教育政策決定教育改革的方向，如何精準地掌握教育改革政策的預測性，建立 預測模式，值得國內於教育改革聲中進一步深入探究。本研究採用美國西部高等教 育協會（Western Interstate Commission for Higher Education, WICHE）公布的九筆 時間序列資料，以樣本外預測能力的比較研究方法，來檢驗灰預測模型之預測能力 是否較Prescott（2008）研究中使用的平均世代生存比例法為佳。本研究建立兩種 灰預測模式，分別為GM（1, 1）和GM（1, 1）滾動模型，與平均世代生存比例法 比較。為檢視模型預測之能力，採用MAD、MAPE及RMSE三種常用誤差值檢驗各 模型樣本之預測能力。研究結果指出，GM（1, 1）之預測能力分別在六筆教育成 就變數之時間序列預測力較GM（1, 1）滾動模型及平均世代生存比例法誤差值 小；而平均世代生存比例法則在兩項時間序列預測結果中略為精確；GM（1, 1） 滾動模型之預測能力則最不具競爭力。研究結果建議，灰預測模型可供台灣教育政 策及教育行政管理等相關短期預測研究參採使用，期有助於台灣教育政策決策及教 育行政管理更精準的預測分析，減少不必要的紛擾，有助於教育改革的順利推展。

Projecting educational composition of a population is generally regarded as having a wide array of contributions to policy planning for educational development and reform of a nation. Since better forecasts could translate into better strategic planning and decision making, this study aims at comparatively judging forecasting efficiency of grey prediction models on school graduates vis-à-vis that of weighted average cohort survival ratio (CSR) employed by Western Interstate Commission for Higher Education (WICHE). An out-of-sample forecasting experiment was carried out to compare forecasting performances across GM (1, 1), GM (1, 1) rolling derived from the grey system theory and weighted average CSR. Predictive power of each model were measured based on MAD, MAPE and RMSE. Forecasting accuracy measures revealed that the GM (1, 1) model outperformed the other two models in projecting six categories of school graduates; weighted average CSR yields more accurate results in two categories of school graduate. The result shows that GM (1, 1) model displays promise for use in forecasting school graduates. Since educational attainment, human capital formation and economic growth of a nation are interlinked, the findings of the study would be useful for professional forecasters who are expected to provide government authorities and education policy makers with accurate information for planning future policy directions and optimizing decision making.