OLS and simultaneous equations bias OLS bias due to “measurement errors” Limitations of regression analysis Ragnar Nymoen Department of Economics, UiO 8 February 2009 ECON 4610: Lecture 5 OLS and simultaneous equations bias OLS bias due to “measurement errors” Overview What are the limitations to regression? Simultaneous equations bias Measurement errors in explanatory variables In both cases the explanatory variable is not exogenous in the econometric sense Main reference is G Ch 15.1 and 15.2;. B Ch 8.1, 10.1 and 10.2;K: Ch 9.3,10.2 ECON 4610: Lecture 5 OLS and simultaneous equations bias OLS bias due to “measurement errors” What are the limitations to regression analysis? It is not linearity in variables, as we have seen it is not linearity in parameters, although we have only covered the linear regression model here Remember that by …rst estimating the linear model we can use the results to estimate parameters that are non-linear functions of the estimated model’s parameters (the “delta method” or its equivalent in the Bårdsen method) If the model is non-linear in the parameter from the outset, can use Non-Linear Least Squares to …t the best non-linear curve to the data. Greene Ch 11, not in the syllabus to this course. It si not con…ned to single equation, as we seen with the SURE estimator. The real limitation to the regression model is when the regression function does not contain the parameter of interest ECON 4610: Lecture 5 OLS and simultaneous equations bias OLS bias due to “measurement errors” A simple Keynes model Let Yt denote GDP in period t D 1, 2, ..., T . Ct is “endogenous expenditure” and let Xt denote “exogenous expenditure”. Assume that Ct depends on GDP, then our example model is Yt Ct D Ct C Xt (1) D b1 C b2 Yt C "t , 0 < b2 < 1 (2) "t is a random disturbance term. We assume that it is white noise uncorrelated with Xt . For simplicity we assume normality "t N.0, 2" /. The parameter of interest is the marginal propensity to consume b2 . ECON 4610: Lecture 5 OLS and simultaneous equations bias OLS bias due to “measurement errors” The reduced form of the model (1) and (2) de…nes a simultaneous equations model. Solution for the two endogenous variables: Yt Ct 11 21 b1 1 b2 b1 D 1 b2 D D 11 C D 21 C 12 D 21 D 12 Xt 22 Xt C 1t (3) C 2t (4) 1 1 b2 b2 1 b2 1 1t D 1 2t D 1 ECON 4610: Lecture 5 1 b2 b2 "t "t OLS and simultaneous equations bias OLS bias due to “measurement errors” The distribution of Y and C The Reduced Form written more compactly Yt Ct D yt C 1t (5) D ct C 2t (6) where 1t 2 y N 0, 2t cy cy 2 c j Xt . The conditional distributions of the stochastic variables 1t and are binormal with zero expectations and variance matrix: 2 y cy cy 2 c j Xt . ECON 4610: Lecture 5 (7) 2t OLS and simultaneous equations bias OLS bias due to “measurement errors” Conditional distribution of C It follows that Yt and Ct are normally distributed with the same 0 covariance matrix as . 1t 2t / and expectations yt D 11 C ct D 21 C 12 Xt , 22 Xt . It also follows (Lect 1) that the conditional distribution of Ct is normal with conditional expectation: E [Ct j Yt ] D D 21 D . C c ct y 22 Xt c 21 y yt c y . c C 11 11 / C . 22 Yt (8) y C 12 Xt / C c y 12 /Xt ECON 4610: Lecture 5 c Yt y C c y Yt OLS and simultaneous equations bias OLS bias due to “measurement errors” We see that The macro model implies (8) as the conditional expextation for Ct . It is the valid regression model of Ct on Yt and can be estimated with full e¢ cency by OLS. It will not deliver an estimate of the marginal propensity to consume, b2 ! In sum: The regression function implied by (1) and (2) is (8), not the regression of Ct on Yt and a constant. And the regression function (8) is not helpful for the estimation for the parameter of