Furthering the use of administrative data in sub-annual financial statistics Paper presented at the International Association for Official Statistics Conference, at Da Nang, Viet Nam, 8-10 October 2014 Mathew Page Senior Analyst, Platform Operations, Statistics New Zealand Private Bag 4741 Christchurch, New Zealand +64 3 964 8700 mathew.page@stats.govt.nz www.stats.govt.nz Nicholas Cox Statistical Analyst, Platform Operations, Statistics New Zealand Private Bag 4741 Christchurch, New Zealand +64 3 964 8700 nicholas.cox@stats.govt.nz www.stats.govt.nz Crown copyright © This work is licensed under the Creative Commons Attribution 3.0 New Zealand licence. You are free to copy, distribute, and adapt the work, as long as you attribute the work to Statistics NZ and abide by the other licence terms. Please note you may not use any departmental or governmental emblem, logo, or coat of arms in any way that infringes any provision of the Flags, Emblems, and Names Protection Act 1981. Use the wording 'Statistics New Zealand' in your attribution, not the Statistics NZ logo. Liability The opinions, findings, recommendations, and conclusions expressed in this paper are those of the authors. They do not represent those of Statistics New Zealand, which takes no responsibility for any omissions or errors in the information in this paper. Citation M.J. Page and N.R. Cox, Furthering the use of administrative data in sub-annual financial statistics. International Association for Official Statistics Conference, Da Nang, 2014. Furthering the use of administrative data in sub-annual financial statistics, by Mathew Page and Nicholas Cox Abstract Statistics New Zealand (Statistics NZ) intends to use administrative data as the primary source of data to produce business statistical outputs. Statistics NZ has undertaken an investigation into developing sub-annual financial business collections based on Goods and Services Tax (GST) administrative data. GST is a comprehensive ‘value added’ tax collection undertaken by New Zealand’s Inland Revenue department. To maximise the use of GST sales in the sub-annual collections we have developed an assessment model that enables decisions to be made and ensures any GST data used meets our quality needs. The underlying structure of the business units and their GST data characteristics are diverse, so different methodological options are suitable for different parts of the business population. The application of the GST sales assessment model has presented us with a number of statistical challenges that need to be resolved, including: (1) identifying conceptual differences between GST sales and our current survey variable, (2) developing a postal collection strategy where the use of GST data is not suitable, and (3) estimating inventories - a variable that does not exist in the GST data. Keywords: fitness for use, administrative data, quality, GST (VAT), managed collection, inventories. 1 Introduction 1.1 Statistics New Zealand’s use of administrative data in economic statistics Over the past decade we have seen advances in Statistics NZ’s use of administrative data, particularly in the economics area. The Statistical Architecture [6] and wider Statistics 2020 [9] programmes have been created to provide the strategic foundation upon which Statistics NZ will transform the collection and production of statistics. As part of this transformation, Statistics NZ’s vision is to have administrative data as the primary source of data to produce its statistical outputs. Data will be directly collected by Statistics NZ (eg by survey) only when necessary [10]. This is a significant paradigm shift in the collection and production of business statistics. This 3 Furthering the use of administrative data in sub-annual financial statistics, by Mathew Page and Nicholas Cox need is being driven by increased customer demands for data, to create efficiencies in the production of official statistics, and the need to decrease Government compliance costs faced by the business population. 1.2 Historic use of GST in sub-annual financial statistics There are two main variables required as part of the Statistics NZ sub-annual business collections – sales and inventories. These have historically been collected in three separate surveys – the Economic Survey of Manufacturing (ESM), the Wholesale Trade Survey (WTS), and Retail Trade Survey (RTS). While the sales and inventories information is published by Statistics NZ directly from the sub-annual business collections, this data also flows through to the National Accounts and is used in the production of quarterly Gross Domestic Product (GDP). Prior to 2001 Statistics NZ’s sub-annual financial collections used a postal survey design only (stratified random sample). The design included a full coverage strata for large businesses, where all businesses above a size threshold were surveyed. This is illustrated in Figure 1. In 2001, a Goods and Services Tax (GST) administrative data component was added to the design for the first time as a supplement to the survey data. This resulted in significant reductions in the postal sample. GST is a comprehensive ‘value added’ tax collection undertaken by New Zealand’s Inland Revenue Department. As GST data is essentially accounting information similar in nature to that supplied by respondents