You are on page 1of 6

Marketing Analytics HW1

As per the data given Smartphone sales are dependent on several factors such as competitors
pricing, amount of money spent on advertising, dealer incentives and the time since introduction of
our smartphones. Using the historical data given on these parameters, we could better understand how
smartphone sales are affected by the above factors and better predict the future sales of the
smartphones. Analyzing the data with descriptive statistics such as average and Standard Deviation,
the conclusion is that on an average our smartphone prices are priced at $62.40 and the price of the
competitors average around ~$112, approximately ours is 80% lower when compared to the
competitors. Looking at the Standard deviation values in Figure 1 from the statistics, we observe that
there is a lot of variability in the prices in our smartphone prices when compared to the competitors.
The total average Marketing spend is $998 per 100 with a standard deviation of $131.02.
Figure 1 Pricing Mean and Standard Deviation

Figure 1 also shows the variation in the prices in the form 2 (95% Upper and lower limits)
of the smartphone prices of our smartphone and that of the competitors.
Analyzing the historical data shows that on an average, the dealer has offered incentives 74
times of the 400 months i.e. 18.5% of the time. Adding a new variable to the data and creating a pivot
and chart on top of it shows the trend in the number of incentives offered in every year in the data. In
few of the years there were incentives offered for 5 times in year 5 and year 31 and couple of years the
incentives were offered 4 times etc. Below is the graph of the distribution. The year has been
calculated using the CEILING(Month/12,1) formula in Excel
Figure 2 - A pivot table is created and then a Line chart is used for the Graph to show the trend.

Total
6
4
2
0

Total

In order to gauge the price sensitivity of our customers, statistics techniques such as
Correlation and Regression Analysis has been employed. Shown below in Figure 3 is the co-relation
and regression analysis values that were developed to find the relationships between our phone Price
and the price of CompA, price of CompB, price of CompC, advertising Spend, dealer Incentive and
time since the handset was Introduced. In general correlation is usually between -1 and +1. If the
correlation between two variables is -1 then this implies that the two variables are negatively
correlated. The results below show that none of the variables are strongly correlated (i.e. closer to +1).
This type of situation in the data is called Multi-collinearity where two independent variables look to
be correlated (but not highly correlated) and we cannot judge if there are really correlated. Additional
analysis is to be done to test the relationship. However, Advertising spend has the highest positive
value (+40) and is expected that Advertising spend would definitely drive additional sales. Time since
introduction has the highest negative correlation that signifies that with more time older products tend
to move slow in the market. The dealer incentive in this case is a dummy variable (with binary 1 and 0
) and the positive correlation coefficient value of 0.33 in this case. It should not be interpreted as to
having a positive correlation since it does not have any meaning in this type of correlation analysis.
Figure 3 - Correlation

We could also perform a linear regression to find additional meaningful relations in this data.
From the regression results in Figure 4 below, we can see that the price of competitor B and C have
high P-Values greater than 0.05. This indicates that our null hypothesis is valid and these two variables
dont have a significant impact on the sales of our smartphone. Advertising spend P-value 0.08 is
borderline and inconclusive here. The R square value is ~ 0.36. The coefficients of the regression
should be interpreted as the below. If our phone price increase by $1 then our sales fall by ~14,268
units on average. If CompA price increase by $1 then our phone sales increase by ~9196. Similarly, if
CompB increases then we see a gain of 365 only and with increase in CompC we sell 3060 units more
respectively.
From the residual plots in Figure 5 in appendix, we can see that that the values are
concentrated around the X axis without much deviation. This suggests that the data might be nonlinear. To correct for this anomaly, we could perform regression analysis on the natural log values of
the price data. One thing to note is that since dealer incentive and time since introduction are not
related to price of the phone we should not take the natural log values for this analysis. Figure 6 in
Appendix shows the results of the Regression Analysis on Natural Log values of the data. From the
natural log regression model results, our phone coefficients indicate the impact the price on the sales
since it is now our elasticity (i.e the dependent variable changes by the % change in the independent
variable x the coefficient from the natural log regression analysis). These elasticities can be used to
explain the price sensitivity of our customers. If our price increases by 1% then our sales would

decrease by -1.69%. In other words, if our price increased from $62.40 to $63.02 (a 1% increase) our
sales would fall by ~8,190 units (reduction from ~485,316 units to ~477,274 units). Similarly, we
could see the effect of our competitors price on our sales. If competitor A increased their price by 1%
we will see an increase in sales of ~7,413 units and if competitors B and C increase their prices by 1%
we will see an increase in sales of ~1,506 units and ~4,070 units respectively. For advertising
spending, dealer incentive and time since introduction, since we didnt take the natural log of those
variables, the regression coefficient is the percentage change in the independent variable for every unit
change in the dependent variable.
From the linear regression model coefficients generated and simple calculations, we could
estimate/predict what our future sales would be given the other parameters. To do this we have created
a Excel Model with Inputs varying in the green cell. This Model would give the results that are
reported below in Figure 8. The excel formulas used in preparing the model is shown in Figure 9 of
the Appendix. If competitor A decreases prices by 10% we will see a fall in sales by ~52,850 units. But
when the dealer offers an incentive it would result in an overall sales of ~549,108 units.
One of the major disadvantage of regression models is their ability to predict outside the data
intervals. Because of this prediction of the future sales values by extrapolation may be very unreliable.
With that said and the lot of variability in the price of our phones that we see from Figure 1 between
$32 and $94, I would not ideally recommend predicting sales units at the two price points requested.
Figure 8 Sales Prediction

Appendix
Figure 4 Regression on Smartphone sales data

Figure 5 Associated Residual Plots

Figure 6 Regression on Natural Log values of Smartphone sales data

Figure 7 Residual Plots Regression on Natural Log values of Smartphone sales data

Figure 9 - Excel Formulas in the Model are shown Below.