Download presentation

Presentation is loading. Please wait.

Published byKaleb Sowle Modified about 1 year ago

1
1 THE DISTURBANCE TERM IN LOGARITHMIC MODELS Thus far, nothing has been said about the disturbance term in nonlinear regression models.

2
2 For the regression results in a linearized model to have the desired properties, the disturbance term in the transformed model should be additive and it should satisfy the regression model conditions. THE DISTURBANCE TERM IN LOGARITHMIC MODELS

3
3 To be able to perform the usual tests, it should be normally distributed in the transformed model. THE DISTURBANCE TERM IN LOGARITHMIC MODELS

4
4 In the case of the first example of a nonlinear model, there was no problem. If the disturbance term had the required properties in the original model, it would have them in the regression model. It has not been affected by the transformation. THE DISTURBANCE TERM IN LOGARITHMIC MODELS

5
5 In the discussion of the logarithmic model, the disturbance term was omitted altogether. THE DISTURBANCE TERM IN LOGARITHMIC MODELS

6
6 However, implicitly it was being assumed that there was an additive disturbance term in the transformed model.

7
7 THE DISTURBANCE TERM IN LOGARITHMIC MODELS For this to be possible, the random component in the original model must be a multiplicative term, e u.

8
8 THE DISTURBANCE TERM IN LOGARITHMIC MODELS We will denote this multiplicative term v.

9
9 THE DISTURBANCE TERM IN LOGARITHMIC MODELS When u is equal to 0, not modifying the value of log Y, v is equal to 1, likewise not modifying the value of Y.

10
10 THE DISTURBANCE TERM IN LOGARITHMIC MODELS Positive values of u correspond to values of v greater than 1, the random factor having a positive effect on Y and log Y. Likewise negative values of u correspond to values of v between 0 and 1, the random factor having a negative effect on Y and log Y.

11
v f(v)f(v) 11 Besides satisfying the regression model conditions, we need u to be normally distributed if we are to perform t tests and F tests. THE DISTURBANCE TERM IN LOGARITHMIC MODELS

12
v f(v)f(v) 12 THE DISTURBANCE TERM IN LOGARITHMIC MODELS This will be the case if v has a lognormal distribution, shown above.

13
v f(v)f(v) 13 THE DISTURBANCE TERM IN LOGARITHMIC MODELS The mode of the distribution is located at v = 1, where u = 0.

14
v f(v)f(v) 14 THE DISTURBANCE TERM IN LOGARITHMIC MODELS The same multiplicative disturbance term is needed in the semilogarithmic model.

15
v f(v)f(v) 15 THE DISTURBANCE TERM IN LOGARITHMIC MODELS Note that, with this distribution, one should expect a small proportion of observations to be subject to large positive random effects.

16
16 Here is the scatter diagram for earnings and schooling using Data Set 21. You can see that there are several outliers, with the four most extreme highlighted. THE DISTURBANCE TERM IN LOGARITHMIC MODELS

17
17 Here is the scatter diagram for the semilogarithmic model, with its regression line. The same four observations remain outliers, but they do not appear to be so extreme. THE DISTURBANCE TERM IN LOGARITHMIC MODELS

18
18 The histogram above compares the distributions of the residuals from the linear and semi- logarithmic regressions. The distributions have been standardized, that is, scaled so that they have standard deviation equal to 1, to make them comparable. 0–1–2– THE DISTURBANCE TERM IN LOGARITHMIC MODELS

19
19 0–1–2– THE DISTURBANCE TERM IN LOGARITHMIC MODELS It can be shown that if the disturbance term in a regression model has a normal distribution, so will the residuals.

20
20 0–1–2– THE DISTURBANCE TERM IN LOGARITHMIC MODELS It is obvious that the residuals from the semilogarithmic regression are approximately normal, but those from the linear regression are not. This is evidence that the semi- logarithmic model is the better specification.

21
21 What would happen if the disturbance term in the logarithmic or semilogarithmic model were additive, rather than multiplicative? THE DISTURBANCE TERM IN LOGARITHMIC MODELS

22
22 If this were the case, we would not be able to linearize the model by taking logarithms. There is no way of simplifying log( 1 X + u). We should have to use some nonlinear regression technique. 2 THE DISTURBANCE TERM IN LOGARITHMIC MODELS

23
Copyright Christopher Dougherty These slideshows may be downloaded by anyone, anywhere for personal use. Subject to respect for copyright and, where appropriate, attribution, they may be used as a resource for teaching an econometrics course. There is no need to refer to the author. The content of this slideshow comes from Section 4.2 of C. Dougherty, Introduction to Econometrics, fourth edition 2011, Oxford University Press. Additional (free) resources for both students and instructors may be downloaded from the OUP Online Resource Centre Individuals studying econometrics on their own who feel that they might benefit from participation in a formal course should consider the London School of Economics summer school course EC212 Introduction to Econometrics or the University of London International Programmes distance learning course EC2020 Elements of Econometrics

Similar presentations

© 2017 SlidePlayer.com Inc.

All rights reserved.

Ads by Google