Contents

1 High low method

2 Linear regression analysis

3 Time series analysis

4 The learning curve

5 Uncertainty in budgeting
 High low method

• Using quantitative aids in budgeting

• Estimating fixed and variable costs: high low method
• Using the high low method

1 High low method

1.1 Using quantitative aids in budgeting

Some quantitative methods might be used in budgeting. These include quantitative methods for:

• analysing fixed costs and variable costs per unit, and

• preparing forecasts, on the assumption that the future will continue on in the same trend
as in the past.

1.2 Estimating fixed and variable costs: high low method

Organisations that budget their expenditure using a marginal costing approach must be able to
estimate the fixed costs and the variable costs per unit in the budget period.

Direct materials costs and direct labour costs are usually treated as variable costs, and fairly
accurate estimates o€ these direct costs can often be prepared. However, it is often much more
difficult to analyse overhead costs, for production, administration and selling and distribution,
into fixed and variable overheads.

One assumption is that the fixed costs in each period and the variable cost per unit will continue
to be the same as in the past, costs can be analysed into fixed and variable components using.

• the high low method, or

• linear regression analysis.

1.3 Using the high low method

The high low method is a basic method to estimate the formula for a line:

y=a+bx

When estimating fixed and variable costs:

y = total costs
a = fixed costs in the period
b = variable cost per unit (or variable cost per $1 of sales)
x = number of units produced/ sold (or total sales)
The high low estimate is obtained by taking two historical records of total costs and the
associated total number of units. The two records used for the estimate are the costs for
the highest volume of activity (output or sales) and the costs for the lowest volume of
activity (output or sales) from amongst the records available.

• It is then assumed that these records of cost for the highest and lowest volumes
of activity are representative of costs at all levels of activity.
• The difference between the total cost at the high volume of activity and the total
cost at the low volume of activity must consist entirely of variable costs, because
the fixed costs are the same at both volumes of activity.
• The difference between the total cost-. a± the two levels of activity, divided by the
difference in activity level, gives us an estimate of the variable cost per unit.
• Having calculated a variable cost per unit, we can use either the high volume or
the low volume of activity and the total costs at that activity level to estimate the
total fixed costs
Example

A company is trying to estimate its fixed and variable production overhead costs in each
month. It has the following historical data of costs in the previous five months:

Production volume Total cost

000 units $000
8 25
6 22
6 19
9 24
5 16

Required

Use this data to estimate fixed costs each month and the variable cost per unit, using
the high low method.

What should be the estimated costs for production overheads next month if the
expected volume of production is 8,000 units?

Answer

With the high low method, the total cost for the highest volume of production and the
total cost for the lowest volume of production are used. These are the costs for 9,000
units and 5,000 units.

Since fixed costs are the same at both volumes of activity, the difference in total cost
between the high and the low volumes must be the variable costs of the difference in
output.
$
Total cost of 9,000 units 24,000
Total cost of 5,000 units 16,000
Therefore variable cost of 4,000 units 8,000
The variable cost per unit is therefore $ 8,000/4,400 units = $2 per unit.

This variable cost can be used, with either the high cost or the low cost, to estimate the fixed costs for the
period.
$
Total cost of 9,000 units 24,000

Variable cost of 9,000 units (x $2) 18,000

Therefore fixed costs each month 6,000

The estimate of costs using the high low method is therefore:

• fixed costs = $6,004

• variable cost= $2 per unit.

If estimated output is 8,000 units, the estimated total costs for the month are: $6,000 + (8,000 x $2) =
$22,000.
 Linear regression analysis
• The formulae
• Using the formulae
• Correlation
• The correlation coefficient r
• Linear regression analysis and forecasting

2 Linear regression analysis

2.1 The formulae

Linear regression analysis is a more accurate forecasting method than the high low method.
Like the high low method, it assumes that there is a straight-line formula y = a + bx. It also
uses historical data to produce an estimate for the values of a and b.

However, unlike the high low method, it uses any number of historical data items, not just
two (the high and the low data items).

The formulae for estimating fixed costs and variable costs with linear regression analysis
are given to you in the examination. You do not need to remember them, but you might be
required to use them.

The formulae are, for y = a + bx:

Σy bΣx
a= -
n n
nΣ x y-Σ x Σy
b=
nΣ x²-(Σx) ²

Where:
n = the number of data items used.

To calculate the value of a, you must fast know the value of b. The estimated value for b must
therefore be calculated first.

2.2 Using the formulae

Using the formula is probably best explained with an example. The following example is the
same as the example used earlier to demonstrate the high low method.
Example

A company is trying to estimate its fixed and variable production overhead costs in each month. It
has the following historical data of costs in the previous five months.

