【摘要】：美元自量化寬松后走勢模糊,一直在多空之間震蕩；我國的外匯儲備逐年上升,對外貿(mào)易、對外投資與吸引外資投資量處于上升周期,經(jīng)濟發(fā)展和美元強弱聯(lián)系密切。因此,研究美元指數(shù)的預測方法對于國家、金融機構和個人都具有重要的現(xiàn)實意義。 本文主要通過支持向量機模型對美元指數(shù)進行預測研究。本文以匯率決定理論為基礎,篩選出影響美元指數(shù)的合適指標,建立指標體系。本文分別用三種不同的變量降維方法對數(shù)據(jù)進行預處理,并選擇出最好的變量降維方法；再用三種不同的參數(shù)優(yōu)化方法對模型進行優(yōu)化,并選擇出最好的優(yōu)化方法,將兩種方法結合,構建出一個優(yōu)化后的支持向量機模型,完成對短期內(nèi)美元指數(shù)的預測。 在實證部分,本文分別使用粒子群優(yōu)化算法(簡稱PSO)、遺傳算法(簡稱GA)和網(wǎng)格遍歷法(簡稱GRID)進行了參數(shù)優(yōu)化,并比較了預測結果,發(fā)現(xiàn)遺傳算法的優(yōu)化效果最佳,平均誤差率只有0.393556%,最低誤差率達到了0.004165%,均方差也只有0.21761；之后分別用因子分析(FA)、粗糙集方法(RS)和主成分分析(PCA)對變量進行降維,得出在該特定問題上因子分析的變量降維效果最好的結論。最后,為了證明本文所建立的FA-GA支持向量機模型的有效性,本文還用BP神經(jīng)網(wǎng)絡模型和未進行變量降維的普通支持向量機模型對相同的數(shù)據(jù)進行了預測,預測結果發(fā)現(xiàn)還是優(yōu)化后的支持向量機的預測結果最好,這一方面證明了本文建立指標的可行性和模型的優(yōu)越性,另一方面證明了對數(shù)據(jù)進行變量降維能大大提高支持向量機的運算速度和準確率。
[Abstract]:After quantitative easing, the trend of the US dollar is vague and has been fluctuating between many short periods. China's foreign exchange reserves are rising year by year, foreign trade, foreign investment and attracting foreign investment are in an upward cycle, and economic development is closely related to the strength of the US dollar. Therefore, it is of great practical significance for countries, financial institutions and individuals to study the prediction method of dollar index. In this paper, the support vector machine model is used to predict the dollar index. Based on the theory of exchange rate determination, this paper selects the appropriate indexes that affect the dollar index and establishes the index system. In this paper, three different variable dimension reduction methods are used to preprocess the data, and the best variable dimension reduction method is selected. Then three different parameter optimization methods are used to optimize the model, and the best optimization method is selected. Combining the two methods, an optimized support vector machine model is constructed to predict the dollar index in the short term. In the empirical part, the particle swarm optimization algorithm (PSO), genetic algorithm) and grid ergodicity method (GRID) are used to optimize the parameters, and the prediction results are compared, and it is found that the genetic algorithm has the best optimization effect. The average error rate is only 0.393556%, the minimum error rate is 0.004165%, and the mean variance is only 0.21761. Then the factor analysis (FA), rough set method (RS) and principal component analysis (PCA) are used to reduce the dimension of variables respectively, and it is concluded that the variable dimension reduction effect of factor analysis is the best on this particular problem. Finally, in order to prove the effectiveness of the FA-GA support vector machine model established in this paper, the BP neural network model and the general support vector machine model without variable dimension reduction are also used to predict the same data. The prediction results show that the optimized support vector machine has the best prediction results, which proves the feasibility of establishing the index and the superiority of the model in this paper. On the other hand, it is proved that variable dimension reduction can greatly improve the operation speed and accuracy of support vector machine.
【學位授予單位】：浙江大學
【學位級別】：碩士
【學位授予年份】：2014
【分類號】：F827.12;F832.6

【參考文獻】

相關期刊論文 前10條

1 喻勝華;肖雨峰;;基于信息粒化和支持向量機的股票價格預測[J];財經(jīng)理論與實踐;2011年06期

2 彭民;孫彥彬;李鳳升;姚麗霞;李臣;;美元指數(shù)對國際原油價格影響的實證分析[J];大慶石油學院學報;2010年06期

3 彭望蜀;;基于BP神經(jīng)網(wǎng)絡與支持向量機的股票指數(shù)預測模型比較[J];南方金融;2013年01期

4 劉q，

本文編號：2477570