It sounds like you are doing some kind of multivariate time series analysis. You need to first establish what are your dependent variables and what are your independent variables and whether your data values are independent. Time series data readings are not independent and therefore they violate one of the assumptions of multiple linear regression.

Dear Peter Samuels,

-Thank you for the answer

-I am beginner in statistics

-What I would like to say is that I only have one record contains monthly rainfall records the values are between(0-350)

-I want to make a statistical forecasting model to compare it's MSE with Neural network MSE.

- if I made my data contains 4 columns[X,lag1,lag2,lag3],and then applied these data to SPSS Multiple regression(X as dependent variable) and (lag1,lag2,lag3 as independent variable, I will get the following equation:

X=b0 + b1*lag1 + b1*lag2 + b3*lag3

What is the name of this model is it:

a-multiple regression

b-vector auto regressive

c-auto regression

or, what is the name of this model????

It's an autoregressive model -- you're regressing X on its lagged values.  But as Peter Samuels indicated, there may be independent variables affecting rainfall.  For example, what about seasonal variations?  

Dear Dr.Mitchell,

-Thank you for your answer.

- Actually I have already developed a NN network using 6 independent variables and I have compared the result with a multiple regression equation with same 6 independent variables, but now I want to compare using only the pervious data of the dependent variable.

- Can you please lead me to the name of Article or book that I can cite from that the autoregressive is same as multiple regression if the regression was on the lagged values?

Dear Yasir,

as Dr. Mitchell points: this is an autoregressive model.

i would like also to bring to your attention that the selection of lag length is an important issue that will affect significantly your results (e.g. including more lags than necessary will result in over-fitting -see below reference-)

this is important. for example, [1] indicates that over-fitting leads to a higher mean-square forecast error of the AR and that under-fitting the lag length often produces auto-correlated errors. 

in [3] there is a  prove that the precision of forecasts from VAR models differs

significantly for alternative lag lengths.

lag-length selection should be based on formal criterion which is supposed to make

equilibrium between the number of included variables on one side and the accuracy of

AR equations on the other side. (see [2)]

furthermore, as Dr. Mitchell stated, you should also take into concern the seasonal variations. this is related to the computation of ACF, PACF (as you probably know).

hope this helps.

regards; 

[1] Lütkepohl H.: ¨ Introduction to Multiple Time Series Analysis. 2nd ed. (1993)

New York: Springer.

[2] CANOVA F. Methods For Applied Macroeconomic Research PRINCETON

UNIVERSITY PRESS. (2007) 110-113

[3] Hafer, R. W. and Sheehan, R. G.: “The Sensitivity of VAR Forecasts to

Alternative Lag Structures.” International Journal of Forecasting 5 (1989), 399-

408