 |
 | Home | Contact Us | FAQ | Search & Site Map | Link to Us |
 | Sign In | Join | Other 45 Sites in Network |
Math Forum / Mathematics / Numerical Analysis / October 2008Using ARMA: some questions
Hi All, I am not a statistician so please be patient. I am simulating an ARMA time series model. 1. Now to find the realistic parameters, do you think I should use the preiously collected data (which has been recorded for every hour)? 2. Do you think its right if I divide the data into two halves and use the first half for parameter estimation and run the simulation and then compare the results with the later half of the data? 3. Also, when the simulation will be run, will I get the results in half hourly format? TIA Leo > Hi All, > [quoted text clipped - 13 lines] > TIA > Leo Leo, As I understand it you have actual readings at 1/2 hour intervals ....say 48 readings per day ....say for 104 weeks and you wish to build/identify an appropriate ARIMA model. To begin with you need to include events ...e.g. holidays to deal with their effects then you might wish to include both wothin day ( seasonality 0f 48 ) ARIMA structure and possible day-of-the-week effects. You might also have week-of-the-year effects in your data. You might also need to incorporate or account for PULSES , SEASONAL PULSES, LECEL SHIFTS and LOCAL TIME TRENDS. Care should be taken to ensure that the variance of the errors is constant AND that paramaeters don't change over time. We write software (AUTOBOX) to automatically do that. We also provide a consultancy to help you do this with your own software. Regards Dave Reilly Automatic Forecasting Systems http://www.autobox.com 215-675-0652 > Hi All, > [quoted text clipped - 13 lines] > TIA > Leo 3. Unless you take control as the programmer, I suggest you refrain from this because you don't know what is under the hood of some canned program. 2. No, that is almost like trying to compare two random noise sequences. The period for meaningful forecasts is relatively short compared to the minimum length sufficient for parameter estimation. 1. Yes. Besides parameter estimates, you might also use the noise estimates in your simulation (random sample the noise estimates). > > Hi All, > [quoted text clipped - 26 lines] > > - Show quoted text - Absolute and uneiquivocal NONSENSE ... autobox IS TRANSPARENT and shows all steps ...thus the user is inviyed and coaxed to look under the hood .. dave r > > > Hi All, > [quoted text clipped - 35 lines] > > - Show quoted text - Assuming that's true, how does that make my answers to 1 and 2, which you neglected to answer, nonsense? So, your Autobox discloses the exact algorithm it uses for ARMA parameter estimation, e.g., LS, Gaussian ML, LAD, Yule Walker, etc., whether the transfer function poles and zeros are constrained/ unconstrained to lie inside the unit circle, how the residuals or noise estimates are calculated, etc.? If so, I bet it would be just as easy and cheaper for someone to do their own programming than understand your instructions. > > > > Hi All, > [quoted text clipped - 42 lines] > parameter estimation, e.g., LS, Gaussian ML, LAD, Yule Walker, etc., > whether the transfer function poles and zeros are constrained/ Yes they are constrained .. > unconstrained to lie inside the unit circle, how the residuals or > noise estimates are calculated, etc.? If so, I bet it would be just > as easy and cheaper for someone to do their own programming than > understand your instructions.- probably not ! when you actually try it you might have something to say .. hope this helps you ( but I doubt it from your comments ! ) dave Hide quoted text - > - Show quoted text - > > > > > Hi All, > [quoted text clipped - 65 lines] > > - Show quoted text - So, then your software can't identify zeros on the unit circle. No, thanks! > > > > > > Hi All, > [quoted text clipped - 70 lines] > > - Show quoted text - AUTOBOX can identify the need for differencing and it automatically incorporates differences of the appropriate order. You are welcome .. > > > > > > > Hi All, > [quoted text clipped - 77 lines] > > - Show quoted text - You are confusing transfer function poles and zeros. Differencing is for certain poles on the unit circle, not zeros. |
Reply to this Message |
 Sign In
 Join
 My Latest Posts My Monitored Threads My Blog My Photo Gallery My Profile My Homepage Start New Thread Enable EMail Alerts Rate this Thread
|
©2008 Advenet LLC Privacy Policy - Terms of Use
This website includes both content owned or controlled by Advenet as well as content owned or controlled by third parties.