An Update of the Awm Database 1

AN UPDATE OF THE AWM DATABASE[1]

This note documents the current update of the Area-wide Model[2] (AWM). The database used for the Area-wide Model (AWM) has evolved over the course of its development history. This is a major new release of the database. It has been constructed from both euro area Monthly Bulletin data and Eurostat data where available. It has then been backdated with aggregated country data from various sources. The changes compared to the previous database are mainly due to:

• the inclusion of Greece,
• new availability of data including ESA95 data
• revisions to historical data and
• interpolation of quarterly historical data using the Chow and Lin methodology.

The database covers a wide range of quarterly euro area macroeconomic time-series. The updated database starts in 1970q1 (for most variables) and is now until 2002q4. This note explains the method and procedures used to create the Area-wide Model database. The first section explains that the historical series for the AWM database are an aggregate of country data. The second section explains the method used to aggregate the historical series. The third section explains how the data for earlier periods are re-scaled to bring the data in line with the Monthly Bulletin data published by the ECB. The last section summarises the differences with the previous version. This note should be considered as a replacement of Annex Two of the AWM working paper.

1country data

The country series come from a variety of sources, but chiefly from Eurostat data, wherever possible. Unfortunately, Eurostat data does not exist for all countries for all series; therefore a number of other sources were also used – chief amongst these was the OECD National Accounts or Main Economic Indicators, or where necessary the BIS database. In addition, series from the European Commission (AMECO) database were also used, in particular for the fiscal variables. Recourse to data not in the public domain has been limited to the bare minimum. Where one source does not provide data for a series from 1970q1 then two series are combined to create a longer historical series. In this case, the series are re-based to the same year (i.e. 1995) and then joined.

1.1 Seasonal adjustment (working day adjustment)

The AWM database consists of seasonally adjusted data (except for HICP which is non-seasonally adjusted). The country data is normally seasonally adjusted. If only non-seasonally adjusted data is available for an original country series then the series is seasonally adjusted using the SABL method (SABL = Seasonal Adjustment, Bell Labs).[3]

1.2Treatment of German Reunification

For the majority of variables, the whole (East plus west) German series start in 1990 or 1991. In these cases, the West German series is used as the historical series. In order to remove the break in the joined series, the West German series have been re-scaled to the new whole German series, by the ratio of the two series at the start date of the whole German series.

1.3 Base years

Variables are re-based to the year 1995. The HICP base year is 1996.

1.4 Updating the databases

Each update of the database is frozen, and any improvements, changes or updates are encompassed in the next version of the database.

2aggregation method

2.1Index method

The method of aggregation used for most variables is the so called “Index method”. The log-level index for any series X is defined as follows:

ln Xz = z w i .ln Xi

where w is the weighting vector. A full explanation of this method can be found in Fagan and Henry’s paper “Long run money demand in the EU: Evidence for area-wide aggregates”. This method is used for both the nominal and real national accounts variables. The deflators are subsequently derived. This method is also used for GDP income variables (e.g. compensation to employees and disposable income) as well as for HICP and components.

2.2Weighted sum

For some other variables, for example ratios, the aggregate is simply calculated as a weighted sum of the variables (without expressing in logarithms). Variables created using this method include: the trade balance as a ratio of GDP; the current account balance as a ratio of GDP, and interest rates.

2.3Simple sum

Finally, there are some series that are just summed, e.g. employment and unemployment.

2.4Weights

The weights used in aggregating most of the individual country series are constant GDP at market prices (PPP) for the EU12 for 1995. If not all countries are available then the weights are re-scaled from the original EU12 weights. For HICP variables, HICP weights are used.

Table 2.1Weights used in aggregation

3Interpolating annual to quarterly

For GDP and components (real and nominal), compensation per employees, household disposable income and employment, two databases were produced, one annual data which is the aggregation of all the countries, and a second quarterly database which is the aggregation of those countries for which quarterly data exists. Using the aggregated quarterly data as an indicator, the aggregated annual data was converted into quarterly.

This has been implemented as Kalman filters, which is similar, but not necessarily the same interpolation as Chow-Lin (as done in Eurostat). The idea underlying the procedure is that the rate of growth of indicators approaches the true underlying rates of growth of true quarterly values. It is assumed that underlying quarterly series grow at a similar pace to indicator series, and that they precisely add up to the original annual aggregate. Another routine performs the identity-adjustment step wherever necessary. This is again done by solving a Kalman filter in which the true underlying components of an aggregate add up to this aggregate, and are imprecisely measured by the initial quarterly version of the components. In practical terms, the underlying adjusted series are assumed to grow at a pace similar to that of the initial (interpolated) quarterly estimates. Besides, an identity linking all the series must hold precisely.

For other series, either quarterly data existed for all countries or the Spline technique was used to interpolate the country data into quarterly (for example for all of the fiscal series).

