Sunday, June 21, 2009

Demand Planning - Essence of Forecasting - Part II

In my earlier blog, I was explaining about forecast fundamentals and in this article we will explore more about Forecasting techniques on Quantitative and Casual with various methods along with time series components.

Quantitative Technique

Uses historical data. In this technique single variable (eg., Sales / Demand) is used.

Time Series Methods consist of

Moving Average
Exponential Smoothing
Holt's Model - Exponential Smoothing Adjusted for Trend
Winter Model – Exponential Smoothing Adjusted for Trend and Seasonal
Times series decomposition
Times series extrapolation

Casual Techniques

Uses the relationship between demand and some other factors (income, population etc) to develop forecast. This uses several independent variables rather than the historical pattern of the time series. These independent variables contain information useful to explain the sales patterns of the dependent variable. For example increase in disposable income (independent variable) lead to increase in sales (dependent variable). There is a correlation between disposable income and sales. This information may help to build a forecasting model more accurate than dependent on sales history alone.

Methods used in Casual Technique

Correlation Method
Regression Model
Econometric Model

However in some cases more than one technique is used . For example ARIMA uses casual (Auto Regression) and Quantitative (Moving Average) techniques. This method is used in advanced forecasting tool.

Time Series

Time Series is a set of evenly spaced or continuous numerical data which is historic in nature, where the basic demand pattern varies little between years. It assumes the factors influencing past will continue in future. It uses statistical models as forecasting tools.

Components of Time Series

Trend – Long term (several years) tendency of a series to rise or fall (upward or downward trend), due to population, technology etc.
Seasonal – Periodic fluctuation in the time series within a certain time period (within a year). These fluctuations form a pattern that tends to repeat from one seasonal period to another. This is due to weather and customs (Diwali, Christmas)
Cyclical – Generally occur over a larger time interval (2 – 10 years duration), and the length of time between time successive peaks or troughs of a cycle are not necessarily the same. Due to factors influencing the economy.
Random – Eratic, unsystematic, "residual" . Random noise or error in a time series. Due to unforeseen events like strike etc.

We focus our attention to Trend, Seasonal, Random components rather than cyclical as it occur over larger time interval.

Times series data contains systematic and random components. Systematic is a combination of Trend, Seasonal and Level. Level is a time series data without having Trend and Seasonal components.

Before using the sales / demand data in a time series one must understand the components present in the data. It is not necessary that all components like Trend, Seasonal, Cyclical, Random should be present in the data. This understanding is critical in choosing the right forecast method.

Given below the table which shows the forecast method to be followed against the given components in the data.

Forecast Method -----------------Applicability
Moving Average --------------------No Trend or Seasonality
Simple Exponential Smoothing ------No Trend or Seasonality
Holt’s Model ------------------------Trend but no Seasonality
Winter Method----------------------Trend and Seasonality

How to choose right Forecast technique and method ? Given below the diagram which shows how the forecast technique and method to be chosen based on the available data.



Acid test for the forecast personal lies in selecting correct forecasting technique and “best fit” forecast method.

Let us compare car buying process against forecast technique and method selection process. For example if a person would like to buy a car, first he will decide on segment like small size, medium, luxury, SUV and then he will choose the brand based on performance (trouble free) and other criteria. Trouble free performance can be compared against “Least Error” in statistical parlance. Hence Forecast personal will choose the right technique first and then select the “best fit” method among various methods which gives least error.

Forecast Technique : At the outset it may look easy to select Time Series as one need to forecast on historical sales data. But after careful consideration, other inputs from Marketing (promotion, Pricing) and sales (Field activity, competition) need to be considered with weightage while forecasting. This observation may lead to opt for casual method due to usage of many related variables.

Forecasting method : Selecting right forecast method is cumbersome process due to availability of wide variety of methods. For example in Time Series lot of methods, as explained above are available. User need to plot the data in all methods chosen and then select the "best fit" method based on result.

What is “Best Fit” Forecast method ?

Forecast personal plot the historical data in the various forecast methods chosen based on technique. He compares the result / output of each method and choose the method which gives least error. Basically he used to compare the MSE (Mean Square Error) & MAPE (Mean Absolute Error Percentage) of each method and then select the best fitted forecast method. This opens up a project scope in Six Sigma on Forecast method.

Above discussion clearly indicate that the forecast personal should familiarize with Statistical techniques, able to understand the output and interpret the results clearly. This helps the top management to plan their Sales and Marketing strategy effectively.

In my next session we will explore about CPFR (collaborative Planning Forecasting Replenishment), Consensus Forecast and analyze right forecast method during recession period.

4 comments:

  1. Correlation Method and Regression Model whats the difference?

    ReplyDelete
  2. Correlation

    Correlation is a measure of association/relationship between two random variables. The variables are not designated as dependent or independent. Correlation quantifies the degree to which two variables are related. A correlation coefficient tells you how much one variable tends to change when the other one does. In correlation analysis you do not consider cause and effect relation between variables. The two most popular correlation coefficients are: Spearman's correlation coefficient and Pearson's product-moment correlation.

    Regression

    Regression is used to examine the relationship between one dependent and one or more independent variable. After performing an analysis, the regression statistics can be used to predict the dependent variable when the independent variable is known. Regression goes beyond correlation by adding prediction capabilities.
    For example, an economist can use personal income details (independent variable - Y) to predict the consumption pattern (dependent variable- X). The purpose of running the regression is to find a formula that fits the relationship between the two variables. Then you can use that formula to predict values for the dependent variable when only the independent variable is known. With regression, you do have to think about cause and effect as the regression line is determined as the best way to predict Y from X.

    Difference
    1. Correlation measures the relationship between two random variables whereas regression is used to examine the relationship between dependent and independent variables.
    2. Regression goes beyond correlation by adding prediction capabilities.
    3. Correlation does not talk about cause and effect relationship whereas regression does.
    4. With correlation, the variables are random, it doesn't matter which of the two variables you call X and which you call Y. You'll get the same correlation coefficient if you swap the two. With linear regression, the decision of which variable you call X (independent) and which you call Y (dependent) matters a lot, as you'll get a different best-fit line if you swap the two. The line that best predicts Y from X is not the same as the line that predicts X from Y.
    5. Correlation is almost used when you measure both variables and normally one will not experimentally manipulate variable. With linear regression, the X variable (income) is often something you experimentally manipulate and the Y variable (consumption) is something you measure.

    ReplyDelete
  3. superb - i love your blog.

    ReplyDelete
  4. Should the auto industry base its sales forecasts more heavily on qualitative or quantitative techniques? what is ur input

    ReplyDelete