Syllabus Content
The Gigafactory
Just outside Sparks, Nevada, the second largest building in the world by volume is being constructed. The owners, electric car company Tesla, call it the Gigafactory. In 2020, when it is fully complete, it will manufacture half of all the world's lithium-ion batteries.
This may seem a little extravagant for a company that currently only sells above 50,000 cars a year. However, Tesla has big plans, and its sales forecasts predict that they will be selling at least 500,000 a year by the time the Gigafactory is complete. The questions have to be asked: Why does Tesla think it is going to grow so fast? Which Business Management tools did it use to come up with the number of 500,000 cars? And finally, what external factors could cause Tesla's predictions to be wrong?
Sales forecasting
Sales forecasting is an attempt by companies to predict what levels of sales they may expect in future years. If this can be done accurately, the company can enjoy some benefits. If sales are expected to grow, then steps can be taken to ensure this extra demand is met. Inventory levels can be expanded. If necessary, additional staff can be recruited, or production capacity can be increased. If companies feel they will not be able to meet their expected demand, then prices can be increased, so that profits can be maximised.
If a drop in sales is forecasted, then a company may choose to rationalise production. Staff may be made redundant and spare land and capital can be reallocated or sold. However, a company may want to react to lower forecasted sales by increasing marketing budgets, in an attempt to fight off the predicted decline.
This subtopic will explain how moving averages may be used to forecast future sales. This technique first looks for trends in past sales data, which can then be extrapolated to make sales predictions. It is important to compare a period of time with a similar period of time previously.
In a simple sense, it would be wrong for an ice cream parlour to use normal averages to forecast its sales, because these change over the course of a year. Summers will be vastly more successful periods of time, so averages must be calculated for this time, as well as tracking increases in sales over the entire time the business has been operating.
This is just one possible method of sales forecasting. Companies could also choose to look at product lifecycle analysis, economic forecasts or market research results, to help them plan for the future.
Trend analysis
Trend analysis looks for underlying patterns in time series data and attempts to make future predictions. There are two steps to trend analysis: calculating a moving average and extrapolation.
Time series data are statistics that are recorded over time. The independent variable is time, and the dependent is whatever is being recorded.
Analysing data in this way allows us to understand a number of things. First, we can know the trend of the sales the business is making; whether this is rising or falling over time. Second, we can know about any seasonal fluctuations. This is important for businesses that sell seasonal products, like ice creams, holidays and clothing. Third, we can pay attention to any cyclical fluctuations. This means those that are the result of increases or decreases in economic growth. The financial crisis of 2008-09 would have affected most businesses in Europe and in the US, as well as Asia, South America and Africa (perhaps to a lesser extent). Finally, we can know which fluctuations are only happening randomly, and are less worrisome if occurring only sporadically.
A trend is a pattern that is happening often over time. Trends occur in fashion, with clothing that is commonly worn by customer groups. We also mean patterns in sales, for example increasing sales.
The easiest way to understand this concept is by use of an example. So let's imagine a business that specialises in selling second-hand cars. The business has a number of loyal customers, who on average replace their cars once every three years (we will use this information in the next subsection). The company is thinking of expanding. However, it will only be profitable to do so if forecasted sales for 2018 are above 100 cars a year. Below are the company’s sales figures for the last nine years.
STEP 1: Calculating the moving average
A moving average attempts to 'smooth out' any peaks or troughs in sales data so that underlying trends in data can be seen. We know our sales data goes through a three-year cycle, therefore we can use a three-part moving average. The 3-part moving average has been added to our data in Figure 3:
The three-part moving average is calculated using a mean. For example, the first figure of 80 was calculated by taking the average of the sales figures from 2007, 2008 and 2009:
Three-part moving average = (34 + 110 + 96) / 3 = 80
After this, we move along to the next three sets of data. This time using data from 2008, 2009 and 2010:
Three-part moving average = (110 + 96 + 43) / 3 = 83
This process is repeated until we reach the end of our data series – in this case, until we use the sales figure from 2015. The final three-part moving average is calculated using data from 2013, 2014 and 2015:
Three-part moving average = (49 + 122 + 102) / 3 = 91
We cannot go beyond this point as we will not have three different years' sales data to use.
