I have sales data for one product for 24 months. It gave me the impression there is a seasonal effect so I used the Holt-Winters method. Using gretl I run the Dickey-Fuller augmented test and apparently the series is non stationary. But since the series has only 24 month data there was no point in trying the ARMA method for forecasting.

I used a software specialized in Holt-Winters method and contrasted the forecast also run on Excel (although Excel only gave me one month forecast). The former uses initially alpha, beta and gamma coefficients in the following order: 0.3, 0.1 and 0.1. Unfortunately, I do not know what are the optimized parameters used for the forecast solution. This is why I also used Excel. I started in Excel with 0.5, 0.5, 0.5 but the Solver gave me the following solution: alpha=0, beta=0, gamma=0.5. Out of curiosity I also used Excel to run different methods in the event the supposedly seasonal effect is not substantial: moving average (N=6), weighted moving average (Solver gave me these coefficients: 0.37, 0.47, 0.16), simple exponential smoothing (Solver gave me 0.93).

The question is the following: Am I interpreting correctly if there is a seasonal effect in the series with a slight tendency to justify the use of the additive Holt-Winters method? Maybe I am misinterpreting the series pattern. Any idea why Solver gave me an alpha=0 and a beta=0? Is that because there is not trend effect at all?

In case any of you is interested in making suggestions, I can pass the monthly sales data for years 2013 and 2014 so that data can be graphed. All I need is a suggestion as far as the forecasting technique you would use. I would be appreciative of suggestions.

Lobo