This action might not be possible to undo. Are you sure you want to continue?
RADMILA KOCURKOVÁ Silesian University in Opava School of Business Administration in Karviná Department of Mathematical Methods in Economics Czech Republic Tel.: + 420 69 63 98 283 Fax: + 420 69 63 12 069 e-mail: email@example.com
Key Words: time series analysis, process ARIMA, unemployment, programme SPSS
In my lecture I would like to tell you something about the time series, respectively about trends in the number of job applicants registered by labour offices in the Czech Republic. On the basis of results I will forecast the number of job applicants registered by labour offices till December 2000. My data was taken from statistic source: Ministry of Labour and Social Affairs. At our university I teach the course Time Series Analysis for the students of system engineering and information specialisation. The main aim of this subject is: investigation of the regular course of time series; deviation investigation of expected regularity; prediction of the future time series; regular determination of the convenient control directions for influence of time series. During the lessons I usually use MS Excel and statistic programme SPSS (Statistical Package for the Social Sciences). SPSS makes some operations with the time series possible in the basic module – Base; for deeper analysis it is necessary to join module Trends. For SPSS programme is the time series normal data file and it is presupposed that one row of the date nut contains the observation in one time and the rows ground in the way, that the oldest observation is the first, the youngest observation is the last row of the nut.
Time series analysis
Firstly, we have to describe the trend of time series of the applicants registered by labour office in the Czech Republic. From the Graph 1 we can say that January 1997 the number of job applicants grows much more than in the previous years. This time series is not stationary (which can be deleted by the seasonal difference of the time series). 836
We can do it by using autocorelate (ACF) and partial autocorelation (PACF) function of the stationary time series. From these values we can determine if the parameters can be used in this model or not. yt "1 . I (Integrative) – random walk. Graph 1 Trends in the number of job applicants registered by the labour office: January 1990 – January 2000 550000 500000 450000 400000 The number of job applicants 350000 300000 250000 200000 150000 100000 50000 0 Source: Ministry of Labour and Social Affairs At first. 837 99 19 9 C 99 DE 1 Y 98 A M 19 8 T C 99 O 1 7 R A 99 M 1 G 7 U 9 A 19 N 96 JA 19 5 N 99 JU 1 V 95 NO 19 PR 94 A 19 P 94 SE 19 B 93 FE 19 L 92 JU 19 2 C 99 DE 1 Y 91 A M 19 1 T C 99 O 1 R 90 A M 19 G 0 U 9 A 19 N JA . yt !1 . Autocorelate means the correlation’s between time series and the same time series lag. This coefficient presents only about direct structure. There are three parts (they do not have to contain always all of these): AR (Autoregressive) – linear combination of the influence of previous values. MA (Moving average) – linear combination of previous errors. for example between yt and yt ! 2 with the elimination of transmission over the observation yt !1 . In ARIMA models.. quite hard for computing and for the understanding of the results.. yt ! 2 . At the end we verify if the residual component is the white noise. Partial autocorrelations are also correlation coefficients between the basic time series and the same time series lag and we will eliminate the influence of the members between.Theory Trend and prediction of time series can be computed by using ARIMA model. In output of SPSS we compute signitications for the individuals parameters in ARIMA model. yt .d. In the case of computing more models we choose the model where AIC (Akai information criteria)..q) model is complex a linear model. respectively SBC (Schwartz-Bayes criteria) are minimal and Log likelihood is maximal. ARIMA (p. They are demand quality and a large number of dealing dates (it is assumpted at least 50 dealing or observations). we assume dependence between the quantities yt ! 2 . These models are very flexible. we have to identify the type of the model and the values of the parameters.
yt ! 2 s . This process is called SARIMA (p. Because time series is not stationary we have to differentiate it. yt " s .0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 -1. is order of seasonal process AR. . where s is the length of the period (in this case 12). is order of seasonal difference. by the first order.Q)s . The equation of this model is: ) P B s ( p #B $#1 ! B $d 1 ! B s ( p #B $ &q #B $ )P B %Q B s is autoregressive operator. is seasonal autoregressive operator. is seasonal operator of moving averages. is the order of process MA.5 Confidence Limits ACF -1.0 0.D.0 Partial ACF -. We also assume the season dynamics that is why we differentiate time series also by seasonal fist order. yt ! s . we can expect also the dependence seasons: yt ! 2 s . where p q d P Q D s is order of process AR. is the length of seasonal period.5 0. is the operator of moving averages.q) (P.5 -.If this process contains the seasonal fluctuation.d. is the order of MA. 3 Autocorrelation and partial autocorrelation function time series of the number of job applicants 1..0 1. yts ..0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Coef ficient Lag Number Transforms: difference (1). # $ # $D yt ' &q #B $%Q #B s $at . Graph 2. is white noise.0 .. seasonal difference (1. as it is in this model. where *at + # $ # $ s Practice For the fist we have to estimate the parameters of the model by using autocorrelation and partial autocorrelation functions. period 12) Source: own calculations 838 .5 . is the order of difference.
