1

I have data in the form given below, and I want to perform forecasting using multiple regression. I found definition of multiple regression from this link: http://otexts.com/fpp/5/1/ . I have these questions: (1) Can we forecast value of Y (the dependent variable) based on the given values of the independent variables (X1 and X2), as in below table? or it is mandatory to have previous values of Y along with X1 and X2 to forecast the future value of Y, (2) I want to forecast the value of Y at T8, by providing two inputs, i.e., value of Y_at_T8 = forecast(X1=22, X2=33), how to proceed on that? , (3) Is it mandatory for multiple regression that Time must be in fixed intervals, can it increase in varying intervals?

enter image description here

Osman Khalid
  • 237

1 Answers1

1

(1) To construct the regression equation you must know the correspending values of y because of the nature of the problem. Otherwise you can't construct the regression equation.

(2) Without knowing y values you can't continue.

(3) The only restriction is that the errors have mean zero and uncorrelated with eachother. You can take any interval you want.

AnilB
  • 4,202
  • Thank you. So Y must represent some subjective rating, or it can be the result of any fixed formula applied on X1 and X2. – Osman Khalid Mar 14 '13 at 22:46
  • @OsmanKhalid That's right. To be more rigid Y is the dependent variable in the system. – AnilB Mar 14 '13 at 23:13
  • Thank you, actually in some example data sets of multiple regression, I saw that Y was assigned values based on some subjective rating criteria, depending on values of independent variables (e.g., http://www.statsoft.com/textbook/multiple-regression/). So I want to know that in the data sets, the value of Y is the result of formula applied on Xs (e.g., gpa calculation based on subject scores). OR, Y must be the subject rating. – Osman Khalid Mar 14 '13 at 23:43
  • As long as there is a relationship between Y and Xs it can be both. I can give more details. What is your background? – AnilB Mar 15 '13 at 00:23
  • My background is computer sciences and software development. It would be really kind of you, if you can provide more details, as I am very much new in forecasting, although I have understood simple linear regression techniques. – Osman Khalid Mar 15 '13 at 04:34
  • Forecasting is a very broad topic. There are lots of methods that can be used such as function approximation, clustering, autoregressive methods, GARCH, neural network, fuzzy algorithms, etc... What do you want to do exactly? – AnilB Mar 20 '13 at 23:27