Restoring the Credibility of Japanese GDP Statistics

In December last year, Japan made a drastic downward revision to its quarterly gross domestic product (GDP) for July-September, while in fiscal 2008, an estimation error occurred in the revised figures. Such incidents have been undermining confidence in Japan's national accounts statistics. In a discussion held in February between the Statistics Commission and economists from the private sector, the GDP statistics came under sharp criticism from some participants, who made comments such as: "The current GDP estimations may seriously damage international trust in Japan's economic statistics and even in the Japanese government" and "economists rely on the Indices of Industrial Production rather than GDP figures."

The "Basic Plan Concerning the Development of Official Statistics (Basic Plan)" approved in a Cabinet meeting last year also points out that Japan's System of National Accounts (SNA) statistics require fundamental reform and lists over 40 issues that need to be addressed within the coming five years. This article examines the issues and future perspectives concerning Japan's SNA statistics.

* * *

GDP can be calculated from three sides, namely, the expenditure, production, and income sides of the economy, and in theory the figures from the three sides should be equal. Therefore, we can improve the accuracy of the estimations by estimating GDP using all three approaches and cross-checking the results.

However, Japan's annual GDP figures are estimated without taking this principle of equivalence into account. That is to say, the difference between GDP estimates from the expenditure side, which measure consumption, investment, etc., distinguishing how much of the supplied goods and services by detailed category are for intermediate and final use, and estimates from the production side, calculated by subtracting intermediary inputs from final output in each industry, is simply treated as "statistical discrepancy."( table )

Now look at the trend in the statistical discrepancy - calculated by subtracting the value of GDP from the production side from the value of GDP from the expenditure side - as a percentage of GDP from the expenditure side. From this, we can see that the discrepancy was small around fiscal 2000, the benchmark year, in which estimates were made based on information from the input-output table by the Ministry of Internal Affairs and Communications and other sources ( figure ). However, since then, the discrepancy has been on an upward trend. This may be due to the increasing inaccuracy and inconsistency of information on intermediary input and other factors. In fiscal 2008, the discrepancy exceeded 11 trillion yen, which is equivalent to 2.3% of GDP from the expenditure side.

Figure : Statistical discrepancy as a percentage to GDP from the expenditure side (n a fiscal year basis)

(Source)System of National Accounts, Cabinet Office, Government of Japan

(Note)Figures in the table are on a calendar year basis. Figures in the chart ae on a fiscal year basis.

In studies in the field of quantitative economic history estimating GDP for the pre-war period, GDP is estimated separately both from the production and the expenditure side, and the results are considered satisfactory if the difference is within 5%. The chart below reminds us of these kinds of historical statistics rather than the SNA statistics of a developed country.

Efforts to identify the sources of the difference and to reduce the discrepancy through the adjustment of estimates from the expenditure and production sides are not satisfactory. In addition, looking at GDP estimates from the income side, operating surplus and mixed income, which are income components, are calculated by subtracting employee compensation and estimates of other types of income from output estimates. This means that GDP estimated from the income side will always be equal to GDP estimated from the production side and therefore do not represent independent estimates, as can be seen in the table above.

In principle, obtaining GDP estimates from the expenditure, production, and income sides that yield the same result should be possible through the estimation of a system of consistent input-output tables. In fact, many other advanced countries prepare such input-output tables on an annual basis. These are then used to cross-check primary statistics on expenditure, production, and income to produce GDP estimates with no statistical discrepancy. However, in Japan, although production and input matrices by economic activity used for the SNA are estimated in a simple fashion, they are used mainly for GDP estimates from the output side and are not used for cross-checking GDP estimates from the expenditure, production, and income sides.

In this context, another point worth noting is the marked decline in Japan's saving rate from 7.5% in fiscal 2007 to 2.0% in fiscal 2008 suggested by the SNA statistics. The saving rate is calculated by subtracting final consumption expenditure from disposable income, etc., and dividing the result by disposable income. While final consumption expenditure is a component of expenditure GDP, Japan's disposable income estimates are based on GDP from the production side. It is therefore possible that part of the apparent decline in the saving rate may actually reflect the increase in the statistical discrepancy.

These observations suggest that it would be desirable, as stated in the Basic Plan, to estimate annual input-output tables for the SNA and move to a system that allows for the consistent estimation of GDP from the three sides.

