| المستخلص: |
Accuracy is an essential ingredient of data quality. Accuracy Data can arise from different sources such as sampling error, non-response, non-coverage, measurement error, and processing error. Hence, research is needed on all these sources of inaccuracy and the results should be communicated to users in order to help them to take a decision whether to use such data or not. Checking data accuracy is usually implemented using naive procedures that can be done manually such as comparing the last two period-to-period growth, or the annual differences, checking if the new observation is an overall minimum or maximum, etc. However, these procedures have several disadvantages. First, they use part of the data not the whole series. Second, they can be strongly affected by seasonality, noise, or special events. Third, manual inspection limits strongly the number of series that can be checked, and hence the inspection has to be made at a relatively high level of aggregation. All these contaminate most of the information contained in the series and hide many facts about them. Therefore, an efficient automatic procedure that considers the full information in the time series and permits inspection on more disaggregate level is needed. TERROR is a reliable program to do such an automatic inspection. Unlike naive techniques to check data accuracy, TERROR methodology considers the full information in the time series. Program TERROR centers on data bases of time series and considers the following basic quality control problem: When the data for the past and present periods become available, which of the new observations are likely to contain an error? The TERROR judging criterion for the evaluation of a new observation is very far from what could have been expected looking at its past history. The program is working in an automatic manner with the following steps: 1. Identifies a suitable model to describe the series and estimates its parameters by exact maximum likelihood (or unconditional least squares). 2. Detects and corrects for several types of outliers and special effects. 3. Yields optimal interpolators of the missing observations with their associated MSE. 4. Computes optimal forecasts for the series, together with their MSE. When the forecast error is, in absolute value, larger than some a priori specified limit, the new observation is identified as a possible error. Summary results for all series and for the aggregate set are also provided. This paper is concerned with checking the accuracy of monthly economic data on the National Data Store (NDS) of the Cabinet's Information and Decision Support Center (IDSC). This is done by testing the existence of potential errors in the last observations of these series using program TERROR. When the program has been applied to the monthly economic data available on the NDS, we can conclude that the major reasons for outliers' existence and or unsatisfactory forecasts can be summarized in one or more of the following problems: 1. Changing the base year for some series. 2. High percentage of missing values in some series. 3. Changing the definition or the methodology of some indicators. 4. Some series end with estimated values rather than true ones.
|