in Statistics NZ’s surveys, it was assumed to be suitable for use within our collections. This design is still currently used in statistical production. The GST component has been constrained to contribute no more than 15 percent of sales value for manufacturing and wholesale trade industries, and 10 percent for retail trade industries to minimise the impact of any problems when increasing our use of GST data. Our proposed new design uses GST data wherever possible, supplemented by a postal collection of large and complex businesses where the use of GST data is not suitable. These businesses will be included in a future Statistics NZ quarterly ‘managed collection’ where the collection of sales, inventories and other key economic variables is required. The managed collection strategy is fully described in section 3.2. 4 Furthering the use of administrative data in sub-annual financial statistics, by Mathew Page and Nicholas Cox An important benefit of the proposed design is that the use of GST allows ‘census-like’ coverage, as opposed to the sample survey approach used in the previous designs. There is also a significant reduction in respondent burden where the number of businesses receiving postal survey forms will decrease. While our current focus is to realise these efficiencies in the existing manufacturing and wholesale trade sub-annual outputs, in the future we intend to apply this design to measure sub-annual financial statistics for industries not currently covered by Statistics NZ, or other sub-population estimates. Figure 1 – History of sub-annual financial statistical designs 1.3 Structure of paper The paper commences by describing the approach we are proposing for maximising GST use to produce subannual sales statistics. We then move into discussing the challenges we have in moving to this approach, including: (1) identifying conceptual differences between GST sales and our current survey variable, (2) developing a managed collection where the use of GST data is not suitable, and (3) estimating inventories - a variable that does not exist in the GST data. 2 Maximising administrative data use To maximise the use of GST to produce sub-annual sales statistics we have established a GST sales assessment model that enable decisions to be made. The GST sales assessment model is presented in 5 Furthering the use of administrative data in sub-annual financial statistics, by Mathew Page and Nicholas Cox Figure 2. This section summarises the process of assessment. Further information on the GST sales assessment model can be found in Page et al [8]. The assessment model looks at three main facets. We assess: why the GST sales data should be used in a particular way (ie the ‘desired outcomes’ of GST use), what GST sales data should be used (ie assess the data’s fitness for use), and how the data should be used (ie the appropriate method). Figure 2 – GST sales assessment model 2.1 The desired outcomes from GST use In reviewing the approaches for the use of GST sales data, it became apparent there are a number of desired outcomes that need to be considered. It is recognised that not all desired outcomes are of equal importance and potential trade-offs between outcomes would need to be considered as part of any use of GST sales data. Maximising the use of GST sales data whilst also maximising the outcomes from its use should be our goal. Key outcomes identified include: Minimise complexity/cost – a higher level of complexity will likely result in a higher production cost. 6 Furthering the use of administrative data in sub-annual financial statistics, by Mathew Page and Nicholas Cox Minimise respondent burden – a key aim is to minimise respondent burden as much as possible. Maximise measurable quality – the options for using GST data must maintain the need for suitable quality measures. Maximise flexibility – a strength of GST data is its level of near ‘census’ coverage. This has the potential to deliver sub-population estimates at lower levels of detail for the likes of ad-hoc research. Maximise scalability – scalability refers to the ability of an option to be used in the development of new statistics. For example, the development of statistics for industries in the economy which are not covered by our existing collections (eg services industries). Maximise unit record availability – there is an increasing need for micro-data which will underpin future statistical analysis. 2.2 Data assessment Our data assessment has two steps. The first step is to assess whether the GST sales data is able to meet our quality criteria. Tax system sourced administrative data like GST is based on business accounting data that is expected to be fit for use within our collections. However, GST data has not been collected by Statistics NZ and an assessment is required to determine if the unit and variable definitions meet our requirements for the data to be fit for use. The second step determines how data will be used to produce statistical outputs. To maximise the use of the GST sales data we need to recognise the different characteristics within the business population. Different ‘parts’ of the data may be fit for use in different ways. As part of the data assessment we look at four different aspects of the GST sales data: GST reporting structures – administrative data is