* * *

Meanwhile, quarterly GDP at present is estimated based on the expenditure approach by gauging the supply situation by product category using statistics such as the Current Survey of Production and taking into account demand-side information from sources such as the Family Income and Expenditure Survey. The quick (first preliminary) estimate is announced within six weeks after the end of the quarter, followed by the second (revised) preliminary estimate announced within 70 days after the end of the quarter, the final estimate within eight months of the end of the fiscal year, and the revised final estimate within 20 months of the end of the fiscal year.

However, the estimates often involve drastic revisions. In December last year, for example, the annualized quarterly growth rate of real GDP for July-September 2009 was revised from 4.8% in the first preliminary estimate to 1.3% in the second preliminary estimate ? a downward revision of 3.5 percentage points (the figure was further revised to ?0.6% in the latest revision). But this was not out of the ordinary, as revisions of the first preliminary estimate of the annual growth rate in the order of two percentage points have often been made in the past.

Moreover, there are also large discrepancies between the second preliminary estimates and the final and revised final estimates. According to Mr. Katsuaki Ochiai, Associate Senior Economist at the Japan Center for Economic Research, the average difference (absolute value) between the first and second preliminary estimates of the annualized quarter-on-quarter growth rate from January-March 2005 to October-December 2009 was 1.0 percentage point, and that between the second preliminary estimates and the revised final estimates (from January-March 2005 to October-December 2007) as high as 2.3 percentage points. These discrepancies are about twice as high as those in the United States for the same period (0.6 and 1.1 percentage points).

In February, the Cabinet Office announced plans aimed at restoring confidence in the GDP statistics. Key components include (1) increasing staff levels in the relevant sections; (2) reviewing the estimation methodology, in particular with regard to private investment in plant and equipment and inventory; (3) revising the seasonal adjustment method; and (4) implementing more drastic measures in the coming 2-3 years.

To improve the quality of the estimates, it is also essential to make a sustained effort to reform the primary statistics on which the quarterly estimates are based. Fundamental demand-side statistics that play an important role in the preliminary estimates, such as the Financial Statements Statistics of Corporations by Industry, Quarterly, and the Family Income and Expenditure Survey, are subject to fluctuations caused by small sample sizes and changes in samples. Moreover, there are large discrepancies between the Current Survey of Production used for the quarterly estimates and the Census of Manufactures used for the annual estimates. As these examples indicate, there are many issues that need to be resolved with regard to primary statistics.

Furthermore, it is necessary to increase the accuracy of quarterly estimates by estimating GDP from the production side using input-output tables and cross-checking expenditure and production approach estimates. This was also highlighted in a report by the International Monetary Fund (IMF) in March 2006 assessing Japan's macroeconomic statistics, which recommended that the quarterly preliminary estimates should be expanded to include production series.

Since preliminary estimates are prepared based on limited fundamental data and under time constraints, it is important to filter out noise in order to improve the accuracy of these estimates. When I recently spoke with the staff responsible for GDP estimates at the U.S. Bureau of Economic Analysis, I was impressed not only by their well-established estimation system, but also by the emphasis they put on the knowledge and resources necessary to identify and resolve issues through interviews and micro-data whenever they find any doubtful primary statistics.

* * *

The 490-page manual for estimating expenditure GDP published in 1976 and other bulky volumes at the time (which summarily came to be labeled "Estimation Materials for the New SNA Statistics") convey the fresh enthusiasm of those in charge, perhaps because these volumes were published shortly after the adoption of the new SNA system (1968 SNA). However, Handbook on the Methodology for the Estimation of SNA Statistics" currently made externally available devotes only 27 pages to the explanation of the methodology for the estimation of expenditure GDP. As various revisions have been made over time, the current estimation system may have deteriorated in terms of consistency and in terms of having a systematic manual that forms the basis for such estimation.

Under the Basic Plan, some difficult issues are being addressed such as the estimation of Financial Intermediation Services Indirectly Measured (FISIM) and stock-based statistics using the current value of capital. But the most urgent need is to accelerate fundamental reform of the estimation system.

The biggest obstacle for achieving this objective is the lack of human resources. The Department of National Accounts of the Cabinet Office, which is the section in charge of estimation of SNA statistics, has only 58 staff, and that is after the increase in staff levels announced in February. This number is far lower than those in Britain and Canada, which employ over 200 staff, or the United States, France and Italy, which each employ more than 100 staff. The SNA statistics represent one of the most important tools for the formulation of economic policy. Against this background, fundamental reform and greater resources for the preparation of reliable GDP statistics is highly desirable.

>> Original text in Japanese

* Translated by RIETI.