Production volume Total cost

000 units $000
8 25
6 22
6 19
9 24
5 16

Required

Use this data to estimate fixed costs each month and the variable cost per unit, using the linear
regression formulae.

What should be the estimated costs for production overheads next month if the expected volume of
production is 5,000 units?

Answer

The number of items of data: n = 5

The other values for the formulae are calculated as follows:

Production volume Total cost

X y xy x²
8 25 200 64
6 22 132 36
6 19 114 36
9 24 216 81
5 16 80 25
34 106 742 242

The values for y are the total costs and the values for x are the activity/output volumes.
Σx = 34
Σy = 106
Σxy = 742
Σx² = 242
N = 5

The formulae can now be used to establish a value for b and then a.
5(742)-(34)(106)
B=
5(242)-(34)²

3,710-3,604 106
= =
1,210-1,156 54
b= 1.96
106 1.96(34)
a= -
5 5
a=21.2-13.3=7.9

The estimate of costs is therefore = 7,900 + 1.96x.

When output x = 8,000 units, the estimated total costs will be:
Total costs = $7,900 + (8,000 X $1.96) = $23,580, say $23,600.
2.3 Correlation

The linear regression formulae can be applied to any data, where there are
pairs o€ data for x and y.
They do not indicate, however, how reliable the estimates might be. The
values for a and b that are estimated from linear regression analysis might
not be reliable at all. On the other hand, they might be very reliable.

Correlation refers to the extent to which values for y can be predicted from
any given value for x, using the values for a and b obtained from linear
regression analysis.
• A high degree of correlation indicates that values for y can be
estimated with a reasonable degree of confidence from values of x.
• A low degree of correlation indicates that values for y estimated
from any given value of x will not be particularly reliable.

The extent of correlation between values of x and values of y can be

measured by a correlation coefficient, r.

2.4 The correlation coefficient r

The correlation coefficient is calculated using the following formula:

n Σxy-ΣxΣy
R=
[n Σx² - (Σx)²] [n Σy² - (Σy)²]

This formula is given to you in the examination. You do not have to learn it, but you might be
required to use it.

If you look at the formula carefully, you should notice that most of the values in the formula are
the same as the values used to calculate b in the linear regression analysis formula. The only
additional values you'need are for n E y 2 and (Ey)2.
 The value of r

The value of r produced by this formula must always be within the range -1 to +
1.
● When r is close to + 1, there is a high degree of positive correlation. When
r = + 1, there is perfect positive correlation, and all the pairs of data for x
and y that have been used to estimate the values for a and b he on a
straight line, y = a + bx, when drawn graphically.
● When r is close to -1, there is a high degree of negative correlation. When r
= -1, there is perfect negative correlation, and all the pairs of data for x
and y that have been used to estimate the values for a and b he on a
straight line, y = a - bx, when drawn graphically.
● When r is 0, there is no correlation at all between the values of x and the
values of y, and the linear regression formula would be completely
unreliable. Correlation probably needs to be at the very least between 0.80
and 1.00 or between - 0.80 and -1.00 to be considered of much
significance.

Example

The correlation coefficient can be calculated for the previous example, as follows:
Production
Volume Total cost
X Y XY X² Y²
8 25 200 64 625
6 22 132 36 484
6 19 114 36 361
9 24 216 81 576
5 16 80 25 256
34 106 742 242 2.302

106 106
R= =
[54] [5(2,302) – (106)²] (54) (274)

106
r= =+0.87
121.6

Here, there is a reasonably strong positive correlation between the values of x and
the values of y that are estimated using the linear regression formula.

2.5 Linear regression analysis and forecasting

Linear regression analysis can also be used in forecasting, where it can be
assumed that there has been a linear trend in the past, and this same linear
trend will continue into the future.
 Exactly the same method is used in forecasting as for estimating fixed and
variable costs. The trend line is a formula y = a + bx, where x is the year or
month.
To simplify the arithmetic, you should number the years 1, 2, 3, 4 and so on (or
even start at year 0, and number the years 0, 1, 2, 3 and so on).

Time series analysis

▪ The nature of a time series

▪ Moving averages
▪ Seasonal variations with an additive model

▪ Using the trend line and seasonal variations to make forecasts

▪ Problems with seasonal variation analysis

▪ Estimating seasonal variations with the proportional model

3 Time series analysis

3.1 The nature of a time series

A time series is a record of data over a period of time. A problem arises with
forecasting when a time series fluctuates. If the fluctuation is regular then this
can be included in forecasts to make these more accurate.

