NEAS


Group: Administrators
Posts: 4.2K,
Visits: 1.2K

Module 2: Basics of regression analysis (The attached PDF file has better formatting.) Homework Assignment: attributes of classical regression analysis Claim severity and speed Suppose a regression of Y = the logarithm of claim severity on X = the speed of the car satisfies the five attributes of classical regression analysis on pages 1517. Explain whether of regression of Yʹ = claim severity on X = the speed of the car satisfies each attribute. Jacob: What is this homework assignment asking? Rachel: Yʹ = e^{Y}. If the conditional distribution of Y, given X, is symmetric, is the conditional distribution of Yʹ, given X, symmetric or skewed? Answer this question for each of the five attributes on page 1517:
symmetric vs skewed single mode vs multiple modes normal vs heavy tailed equal vs unequal spread linear vs nonlinear
For four of these five attributes, the relation assumed in classical regression analysis does not hold for Yʹ if it holds for Y. Jacob: Are the five attributes explicitly listed? Rachel: The five attributes are implicit in Fox’s discussion: symmetric, unimodal, normal distribution, constant variance, and linear relation.



CalLadyQED


Group: Forum Members
Posts: 62,
Visits: 2

I'm having a hard time with this HW. Any thoughts on how to go about reasoning this? How long an answer are we needing to give?



KenShun


Group: Forum Members
Posts: 8,
Visits: 1

I also found this problem difficult. [NEAS: Jacob: What are we supposed to explain for this homework assignment? Rachel: If Y is normally distributed, Yʹ is lognormally distributed. If Y is symmetric, Yʹ is skewed, and the assumption of classical regression analysis does not hold. This problem is true for much actuarial work. In later modules, Fox explains how to solve this problem by generalized linear models. This pattern holds for all five items.]



CalLadyQED


Group: Forum Members
Posts: 62,
Visits: 2

I skipped this homework and came back to it after getting to the module on transformations. By considering the discussion of the log transformation in Module 6 (Chapter 4), I believe I've been able to figure out which characteristic is preserved by exponentiation. For example, if we use a log transformation to change the skewness and create symmetry, we would assume that the original function is skewed.
Nevertheless, I'm still at a loss as to how much of a reason is wanted. It all gets down to the nature of the function f(x)=e^x. [NEAS: Correct. We are not looking for much explanation. These items become clear in the chapters on generalized linear models.]



noturbizniss


Group: Forum Members
Posts: 13,
Visits: 3

Is a one sentence answer sufficient? I.E. to list which attributes it does and does not hold for? [NEAS: A concise answer is sufficient.]



anne26


Group: Forum Members
Posts: 4,
Visits: 1

I am having a hard time answering this homework. There are no examples given even in the reference book. In my previous Microeconomic VEE, examples were given first in the modules and this will give us idea in answering our homework.
Adding examples will help us a lot. I hope NEAS will do this.
[NEAS: Fox discusses five attributes: symmetric, unimodal, normal distribution, constant variance, and linear relation. Explain (in a phrase) how each attribute relates to lognormal distributions. For example, the lognormal distribution is skewed, not symmetric.]



Michelle2010


Group: Forum Members
Posts: 18,
Visits: 1

NEAS: Could someone help me with the equal vs unequal spread? Here are my thoughts: If Y is normal with equal spread, then the conditional variance of Y, sigma, is constant for all the Xs. I realize that this does not necessarily mean the conditional variance of Y' will be constant for all the Xs, but wouldn't it be constant in the case where the mean of Y is constant for all the Xs? Is their someway I can insert formulas on this discussion board to make my question clearer? [NEAS: If the mean of Y is constant at all X points, the estimated beta is zero: Y = alpha + 0 * X + epsilon. Classical regression analysis assumes the means of Y differ by X (so beta is not zero), but the variance of the distribution of Y does not differ by X.]



FrequentlySevere


Group: Forum Members
Posts: 8,
Visits: 1

If you take a look at the lognormal pdf (http://en.wikipedia.org/wiki/Lognormal_distribution) here, you can get an idea of how the lognormal distribution might be described in terms of skewness, heavy tails, etc...



dwhat


Group: Forum Members
Posts: 1,
Visits: 1

I'm a little confused by one of the responses in a past post, that said we should comment on the features (e.g. skewness vs. symmetry) of a lognormal variable. The third feature is normal / heavy tailed so from that I thought that we would not assume that the variable Y' is lognormal except in that one part of the question. Should we assume Y is symmetric, has a single mode, is normal, has equal spread AND is linear? Or should we comment on each separately (e.g. Y given X has a single mode but is not necessarily normal)?
[NEAS: The normal distribution has the five attributes you mention. If the distribution is lognormal, which of the five attributes change?]



newFrad


Group: Forum Members
Posts: 2,
Visits: 1

I believe I've been able to figure out which characteristic is preserved by exponentiation. For example, if we use a log transformation to change the skewness and create symmetry, we would assume that the original function is skewed.
HP Support


