Difference between Median and Exponential Smoothing Methods

[Former Member]

Dear All,

Please do let me know the difference between Median and Exponential Smoothing Methods in Forecast.

And on which conditions do I need to select these forcast methods?

Thanks,

Siva.

[Former Member · Accepted solution]

Hi Siva,

Median method basically tries to calculate the median values of the difference in history values from one period to other. It checks for the median values for base and trend and generally takes seasonal index as 1.It is not an advanced method for forecasting but better than moving average method. As it automatically takes the median values for the history differences so any outlier values in history is automatically corrected to median values. As far as application is considered , these should be used where you do not see much changes in pattern of demand . The only piece seen could be trend and some effect of seasonality. E.g could be demand for grossery items in retail industry. They can be considered horizontal and each year you could see some trend in demand variantions. Hence Moving average or median methods could be useful.

Simple Exponential smoothening can be basically of two types - With manual adoption of alpha or automatic adaptation of alpha parameter. The advantage of smoothening method is that you can decide whether you would like to give wightage to recent history values or past history depending on values of alpha. Recommended value is 0.3 however it can vary between 0 and 1.

Simple Exponential smoothening with manual alpha adoption is most basic and used when very less number of historical data points are available and pattern is horizontal. The alpha value is determined on gut feeling of planner depending on whether s/he wants to give more or less weight to recent points.

Simple Exponential smoothening with automaticl alpha adoption . You can use this when you are fairly certain that Simple Exponential Smoothening will fit the data but not sure of the value of alpha factor. The system iteratively goes through various values of alpha and comes up with value that gives lowest error measure selected.You need ti give the starting and ending alpha values with increment values required.The default forecast accuracy measure is MAD but you can select the measure of your choice.

Let me know If it helps.

Regards

Gaurav

[Former Member]

Hi Siva,

1) Median method:- It takes care of basic, trend and seasonal parameter levels

of the forecast. 3 seasons of historical value is must to run this method.

The advantage of this method is not to execute outlier correction which will be

taken care automatically.

2) Exponential smoothing:- System when calculates the forecast using normal

methods, does some extensive calculatiions based on the available historical

values using optimistic upward trend which will be smoothened using this

method.

This usually react very slowly to extreme structural changes in historical data.

Regards

R. Senthil Mareeswaran.