In time series analysis it is assumed that actual data is made up of four

components:

▪ A trend ('I') - This is the long term underlying movement of the data. If this
is linear it can be forecast using the high low method, linear regression
analysis, by plotting the data and estimating a line of best fit or by
using moving averages which will be explained later in the chapter.
 ▪ A cyclical variation (C) - This is a long-term regular fluctuation around the
trend and is caused by the economic cycle of boom, recession and
recovery.
▪ A seasonal variation (S) - This is the short-term regular fluctuation around
the trend. Examples include actual sales which vary by the quarter of the
year
(skiing holidays, ice cream sales), usage which varies by the time of the day
(electricity, telephone), usage which varies by the days of the week (rail
transport, hotel accommodation).

▪ A random variation (R) - Unpredictable affects like strikes, adverse

weather conditions or stock market crashes.

In time series analysis, cyclical variations are often ignored as they require too
much data to analyse. Random variations are, by definition, unpredictable so
the time series model concentrates on identifying the trend and seasonal
variations which affect data. These can then be used to make forecasts.

There are two different models used;

▪ The additive model

Actual data = T + S

▪ The proportional model

Actual data = T X S

The additive model assumes that a seasonal variation is an absolute figure

which can be added or subtracted to the trend each period to find the actual
data. The proportional model assumes that the seasonal variation is an index or
percentage which is multiplied by the trend to find the actual data. The
proportional model is often considered to be more accurate when considering
sales or production data when the trend is increasing.

3.2 Moving averages

Moving averages can be used to estimate help estimate the trend and seasonal
variation. This is done as follows:

Step 1. Decide the length of the cycle. For example, the cycle will be seven
days when historical data is collected daily for each day of the week. The cycle
will be one year when data is collected monthly for each month of the year, or
quarterly for each season.
Step 2. Use the historical data to calculate a series of moving averages. A
moving average is the average of all the historical data in one cycle. For
example, suppose that historical data is available for daily sales over a period
Day 1- Day 21, and there are seven days of selling each week. A moving
average can be calculated for Day 1- Day 7. Another moving average can be
calculated for Day 2 - Day 8. Another moving average can be calculated for
Day 3 - Day 9, and so on up to a moving average for Day 15 - Day 21.
Step 3. Match each moving average with an actual time period. The moving
average should be matched with the middle time period of the cycle. For
example a moving average for Day 1 - Day 7 is matched with Day 4, which is
the middle of the period. Similarly, a moving average for Day 2 - Day 8 is
 matched with Day 5, and a moving average for Day 15 - Day 21 is matched
with Day 18.
Step 4. Use the moving averages (and their associated time periods) to
calculate a trend line, using simple averaging, high low method or linear
regression analysis. It is also often useful to plot the data on a graph and
extend a line of best fit.

Example
A company operates for five days each week. Sales data for the most recent three weeks are as follows:
Sales Monday Tuesday Wednesday Thursday Friday
Units Units Units Units Units

Week 1 78 83 89 85 85
Week 2 88 93 99 95 95
Week 3 98 103 109 105 105

For convenience, it is assumed that Week 1 consists of Days 1-5, Week 2 consists of Days 6-10, and Week 3
consists of Days 11-15.
This sales data can be used to estimate a trend line. A weekly cycle in this example is 5 days, some must calculate
moving averages for five day periods, as follows:

Period Middle day Moving average

Days 1-5 Day 3 [78+83+89+85+85]/5 84

Days 2-6 Day 4 [83+89+85+85+88]/5 86
Days 3-7 Day 5 [89+85+85+88+93]/5 88
Days 4-8 Day 6 [85+85+88+93+99]/5 90
Days 5-9 Day 7 [85+88+93+99+95]/5 92
Days 6-10 Day 8 [88+93+99+95+95]/5 94
Days 7-11 Day 9 [93+99+95+95+98]/5 96
Days 8-12 Day 10 [99+95+95+98+103]/5 98
Days 9-13 Day 11 [95+95+98+103+103]/5 100
Days 10-14 Day 12 [95+98+103+109+105]/5 102
Days 11-15 Day 13 [98+103+109+105+105]/5 104

In this example, all the moving average figures lie on a perfect straight line. It can be seen that each day the trend
increases by 2. If x = the day number, the formula for the trend can be calculated by taking any day, say day 12
A + 2+ x 12 = 102 soa = 78.
The formula is daily sales = 78 + 2x.
This trend line can be used to calculate the seasonal variations. The approach will be slightly different depending on
whether an additive or proportional model is being used.
3.3 Seasonal variations with an additive model
The trend line on its own is not sufficient to make forecasts for the future. We
need estimates of the size of the 'seasonal' variation for each o£ the different
seasons. In the example above, we need an estimate of the amount of the
expected d variation in sales, for each day of the week.

