《中国外汇储备的影响因素分析(共15页).doc》由会员分享,可在线阅读,更多相关《中国外汇储备的影响因素分析(共15页).doc(15页珍藏版)》请在taowenge.com淘文阁网|工程机械CAD图纸|机械工程制图|CAD装配图下载|SolidWorks_CaTia_CAD_UG_PROE_设计图分享下载上搜索。
1、精选优质文档-倾情为你奉上中国外汇储备的影响因素分析 金融双语 张翔 张程一、分析与提出问题选题意义:最近一些国家以我国外汇储备增长迅速为由向人民币汇率施压。那些因数对我国外汇储备增长有实际贡献,有助于制定合理的外汇储备和人民币汇率政策。本报告利用中国自1984年年度国际收支数据,通过建立外汇储备的影响因数模型,适当的扣除出口和外商直接投资对国际收支余额增长的重复贡献,并考虑到国际收支规模的影响以及政治的因数,对进出口和外商直接投资对我国外汇储备增长的实际贡献进行测算。主要内容:要求人民币升值观点的依据之一是中国快速增长的外汇储备。截止到2003 年6月份中国外汇储备达到了历史上的高点3465
2、亿美元比去年猛增了601亿美元。于是一些人很自然地把中国的巨额外汇储备与中国对外出口联系起来。认为外汇储备的激增来自长期的贸易顺差,而长期的贸易顺差是由于人民币人为低估造成的。与对经常项目的关注相反,资本流入对外汇储备的贡献没有得到足够的注意。探讨外汇储备增长的主要原因对考量人民币汇率升值压力有重大意义。如果外汇储备增长主要来源于我国出口创汇,则说明我国经济的基本面有很大的发展,带动人民币成为强势货币,人民币确实存在着较大的升值压力。但是如果我国外汇储备的激增很大程度上是由于外来资本的流入造成的,在考察人民币升值压力时就应该将这部分因素分开考虑。我们知道,外来资本有着很大的不确定性。当本国经济
3、发展良好时,大量的外来资本流入,导致外汇储备增加。当本国经济恶化时,则会出现大量外来资本流出,出现外汇储备流失,可能会影响正常的经济秩序,给本国经济发展带来很大困难。因此,由贸易顺差所导致的储备增加与资本流入所导致的储备增加对汇率水平的压力是不同的。在经常项目中,我国的服务贸易比例很小,仅占经常项目总额的八分之一左右,而且长期处于逆差。因此在经常项目中,我们主要考察出口对储备增长的贡献。 对于资本项目,按照Paul R. Krugman的分类“外来资本的流入主要有五种方式:外商直接投资、证券投资、债券融资、银行贷款、官方借贷。”自90年代以来,外商直接投资成为我国利用外资的主要形式。1993年
4、,外商直接投资占全部利用外资的比重为69,1994年为78.2,1995年为78%,1996年为76.1%,1997年由于受到东南亚金融危机等的影响,该比例有所下降,为70.3%。1998年又上升到77.6%。截止1998年底,全国累计批准设立外商投资企业324,620家,合同外资金额5724.95亿美元 实际使用外资金额2671.09亿美元(对外经济贸易合作部1999),外商直接投资成为我国权益性资本流入的主要形式。二、收集数据通过查阅中国统计年鉴,上网(中国统计局)收集数据X1:进口 X2:出口 X3:国外直接投资 Y:外汇储备 ZZ:虚拟变量 单位:亿元obsX1X2X3YZZ1985
5、273.5000 422.5000 19.60000 26.40000 0.1986 309.4000 429.0000 22.40000 20.70000 0.1987 394.4000 432.2000 23.10000 29.20000 0.1988 475.2000 552.7000 31.90000 33.70000 0.1989 525.4000 591.4000 33.90000 55.50000 0.1990 620.9000 533.5000 34.90000 110.9000 0.1991 719.1000 637.9000 43.70000 217.1000 0.1992
6、 849.4000 805.9000 110.1000 194.4000 0.1993 917.4000 1039.600 275.2000 212.0000 0.1994 1210.100 1156.200 337.7000 516.2000 0.1995 1487.800 1320.800 375.2000 736.0000 1.1996 1510.500 1388.300 417.3000 1050.300 1.1997 1827.900 1423.700 452.6000 1398.900 1.1998 1837.100 1402.400 454.6000 1449.600 1.199
7、9 1949.300 1657.000 403.2000 1546.800 1.2000 2492.000 2250.900 407.2000 1655.700 1.2001 2661.000 2435.500 468.2000 2121.600 1.2002 3255.700 2952.000 527.4000 2864.100 1.2003 4382.300 4127.600 535.0467 4032.510 1.2004 5933.600 5613.800 620.0000 6099.000 1.三、分析数据初步回归Dependent Variable: YMethod: Least
8、SquaresDate: 04/28/05 Time: 23:47Sample: 1985 2004Included observations: 20VariableCoefficientStd. Errort-StatisticProb. C-264.2837143.4186-1.0.0852X11.0.3.0.0016X2-0.0.-1.0.1824X30.0.0.0.9646R-squared0. Mean dependent var1218.531Adjusted R-squared0. S.D. dependent var1584.775S.E. of regression168.8
