Short-term Distributive Trade Statistics Workshop for African countries on the Implementation of International Recommendations for Distributive Trade Statistics 27-30 May 2008, Addis Ababa, Ethiopia UNITED NATIONS STATISTICS DIVISION Trade Statistics Branch Distributive Trade Statistics Section Outline of the presentation Overview of short-term statistics (STS) Indices of Distributive Trade Seasonal Adjustment Benchmarking Overview of STS (1) STS are an important source of information for: Developing and monitoring effectiveness of economic policy Carrying out business cycle analysis Priorities of short-term DTS Production of monthly or quarterly indicators for distributive trade sector in the most timely manner Characteristics of short-term DTS Presented in the form of indices, growth rates and in absolute figures (levels) If compared to structural DTS, short-term statistics have: lower accuracy less details reduced scope Produced according to a strict timetable Subject to revisions Overview of STS (2) Analyses performed with short-term DTS fall into two categories Comparison of activities of distributive trade units between two different points in time Comparison within one reference period of two or more different sub-populations of units Compilation of short-term DTS requires from countries development and implementation of appropriate statistical techniques Structural and short-term statistics should be reconciled so as to combine the relative strengths of each type of data Requirements for compilation of short-term DTS (1) To be based on the identical with structural statistics concepts, measurement principles, statistical units, classifications and definitions of data items To be built on a foundation of timely and accurate infra-annual data sources that cover an adequate proportion of units (size of the samples) To be made consistent with their annual equivalents For convenience of users For proper implementation of benchmarking techniques Requirements for compilation of short-term DTS (2) Econometric methods and indirect estimation procedures should not substitute the collection of short-term DTS by countries Flash estimates – require use of econometric methods If the econometric methods have been used, countries are advised to: Make available to users both the methods used and the reliability of the estimates Revise the estimates as soon as new and more accurate information becomes available Indices of distributive trade (1) Purpose To describe the short-term changes in value and volume of: Wholesale and retail trade turnover Output of distributive trade sector as a whole and of its activities To complement indices of other economic activities for short-term analysis of the economy To provide a key input in the compilation of quarterly national accounts Indices of distributive trade (2) Types of distributive trade indices Indices of turnover changes in nominal terms (value index) Indices of turnover volume and output of distributive trade sector (volume index) Periodicity Monthly indices produced without significant time lag are recommended, however Quarterly indices are also acceptable if a NSO does not have sufficient capacity to produce them Indices of distributive trade (3) Recommendations for compilation of distributive trade volume indices Preferred approach Chained-linked Laspeyres index with weights being updated at least every five years Annual chain-linking takes better account of changes in relative prices and thus recommended for indices of distributive trade services whose structure of weights evolve rapidly Alternative for countries using Laspeyres volume index with fixed weights The periods between which weights are updated should be as close to five years as possible While updating the weights countries are encouraged to make every effort and to chain-link the series with the new weights Indices of distributive trade (4) Indices of wholesale and retail trade turnover Value index Compares the value of turnover in the current period (at current prices) with the value of turnover in the base period (at base year prices) Volume index Compares the value of turnover in the current period (at base year prices) with the value of turnover in the base period (at base year prices) Deflation Price effect in current period values of turnover should be eliminated - CPI, PPI, WPI Output volume indicators (quantity of goods sold) Input indicators (labour) Indices of distributive trade (5) Turnover volume index vs. index of output of wholesale and retail trade Both indices important in their own right Turnover index recommended within the framework of shortterm statistics Output index meaningful within the framework of national accounts Conceptual differences Goods bought for resale in the same condition as received Goods produced (or purchased) and stocked before sale Quality of trade service supplied Seasonal Adjustment for DTS (1) Need for SA Infra-annual data on DTS represent a key tool for policy making, modeling and forecasting DTS data are often contaminated by seasonal fluctuations and other calendar/trading day effects that can mask relevant features of the time series SA goal is to remove these influences to achieve a better knowledge of the underlying behavior of the time series Seasonal Adjustment for DTS (2) Advantages of SA Provide more smooth and understandable series for analysis Supply the necessary inputs for business cycle analysis, trend-cycle decomposition and turning points detection Facilitate the comparison of long-term and short-term movements among sectors and countries Seasonal Adjustment for DTS (3) Drawbacks of SA Quality of SA strongly depends on quality of raw data SA depends on ‘a priori’ hypotheses on the model and the data generation process Information contained on the hypothesized seasonal components and correlation with other components are lost after SA SA data are often inappropriate for econometric modeling purposes Seasonal Adjustment for DTS (4) Main Principles of SA SA is performed at the end of the survey process on the series of original estimates Fundamental requirement No residual seasonality Lack of bias in the level of the series Stability of the estimates Seasonal Adjustment for DTS (5) Time series Data collected at regular intervals of time (example: turnover of retail trade for each sub-period of the year) Data collected irregularly or only once is not a time series Types of time series Stock (activity at a point in time) Flow ( activity over a time interval) SA is mainly performed on flow series Seasonal Adjustment for DTS (6) Components of time series Trend: associated with long-term movements lasting many years Cycle: associated with the fluctuations around the trend characterized by alternating periods of expansion and contraction (business cycle) In much analytical work, the trend and the cycle are combined to form the trend-cycle Seasonal component: movements within the year associated with events that repeat more or less regularly