For the purpose of forecasting, the seasonal variation (in the above example,
daily variation) is the difference between:
▪ the trend line value, and
▪ the forecast or expected value (allowing for the seasonal variation).

The size of the seasonal variations is estimated from the historical data. Seasonal
variations for each quarter of the year or each day of the week can be estimated
from the difference between:

▪ the actual historical value for each time period, and

▪ the moving average value for the same time period

The seasonal variation for each season (or daily variation for each day) is
estimated as follows:

▪ Use the moving average values that have been calculated from the
historical data, and the corresponding historical data. ('actual' data)
-
for the same period.
▪ Calculate the difference between the moving average value and the a
historical figure for each time period. This is a seasonal variation.
You will now have a number of seasonal variations, covering several
weekly or annual cycles.:
▪ Group these seasonal variations into the different seasons of the year (or
days the week). You will now have several seasonal variations for
each day of week or season of the year.
▪ For each season (or day), calculate the average of these seasonal
variations.
▪ This average seasonal variation for each day of the week or season of the
year used as the seasonal variation for the purpose of forecasting.

The seasonal variations can then be used, with the estimated trend line, to
forecasts for the future.
Example
Using the previous example, the seasonal variations are calculated as follows:

Variation
Day of the Moving average (Actual -
Middle day Actual sales
week value moving
average)
Day 3 Wednesday 84 89 +5
Day 4 Thursday 86 85 -1
Day 5 Friday 88 85 -3
Day 6 Monday 90 88 -2
Day 7 Tuesday 92 93 +1
Day 8 Wednesday 94 99 +5
Day 9 Thursday 96 95 -1
Day 10 Friday 98 95 -3
Day 11 Monday 100 98 .2
Day 12 Tuesday 102 103 +1
Day 13 Wednesday 104 109 +5

The seasonal variation (daily variation) is now calculated as the average seasonal variation for each day, as follows:

Monday Tuesday Wednesday Thursday Friday

Variation
Units units units units Units
Week 1 +5 -1 -3
Week 2 -2 +1 +5 -1 -3
Week 3 -2 +1 +5
Average -2 +1 +5 -1 -3

Points to note

(1) In this example, the average seasonal variation for each day of the week is exactly the
same as the actual seasonal variations. This is because the historical data in this example
produces a perfect trend line.

(2) The total of the seasonal variations for each day of the week is 0. (- 2 + 1 + 5 -13 = 0.)
When seasonal variations are applied to a straight-line trend line, they must always add
up to zero. If the seasonal variations did not add up to 0, the trend line would not be
straight. It would 'curve' up or down, depending on whether the sum of the seasonal
variations is positive or negative.

3.4 Using the trend line and seasonal variations to make forecasts
 When the trend line and seasonal variations have been estimated, we can make forecasts
for the future. In the example above, suppose that we wanted to forecast sales in Week
4 (days 16 - 20). The trend line is 78 + 2x. The daily forecasts are as follows:

Day Trend line value Seasonal Forecast

(78 + 2x) Variation

16 Monday 110 -2 108

17 Tuesday 112 +1 113
18 Wednesday 114 +5 119
19 Thursday 116 -1 115
20 Friday 118 -3 115

3.5 Problems with seasonal variation analysis

There are two problems with using this technique to make forecasts with seasonal
variations.

(1) The historical data will not provide a perfect trend line. The trend line that is estimated
with the historical data is a `best estimate', not a perfect estimate. The seasonal
variations measured from historical data will not be a constant amount for each day
of the week or. season of the year. In practice, seasonal variations are estimated as
average values from the historical data.

(2) The number of seasons in each cycle might be an even number (2, 4, 6 or 8). A moving
average is an average for the middle time period in the cycle. When there is an
even number of periods in one cycle, the moving average does not match any specific
time period. For example, the moving average of Season 1 - Season 4 is for Season
2.5, which does not exist.

There are techniques to deal with each of these problems.

Historical data does not produce a perfect trend line

When historical data does not produce a perfect trend line, the same technique is used to
prepare forecasts, with one difference.

The sum of the seasonal variations must be 0 for a straight trend line, but the actual seasonal
variations calculated from the historical data will not add up to 0. They must therefore be
adjusted so that they do add up to 0.

Example

Sales Monday Tuesday Wednesday Thursday Friday

Units Units Units Units
Units Week 1 55 49 54 60
65 Week 2 70 65 65 76
 80 Week 3 88 81 85 91
93

For convenience, it is assumed that Week 1 consists of Days 1- 5, Week 2 consists of Days 6 -
10, and Week 3 consists of Days 11-15.