9、867 Akaike info criterion13.30865Sum squared resid.5 Schwarz criterion13.55758Log likelihood-128.0865 F-statistic414.5020Durbin-Watson stat2. Prob(F-statistic)0.图示从线性图示中我们看出变量间的趋同性很强,并且具有有钱的联系与相关性。同时为了让数据更加平稳,我们假设其模型是非线性,所以取对数对其进行分析与初步的研究。期望效果更好。obsLX1LX2LX3LY1985 5. 6. 2. 3.1986 5. 6. 3. 3.1987 5.
10、6. 3. 3.1988 6. 6. 3. 3.1989 6. 6. 3. 4.1990 6. 6. 3. 4.1991 6. 6. 3. 5.1992 6. 6. 4. 5.1993 6. 6. 5. 5.1994 7. 7. 5. 6.1995 7. 7. 5. 6.1996 7. 7. 6. 6.1997 7. 7. 6. 7.1998 7. 7. 6. 7.1999 7. 7. 5. 7.2000 7. 7. 6. 7.2001 7. 7. 6. 7.2002 8. 7. 6. 7.2003 8. 8. 6. 8.2004 8. 8. 6. 8.四、统计检验 (1)单位根检验A对LYA
11、DF Test Statistic-4. 1% Critical Value*-4.5743 5% Critical Value-3.6920 10% Critical Value-3.2856*MacKinnon critical values for rejection of hypothesis of a unit root.Augmented Dickey-Fuller Test EquationDependent Variable: D(DLY)Method: Least SquaresDate: 06/15/05 Time: 23:28Sample(adjusted): 1987
12、2004Included observations: 18 after adjusting endpointsVariableCoefficientStd. Errort-StatisticProb. DLY(-1)-0.0.-4.0.0011C0.0.2.0.0250TREND(1985)-0.0.-0.0.4349R-squared0. Mean dependent var0.Adjusted R-squared0. S.D. dependent var0.S.E. of regression0. Akaike info criterion0.Sum squared resid1. Sch
13、warz criterion0.Log likelihood-0. F-statistic8.Durbin-Watson stat1. Prob(F-statistic)0.从检验的结果看其满足一阶单整B对LX1ADF Test Statistic-3. 1% Critical Value*-4.5743 5% Critical Value-3.6920 10% Critical Value-3.2856*MacKinnon critical values for rejection of hypothesis of a unit root.Augmented Dickey-Fuller Te
14、st EquationDependent Variable: D(DLX1)Method: Least SquaresDate: 06/15/05 Time: 23:20Sample(adjusted): 1987 2004Included observations: 18 after adjusting endpointsVariableCoefficientStd. Errort-StatisticProb. DLX1(-1)-1.0.-3.0.0020C0.0.2.0.0475TREND(1985)0.0.0.0.6143R-squared0. Mean dependent var0.A
15、djusted R-squared0. S.D. dependent var0.S.E. of regression0. Akaike info criterion-1.Sum squared resid0. Schwarz criterion-1.Log likelihood17.88252 F-statistic7.Durbin-Watson stat1. Prob(F-statistic)0.从检验结果来看其也符合一阶单整C对LX2ADF Test Statistic-3. 1% Critical Value*-4.6193 5% Critical Value-3.7119 10% Cr
16、itical Value-3.2964*MacKinnon critical values for rejection of hypothesis of a unit root.Augmented Dickey-Fuller Test EquationDependent Variable: D(DLX2)Method: Least SquaresDate: 06/15/05 Time: 23:21Sample(adjusted): 1988 2004Included observations: 17 after adjusting endpointsVariableCoefficientStd