each year (climatic and institutional events) Irregular component: associated with unforeseeable movements related to events of all kinds Seasonal Adjustment for DTS (7) Decomposition models Additive model Assumes that the components of the time series behave independently The size of the seasonal oscillations is independent of the level of the series Multiplicative model (default model) Assumes that the components are interdependent The size of the seasonal variations increases and decreases with the level of the series Seasonal Adjustment for DTS (8) Main effects of the seasonal component Seasonal effects narrowly defined Calendar effects Stable in terms of magnitude and timing (e.g. Christmas) Variations associated with the composition of the calendar (not stable in time) Moving holidays Trading days Length-of-month and Leap year effects Original series should be adjusted for all ‘seasonal variations’ and not only for the seasonal effects narrowly defined Seasonal Adjustment for DTS (9) Moving holidays Holidays that occur at the same time each “year” based on calendars other than the Gregorian calendar Their exact timing shifts systematically each Gregorian calendar year Examples: Easter, Chinese New Year, Ramadan, Korean Thanksgiving, etc. Types of effects Immediate effects: some retail stores are closed during the holidays Gradual effects: the level of trade activity is affected during several days before the holidays and Leap year effects Seasonal Adjustment for DTS (10) Trading days effect TD effect is present when the level of activity varies with the days of the week TD effect is due to the number of times each day of the week occurs in a given period (month/quarter) Number of TD may differ: From period to period Between same periods in different years TD effect is found in many economic time series, especially in Distributive Trade Other calendar effects Length of Month effect: different months of the year have different lengths (28, 29, 30 and 31 days) Leap Year effect: February has 29 days every four years. Seasonal Adjustment for DTS (11) SA methods and software packages Moving average (filtering) methods Model based methods Mainly descriptive, non parametric and iterative estimation procedures Main packages: X-11-ARIMA, X-12 and X-12-ARIMA Components are modeled separately using advanced time series methods (e.g. Kalman Filter) Assume that the irregular component is “white noise” Main packages: TRAMO-SEATS, STAMP, BV4, etc. New tendency: combination of the main two approaches DEMETRA (Eurostat) X-13-SEATS (U.S. Census Bureau) Seasonal Adjustment for DTS (12) Main recommendations Production of seasonally adjusted data should be considered an integral part of countries program of quality enhancement of DTS SA should be performed at the end of the survey process when final estimates are produced All three types of data should be made available to users Raw series (original) Seasonally adjusted series Trend-Cycle Seasonal Adjustment for DTS (13) Main recommendations (cont.) Revisions of SA data should be scheduled in a regular way according to the release calendar Re-identification of the ARIMA models should be undertaken once per year while re-estimation of the parameters every time SA is performed Country-specific calendars to be used in order to ensure more accurate results in trading day adjustment Direct adjustment is preferred when the components of the aggregate series have similar seasonal patterns Indirect adjustment is recommended in the opposite case Benchmarking (1) What is benchmarking? Process by which the relative strengths of low and high frequency data are combined The short-term movement is preserved under the restriction of annual data Process ensuring an optimal use of annual and sub-annual data in a time series context Example: It is important to have consistency between annual and infra-annual estimates of levels of any variable. However, the turnover of distributive trade sector derived from monthly (quarterly) surveys differs from that derived from annual sources Benchmarking (2) Main aspects of benchmarking Interpolation – no genuine monthly (or quarterly) measurements exist, and annual totals are distributed across months (quarters) Extrapolation - time series are extended with the estimates for months/quarters for which no annual data are yet available Benchmarking (3) Basic concepts: Benchmark-to-indicator (BI) ratio framework Basis for the reconciliation of statistical data derived from different data sources Defines the relationship between the corresponding annual and monthly/quarterly data Benchmark series: the original low frequency (annual) data Indicator series: the original high frequency data (monthly/quarterly data) In practice BI ration differs from 1, so adjustments are necessary to be made to bring it to 1 Benchmarking (4) Benchmarking methods Numerical approach - the model that a time series is supposed to follow is not specified Pro-rata distribution method Family of least squares minimization methods (Denton family) Statistical modeling approach ARIMA model-based methods Various regression models Benchmarking (5) Pro-rata distribution method Distributes the annual level data according to the distribution of monthly/quarterly indicator BI ratios for adjacent years are different Introduces a “step problem” - discontinuity in the growth rate from the last month/quarter of one year to the first month/quarter of the next year Denton family of benchmarking methods Based on the principle of movement preservation Month-to-month (or quarter-to-quarter) growth in the original and adjusted time series should be as similar as possible The adjustment for neighboring months (or quarters) should be as similar as possible Incorporation of new annual data for one year requires revision of previously published monthly/quarterly estimates Proportional Denton method – the most preferred method in this family Benchmarking (6) Recommendations Countries are encouraged to: Consider benchmarking an integral part of the compilation process of short-term DTS Perform it at the sufficiently detailed compilation level Benchmarking and revisions At least two to three preceding years have to be revised each time new annual data become available in order to maximally preserve the short-term movements of the infra-annual series Benchmarking (7) Recommendations (cont.) Benchmarking and quality Benchmarking techniques play a key role in improving the quality of distributive trade statistics In the short-to-medium term, benchmarking techniques often succeed in filing the gaps of missing data and solving shortcomings In the longer term, benchmarking techniques play an important role in optimizing the use of available data Benchmarking and seasonal adjustment Benchmarking should be performed at the end of the survey cycle when data has been collected, processed and edited; and estimates are produced In most cases, benchmarking is performed before seasonal adjustment Thank You