本文主要在探討非線性迴歸模式的推定問題。 首章為導論。敘述線性與非線性迴歸模式的定義及基本假設，討論最小平方推定，並 闡述非線性迴歸模式一般解的特性。 第二章為非線性迴歸模式的解法。討論最陡下降法、線性化法及Marquardt 折衷法， 並舉一實例以說明實際運算的過程。 第三章及第四章分別討論線性及非線性最小平方推定在幾何學上的意義，在樣本空間 及參數空間上探討誤差平方和等值線的特性。 第五章建立一個修正型的羅吉斯成長模式，以討論台灣地區電話需求成長的模式。 第六章為實例分析。以第二章所討論的方法，設計計算機程式，解決一電話非住宅用 戶所佔百分數的迴歸問題。 第七章為結論。對全文整體做一總結。 電子計算機程式及執行結果列於附錄中。

本論文共分三章，全數圍繞在消費尤拉方程式中，消費成長的高階動差在理論與實證上的重要性。分別說明如下： 本論文第一章討論消費高階動差在實證估計消費結構性參數之重要性。消費尤拉方程式是消費者極大化問題的一階條件，而自Hall (1978)起，估計消費結構參數如跨期替代彈性時，也多是利用這個尤拉方程式所隱涵的消費動態關係，進行估計。但是由於消費資料存在嚴重的衡量誤差問題，實證上多將尤拉方程式進行對數線性化，或是二階線性化後進行估計。 然而前述一、二階線性化，固然處理了資料的衡量誤差問題，卻也造成了參數估計上的近似誤差(approximation bias)。其原因來自於線性化過程中所忽略的高階動差實為內生，而與迴歸式中的二階動差相關。這使得即便用工具變數進行估計，仍然無法產生具有一致性的估計結果。這當中的原因在於足以解釋二階動差，卻又與殘差項中的高階動差直交的良好(valid)的工具變數無法取得。 我們認為在資料普遍存在衡量誤差的情況下，線性化估計尤拉方程式不失為一可行又易於操作的方法。於是我們嘗試在線性化的尤拉方程式中，將高階動差引入，並檢視這種高階近似是否能有效降低近似誤差。我們的模擬結果首先證實，過去二階近似尤拉方程式的估計，確實存在嚴重近似誤差。利用工具變數雖然可以少部份降低該誤差，但由於高階動差的內生性質，誤差仍然顯著。我們也發現，將高階動差引入模型，確實可以大幅降低近似誤差，但是在偏誤降低的同時，參數估計效率卻也隨之降低。 高階動差的引入，除了降低近似偏誤外，卻也必須付出估計效率降低的代價。我們因此並不建議無限制地放入高階動差。則近似階次選取，乃為攸關估計績效的重要因素。本章的第二部份，即著眼於該最適近似階次選取。我們首先定義使參數估計均方誤(mean squared error, MSE)為最小的近似階次，為最適近似階次。我們發現，該最適階次與樣本大小、效用函數的彎曲程度都有直接的關係。 然而在實際進行估計時，由於參數真值無法得知，MSE準則自然無法作為階次選取之依據。我們於是利用目前在模型與階次選取上經常被使用的一些準則進行階次選取，並比較這些不同準則下參數估計的MSE。我們發現利用這些準則，確實可以使高階近似尤拉方程式得到MSE遠低於目前被普遍採用的二階近似的估計結果，而為估計消費結構參數時更佳的選擇。 本論文第二章延續前一章的模擬結果，嘗試利用消費高階動差間的非線性關係，進一步改善高階近似消費尤拉方程式的估計表現。由第一章的研究結果，我們發現高階近似估計確有助大幅降低近似誤差，但這其中可能產生的估計效率喪失，卻是輕乎不得的。這個效率喪失，很大一部份來自於我們所使用的工具變數，雖然可以有效掌握消費成長二階動差的變動，但是當這同一組工具變數被用來解釋如偏態與峰態等這些更高階動差時，預測力卻大幅滑落。這使待得當我們將這些配適度偏低的配適後高階動差，放到迴歸式中進行估計時，所能提供的額外情報也就相當有限。而所造成的共線性問題，也自然使得估計效率大幅惡化。 於是在其他合格的工具變數相對有限的情況下，我們利用高階動差間所存在的均衡關係，將原來的工具變數進行非線性轉換，以求得對高階動差的較佳配適。由於消費動差間之關係，尚未見諸相關文獻。於是我們首先透過數值分析，進一步釐清消費高階動差間之關係。這其中尤為重要的是由消費二階動差所衡量的消費風險，與更高階動差間之關係。因為這些關係將為我們轉換工具變數之依據。 我們發現與二階動差相一致地，消費者對這些高階動差之預期，都隨其財富水準的提高而減少。這隱涵消費風險與更高階動差間之正向關係。更進一步檢視消費風險與高階動差間之關係也發現，二者間確實存在非線性之正向關係。而這也解釋了何以前一章線性的工具變數，雖可適切捕捉消費風險，但對高階動差的解釋力卻異常薄弱。 利用這些非線性關係，我們將原始的工具變數進行非線性轉換後，用以配適更高階動差。透過模擬分析，我們證實了這些非線性工具變數，確實大幅改善高階近似尤拉方程式的估計表現。除了仍保有與線性工具變數般的一些特性，諸如隨樣本的增加，最適近似階次也隨之增加之外，相較於線性工具變數，非線性工具變數可以在較低的近似階次下，就使得估計偏誤大幅下降。在近似階次愈高估計效率愈低的情況下，這自然大幅度地提高了估計效率。比較兩種工具變數估計結構數參數所產生的MSE也證實，非線性工具變數確實有遠低於原始線性工具變數的MSE表現。 然而我們同時也發現，利用非線性工具變數估計，若未適當選擇近似階次，效率喪失的速度，可能更甚於線性工具變數時。這凸顯了選擇近似階次的重要性。於是我們同樣檢視了前述階次選擇準則在目前非線性工具變數環境下的適用性。而總結第一、二章的研究結果，我們凸顯了高階動差的重要性，確實助益重要消費結構參數估計。而利用過去尚未被討論過的高階動差間非線性關係，更可大幅度改善估計績效。 本論文的最後一章，則旨在理論上建立高階動差的重要性。我們在二次式的效用函數(quadratic utility function)設定下，推導借貸限制下的最適消費決策。二次式的效用函數，由於其邊際價值函數(marginal value function)為一線性函數，因此所隱涵的消費決策，具有確定相等(certainty equivalence)的特性。這表示消費者只關心未來的期望消費水準，二階以上的更高階動差，都不影響其消費決策。然而這種確定相等的特性，將因為借貸限制的存在而不復存在，而高階動差的重要性也就因此凸顯。 我們證明，確定相等特性的喪失，其背後的理論原因在於，借貸限制的存在，使得二次式效用函數的邊際價值函數，產生凸性。消費者因而因應未來的不確定性，進行預防性儲蓄。透過分析解的求得，我們也得以進一步分析更高階動差的對消費決策的理論性質。同時我們也引申理論推導的實證意涵，其中較重要者諸如未受限消費者因預防性儲蓄行為所引發的消費過度敏感性現象，實證上樣本分割法的選取，以及高階動差的引入模型。 / The theme of this thesis seeks to explore the importance of higher-order moments in the consumption Euler equation, both theoretically and empirically. Applying log-linearized versions of Euler equations has been a dominant approach to obtaining sensible analytical solutions, and a popular choice of model specifications for estimation. The literature however by now has been no lack of conflicting empirical results that are attributed to the use of the specific version of Euler equations. Important yet natural questions whether the higher-order moments can be safely ignored, or whether higher-order approximations offer explanations to the stylized facts remain unanswered. Such inquires as in the thesis thus can improve our understanding toward consumer behaviors over prior studies based on the linear approximation. 1. What Do We Gain from Estimating Euler Equations with Higher-Order Approximations? Despite the importance of estimating structural parameters governing consumption dynamics, such as the elasticity of intertemporal substitution, empirical attempts to unveil these parameters using a log-linearized version of the Euler equation have produced many puzzling results. Some studies show that the approximation bias may well constitute a compelling explanation. Even so, the approximation technique continues to be useful and convenient in estimation of the parameters, because noisy consumption data renders a full-fledged GMM estimation unreliable. Motivated by its potential success in reducing the bias, we investigate the economic