17、. Errort-StatisticProb. DLX2(-1)-1.0.-3.0.0044D(DLX2(-1)0.0.1.0.1532C0.0.1.0.3239TREND(1985)0.0.1.0.1799R-squared0. Mean dependent var0.Adjusted R-squared0. S.D. dependent var0.S.E. of regression0. Akaike info criterion-1.Sum squared resid0. Schwarz criterion-1.Log likelihood14.50282 F-statistic4.Du
18、rbin-Watson stat1. Prob(F-statistic)0.从检验结果来看其符合一阶单整DLX3ADF Test Statistic-2. 1% Critical Value*-3.8877 5% Critical Value-3.0521 10% Critical Value-2.6672*MacKinnon critical values for rejection of hypothesis of a unit root.Augmented Dickey-Fuller Test EquationDependent Variable: D(DLX3)Method: Leas
19、t SquaresDate: 06/15/05 Time: 23:23Sample(adjusted): 1988 2004Included observations: 17 after adjusting endpointsVariableCoefficientStd. Errort-StatisticProb. DLX3(-1)-0.0.-2.0.0121D(DLX3(-1)0.0.1.0.1033C0.0.1.0.0850R-squared0. Mean dependent var0.Adjusted R-squared0. S.D. dependent var0.S.E. of reg
20、ression0. Akaike info criterion0.Sum squared resid0. Schwarz criterion0.Log likelihood1. F-statistic4.Durbin-Watson stat1. Prob(F-statistic)0.从检验结果来看其也符合一阶单整(2)GRANGER因果关系检验Pairwise Granger Causality TestsDate: 06/15/05 Time: 23:24Sample: 1985 2004Lags: 2 Null Hypothesis:ObsF-StatisticProbability X1
21、 does not Granger Cause Y18 1.40307 0.28071 Y does not Granger Cause X1 4.21887 0.03872 X2 does not Granger Cause Y18 9.27388 0.00314 Y does not Granger Cause X2 2.18682 0.15182 X3 does not Granger Cause Y18 5.83756 0.01552 Y does not Granger Cause X3 1.02952 0.38455 X2 does not Granger Cause X118 4
22、.83349 0.02695 X1 does not Granger Cause X2 0.07556 0.92763 X3 does not Granger Cause X118 2.85820 0.09358 X1 does not Granger Cause X3 2.16285 0.15458 X3 does not Granger Cause X218 2.14288 0.15691 X2 does not Granger Cause X3 2.99361 0.08524从检验的结果看其服从单项因果关系,为Y分别与X1、X2、X3有因果关系,且X1、X2、X3相互之间也存在因果关系。
23、(3)协整检验AADF Test Statistic-3. 1% Critical Value*-4.5743 5% Critical Value-3.6920 10% Critical Value-3.2856*MacKinnon critical values for rejection of hypothesis of a unit root.Augmented Dickey-Fuller Test EquationDependent Variable: D(EX)Method: Least SquaresDate: 06/15/05 Time: 23:41Sample(adjusted
24、): 1987 2004Included observations: 18 after adjusting endpointsVariableCoefficientStd. Errort-StatisticProb. EX(-1)-1.0.-3.0.0017D(EX(-1)0.0.1.0.1439C-0.0.-1.0.1804TREND(1985)0.0.1.0.2037R-squared0. Mean dependent var-0.Adjusted R-squared0. S.D. dependent var0.S.E. of regression0. Akaike info criter
