually (Australia), at the last quarterly release for the financial year (New
Zealand, Ukraine) and twelve-monthly (Chile). A common experience emerging out of- 5 -
the country experience is that revisions are a process, not an event. The underlying
drivers are arrival of late returns (New Zealand, Estonia, South Africa) different
frequencies and lags for sources (Chile, Korea), integration with national accounts (Chile,
Australia, Norway), replacing estimates with actual data (Chile, USA), correction of
errors (New Zealand, Chile) and new data sources/improvement in reporting or
estimation procedures/concepts (Chile, Italy, South Africa, Ukraine). Accordingly, only
a third of SDDS countries are reported to complete revisions in one year largely with the
hope that as time passes, the significance of revisions diminishes. Twenty per cent of
SDDS countries revise up to two years and 15 per cent up to three to five years (IMF,
2002).
Extraordinary revisions overriding the established revisions cycle are fairly
common in the country experience. The underlying causes range from the candid
detection of errors (New Zealand) to ‘significant revisions’ (Australia) and
methodological change (Chile, Estonia, Italy, Norway, Ukraine). More often than not,
the reasons for extraordinary revision are not clearly discernible.
In most countries, studies of revisions are either not conducted or are not made
public (IMF, 2002). Evidently, the desire to abstain from confusing users in the face of
scanty public interest in such analyses is a major inhibiting factor. In a few countries,
revisions studies are made available when major methodological changes occur or when
there is a ‘structural break’ due to new data collection methods (Italy). Some revision
analyses include cross-country comparisons (New Zealand). In one instance, revision
studies seemed to follow a regular pattern (Australia). In the majority of countries
examined, revisions are notified in the public release; in some cases, causes of revision
are explained in footnotes or accompanying text.
Information on the occurrence of historical revisions is difficult to identify in
country reports. Such revision exercises are typically associated with countries which
have stabilised their revision cycles (South Africa, Australia, New Zealand).
A summary of these findings from a sample of 23 developed and emerging market
economies is set out below (Table 1). - 6 -
Table 1: Cross Country Comparison of BoP Revision Practices
Country Pre-determined
Revision Cycle
Lag in Release of
Final Data
Escape Clauses
(Extraordinary Revisions)
(Reasons)
Argentina Yes 6 months ..
Australia Yes 4 years Significant revisions that are
important enough to require
immediate publication
Brazil Yes 6 months ..
Canada Yes 4 years Historical revisions
Chile Yes 15 months ..
Germany Yes 4 years ..
Hong Kong SAR,
China
Yes 2 years ..
Indonesia Yes 6 months ..
Israel Yes Updated each year ..
Italy Yes 13 months Methodological changes/new
collection system
Japan Yes 4 months ..
Korea Yes 6-7 months Conceptual and
methodological
Malaysia Yes 12 months ..
Mexico Yes 5 months ..
New Zealand Yes 18 months Significant revisions balancing
the need for stability in the
series and integrity of the
statistics
Philippines Yes .. ..
South Africa Yes 4 years Changes in historical data in
the light of new information
Sweden Yes Updated each year ..
Thailand Yes Updated each year ..
Turkey Yes Final at the time of
dissemination of annual
provisional data
Measurement issues
UK Yes 12 months ..
US Yes 6 years Major conceptual and
methodological revisions
.Not available ..
Source: IMF (2002), Revision Policy and Practice: A First Overview of Country Practices.- 7 -
At the other end of the spectrum are countries with no established revision
policies, inspite of considerable sensitivity to the need for such policies. Until September
2004, India belonged in this category. In the Euro area, lack of harmonisation across
members and different reporting requirements to data authorities are the key impediments
to a unified and publicly announced revisions policy. In Japan, the overarching desire to
ensure continuity of data series and thereby their credibility has considerably dampened
the need for a revisions policy even as recognition that it is an international best practice
has grown.