A trend line is calculated using linear regression analysis, giving the following forecast of daily
sales:

Sales = 45 + 3x.
The actual sales and moving average values of sales are compared, to produce
seasonal variations for each day, as follows:

Day 3 4 5 6 7 8 9 10 11 12 13
Moving avg 56 59 5 64 68 71 75 78 82 85 88
Actual 54 60 62 70 65 65 76 80 88 81 85
Variation -2 +1 +3 +6 -3 -6 +1 +2 +64 -4 -3

The average variation for each day is now calculated, but they sum of the
averages in not 0. It is 2.33. To reduce this total of seasonal variations to 0, we
can reduce the average seasonal variation per day (for five days) by 2.33/5 =
0.466, say 0.5. (Remember that minus a minus figure is `plus'.)

Variation Monday Tuesday Wednesday Thursday Friday Total

Units Units Units Units
Week 1 -2 +1 +3
Week 2 +6 -3 -6 +1 +2
Week 3 +6 -4 -3
Average +6 -3.5 -3.67 +1 +2.5 +2.33
Adjustment -0.5 -0.5 -0.5 -0.5 -0.5
Seasonal +5.5, say +5 -4 -4.17, say -4 +0.5, say +1 +2
variation

The estimated variations each day of the week are therefore:

Monday + 5, Tuesday - 4, Wednesday - 4, Thursday + 1 and Friday + 2 -

These add up to zero, so can be used with the trend line to make forecasts of
future sales based on an assumption that sales will rise in a straight line, with
daily variations according to the day of the week.

An even number of seasons

When there is an even number of seasons in a cycle, and the moving averages do
not correspond to an actual season, it is necessary to take -moving averages of
the moving averages. These will correspond to an actual season of the year.

Example

The following sales figures will be used to estimate a trend line for quarterly sales
with seasonal variations:
Sales Quarter 1 Quarter 2 Quarter 3 Quarter 4
$000 $000 $000 $000
Year1 20 24 27 31
 Year2 35 39 44 47
Year3 49 56 60 64

These quarters for the three years will be called quarter 1 - Quarter 12. There are four seasons in the annual cycle, so
moving average values for each quarter are calculated as follows:

Moving average
Period Middle quarter Moving average of moving average
(average of 2)

Quarters 1 -4 Quarter 2.5 25.50

Quarter 3 27.375
Quarters 2 - 5 Quarter 3.5 29.25
Quarter 4 31.125
Quarters 3 - 6 Quarter 4.5 33.00
Quarter 5 35.125
Quarter; 4 -7 Quarter 5.5 37.25
Quarter 6 39.250
Quarters 5 - 8 Quarter 6.5 41.25
Quarter 7 43.000
Quarters 6 - 9 Quarter 7.5 44.75
Quarter 8 46.875
Quarters 7 -10 Quarter 8.5 49.00
Quarter 9 51.000
Quarters 8 -11 Quarter 9.5 53.00
Quarter 10 55.125
Quarters 9 -12 Quarter 10.5 57.25

The moving averages in the right hand column correspond with an actual season. These moving
averages are used to estimate the trend line and the seasonal variations.

3.6 Estimating seasonal variations with the proportional model

A proportional time series model uses the same approach as the additive model to find the trend.
The seasonal variation is then calculated by dividing the actual data by the trend. (Since A = T X
S, then S = A/ T)

Example
A company has used sales records for each quarter of the year for the past few years to prepare
an estimate of the trend in sales each quarter.
The straight-line trend in quarterly sales is

y =150,000 + 10,000x

where
x = the quarter in the time series, where 0 - the sales in the first quarter of 20X1.

The actual and trend sales in each quarter are shown in the table below. The seasonal variation
is found by dividing the actual sales by the trend.

Estimated sales, using trend Seasonal variation =

Time period Actual sales line y - 150,000 + 10,000x AIT

Quarter 1, 20X1 119,200 150,000 0.79

Quarter 2, 20X1 167,500 160,000 1.05
Quarter 20X1 219,200 170,000 1.29
3, 4, 20X1
Quarter 151,500 180,000 0.84
Quarter 1, 20X2 162,000 190,000 0.85
Quarter 2, 20X2 203,700 200,000 1.02
Quarter 3, 20X2 258,400 210,000 1.23
Quarter 4, 20X2 199,700 220,000 0.91
Quarter 1, 20X3 200,400 230,000 0.87
Quarter 2, 20X3 248,000 240,000 1.03
Quarter 3, 20X3 296,200 250,000 1.18
Quarter 4, 20X3 237,200 260,000 0.91
Quarter 1, 20X4 242,000 270,000 0.90
Quarter 2, 20X4 284,800 280,000 1.02
Quarter 3, 20X4 341,200 290,000 1.18
Quarter 4, 20X4 276,400 300,000 0.92

These figures can be used to calculate an average seasonal variation for each quarter of the year.
However, the total seasonal variations must add up to 4, otherwise a straight-line trend does not
exist. If the average does not sum to 4 an adjustment must be made to the average.

