This report has been prepared within the framework of Intervention 3.2 component of the “Strengthening the Impact of Conditional Cash Transfer (CCT) for Increasing High School Attendance Rate in Türkiye (CCT II) Technical Assistance Project” in order to foresee the demand for future Social Assistance/CNT in Türkiye.

In this project, the demand for the Conditional Cash Transfer for Education (CCTe) provided by the state to combat poverty in Türkiye, is examined. The CCTe programme consists of assistance provided to families in need and without social security, on condition that their children continue their education. The main objectives of the CCTE are to increase schooling rates, thus contributing to development by improving human capital.

In order to determine the variables determining the demand for CCTe, a mixed design research method, including qualitative and quantitative methods, was used. In the first stage, 12 cities with a high number of beneficiaries of the CCTe programme and ensuring regional representation were selected. In each city, workshops were organised in which participants who were involved in CCTe implementation or were implementing actors shared their perspectives. In the workshops, firstly, the perspectives of the participants on the factors that constitute the current demand for the CCTe programme were obtained. Thus, a general picture of the concepts that currently determine the demand for CCTE has emerged. Afterwards, participants were asked for their opinions on social, economic, demographic, and cultural phenomena that may increase or decrease the demand for CCTe. These cases were requested to be associated with short, medium, and long term risks. The identification of both current and future variables of demand for CCTE is based on the participants’ experiences and observations. In addition, the specific socio-economic characteristics of each city were also taken into account in the evaluation.

According to the perspectives of the participants in the workshops, the main factors determining the demand for CCTe in the current situation are as follows:

• Neediness
• Unemployment
• Disability
• Household composition
• Educational attainment
• Number of social aid programs benefited from

In the evaluations regarding the demand for CCTe in the future period, both the factors that will increase demand and the factors that will decrease demand were emphasised. While the socioeconomic variables that are thought to be effective in demand differed according to the cities, common variables for demand emerged in 12 cities. These variables are stated as:

• Substance use rate
• Crime rate
• Divorce rate
• Rate of single-parent households
• Migration rate
• Aid amount
• Number of assistance programs
• Education level (Net schooling rate, number of higher education graduates)
• Economic growth
• Educational attainment
• Employment rate
• Unemployment rate
• Child labor rate
• Inflation
• Household composition

* These indicators, which are transformed into variables from qualitative word strings, can be represented by indicators of economy, family structure (number of divorced, single parent, children living with grandparents) and education. In order to use these variables in quantitative analysis, poverty index, human development index, unemployment rate and population were used.

In the quantitative part of this study, a comprehensive analysis was conducted using various statistical models to predict the future demand for CCTE in Türkiye. The Project aimed to forecast the demand for CCTE in 2025, 2030, 2035 and 2040. Time series data and various univariate and multivariate forecasting models are used in the analyses.

Demand for CCTe is measured by the number of applications for conditional education assistance. CCTe demand data for the period of July 2013 – October 2023 were obtained on a monthly basis from the Republic of Türkiye Ministry of Family and Social Services. In the modelling, the number of CCTE applications is not used directly, and the number of CCTe applications per 1 million population is used. In the 2014-2022 period, the number of CCTe applications does not exhibit a clear trend, and the number of applications shows a behaviour relatively parallel to the time axis in terms of value over time. For this reason, modelling the number of applications directly as the explained variable may lead to the absence of a significant explanatory variable. This will pose a problem especially in ARDL and similar models, which are established as a time series regression model, and which are also studied in this study. However, while the number of applications remains relatively static over time, there is a dynamism in the number of applications per population. For this reason, the predicted values of the application demand are indirectly obtained by taking the number of applications per million population as the explained variable.

In the methodology of the study, univariate and multivariate forecasting models were used:

1. Univariate forecasting models
2. Autoregressive Integrated Moving Average (ARIMA)
3. Multilayer Perceptron Artificial Neural Network (MLP)
4. Multivariate predictive models
5. Autoregressive Distributed Lag Model (ARDL)

As an alternative to univariate models, two hybrid models that are expected to improve the prediction success are considered: CEEMDAN-ARIMA and CEEMDAN-MLP. RMSE, MAE and MAPE criteria were used to evaluate the prediction performance of all models.

In all prediction models, the number of CCTe applications per population series was divided into 2 parts as training data for the period between 2013:07-2021:09 and test data for the remaining period between 2021:10-2023:10, and the performance of the prediction models was compared using the test data.

In the multivariate forecasting model ARDL, Sen-Shorrocks-Thon (SST) poverty index, Human Development Index (HDI) and unemployment rate are included in the model as explanatory variables. For all three explanatory variables, projections were made until 2040 with the hybrid CEEMDAN-ARIMA model.