III. A Revisions Policy for India’s BoP
In India, the laying down of stable policies and practices for revisions in BoP data
has been an abiding concern. This has found expression in a mammoth revision exercise
covering data for the period 1950-81, primarily to incorporate reinvested earnings of FDI
enterprises and non-cash inflows of FDI Reserve Bank of India, (RBI, 1993). Although
revisions were not systematically conducted for subsequent years, an indigenous manual
on balance of payments compilation procedures was developed to provide an anchor for
latter day compilers to good practices of the past (RBI, 1987). A paper presented by the
RBI at the fifteenth meeting of the IMF’s BoP Committee in 2002 showed that several
elements of a stable and consistent revisions policy were already in place:
- consultations on revision practices with key users
- strong similarity with the country experience on during the year revisions
- timeliness in preliminary data releases
- clear identification of preliminary and revised data with explanatory
footnotes to Tables
- dissemination of revised data (RBI, 2003)
The principal factor that needs to be reckoned in the setting of a revisions policy
for BoP in India is the varying lags in arrival of data from source entities. The lag
structure of data reported for BoP compilation shows that the provisional data are
received within a time period ranging from 7 to 80 days from the reference date (Table
2). Partial revisions for different items of BoP occur at discrete intervals of varying- 8 -
periods, depending on the source of information. The maximum period up to which
revisions occur is 14 months for data reported by banks for receipts and payments against
merchandise exports/ imports, invisible and financial transactions. The maximum lag of
24 months is reported for reinvested earnings of FDI enterprises which are obtained from
the IIP. In summary, the lag in arrival of final data from different sources varies from 2
months to 24 months.
Table 2: Lags in Receipt of Data for BoP Compilation in India
Items
Lag in Reporting of Data (measured from reference
date) (No. of days/months approximately)
Provisional Partial Revisions Final
(First Release) Revision
Merchandise Trade, Customs 30 days 3 revisions covering 10 months 10 months
Banks’ Reporting of Receipts and
Payments of Merchandise, Invisible and
Financial Transactions
30 days Monthly revisions covering 14
months
15 months
Tabulations for Preliminary data release 80 days Occasional
Baggage Bullion, Customs Imports and
Grants routed through
Embassies/Consulates 30 days 2 months
Official Grants 30 days 2 months
Aid Receipts 60-75 days 12 months
Software Exports 60-75 days 3 revisions in 12 months 12 months
Foreign Direct Investment
Equity 60 days 12 months
Reinvested Earnings Estimates 24 months
Other Capital 80 days 12 months
Portfolio Investment in Stock Exchanges 15 days Occasional
Bond/Equity Issues in International
Stock Exchanges 30 days Occasional
External Commercial Borrowings 60-75 days 3 revisions in 12 months 12 months
Short-term Trade Credits 30 days 3 months
Non-Resident Deposits 60 days 3 months
Foreign Assets and Liability of Banks 20 days 3 months
Other Components of Banking Capital 60 days No revisions
Other Financial Transactions 60-75 days Monthly revisions covering 14
months
15 months
Foreign Exchange Reserves 7 days No revisions- 9 -
IV. Quantitative Analysis of Revisions
Revisions and reliability go hand in hand. Practical considerations emerging out
of the country experience suggest that the former can be sacrificed at the altar of the latter
in the short-run. In the longer run, however, there is a co-integrating relationship. Clarity
about rules and processes of revisions can only complement reliability, which is a
qualitative concept since information on macro economic statistics is a public good.
Reliability is quantitatively defined in the IMF’s DQAF as the closeness of the
initial estimated value to the subsequent estimated value. This involves the assessment of
revisions in terms of size and stability vis-a-vis the earlier releases of data. In this regard,
key indicators have been developed for the Euro area (Eurostat and European Central
Bank, 2003). These indicators are assessed in the Indian context for the period Q2 2000
to Q1 2004 for which revised data on India’s BoP were made available along with the
announcement of a revisions policy.
The basic assessment of the reliability of revised data is the simple calculation of
the differences from the preliminary data. This is set out in Charts 1 and 2 for gross
current receipts and payments which, in the case of India’s BoP, are subject to the
maximum amount of revisions. The graphs reveal close co-movement between
preliminary and revised data. For recent quarters, the revisions suggest that the
preliminary data had a modest downward bias.