Seasonal
variations
Quarter 1 Quarter 2 Quarter 3 Quarter 4

20X1 0.79 1.05 1.29 0.84

20X2 0.85 1.02 1.23 0.91
20X3 0.87 1.03 1.18 0.91
20X4 0.9 1.02 1.18 0.92
Total 3.41 4.12 4.88 3.58
Average 0.85 1.03 1.22 0.9

The trend and seasonal variations can now be used to forecast future sales.

Sales forecast for 20X5

Sales
Estimated sates, using trend Seasonal
Time period forecast
line y -150,000 + 10,000x variation
TxS

Quarter 1, 20X5 (x = 16) 310,000 0.85 263,500

Quarter 2, 20X5 (x =17) 320,000 1.03 329,600
Quarter 3, 20X5 (x =18) 330,000 1.22 402,600
Quarter 4, 20X5 (x =19) 340,000 0.9 306,000

The reliability of forecasts of trends and seasonal variations

Forecasts using estimated trends and seasonal variations are based on the assumptions that the
past is a reliable guide to the future. This assumption may be incorrect. However, forecasts must
be made; otherwise it is impossible to make plans beyond the very short term.

The learning curve

■ Learning curve theory

■ The learning curve model

■ Graph of the learning curve

■ Formula for the learning curve

■ Conditions for the learning curve to apply

4 The learning curve

4.1 Learning curve theory

When a workforce begins a task for the first time, and the task then becomes
repetitive, it will probably do the job more quickly as it learns. It will find
quicker ways of performing tasks, and will become more efficient as knowledge
and understanding increase.

When a task is well-established, the learning effect wears out, and the time to
complete the task becomes the same every time the task is carried out.
However, during the learning period, the time to complete each subsequent task
can fall by a very large amount.

The learning curve effect was first discovered in the US during the 1940s, in
aircraft manufacture. Aircraft manufacture is a highly-skilled task, where:
• the skill of the work force is important, and
• the labour time is a significant element in production resources and
production costs.

The time taken to produce the first unit of a new model of an aeroplane might
take a long time, but the time to produce the next unit is much less, and the
time to produce the third is even less, and so on. Labour times per extra unit
therefore fall.

This has important implications for:

• budgeting/ forecasting production requirements and production costs
• pricing: prices calculated on a 'cost plus' basis can allow for future cost
savings.

4.2 The learning curve model

 The effect of the learning curve can be predicted mathematically, using a
learning curve model. This model was developed from actual observations and
analysis in the US aircraft industry.

The learning curve is measured as a percentage learning curve effect. For

example, for a particular task, there might be an 80% learning curve effect, or a
90% learning curve effect, and so on.
When there is an 80% learning curve, the cumulative average time to produce
units of an item is 80% of what it was before, every time that output doubles.
▪ The cumulative average time per unit is the average time for all the
units made so far, from the first unit onwards.

▪ This means, for example, that if an 80% learning curve applies, the
average time for the first two units is 80% of the average time for
the first unit. Similarly, the average time for the first four units is
80% of the average time for the first two units.

Example

T h e t i m e t o make a new model of a sailing boat is 100 days. It has been

established that in the boat-building industry, there is an 80% learning curve.

Required

Calculate:
(a) the cumulative average time per unit for the first 2 units, first 4 units,
first 8 units and first 16 units of the boat
(b) the total time required to make the first 2 units, the first 4 units, the first
8 units and the first 16 units
(c) the additional time required to make the second unit, the 3rd and 4 t h
units, units 5 - 8 and units 9 -16.

Answer
Cumulative Average time
for
Total Units Average time Total time for Incremental time additional units
(cumulative) per unit all units for additional units
days days days days
1 100 100 100
2 80 160 60 60
4 64 256 96 48
8 51.2 409.6 153.6 38.4
16 40.96 655.36 245.76 30.72

Example
 The first unit of a new model of machine took 1,600 hours to make. A 90%
learning curve applies. How much time would it take to make the first 32 units of
this machine?
Answer

Average time for the first 32 units =1,600 hours x 90% x 90% x 90% x 90% x
90% = 944.784 hours
Total time for the first 32 units = 32 x 944.784 hours = 30,233 hours.

4.3 Graph of the learning curve

The learning curve can be shown as a graph. There are two graphs following.

The left-hand graph shows the cumulative average time per unit. This falls rapidly at first, but
the learning effect eventually ends and the average time for each additional unit becomes
constant (a standard time). This is known as the steady state.

