Some examples of the comparison Direct vs. Indirect adjustment for the Mexican economic indicators Blanca Rosa Sainz López National Institute of Statistics and Geography (INEGI) Introduction The users´ questioning on why seasonally adjusted series calculated by the direct method and the indirect method can show contrary signs about the magnitude of growth in certain periods, led to the analysis of the results of both methods to find the reason for these differences. a) Total Investment: Discrepancy between the aggregate (direct adjustment) and the sum of its components (indirect adjustment). Users´ questioning: in recent quarters it has been observed that the behavior of the directly seasonally adjusted total investment has differed its implicit evolution from the addition of the seasonally adjusted components. In the fourth quarter of 2010 and the first of 2011 the total investment seasonally adjusted directly showed a slowdown, while the total investment obtained by the sum of its components has shown an increase in its quarterly growth rate (see figure 1). Figure 1 Total Investment: Seasonally adjusted series a) Annual percentage variation b) Quarterly percentage variation 15 8 6 10 4 5 2 0 0 -2 -5 Direct -10 Indirect Direct -4 Indirect -6 -8 -15 I II III IV I II III IV I II III IV I II III IV I II III IV I II III IV I II III IV I 2004 2005 2006 2007 2008 2009 2010 -10 I II III IV I II III IV I II III IV I II III IV I II III IV I II III IV I II III IV I II III IV I 2003 2011 2004 2005 2006 2007 2008 2009 2010 2011 Response: when you use the indirect method to the aggregate you are adding the calendar and seasonal effects of the components, and it is not necessarily such way. It is convenient to analyze each case; there are cases where for one or more of the components the calendar effects are not significant and for other components they are significant. When you add the components you assume that the adjustment in the total is in the same way as in its components, but it can be that in the aggregate the calendar effect is more significant than it is in the components. That’s why we consider it is important to analyze the aggregate independently of the effects of its components and define its own model. It may also be happen at the time of adding several components that the effects are cancelled and are not significant in the aggregate. For example, there is an -1- outlier in one of the components; if we calculate the adjustment of the aggregate as a sum of the components adjustments, the outlier of the component will appear in the aggregate decomposition. Probably if we adjust directly the aggregate, maybe that outlier is not significant or less significant or can happen there are outliers in the aggregate that are not in the components. If we analyze the case of Mexican Total Investment which has two components, the Public and Private Investment, in the Public Investment the Easter effect is not significant and in the Private Investment it is significant with a parameter value of 3 days. For 2010 when Easter Sunday was on April 4th, for the Private Investment the adjustment for the Easter effect was equal to the one in 2009 and 2011. As a consequence the Easter adjustment for the aggregate obtained by the sum of its components in 2010 was equal, to the one 2009. However, if we calculate the adjustment directly for the aggregate, we have that the Easter effect is significant with a value of 8 days, greater than the value of its components, which could indicate that if the effect is not significant for a component that doesn’t mean there isn’t any effect in the component. With the value of the Easter effect for the aggregate the adjustment for 2010 and 2011 is in the following way: Total Investment (Annual percentage variation) Period 2010 2011 Original Serie I II III IV I (-)2.73 1.85 3.81 6.40 7.74 Original Serie adjusted for Easter effect, direct method (-)1.12 0.20 3.81 6.40 5.99 Original Serie adjusted for Easter effect, indirect method (-)2.73 1.85 3.81 6.40 7.69 As we can appreciate the adjustment for the Easter effect is very different with the direct method than with the indirect one, and even more in the first and second quarter of 2010 and the first quarter of 2011. So if we use the indirect adjustments this would led to different conclusions than if we use the direct results. Below are the figures 2 and 3 of the Original series adjusted by the Easter effect with the direct and indirect method and the corresponding factors. -2- Figure 2 Total Investment: Original series adjusted by Easter effect Annual percentage variation 15 10 5 0 Direct Indirect Differences -5 -10 I II III IV I 2004 II III IV I II 2005 III IV I 2006 II III IV I 2007 II III IV I 2008 II III IV I II 2009 III IV 2010 I -15 2011 Figure 3 Total Investment: Easter factors 103 102 101 100 99 Direct Indirect 98 I II III IV 2003 I II III IV 2004 I II III IV 2005 I II III IV 2006 I II III IV 2007 -3- I II III IV 2008 I II III IV 2009 I II III IV 2010 I 2011 97 Additionally, there aren’t significant outliers in the components in the study period, but in the total there is an outlier in LS2009.1 (change of level in the first quarter of 2009). It is the figure 4 of the seasonal factors with both methods, where you can see that the factor is substantially changed over time with the indirect method and in recent years seasonality becomes more marked, while the factors calculated with the direct method have remained