# Value of alpha and beta in Holt's exponential smoothing method

How to choose the best values of alpha and beta in Holt's exponential smoothing? Leaving it upon R gives me $\alpha$ =1. Is this appropriate?

Entering different values of alpha and then comparing with the real data shows best result for $\alpha$ = 0.45. But then R calculates $\beta$=0.99. Is this fine?

-

Values of $\alpha$ and $\beta$ close to one suggest the model is mis-specified.

Try using the ets() function in the forecast package instead. It will choose the model for you, and select the best values of the smoothing parameters.

-
Used ets(). There is a slight improvement. Alpha=0.9999 while beta is close to 0.02. I am using daily data for one year for short-term forecasting. using six month data gives alpha=0.9996 with beta in the same range. – Leo Nov 12 '12 at 6:29
I think a value of alpha close to 1 means estimates are based on the recent observations while small beta implies that trend is not changing a lot. Correct me if I am wrong. – Leo Nov 12 '12 at 8:56

I am no expert in forecasting, but to my experience Holt-Winter's method may be highly unstable in R (especially in double exponential smoothing - where high frequency data may have extreme forecasts).

As for the alpha, beta values I think the most reasonable approach is to plot the forecasts--to check whether the value given by R makes sense or not. As I said, factors such as missing values or high frequency may highly effect the output returned by R.

If you are using ggplot2, you might want to look to this Plotting forecast() objects in ggplot part 2: Visualize Observations, Fits, and Forecasts.

-
I have plotted the data and it's constantly trending. Residuals have been checked. Only the value of alpha I get is 1. I thought it should be less than 1. – Leo Nov 12 '12 at 2:33