Chart 1:Current Receipts: Revised and Preliminary Data
17000
22000
27000
32000
2000:Q2
2000:Q3
2000:Q4
2001:Q1
2001:Q2
2001:Q3
2001:Q4
2002:Q1
2002:Q2
2002:Q3
2002:Q4
2003:Q1
2003:Q2
2003:Q3
2003:Q4
2004:Q1
US $ million
Revised Data Preliminary Data- 10 -
Chart 2:Current Payments: Revised and Preliminary Data
18000
20000
22000
24000
26000
28000
30000
2000:Q2
2000:Q3
2000:Q4
2001:Q1
2001:Q2
2001:Q3
2001:Q4
2002:Q1
2002:Q2
2002:Q3
2002:Q4
2003:Q1
2003:Q2
2003:Q3
2003:Q4
2004:Q1
US $ million
Revised Data Preliminary Data
As the Joint ECB/Eurostat Task Force on Quality has pointed out, the evaluation
of revisions in terms of magnitudes hampers comparability across time, across different
variables and across countries. It is, therefore, necessary to employ a relative measure.
The Task Force proposed the Mean Absolute Percentage Error (MAPE) for gross data
which is expressed as percentage changes between revised and preliminary data as a ratio
of the revised data and averaged over time. Table 3 shows that the MAPE for India’s
gross current receipts and payments is low, attesting to the co-movement showing up in
the graphical representation.
Table 3: Mean Absolute Percentage Error Statistics
For Current Receipts and Payments in India BoP
Variable MAPE= 1/N åç(Ri- Pi)/Rçi , R= revised
data, P= preliminary data
Gross Current Receipts 0.029
Gross Current Payments 0.036
The current account balance (CAB) reveals a generally close association between
revised and preliminary data, except in Q1 2004 when methodological changes in the
process of collecting merchandise trade statistics at the customs frontier brought about a
significant upward shift and necessitated revisions in the BoP (Chart 3).- 11 -
Chart 3: Quarterly Current Account Balance: Revised and
Preliminary Data
-3000
-2000
-1000
0
1000
2000
3000
4000
5000
6000
2000:Q2
2000:Q3
2000:Q4
2001:Q1
2001:Q2
2001:Q3
2001:Q4
2002:Q1
2002:Q2
2002:Q3
2002:Q4
2003:Q1
2003:Q2
2003:Q3
2003:Q4
2004:Q1
US $ million
Revised Data Preliminary Data
For the CAB, the MAPE may not yield meaningful results since the CAB is a net
concept and expressed as the difference between exports and imports or foreign exchange
relating to these opposing flows of underlying transactions. The quantitative indicator for
the CAB as for other net transactions would have to express the difference in magnitudes
in relation to the variability of the revised series. Accordingly, as recommended by the
Task Force, the Mean Absolute Relative Error (MARE), the Root Mean Square Relative
Error (RMSRE) and the Mean Relative Error (MRE) were computed to assess the impact
of revisions on India’s CAB.
Table 4: Quantitative Indicators for India’s CAB
Indicator Value
MAPE 0.0430
MARE= 1/N å ç(Ri- Pi)/Var(R)ç 0.0002
RMSRE = Ö
-
1/Nå(Ri-Pi)
2
¤ Ö
-
1/Nå(Ø-Ri)
2
where Ø is the average value of R
0.4747
MRE = [å (R-P)
2
/å (Ø-R)
2
[
1/2
0.4747
Both the MAPE and the MARE attest to the stability and reliability of the revised
CAB indicated by the MAPEs for gross current receipts and payments. The RMSRE and
the MRE yield somewhat higher values, possibly associated with the choice of reference- 12 -
value for the revised series (denominator), since revisions are normally expected to
minimize deviations from the series average. Moreover, the RMSRE and the MRE
coincide, suggesting that the latter is a recommendation of the general case rather than an
alternative to the RMSRE.
Given the ambiguity reflected in the values obtained for RMSRE and the MRE,
data quality assessment was also undertaken in terms of the analysis of variance which
highlights the reasonable degree of stability of revisions. The t-statistics and F-statistics
relating to current account balances reveal that the variance of the revised series is not
significantly different from the original series (Table 5).
