傳統的預測模式建構與探討大都建立在線性與定態之假設上。但是對於一般經濟 時間數列，其走勢卻常呈現非線性與非定態之特性。本研究即探討如何應用神經網路之自由 模式學習概念，對非線型經濟時間數列提出一穩健的預測模型。並應用弱相關定理，證明了 神經網路逆傳導過程中的一個大數法則。 最後以臺灣地區公債發行額為例，將神經網路之預測結果與多變量時間數列模式做一預測效 果之比較分析。結果顯示，神經網路模式對於複雜之非線型時間數列走勢比多變量時間數列 模式能得到更佳的預測效果。

　　　　　Conventionally the research on the statistical forecasting has focused upon the modeling of dynamic data under the assumptions of linearity and stationarity. While in the socioeconomic system, many time series exhibit certain nonlinearity as well as non-stationarity. In this paper we try to use neural networks with feedback connections which provide a model free system to solve a rich class of forecasting problems. A fundamental statistical property. the strong law of large numbers for the weak dependent random variables, in the neural networks is also properly proved. We also present as illustrative example about the Government bonds issue of Taiwan and compare the forecasting performance with state space modeling. The forecasting results demonstrate that it is possible to employ a systematic approach in designing neural networks for forecasting problems and that large-scale neuron networks are capable of yielding high-quality forecasting to complex time series.