interest b1 (in fact since c D 1 it estimates the identity in this special case) ) y ECON 4610: Lecture 5 OLS and simultaneous equations bias OLS bias due to “measurement errors” Simultaneity bias in the macro model example Suppose we estimate the consumption function by OLS regardless. We will estimate “some parameter”. What is it? P P N N Ob2 D PCt .Yt Y / D P Ct .Yt Y / .Yt YN /2 Yt .Yt YN / P where YN D 1/T Yt . X 1 fb1 C b2 Yt C "t gt .Yt YN / (9) bO 2 D P Yt .Yt YN / P "t .Yt YN / D b2 C P .Yt YN /2 We must evaluate the term P in the light of the model. " t .Yt YN / P .Yt YN /2 ECON 4610: Lecture 5 OLS and simultaneous equations bias OLS bias due to “measurement errors” Since Yt depends on the shocks "t to consumption, and Ct depends on Yt , then "t and Yt are correlated. This correlation will not go away as T grows. Using the RF expression for Yt , the denominator can be written as 1 X 1 X 2 XN / C . 1t N 1 / .Yt YN /2 D 12 .Xt T T Take probability limits: 1 X plim .Yt YN /2 D T 1 X 2 XN /2 D plim 12 .Xt T 1 X C 2 12 plim .Xt XN /. 1t N 1 / T 1 X . 1t N 1 /2 C plim T D 212 Var .Xt / C 2y ECON 4610: Lecture 5 OLS and simultaneous equations bias OLS bias due to “measurement errors” bO 2 plim b2 D plim T1 P "t .Yt YN / P plim T1 .Yt YN /2 Cov ."t , Yt / 2 2 12 Var .Xt / C y D From the Reduced Form we also have Cov ."t , Yt / D E ["t D 2 12 Var .Xt / C 2 y D yt ] D E [" t 1 1 b2 "t ] D 2 " 1 b2 1 1 b2 2 Var .Xt / C 2 " ECON 4610: Lecture 5 1 1 b2 Var ["t ] OLS and simultaneous equations bias OLS bias due to “measurement errors” The inconsistency of OLS , 2 " plim bO 2 b2 D D 1 b2 1 1 b2 2 2 " Var .Xt / C .1 b2 / .1 b2 / 2" D Var .X / 2 t Var .Xt / C " C1 2 " The bias is positive Large variance in Xt relative to "t reduces the biases. But it does not kill the bias. The reason is that OLS “assumes the wrong model for Ct ”, one with Cov .Yt , "t / D 0. It is not here. ECON 4610: Lecture 5 OLS and simultaneous equations bias OLS bias due to “measurement errors” Example with an expectations variable Assume the simple regression model (in Greene’s notation again): yi D 1 C 2 xi C " i , i D 1, 2, ..., n. (10) with all the classical assumptions holding. If xi is an expectations variable that we as econometricians cannot observe or cannot measure without error, we can still try to estimate 1 and 2 using the observable (actual) where xi . We then need to make assumptions about the properties of the di¤erence ui D xi xi . ECON 4610: Lecture 5 (11) OLS and simultaneous equations bias OLS bias due to “measurement errors” Assumptions: ui is random, zero mean, variance 2 u Cov .ui , "i / D 0 Cov .ui , xi / D 0 Both ui and "i have the classical properties The model that we estimate becomes: yi i D 1 D "i 2 xi C 2 ui C (12) i (13) But with E [xi i ] D E [.xi C ui /."i 2 ui /] D ECON 4610: Lecture 5 2 2 u OLS and simultaneous equations bias OLS bias due to “measurement errors” OLS gives O2 and we have plim P x/ N i .xi b2 D 2 C P .xi x/ N 2 plim O 2 2 D plim T1 plim T1 P P i .xi .xi x/ N x/ N 2 2 we already have that 2 u goes into the numerator. The denominator is more work (like in the sim eq case) but intuitively it must boil down to the sum of the variances of xi and ui , hence plim ( O 2 2/ D 2 2 u Var .xi / C 2 u ECON 4610: Lecture 5 OLS and simultaneous equations bias OLS bias due to “measurement errors” plim O 2 D 2 1C 2 u < 2 if 2 is positive. Var .xi / It can be shown that by taking the “inverse regression”, xi on yi , gives an overestimation, so OLS de…nes a bound around the true parameter. Measurement errors in yi : No bias problem, but potential for heteroscedasticity. Solution to both classes of bias problems exempli…ed here: Replace OLS with other estimators. IV, 2SLS as we shall see. ECON 4610: Lecture 5