usually collected for the legal unit and this is the case with GST. Integrating the Statistics NZ business register with the legal units used for GST enables us to identify legal units that are most likely to be statistically fit for use. Such legal units are those which correspond to statistical units with activity in a single industry. However, as the unit structures become more complex, the use of GST data becomes more problematic. For these units 7 Furthering the use of administrative data in sub-annual financial statistics, by Mathew Page and Nicholas Cox there is a decrease in quality and potentially an increased need for an alternative to GST data to be used (eg a survey). An example would be where Inland Revenue allow a group of legal units linked by ownership to provide data in a total which is collected against one unit while other units in the group record zero values (referred to as GST groups). Conceptual alignment – as GST data is not collected for the purpose of producing financial or economic statistics we need to assess how the GST sales data aligns to the conceptual and definitional requirements of a sub-annual output. The sub-annual collection of ‘sales’ is defined to be fit for use within the System of National Accounts. The definition of sales for this use is expressed in our existing surveys (the ESM, WTS, RTS, and Annual Enterprises Survey (AES)). GST is designed as part of a value added tax regime with sales defined to measure the income side of value added. Value added is the concept that the National Accounts is trying to measure from our collections, so it is expected that GST sales data should meet our user need. However, validation by comparison to values from other sources such as AES (our most comprehensive source of annual accounting information) may reveal that the definitional differences are significant for some parts of the population. The GST sales data for these parts of the population may be deemed ineligible for use. Reporting frequency – GST data is reported to Inland Revenue on several frequencies depending on turnover size. Higher turnover businesses report monthly, medium turnover businesses two-monthly, and low turnover businesses every six months. For the sub-annual statistics considered in this paper, a monthly reporting frequency is ideal. Sales data for the three months of the quarter can simply be added. However, for the GST reporting frequencies greater than a month obtaining quarterly values requires some manipulation. The issue is addressed through a transformation of the data available for the period (eg first two months of the quarter) using a model that includes the data for the monthly reporters. Timeliness – there is an underlying assumption that to use GST data within our statistical outputs that we must have timely data available. For use within a sub-annual output the GST data must be for the current period being measured to identify turning points and changes within the economy. However, in order to maximise the use of GST data it may not be possible to wait until all the data 8 Furthering the use of administrative data in sub-annual financial statistics, by Mathew Page and Nicholas Cox needed for the production of an output is available. If most GST data is available, data for the missing units can potentially be obtained through the standard Statistics NZ imputation procedures. 2.3 Method options Our data assessment process enables us to determine what method options can be used for different parts of the business population. Can the data be used directly as received? Can it be used in conjunction with some other data source? Or can it not be used at all? Direct approaches – to use the data directly requires that the data meets some basic statistical standards such as an appropriate statistical unit and the relevance of the GST sales measure to our target concept. In our ‘direct unmodified’ approach the GST sales data is used as received. This method is applicable to units with a monthly reporting frequency. The ‘direct transformed’ approach uses one part of the GST data to ‘complete’ another part of the data. The data is used ‘directly’ in that it is not reliant on any other data source to produce the output. This method is applicable to two-monthly filers, since some data manipulation (using available monthly filers) is needed to adjust for the fact that the data cannot be added directly to produce a quarterly result. Combined sources approach – the ‘combined sources’ approach combines GST with data from other sources (eg another administrative data source) to produce statistical output. The approach can be used when there is some type of relationship between the data sources, despite their concepts not being closely aligned. For example, this method is applicable to small and medium sized GST groups, where employment data from the Employer Monthly Schedule (EMS)1 can be used to derive ratios to apportion the GST sales data to the individual group members. Other sources approach – the ‘other sources’ approach is the use of non-administrative data sources. This may include the use of the managed collection of large and complex enterprises where the use of 1 Employer Monthly Schedule data is collected on a monthly frequency by Inland Revenue for the purpose of administering New Zealand’s taxation system. It contains details of employee numbers, earnings, tax type, and tax deducted. 