1. ——Comparing between the methods based on X-12ARIMA and TRAMO/SEATS By Ming-Fang Zhang Suo-Xian Pan IAOS Conference, Shanghai, 14-16 October, 2008 Study of measuring the season change mode for main price indexes in China

2. Background Up to now, official statistic organizations of many countries in the world have studied and published seasonally adjusted macro-economic indicators. However, no seasonally adjusted macro-economic indicators have been published by Chinese government statistics agency. Neither have these indicators been deeply measured. Are there any season changes in China’s main price indexes? How to measure it? This paper studies these two questions. The paper uses X-12ARIMA and TRAMO/ SEATS to measure the season change mode of China’s main price indexes.

3. Defects of Present Methods  In china, the most recent values are compared with the corresponding values a year ago to obtain a time series free of seasonal fluctuations. However, an approach based on year-on-year comparisons has following defects:  1.Influenced by the seasonal factors of the base period.  2.Turning points would always be delayed.  3.Seasonal factors would not be eliminated thoroughly.  4.Accurate seasonal mode cannot be measured

4. Measuring Methods  1.TRAMO/SEATS(model-based) [See Björn FISCHER(1995)]

5.  2.X-12-ARIMA(empirical-based) [See D. F. Findley etc(1998)]

6. Empirical Research Main Prices Indexes  CPI-Residents Consumer Price Index  RCPI-Retail Commodity Price indexes  PPIR-Purchasing Price Index of Raw Materials, Fuels and Power  PPI-Producer Price Index of Industry Products Original Data: China Economic Information Network

7. Data transformation Trend of China’s main prices indexes(base year=1996)

8. Length of time series & Base year  “Increasing the length of the time series will generally increase the accuracy of estimation, but not necessarily in direct proportion to the amount of data. One of the problems with particularly long time series is that frequently the pattern changes with time and using a longer time span may in fact give less accurate results than using only the minimum number of data necessary.”——see Derek Blades(2001)  Examine the presence of seasonality by X-12-ARIMA  Selecting Rules: 1.Larger F s -value in test for the presence of seasonality 2.Easy to compare highly-related indexes 3.Less presence of moving seasonality  1996 as base year for CPI & RCPI (132 months)  1998 as base year for PPI & PPIR (108 months)

9. Measuring the mode of seasonality  Software:Seasonal Adjustment Interface DEMETRA for Tramo/Seats and X-12-ARIMA(User manual), EUROSTAT, February 2000  Adjust using default options  Since China does not celebrate Easter, we are not going to take it in to account in the seasonal adjustment

10. Comparison between two methods  Test for seasonality  Correlation test

11.  Graph Comparison(1) Final seasonal factor of CPIFinal seasonal factor of RCPI

12.  Graph Comparison(2) Final seasonal factor of PPIRFinal seasonal factor of PPI

13. Comparison of seasonal modes among main prices indexes  Range of seasonal fluctuation(See Graph)See Graph  Correlation coefficient of final seasonal factors

14.  Final Seasonal Factor/Component Per Month CPI RCPI PPIRPPI

15. Diagnostic Idempotency criterion  If is the filter output at time t after using the seasonal adjustment filter for the first time and is the filter output at time t after using the filter for a second time, M is number of time series data, a measure is the mean absolute percentage difference(MAPD)

16. Estimated idempotency criterion value(r%) For RCPI, CPI and PPI, the estimated values in T/S is much smaller than those in X-12, which means T/S eliminates more seasonal factor.

17. Sliding spans  There are four estimates S i,j (1),…, S i,j (4) for the seasonal factor of X i,j in the ith month in the jth year. We identifies the ith month in the jth year as having an unreliable seasonal factor if  S(%)=percentage of months with unreliable seasonal factor estimates  MSP=mean value of Sp i,j

18. Measuring results of X-12 are relatively stable. Standard of Sliding spans diagnostic [See D. F. Finley(1990)] Results of Sliding spans diagnostic

19. Conclusions and Suggestions Conclusions  China’s main price indexes have significant seasonality  Different price index have different seasonal mode  1996 as base year, CPI & RCPI have similar seasonal mode: go down overall with rebound in the middle of the year; 1998 as base year, PPI & PPIR have similar seasonal mode: go up overall  Indexes calculated by X-12 and T/S are highly correlative  T/S eliminates seasonal factor more sufficiently, Measuring results of X-12 are relatively stable.

20. Suggestions  Pay more attention to measuring seasonal mode and seasonal adjustment  Establish and perfect statistic system about monthly price index time series  Bring in more techniques about seasonal adjustment and innovate them

21. Thank you for your attention!