significance and empirical relevance of higher-order approximations to the Euler equation with simulation methodology. The higher-order approximations suggest a linear relationship between expected consumption growth and its higher-order moments. Our simulation results clearly reveal that the approximation bias can be significantly reduced when the higher-order moments are introduced into estimation, but at the cost of efficiency loss. It therefore documents a clear tradeoff between approximation bias reduction and efficiency loss in the consumption growth regression when higher-order approximations to the Euler equation is considered. A question of immediate practical interest arises ``How many higher-order terms are needed?'' The second part of our Monte-Carlo studies then deals with this issue. We judge whether a particular consumption moment should be included in the regression by the criterion of mean squared errors (MSE) that accounts for a trade-off between estimation bias and efficiency loss. The included moments leading to smaller MSE are regarded as ones to be needed. We also investigate the usefulness of the model and/or moment selection criteria in providing guidance in selecting the approximation order. We find that improvements over the second-order approximated Euler equation can always be achieved simply by allowing for the higher-order moments in the consumption regression, with the approximation order selected by these criteria. 2. Uncovering Preference Parameters with the Utilization of Relations between Higher-Order Consumption Moments Our previous attempt to deliver more desirable estimation performance with higher-order approximations to the consumption Euler equation reveals that the approximation bias can be significantly reduced when the higher-order moments are introduced into estimation, but at the cost of efficiency loss. The latter results from the difficulty in identifying independent variation in the higher-order moments by sets of linear instruments used to identify that in variability in consumption growth, mainly consisting of individual-specific characteristics. Thus, one major challenge in the study is how to obtain quality instruments that are capable of doing so. With the numerical analysis technique, we first establish the nonlinear equilibrium relation between consumption risk and higher-order consumption moments. This nonlinear relation is then utilized to form quality instruments that can better capture variations in higher-order moments. A novelty of this chapter lies in adopting a set of nonlinear instruments that is to cope with this issue. They are very simple moment transformations of the characteristic-related instruments, thereby easy to obtain in practice. As expected, our simulations demonstrate that for a comparable amount of the bias corrected, applying the nonlinear instruments does entail an inclusion of fewer higher-order moments in estimation. A smaller simulated MSE that reveals the improvement over our previous estimation results can thus be achieved.\ 3. Precautionary Saving and Consumption with Borrowing Constraint This last chapter offers a theoretical underpinning for the importance of the higher-order moments in a simple environment where economic agents have a quadratic-utility preference. The resulting Euler equation gives rise to a linear policy function in essence, or a random-walk consumption rule. The twist in our theory comes from a presence of borrowing constraint facing consumers. The analysis shows that the presence of the constraint induces precautionary motives for saving as responses from consumers to income uncertainties, even there has been no such motives inherent in consumers' preference. The corresponding value function now displays a convexity property that is virtually only associated with more general preferences than a quadratic utility. The analytical framework allows us to be able to characterize saving behaviors that are of precautionary motives, and their responses to changes in different moments of income process. As empirical implications, our analysis shed new light on the causes of excess sensitivity, the consequences of sample splitting between the rich and the poor, as well as the relevance of the higher-order moments to consumption dynamics, specifically skewness and kurtosis.

