流浪教師問題是目前教育界中ㄧ重要問題，流浪教師數的預測精準與否，將會影響教育政策的裁定。本研究中，使用多變量門檻自迴歸模式，預測100年度到103年度的流浪教師數量。結果顯示，多變量門檻自迴歸模式較ARIMA模式更能顯現數列的趨勢，對於預測上有極大的幫助。且多變量門檻自迴歸模式的可用範圍很廣，因為一般的時間數列中或多或少都會有結構改變的現象，時間數列的資料普遍存在有非線性現象，且同時受到多個變數影響，此時加入多個外生變數作為考量，更能精準分析資料和做預測。

The vagabond teachers in elementary schools is an important problem in education administration. An accurate forecast of the number of vagabond teachers in elementary schools may heavily affect educational policy. In this thesis, we use multivariate TAR model analysis to forecast the number of vagabond teachers in elementary schools in Taiwan Area during a period from 100 to 103.
According to the result, multivariate TAR model perform well for prediction. Multivariate TAR model can be widely used in different circumstances, especially complicated situation. As far as common time series data is concerned, it has change point or change period occurs.Structural change of a non-linear time series is auniversal phenomenon. Selecting suitable data variables and using exogenous variables to be a threshold, we could obtain better predictable effect by multivariate TAR model.

中文部分：
[1]. 吳柏林(1995) 時間數列分析導論。台北：華泰書局。
[2]. 楊奕農(2009) 時間序列分析：經濟與財務上之應用。台北，雙葉書廊。
[3]. 歐用生(1998) 展望師資培育法的修訂。載於歐用生著：新世紀的學校。台北：台灣書店。頁255-264。
[4]. 教育部(2011) 中華民國教育統計。台北市：教育部編印。
[5]. 教育部(2009) 師資培育統計年報。台北市：教育部編印。
[6]. 黃昆輝(1975) 台灣省未來六年國小教師需求量之推估研究。台北市：國立台灣師範大學教育研究所。
[7]. 馬信行(1987) 我國各級學校未來學生數之預測。國立政治大學學報。56期，頁111-167。
[8]. 馬信行(1990) 時間數列分析之轉換模式在學生數預測上之應用。國立政治大學學報。61期，頁237-273。
[9]. 吳柏林、許瑞雯(1990) 臺灣地區國中教師數預測模式。教育與心理研究。17期，頁29-44。
英文部分：
[1]. Tong H. and Lim K. S. (1980). Threshold Autoregressive, Limit Cycles and Cyclical Data (with Discussion), Journal of the Royal Statistical Society. Series B, Vol.42, No.3, pp.245-292.
[2]. Subba Rao T. and Gabr M. (1980). A test for linearity of stationary time series analysis, Journal of Time Series Analysis , Vol.1, No.1, pp145-158.
[3]. Haggan V. and Ozaki T. (1980). Amplitude-dependent Exponential AR Model Fitting for Non-linear Random Vibrations, in Time Series, (O. D. Anderson ed.), North-Holland, Amsterdam.
[4]. J. D. Byers and D.A. Peel (1995). Evidence on volatility spillovers in the interwar floating exchange rate period based on high/low prices, Applied Economics Letters, Taylor and Francis Journals, Vol.2, No.10, pp394-396.
[5]. Liu Y, Garceau NY, Loros JJ and Dunlap JC (1997). Thermally regulated translational control of FRQ mediates aspects of temperature responses in the Neurospora circadian clock, Cell, Vol.89, pp477–486 .
[6]. Bai Jushan and Pierre Perron (2003). Computation and Analysis of Multiple Structural-Change Models, Journal of Applied Econometrics, Vol.18, No.1, pp1–22.
[7]. Donald W.K. Andrews and Werner Ploberger (1994). Optimal Tests When a Nuisance Parameter Is Present Only under the Alternative, Econometrica, Vol.62 No.6, pp1383-1414.
[8]. Kumar K and Wu B (2001). Detection of change points in time series analysis with fuzzy statistics, International Journal of Systems Science, Vol.32, No.9, pp1185-1192.
[9]. Zhou H. D. (2005). Nonlinearity or structural break? - data mining in evolving financial data sets from a Bayesian model combination perspective, Proceedings of the 38th Hawaii International Conference on System Sciences.
[10]. Hansen, Bruce E. (1999). Testing for Linearity, Journal of Economic Surveys, Vol.13, No.5, pp551-576..
[11]. Shen Chung-Hua, and David R. Hakes (1995).Monetary policy as a decision-making hierarchy: the case of Taiwan. Journal of Macroeconomics, Vol.17, No.2, pp357-368.
[12]. Sharma S. (1996). Applied Multivariate Techniques, John Wiley & Sons. New York, USA.
[13]. Tsay Ruey S. (1989). Testing and Modeling Threshold Autoregressive Processes, Journal of the American Statistical Association, Vol.84, No.405, pp231-240.
[14]. Akaike, H. (1973). Information theory and an extension of maximum likelihood principle, Second International Symposium on Information Theory, Vol.1, pp267-281.