REVISION AND AUTHORISATION HISTORY

Transcription

1 Pilot Study on the Estimation of Regional Capital Stocks A Final Report for the European Commission (Directorate General Regional Policy) 12 March 2010

2 REVISION AND AUTHORISATION HISTORY Version Date Authorised for/ release by Description /07/09 Ben Gardiner Interim Report /08/09 Ben Gardiner Revised Interim Report /11/09 Richard Lewney Final Report /03/10 Richard Lewney Final Report Cambridge Econometrics Covent Garden, Cambridge CB1 2HT, UK Tel Fax Website Cambridge Econometrics Limited is owned by a charitable body, the Cambridge Trust for New Thinking in Economics. ii

3 CONTENTS page 1 Introduction 1 2 The Perpetual Inventory Method (PIM) 3 3 Gross Fixed Capital Formation 13 4 Other Assumptions 35 5 Analysis of Results 45 6 Conclusion 69 7 References 71 Appendices Appendix A Base-year Capital Stock Estimation A-1 Appendix B RAS Procedure B-1 Appendix C Main Data Sources C-1 Appendix D Raw Data Availability by Country D-1 Appendix E Cohesion Regions E-1 iii

4 1 INTRODUCTION 1.1 Introduction This report is the final report of a project carried out by Cambridge Econometrics on behalf of DG Regional Policy to estimate the capital stock of the EU27 at NUTS2 regional level, or for as many of these countries as possible given the available data. The project was carried out in three stages between January 2009 and November In the initial stage the national statistical offices of the EU27 Member States were surveyed to ascertain what capital stock statistics they currently publish, at what geographic level and what methods and assumptions they use to produce them. This information was used to decide the method by which capital stock estimates should be produced for this project. The second stage tested this approach by calculating estimates for ten of the EU27 using the simplest form of the Perpetual Inventory Method which employs the simultaneous retirement of assets at the end of their service lives. The third and final stage of this project extended this method to fourteen more of the EU27 countries and introduced some more sophisticated, alternative retirement assumptions (the Winfrey S2 and Winfrey S3 retirement functions). This project therefore describes how the capital stock estimates were produced for 24 of the 27 EU Member States at NUTS2 level and presents some analyses of the results. The capital stock estimates were delivered in spreadsheet form 1 in three alternative sets corresponding to the three different retirement functions employed to create them. 1.2 Background The survey of European national statistical offices (NSOs) carried out in the initial stages of the project found that all responding countries that carried out capital stock estimation employed the Perpetual Inventory Method (PIM). However, while there is harmony between the surveyed countries in this respect, there is considerable variation in the assumptions that are used as inputs into this standard approach. Different NSOs employ varying assumptions regarding, for example, average service lives and mortality functions. The sector, institutional sector and asset-type disaggregation of capital stock estimates also vary from country to country, with disaggregation across these dimensions available for some countries but not for others. 1.3 Use of standard PIM It was therefore decided that the best way to produce comparable estimates would be to arrive at a standard set of assumptions which, as far as possible, are representative of the 1 Nominal Capital Stocks Base.csv, Nominal Capital Stocks ws2.csv, Nominal Capital Stocks ws3.csv, Real Capital Stocks Base.csv, Real Capital Stocks ws2.csv, Real Capital Stocks ws3.csv 1

5 most common approach taken by NSOs across Europe in the production of their own estimates, and to apply these standard assumptions to Gross Fixed Capital Formation (GFCF) data from Eurostat. By employing the most common, or modal, approach for each assumption, it is hoped that the estimates finally arrived at will achieve an acceptable amount of support from Member States. Such support is also more likely as the standard PIM is effectively endorsed by the OECD (OECD Manual, 2001) and many Member States of the EU are also members of the OECD. Employing these modal assumptions and carrying out a standard PIM approach using GFCF data from Eurostat should ensure that the final estimates are comparable in that exactly the same methodology will have been applied in every case to GFCF data from the same source in most cases. 1.4 Remaining chapters of the report Chapter 2 provides an example of the standard PIM that has been employed to produce capital-stock estimates at NUTS2 level for twenty-four EU Member States in this report 2. Chapter 3 describes the methods applied to the Eurostat data to arrive at a consistent GFCF time-series for each member state. Chapter 4 sets out the other assumptions used in the production of the capital stock estimates. Chapter 5 provides a comparison of the estimates produced for this project with those produced by some NSOs and as part of other projects at the national level. It also compares the NUTS2 level estimates produced for Spain as part of this project with those produced by Mas et al (2006). Chapter 5 uses the estimates in selected analyses (notably in the construction of capital-to-output ratios) to check their robustness. Chapter 6 presents conclusions. Appendix A contains a brief explanation of the method employed to arrive at a base-year capital stock estimate for each country on which to begin the PIM. Appendix B contains a description of the RAS procedure which has been used to fill missing values when preparing some of the GFCF data. Appendix C notes the data sources used to produce the estimates. 2 Estimates for Greece, Malta and Bulgaria could not be produced due to an absence of data on which to create a starting base-year estimate for

