Making predictions - J Programming

Message from discussion Making predictions

DevonMcC@gmail.com

More options Jan 10, 10:04 am

From: "Devon...@gmail.com" <Devon...@gmail.com>

Date: Thu, 10 Jan 2008 07:04:01 -0800 (PST)

Local: Thurs, Jan 10 2008 10:04 am

Subject: Re: Making predictions

  Reply to author
  |
  Forward
  | Print
  | View thread
  | Show original
  | Report this message
  | Find messages by this author
  

Actually, future oil prices can be extrapolated on even less data: the
supply is finite and consumption is growing.

However, Gosi's exercise here is a nice example of using J to simply
apply a moderately sophisticated technique to a price series and
display the result graphically. To take this a little further, I
would suggest a few enhancements.

1. Use better names - for instance, instead of "a", how about
something like "oilpx" ("px" is a common abbreviation for "price")?
2. Combine better names with some J shortcuts for breaking apart
data. For instance, "x3" and "x4" could be assigned with a statement
like
'polydegree fittype'=. x NB. Assuming my guesses about meaning of
these 2 vars are correct...
3. Show the original data in the same picture as the fitted line.

Also, I might not bother with the intermediate variables y1 through y5
- their values could be calculated in-line just as easily, but that's
a matter of taste.

As to the validity of forecasting this way, I have a library of more
than a hundred books about this subject and have been studying it for
years, so it's far too involved to get into here. Suffice to say that
this is a good pedagogical example.

On Jan 10, 9:28 am, Lynn & Bob Graf <lbg...@yahoo.com> wrote:

> Do you really believe that oil prices can be extrapolated into the future based on historical pricing data alone? Incredible! I daresay that without being able to factor in issues of international politics, future value of currencies and possibly gold, technical inovation, mass psychology, among other things, this whole prediction business is no more than a mathematical curiosity. If it's really relevant, perhaps you should be publishing this stuff elsewhere....

> What is being done here seems to a person like me with no economics knowledge, as the equivalent of trying to fit any data to a polynomial equation. That works well for interpolation, but for extrapolation - you've got to be kidding. The equations used to fit data for extrapolation must be based on scientific reality, which usually results in a non-linear regression analysis using something like a Marquardt algorithm to an equation which can be virtually anything depending on discipline. Considering that economics is, I believe, at best, a soft science, and validity to this prediction would be dumb luck.

> gosi <gos...@gmail.com> wrote:

> From the page about the price of oilhttp://oilnergy.com/1opost.htm

> I read in the points from the yearly prices

> a=:30 22 21 20 20 22 23 21 21 21 21 15 16 15 19 19 19 18 19 19 19 20
> 20 19 19 20 20 20 20 22 25 30 30 30 30 31 32 31 31 32 32 32 33 35 35
> 35 35 35 36 36 36 36 38 36 36 36 36 36 36 36 36 36 37 37 37 37 37 38
> 38 39 39 40 75 78 80 52 90 140 220 320 310 305 290 130 155 130 160 200
> 170 165 148 140 149 180 168 110 160 270 230 240 280 370

> a will then contain data until 2005 since 1920

> (3;1) spall a

> The older data may not be so interesting for the predictions so I take
> the last 30 years since 1975
> (3;1) spall _30{.a

> I try out different options and find the 3 gives good fit

> (3;3) spall _30{.a

> It gives quite a good fit and the two more points available fit well
> into the estimate

> I then add the two remaining actual points in

> b=: a,500 600

> (3;3) spall _30{.b

> Whatever option I use then for the foreseeable future it looks like
> the prices of oil are going up

> ---------------------------------
> Never miss a thing. Make Yahoo your homepage.

Reply to author Forward