The Analysis and Forecasting of Unemployment in Croatia
Professor Ante Rozga, Ph.D., University of Split
Ekonomski fakultet Split, Matice Hrvatske 31, 21000 Split, Croatia
E-mail:
Ana Secer, BSc. Econ
Metula, Put sv. Lovre 55c, 21000 Split, Croatia
E-mail:
Josip Arneric, MSc. Econ., University of Split
Ekonomski fakultet Split, Matice Hrvatske 31, 21000 Split, Croatia
E-mail:
Abstract
Many factors influence trends in unemployment in Croatia. After proclaiming independence in 1991 Croatia had to battle the war for independence with devastating consequences particularly on unemployment which soared. During and after the war government decided to privatize social-owned companies which on the other side pushed once again unemployment upwards since in socialist economy productivity was much lower than in capitalist one. Foreign and domestic private owners of privatized companies tried to minimize the cost of labor force and they dismissed many workers.
The consequence of introducing market-oriented economy in Croatia has been ever increasing percentage of women among unemployed persons which is now 61.6 % (in 1991 it was 52.2%), compared with 51.5% in the total population. Also, the number of unemployment benefit recipients has fallen dramatically because welfare state in market-oriented economy is not important as in socialist one. The educational background of unemployed persons has been also changed due to huge turmoil in Croatian economy.
When analyzing trends and forecasting the unemployment in Croatia we must take into account its seasonal character because Croatian economy relies very much on tourism and related industries with very strong seasonal variations. This could be helpful for unemployed persons to find a job for several months per year but it is not good for economy as a whole to rely too much on seasonal industries. Also, calendar variations are also very important. We tried several methods for seasonal adjustment and forecasting time series in order to find the most suitable one. Traditional methods of X-11-ARIMA from Statistics Canada and X-12-ARIMA from Census Bureau were confronted to model-based method TRAMO/SEATS. The newest method called X-13-ARIMA-SEATS tried to overcome some problems of both methods and to integrate it. Also, we used structural time series modeling by Harvey and Durbin to see different approach to modeling of time series. TRAMO/SEATS proved to be better when analyzing particular Croatian unemployment series compared with very popular empirically based methods such as X-11-ARIMA and X-12-ARIMA. Structural time series model could not be compared directly with traditional seasonal adjustment methods because of different assumptions, but gave very good results in adjustment and forecasting.
Keywords: unemployment, seasonal variations, seasonal adjustment, trend.
1. The analysis of historical data
We have analyzed number of unemployed persons from September 2000 until September 2007. The peak was in March 2002 with 415 352 unemployed persons. The number has fallen from 359 921 at the beginning of the period observed to 246 191 at the end of time interval, which is the fall of 31.6%. The number of persons seeking first job for has fallen by 47.6% and the number of unemployment benefit recipients has also fallen by 17.1%. The latest figure is due to fact that redundancy jobs have effected mostly older employees with the long lasting unemployment benefits. Regarding professional attainment the things have also changed. In the year 2000 the share of unskilled persons was 18.9%, while in the year 2007 it was only 7.3%. The share of unemployed persons with tertiary education was 7.3% in 2000 compared with 7.4% which is almost the same, but the share of tertiary education among employees has been improved.
Figure 1. Unemployed persons
2. The analysis of seasonal variations
Since the unemployment is affected by seasonal variations we have used two programs for seasonal adjustment and decomposition of time series. The first is very popular one, namely X-12-ARIMA developed by Census Bureau in US. The second method which has been slightly introducing as an official method in some European countries is TRAMO/SEATS method developed in Banco de España. TRAMO/SEATS is model-based method. Scientists from Census bureau are developing X-13-ARIMA-SEATS which is to encompass some disadvantages of X-12-ARIMA combined with TRAMO/SEATS.
Figure 2. Final seasonal factors
As we can see from the figure above there is no big difference between two methods regarding final seasonal factor. Both methods detected one additive outlier (August 2002) and one temporary change (November 2001).
Comparing the quality of seasonal adjustment between two methods, we have used seasonal adjustment quality index. The SA quality index is calculated using all diagnostic statistics that are also applied in usual quality check. Quality index vary from 0 (the optimal value) to infinity (the worst value). If the value exceeds the standard base value of 10 at least one test statistic must have been significant. In our example, SA quality index for X-12-Arima was 2.67 and for TRAMO/SEATS 1.492, thus giving the advantage to TRAMO/SEATS method regarding the quality of seasonal adjustment.
It is clearly seen that seasonal factors are getting higher over time, which means the greater seasonality in more recent years. This is the consequence of booming in tourism, trading, travel and leisure industry which are labor intensive industries and as a such giving significant seasonal variations on unemployment. It is not so good for labor market since tourist season in Croatia is rather short, but still improving and enlarging.
When analyzing irregular factors we could see more differences between two methods.
Figure 3. Final irregular factors
We can conclude that X-12-ARIMA produces higher irregularities, particularly at the beginning and at the end of time series.
Figure 4. Final trend in unemployment in Croatia
Unemployment has been falling since 2002 and the trend is like that, showing downward tendency.
Figure 5. ACF of residuals
3. Forecasting of unemployment in Croatia
Using both methods we tried to forecast the number of unemployed persons in Croatia for the next twelve months.
Figure 6. Forecasts of unemployment in Croatia
Two methods are quite similar when forecasting trend in unemployment.
Figure 7. Forecasts of seasonal factors
Forecasts of final seasonal factors for the next twelve months show their slightly rising as it was the tendency over the historic time series. The differences between two seasonal adjustment methods were not very large.
4. Conclusion remarks
After the outbreak of the war in Croatia the unemployment in Croatia soared, pushed by destruction of industries as well as the economic transition from socialist type economy to market oriented one. When the war finished and the economy has been consolidated the number of unemployed persons has been ever falling. Booming of service industries pushed by direct foreign investments contributed very much for that fall. We must also include demographic changes with the births dramatically falling.
Regarding methods for seasonal adjustments we have concluded that method TRAMO/SEATS is slightly better when focusing on seasonal adjustment quality index. The other differences regarding trend and forecasts were not so big.
References
Bell, W.R. and Nguyen, T.T. (2002): Comparison of Time Series Characteristics for Seasonal Adjustment from SEATS and X-12-ARIMA. American Statistical Association, Proceedings of the Business and Economic Statistics Section (CD-ROM).
Dent, A.Y., Hood, C.C.H., McDonald-Johnson, K.M., and Feldpausch, R.M. (2005): Comparing the ARIMA Model Selection Procedures of X-12-ARIMA Versions 0.2 and 0.3. American Statistical Association, Proceedings of the Business and Economic Statistics Section (CD-ROM).
Feldpausch, R.M., Hood, C.C.H., and Wilks, K.C. (2004): Diagnostics for Model-Based Seasonal Adjustment. American Statistical Association, Proceedings of the Business and Economic Statistics Section. (CD-ROM).
Hood, C.C.H. (2002): Comparing the Automatic ARIMA Model Selection Procedures of TRAMO and X-12-ARIMA Version 0.3 and the Seasonal Adjustment of SEATS and X-12-ARIMA. Unpublished work presented at the Eurostat Working Group on Seasonal Adjustment Meeting. Luxembourg, April 2002.
Monsell, B.C., Aston, J.A.D., and Koopman, S.J. (2003): Towards X-13. American Statistical Association, Proceedings of the Business and Economic Statistics Section. (CD-ROM).
1