Singapore’s Advance GDP Estimates International Seminar on Timeliness, Methodology & Comparability of Rapid Estimates of Economic Trends 28 May 2009.

Singapore’s Advance GDP Estimates

International Seminar on Timeliness, Methodology & Comparability of Rapid

Estimates of Economic Trends

28 May 2009

2

Outline

Compilation of Output-based Quarterly GDP Estimates Timeliness Methodology

Assessing the Quality of Advance GDP Estimates Methodology Dataset Results

Conclusion

Compilation of Quarterly Output-based GDP

Estimates

4

Compilation Cycle

Incomplete and limited data

Detailed disaggregation may not be possible

More disaggregation possible. Quarterly and earlier annual estimates revised

Estimates reconciled and benchmarked with I-O tables

Advance QtrlyEstimates

Prelim QtrlyEstimates

AnnualEstimates

PeriodicRebasing

5

Jan Feb Apr MayMar

Compilation cycle for 1Q GDP

estimatesAdvance

Released not later than 2 weeks after end of reference qtr

Preliminary

Released 8 weeks after end of reference qtr

Example: Advance 1Q09 is released on 14 Apr 09 Preliminary 1Q09 is released on 21 May 09

Timeliness

6

Industry BreakdownAdvance GDP Release Preliminary GDP Release

Timeliness Not later than 2 weeks after end of reference quarter

Not later than 8 weeks after end of reference quarter

Industry Breakdown Overall GDP Manufacturing ConstructionServices Producing Industries

Overall GDP Goods Producing Industries

Manufacturing Construction Utilities Other Goods Industries

Services Producing Industries Wholesale & Retail Transport & Storage Hotels & Restaurants Information & Communications Financial Services Business Services Other Services Industries

Ownership of Dwellings

7

Use of Indicators for GDP Compilation

Base year (reconciled) nominal VA estimates

Constant Price

GDP

Current Price

GDP

Price Indicators

Volume Indicators Value Indicators

8

Methodology

Indicators used Examples

Deflated turnover Turnover estimates from monthly or quarterly industry surveys (e.g. catering trade, retail trade)

Deflated current price indicators

Progress payments for the construction industry

Volume indicators Container throughput, visitor arrivals, mobile call minutes

Input indicators Employment, wages

9

Methodology

Tools for compiling the Advance GDP estimates Forecasting

ARIMA forecasts generated by X12-ARIMA software (developed by US Census Bureau)

Excel Interface Allows quick and easy forecasting Multiple series can be forecasted simultaneously

Inputs from data providers/major industry players Professional judgement

10

Advance Estimates for Manufacturing

Forecasting

Inputs from data providers

Professional judgement

2 months of the Index of Industrial Production

Methodology

How the Advance Estimates for Manufacturing are compiled

Assessing the Quality of Advance GDP Estimates

12

Methodology

To assess the quality of Advance GDP Estimates using revision analysis

Examine:1) Whether Advance GDP is a biased estimate of

the Prelim GDP2) Whether information are efficiently used in the

Advance GDP

Revision refers to Prelim GDP – Advance GDP, i.e. later estimate minus earlier estimate

13

1) To examine whether Advance GDP under- or over-estimate Prelim GDP

a) Mean Revisions and its statistical significance (using HAC-variance-based t-test at 5 % level): where significant mean revisions imply possible under- or over-estimation in Advance GDP

Follows the approach described in Di Fonzo (2005)

Methodology

14

Methodology

2) To examine whether information are efficiently used in the estimation of Advance GDP

a. Correlation between revisions and earlier estimate, and its statistical significance: where significant correlation indicates that information are not efficiently utilized in earlier estimate, i.e. part or all of the revisions are corrections to earlier estimates, i.e. revisions reflect ‘noise’

15

b. Correlation between revisions and later estimate, and its statistical significance: where significant correlation indicates that part or all of the revisions reflect new information i.e. revisions reflect ‘news’

Follows the approach described in Mckenzie, Tosetto and Fixler

Methodology

16

Published 2002 Q4 – 2008 Q4 year-on-year growths of the GDP Advance Estimates (E) and the GDP Prelim Estimates (L): Total GDP Manufacturing Services Producing Industries

Dataset

17

Revisions to Total GDP Growth

18

Sample Size 25

Mean Revisions 0.3% Mean Rev is not significant at 5% levelHAC-based p-value 0.07

Corr( Rev,Advance) 0.3 No evidence of noise

Clear evidence that revisions are due to news

P-value 0.15

Corr( Rev,Prelim) 0.45

P-value 0.02

Revisions to Total GDP Growth

19

Revisions to Manufacturing Growth

20

Sample Size 25

Mean Revisions 0.7% Mean Rev is not significant at 5% levelHAC-based p-value 0.15

Corr( Rev,Advance) 0.22 No evidence of noise

Strong evidence that revisions are due to news

P-value 0.28

Corr( Rev,Prelim) 0.44

P-value 0.03

Revisions to Manufacturing Growth

21

Revisions to Services Growth

22

Sample Size 25

Mean Revisions 0.2% Mean Rev is not significant at 5% levelHAC-based p-value 0.36

Corr( Rev,Advance) 0.20 No evidence of noise

Clear evidence that revisions are due to news

P-value 0.35

Corr( Rev,Prelim) 0.41

P-value 0.04

Revisions to Services Growth

23

Conclusion

Advance GDP estimates are good early indicators of the aggregate economic activity

Advance GDP estimates are generally unbiased

Information is efficiently used in Advance GDP estimates. Revisions to Advance GDP reflect new information not available at the time.

24

References

Di Fonzo, T. (2005), The OECD project on revisions analysis: First elements for discussion, paper presented at the OECD STESEG Meeting, Paris, 27-28 June 2005 http://www.oecd.org/dataoecd/55/17/35010765.pdf

Mckenzie, Tosetto and Fixler Assessing the efficiency of early estimates of economic statistics, http://www.oecd.org/dataoecd/20/13/41009155.pdf

Thank you