25、ion-0.Sum squared resid0. Schwarz criterion-0.Log likelihood6. F-statistic5.Durbin-Watson stat2. Prob(F-statistic)0.LY = -5. + 2.*LX1 - 1.*LX2 + 0.*LX3我们主要是从对数的回归结果中发现每个变量都服从一阶单整,且把其回归的参差项进行单位根检验发现其满足零阶单整,从而我们也就得到了上面的表达式。BDate: 06/16/05 Time: 01:12Sample: 1985 2004Included observations: 19Test assum
26、ption: Quadratic deterministic trend in the dataSeries: Y X1 X2 X3 Lags interval: No lagsLikelihood5 Percent1 PercentHypothesizedEigenvalueRatioCritical ValueCritical ValueNo. of CE(s) 0. 99.74420 54.64 61.24 None * 0. 50.25664 34.55 40.49 At most 1 * 0. 22.75529 18.17 23.46 At most 2 * 0. 7. 3.74 6
27、.40 At most 3 * *(*) denotes rejection of the hypothesis at 5%(1%) significance level L.R. test indicates 4 cointegrating equation(s) at 5% significance level Unnormalized Cointegrating Coefficients:YX1X2X3-9.35E-05-2.69E-05 0.-9.47E-05-0. 0.-0. 0.-0.-0. 0. 0. 0.-0.-0.-0. Normalized Cointegrating Co
28、efficients: 1 Cointegrating Equation(s)YX1X2X3TREND(86)C 1. 0.-7. 1. 848.2637-667.3535 (3.68934) (11.0582) (3.11995) Log likelihood-447.3716 Normalized Cointegrating Coefficients: 2 Cointegrating Equation(s)YX1X2X3TREND(86)C 1. 0.-6. 0. 773.2782-590.8337 (3.74283) (2.44391) 0. 1.-2. 0. 260.5108-265.
29、8411 (1.46509) (0.95665) Log likelihood-433.6209 Normalized Cointegrating Coefficients: 3 Cointegrating Equation(s)YX1X2X3TREND(86)C 1. 0. 0.-0.-166.1277 794.5952 (0.90009) 0. 1. 0.-0.-179.7503 383.4527 (0.53511) 0. 0. 1.-0.-151.5773 223.5450 (0.45527) Log likelihood-426.0365这是黎老师教给的方法,知道其最多有三个协整关系(
30、4)回归结果Dependent Variable: LYMethod: Least SquaresDate: 06/17/05 Time: 00:05Sample: 1985 2004Included observations: 20VariableCoefficientStd. Errort-StatisticProb. C-5.0.-5.0.0000LX12.0.5.0.0001LX2-1.0.-2.0.0316LX30.0.2.0.0289ZZ0.0.2.0.0105R-squared0. Mean dependent var5.Adjusted R-squared0. S.D. dep
31、endent var1.S.E. of regression0. Akaike info criterion0.Sum squared resid0. Schwarz criterion0.Log likelihood2. F-statistic256.3780Durbin-Watson stat1. Prob(F-statistic)0.这里我们引入了虚拟变量希望能解释一定的现象(回归结果良好)(5)异方差检验ARCH Test:F-statistic0. Probability0.Obs*R-squared0. Probability0.Test Equation:Dependent Va
32、riable: RESID2Method: Least SquaresDate: 06/17/05 Time: 00:06Sample(adjusted): 1987 2004Included observations: 18 after adjusting endpointsVariableCoefficientStd. Errort-StatisticProb. C0.0.1.0.1386RESID2(-1)0.0.0.0.9583RESID2(-2)0.0.0.0.9339R-squared0. Mean dependent var0.Adjusted R-squared-0. S.D.