Abstract In time series analysis, we often find the trend of dynamic data changing with time. Using the traditional model fitting can't get a good explanation for dynamic data. Therefore, many scholars developed various methods for model construction. The major drawback with most of the methods is that personal viewpoint and experience in model selection are usually influenced in them. Therefore, this paper presents a new approach on genetic-based modeling for the nonlinear time series. The research is based on the concepts of evolution theory as well as natural selection. In order to find a leading model from the nonlinear time series, we make use of the evolution rule: survival of the fittest. Through the process of genetic evolution, the AIC (Akaike information criteria) is used as the adjust function, and the membership function of the best-fitted models are calculated as performance index of chromosome. Empirical example shows that the genetic model can give an efficient explanation in analyzing Taiwan exchange rates, especially when the structure change occurs.

在製藥工業中，藥量特性之研究常被應用於藥理學、毒物學及臨床試驗中。藥量特性之研究同時在臨床試驗中的第一階段的藥物安全性及第二階段的藥物有效性中扮演著重要的角色。透過藥量特性的研究，使我們對於藥物的開發有著更深一層的認識，並可藉此縮短藥物核准上市的時間。 在多數的情況下，我們對於藥物動力學之參數與藥物劑量間的線性關係，有著特別的興趣。基於這個目的，一些典型的方法即是由一般線性模式發展而成的。然而，這些典型的統計方法常遇到下列的難題而違反其假設：（一）不同藥物劑量間的異質變異性。（二）不滿足常態性的假設。針對這些問題，我們藉由比較不同劑量間的斜率關係的無母數檢定程序來評估其線性關係並刻劃出藥物的反應曲線，文中並藉此方法舉出交叉實驗之實例。 / The problem of characterizing dose response curve for a pharmacokinetic parameter over a specific dose range is considered. In many cases, it is of interest to determine dose linearity (or dose proportionality) between the pharmacokinetic parameters and dose levels. For this purpose, several classical methods based on a general linear model procedure are available. However, two difficulties commonly encountered, namely (i) heterogeneity of the varibility at different dose levels and (ii) violation of the normality assumptions, often make the classical methods not applicable. To account for these problems, we propose a general nonparametric test procedure by comparing the slopes at different dose level to asses dose linearity and to characterize dose response curve. An example concerning the study of dose response of a compound based on a four-way crossover experiment is presented.