6 2 THE PERPETUAL INVENTORY METHOD (PIM) 2.1 Introduction There is no single correct way to measure the capital stock. Since the purpose for which estimates are produced varies between countries, so does the method used for producing the estimates. Different countries employ approaches that are similar but which are designed to produce capital stock estimates that are adequate for their own purposes. There is, however, increasing convergence towards internationally standard measures with many countries now adopting the standard application of the Perpetual Inventory Method (PIM). Indeed, the standard PIM, as described in the OECD Manual (2001), is that which is most commonly employed across the EU, as evidenced by the survey of National Statistical Offices presented in the inception report for this project. A version of this standard PIM was applied to create the capital-stock estimates produced as part of this project. Using this approach, the gross capital stock is, simply put, made up from accumulated investment at historical prices with a linear depreciation applied in order to estimate capital consumption and arrive at the net capital stock. In addition, when producing estimates for this final report three different retirement functions have been applied: the Winfrey S2 function, the Winfrey S3 function and simultaneous exit. In each case the retirement function has been integrated with linear depreciation. This chapter describes below the method by which this integration was achieved. However, for further information on these retirement functions and the method by which they are combined with depreciation see pages of the most recent version of the OECD Manual produced in January 2009 (OECD Manual, 2009). Chapter 4 of this report demonstrates that linear depreciation is by far the most common depreciation assumption employed by EU national statistical offices (as shown by the survey evidence collected as part of the initial stage of this project). By employing the linear depreciation assumption for this report, we are therefore replicating the depreciation approach taken by the majority of EU countries. Furthermore, it has been shown by both Maddison (1992) and Ward (1976) that the linear depreciation assumption produces estimates that are an adequate reflection of reality. Indeed, as is shown in Chapter 5, employing a simple linear depreciation pattern alongside simultaneous exit already replicates well the estimates at the national and NUTS2 level produced by NSOs and other organisations even without additional, more sophisticated assumptions, for example for retirement. Ward (1976) has stated that for most assets, especially those with long lives and those like buildings where the services they provide do not change materially over time, the conventional straight-line estimate of annual factor services used up seems appropriate. Maddison (1992) has commented similarly and has also employed a linear depreciation pattern as noted in Verbiest (1996). Based on these conclusions, Verbiest (1996) has produced capital-stock estimates for the Netherlands national statistical office using linear depreciation. Statistics Netherlands (1997) provides evidence of this. 3

7 In this chapter we define the general concepts and terms used in capital stock measurement, and this provides a framework for our subsequent discussion of the PIM and its different applications. The chapter ultimately presents a mathematical formalisation of the standard PIM employed to construct capital stock estimates in this report. The method followed in this report is similar to the method recently employed by TNO to produce NUTS2 estimates for the Netherlands. Following the formalisation of the standard PIM, the chapter goes on to describe the Winfrey S2 and Winfrey S3 retirement functions and how these are combined with linear depreciation to produce capital stock estimates. 2.2 Glossary List of concepts and terms Gross capital stock Net capital stock Benchmark capital stock Rentals Average service life Values all assets in use as if they are still new (at the current price for new assets of the same type) regardless of age or condition. The as new prices are obtained for assets acquired in earlier time periods by revaluing them using price indices for the relevant type of asset. Assets are valued at their market prices (ie the as new price) minus an estimate for the consumption of fixed capital since purchase, which is in turn a reflection of the asset s declining productive capacity as it ages. An initial total capital stock from which investments and retirements can subsequently be added and removed and to which price indices can be applied resulting in the capital stock for subsequent periods. The quantity of annual capital services generated by an asset multiplied by the price of those services. Under the alternative Perpetual Inventory Method the rentals accrued from the asset are then discounted to provide the asset s value in a given year. The average number of years for which an asset is expected to be productive. Mortality function (also referred to as the retirement pattern) The method by which assets are assumed to exit the capital stock in relation to their assumed average service life. This is important because in reality some assets will not reach this expected average service life and will be retired earlier, while some assets will continue to be productive some time after this average. Discount rate Using the more sophisticated alternative PIM, future income streams (capital services) derived from the use of the asset are discounted by a discount rate in order to convert them into their present value. The discount rate employed is often the interest rate on long-term bonds (OECD Manual, 2001, para. 2.4). 4

