Scholarlink

Background <p>Somalia has to be properly prepared to plan its post-conflict operations because the country is in the midst of a major demographic transition. This paper is driven by the fact that there is a dire need to have accurate forecasting frameworks that can guide the regional policy and socioeconomic development of the Horn of Africa.</p> Purpose <p>This study assesses the predictive performance of seven advanced univariate time-series models and twelve ensemble methods derived from various hybrid combinations. It particularly responds to the gap in comparative forecasting literature to do so in a volatile post-conflict situation of Somalia.</p> Methods <p>The study used the World Bank data between 1960 and 2023, which was preprocessed using linear interpolation and second-order differencing to stabilize structural breaks that are detected. Models were tested on a 90/10 training-test split and benchmarked using measures like MAPE and Diebold-Mariano test.</p> Findings <p>Parsimonious ETS and hybrid ARIMA-ETS models were the best and most accurate with a MAPE below 1% with a vastly superior performance compared to complex machine learning architectures. According to the study, a consistent reduction of the TFR of Somalia is expected, to about 5.34 births per woman in 2030.</p> Conclusions <p>The results show that Somalia has been experiencing a sound downward fertility trend, which has marked the termination of a period characterized by stagnant transition. Complex algorithm architectures are less robust and reliable in data scarce and volatile environments, where well-specified linear and state-space models are used.</p> Contribution <p>The research offers a confirmed methodological framework of demographic forecasting in the Horn of Africa through the use of the latest data (up to 2023). It shows that parsimonious hybrid modeling is better than machine learning complexes in demographic projections in post-conflict environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Forecasting Somalia’s total fertility rate: a comparative analysis of time series models in a post-conflict context

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月1日

查看全文
  • Abdifatah Mohamud Abdi,
  • Mohamed Abdirahim Omar

摘要

Background

Somalia has to be properly prepared to plan its post-conflict operations because the country is in the midst of a major demographic transition. This paper is driven by the fact that there is a dire need to have accurate forecasting frameworks that can guide the regional policy and socioeconomic development of the Horn of Africa.

Purpose

This study assesses the predictive performance of seven advanced univariate time-series models and twelve ensemble methods derived from various hybrid combinations. It particularly responds to the gap in comparative forecasting literature to do so in a volatile post-conflict situation of Somalia.

Methods

The study used the World Bank data between 1960 and 2023, which was preprocessed using linear interpolation and second-order differencing to stabilize structural breaks that are detected. Models were tested on a 90/10 training-test split and benchmarked using measures like MAPE and Diebold-Mariano test.

Findings

Parsimonious ETS and hybrid ARIMA-ETS models were the best and most accurate with a MAPE below 1% with a vastly superior performance compared to complex machine learning architectures. According to the study, a consistent reduction of the TFR of Somalia is expected, to about 5.34 births per woman in 2030.

Conclusions

The results show that Somalia has been experiencing a sound downward fertility trend, which has marked the termination of a period characterized by stagnant transition. Complex algorithm architectures are less robust and reliable in data scarce and volatile environments, where well-specified linear and state-space models are used.

Contribution

The research offers a confirmed methodological framework of demographic forecasting in the Horn of Africa through the use of the latest data (up to 2023). It shows that parsimonious hybrid modeling is better than machine learning complexes in demographic projections in post-conflict environments.