9 Furthering the use of administrative data in sub-annual financial statistics, by Mathew Page and Nicholas Cox GST is not suitable. It may also include the use of a sample survey more widely across an industry, if there are large conceptual differences between the GST sales data and the target concept. 2.4 Application of the assessment model The data assessment undertaken ensures that any GST sales data used will meet our quality needs. From this assessment the most appropriate method can be selected to ensure we make maximum use of the available GST data. The methods in turn are assessed against the desired outcomes to identify the potential benefits that arise from the use of the GST sales data. When reviewing the method options, there are some general points that need to be recognised. Options should not necessarily be considered mutually exclusive. We are not looking for a ‘one size fits all’ solution. The methods presented are 'building blocks' that could be combined in different ways to produce the desired outputs for each industry. This GST sales assessment model has been used to produce an experimental sales series of Christchurch city retail trade and hospitality activity, following the Canterbury earthquakes of 2010 and 2011 [4]. 3 Challenges in maximising GST data use The application of the GST sales assessment model for the manufacturing and wholesale trade industries has presented us with a number of statistical challenges that need to be resolved. These are currently being investigated by Statistics NZ and are described below. 3.1 Conceptual differences in the GST data To maximise the use of GST sales data in the sub-annual outputs we need to ensure the GST data closely aligns to the conceptual and definitional requirements of the existing manufacturing and wholesale trade subannual outputs. Statistics NZ have developed two statistical processes to assess this. 10 Furthering the use of administrative data in sub-annual financial statistics, by Mathew Page and Nicholas Cox 3.1.1 Aggregate industry assessment The GST sales assessment model describes the processes used to check the conceptual alignment of GST legal units. When examining the legal units across an aggregate industry, it becomes clear what industries are suitable for extensive GST data use, and industries where the use of GST data will be limited. The assessment process compares annualised GST sales data at a unit-record level with that from AES. The sales data from AES is defined and controlled to meet user needs through the questionnaire and survey design process. Our work to date has shown extensive use of GST data will be feasible across most manufacturing and wholesale trade industries. Figure 3 shows the aggregate industry assessment for two industries using 2012 data – ‘Printing’ and ‘Commission-based wholesaling’. The scatterplots present all units in the industry, excluding those with complex GST reporting structures. The annualised GST values are represented on the x-axis and the AES data on the y-axis. No outliers have been removed from the analysis. While sales values have been removed from both axis for confidentiality reasons, it should be noted the endpoint of the x-axis and y-axis is represented by the same maximum value. The ‘Printing’ industry shows a strong conceptual relationship between annualised GST and AES sales values. This means that extensive use of GST data for this industry is possible. The scatterplots are supplemented by linear regression diagnostics such as the R-square and slope. In this first example, both the R-square and slope statistics are close to 1. 11 Furthering the use of administrative data in sub-annual financial statistics, by Mathew Page and Nicholas Cox Figure 3 – Aggregate industry assessment graphs – 2012 year Printing Commission-based wholesaling In contrast, the ‘Commission-based wholesaling’ industry shows a large level of disparity between annualised GST and AES sales for a significant number of units. Businesses within this industry can record their sales on either a gross or net (commission) basis. Our target measure of sales is on a net basis. The GST sales assessment model indicates a large proportion of this industries value would be required from ‘other sources’, so we may need to continue a sample survey to collect data for this industry. 3.1.2 Identifying capital sales There are two known transactions that can be included in the GST data that are not part of the conceptual measure of sales required for National Accounting purposes – sales of capital items and of going concern 12 Furthering the use of administrative data in sub-annual financial statistics, by Mathew Page and Nicholas Cox businesses. Statistics NZ have developed editing methodologies to attempt to identify these transactions. This is to ensure the GST data is fit for purpose in the production of sub-annual financial outputs. Unlike the previous section which examines the applicability of GST sales data use across an industry, this editing process looks at each business in the GST population on an individual basis. The proposed methods reflect internationally recognised practice, being based on Statistics Canada’s Banff editing and imputation processes [1]. Banff