8 Gross fixed capital formation Investment in new capital minus disposals of capital that has come to the end of its service life. 2.3 The Perpetual Inventory Method The standard PIM A benchmark estimate of capital stock in the chosen year The estimation of the gross capital stock via the standard Perpetual Inventory Method requires five inputs. As noted in subsequent chapters, there is some variation among the EU27 in the way these five inputs are constructed. These five inputs, and the common ways by which they can be constructed, are described in the following subsections. As indicated in the OECD Manual (2001, para. 6.7), it is possible to build estimates of the capital stock without having an initial benchmark on which to start. However, this requires records and investment time-series which stretch back 100 years or more. This is because the longest-lived assets, usually buildings and other structures, often have service lives in excess of 100 years. For this reason, all countries in the EU that produce capital stock estimates, whether at regional or national levels, start with an initial capital stock benchmark. The most popular sources for constructing this initial benchmark are population censuses, fire insurance records, company accounts, administrative property records and share valuations. None of these sources provide a highly accurate initial benchmark. However, the inaccuracy associated with the initial benchmark works its way out of the capital stock estimates over a number of years due to the depreciation and eventual retirement of the assets associated with the initial benchmark. For this reason it is important that the period of interest for which the capital stock estimates are to be employed (in most cases, presumably, the current period) is some years after the initial benchmark. For this pilot study the initial benchmark is 1995 and this is considered to be a sufficient number of years from the period of use (approximately 2010) for many of the errors introduced as part of initial benchmarking to have worked their way out of the country-specific estimates achieved. Data on Gross Fixed Capital Formation which extend back to the chosen base year An obvious requirement for the production of capital stock estimates is a time-series of the flow of Gross Fixed Capital Formation (GFCF). The description Gross can be misleading. It is net of disposals of fixed assets, but it is referred to as Gross because it does not yet incorporate any depreciation. The depreciation estimates are applied in the capital stock estimation process to which GFCF is an input. Asset price indices with which to re-value the accumulated gross fixed capital formation from previous years As noted earlier, the capital stock estimation process requires the initial benchmark on which the estimate is based to be some years prior to the period for which the statistics are eventually to be used. However, this poses the problem that an estimate of capital stock from, say, ten years ago must be deflated to reflect the changing value of money. This is very important as not to carry out this revaluation exercise could lead to very misleading results. It would be impossible to say whether an increase in the capital stock actually represented an increase in the volume of capital in place and contributing to 5

9 production in the region in question, or whether the apparent increase was simply due to changing prices. For this reason it is important to employ price deflators that are as accurate as possible in order to gain a true picture of the actual increasing and decreasing volume of capital. The OECD Manual (2001, para. 6.17) states that if errors are introduced through the application of incorrect price deflators, it is possible for these to affect the final results as much as would the use of the wrong average service life estimates and mortality functions. In order to achieve the required accuracy it is highly advisable to apply separate deflators for each asset type for which the capital stock is to be estimated. An estimate of average service life for each type of asset Common sources of information on asset service lives are tax authorities, company accounts, empirical surveys and expert advice from firms that employ the asset in the production process. The OECD Manual (para. 6.48) suggests that service life assumptions have a significant impact upon the capital stock estimate finally arrived at. Indeed, some capital stock economists have gone as far as to suggest that service life assumptions are the only relevant parameter in the PIM (Meinen et al, 1998, page 3). The impact of differing service life assumptions on the resulting capital stock estimate has been demonstrated by a number of sensitivity studies. The OECD Manual (2001, para. 6.48) lists the results of two major sensitivity studies on the impact of differing service life assumptions on capital stock estimates as: longer service lives always increase the size of the gross capital stock longer service lives usually reduce consumption of fixed capital longer service lives usually increase the size of the net capital stock and by relatively more than in the case of the gross capital stock longer service lives have an unpredictable effect on growth rates longer service lives reduce the volatility over time in the growth of stocks and capital consumption A Canadian study (OECD Manual, 2001, para. 6.42) varied average service life estimates for asset types by +/- 50% and found that increasing average service life estimates by 50% results in a 40% increase in the eventual capital stock estimate arrived at. A similar effect results from decreasing service life estimates by this much. This study drew the conclusion, however, that most service life assumptions are not likely to show this degree of error. Service life estimates are likely to be wrong by no more than 10%. Nevertheless, even this margin of error means that capital stock estimates are likely to have an error margin in the range +/- 8%. However, as noted in Chapter 4 below, varying the average service life assumptions by +/- 20% does not impact the final results greatly, although this is partly because of the relatively short period from the base-year (1995) to the final year of estimates (2007). The results achieved are still comparable to those achieved by other projects at the national level (EU KLEMS) and those produced by NSOs. 6