33、 dependent var0.S.E. of regression0. Akaike info criterion-2.Sum squared resid0. Schwarz criterion-2.Log likelihood23.20457 F-statistic0.Durbin-Watson stat1. Prob(F-statistic)0.White Heteroskedasticity Test:F-statistic1. Probability0.Obs*R-squared8. Probability0.Test Equation:Dependent Variable: RES
34、ID2Method: Least SquaresDate: 07/01/05 Time: 00:07Sample: 1985 2004Included observations: 20VariableCoefficientStd. Errort-StatisticProb. C-3.5.-0.0.4657LX1-0.2.-0.0.6846LX120.0.0.0.7045LX22.2.0.0.4281LX22-0.0.-0.0.4635LX3-0.0.-0.0.7231LX320.0.0.0.8244ZZ-0.0.-0.0.7893R-squared0. Mean dependent var0.
35、Adjusted R-squared0. S.D. dependent var0.S.E. of regression0. Akaike info criterion-2.Sum squared resid0. Schwarz criterion-1.Log likelihood28.05046 F-statistic1.Durbin-Watson stat2. Prob(F-statistic)0.没有异方差;同时自相关也没有四、其它的结果其实我们开始的很早,所以向平稳性检验也是后面加进去的。前期的工作中我们考虑最多也最难为我们的是关于模型的建立与多重共线性的克服,其中的一些“成就”望与大家
36、分享最早的模型是四个变量X1出口,X2进口,X3国外直接投资,X4外债。其中我们发现进出口对于外汇储备的影响十分明显,外债也不错,国外投资稍稍差点,问题是我们把它们放在一起时其回归结果相当糟糕。当时我们考虑可能进出口与国外投资之间有着联系(数据的问题)。所以我们的目的是消除国外直接投资的影响,后在建立与外债相联系的联立方程对其经济意义进行解释。(A)W检验Wald Test:Equation: UntitledNull Hypothesis:C(2)=1.08*C(3)C(4)=0.12*C(3)F-statistic2.Probability0.Chi-square5.Probability
37、0.从图中我们应该可以看出其并没有通过检验,所以我们没有再往下面作步骤:分别对X1、X3和X2、X3进行回归,把X2看作中间变量对其进行约束,如C(2)=1.08*C(3) C(4)=0.12*C(3) 。这样就能使X3包含于新的变量中,把X3“消除”(B)除法这也是一种消除变量的方法,不同的是它在Y=(参数)X1+等式的右边除以你要消除的变量Last updated: 05/30/05 - 20:26Modified: 1985 2004 / x11=x1/x3Last updated: 05/30/05 - 20:26Modified: 1985 2004 / x22=x2/x3Last
38、updated: 05/30/05 - 20:33Modified: 1985 2004 / x7=0.76*x11+x22Dependent Variable: YMethod: Least SquaresDate: 06/14/05 Time: 23:42Sample: 1985 2004Included observations: 20VariableCoefficientStd. Errort-StatisticProb. C-3257.906416.2707-7.0.0000X774.6712613.085355.0.0000X43.0.14.334260.0000R-squared
39、0. Mean dependent var1218.531Adjusted R-squared0. S.D. dependent var1584.775S.E. of regression432.2942 Akaike info criterion15.11357Sum squared resid. Schwarz criterion15.26293Log likelihood-148.1357 F-statistic119.1732Durbin-Watson stat0. Prob(F-statistic)0.如图示定义X11=X1/X3(X1出口,X3国外直接投资),来生成新的数据在对其进行回归,但是最后还是经济意义的解释上出问题。(C)其它思考现在衡量国际收支各项目对外汇储备增长的贡献最直观的做法是对国际收支会计平衡方程直接进行计算。但由于没有扣除各项目之间对外汇储备的重复贡献,不仅高估了各项目的贡献率,而且掩盖了各项目之间实际贡献的相对比率。所以从外汇储备的会计平衡方程出发,考虑影响外汇储备诸因素之间的相互关系,构造国民收入