Advanced Management Accounting

Session 1

Forecasting

10/04/2018 1
 Forecasting

Text Book References

Drury – Eighth edition- Chapter 23

2
 Forecasting

Main topics for this session are :-

• The hi-lo method.

• Regression analysis.
• Tests of reliability.
• Time series analysis

3
 Forecasting - High/low Method

This involves selecting the past cost periods with the highest and
lowest activity levels. Past total costs are separated into fixed
and variable elements in the form of an equation

y = a + bx

where a = fixed Cost

b = variable Cost
 Example
The management accountant of a company is trying to estimate
the company’s anticipated costs in respect of fuel for the next
few months. The company has been operating for only six
months and data relating to this period is as follows:
Month Production hours Fuel cost (£)
March 60 1700
April 90 2300
May 75 1960
June 80 2050
July 85 2180
August 65 1810
 Solution
Hours Costs (£)
High 90 2300
Low 60 1700
Difference 30 600

Assuming fixed costs are constant, the

difference between the two levels of cost
is variable cost of £600.

Variable cost per hour =

£600/30hrs = £20/hour
 Solution
We can then substitute this into, for example, the “high” period
to get

£2300 = fixed costs + ( 90 hours x £20/hr)

£2300 = fixed costs + £1800
£2300-£1800 = fixed costs = £500

The hi-lo method is easy to apply and is a useful examination

technique. However all other observations are ignored and the
highest & lowest values are unlikely to be typical.
 Solution
• We can then substitute this into, for example, the
“high” period to get
£2300 = fixed costs + (90 hours x £20/hour)
£2300 = fixed costs + £1800
£2300 - £1800 = fixed costs = £500
• The hi-lo method is easy to apply and is a useful
examination technique. However all other
observations are ignored and the highest & lowest
values are unlikely to be typical.
 The least squares method
• The line of best fit can be more accurately
derived by using a mathematical technique to
calculate the regression line of best
• This can be calculated using the same data as
before by applying formulae.
• If our regression line of best fit has the equation:

y = a + bx

Where: y = total costs, x = hours,

a= fixed costs and b = variable costs
 The least squares method

The formula for a is a= ∑y - b ∑x

n n
where n = the number of observations made.

The formula for b is b = n∑xy - ∑x∑y

n∑x² -(∑x)²
 The least squares method
The calculations are as follows:

2 2
x y y xy xy x² x y² y
60 1,700 102,000 3,600 2,890,000
90 2,300 207,000 8,100 5,290,000
75 1,960 147,000 5,625 3,841,600
80 2,050 164,000 6,400 4,202,500
85 2,180 185,300 7,225 4,752,400
65 1,810 117,650 4,225 3,276,100
∑ x=455 ∑ y=12,000 ∑ xy=922,950 ∑ x² =35,175 ∑ y²=24,252,600
 The least squares method

b= n∑xy - ∑x∑y a= ∑y - b ∑x
n∑x2 – (∑x)2 n n
= (12,000/6) - (19.3043 x 455/6)
= (6 x 922,950) - (455 x 12,000) = (12,000 / 6) - (19.3043 x 455 / 6) =
(6 x 35,175) - 4552 2,000 - 1463.9094
= 5,537,700 - 5460 000 = 536.09
(211,050 - 207,025)
= 77,000 Fixed costs are £536.09
4,025
= 19.3043

Variable costs are £19.30 per hour

 Graphical Representation
Fuel Cost
3000

2500

2000

1500

1000

Fuel Cost
500 Linear (Fuel Cost)

0
Production hours 60 65 75 80 85 90
 The Correlation Coefficient
( termed ‘r’)
Tests if two sets of data appear to have a relationship and
indicates the extent of this.

The measure can have values in the range

+1 perfect positive correlation

0 no correlation
-1 perfect negative correlation.
The formula for the correlation
coefficient, r

The formula for the correlation coefficient =

r= n∑xy - ∑x∑y
SQRT([n∑x2 – (∑x)2] [n∑y2 – (∑y)2])

r= (6 x 922,950) – (455 x 12,000)

SQRT([(6 x 35,175) – 4552] x[(6 x 24,252,600 )-12,0002])

r= 5,537,700 – 5,460,000
SQRT([211,050 – 207,025] [145,515,600 – 144,000,000])

r= 77,700 = + 0.9948229
78,104.35
 Coefficient of Determination (r2)

Tests how much of the variation in the ‘y’ values can be explained by changes
in the ‘x’ values.

In this example - how much of the variation in cost can be explained by the
variations in activity level, expressed in hours.

The value must be between 0 and 1,

Can be found by squaring the correlation coefficient, by formula,

or by the RSQ function on Excel.

r² = 0.9849672

This means that 98.49672% of the change in fuel costs is explained by the
change in production hours.
 Limitations

The estimates of cost behaviour were made using a range of

observations, in this case from 60 to 90 hours. If we were to estimate
fixed and variable costs using this data and use these values to
estimate costs for data outside this range (for example 100 hours) then
this could be misleading. The costs may only be linear over the
relevant range.