The right hand graph shows how total costs increase. The total cost line is a curved
line initially, because of the learning effect.

4.4 Formula for the learning curve

The learning curve is represented by the following formula (mathematical model):

Learning curve: y = axb

Where
y = the cumulative average time per unit for all units made
x = the number of units made so far (cumulative number of units) a = the time for the first unit
b = the learning factor.

The learning factor b = Logarithm of learning rata

Logarithm of 2

The learning rate is expressed as a decimal, so if the learning curve is 80%, the learning factor is:
(logarithm 0.80/logarithm 2)

To use this formula you must be able to calculate logarithms. Make sure that you know how to
use the logarithms function on your calculator.
Example

If there is an 80% learning curve, the learning factor is calculated as follows:

Logarithm 0.80 - - 0.09691= _ 0.32193

Logarithm 2 0.30103

The learning curve formula is therefore: y = ax -0.32193

1
It might help to remember that x-032913 is another way of writing
x-0.32193
Going back to the previous example, the cumulative average time to produce 8 units can
therefore be calculated as:
I
Y =100X 80.32193

=100 (0.512)

= 51.20

Example

It will take 500 hours to complete the first unit of a new product. There is a 95 learning

curve effect.

Calculate how long it will take to produce the 7h unit.

Answer

The time to produce the seventh unit is the difference between:

▪ the total time to produce the first 6 units, and
▪ the total time to produce the first 7 units.
(1)Learning factor
Logarithm 0.95 = -0.02227639 = 0.074
Logarithm 2 0.30103

(2) Average time to produce the first 6 units

y =500x 1
 6 0.074

= 500 (0.8758239)

= 437.9 hours per unit

(3) Average time to produce the first 7 units

y = 500 x 1
70.074

= 500 (0.86589)

= 432.9 hours per unit

(4) Time to produce the 7th unit

Total time for the first 7 unfts (7 x 432.9) 3,030.3

Total time for the first 6 units (6 x 437.9) 2,627.4
Time for the 7th unit 402.9

4.5 Conditions for the learning curve to apply

The learning curve effect will only apply in the following conditions:

▪ There must be stable conditions for the work, so that learning can take
place. For example, labour turnover must not be high; otherwise
the learning effect is lost. The time between making each subsequent
unit must not be long; otherwise the learning effect is lost because
employees will forget what they did before.

▪ The activity must be labour-intensive, so that learning will affect the time
to complete the work.

▪ There must be no change in production techniques, which would require

the learning process to start again from the beginning.

▪ Employees must be motivated to learn.

In practice, the learning curve effect is not used extensively for budgeting or
estimating costs (or calculating sales prices oil a cost plus basis). In a modern
manufacturing environment production is highly mechanised and therefore
the learning curve effect does not apply.
 Uncertainty in budgeting

▪ The nature of uncertainty in budgeting

▪ Flexible budgets

▪ Probabilities and expected values

▪ Spreadsheets and 'what if' analysis

5 Uncertainty in budgeting

5.1 The nature of uncertainty in budgeting

Uncertainty arises when there is a lack of reliable information. In budgeting, there is uncertainty
because estimates and forecasts may be unreliable. Information is almost never 100% reliable (or
'perfect'), and some uncertainty in budgeting is therefore inevitable.

Risk arises in business because actual events may turn out better or worse than expected. For
example, actual sales volume may be higher or lower than forecast. The amount of risk in
business operations varies with the nature of the operations. Some operations are more
predictable than others. The existence of risk means that forecasts and estimates in the budget,
which are based on expected results, may not be accurate.

Both risk and uncertainty mean that estimates and forecasts in a budget are likely to be wrong.

Management should be aware of risk and uncertainty when preparing budgets and when
monitoring performance.
▪ When preparing budgets, it may be appropriate to look at several different forecasts and
estimates, to assess the possible variations that might occur. In other words,
managers should think about how much better or how much worse actual results
may be, compared with the budget.

▪ When monitoring actual performance, managers should recognise that adverse or

favourable variances might be caused by weaknesses in the original forecasts,
rather than by good or bad performance.

Several approaches may be used for analysing risk and uncertainty in budgets.. These include:
▪ flexible budgets
▪ using probabilities and expected values
▪ using spreadsheet models and 'what if' analysis (sensitivity analysis).
5.2 Flexible budgets

Flexible budgets may be prepared during the budget-setting process. A flexible budget is a
budget based on an assumption of a different volume of output and sales than the volume of
output and sales then the volume in the master budget or fixed budget,.
For example, a company might prepare its master budget on the basis of estimated sales of
$100 million. Flexible budgets might be prepared on the basis that sales will be higher or lower --
say $80 million, $90 million, $110 million and $120 million. Each flexible budget will be
prepared on the basis of assumptions about fixed and variable costs, such as increases or
decreases in fixed costs if sales rise above or fall below a certain amount, or changes in variable
unit costs above a certain volume of sales.

