How we can decide forecast model for Steel industry

[Former Member]

Hi Experts,

We are implementing DP and would like to know how we can decide upon the various forecating models like constant, 1st order exponential smoothing, trend etc....

There is one basic funda behind is towards our historical data that how it behave like it was included trend, season etc.. but here in steel industory which params should we look upon for deciding best model.

thanks for your help

Hitesh

[Former Member · Accepted solution]

Hi Hitesh,

For Forecast model selection, the best approach is to form core team

with different members to study the historical data pattern, analyse

it and compare with the various models available in system

Your selection will depend on your knowledge of the data pattern to-date.

Historical values are available and you can make conclusions from these

regarding the most appropriate forecast model.

The model selection approach includes

a) Manual model selection:- You determine the forecast according to your

preferred model. This is suitable when you have a clear idea of the future

development of the key figures although this development will have a

different pattern to their development in the past. In this case, the future

data cannot be derived mathematically from the historical data and you

must let the system know how to proceed.

b) Automatic selection:- The pattern to-date enables you to determine the

presence of seasonality or another trend and, therefore, you run a test to

find the closest development pattern

Regards

R. Senthil Mareeswaran.

[Former Member]

Got an idea how we can select best method of forecasting. Thanks experts

[Former Member]

Hi,

You can create a matrix to map the product segmentation. Based on the historical data and the analysis you have so far, you can subdivide your products in a portfolio matrix according to products with stable sales and products with unstable sales. In addition, you can separate the historical data of your products according to their degree of completeness.

Based on the data that is available to you up to this point, you can make a preliminary assignment of the forecast methods to the products. To do that, you must compare the product segmentation matrix with the matrix of the forecast techniques.

For products with an unstable sales behavior and a short or incomplete history, the Moving Average Method, a Manual forecast, the Bass model, or a Composite forecast would be the most appropriate solutions. The use of a Forecast Team instead of an individual planner would also be conceivable.

For products with a stable sales behavior and a short or incomplete history, you should use the 1st or 2nd Order Exponential Smoothing methods, the Winters method, or the Box-Jenkins method.

For products with a stable sales behavior and a complete history, the most appropriate methods are the Multiple Regression and Lifecycle Method.

Hope this will help you.

Thanks,

Dipankar