applies the statistical process control method, which means for a given business in the GST data collection, all values of one variable are expected to fall within a certain range; the range being a function of the values previously observed. Any values that fall outside the range are then identified as being potentially anomalous. Our analysis (which again has mainly focused on the manufacturing and wholesale trade industries) has shown only a relatively small number of enterprises are identified as potentially anomalous on a quarterly basis. Therefore, in a future statistical production process we anticipate analyst input will be required before an adjustment is made. The factors that will be considered before any adjustment is made include: How large is the anomalous value and what material impact will the adjustment have on aggregate industry level outputs? What GST item is the anomalous value sourced from? For example, a sale of a going concern business is ‘zero-rated’ for GST purposes and recorded separately. Capital sales items are more difficult to identify as they are recorded with general sales transactions. What is the behaviour of other administrative variables? For example, is a large increase in GST sales matched by a similar rise in GST purchases, or employee numbers in Inland Revenue’s EMS? 3.2 Development of a managed collection A managed collection is required as the GST sales data for large and complex businesses is at times not suitable (as described in the GST sales assessment model). These businesses will be included in the Statistics NZ quarterly managed collection where we will continue to collect sales, inventories and other key economic variables that are required on a sub-annual basis. 13 Furthering the use of administrative data in sub-annual financial statistics, by Mathew Page and Nicholas Cox Very large businesses with simple reporting structures may also be included in the managed collection so we can retain control over their reporting. This is so we can continue to contact these businesses with queries about their data, collect any additional variables that are not available from the administrative data on a subannual basis, and minimise any risk associated with using their GST data directly. The risks from using GST directly for large businesses include the inclusion of items outside the true conceptual measures for National Accounting purposes, or any GST processing lags that do not meet our publication timeliness requirements. Statistical adjustments of the GST sales data to deal with both of these facets for large businesses may have a significant material impact on our outputs. The key statistical issue for defining a managed collection strategy is determining the boundary to maximise the use of administrative data while continuing to maintain the quality of our outputs. This is illustrated in Figure 4. Figure 4 – Managed collection strategy The following business rules were used to define a trial managed collection. These rules were generated following what-if analysis which tested a number of different options. A $100 million significance rule - if an enterprise, or group of enterprises linked by ownership, have an annual GST turnover of more than $100 million. A 3 percent industry dominance rule - if an enterprise makes more than a 3 percent contribution to annual total income for an industry. 14 Furthering the use of administrative data in sub-annual financial statistics, by Mathew Page and Nicholas Cox All enterprises that have a significant level of activity across more than one industry. Figure 5 shows two of our industry outputs using a ‘pure GST’ approach and compares it to a series using the trial managed collection business rules. These are both compared against our current published output The ‘pure GST’ series uses GST sales data for every business in the population (no survey data is used). This means the statistical processes described as part of the GST sales assessment model are largely ignored. There are still data transformations made to deal with businesses with complex GST reporting structures - here we use employment data to apportion the sales data to the individual businesses (using the combined sources approach). We also have included statistical processes to produce quarterly data for businesses with two and six-monthly GST filing frequencies. The ‘trial series’ applies the GST sales assessment model to utilise GST data wherever possible. However, this is supplemented by existing sub-annual survey data from the trial managed collection. As businesses in the trial managed collection are large and complex, the majority were already included in our existing sub-annual survey designs. So we had a valuable supply of historic survey data available to use as part of our analysis. For the example industries presented in Figure 5, the pattern of change over time is similar for the three series. However, the quarterly movements in the pure GST series for ‘machinery and equipment wholesaling’ are closer to the trial and published series than in the ‘fruit, oil, cereal and other food manufacturing’ industry. This wholesaling series is dominated by small and medium sized enterprises where extensive use of GST data is possible. However, the manufacturing industry has a number of large and complex businesses where we cannot be confident our transformations of the GST data produce robust estimates. This is where the use of a managed collection is important given the material impact these large businesses have on our statistical results. 