For detailed report http://www.scribd.com/doc/99930372/Sales-Forecast-for-Bhushan-Steel-Limited
For the Matlab code of the same http://www.scribd.com/doc/99930806/Matlab-Code-Sales-Forecasting-for-Bhushan-Steel-Ltd
2. Objective
• Forecast sales: Data Driven – with statistical
techniques
• Compliment Current Process: Heuristiques;–
Substantiate Quantitatively
Its easy to lie with #s – its easier without them!
3. Why forecast sales?
• Sales – Pillar of a business
• Knowing Future Helps BSL Growing well!
(Operations; Financials;
Inventory; Pricing; Investor
relations)
Customers, Sales, Production & Resources – Pillars of business
4. Forecasting Techniques
• Qualitative - rely more on opinion
• Quantitative – primarily statistical and
economics tools
Data Driven Decisions: Combination of qualitative with
quantitative
5. Statistical Forecasting Techniques
• Regression Analysis – Form relation between a
dependent variable and one or more independent
variables
At time of projections we assume independents are known!
Project variable with behavior of others!
6. Statistical Forecasting Techniques
• Time Series Analysis – Form relation between
current value of a variable with its past value
Project future using its past behavior!!
8. Forecasting Sales for BSL – Models
Used
1. Time Series Model – 1,2
2. Regression Model
3. Combined Model
Tools Used: Microsoft® Excel and MathWorks MATLAB®
Other available: SAS, SPSS, SAP
9. TS Model -1
• Time Series Model - 1
– Relation between current sales and sales of past
3 quarters
– Data Crunching produces following relationship…
Performance this quarter – 60% dependant of immediate past!
10. Forecasted Sales - TS Model -1
` 2939.92 Cr
Note a substantial difference at start of Recession !
11. Error Plot – Model -1
Magnitude of errors is increasing with time – not acceptable!
12. Improved TS Model
• Time Series Model - 2
– Do log transformation on sales data
– Data Crunching produces following relationship…
Build model – Validate – Improve – Rebuild!!
14. Error Plot – TS Model -2
Errors don’t demonstrate relation with time – as should!
15. Regression Model
• Dependent = BSL Sales; Independent variables
– sales of BSL customers, Tata Motors and Metal Index
• No or Negative correlations with Tata Motors
(variable 7) and Maruti (variable 5)– Opportunity to
improve!!
Qualitative Analytics can be used to choose independents!!
Opportunity – coefficient to TM and Maruti would change the game!
17. Error Plot - Regression Model
Errors graphs don’t demonstrate relation to time – as should!
18. Combined Model
• Current value of Sales with current values of
independent variables
• Following relation is obtained
• Please note – negative correlation of BSL Sales
with TM and Maruti!! Opportunity to improve..
Please note – variables on right side need to be estimated using
TS and then use!!
21. Summarzing
Model Desciption Std. Error Error Distribution Forecast for June-
2012 quarter (Cr)
TS – 1 Relation with past 3 125.24 Magnifies with 2939.92
quarter sales time
TS – 2 Relation with past 3 128.64 Random 2833.77
quarter sales (after
log transform)
Regression Relation with other 111.72 Random 3076.327
variables
Combined TS on all variables, 121.25 Random 2949. 92
then Regression
22. Opportunity of implementation
• More accurate assesment of future – helps
pricing / inventory etc
• Comparatives with competition – helps
identifying low hanging fruits
23. REFERENCES
• www.moneycontrol.com
• www.fem.uniag.sk
• managementinnovations.wordpress.com
• www.businesslink.gov.uk
• money.howstuffworks.com
• www.wikipedia.org
Data Gathering – Most public listed company data was available
on net. BSL data too was faster on net than A/C department!