lding period for returns is greater than the sampling frequency of the data.
⇒Overlapping data problem(see VerbeekChp4.11.3 for an illustration of this problem in the FX market and see below).Warwick Business School 7
Overlapping data problem
Suppose we have a sample of daily datafor returns but our model is for m-period holding returns(m > one day)Even if the one-period returns are independent the m-period returns for different periods consist of ‘overlapping’one-period returns ⇒the m-period returns are correlated: Any static model involving the m period returns will therefore have an autocorrelated[MA(m-1)] error term ⇒OLS inferences are invalid.Use of NeweyWest standard errors is an appropriate remedy in this case.()()()mtttmtmtttttmttmtrrrppppppppr−+−−−+−−−−+++=−++−+−=−=111211......()()()()()()()()0,covvar,cov...var1,covvarvar11==−==−−+−mmtmttmmtmttmtmttmtrrrrrrmrrrmrThe autocorrelations decay to zero after m-1 lags.This is indicative of an MA(m-1) process.... ... ...1111111++=+++=+++=−−+−+−−+−−−+−mtmtmmtmtmttmtmtttmtrrrrrrrrrrrKlog pricesWarwick Business School 8
Method of Moments Estimation (MME)
OLS as a MMEThe CLRM requires the following populationmoment conditionsThe MME finds by solving the sample moment conditionsThere are k sample moment conditions and kunknown parameters ⇒possible to find a unique solution for . εβ+=Xy()0=′εXE()0ˆ1ˆ1=−′=′βεXyXTXTβˆ
This is a k×1 vector.⇒There are k moment conditions which the OLS estimator must satisfy.These moment conditions imply the values of Xare determined outside of the model:X is exogenousβˆ
()()error term theoft independenlinearly is X00 :A2=′⇒=εεXEXEWarwick Business School 9
OLS as a MME()()yXXXXXyXXyXT′′=⇒′=′⇒=−′−1ˆˆ0ˆ1βββThe MM estimator for the CLRM is identicalto the OLS estimator.Warwick Business School 10
Properties of MME
MME is a general approach to estimation which imposes population moment conditions (required by the statistical model) to hold exactly in the sample.
These moment conditions are then solved for the unknown parameters in the model (example above).
MME has 3 attractive features:
1.It makes no distributional assumptions.
2.It is a consistent estimator.
3.It is a verygeneral technique (e.g., applicable to non-linear models).Warwick Business School 11
Endogenous regressors
In many instances in economics/finance there is a two way or simultaneous relationship between X and y.⇒both X and y are determined insidethe model.⇒X is endogenous.Endogeneityis common due to the non-experimentalnature of economic/finance data ⇒In that case OLS/MME estimation (assuming ) is invalid. The estimator is biased (and inconsistent)⇒()0=′εXE()()()()εβεββXXXXXXXyXXX′′+=+′′=′′=−−−111ˆ()()()().0 unless ˆ1=′≠′′+=−εβεββXEXEXXE()0≠′εXEWarwick Business School 12
Examples of endogeneityin finance: testing CIP and UIP (Cuthberston& NitzscheChps24.3/4, 25.1/2)
Covered Interest Parity
Where Fh(h-period forward exchange rate), S(spot exchange rate) are denominated in terms of the domestic currency price of a unit of foreign exchange.
r(domestic interest rate), r*(foreign interest rate) (interest rate on h-period T-Bills).*1
本论文由英语论文网提供整理,提供论文代写,英语论文代写,代写论文,代写英语论文,代写留学生论文,代写英文论文,留学生论文代写相关核心关键词搜索。