10 An understanding of how assets are retired from the capital stock in relation to this average service life Another source of variation between country-specific applications of the standard Perpetual Inventory Method results from a fifth required input. It is necessary to apply an assumed mortality function to the service life assumptions described above in order to remove assets from the capital stock upon completion of their service lives. The OECD Manual (para. 6.50) describes four mortality functions: simultaneous exit, linear, delayed linear and bell-shaped. The simultaneous exit function is probably the easiest to apply as it simply assumes that all assets of the same type and age will exit immediately upon completion of their service life. In addition, according to Mas et al (2006, page 71), substituting a simultaneous exit mortality function for a Winfrey S-3 function appears not to result in a significant change in the resulting gross capital stock in the way substituting with a linear function does. However, the OECD Manual (para. 6.64) suggests that the assumption of simultaneous exit is simply not plausible. It implies that the use of a Winfrey S-3 bell-shaped curve is likely to be most accurate. For the purpose of this final report three sets of capital stock estimates have been produced corresponding to three different mortality functions: simultaneous exist, Winfrey S-2 and Winfrey S-3. A comparison is made between the estimates achieved using the two alternative Winfrey mortality functions and those achieved using simultaneous exit. 2.4 Specification of standard PIM applied in this report There are a number of variations on the Perpetual Inventory Method which can be employed to produce capital stock estimates. In addition to the standard approach employed here there is also an alternative method which is more intricate. It was described in the Interim Report and is described in detail in the OECD Manual (2001). However, even within this standard approach there are a number of possible variations because there are an array of different depreciation and retirement approaches that can be applied. The estimates produced for this report were achieved by assuming linear depreciation and then integrating three different types of mortality function with this. The basic gross capital stock formula applied in this report is shown in the following equation: d 1 i 0 CS t I t 1 where CS t = gross capital stock in year t (valued at historical cost) I t = investment in year t d = life expectancy This values capital at its historic cost (ie its cost at purchase). To revalue this to the prices prevailing in year t the equation is modified slightly as follows: d 1 t, t I t i* Pt i t i 0 CS, 7

11 P t i, t = price index in year t-i with year t as the price-base year CS t, t = Capital stock of year t in prices of year t The net capital stock is defined as the gross capital stock minus the accumulated consumption of fixed capital: NCS d 1 t, t t i* t i, t / i 0 I P * 1 i d NCS t, t = Net capital stock in year t in prices of year t d = the assumed service life of the asset The assumed method of writing-off the consumed capital is crucial to the process of constructing estimates and is represented by the second half of the equation above as shown below. 1 i/ d d = the assumed service life of the asset Here the write-off is assumed to be linear in nature. Under this approach, once the asset has been depreciated to zero it has effectively been entirely retired from the capital stock. It can therefore be seen that using this approach the depreciation and retirement patterns are essentially one and the same or, more accurately, what is being assumed for retirement is simultaneous exit. All items of capital stock are retired immediately at the end of their assumed service life. Alternative, more realistic retirement assumptions are examined in the next section. From the above equations it is clear that for any one year the capital stock is essentially a three-way interaction between a revaluation to this year s prices of the capital stock from the previous year, the depreciation and retirement of capital due to its consumption, and the addition to the stock brought about through investment in this year. 2.5 Incorporating the Winfrey retirement functions The Winfrey S2 and S3 retirement functions When employing simultaneous exit as the method of retirement, an entire cohort is assumed to be retired at exactly the same time, simultaneously. Once a cohort s average service life has expired the probability that a randomly chosen asset will be retired is therefore equal to unity. By comparison, under the Winfrey set of retirement functions, assets of the same cohort have a differing probability of retirement around the average service life and, therefore, do not all retire at the same time. The retirement pattern is bell-shaped around the average service life. The bell-shaped curve has various levels of skewness depending upon which of the Winfrey set of retirement functions is employed. There are 18 Winfrey curves/functions in total. The S2 and S3 curves are the most commonly employed and have been adopted to produce alternative estimates for this report. Altogether, there are six S or symmetrical curves, six L or left skewed curves and six R or right skewed curves. The number which occurs in each curve s name (eg S2) relates to the relative flatness of the curve, S1 being the flattest symmetrical curve and S6 being the tallest (ie the most narrowly distributed around the average service life). 8

