| 摘要: | 摘要 本研究利用時間數列分析ARIMA模型建立全民健保應收保費預測模型，亦即利用單變量ARIMA模型，來建立全民健康保險應收保費預測模型；但由於單變量模型缺乏先驗情報，對模型之解釋能力較難令人信服，而在許多的例子中有可能發生一筆資料其目前的觀測值受到過去的觀測值影響，並且與另一筆（或多筆）時間數列資料具有相關性，則考慮利用轉換函數模型（Transfer Function Model）來建立全民健康保險應收保費模型。另應收保費時間數列會受到政策之修訂之影響，如平均眷口數及眷口數上限的修正，而使時間數列受到干擾，影響模型建構進而降低預測之精確結果。因此，利用介入模型（intervention model）期更能建立合適之模式。同時針對上述所建構之單變量模型、轉換函數模型、介入模型作比較，何者預測效果較準確。研究資料以SCA套裝軟體進行分析，所得結果如下： 一、 從單變量模型中來考量，其模型顯示全民健保應收保費資料符合ARIMA（1,1,0）的模型，亦即應收保費受本身前一期之影響。 二、 利用介入模型預測來建立預測模式，除了考量本身前一期之影響，亦應考量平均眷口數修訂因素。 三、 從二元轉換函數模型來建立預測模式，顯示應收保費除了受本身前一期之影響，亦應考量納保人口當期之影響，以及人口月增加率之前一期影響。由多元轉換函數模型來建立預測模式，研究發現應收保費除了與本身前一期有關外，亦應考量納保人口之當期與前一期以及平均眷口數第一、二次修訂。 四、 以RMSE及MAPE比較本研究所決定之預測模型，從結果中我們可以發現，所建立之五種模型效果均可接受，而且MAPE均小於10﹪，顯示各模型均具有高精確預測能力，其中又以轉換函數+介入模型較佳。 預測的目的在「估計」未來，而其動機則是探究影響因素欲「控制」現象。因為祇有透過準確的預測，瞭解應收保費結構變化，再配合組織系統運作，才能充分發揮經營決策之效率。 關鍵字：應收保費 /時間數列分析/ARIMA模型/介入模型/轉換函數模型; ABSTRACT This research intends to apply the ARIMA model of time series to set up a model to estimate the premium receivable of the National Health Insurance（NHI） system. It is to say that the model will be in an univa- riate one; but due to the lack of previous information of the univariate model, the explanation of the model is not rather convincing. There may be one of numerous subjects whose present observation data affected by its past record and then related to one or more other time series, so we consider applying Transfer Function Model （TFM）to establish the model of the premium receivable of the NHI system. Furthermore, the amendm- ent of the NHI ACT would influence the premium receivable of NHI; for instance, the averaged number of the dependents and the ceiling number, and then influence the time series and reduce the accuracy of the interve- ntion of the establishment of the model. Therefore, by making use of the Intervention Model（IM）can we expect to set up an even more suitable model. Meanwhile, we compare the univariate model, TFM, and IM in order to find out which one is the most effective one. We implement SCA to develop our analysis, and the result of each model is as follows: 1. According to the univariate model, the premium receivable data of NHI fit into ARIMA (1,1,0 ). It is to say that the premium receivable is affected by its first period. 2. We should consider the influence of the previous period and the amen- dment of the ACT in case of using the Intervention Model to set up the estimating model. 3. The estimating mode is established by bivariate TFM. The result sho-ws that the premium receivable is related to both its previous period and the present influence of the insured and the affection of the popul-ation growth ratio of the previous period. The estimating mode is esta-blishhed by multiple TFM and the result implies that the premium receivable is related to its previous period and the population of the insured of the first period and the two amendment of NHI ACT stated at the begining. 4.The MAPE of the models are all less than 10% and this proves that the five models are good enough to fit the data. However, by comparing the RMSE and MAPE in each model did we find those percentages in Transfer Function Model and Intervention Model are smaller than those in the univariate model. It implies that these two models can generate more accurate estimating especially the Transfer Function Model +Intervention Model proves to be the optimum one. The purpose of estimating is to predict what will happen. In addition, its priority is to break down the factors of the estimated. Only by applying accurate estimating, properly understanding the change of premium struc- ture can we raise the efficiency of the operation strategy under the NHI organization. |