The above assumes that the past can be used to predict the future
which may not be the case. Also we are assuming that cost is
determined by only one variable. There may be many variables
affecting cost, in which case multiple regression (which is not covered
on this course)would be more appropriate.
 Time Series Analysis

A time series can be thought of as a set of quantitative

data that are obtained at regular periods of time.

The objectives of time series analysis are to isolate and

identify those factors that have influenced patterns of
activity in the past and present and will continue to
influence activity in the much the same manner in the
future. The values derived from the analysis are used in
predicting future activity and assisting with
management planning and control.
 Example
Sales
Year Quarter Leads
Consider the 20X0 1 20
2 15
following data that 3 10
4 18
relates to quarterly 20X1 1 24
sales leads 2 18
3 13
generated by a 4 21
company over a 4 20X2 1
2
28
22
year period. 3 19
4 25
20X3 1 32
2 26
3 21
4 29
If we wish to use
Y ear x y
regression analysis, we
20X 0 1 20
need to assign the x and 2 15
y values. 3 10
4 18
20X 1 5 24
The y value is simply the
6 18
sales leads data, and the 7 13
quarter number is the x 8 21
value. 20X 2 9 28
10 22
11 19
We do this by assigning 12 25
a numeric value to each 20X 3 13 32
quarter, starting at 1 14 26
15 21
and in this case running 16 29
to 16.
If we wish to use
Year x y
regression analysis, we
20X0 1 20
need to assign the x and 2 15
y values. 3 10
4 18
The y value is simply the 20X1 5 24
6 18
sales leads data, and the 7 13
quarter number is the x 8 21
value. 20X2 9 28
10 22
11 19
We do this by assigning
12 25
a numeric value to each 20X3 13 32
quarter, starting at 1 14 26
and in this case running 15 21
to 16. 16 29
Line of best fit – graphical solution
Sales Leads Regression Line
35

30

25
Sales Leads

20

15 Sales Leads

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Quarter
We can apply the formulae to determine the regression equation
i.e.:

y = a + bx

where y = sales leads, x = month number, a= starting value and b = variable element.

The components are found by the following formulae:

a = ∑y - b ∑x
n n
where n = the number of observations made and:

b = n∑xy - ∑x∑y
n∑x² -(∑x)²
• You can use an EXCEL spreadsheet rather than
calculating the line of best fit manually.
• Applying the LINEST and INTERCEPT statistical
functions to the data, gives the answer:

Y = 14.3 + 0.825x

• We can now substitute the quarter number for the x

value in the formula to calculate the trend value for
each subsequent quarter. For quarter 18:
Y = 14.3 + 0.825x
= 14.3 + (0.825 x 14) = 25.85
 Seasonal Adjustment
• The regression equation allows us to forecast future
values based on historical trends.
• However, a problem arises when dealing with data
which is affected by seasonal fluctuations during the
year (for example ice cream sales would be expected
to be higher during the summer months and lower
during the winter months).
• To deal with this, we need to adjust the forecast
value by a seasonal adjustment.
 Time Series – the Additive Model
• The additive model for time series analysis is:
TS = T + SV + R
Where TS = the actual time series,
T is the trend series,
SV is the seasonal component and
R is the random (non recurring component)
• This can also be expressed as:
TS – T = SV and R
• Meaning that if the trend series is deducted from the actual
series, the seasonal and residual components of the time
series will remain.
• For the purposes of this module, we will assume that the
residual element is negligible and the seasonal component
can be taken as:
SV = TS - T
 Using Moving Averages To Find The
Trend
• Moving averages can be used as an alternative to
regression analysis to find the trend.
• Applying this approach to the sales leads data, we
calculate a moving average of each of the four
quarters, starting with quarters 1,2,3 and 4 with
the total and average plotted in quarter 3.
• As quarter 3 is not the real mid-point, the moving
average of this and the next quarter is averaged
to establish the trend.
 Using Moving Averages To Find The
Trend
• Moving averages can be used as an alternative to
regression analysis to find the trend.
• Applying this approach to the sales leads data, we
calculate a moving average of each of the four
quarters, starting with quarters 1,2,3 and 4 with
the total and average plotted in quarter 3.
• As quarter 3 is not the real mid-point, the moving
average of this and the next quarter is averaged
to establish the trend.
Using Moving Averages To Find The
Trend
S ales 4 Qtr 4 Qtr Mid point
Y ear Quarter L eads Total Average (Trend)
20X 0 1 20
2 15
3 10 63 15.75 16.25
4 18 67 16.75 17.125
20X 1 1 24 70 17.5 17.875
2 18 73 18.25 18.625
3 13 76 19 19.5
4 21 80 20 20.5
20X 2 1 28 84 21 21.75
2 22 90 22.5 23
3 19 94 23.5 24
4 25 98 24.5 25
20X 3 1 32 102 25.5 25.75
2 26 104 26 26.5
3 21 108 27
4 29
Using Moving Averages To Find The
Trend
S ales 4 Qtr 4 Qtr Mid point
Y ear Quarter L eads Total Average (Trend)
20X 0 1 20
2 15
3 10 63 15.75 16.25
4 18 67 16.75 17.125
20X 1 1 24 70 17.5 17.875
2 18 73 18.25 18.625
3 13 76 19 19.5
4 21 80 20 20.5
20X 2 1 28 84 21 21.75
2 22 90 22.5 23
3 19 94 23.5 24
4 25 98 24.5 25
20X 3 1 32 102 25.5 25.75
2 26 104 26 26.5
3 21 108 27
4 29
 S ales 4 Qtr 4 Qtr Mid point S eas onal
Y ear Quarter L eads Total Average (Trend) Var
20X 0 1 20
2 15
3 10 63 15.75 16.25 -6.25
4 18 67 16.75 17.125 0.875
20X 1 1 24 70 17.5 17.875 6.125
2 18 73 18.25 18.625 -0.625
3 13 76 19 19.5 -6.5
4 21 80 20 20.5 0.5
20X 2 1 28 84 21 21.75 6.25
2 22 90 22.5 23 -1
3 19 94 23.5 24 -5
4 25 98 24.5 25 0
20X 3 1 32 102 25.5 25.75 6.25
2 26 104 26 26.5 -0.5
3 21 108 27
4 29

