I would try Winter's method. You have seasonality, but no trend and no cyclicity, so there is no need to get complicated (i.e., ARIMA).
Hi everyone,
I do have a time series composed by 45 monthly observations and I am trying to figure out which kind of forecasting method is best suited for this series.
I am attaching an xlsx file with the data table, the graphic plot of the series as a whole and a second graphic which shows the “seasonality”, comparing the “yearly” series.
I also made a corelogram analysis through ACF and PACF but I am a sort of newbie about this kind of analysis and the only things I guess I realized are the following:
- The series doesn’t seem to have a trend
- It has a seasonal pattern (12 months)
- The previous points suggest, in my opinion, that both Moving average method and Exponential smoothing are not the right ones to use.
- Maybe either ARIMA or SARIMA would be the way to go, but in this case I don’t know how to deal with the right p, q parameters.
Would you please help me to understand which method would be best for this kind of series and how I could set up a model in Excel 2016 ( I am a power user with this software).
Thanks you so much for any help
Mark
I would try Winter's method. You have seasonality, but no trend and no cyclicity, so there is no need to get complicated (i.e., ARIMA).
But, if I am not mistaking, Holt-Winter's method supposes there is some trend, which can be either damped or not, but a trend is accounted for in the equation.
Since my time series does not have a trend, is this method suitable? In case it is, which kind of holt-winter's should I apply?
Winter's will work with or without a trend. I use Minitab, so I don't know how you would specify this in your software, I just set the gamma (trend) parameter to 0 to specify there is no/minimal trend.
I am trying to build up a model in Excel 2016 but I see its is quite difficult, even if I am a real power-user with Excel....I am considering to start using R ...
In your opinion Miner, a multiplicative model is better than an additive one for my time series?
I tried both and got better MAPE/MAD/MSE with the multiplicative model. It was a small, but noticeable difference.
GretaGarbo (10-12-2016)
Very good Miner, thanks a lot.
I will try to implement it through R...I'll let you know
Here is an introduction to time series in R. Go down almost half to find Holt-Winter. You can set the trend to be off.
Im trying to perform some time series analysis on 2015 and 2016 monthly recorded data to see what method is best for forecasting 2016 monthly values for the remainder of the year. The data has an annual cycle and I already have values for the first 9 months out of 2016. Using Winter's method, I forecasted values for the first 9 months of 2016 and summed up the absolute differences between these forecasted values and the actual 2016 values I already had. From there I calculated the 2016 monthly forecasted value to be off from the 2016 monthly actual value by a 12% average. I then tried a second method where I summed up the absolute differences between the 2015 actual monthly average 2016 actual monthly value. Long story short this elementary method resulted in predictions being off by an average of 6%. My question is how is the Winters error percentage so much higher than using a straghtforward average method. I should also note the MAPE predicted in Minitab using Winters for fitted 2015 monthly data was 2%. Can anyone make sense of all this for me? It would be greatly appreciated.
Your time series has trend: you bet!
It's only stable, not increasing and not decreasing.
You can decompose the series in its parts and then using multiplicative recomposing as forecast method.
Hope I helped you.
Winter's makes assumptions as do all ESM methods. It assumes a trend and it assumes seasonality (which can be additive or multiplicative). If there is no trend, a trend different than the past, no seasonality or the wrong time of seasonality your results will be way off. If you assumed a trend that does not exist, a straight line drawn from the last value which does not assume this trend could be a better prediction.
Time series is not simple, if it was (to quote Loki) anyone could do it As it is even experts struggle with it. Generally speaking you need 50 months of data, not 24, to use Winters. It seems to me you have way to few data points to be doing time series.
"Very few theories have been abandoned because they were found to be invalid on the basis of empirical evidence...." Spanos, 1995
Tweet |