1

I have 4 years of daily data. there is a decreasing trend for the data for the first 3 years but the trend increase for the 4th year. I wanted to find a fitted model using the first 3 years and then test the model for 4th year but I think I can not do so as the best fitted model shows a decreasing trend and predicts the 4th year to have lower values but the actual data shows a different trend. How can I treat these kinds of data? I can not use 4 full years as I dont have the data for the 5th year to test my model so I had to use the first 3 years as the historic data and the fourth year as the test data.

user12
  • 129
  • 10

1 Answers1

1
    "A trend is a trend is a trend, 
But the question is, will it bend? 
Will it alter its course 
through some unforeseen force 
And come to a premature end ? - Cairncross (1969)

This is quote that I got from the book principles of forecasting by J Scott Armstrong. Following the above quote Armstrong writes:

    Will the trend bend ? Some statisticians believe that the data can reveal this. 
In my judgement, this question can be best answered by domain knowledge. 
Experts often have a good knowledge of the series 
and what causes it to vary. - J Scott Armstrong, Principles of forecasting (2002)

Forecasting trend is very hard, unless you have domain knowledge and expertise to know on what causes trend to increase/decrease/stay the same it is going to very difficult to forecast the trend irrespective of using any algorithms.

forecaster
  • 7,349
  • 9
  • 43
  • 81
  • Thank you for your quote. Does it mean that I can not the existing models to forecast future data for a historic data with this feature? – user12 Sep 22 '14 at 19:20
  • I would say, avoid data mining to forecast trends. You need to know /understand "why" the trends behave as they do, and then use an empirically validated model/method such as analog time series to predict future trends. – forecaster Sep 22 '14 at 19:23