時間數列分析過程，常常發現其走勢，隨著時間過程而演變，應用傳統的線性模式來配適，往往很難獲得合適預測值。因此近幾年來，非線性時間數列結構性改變的研究越來越受到重視，也一直是時間數列及計量經濟學學者所熱衷的研究主題之一。本文利用模糊理論的觀念，以模糊炳找出配適ARCH模式數列之轉折區間，分別以轉折區問起始點及結束點作為截取點，去配適ARCH(1)模式，稱之為截取式自迴歸條件變異數分析法(Trimmed ARCH(1) model)。針對台幣對美元銀行間每日收盤匯率，分別以單變量ARIMA、ARCH(1)、Trtmmed ARCH(1)來建構模式，並做比較分析。比較結果發現，以轉折區間結束點作為截取點之Trimmed ARCH(1)模式，其預測值最為準確，大為改善了原來ARCH(1)模式之預測水準。 / In time series analysis, we often find the trend of which changing with time. Using the traditional model fitting can't get a good prediction. Hence the research of structure change of non-linear time series is attentive in recent years, and non-linear time series analysis is a research topic which the scholars of time series and econometrics are intent on. This article tries to use the theory of fuzzy ,to recognize the structure change period by the fuzzy classification, let the first point and the last point of the structure change period be the cute points, to fit ARCH(1) mod ie which we called the Trimmed ARCH(1) model. We use the data of the exchange rate between N.T dol liars and U.S dollars to compare the ARIMAwith ARCH(1) and Trimmed ARCH(1), the forcasting performance shows that Trimmed ARCH(1) model takes a better prediction result.

近幾年來，非線性時間數列轉折點的研究愈來愈受到重視，學者們也提出許多關於轉折點的偵測及檢定方法。若考慮實際資料走勢轉變的情形，“轉折區間”的概念更可以解釋結構改變的現象。但文獻中對於如何找尋時間數列結構改變之轉折區間的研究並不多。本文擬以時間數列統計模式及模糊學理論的角度來研究，並結合遺傳演算的規則而提出主導模式的概念，來架構出時間數列遺傳模式，再藉由轉折區間決策法則來找出數列的轉折區間。其中，我們以統計模式為遺傳演化過程中的染色體，而以候選模式之隸屬度函數為衡量染色體適應能力的指標。最後，我們舉出臺灣股價收盤指數之實例，分別以我們所提出的方法及其他方法找出數列的轉折區間及轉折點，並做比較。 / For recent years, the research of change point in nonlinear time series has been considered to be more and more important. Scholars have proposed a lot of detecting and testing methods about change points.If considering the trend of real situation, the concept of change period will show the phenomena of structure change.But there are not many researches about how to find change period in time series.My paper is based on the points of time series models and fuzzy theory.Besides,it combines the rules of genetic algorithm and provides the concepts of leading model to construct time seriep genetic model and to find out change period by decision rule.ln this paper, we use time series statistical models as chromosome in procedure of genetic evolution, and we also use membership function of selected models as pereformance: index of chromosome.Finally, the empirical application about change periods and change points detecting by our method and other's for Taiwan stock closing prices is demonstrated and make a comparision with these results.

近幾年來，非線性時間數列分析有快速的發展。其中的門檻自迴歸模式(SETAR)，以具有許多線性ARIMA模式所不能配適的特性而受到重視。但是，自1978年Tong建立SETAR模式以來，門檻參數估計的問題一直是SETAR模式在發展應用上的一個瓶頸。本文將探討以實數編碼遺傳演算法，結合統計學上的模式選取準則，建構SETAR模式門檻與延遲參數估計程序的可行性。並從這個基礎上，進一步地研究較精確的門檻參數估計法。 / Non-linear time series analysis has rapidly developed in recent years. Self-exciting threshold autoregression(SETAR) model of non-linear time series models is attentive, because it has some characters which linear ARIMA model fail to fit. But, It has not yet been applied widely because the question of estimation of threshold parameter limits its development and application since Tong proposed SETAR model in 1978. In this paper, we will study the feasibility which constructs a procedure of estimation of SETAR's threshod and delay parameters with real-coded genetic algorithm and statistical criterion of model selection, and develop a more precise estimation of threshold parameter in the basis.