The seasonal variation can now be calculated

using the formula:
SV = TS - T
 Proportional Method
• The actual time series can also be expressed
as a %age of the trend as shown in the next
slide (this is known as the Proportional
Method).
 S ales Mid point S eas onal S eas onal
Quarter n L eads (Trend) Var % age
20X 0 1 1 20
2 2 15
3 3 10 16.25 -6.25 0.6154
4 4 18 17.125 0.875 1.0511
20X 1 1 5 24 17.875 6.125 1.3427
2 6 18 18.625 -0.625 0.9664
3 7 13 19.5 -6.5 0.6667
4 8 21 20.5 0.5 1.0244
20X 2 1 9 28 21.75 6.25 1.2874
2 10 22 23 -1 0.9565
3 111 19 24 -5 0.7917
4 12 25 25 0 1.0000
20X 3 1 13 32 25.75 6.25 1.2427
2 14 26 26.5 -0.5 0.9811
3 15 21
4 16 29
 Q1 Q2 Q3 Q4
0 0 0.6154 1.0511
1.3427 0.9664 0.6667 1.0244
1.2874 0.9565 0.7917 1.0000
1.2427 0.9811 0 0

Total 3.87273 2.904 2.074 3.075 Total

Average 1.29091 0.968 0.691 1.025 3.98
Adjusted 1.29892 0.974 0.696 1.032 4

Using the additive model, variations around the trend

should cancel each other out.

When using the proportional method, the averages

should sum to 4 overall. In the case of the
proportional method the values must be adjusted as
shown by multiplying by (4/3.98) as shown.
 Forecasting - Example
• You are required to forecast the sales leads for the
first quarter of 20X4 (quarter 17).
• The trend rose from 21.75 in Q1 20X2 to 25.75 in Q1
of 20X3 – a rise of 4.
• We can therefore estimate that Q1 of 20X4 will show
an increase of 4 to 29.75.
• The seasonal adjustment we calculated for Q1 was
1.29892.
• The seasonally adjusted forecast for Q1 20X4 is thus:
29.75 x 1.29892 = 38.64 (rounded)
 Time Series Example
Q ltd has prepared the following regression equation as
the basis for estimating sales (y) in units for period x.
Y = 27x - 24
Quarterly seasonal variations affecting the company’s
sales are as follows:
Q1 Q2 Q3 Q4
-25% -25% +15% +35%

Period 12 is in quarter 4. What are the forecast sales for

period 12?
 Solution
Expected sales = ((27 x 12) – 24) x 1.35
= 405 units

(Note: remember to increase the value by the seasonal

adjustment.)
 Further example
The overhead costs of Beta ltd have been found to be
accurately represented by the formula:
y = £10,000 + £0.25x
Where y is the monthly cost and x represents the level of activity in units.
Monthly activity levels can be estimated using a
combined regression analysis and time series model as
follows:
a = 100,000 + 30b
Where a represents the de-seasonalised monthly activity level and b
represents the month number.
If the seasonal index value for month 240 is 108, what
is the overhead cost for the month?
 Solution
Number of orders:
= (100,000 + (240 x 30)) x index value
= (100,000 + (240 x 30 )) x 1.08
= 115,776
The overhead cost is thus:
Y = £10,000 +(£0.25 x 115,776)
= £38,944
 Session Summary

In this session we have examined statistical approaches

to forecasting including:

• The hi-lo method.

• Regression analysis.
• Tests of reliability.
• Time series analysis.

40