사망원인에 따른 국내 월별 사망자 수 적합 및 예측을 위한 정수값을 갖는 시계열 모형들 간의 비교연구

다국어 초록 (Multilingual Abstract)

In this study, we compare the fitting and forecasting performance of integer-valued time series models using monthly mortality data in Korea. The data were obtained from the Korean Statistical Information Service （KOSIS） and consist of monthly death counts from January 1995 to December 2024.

Three count time series models were considered: the INGARCH（1,1） model, the NBINGARCH（1,1） model, and the log-linear Poisson autoregressive model. The analysis was conducted for three groups of causes of death based on the International Classification of Diseases （ICD）: neoplasms （C00–D48）, diseases of the respiratory system （J00–J98, U04）, and external causes of morbidity and mortality （V01–Y89, U12）.

Model parameters were estimated using the maximum likelihood method. Model fitting performance was evaluated using the Akaike Information Criterion （AIC）. Forecasting performance was assessed through a rolling one-step-ahead forecasting scheme based on the last 50 observations of each series. Mean Squared Error （MSE） and Mean Absolute Error （MAE） were used as measures of forecasting accuracy.

The results indicated that the NBINGARCH（1,1） model produced the smallest AIC values for all datasets, suggesting superior goodness-of-fit compared with the other models. However, forecasting performance differed across datasets. For the C00–D48 dataset, the INGARCH（1,1） model yielded the lowest forecasting errors. For the V01–Y89,U12 dataset, the NBINGARCH（1,1） model achieved the best forecasting performance. In the J00–J98,U04 dataset, the forecasting performances of the three models were very similar, with no clear superiority among them.

These findings suggest that the model with the best fitting performance does not necessarily provide the most accurate forecasts. Therefore, selecting an appropriate integer-valued time series model according to the characteristics of the data is important for the analysis and forecasting of mortality counts.

Keywords: Count time series, INGARCH model, NBINGARCH model, Log-linear Poisson autoregressive model, Forecasting, AIC.

목차 (Table of Contents)

• 1. 서론 1
• 2. 연구 방법 4
• 2.1. 모형 소개 4
• 2.1.1. INGARCH(1,1) 모형 5
• 2.1.2. NBINGARCH(1,1) 모형 6
• 2.1.3. Log-Linear 포아송 자기회귀 모형 8
• 2.2. 모형의 비교: AIC 9
• 2.3. 예측 분석 10
• 3. 실증 분석 12
• 3.1. 데이터 소개 12
• 3.2. 분석 결과 19
• 4. 토의 및 결론 31
• 4.1. 연구 결과 요약 31
• 4.2. 연구의 의의 32
• 4.3. 연구의 한계점 및 향후 연구 32
• 참고문현 34
• Abstracts 35