近幾年來，非線性時間數列分析一直是時間數列及計量經濟學者所熱衷的研究主題之一，而非線性時間數列結構改變的研究也越來越受到重視。其中的門檻自迴歸模式，雖具有線性模式所不能配適的特性，但模式建構的問題，一直是其在發展應用上的瓶頸。本研究擬以門檻自迴歸模式建構的流程並結合遺傳演算法的最佳化搜尋技術，架構出時間數列遺傳演算法，藉此演算法則及程序，全域性地搜尋最佳的門檻自迴歸模式。 / Non-linear time series analysis is a research topic which the schalors of time series and econometrics are intent on, and the research of structure change of non-linear time series is attentive. Threshold autoregressive model (TAR model) of non-linear time series has some characters which linear model fail to fit while the problem of how to find an appropriate threshold value is still attracted many researchers attention. In this paper, we present about searching the parameters for a TAR model by genetic algorithms.

公共支出與公共收入乃為一體兩面之事。A.C. Pigou於1947年就曾提出租稅融通方式不同，將影響公共支出水準的看法，他並且認為當政府採用扭曲性租稅來融通公共財時，會產生超額負擔使得公共財提供的邊際成本增加。基於這項觀點，就不宜再以公共財的直接成本作為最適公共支出水準的依據，而應將扭曲性租稅本身可能造成的扭曲效果納入考慮。而這也表示了P.A. Samuelson於1954年提出的最適公共財提供條件ΣMRS=MRT應該被修正，且視融通公共財的租稅方式所產生的影響而應有不同的最適公共財提供條件。 　　本篇論文的目的，即在討論各種不同的租稅制度融通下，以最適公共財提供條件的修正，分析修正項目對最適公共支出水準的影響，並與過去的文獻作一比較，以期得出不同的租稅制度融通和最適公共支出之間的關係。

本文採用傅利葉轉換與小波轉換以探討非線性現象：長期相依的碎形結構與混沌現象。藉由傅利葉轉換與小波轉換兩種研究方法，所得到臺灣股票市場加權股價指數的實證結論如下：1.藉由傅利葉轉換所得到的H值為0.4632；藉由小波轉換所得到的H值為0.4750。這兩種研究方法皆顯示臺灣股票市場具有負的長期相依的碎形結構。2.藉由傅利葉轉換的研究方法，臺灣股票市場加權股價指數的頻譜由初始向下與寬的連續的頻帶所組成；臺灣股票市場加權股價指數的自我相關函數則隨著時間差距的增加而遞減。此顯示臺灣股票市場具有混沌現象。3.小波轉換可以檢測出臺灣股票市場加權股價指數的奇異之處，並且指出存有一能說明臺灣股票市場碎形結構的複雜性的機制。藉由以上的實證結論，可以得知臺灣股票市場具有反持續性的碎形結構，股票價格的變動來自於臺灣股票市場尺度上的自我相似性。即使如此，由於混沌不可預測性的本質，使得股票價格的預測似乎是不可能的。 / The Fourier transform and the wavelet transform are utilized in this research to explore the nonlinear phenomena: the fractal structure of long trem dependence and the phenomenon of chaos. 　　In terms of the two research methods of the Fourier transform and the wavelet transform, the empirical conclusions of the Taiwan stock exchange weighted stock index are derived as follows: 　　1. The $H$ value of the research method of the Fourier transform is 0.4632; the $H$ value of the research method of the wavelet transform is 0.4750. The two research methods show that the Taiwan stock market has a fractal structure of negative long term dependence. 　　2. In terms of the research method of the Fourier transform, the power spectrum of the Taiwan stock exchange weighted stock index consists of initially downward and wide continuous band of frequencies; the autocorrelation function of the Taiwan stock exchange weighted stock index decreases as the time lag increases. These observations show that there exists the phenomenon of chaos in the Taiwan stock market. 　　3. The wavelet transform can detect out the singularities of the Taiwan stock exchange weighted stock index and can point out the heirarchy that illustrates the complexity of the fractal sturcture in the Taiwan stock market. 　　By the above empirical conclusions, there exists the antipersistent fractal structure in the Taiwan stock market. The variations of stock prices result from the self-similarity of the scales of the Taiwan stock market. Even so, the prediction of stock prices seems very impossible as a result of the unpredictability of chaotic nature.