12 Winfrey curves are described by the same general equation, shown below. 2 2 F F (1 T / a ) T 0 m where F T is the proportion that retires in time period T F 0 is the proportion retiring at the average retirement age the parameter a determines how the time periods correspond fractionally to the average service life (eg a = 10 means the time periods are deciles around the average service life) and the parameter m determines the relative flatness of the function The 18 Winfrey curves are distinguished by the values applied to the parameters F 0 and m, representing varying kurtosis. F 0 The Winfrey S2 curve has = , and m = F 0 The Winfrey S3 curve has = , and m = Given these parameters, if the assumed average service life of an asset (the parameter a in the above equation) were 5 then the resulting probability of retirement would be as in Table 2.1. Combining the S2 and S3 retirement functions with straight-line depreciation When employing the Winfrey S2 or S3 functions, only a fraction of a cohort of an asset with an average service life of 5 years will actually be in service for exactly 5 years. The depreciation function must be combined with the retirement function to reflect this. For example, if there is an asset type with a 5-year average service life then, as shown in Table 2.1, according to the Winfrey S2 function 0.5% of them will be retired after 1 year, 4.6% after two years, 12.5% after three years, 20.5% after four years and so on. So after, say, 2 years of the cohort s existence, 4.6% of the stock will have the fraction (2-2)/2 = 0 of its initial value remaining (ie it is retired), 12.5% will have (3-2)/3 = 1/3 of its initial value remaining, 20.5% will have (4-2)/4 = ½ of its value remaining, and so on. In this way depreciation and retirement are combined. This combining of depreciation and retirement functions is expressed as follows: n T max h g ( T ) F T n n T where Tmax is the maximum possible length of service life of a particular asset type. TABLE 2.1: COMPARISON OF PROBABILITY OF RETIREMENT UNDER ALTERNATIVE APPROACHES Years (T) Winfrey S Winfrey S Simultaneous exit Note(s) : Average service life is 5 years in this example. 9

13 where g n (T ) is the depreciation schedule of an asset with service life T where is the probability of retirement at age T where F T h n is the combined depreciation and retirement function This shows that the value of a capital stock in time period n is calculated as the sum of the values of the remaining assets of different service lives, weighted by their probabilities according to the retirement function. Table 2.2 shows how h n is calculated for an asset with an average service life of 5 years for the Winfrey S2 retirement function and linear depreciation. The first two columns represent the retirement function, with the numbers 1-10 in the first column (T) being the asset service life and the numbers in the second being the probability of an asset retiring at this age. The value of the capital stock h n in a particular year n is calculated as the sum of the values in the column n. The values in each column of Table 2.2 are calculated by multiplying the probability of retirement by the depreciation function. An example of the calculations are shown in column 3 (n=3) of the table. Column 3, row 6 corresponds to the value of a 6-year asset after 3 years weighted by the proportion of 6-year assets in the total stock. By summing up all these cells in column 3 we get the total value of that cohort of capital stock in year 3. Table 2.3 gives the same demonstration but for the Winfrey S3 function. Note that the numbers in the second column and hence subsequent columns have changed. TABLE 2.2: COMBINING WINFREY S2 AND LINEAR DEPRECIATION T Probability of retirement /3*12.5= ¼*20.5= /5*23.8= /6*20.5= /7*12.5= /8*4.6= /9*0.5= /10*10= Value (h n ) Note(s) : Average service life is 5 years in this example. 10

14 TABLE 2.3: COMBINING WINFREY S3 AND LINEAR DEPRECIATION T Probability of retirement Value (h n ) Note(s) : Average service life is 5 years in this example. 2.6 Member State coverage We present capital stock estimates for 24 EU countries. The three missing countries are Greece, Bulgaria and Malta. It was not possible to produce estimates for these three countries because capital stock estimates capital stock estimates are not available to serve as a base-year estimate for Conclusion This chapter has described the PIM and its inputs. formalisation of a simple form of the PIM. It has provided a mathematical Employing the approach outlined in this chapter, the estimates produced as part of this project can be characterised as a snapshot of accumulated past investment in assets, from which is deducted a part of the initial value of assets part-way through their service life, and the entire remaining value of assets which have reached the end of their service life. There are several ways that these deductions from accumulated investment can be carried out, as has been explained in this chapter. The estimates produced as part of this project therefore view capital as a stock at a single point in time (the middle of each year for which estimates have been produced). Capital can also be viewed as a flow of services to production stemming from accumulated past investment in assets. Under this approach a Volume index of capital services to production is estimated. This alternative approach is not taken in this study as the focus here is on capital as a stock rather than a flow. The linear depreciation assumption is central to the approach that has been outlined and is the method by which part of the initial value of assets part-way through their service 11

15 life is deducted from accumulated investment, depending on how much of the asset s service life is completed. Linear depreciation has been shown to be a reasonable initial approximation of reality by Ward (1976) and Maddison (1992). It has been employed by Verbiest (1996) to produce national capital stock estimates for the Netherlands national statistical office and, more recently, by TNO to produce NUTS2 level estimates for the Netherlands. Three alternative methods have been outlined in this chapter for the deduction from the stock of the entire remaining value of assets which have completed their service life. The simplest method, simultaneous exit, assumes all assets of the same cohort are retired simultaneously and their remaining value removed from the stock at the same time. What has been shown is that the capital stock for a particular year is a three-way interaction between the revaluation of the capital stock from the previous year, the depreciation and retirement of capital due to its consumption, and further addition to the stock brought about through Gross Fixed Capital Formation. 12