0) (0.0526 Standard error 4950.84 yt ! 26 " at We can say.79 yt ! 25 ! 0.43481456 SEB .0)12.84 yt !1 ! 0.1.8442 2149.0)12 Variable: The number of unemployment applicant Seasonal differencing: 1 No missing data.0) (1. The shorter SPSS output for model SARIMA (1.48 yt !14 " 0.00000000 . PROB. we get the following equation and it describes the dynamics of our time series: yt ' 1.0)12 or model SARIMA (1.653 12 11 new cases have been added.1.57 yt !12 ! 1.285775 -4.1.On the basis of the form functions we choose model SARIMA (1. .0) (1.1yt !13 " 0.43 yt ! 24 ! 0.83996600 -.1. Regressors: NONE Non-seasonal differencing: 1 Length of Seasonal Cycle: FINAL PARAMETERS: Number of residuals Log likelihood AIC 108 -1072.1. General equation of the chosen model is: 1 )1 B12 (1 #B $#1 ! B $1 1 ! B12 yt ' at .1.810591 APPROX.05157667 . that the number of job applicants this month is much more influenced by the number of job applicants in the last month. 839 .0) (1.00000501 SBC 2155.1. we choose this model.0)12.09038694 T-RATIO 16.84 yt ! 2 " 0. # $ # $ After the modification of the equation and substitution of estimated values.1.6884 Variables in the Model: B AR1 SAR1 . Akai information criteria and Schwartz-Bayes criteria are lower at the model SARIMA (1.
Résumé We can describe graphically the original time series of the number of job applicants and the predicted time series by SARIMA (1. Graph 4 The original and estimation values of the number of job applicants 60 0000 55 0000 The n umber of job applic ants 50 0000 45 0000 40 0000 35 0000 30 0000 25 0000 20 0000 15 0000 10 0000 5 0000 0 0 0 0 2 9 L 99 9 JU 1 99 C E 1 D Y 98 A 19 8 M T 99 C 1 O R 97 A 19 M G 97 U 9 A N 1 96 9 5 JA 1 99 N 1 JU V 95 O 9 N 1 94 R P 9 A 1 94 P E 9 S 1 93 B E 9 F 1 2 L 99 2 JU 1 99 C E 1 D Y 91 A 9 M T 1 91 C 19 0 O R 99 A 1 M G 90 U 9 A 1 N JA Original time s eries Estimation Source: own calculations Table 1 shows forecasting the time series in the number of job applicants registered by labour office in the Czech Republic. Table 1 Expected numbers of job applicants in year 2000 Month February March April May June July August September October November December The number of job applicants 512 094 509 978 499 388 495 024 507 073 528 465 539 225 549 669 546 019 551 165 573 290 Source: own calculations 840 .0) (1.1.0)12.1.
Very dangerous is the situation in some place in the Czech Republic. ISBN 80-04-25556-6. J. 290. S.1. where industry was the main sector of employment. 650 applicants. there are some jobs for construction workers. School of Business Administration is in one of such parts country (Karviná) and there is about 69. 667 applicants. especially in its three previous years. 024. M. FISCHER. Chicago: SPSS Inc. The lowest number of job applicants we can expect in April 2000 – around 10. ISBN 0-13-656935-8. Summary The unemployment is a big problem for almost every country. Graph 5 Expected monthly addition of the numbers of job applicants in year 2000 25000 Expected monthly addition 20000 15000 10000 5000 0 Au gu st Se pt em be r O ct ob er N ov em be r D ec em be r Fe br ua ry M ar ch Ap ril M ay Ju ne Ju ly -5000 -10000 -15000 Source: own calculations In year 2000 we can expect the highest addition number of job applicants in December around 22. Praha: SPN.4 applicants to one work place. In comparison with the same period of year 1999 the number of job of applicants has extended by about 85. 1997. July around 21. 1994. The lowest number of job applicants we can expect in May 2000 and that is 495. agriculture workers. Chicago: SPSS Inc.5. R. 392 (in this period graduated students from every school are registered by labour offices)... This is the consequence of the seasonal unemployment. 590 and in October 3.In year 2000 we can expect the highest the numbers of job of applicants in December 573. ISBN 0-13-201055-0. In this period. SPSS Professional Statistics 7. and young people will go abroad for a job. the Czech Republic has the problem with unemployment. Literature NORUŠIC. Makroekonomie. After the Velvet revolution. 841 . That is way I decided to use SPSS software for the analysis of this problem in the conditions of the Czech Republic. DORNBUSCH. 1994. SPSS INC. 125.. SPSS Trends 6.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.