15 Furthering the use of administrative data in sub-annual financial statistics, by Mathew Page and Nicholas Cox Figure 5 – Sales series using a trial managed collection Fruit, Oil, Cereal and Other Food Manufacturing 2,100 Sales ($m) 2,000 1,900 1,800 1,700 1,600 1,500 J 10 S 10 D 10 M 11 J 11 Statistics NZ published S 11 D 11 M 12 J 12 S 12 Trial series D 12 M 13 J 13 S 13 D 13 Pure GST Source: Statistics New Zealand These examples are typical of most industries across the manufacturing and wholesale trade industries. However, there are some examples where the series are moving in different directions and/or there is a temporary or consistent level difference between the trial series and published results. Our investigations have identified a number of common themes which help explain the differences in the series: 16 Furthering the use of administrative data in sub-annual financial statistics, by Mathew Page and Nicholas Cox The published series are largely based on postal samples which are designed to produce statistics to a certain level of accuracy. The trial series are based on the entire business population with no sampling used. This sometimes results in a consistent level difference between the two series if there is some degree of statistical sample bias in the published estimates. The trial series includes capital transactions and sale of going concern businesses which can be included in the source GST data as described in section 3.1. These transactions are not included in the published series. The managed collection in the trial series uses a ‘snapshot’ of enterprises as at the June 2011 quarter which is applied across the entire timeseries. While our analysis has found the group of large and complex businesses in the NZ economy remain relatively stable over time, there are sometimes changes to these businesses in the real world which can result in disparities between the two series. We are currently developing statistical processes for maintaining the managed collection over time. This is to ensure it continues to reflect economic reality on a quarterly basis, and takes account of any business restructures, mergers, split-offs etc, as well as significant growth or reduction in the size of large businesses. . 3.3 Measurement of inventories Quarterly inventory measures are required by National Accounts in the production of GDP. In the current subannual design inventories are primarily measured from data collected in the postal survey. In our proposed design, we will continue to collect inventories data for businesses in the managed collection. However, there will be no quarterly inventories data available for businesses outside the managed collection. This is illustrated in Figure 6. Statistics NZ have developed methods for measuring quarterly inventories in the proposed design for the administrative data component of the population (which will be considerably larger than in the current design). The work has found different methods will be required across the manufacturing and wholesale trade subindustries. Statistics NZ has developed an inventories assessment framework to determine what method 17 Furthering the use of administrative data in sub-annual financial statistics, by Mathew Page and Nicholas Cox options are most appropriate for each industry. The aim is to use the best method available for each industry. The framework takes into account the significance of inventories for each industry. Figure 6 – Measurement of inventories Using historic GST and sub-annual survey data, trial series using the new methods have been calculated across the timeseries and compared with Statistics NZ published results. The trial managed collection described in the previous section was also used in the production of these results. To calculate the trial series: Total inventories for industry x = inventories for the managed collection enterprises + inventories estimate for the administrative data component of the population. Figure 7 shows the comparisons between the trial series and the currently published series for finished goods inventories for the June 2010 quarter to the June 2013 quarter, illustrating three of our methods: The first graph illustrating ‘Petroleum and coal product manufacturing’ uses the ‘direct publish’ approach. Here we use the managed collection inventories data to calculate statistical estimates directly. The managed collection enterprises are at times responsible for most (if not all) of industry inventory estimates. In this method the administrative data component of the population effectively contributes ‘zero’ to the inventories calculation. This is the preferred method. The ‘benchmark to annuals’ approach is used where the managed collection enterprises captures quarterly change in inventories (from one quarter to the next quarter), but does not capture the inventories level. Inventory levels will be obtained by ‘rating up’ the aggregate inventories series using annual survey data. As the new annual values become available they are interpolated using the quarterly indicators. The ‘Chemical, polymer, and