During the financial year covered by the budget, it may become apparent that actual sales and
production volume will be higher or lower than the fixed budget forecast. In such an event,
actual performance can be. compared with a suitable flexible budget.

Flexible budgets can be useful, because they allow for the possibility that actual activity
levels may be higher or lower than forecast in the master budget. The main disadvantage of flexible
budgets could be the time and effort needed to prepare them. The cost of preparing them
could exceed the benefits of having the information that they provide.

5.3 probabilities and expected values

Estimates and forecasts in budgeting may be prepared using probabilities and expected
values. An expected value is a weighted average value calculated with probabilities.

Example
A company is preparing a sales budget. The budget planners believe that the volume of
sales next year will depend on the state of the economy.

State of the economy Sales for the year

$ million
No growth 40
Low growth 50
Higher growth 70

It has been estimated that there is a 60% probability of no growth, a 30% probability of low
growth and a 10% probability of higher growth.

The expected value (EV) of sales next year could be calculated as follows:

State of the economy Sales for the year probability EV of sales

 $ million $ million
No Growth 40 0.6 24
Low Growth 50 0.3 15
Higher Growth 70 0.1 7
EV Growth 46

The company might decide to prepare a sales budget on the assumption that annual
sales will be $46 million.

The problems with taping probabilities and expected values

There are two problems that might exist with the use of probabilities and expected values:
▪ The estimates of probability might be subjective, and based on the judgement or
opinion of a forecaster. Subjective probabilities, might be no better than educated
guesses. Probabilities should have a rational basis.
▪ An expected value is. most useful when, it is a weighted average value for an
outcome that will happen many times in the planning period. If the forecast event
happens many times in the planning period, weighted average values are suitable for
forecasting. However, if an outcome will only happen once, it is doubtful whether an
expected value has much practical value for planning purposes.

This point can be illustrated with the previous example of the EV of annual sales. The
forecast is that sales will be $40 million (0.60 probability), $50 million (0.30 probability) or
$70 million (0.11y probability). The EV of sales is $46 million.

▪ The total annual sales for the year is an outcome' that occurs only once. It is doubtful
whether it would be appropriate to use $46 million as the budgeted sates for the year.
A sales total . of $4Gndtbw is not expected to happen.

▪ It might be more appropriate to prepare a: fixed budget 6n the basis that sales will be
$40 million (the most likely outcome) and prepare flexible budgets for sales of $50
million and $70 million. . .

When the forecast outcome happens many times in the planning period, an EV might be
appropriate. For example, suppose that the forecast of weekly sales of a product is as follows:

Weekly sales Probability EV of weekly sales

$ $
7,000 0.5 3,500
9,000 0.3 2,700
12,000 0.2 2,400
8,600
Since there are 52 weeks in a year, it would be appropriate to assume that weekly sales will
be a weighted average amount, or EV. The budget for annual sales would be (52 x $8,600) =
$447,200. If the probability estimates are fairly reliable, this estimate of annual sales should
be acceptable as the annual sales budget.

5.4 Spreadsheets and `what if analysis

 Preparing budgets is largely a `number crunching' exercise, involving large amounts of
calculations. This aspect of budgeting was made much easier, simpler and quicker with IT
and the development of computer-based models for budgeting. Spreadsheet models, or similar
planning models, are now widely used to prepare budgets.

A feature of computer-based budget models is that once the model has been constructed, it
becomes a relatively simple process to prepare a budget. Values are input for the key
variables, and the model produces a complete budget.

Amendments to a budget can be made quickly. A new budget can be produced simply by
changing the value of one or more input variables in the budget model.

This ability to prepare, new budgets quickly by changing a small number of values in the
model also creates opportunities for sensitivity analysis and stress testing. The budget planner, can
test how the budget will be affected if forecasts and estimates are changed by asking 'what
if, questions. For example:

▪ What if sales volume is 5% below the budget forecast?

▪ What if the sales mix of products is different?

▪ What if the introduction of the new production system or the new IT system is
delayed by six months?

▪ What if interest rates go up by 2% more than expected?

▪ What if; the fixed costs are 5% higher and variable costs per unit are 3% higher?

The answers to 'what if questions can help budget planners to understand more about the risk
and uncertainty in the budget, and the extern to which actual results might differ from the
expected outcome in the master budget. This can provide valuable information for risk
management, and management can assess the 'sensitivity' of their budget to particular
estimates and assumptions.