Question for mathtalk-ga please.
Some time ago you answered a question on implementing Kalman filtering
(posed by asiatechnicals-ga). I am developing an implementation of a
model for commodity prices, and I need some help in developing an
agorithm to support testing of my implementation with real data.
I am proficient in Excel and VBA (so would prefer to work in that
platform) and I have some background in econometrics and mathematical
statistics (but very little background with Kalman filtering).
My question is: Help me implement a Kalman filter in Excel/VBA that
will estimate the parameters of given stochastic process with Maximum
Likelihood techniques. I have worked through and understood the model
you developed for asiatechnicals-ga, and this could be an extension of
that. The differences in this case are:
1) the data needs to be a time series of commodity futures prices,
with several different maturities for each time. I can send you an
example data set.
2) the stochastic process for the evolution of futures prices is
known, but the parameters need to be estimated (using maximum
likelihood). I will give you the state-space formulation of the
particular stochastic process I am using.
The deliverables are: Similar to what you developed for
asistechnicals-ga - A file, VBA listing, detailed and simple
instructions for entering formulae, or whatever it takes for me to
have a working, modifiable, Excel implementation.
Fees
====
I am offering a rate of $20 per hour, on the basis of five chargeable
hours work ($100). I am willing to negotiate if the work takes longer
than I have anticipated. |
Request for Question Clarification by
mathtalk-ga
on
07 Aug 2004 14:20 PDT
Hi, finmodeler-ga:
Estimating the parameters is the hard part, isn't it? I'm not sure
how I can best help. If you describe the stochastic process that
models the pricing and post a link where I can download a data sample,
I'll be glad to take a look.
regards, mathtalk-ga
|
Clarification of Question by
finmodeler-ga
on
07 Aug 2004 15:06 PDT
I have an excel file with data, as well as a pdf file that contains
the information about the stochastic process, but can you tell me how
best to get them to you? Is there a google address I could e-mail
them to?
|
Request for Question Clarification by
mathtalk-ga
on
07 Aug 2004 20:46 PDT
Alas, there is no email address to which you can send these things.
The approach I take, as in the Question posed by asiatechnicals-ga, is
to place the files on a Web server from which they can be downloaded.
There are some sites which will allow you to upload files where others
can download them. I guess the first thing to clarify is how big your
files are. Then we can find a suitable "host" for them.
regards, mathtalk-ga
|
Clarification of Question by
finmodeler-ga
on
08 Aug 2004 07:01 PDT
The data file is not very big (37KB). The pdf file contining the paper is 0.85MB.
|
Request for Question Clarification by
mathtalk-ga
on
09 Aug 2004 07:22 PDT
Hi, finmodeler-ga:
Is it possible to post a description of the pricing model here? I
realize that it can be a challenge to use simple ASCII text to
represent mathematical formulas; you may get some ideas for doing this
from my earlier Answer on Kalman filtering.
There are a number of sites that provide "free file hosting", and I'm
asking around to see if there are some recommendable ones.
regards, mathtalk-ga
|
Clarification of Question by
finmodeler-ga
on
09 Aug 2004 07:55 PDT
I would first rather exhaust the possibilities for getting the
information to you in a more user-friendly format, because there is a
fair amount of it and it is critical to have it all correct.
I sent an e-mail to google user support to ask it they could provide
me with an address to send the attachments, which could then be
forwarded to you. I'm sort of surprised they haven't already set
something like this up. The second option is for me to set up a
temporary web page on my DSL provider's server to post the files to
and then provide you with a link.
If I don't hear back from Google user support, I'll proceed with the
web page later today.
|
Clarification of Question by
finmodeler-ga
on
09 Aug 2004 18:01 PDT
mathtalk-ga,
Here is a link to the data. I has been cut and pasted into the model
you created earlier: http://finmodler.4t.com/Kalman%20filter.xls
The information on the stochastic process and state/space formulation
is in a pdf file at this link:
http://finmodler.4t.com/Parameter%20Estimation%20-%20S%26S.pdf
Please let me know if you have any trouble accessing these, or if I
can clarify anything.
Thanks,
finmodler-ga
|
Clarification of Question by
finmodeler-ga
on
09 Aug 2004 20:00 PDT
There were errors in the above links - please use this instead:
http://finmodler.4t.com
|
Request for Question Clarification by
mathtalk-ga
on
09 Aug 2004 20:15 PDT
Okay, nice work, I got them!
regards, mathtalk-ga
|
Clarification of Question by
finmodeler-ga
on
15 Aug 2004 20:08 PDT
Hi Mathtalk-ga,
At the link above I have posted an update to the Kalman Filter model.