According to the results of the analyses, among the univariate forecasting models, the hybrid CEEMDAN-MLP model has the best forecasting performance according to RMSE, MAE and MAPE criteria. Considering the MAPE value, the prediction success of the CEEMDAN-MLP model is 7.69% higher than the MLP model. This result confirms the expectation that hybrid models improve the forecasting performance.

On the other hand, among the multivariate forecasting models, the ARDL model, i.e. the model with SST, HDI and unemployment rate as explanatory variables, has the best forecasting performance. Considering the MAPE criterion, the predictive performance of the ARDL-2 model improves by 11.86% compared to the ARDL-1 model without the unemployment rate. This result shows that the unemployment rate has a successful effect on CCT demand forecasting.

The demand for CCTe application is foreseen under three scenarios. These scenarios are labelled as normal, lower and upper. Forecasts for the scenarios are given for CEEMDAN-MLP, which has the best performance, as well as for ARDL, which is based on the structural relationship between variables and has a very satisfactory forecasting performance. Each scenario is created separately for population and unemployment rate. In the normal scenario, demand for CCTe is projected using point estimates for median population and unemployment rate. In the lower and upper scenarios, the prediction intervals of the population and unemployment rate were used.

• As a result of population-based scenarios, the number of CCTe applications in 2040 is expected to reach 420 thousand in the lower scenario, 441 thousand in the normal scenario and 463 thousand in the upper scenario.
• As a result of the unemployment rate-based scenarios, the number of CCTe applications in 2040 is expected to reach 337 thousand in the lower scenario, 479 thousand in the normal scenario and 680 thousand in the upper scenario.
• For the scenarios where both population and unemployment rate are considered together, the number of CCTe applications in 2040 is projected to increase to 320 thousand in the lower scenario, 478 thousand in the normal scenario and 713 thousand in the upper scenario.

In conclusion, the findings of this project provide important information to improve the effectiveness of the CCTe programme in Türkiye and to prepare for future increases or decreases in demand. Using these analyses, the government and relevant institutions can plan and implement social assistance policies and resource allocation more effectively. In the findings obtained as a result of the qualitative research, it was observed that the concepts of unemployment, employment and population structure in terms of economic and social indicators are determinant in the demand for CCTe. In this context,

• Since unemployment and employment characteristics are determinant for the interest of social assistance beneficiaries towards the CCTe demand, job and occupational profiles of household members should be created.
• The process of directing social assistance beneficiaries to employment, which is currently carried out with İŞKUR, should be strengthened in such a way that the household does not need assistance.
• The condition of no social security in the household in order to be a CCTe beneficiary creates a problem in the link between assistance and employment. The link between employment and the CCTe should be strengthened, as the rate of informal employment is expected to be one of the variables that will determine the demand for assistance in the future if the eligibility criteria remain the same.
• The expectation that the increase in the unemployment rate will also increase the demand for the CCTe, the employment process should be monitored in the long term, as well as the employment of household members who are able to work in households benefiting from social assistance programmes, including the CCTe.
• Within the scope of population structure, different and various indicators such as composition of households, crime rate, substance dependency rate, habits of receiving aid, health indicators and education level are expected to affect the demand for CCTE. Therefore, households should be monitored and supported holistically and through social services, especially in social assistance programmes such as CCTe where children are the subjects.
• Since the increase in the number of single-parent households and divorce rate is expected to be reflected on the demand for CCTE in the medium and long term, family trainings are recommended for parents in households benefiting from assistance.
• There is a need for practices to strengthen the socio-economic characteristics of households that are CCTe beneficiaries and long-term beneficiaries of many social assistance programmes. The habit of benefiting from social assistance, which is expressed as a factor that increases and is expected to increase the demand for CCTe, makes poverty a long-term permanent feature for a family.
• There is a consensus that increasing the amount of the CCTE would increase the demand for this assistance programme. Rather than increasing the amount of this aid, efforts should be made to find a threshold value that will facilitate the schooling of children.
• Currently, the education level of parents in CCTe beneficiary households is low. Improving the awareness of families on education is expected to reduce the demand for CCTE in the long run. For this purpose, families eligible for CCTe should be informed about the content and importance of the assistance at regular intervals.
• The development in the industry and tourism sectors in cities, which is expressed as a factor expected to reduce the demand for CCTe, brings along the problem of child labour. Children from low-income families or families who prefer vocational education instead of academic education should be directed to MESEMs.
• Since health indicators of household members affect household income and welfare level, this situation is indirectly reflected on the demand for CCTe. Households with individuals with chronic diseases, in need of care, elderly and/or disabled people are currently highly interested in CCTE. Based on the prediction that households with high vulnerability level in terms of health indicators will increase the demand for CCTe in the coming period, all health-related processes of vulnerable individuals in these households should be monitored.