STEP 2: Extrapolate the trend
We can now use our trend line to make predictions about future sales. In the graph below the moving average or trend line has been extrapolated (extended) out as far as 2018.
One way of improving the accuracy of this trend line is to work out coordinates for one point that you know the line must cut through. You do this by working out the average of all the points along the blue line:
X̄ = ∑X (the total years) / N (number of years)
X̄ = 2008 + 2009 + 2010 + 2011 + 2012 + 2013 + 2014 / 7 = 2011
Ӯ = ∑Y (the total sales in the trend)/N (number of years)
Ӯ = 80 + 83 + 85 + 86 + 88 + 90 + 91 / 7 = 86.14
Therefore, one point through which the line of best fit should pass has the coordinates (2011, 86.14).
A line of best fit is a graphical tool that draws a straight line through points plotted on a graph that expresses a relationship between two variables. The straight line will give an indication of the nature of the relationship between the two variables.
You can now see clearly the difference between the red sales line, and the blue line that we have plotted with the 3-period moving average data. The red line is very jagged, and moves up and down frequently. Every third year, sales drop quite dramatically. However, the general trend for the business is still upwards over the period of time.
Using the extrapolated line of best fit the forecasted sales figure for 2015 to 2018 have now been added to our table below.
We are now in a position to answer our original question: should the business expand? Based on our analysis, we predict the business will have sales of just 98 cars in 2018. This is less that the desired figure of 100. Therefore, based on our analysis, we would not recommend expansion at this time.
Task 1: Number of Tesla vehicles delivered worldwide from 3rd quarter 2015 to 1st quarter 2019 (in units)
https://www.statista.com/statistics/502208/tesla-quarterly-vehicle-deliveries/
(a) Access the web page above and input the values for Tesla's quarterly sales from 2015 (Q3) to 2019 (Q1).
(b) Construct a relevant line-graph
Task 2 Sales Trend analysis
PF
The manager of PF with responsibility for monitoring sales of product XN30 is convinced that the levels of sales is on a rising trend, but that it is seasonal with more sales at some times of the year than at others. The manager has gathered the following data about sales in recent years. Trend values are shown in brackets.
(a) Calculate the 3-part moving average for the above data
(b) Illustrate the actual sales level and the trend line on a line graph
(c) Extrapolate the trend for the next two quarters
Task 3:
Variations from the trend
It is rare that sales figures follow a smooth pattern. Businesses will experience good years and bad years. Sometimes sales will be better than average and other times they will be worse. This subtopic explores how firms can use these variations to improve the accuracy of their sale forecasts.
There are three types of variation that a business may face:
· Seasonal variations – Products that experience higher sales volumes at certain times of the year are said to be seasonal. One example is children's toys, for which sales peak at Christmas. Other products may experience a peak in the summer months, such as sun cream, certain clothes and holidays;
· Cyclical variations – Cyclical variations are affected by the economic cycle. Sales of normal goods, such as new cars and televisions, grow in recovery and boom periods and fall during recessions and slumps;
· Random variations – These may occur at any time, and for any reason: A natural disaster, a major sporting event or political unrest can all affect sales of various products, in unpredictable ways.
One of the problems with using moving averages is that it does smooth out the sales data (which we wanted to do). However, the example on the previous page predicted car sales of only 98, but we could have expected more. This is especially true when we consider the volatility of the sales data in that example!
In addition, the economic cycle can experience peaks and troughs that affect the sales of a business. Figure 1 shows the economic or business cycle. This is the real output or real GDP of a country plotted over time.
Using the example from the last subsection, we can use cyclical variations to improve the accuracy of our forecast. This example focused on a business that specialises in selling second-hand cars. The business has a number of loyal customers, who, on average, replace their cars once every three years. The company is thinking of expanding. However, it will only be profitable if forecasted sales for 2018 are above 100 cars a year. Below is a table of data from the last subsection. Two rows have been added: cycle stage and variation.
The method for calculating the three-part moving average and the extrapolation were explained in the previous subsection. We will now extend our analysis to include the cyclical variations.
