BUS 520 Module 4 Case (Revisions)

BUS 520 Business Analytics and Decision Making

Dr. Margaret Sabe

29 June 2016

Background of the Problem

A client of Excellent Consulting Group (ECG) wants a way to forecast future sales given past sales data. Linear regression forecasting has been done using the sales data and previous months’ website hits but my consulting manager wants to make sure that the client is given the most effective methodology and has thus tasked me with doing a time series analysis on the sales data. The data involves the sales details of a lottery app for smartphones and to comply with my manager, a simple exponential smoothing method has been used to forecast future sales of the lottery app. Alphas of 0.15 and 0.9 were used for different analysis and the results are as presented below.

Excel Analysis

Part 1

Part 2

Linear regression forecasts

Analysis of Excel

Observations

Using an alpha of 0.15 and an alpha of 0.9 in the SES analysis of the first twelve months, it can be observed that an alpha of 0.9 produces lower Mean Error (ME) and Mean Percent Error (MPE) compared to the ME and MPE of the 0.15 alpha. However, when using 0.15 as alpha, the Mean Absolute Percent Error (MAPE) for the data is lower than when using 0.9 as the alpha. It can also be noted that using the 0.9 alpha produces a negative MPE while the 0.15 alpha gives a positive MPE.

An SES analysis for the three months, that is, February, March and April gives positive ME, MPE and MAPE and the 0.15 alpha provides values that are consistently lower than those gotten from use of the 0.9 alpha.

Assumptions, Logic and Rationale

· SES forecasting assumes that there are no trends or seasonal patterns to the data provided and that assumption is also applicable to this case. This assumption is important because the single exponential smoothing time series forecasting method does not consider any other data as important except the current observation. This means that if any trend or seasonality of sales exists, this forecasting method does not incorporate that possibility.

Looking at the actual sales data, the sales number decreases in March and then dramatically increases in April but under this assumption, the sales number is assumed to show such volatility as a result of other factors rather than a trend or seasonal pattern. Rationally speaking, a lottery app is not a seasonal consumer good thus the assumption made for this case is quite logical.

Discussion and Recommendation

Using the regression method, the ME, MPE and MAPE can be calculated to be 43.82, 0.077 and 0.077 respectively. Notice that the MPE and MAPE are similar, this means that all the sales forecasts were less than the actual sales for the months of interest. In the same vein, the MPE and MAPE found for both alphas using the SES method are different which means that the forecasts are sometimes greater than the actual sales. Despite this, the computations show that the SES method produces less ME, MPE and MAPE than the linear regression method. Another advantage of the SES model is the flexibility to experiment with different alphas until one is found that is most suited for the data under investigation.

Generally, the method that has the lowest MAPE is the most accurate since it shows that the forecasted numbers are closer to the actual values than a high MAPE and in this case, the three month SES using a 0.15 alpha is the most appropriate method to use for forecasting sales numbers for the lottery app for smartphones. Between the linear regression forecasts and the SES forecasts, the latter forecasts are more accurate.