4re-scaling of area-wide data to monthly bulletin data

As a general principle euro area data available in the ECB Monthly Bulletin and in Eurostat has been used. This has been done to ensure consistency between the data used in the model, and the data used in the ongoing monitoring of the area-wide economy. This was achieved by either replacing original country by published euro area series, or by linking them with the data contained in the original AWM database, i.e. the database including only information directly aggregated from country data. The linking takes, as a general rule, the available euro area data from the Monthly Bulletin or Eurostat as far back as possible. This series was then joined with the aggregation of country data.

The variables are re-scaled as follows.

• Real GDP and components are taken from Eurostat, the original source of the corresponding Monthly Bulletin data; they are then backdated with rates of growth of the AWM’s original series.

• GDP deflators are taken directly from the corresponding Monthly Bulletin series (which are calculated by ECB staff as a weighted average of the national deflators using PPP weights[4]), and are linked to the older AWM deflators.

• The Unemployment rate is taken from Eurostat, the same series as reported in the Monthly Bulletin. Backdating is done, as for the rest of the variables.[5]

• Total employment/employees, total compensation to employees and gross operating surplus are taken from the Monthly Bulletin and backdated in rates of growth.

• HICP and components are taken from the Monthly Bulletin and backdated in rates of growth.

• Fiscal series, in the form of ratios over GDP, are taken from the Monthly Bulletin and interpolated.[6] The series are then backdated with aggregated interpolated country data from the AMECO database. These ratios are further treated to adapt them to the accounting framework of the model.

• Finally, interest rate data are taken from the Monthly Bulletin. They are backdated with the corresponding series contained in the original database (source: BIS and AMECO). Again it should be noted that the aggregation weights were changed according to the number of countries available for each period.

Outside these broad categories, officially published area-wide series are still relatively scarce. As a result, it is necessary, for the purpose of the AWM, to use data which is only available in the original database of the model.

In the AWM (as in the Eurostat national accounts data), exports and imports of goods and services are a gross concept (i.e. do not net out intra-area trade flows). While, in principle, this does not affect net trade and other ‘balance’ items of the current account, it does mean that both export and import figures overstate significantly the true trade of the area (since intra-area trade accounts for about half of gross exports). For the time being, this is dealt with by explicitly including euro-area final demand as a variable in the aggregate export equation. Work is currently ongoing to disaggregate trade into intra and extra euro area trade and to re-estimate the trade side.

5UNITs

The units of the series generally follow Eurostat:

• Real GDP and components are in millions of euros/ecu with base year 1995.

• Nominal series typically millions of euros/ecu, including compensation to employees, and gross operating surplus.
• Deflators are generally set to 1.0 in 1995 (with the exception of YFD).
• HICP and components are an index with base year 1996=100.
• Employment/Employees are thousands of persons.
• Unemployment rate is ratio to the civilian workforce.
• Commodity prices, World GDP are in dollars.

6comparison with the original published awm database

The major differences compared to the database released with the ECB working paper number 42 are the following:

• Historical series have been updated to include Greece to the extent possible. Hence the data set should be considered an EU12 data set.

• For nearly all countries ESA95 data was used (where available).

• Historical data has in many cases been revised since the first release of the AWM database.

• Additional variables have been added, including a breakdown of employment into employees and HICP into HICP excluding energy and HICP energy.
• Some variables have been re-scaled, notably the levels of Stocks.

• Some variables that were in the original database are no-longer part of the dataset, - for example M3.

5summary

This note explains the new release of the AWM. This database has taken publicly available ECB Monthly Bulletin data and Eurostat data, supplemented with aggregated country data. Whilst the differences with the initial AWM database appear to be not so large there are some differences, especially for the fiscal variables. The differences are mainly due to changing trends in the series because of differences in the construction of ESA79 data and ESA95 data. This database is in the process of being used at the ECB for the AWM. Preliminary work suggests that, typically only a re-scaling of the constants of the stochastically estimated equations in the AWM is needed.

6ANNEX – LIST OF VARIABLES

Attached

Page 1 of 6

[1] For questions on construction of the database please email: .

[2] For a description of the model see ECB working paper No. 42: ‘An Area-wide Model (AWM) for the euro area’ by Gabriel Fagan, Jérôme Henry, and Ricardo Mestre (January 2001).

[3]An overview of this method can be found in Cleveland, Devlin and Terpenning, “The SABL Seasonal and Calendar Adjustment Procedures”, Time Series Analyses: Theory and Practice 1. Proceedings of the International Conference held at Valencia, Spain, June 1981. Ed. O.D. Anderson, J.G. De Gooijer and K.D.C Stoodley. Amsterdam: North-Holland Publishing Company, 1982. pp 539-564.

[4] This procedure is necessary since the published Eurostat figures for nominal GDP and its components are expressed in terms of the current exchange rate (in ECU terms) which implies that, for earlier years, the implicit deflators calculated from the Eurostat data would be distorted by exchange rate movements.

[5] Labour force re-scaling is less troublesome if the unemployment rate is linked in rates of growth.

[6] Interpolation of series done by Pablo Garcia-Luna DGE, ECB.