rubber product manufacturing’ industry has been 18 Furthering the use of administrative data in sub-annual financial statistics, by Mathew Page and Nicholas Cox selected to illustrate this approach. The graph shows the published series and the trial series prior to the ‘rating up’ process. The pattern of change over time is similar for the two series. The ‘model from annuals’ approach is used for some smaller industries where the inventories levels remain relatively consistent over time. A model is used to calculate the sub-annual inventory estimates –using quarterly GST sales or purchases in combination with inventories obtained from our annual collection. The model is updated on an annual basis when new data becomes available. The ‘Printing’ industry illustrates the trial series using this method. The trial series and the current published series remain flat throughout the timeseries. If none of the above three methods are fit for purpose, a traditional stratified random sample survey maybe required. This is a well-established approach and will not be discussed in this paper. Figure 7 – Trial series for inventories 19 Furthering the use of administrative data in sub-annual financial statistics, by Mathew Page and Nicholas Cox 4 Summary This paper presents an approach for maximising the use of GST sales data in the production of statistics within the sub-annual financial business collections. We describe how an assessment model can be used to determine what method options are suitable for maximising GST sales data use in different parts of the business population. The application of the assessment model in the manufacturing and wholesale trade industries has presented us with a number of statistical challenges which have mostly been resolved. 20 Furthering the use of administrative data in sub-annual financial statistics, by Mathew Page and Nicholas Cox Statistics NZ has developed statistical processes to check if the GST sales data aligns to the conceptual and definitional requirements of the sub-annual outputs. Our investigations have shown extensive use of GST sales data will be feasible across most industries. The use of GST sales data for large and complex businesses is at times not suitable. These businesses will be included in a Statistics NZ quarterly managed collection. Our analysis has provided us with an appropriate set of business rules to create and maintain a managed collection on a quarterly basis which compares well with published Statistics NZ results. Quarterly inventory measures are required by National Accounts in the production of GDP. This variable is not available in the GST data. Statistics NZ have developed a number of methods for measuring quarterly inventories for the administrative data component of the population. We illustrated three such methods which are shown to produce similar inventory estimates as the current published series. Acknowledgements The authors thank John Stewart, Chen Chen and Vera Costa for their contribution into the development of the ‘assessment model for maximising GST use’ described in this paper. Section 2 is largely based on previous work presented at the International Conference on Establishment Surveys in June 2012. References [1] Banff Version 1.04, Generalized System Methods Section & Business Survey Methods Division, Statistics Canada, Montreal, 2005. [2] F Brisebois, J Girard, and H Xie, Challenges surrounding the use of tax data in the development of quarterly indicators for various services industries, Statistical Society of Canada Annual Meeting, St John’s, 2007. [3] M Brodeur and D Ravindra, Statistics Canada’s new use of administrative data for survey replacement, Conference on Administrative Simplification in Official Statistics, Ghent, 2010. [4] Christchurch retail trade indicator (information release), Statistics New Zealand, Wellington, 2014. 21 Furthering the use of administrative data in sub-annual financial statistics, by Mathew Page and Nicholas Cox [5] R McKenzie, Managing the quality of administrative data in the production of economic statistics, International Statistical Institute, Durban, 2009. [6] R McKenzie, Statistical architecture, Statistics New Zealand, Christchurch (internal paper), 2008. [7] C Orchard, K Moore and A Langford, Practices for using VAT turnover data within the UK to produce estimates of growth and monthly turnover, Office for National Statistics, Newport, 2010. [8] M Page and N Cox, Maximising the use of administrative data in sub-annual business collections, New Zealand Association of Economists Conference, Auckland, 2014. [9] Statistics New Zealand Strategic Plan 2010-20, Statistics New Zealand, Wellington, 2010. [10] Statistics New Zealand Collection Strategy 2011, Statistics New Zealand, Wellington, 2011. [11] J Stewart, V Costa, M Page and C Chen, Towards an architecture for sub-annual business collections, Statistics New Zealand, Christchurch (internal paper), 2011. [12] J Stewart, V Costa, M Page and C Chen, Maximising the use of administrative data in sub-annual business collections, International Conference on Establishment Surveys, Montreal, 2012. [13] W Yung and P Lys, Use of administrative data in Statistics Canada’s business surveys – the way forward, International Association for Official Statistics Conference, Shanghai, 2008. 22