STEP 3: Calculate the annual variation
The annual variation is the difference between the sales and trend figures. To calculate this the following formula is used:
Variation = Sales – Three-part moving average (Trend)
This was how the first figure of 30 was calculated:
Variation = Sales – Three-part moving average (Trend)
30 = 110 – 80
This process is then repeated for each year we have data for. Notice that the first entry is x for both the three-part moving average and the variation. This is because we do not have data in our table for 2006, which we would need to calculate the three-part moving average for 2007. That means we also cannot calculate the variation for 2007.
STEP 4: Calculate the cyclical variation
Looking at the data, clear patterns can now be seen. Car sales in Year 1 tend to be much lower than the trend, while the opposite is true for sales in Year 2.
A cyclical variation is just an average of all of the annual variations for that cycle stage. We must add the variations together and divide by the number of years:
(30 + 13 – 42 + 30 + 11 – 41 + 31 + 9) / 8 = 5.125
STEP 5: Adjust the sales forecast
Now all that remains is to add the cyclical variation to the sales forecast we calculated in the previous subsection.
The cyclical variation is +5.125, which means that we have to add it to the predicted figure of 98 cars for the year 2018, so that our prediction might be more accurate. We are going to round down the figure, since we are talking about cars.
To remind ourselves about the function of this entire process: the data needed to be smoothed out to be able to predict a reasonable forecast for the future. The trend was 'smoothing' the raw data of the sales figures. Then the line of best fit was drawn, and as it is a straight line, there was more smoothing of the data. When we predict, we try to make a more accurate prediction and try to account for the previous 'smoothing' of the data.
98 + 5 = 103 cars
Our calculations are now complete, and we are now in a position to answer the original question, which was:
Should the business invest in expansion?
Remember the condition of expansion was that forecasted sales have to be over 100 cars per year in 2018. Our prediction estimates sales of 103 cars in 2018 so based on this, we would recommend expansion.
Seasonal variations
Corporations are required to report their earnings to the stock market every three months. These are referred to as 'quarterly earnings reports'. Because of this, most sales data is published in quarters. In addition, it is a good idea to divide the year into periods of 3 months each – the quarters – in order to plan better, and to control the execution of these plans promptly. For example, variance analysis can be done by managers at the end of every quarter, rather than just once a year. Every month would probably be a little too often, especially for large companies.
Quarterly reporting explained
A quarter refers to a period of three months. Normally the calendar is broken into the following quarters:
· January, February and March = Quarter 1.
· April, May and June = Quarter 2.
· July, August and September = Quarter 3.
· October, November and December = Quarter 4.
Sales forecasting
Quarterly sales figures provide a problem when sales forecasting. Because the year is split into four parts, it might make sense to use a four-part moving average. However, this would mean that the trend figure would sit in between two quarters. (If the average of Q1 to Q4 was calculated, the result would have to sit in between Q2 and Q3, for example.) This is unsatisfactory for calculating variations. As a result, the system called centring has been developed.
Table 1 contains some sales data that has been split into quarters. To keep things simple, all figures have been rounded to the nearest million. This section will explain the process of centring and how to calculate a seasonal variation. Let's say that a company wanted to predict their sales for each quarter next year. To do this, many of the steps are similar to those we have studied in the previous two sections. In fact, it is only the first two steps that differ. Below is a worked example of how to 'centre' quarterly sales data so that it can be used as a sales forecast.
STEP 1: Calculate the four-quarter moving total
This first step is different to those we have studied before. When working with quarterly data, we begin by calculating a four-quarter moving total. This number represents the total sales that were made by a company over a twelve-month period. In Figure 2, the four-quarter moving total has been added to our data.
The first figure of 50 is the sum of the first four sales figures (2013 Q1 to 2013 Q4):
· Four-quarter moving total = 21 + 11 + 9 + 9 = 50
Now, we move down the data set. The next figure of 48 is the sum of the next four sales figures (2013 Q2 to 2014 Q1);
· Four-quarter moving total = 11 + 9 + 9 + 19 = 48
This process is repeated until we use the final sales figure of 2016 Q4.
