12.07.2015 Views

The Limits of Mathematics and NP Estimation in ... - Chichilnisky

The Limits of Mathematics and NP Estimation in ... - Chichilnisky

The Limits of Mathematics and NP Estimation in ... - Chichilnisky

SHOW MORE
SHOW LESS
  • No tags were found...

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

28Advances <strong>in</strong> Econometrics - <strong>The</strong>ory <strong>and</strong> Applications<strong>of</strong> near-unit root bias <strong>and</strong> result<strong>in</strong>g the lack <strong>of</strong> distribution, the empirical studies generallyapply Andrews’ (1993) median unbiasedness method to estimate the AR(1) coefficient to<strong>in</strong>vestigate the persistence behavior.For AR(1) case, this paper contributes to the literature by apply<strong>in</strong>g the IGF approach <strong>of</strong>Phillips et al. (2004) to estimate the coefficients <strong>of</strong> near unit root process, IGF estimator isproved to be normal asymptotically, hence it is very easy to construct confidence <strong>in</strong>tervals.For AR(p) case, moreover, <strong>in</strong>stead <strong>of</strong> recalculation, we propose a unrestricted FM-AR(p)model, a slight extension <strong>of</strong> Phillips’ (1995) FM-VAR, to estimate coefficients directly.Our empirical illustration <strong>of</strong> real effective exchange rate <strong>in</strong>dicates that FM-AR(p) is a useful<strong>and</strong> easy-to-use method to exam<strong>in</strong>e the econometric persistence.5. Appendix: <strong>The</strong> f<strong>in</strong>ite sample properties <strong>of</strong> FM-ARThis paper shows that a FM estimator can ameliorate the small sample biases that arise fromnear unit root bias. Our attention here is focused on the class <strong>of</strong> dynamic AR(p). A MonteCarlo simulation is used here to <strong>in</strong>vestigate the performance, our results <strong>in</strong>dicate that FMestimator successfully reduces the small-sample bias.Assum<strong>in</strong>g { yt} t 0 is governed by a AR(3) time series process generated from (, 2 ) = (0, 1),which satisfies the data-generat<strong>in</strong>g process specified below:y y y y (1)t 1 t1 2 t2 3 t3twhere t i.i.d.. <strong>and</strong> ’s represent the associated parameters, <strong>and</strong> is the st<strong>and</strong>arddeviation. In our simulation, we generate an AR(3) process. <strong>The</strong> summation <strong>of</strong> the threecoefficients measures the degree <strong>of</strong> persistence, the vector is parameterized belowDegree <strong>of</strong> Persistence 1: { 1 , 2, 3 } = {0.90, 0.085, 0.015},Degree <strong>of</strong> Persistence 2: { 1 , 2, 3 } = {0.90, 0.050, 0.015}Degree <strong>of</strong> Persistence 3: { 1 , 2, 3 }= {0.85, 0.050, 0.015}<strong>and</strong>T{ 200, 400, 800, 1500, 3000}where T represents the vector <strong>of</strong> sampl<strong>in</strong>g size that are used <strong>in</strong> practice. <strong>The</strong> DGPs arecharacterized by a modest change <strong>in</strong> the <strong>in</strong>novation variance but allow for drastic changes<strong>in</strong> others.Table A1 reports the characteristics <strong>of</strong> the f<strong>in</strong>ite-sample distribution <strong>of</strong> both estimators <strong>of</strong> theelements <strong>of</strong> estimates. <strong>The</strong>se <strong>in</strong>clude the deviation <strong>of</strong> the estimate from the true parametervalue, or bias, as well as measures <strong>of</strong> skewness <strong>and</strong> kurtosis. I compare the bias <strong>and</strong>normality to illustrate the problem. <strong>The</strong>re are several ma<strong>in</strong> results.Firstly, the biases are decreas<strong>in</strong>g function <strong>of</strong> sample sizes. Even <strong>in</strong> small samples around 200<strong>and</strong> 400, the biases are <strong>in</strong> the range <strong>of</strong> 10 -2 . <strong>The</strong> bias for FM-AR is quite small.Secondly, the variance bias exhibits the similar conclusion. AR(3) is generated from (0,1), theempirical bias is <strong>in</strong> the range <strong>of</strong> 10 -3 , <strong>and</strong> is a decreas<strong>in</strong>g function <strong>of</strong> sample size.F<strong>in</strong>ally, the normality property <strong>of</strong> distribution is drawn from skewness <strong>and</strong> kurtosis.Unfortunately, no regular pattern is found among three parameter estimates <strong>and</strong> is relatedto the persistence <strong>of</strong> parameter vector designed; <strong>in</strong> general, skewness is close to zero whichgives normality an acceptable condition, although the excess kurtosis (>3) is found.As a result, FM-AR is a feasible estimator to directly estimate AR(p), whose empiricalapplications also calls for further studies <strong>in</strong> the future.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!