Statisticians_group : Message: Re: [Statisticians_group] Problem related to time series analysis

Swathy,

1. U must know or hypothesie the nature of data and accordingly, in principle, confirm such nature by use of fitting function.

2. R-sqr will increase with increase in number of degrees of polynomial, as the overfitting of the data occurs.

3. Further, R-sqr of 100% means the curve fits the data points as they r. Thus, if data points have some random error associated with them, then this is obviously wrong.

Therefore, if ur data is from a source which has some random flactuation aroudn a trend line, u can choose before hand (a) the function to be fit to the data baed on ur hypothesis and (b) the value of R-sqr at some pre-determined level of acceptability.

typically, 70% and above will be a good value.Typically, one should not go beyond cubic polynomial.

finally, There are no hard and fast rules.

surendra

----- Original Message ----
From: swathi shetty <swathisona_shetty@...>
To: Statisticians_group@...
Sent: Wednesday, October 11, 2006 11:26:41 AM
Subject: [Statisticians_group] Problem related to time series analysis

Hi all,

Can anybody help me in solving the below mentioned problems

Excel has a number of tools to assist in analyzing trends in data. Excel offers six choices of types of trend lines that might be applied to the charted data, namely

Linear
Logarithmic
Polynomial
Power
Exponential
Moving Average

And along with these trend line, if necessary excel will also display the R² value and the equation.

My problem here is in interpreting the R²value, based on the R² value we select one of the trend line which best fits the data and use it to predict the future values, but the problem in case of polynomial regression is when we increase the degree of the polynomial the R² value increases. What decision has to be taken in this case? Do we have to select the model with highest degree, since it has the highest R² value?

In case of time series analysis with one set of study variable, when we fit a trend line using polynomial regression

Y=b+c₁X+c₂X²+------------+c₆X⁶in excel the values for X are taken as 1, 2, 3, 4, 5, --------------, as the number of observation increases the X value also increases and therefore the future predicted response values(Y) will be too big for the model with positive coefficient which is far from reality. How to over come this problem?

Looking forward for the solutions.

Kind Regards

Swathi.

Find out what India is talking about on - Yahoo! Answers India
Send FREE SMS to your friend's mobile from Yahoo! Messenger Version 8. Get it NOW

Yahoo! Groups Tips

Did you know...

Messages