We help IT Professionals succeed at work.

Future value prediction

Medium Priority
365 Views
Last Modified: 2012-05-06
Hi,

I'm looking for ways to predict short-term value given a long term historical value. I've tried a simple method which perform curve fitting through data-points and find the equation. However, this method doesn't seems to be working well.

Any suggestions on what algorithm or methods should I do to get a pretty good approximation?

Thanks
Comment
Watch Question

CERTIFIED EXPERT

Commented:
"looking for ways to predict short-term value given a long term historical value."

Some historical data do not predict future results, the stock market for example, because there are too may independent variables involved.

What, for example, are you trying to predict?

Author

Commented:
Hi,

I'm trying to predict a total cost of operation for the company given historical data. So along the x-axis will be quarter of years (i.e  q1_1980, q2_1980....q4_2008) and along the y-axis i will have operation cost total.

Given those information, I need to find ways to predict q1_2009, q2_2009... etc.

Feel free to ask me any other questions.

Thank you in advanced
CERTIFIED EXPERT
Commented:


If you have Excel then enter your quarter data in Column A (A1, A2, A2...An) and
enter you corresponding data in column B (B1, B2, B3...Bn)

the last row should have q1_2009 in the A column and its data in the B column

then add q1_2009, q2_2009... etc. in the A column.

Then in the B column next to q1_2009, insert the Trend function (its a statistical function)
Following the example given in the first posting at http://www.experts-exchange.com/Software/Office_Productivity/Office_Suites/MS_Office/Excel/Q_21465810.html

After you have the first trend point then copy and past the trend function into the cells below it in the B column.


Note: remember to do the ctl-shift-enter, as described.  And use the Round function, in order to round the data


Not the solution you were looking for? Getting a personalized solution is easy.

Ask the Experts

Author

Commented:
Hi, I'm acturely implementing it into company program so I need somehow know what algorithm can be used to perform this task. It'll be great if you can provide me the name and possiblely algorithm.

Thank you in advanced
ozo
CERTIFIED EXPERT
Most Valuable Expert 2014
Top Expert 2015
Commented:
do you have a model for what relationships you would expect to hold between  short-term values and long term historical value?
Would you expect values to be periodic?  Would you expect values look like Gaussian noise around a linear function?
Each model may yield a predictive algorithm, but you need to have some model in order to choose the algorithm.
Commented:
A few ideas on how to improve your results.

1) Your short term trends are more likely to follow recent results, than long term ones. To compensate for this you should run your predictor formula over different periods 3,5,10 years, and take a weighted average of results, giving more weight to the short periods.

2) For most companies there is also a cyclic trend from quarter to quarter, rather than an even performance over the year. you can compensate for this, by running seperate predictor formulae for each quarter, and then again taking a weighted average with more weight being given to the current quarters results.

A tailored combination of the two methods should give you much more reliable predictions, once you find the correct weightings to use for your company. Afraid this bit is really just trial and error, but if you use method to try and predict the last couple of years results, from previous history, and tune weightings to give reasonable results for all quarters, you should have much more luck predicting future trends.
ozo
CERTIFIED EXPERT
Most Valuable Expert 2014
Top Expert 2015
Commented:
whether or not short term trends are more likely to follow recent results or whether there is also a cyclic trend from quarter to quarter depends on the nature of the values one is predicting, which is why it is necessary to have a model for how the values behave before one can chose and tailor an appropriate algorithm

Author

Commented:
Thank you for answering this broad question. After merge ideas from all of you I came up with a pretty good solution. Again, this solution is derived from all of your effort. Thank you again for your time
Access more of Experts Exchange with a free account
Thanks for using Experts Exchange.

Create a free account to continue.

Limited access with a free account allows you to:

  • View three pieces of content (articles, solutions, posts, and videos)
  • Ask the experts questions (counted toward content limit)
  • Customize your dashboard and profile

*This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.

OR

Please enter a first name

Please enter a last name

8+ characters (letters, numbers, and a symbol)

By clicking, you agree to the Terms of Use and Privacy Policy.