北大老师概率论复习课件.ppt

北京大学中国经济研究中心 沈艳 * Covariance measures the linear dependence between two random variables. 北京大学中国经济研究中心 沈艳 * Two variables can be uncorrelated but they are related nonlinearly. 北京大学中国经济研究中心 沈艳 * Covariance and Correlation Property For any constants a1 ,b1 , a2 , b2 , cov(a1X+ b1,a2Y+ b2)=a1a2cov(X,Y). Implication: covariance between X and Y can be altered by multiplying one or both by a constant. Important as we often need to change the units of variables, like inflation rates, amount of money, etc. Cauchy –Schwarz inequality: The bounds of covariance: |cov(X,Y)|≤sd(X)sd(Y) 北京大学中国经济研究中心 沈艳 * Correlation Coefficient The second property of covariance indicates that its value is related to the units of measurement. However, how strong two variables are related should not depend on how they are measured. Correlation coefficient solves this problem: corr(X,Y)=cov(X,Y)/{sd(X)sd(Y)}. The sign of corr(X,Y) is determined by that of cov(X,Y). corr(X,Y)=0 iff cov(X,Y)=0. Also, corr(X,Y)=0 does not necessarily imply X and Y are independent. 北京大学中国经济研究中心 沈艳 * Correlation Coefficient Property of correlation coefficient -1≤ corr(X,Y) ≤1. Correlation of X and Y are invariant to the units of measurement, that is, for any constants a1 ,b1 , a2 , b2 , corr(a1X+ b1,a2Y+ b2)=corr(X,Y) if a1 a20 =–corr(X,Y) if a1 a20. 北京大学中国经济研究中心 沈艳 * Sums of Random Variables Property V(aX+bY)=a2V(X)+b2V(Y)+2abcov(X,Y). V(X+Y)=V(X)+V(Y) if cov(X,Y)=0, V(X-Y)=V(X)+V(Y) if cov(X,Y)=0. If {X1 , X2 ,…Xn} are pairwise uncorrelated, and ai are constants, then V(a1X1+…+anXn )=a12 V(X1)+…+ an2 V(Xn). 北京大学中国经济研究中心 沈艳 * Conditional Expectation Covariance and correlation treats X and Y symmetrically. We often need to explain one variable in terms of other variable(s). We wish to summarize the information of the conditional distribution. One way to summarize is to calculate the conditional expectation of Y given X. We wish to calculate E(Y|X=x), which should be a function of X. 北京大学中国经济研究中心 沈艳 * Trac