Table 5: Analysis of Variance of Preliminary and Final Data Series
on Current Account Balances
Sample period: 2000:Q2 to 2004Q1
t-Test: Two-Sample Assuming Equal Variances
t Statistics 0.7360
t Critical two-tail 2.0423
Null Hypothesis: Two series have the same variance
F-Test Two-Sample for Variances
F-Statistics 1.5069
F Critical one-tail 2.4034
Null Hypothesis: Two series have the same variance
Tests were also conducted for internal consistency, based on the errors and
omissions series. Revisions are expected to result in cancellation/reduction of estimation
errors and consistency in the direction of errors associated with individual items. The two
measures for errors and omissions (EO) recommended by the Task Force are computed
i.e., average absolute error (AAE) and root mean square error (RMSE), the latter being
amenable to further decomposition into bias and variance.
Table 6: Indicators for Errors and Omissions
AAE = å çEO ç/ (N+1), where EO refers to
errors and omissions as a proportion to
gross current receipts (figures in brackets
are EO expressed as proportions to gross
current receipts and payments
0.009
(0.00003)
RMSE = Ö
-
å(EO)
2
/(N+1) 0.011
(0.00550)- 13 -
Quantitative indicators of data quality need to be back-tested more rigorously and
widely before they can be adopted as standards in revisions studies. Above all, it needs to
be underscored that they can only reinforce ‘….. qualitative assessment of the data, not as
a substitute for it’ (Eurostat and the European Central Bank, 2003).
V. Conclusion
In India, a clear understanding of the lag structures associated with the key factor
– arrival of data – has been enabled by the institution of an electronic reporting system
for banking data (FET-ERS), primarily with a view to complying with SDDS
requirements for release of preliminary BoP data. Benchmarking national requirements
against the cross – country experience reveals reasonably close co-movement. The
Revisions Policy for India’s Balance of Payments Data pensively reflects the journey
along the learning curve:
“India’s balance of payments statistics are published as ‘preliminary’,
‘partially revised’ and ‘revised’ data. Preliminary data are quarterly and
are released with a lag of three months from the reference date (e.g., data
for the quarter ending March 2004 are available at the end of June 2004).
Preliminary data are subjected to some revisions during the year and
partially revised data are released with lags of six months, nine months
and twelve months from the reference date, alongside preliminary data
for the relevant quarter(s). Thereafter, the data are ‘frozen’ and final
revisions are incorporated in the revised annual data, which are released
within a lag of twenty-four months from the reference date.
Extraordinary revisions may be undertaken within this cycle in the event
of methodological changes in respect of data collection and compilation
procedures and/or significant changes indicated by data sources that
cause structural shifts in the data series. These extraordinary revisions
are documented at the time of release. Preliminary, partially revised and
revised data are clearly identified in the text and tables”(RBI, 2004). - 14 -
References
Australian Bureau of Statistics (2002), Revisions in Australia’s Balance of Payments
(BOP) Statistics, Fifteenth Meeting of the IMF Committee on Balance of Payments
Statistics, Canberra, Australia, October.
Carson, C.S., Khawaja, S and Morrison, T.K. (2003), Revisions Policy for Official
Statistics, 54
th
Session of the International Statistical Institute, Berlin, Germany, August.
Central Bank of Chile (2003), Revisions in Chile’s BOP Statistics, Sixteenth Meeting of
the IMF Committee on Balance of Payments Statistics, Canberra, Australia, December.
Eurostat and the European Central Bank (2003), Task Force on Quality: Report on the
Quality Assessment of Balance of Payments and International Investment Position
Statistics, Sixteenth Meeting of the IMF Committee on Balance of Payments Statistics,
Canberra, Australia, December.
International Monetary Fund (2002), Revision Policy and Practice: A First Overview of
Country Practices.
Reserve Bank of India (1987), Balance of Payments Compilation Manual.
Reserve Bank of India (1993), India’s Balance of Payments, 1948-49 to 1988-89.
Reserve Bank of India (2004), A Revisions Policy for India’s Balance of Payments Data,
September (www.rbi.org.in).
Shirley Nesbit (2002), Revisions in the New Zealand Balance of Payments, Fifteenth
Meeting of the IMF Committee on Balance of Payments Statistics, Canberra, Australia,
October.