STEP 2: Calculate the eight-quarter moving total
Unsurprisingly, the eight-quarter moving total is a sum of two four-quarter moving totals. This has been added to the table below:
The first figure of 98 is the sum of the first two four-quarter moving totals;
· Eight-quarter moving total = 50 + 48 = 98
Again, this process is repeated until all eight-quarter moving totals have been used. The final figure of 56 was calculated as below;
· Eight-quarter moving total = 29 + 27 = 56
STEP 3: Calculate the eight-quarter moving average (Trend)
We are now in a position to calculate the trend. The table below has had the trend data added.
To get the eight-quarter moving average we divide the eight-quarter moving total by eight. For example, the first figure of 12 was calculated by;
· Eight-quarter moving average (trend) = 98 / 8 = 12.25
This raises two questions: why did we divide by 8 and why was this figure placed opposite 2013 Q3?
The figure of 98 represents sales figures from eight quarters: 2013 Q1, 2010 Q2 (twice), 2013 Q3 (twice), 2013 Q4 (twice) and 2014 Q1. We divide by eight to get the average of these sales figures.
Although 98 represents sales from eight quarters, only five of these are unique (2013 Q1 to 2014 Q1). The mid-point of these five quarters is 2013 Q3, so we must enter the result from our calculation in the cell for the eight-quarter moving average for 2013 Q3.
STEP 4: Calculate the quarterly variation
As before, this is the difference between the actual sales figure and the trend. The quarterly variations have been added to the table below.
This can only be carried out for periods where we have both sales and trend data. The first figure of -3m was calculated using the sales data and moving average for Q3 2013, and the following formula:
Quarterly variation = Sales – Eight-quarter moving average (Trend) = 9 – 12 = -3
STEP 5: Calculate the seasonal variation
Now let's look at all the first quarters in the time series. In our data, Q1 for every year is the most variable period on average, with 8 in 2014, 6 in 2015 and 5 in 2016. To calculate the seasonal variation, we must take the data for the same quarters across the years and average them. The calculations for seasonal variations from our data are as follows:
Quarter 1 seasonal variation = (8 + 6 + 5) / 3 = 6.3
Quarter 2 seasonal variation = (-2 + -1 + -1) / 3 = -1.3
Quarter 3 seasonal variation = (-3 + -2 + -2) / 3 = -2.3
Quarter 4 seasonal variation = (-3 + -3 + -3) / 3 = -3
Whatever number we predict by extrapolation of the trend, we should add or subtract the figure for the seasonal variation, in order to make more accurate predictions.
STEP 6: Plot the data and extrapolate the trend
The sales figures and eight-quarter moving average (trend) have been plotted on the graph below. The trend line has been extrapolated to Q4 2017, using a line of best fit. Remember that previously we needed to calculate the coordinates of the point through which the line of best fit passes.
STEP 7: Make the sales forecast
In the table below, the data in the second column has been taken from the graph we plotted in Step 6. This has been added to the seasonal variation to give our sales forecast.
Based on our analysis, it is predicted that the company will sell a total of $19.2 million in 2017, with $12.3 million in Q1, 3.7 million in Q2, $2.2 million in Q3 and, finally, $1 million in Q4.
Task 4: The following sales data has been collected three times a year over the last four years
(a) Calculate the trend and the seasonal components
Task 5: The figures below relate to the sales trend for Pladex Ltd
Task 6: Time Series Analysis and seasonal adjustments
Captain Cook’s frozen foods have run two sales promotions in the last year (each costing €40,000). One was an on-pack competition in February, and the other was a 25c off on-pack offer in September. The company wants to know which the more effective was in order to set their marketing plans for next year.
Past measurement of their sales figures has already enabled them to calculate their monthly seasonal adjustment factors.
(a)Calculate the estimated actual sales
(b)Calculate the sales volume variance
(c) What was the implied effect of the promotions in February and September?
Task 7: A company is investigating the number of orders of four of its products A, B, C and D and has quarterly data available for the past three years
(a) For each product, calculate the moving total and centred moving averages
(b) Calculate the seasonal components and adjust if necessary.
(c) Deseasonalise the data.
Evaluation of sales forecasting
Files to download