16 3 GROSS FIXED CAPITAL FORMATION 3.1 Introduction In order to produce capital stock estimates, the PIM requires two main data inputs. The first is an initial set of estimates for capital stock in the base year. The process by which these were produced in this study is described in the first part of this chapter with additional detail in Appendix A. The second input is a set of estimates for gross fixed capital formation (GFCF) and these are the focus of the rest of this chapter. The objective for this stage of the project was to produce GFCF estimates in constant and current prices for the years from 1995 to 2007, disaggregated by industrial sector, asset type and institutional sector. It was decided that the method should employ data from Eurostat primarily and to make use of data from other sources, particularly national statistical offices, only where the data from Eurostat are insufficient. The estimates produced are complete across most dimensions and are internally consistent. This chapter describes the availability of raw data and outlines the methodology that was implemented. Where GFCF data on Eurostat are missing, incomplete or insufficient for a particular country or one of its NUTS2 regions, this chapter describes the procedures employed to produce a complete set of GFCF data to use as an input to the PIM. 3.2 Establishing a base-year estimate Unless a time-series of GFCF is available which goes back further than the longest-lived asset, it is necessary to construct a base-year capital stock estimate on which to begin the FIGURE 3.1: PROCESSING OF NATIONAL GFCF DATA Eurostat GFCF by asset Eurostat GFCF totals AMECO GFCF Eurostat GFCF by industry GVA by industry Investment ratios methodology (see Figure 2) Filled GFCF by asset Apply AMECO growth rates Filled GFCF totals Divide constant by current price GFCF to create deflators Investment-output ratios methodology (see Figure 2) Filled GFCF by industry (current prices only) Filled GFCF by industry (constant and current prices) Data Data filling Data creation RAS methodology (see Appendix 2) GFCF by asset by industry 13

17 PIM described in the previous chapter. Since the longest-lived assets (usually dwellings) have assumed service lives approaching 70 years or more, it is highly unlikely that a full GFCF time-series will be available. Indeed, Eurostat only has GFCF time-series at NUTS2 level going back to For this reason the estimates to be produced as part of this project begin in this year and it is therefore necessary to produce a base-year estimate for 1995 on which to begin the PIM s process of adding GFCF and subtracting consumed capital and retirements. In keeping with the objectives of the rest of the project, the objective was to create base-year estimates that were as comparable as possible. For eleven countries capital stock estimates have recently been produced as part of the EU KLEMS project. Since the method used to produce them was the same for all countries it was considered useful to employ this EU KLEMS data to create base-year estimates in these eleven cases. For the remaining countries the base-year has been constructed in other ways, usually based on capital stock estimates produced by the national statistical office. Figure 3.4, below, summarises the methods for creating base-year estimates for The specific details for each country can be found in Appendix A. Figure 3.4 shows that the option of last resort was to use the output-capital ratio of similar countries to produce the base-year estimate where no other data is available. It was only necessary to resort to this method for three countries: Bulgaria, Greece and Malta. The similar countries chosen and the methods used for these three countries are described in detail in Appendix A too. 3.3 GFCF raw data availability Eurostat defines GFCF as follows: Million EUR (SA). Gross fixed capital formation (GFCF, ESA 1995, 3.102) consists of resident producers acquisitions, less disposals, of fixed assets during a given period plus certain additions to the value of non-produced assets realised by the productive activity of producer or institutional units. GFCF includes acquisition less disposals of, eg buildings, structures, machinery and equipment, mineral exploration, computer software, literary or artistic originals and major improvements to land such as the clearance of forests. Values are seasonally adjusted (SA). National level data At the national level, Eurostat produces GFCF estimates for the EU27 broken down by asset type and industrial sector. However, GFCF by institutional sector are unavailable and for this reason estimates of capital stock cannot be produced with disaggregation by institution. The GFCF data by industry are divided into categories according to the industry in which the assets are being used as means of production. The three industries for which the data are available at national level are summarised in Table 3.1. The GFCF data from Eurostat are also available according to the type of asset that has been invested in. There are five asset types available as shown in Table 3.2. The composition of the other assets category in Table 3.2 varies with the data source. For those countries where the base-year is calculated from EU KLEMS data (see Appendix A) other assets only includes software. These countries are Austria, the Czech Republic, Germany, Denmark, Finland, Italy, the Netherlands, Portugal, Sweden, Slovenia and the United Kingdom. For those countries where the base-year estimates are 14