approximately with the same pattern only slight changes. Moreover, it is also reflected in an irregular factor with more volatility obtained with the indirect method (see figure 5). Figure 4 Total Investment: Seasonal factors 108 106 104 102 100 98 96 Direct Indirect I II III IV 2003 I II III IV 2004 I II III IV 2005 I II III IV 2006 I II III IV 2007 -4- I 94 II III IV 2008 I II III IV 2009 I II III IV 2010 I 2011 92 Figure 5 Total Investment: Irregular Factor 102 101 100 99 98 Direct Indirect I II III IV 2003 I II III IV 2004 I II III IV 2005 I II III IV 2006 I II III IV 2007 I II 97 III IV 2008 I II III IV 2009 I II III IV 2010 I 96 2011 We also compared the diagnostics of the direct and indirect adjustments of the spectral figures to see if there was residual seasonality or trading day in the results. In both cases there isn’t any residual seasonality nor trading day. We couldn’t calculate the sliding spans and the revision history diagnostics because the series is not long enough. b) Demand for goods and services: Discrepancy between the aggregate (direct adjustment) and the sum of its components (indirect adjustment). Users´ questioning: In the last three quarters (3rd and 4th quarter 2010 and 1st quarter 2011) the direct seasonally adjusted aggregate demand for goods and services and the one obtained by adding their components adjusted have showed contrary signs about the magnitude of the growth (see figure 6). -5- Figure 6 Demand for goods and services: Seasonally adjusted series a) Annual percentage variation b) Quarterly percentage variation 15 8 10 6 4 5 2 0 0 -2 -5 Direct Indirect Direct Indirect -10 -4 -6 -15 -8 -20 I II III IV I II III IV I II III IV I II III IV I II III IV I II III IV I II III IV I 2004 2005 2006 2007 2008 2009 2010 -10 I II III IV I II III IV I II III IV I II III IV I II III IV I II III IV I II III IV I II III IV I 2011 2003 2004 2005 2006 2007 2008 2009 2010 2011 Concerning the case of aggregate demand, if you analyze the components we have that only in the Gross Fixed Capital the Easter effect is significant, with a value of 8 days, which added to obtain the indirect method would indicate, since this component represents only 14.9%, that the Easter effect in the demand is less or not significant. But the Easter effect is more significant when analyzing the total directly, with a value of 10 days. This could indicate that if the effect is not significant for the other components that does not mean there isn´t any effect in those components. With the value of that parameter for the Easter effect the adjustment in 2010 and 2011 is as follows: Demand for goods and services (Annual percentage variation) Period 2010 2011 Original Serie I II III IV I 8.23 13.17 9.42 7.36 6.16 Original Serie adjusted for Easter effect, direct method 8.84 12.54 9.42 7.36 5.57 Original Serie adjusted for Easter effect, indirect method 8.53 12.87 9.42 7.36 5.88 As you can appreciate in the table, there are differences in the adjustment of the Easter effect between the direct and indirect method, up to 0.3 percentage points in the first and second quarter of 2010 and in the first quarter of 2011, causing differences in the seasonally adjusted series also. Below are figures 7 and 8 of the original series adjusted for the Easter effect with the direct and indirect method and the corresponding factors. -6- Figure 7 Demand for goods and services: Original series adjusted by Easter effect Annual percentage variation 15 10 5 0 -5 Direct Indirect Differences I II III IV I 2004 II III IV I II 2005 III IV I 2006 II III IV I 2007 II -10 III IV I II 2008 III IV I II 2009 III IV -15 I 2010 2011 Figure 8 Demand for goods and services: Easter factor 100.6 100.4 100.2 100 99.8 Direct 99.6 Indirect I II III IV 2003 I II III IV 2004 I II III IV 2005 I II III IV 2006 I II III IV 2007 I II III IV 2008 I II III IV 2009 I II III IV 2010 I 99.4 2011 Additionally Total Consumption has two significant outliers LS2009.1 (change of level in the first quarter of 2009) and AO2009.2 (additive in the second quarter of 2009), Exports and Gross Fixed Capital Formation have an outlier LS2009.1 and Change in inventories -7- doesn´t have significant outliers, while for the aggregate in the addition of the outliers mentioned for the components also has a LS2008.4 (change of level in the fourth quarter of 2008). Seasonal factors are shown in figure 9 using both methods. You can see that factors calculated with the indirect method differ in the first and second quarters of each year from the factor obtained by the direct method and this is due to the difference in the Easter effect adjustment. This is also reflected in an irregular factor more volatile with the indirect method. Figure 9 Demand for goods and services: Seasonal factor 104 102 100 98 96 Direct Indirect I II III IV 2003 I II III IV 2004 I II III IV 2005 I II III IV 2006 I II III IV 2007 -8- I II III IV 2008 I II III IV 2009 I II III IV 2010 I 2011 94 Figure 10 Demand for goods and services: Irregular Factor 102 101 100 99 98 Direct Indirect 97 I II III IV 2003 I II III IV 2004 I II III IV 2005 I II III IV 2006 I II III IV 2007 I II III IV 2008 I II III IV 2009 I II III IV 2010 I 96 2011 Finally, we have to take into account that if we want to preserve the additivity between the aggregate and the seasonally adjusted components, we may not have consistent results and we may even get opposite conclusions to the ones we obtain analyzing the aggregate directly. -9-