I'm still working with the template you provided in your earlier
answer on this topic. I don't think there's much chance I can work
this out completely without assistance, but at least the exercise will
help with my understanding.
What I first did was to link the data on two of the time series (1 mo.
futures, 3 mo. futures) into the model, and then begin to change the
model from a smoothing/fitting application to one for estimating the
parameters of the given stochastic process. You'll also note I
switched your nomenclature a bit to match the process information I
gave you in the pdf file.
I have attempted to label sections of the sheet to show the flow of
updating the state value estimates. There are new columns added for
the mean and covariance of the state variables, as well as for the
measurement errors and their covariance matrix. At the top of the
sheet, I also added a section for the parameters of the stochastic
process, and then set the parameters and covariance matrices of the
measurement and transition equations to calculate directly from the
parameters. So now everything starts with the parameters. Finally,
the log likelihood equation is entered as the target for maximization
(by varying the parameters). Unfortunately, when I tried to solve
using the Solver function in Excel, I got results that are obviously
wrong, so there is a bust somewhere in my approach or calculations.
I doubt that any of this will be of any use to you, but I'm sending
the link just in case. Let me know if you have made any progress, or
have any needs for clarification.
Regards,
finmodler-ga
|
Request for Question Clarification by
mathtalk-ga
on
16 Aug 2004 06:09 PDT
Hi, finmodeler-ga:
I got the updated spreadsheet. I'll post some requests for
clarification later today once I have a chance to look over it.
thanks, mathtalk-ga
|
Clarification of Question by
finmodeler-ga
on
17 Aug 2004 19:59 PDT
Mathtalk-ga,
I don't know if I am close to a solution, or off in the weeds, but I
am definitely stuck. Let me know how you are progressing.
Thanks,
finmodler-ga
|
Request for Question Clarification by
mathtalk-ga
on
18 Aug 2004 04:48 PDT
It's a bit basic, but is there a formula missing at the top of page 2
in the PDF file? It looks like at the first "break" in the text where
the "mean rebounding" effect is going to be modelled, there is a
missing formula, followed by "where [kappa] represents the
mean-reversion rate".
The "missing" formula seems to be called the "Ornstein-Uhlenbeck process".
Of course the formula might be there, but I'm missing a font that
Acrobat Reader needs?
regards, mathtalk-ga
|
Clarification of Question by
finmodeler-ga
on
18 Aug 2004 05:05 PDT
mathtalk-ga,
The formula that is supposed to follow the : on the very top of page 2
somehow floated back to the very bottom of page 1 (perhaps in the
translation to pdf format). So, the formula dX=-KXdt+sigma*dz should
appear after the :
Sorry for the confusion - I should have caught this before posting.
Regards,
finmodler-ga
|
Clarification of Question by
finmodeler-ga
on
28 Aug 2004 12:25 PDT
Hi Mathtalk-ga,
Have you been able to make any progress on this? If you're still
working on it, then I have underestimated the time required and need
to revise the price. If you're not, then I need to withdraw the
question and figure out how to compensate you for your time.
Thanks,
finmodler-ga
|
Request for Question Clarification by
mathtalk-ga
on
28 Aug 2004 18:28 PDT
Hi, finmodeler-ga:
I am primarily motivated to continue working on a problem of this
type, which requires some "learning" on my part, because of the
opportunity not only to learn something new for myself but to teach
others.
The price you posted is quite generous in the scheme of Google Answers
requests, and I would refer you to the guidelines which state that
Researchers are not to post partial answers. So bear with me as we
get through the end of the month, and I'll try to produce something we
can both be happy with.
Of course you are able to Close (Expire) the Question at any time
prior to an Answer being posted, and if you are no longer interested
in an Answer, that would be the appropriate action. The guidelines
being what they are, one can only "afford" to undertake a problem like
this one on the basis of a special interest.
regards, mathtalk-ga
|
Clarification of Question by
finmodeler-ga
on
28 Aug 2004 21:05 PDT
Mathtalk-ga,
That's fine, just wanted to get some idea on whether you were
progressing and the the amount of your time that is being required. I
continue to be interested in the problem, but I also recognize that
this is a fairly challenging one. Let me know if and when it would be
appropriate to extend the expiration date (possible?) or to re-pose
the question.
Regards,
finmodler-ga
|
