Consumption based asset pricing models

[ranveersk]

Hi
I'm doing my dissertation on empirically comparing asset pricing models through GMM, I'm having a bit of trouble estimating and testing the models.
I'm comparing the standard expected utility model with constant relative risk aversion, a recursive utility model (Epstien & Zin), two habit persistence models: internal habit persistence (Ferson & Constantinides) and an external habit model (Campbell and Cochrane). The topic is relatively new to me but I've read up on the theory and papers that compare the models but now I have to compare them for my self with my dataset on eviews and none of the papers I've seen use Eviews. these are my problems:

I've set up a system with the moment conditions, on the estimation options I choose Time series HAC,I think I use 2SLS? from what i've understood I need to estimate with the identity matrix first then the matrix updates to the "optimal" weighting matrix ( as described by hansen and singleton) the inverse of the variance- covariance matrix of the sample counter part of the pricing equation E(m R)-1=0, m= stochastic discount factor for a model, R= returns of assets.
I want to use the optimal matrix as above but also get results with just the identity matrix for comparison purposes, what settings do I choose in Eviews to conduct it both ways?
I'm a bit confused about the kernel and bandwith options, none of the papers I've looked at say anything about this.

I also want to conduct Hansen and Jagannathans Volatility bounds and Specification error tests as non parametric evaluations. If anyone's familiar with this, how can I do it on eviews?
Volatility bounds : Std(m)>Std(x)= {[E(q)-E(m)E(x)]' *(unconditional variance covariance matrix of x)^-1[E(q)-E(m)E(x)]}^(1/2)
E(m)= unconditional mean of candidate stochastic factor Std(m)= standard deviation of candidate stochastic discount factor, x= pay off vector, p= asset price vector

Specification error tests: d= {[E(q)-E(mx)]' E(xx')^-1[E(q)-E(mx)]}^(1/2), E(xx')^-1 which is fixed is supposed to be a weighting matrix but by using the pre fixing the matrix to that for comparison

umm in the internal habit model, the SDF has expectations operators in it, how do I define that in the system? for exapmple the numerator in the SDF function is
(Ct+1-bCt)^-y - bFEt+1(Ct+2-bCt+1)^-y the t's and t+1's are supposed to be subscripts, the denominator is the same but lagged vales of c. Do I just type it out as though they arent there? is there an Expectations operator command in eviews?

let me give you an idea of what my problem is exactly, the topic was suggested by a lecturer, the topic isnt covered in our curriculum but it appealed to me cause it was a straight forward comparison that was just demanding technically (which i don't mind) compared to the other topics. but my supervisor hasn't replied to any of my emails or anything. So I've had to learn this on my own from the articles on each model and what ever papers that go into the topic (I've found that there so many different takes on how people estimate them, most of them do it through GMM but even in that they take different approaches from how they set different restrictions or additional conditions to different techniques in GMM) , I've wasted alot of time getting confused reading papers which I now realise are out of bounds for my disso, In fact all I have are other peoples studies and articles to base what my disso should be structured as. I think I have a decent grasp of the models now and an alright idea of techniques I'm supposed to use but on actually doing the empirical work on eviews i'm screwed, this method is totally new to me. If anyone could guide me on a framework with in Eviews It would save my life, literally. I can upload a doc breifly explaining each model and the non parametric evaluations if someone isn't familiar with topic but would still like to help......
oh one last thing, after I get estimates for the coefficients ,run the estimations will different pre specified values and instrument sets, compare jstats.... etc. The most basic thing: how do I graph the fitted results with the actual asset data.

Thanks in advance

[ranveersk]

ok no one seems to be helping out with the larger framework of it

could some one please explain how I should estimate the internal habit model, cause it has expectations operator in it, how do i treat that?
I type out the moment conditions E(m*R)-1=0 where m is the stochastic discount factor, the SDF of the internal habit model is:

D*[(Ct+1-bCt)^-y - bD*Et+1(Ct+2-bCt+1)^-y]/[(Ct-bCt-1)^-y- bD*Et(Ct+1-bCt)^-y]

D= subjective discount facot,C= consumption of nondurables and services, b= time non separable paprameter( time preference) y=coefficient of relative risk aversion, Et is the expected value in the current period, Et+1 is the expected value in the next period. do i ignore the expectations operator?

can anyone explain what bandwidth and kernel options I should use? actually i think I use the quadratic option cause I'm supposed to minimize G'WG which is quadratic right, G= the sample counter part of the Euler equation restrictions with instrumental variables, W is the weighting matrix. which leads me to my next problem:

lets say I want to pre specify the weighting matrix so i can keep it the same when testing the different models. how do i do this?
I want to use the identity matrix and a matrix defined as E(xx')^-1 = the second moment of payoffs x=payoffs or returns on the assets
The first (identity matrix) is just so that all assets are priced equally so I can compare the different models based on that, rather than when the weighting matrix updates the resulting gmm criterion may be affected by changes in the weighting matrix rather than the models ability to price assets. I understand that using the identity matrix is the same as OLS cross sectional regression, can anyone explain how i do this with my models?
The second( second moment of payoffs) Is so the GMM criterion reflects the true distance between the candidate stochastic discount factor and the space of admissible discount factors

anyone? please.....

[BradleyMorris]

I'm sorry Ranversk, I'm not able to help you, I just started with eviews, but if you don't mind telling me how to set eviews up in order to do non-parametric tests (it seems like you had some experience in that) that would be great.

I'm want to run a spearman's correlation and the question was which factor specifications do I need to enter for this in eviews? In the tab "factor estimation" it asks you for the method, the number of factors, and initial communalities. All are a mystery to me..

I'm running a panel data regression, with non-normally distributed data, and what I what to see is the correlation between the factors..

I hope you can find the time to answer me, and good luck with your own work.!

-Bradley