18 TABLE 3.1: CLASSIFICATION OF INDUSTRIES Industry SIC code Description Agriculture A-B Agriculture, forestry and fishing (A) and mining and quarrying (B) Manufacturing C-F Manufacturing C), electricity and gas supply (D), water supply (D) and construction (F) Services G-P Wholesale and retail trade (G), transport (H), accommodation and food service activities (I) information and communication (J), financial and insurance activities (K), real estate activities, (L) professional, scientific and technical activities (M), administrative and support service activities (N), public administration and defence (O) and compulsory social security and education (P) based on net non-financial fixed assets estimates from the national statistics office the figures for cultivated assets and intangible assets are aggregated into the band other assets. These countries are Estonia, Luxembourg, Latvia, Poland and Slovakia. An exception is Lithuania which has not produced any cultivated assets estimate for fixed assets. Thus the base-year estimates of other assets for Lithuania only includes intangible assets. For countries where the base-year estimates are based on net capital stock produced by national statistics offices or other sources that are disaggregated by asset type, the figure for software, products of agriculture, forestry, fisheries and aquaculture, and other products not included in other categories are aggregated into other assets. These countries are Belgium and Spain. For Ireland, where the base-year estimate is based on net capital stock estimates in Slattery (1975), the other assets category only includes the figure for agricultural machinery as Slattery (1975) did not produce estimates for the other two figures. For countries where the base-year estimates are only available by industry type, the breakdown by asset type is based on the average TABLE 3.2 CLASSIFICATION OF ASSET TYPES Asset type ESA 95 code Description Metal Products & Machinery AN Machinery and equipment not elsewhere classified. Examples include machinery for the production and use of mechanical power (except aircraft, vehicle and cycle engines), other general purpose machinery, agricultural and forestry machinery, machine-tools, office computers and electrical apparatus. Transport Equipment AN Equipment for moving people and objects. Examples include motor vehicles, trailers and semitrailers ships railway and tramway locomotives and rolling stock aircraft and spacecraft and motorcycles, bicycles, etc. Dwellings AN.1111 Buildings that are used entirely or primarily as residences, including any associated structures, such as garages, and all permanent fixtures customarily installed in residences. Other Construction AN.1112 Non-residential buildings (AN.11121) eg warehouse and industrial buildings, commercial buildings, buildings for public entertainment, hotels, restaurants, educational buildings, health buildings, etc., and Other structures (AN.11122) eg highways, streets, roads, railways and airfield runways bridges, elevated highways, tunnels and subways waterways, harbours, dams and other waterworks long-distance pipelines, communication and power lines local pipelines and cables, ancillary works constructions for mining and manufacture and constructions for sport and recreation. Other Assets AN.112, AN.1114 Intangible fixed assets (AN.112) consists of mineral exploration (AN.1121), computer software (AN.1122), entertainment, literary or artistic originals (AN.1123) and other intangible fixed assets (AN.1124), defined as new information, specialised knowledge, etc., not elsewhere classified, whose use in production is restricted to the units that have established ownership rights over them or to other units licensed by the latter. Cultivated asset (AN.1114) includes livestock for breeding, dairy, draught, etc. and vineyards, orchards and other plantations of trees yielding repeat products that are under the direct control, responsibility and management of institutional units. 15

19 asset shares of all the nations for which data are disaggregated by appropriate asset types and industry. These countries are Cyprus, France, Hungary, and Romania. Note that the industrial breakdown and asset type breakdown are published separately by Eurostat, meaning that GFCF for each industry are not available with an asset type breakdown or vice versa. Table 3.3 shows the availability from Eurostat of current price GFCF data broken down by sector and asset for the Member States for which capital-stock estimates have been produced in this interim report. Table 3.4 contains similar information for the constant (2000) price data. There are no data by industry in constant prices for any of the countries. As can be seen from both tables, for national totals there is data covering all the years available for most countries. The TABLE 3.3: RECORD OF RAW DATA AVAILABILITY: CURRENT PRICE INVESTMENT DATA Totals By industry By asset AT * BE * BG CY CZ * DE DK * EE ES FI FR GR HU IE IT LT LU * LV MT NL PL PT RO SE SI SK UK Percentage of all data points 98% 69% 89% Note(s) : * From National Statistics Office. 16

20 TABLE 3.4: RECORD OF RAW DATA AVAILABILITY: CONSTANT PRICE INVESTMENT DATA Totals By industry By asset AT BE BG CY CZ DE * DK EE ES FI FR GR HU IE IT LT LU LV MT NL PL PT RO SE SI SK UK Percentage of all data points 96% 0% 76% Note(s) : * From National Statistics Office. number of countries with some missing years is greater by asset but, still, for the majority data for all the years are available. Table 3.5 describes the methods used to fill missing years. In addition, AMECO produces GFCF data for the EU27 countries at the national level without breakdown by industry or asset type. AMECO is the annual macro-economic database of the European Commission s Directorate General for Economic and Financial Affairs (DG ECFIN). The main data source is Eurostat (the Statistical Office of the European Commission), complemented, where necessary, by other appropriate national and international sources. These data are available in both constant and current prices and are complete for the years for all the EU27 countries. AMECO also produces GDP deflators broken down by industry. Both the GFCF data and the GDP 17

