Marketing Analytics - Group 7
Marketing Analytics - Group 7
Marketing Analytics - Group 7
Introduction ............................................................................................................................................ 2
Moving Average Method ........................................................................................................................ 3
Exponential Smoothing: .......................................................................................................................... 4
Holts method ......................................................................................................................................... 5
Multiplicative model ............................................................................................................................... 7
Comparison of methods ......................................................................................................................... 8
Introduction
Forecasts of sales for one of the companys products affect everything from production
schedules, raw material purchasing, inventory policies, and sales quotas. Consequently, poor
forecasts leads to poor planning and increased costs for the company. To tackle such a grave
issue, how should a company possibly try to develop its forecasting methodology?
Good judgment, intuition, and an awareness of the state of the economy may give the company
an idea or feeling of what is likely to happen in the future, but converting that feeling into a
number that can be used as next years sales forecast is extremely difficult. Forecasting
methods that we can use can be either qualitative or quantitative.
Qualitative methods generally involve the use of expert judgment to develop forecasts. Such
methods are appropriate when historical data on the variable being forecast are either not
applicable or unavailable. Quantitative forecasting methods, on the other hand, can be used
when (1) past information about the variable being forecast is available, (2) the information
can be quantified, and (3) it is reasonable to assume that the pattern of the past will continue
into the future. In such cases, a forecast can be developed using a time series method. We will
focus exclusively on quantitative forecasting methods in our report.
The objective of time series analysis is to discover a pattern in the historical data or time series
and then extrapolate the pattern into the future; the forecast is based solely on past values of
the variable and/or on past forecast errors. The underlying pattern in the time series is one of
the most important factor in selecting a forecasting method. Thus, we first look at the time
series plot to try to determine what forecasting method to use.
However, the measures of forecast accuracy that we see in the report simply measure how well
the forecasting method is able to forecast historical values of the time series. Though measures
of forecast accuracy are important factors in comparing different forecasting methods, we have
to be careful not to rely upon them too heavily. Good judgment and knowledge about business
conditions that might affect the forecast also have to be carefully considered when selecting a
method.
The purpose of this report is to provide an understanding about the time series analysis and
forecasting using various methods, and which one among them may be best suitable in the
given context.
Moving Average Method
There are two type of moving average method that could be used in forecasting:
Simple Moving Average Method:
Moving average is a widely used indicator in technical analysis to smooth out price
action by filtering out the noise from random prices that are observed in data set
Using moving averages could be an effective method for eliminating strong
fluctuations in data.
The major limitation is that data points from older data are not weighted differently
than data points near the beginning of the data set.
MAD
BIAS MSE MAPE
Exponential Smoothing:
Past data should be discounted in a more gradual fashionfor example, the most recent
observation should get a little more weight than 2nd most recent, and the 2nd most
recent should get a little more weight than the 3rd most recent, and so on. The simple
exponential smoothing (SES) model accomplishes this.
For a series y1, y2, yn, forecast function, which gives an estimate of series of l steps ahead
can be written as:
Y(n+l) m lb l=1,2,3,4...
Where mn is the current level and bn is the current slope. Therefore, the one step ahead of the
prediction is given by:
Since, there are two terms to the exponential smoothing, two different smoothing constants are
required, alpha0 for the level and a1 for the slope.
This provides the level at time t. Since the level at time t-1 is already known, it is somehow
possible to estimate of the slope:
These equations can also be transcribed in the form of appropriate error correction form:
This method, known as holts method and requires starting values for mt and bt to be input. It
would be typical to find 0.02<a0, a1<0.2, but they can be estimated by minimising the sum of
squared errors as in the form of single exponential smoothing. Also, it is often found that,
m1=y1, and b1=y2-y1 are reasonable starting values.
Obtaining forecasts
We consider the details of forecasting with the multiplicative models of the form
Where the new components are in the logarithms of those in the earlier equation.
The model is therefore it is fitted by the originally taking the logs of original data values
and fitting an additive model. However, this model will give forecasts of the log data.
To obtain forecasts in the original scale, the inverse transformation must be applied. If
logarithms to the base 10 have been used (as in the diagrams in cast), the appropriate
inverse transformation is
If logarithms are used, the exponential function must be used to return the forecast
constructed on the log data for a forecast in the original units.
In many times series involving quantities (e.g. Money, production,), the absolute
differences within the values are less interest and importance than the percentage
variations.
Assuming that the seasonal and all other components acting proportionally on the series
are the equivalent to a multiplicative model,
Multiplicative models are similarly easy to fit to data as the additive models. To fitt a
multiplicative models take logarithms of both sides of the model,
After taking logarithms (either the natural logarithms or to base 10) the four components
of the time series act additively. It is important to identify when multiplicative models are
appropriate.
Exponential
-593.9148936 23835.71319 918447957.9 32.97%
Smoothing
Multiplicative -
Seasonality 15.43921543 14542.51966 299259738.1 20%
Looking at the comparison table, we can see that the multiplicative seasonality method has the
lowest value of MAPE at 20%. Thus, for the given data, we should use this to forecast future
value of sales.