21 TABLE 3.5: METHODS FOR DATA GAP FILLING Country Data Totals By industry AT nominal Use NSO data, and investment-output ratio real Use deflators BE nominal Investment-output ratio real Use deflators BG nominal Investment-output ratio real Use deflators CY nominal Investment-output ratio real Use deflators CZ nominal Use NSO data, and investment-output ratio real Use deflators DE nominal Investment-output ratio real Use deflators DK nominal Use NSO data, and investment-output ratio real Use deflators EE nominal Investment-output ratio real Use deflators ES nominal Investment-output ratio real Use deflators FI nominal Investment-output ratio real Use deflators FR nominal Investment-output ratio real Use deflators GR nominal AEMCO data growth rate Investment-output ratio real AEMCO data growth rate Use deflators HU nominal Investment-output ratio real Use deflators IE nominal Investment-output ratio real Use deflators IT nominal Investment-output ratio real Use deflators LT nominal Investment-output ratio real Use deflators LU nominal Use NSO data real Use deflators LV nominal Investment-output ratio real Use deflators MT nominal Investment-output ratio real AEMCO data growth rate Use deflators NL nominal Investment-output ratio real Use deflators PL nominal Investment-output ratio real Use deflators 18

22 TABLE 3.5 (CONTINUED): METHODS FOR DATA GAP FILLING Country Data Totals By industry PT nominal Investment-output ratio real Use deflators RO nominal AEMCO data growth rate Investment-output ratio real AEMCO data growth rate Use deflators SE nominal Investment-output ratio real Use deflators SI nominal Investment-output ratio real Use deflators SK nominal Investment-output ratio real Use deflators UK nominal Investment-output ratio real Use deflators deflators produced by AMECO are used in the methodology outlined in Section 3.3. We also use GVA data broken down by industry. The GVA data for are based on Eurostat figures and are filled using data from national statistical offices and other local sources. Where an incomplete series exists at NUTS2 level, interpolation methods have been used which fill gaps in the series from complete series available for aggregates of NUTS2 regions. The totals of regions containing interpolated values are constrained to sum to known totals at higher levels of the spatial hierarchy. In this way, a detailed series has been built up which is consistent with the higher-order regional values available in published statistics. The GVA data for the years 2006 and 2007 are based solely on NSOs and other local sources. These figures are then deflated by industry using the AMECO deflators. Regional level data Additional data from NSOs Summary position For the NUTS2 regions, Eurostat has total GFCF and data broken down by industry in current prices only with incomplete coverage. There are no data by asset type. For most countries, there are at least some data by industry for every NUTS2 region, although not for the whole time period required. For the entire EU27, 71% of the totals figures and 68% of the figures by industry are available at NUTS2 level for the period This means that for our required period of we have 53% of the totals data and 54% of the data by industry. The national statistical offices of several of the EU27 countries produce estimates of GFCF across various dimensions. These are not produced on a comparable basis, and so we only use them to fill gaps in the Eurostat data using their rates of growth. In summary, the Eurostat data gives GFCF in some detail in terms of breakdown by asset type and industry as well as NUTS2 region. However there are many missing values. The AMECO data are far more complete but are only available at the national level with no breakdown by asset or industry. Data from national statistical offices are potentially a useful source of data but the figures are not produced on a comparable basis. The 19

23 methodology section that follows lays out the approach for making optimal use of all the available data in producing a consistent set of estimates. 3.4 Methodology The following methodology describes how GFCF estimates by asset, by industry and totals were produced from the available raw data for the Member States for which estimates have been produced in this report. Totals As shown in Table 3.3 and Table 3.4, there are a few missing years for one or two countries for total GFCF. In these cases the gaps are filled by applying the growth rates from the AMECO data in both constant and current prices. For the NUTS2 regions, total GFCF from Eurostat exists only for current prices. A ratio of these investment figures to GVA was calculated in each area for the periods where data were available. The growth rates of these investment-output ratios were calculated over the period for which data existed. Missing years were then filled in by applying this growth rate to the investment-output ratios. If data were available for only a single year then the resultant investment-output ratio was used in all years. These filled investment-output ratios were then applied to the GVA data to produce a complete set of GFCF estimates at the NUTS2 level. These were then scaled such that the NUTS2 regions of a country summed to their national totals. To create the constant price data, GFCF was deflated in each NUTS2 region using deflators calculated for the country as a whole. These deflators were derived from the constant and current price national data from Eurostat. FIGURE 3.2: INVESTMENT RATIOS AND INVESTMENT-OUTPUT RATIOS METHODOLOGY Investment ratios or investment-output ratios Yes Complete? End No 0 Data points? Yes Data exist in current or constant prices? 1 2+ Apply deflators No Yes NSO data? Use this ratio for all years Use growth rate over existing data to fill missing data End No Apply shares from a similar country 20