Accounting › Section P · Managerial accounting and cost behavior › Topic P.6

Estimating a cost equation

A manager planning next month needs to know what costs will be at the expected level of activity. A cost equation answers that, and when a bill mixes fixed and variable parts, past records can be used to estimate it with a scatter graph, the high-low method or regression.

The cost equation

"Variable, fixed, and mixed costs" showed that a mixed cost has a fixed part and a variable rate. Written as an equation:

Y = a + bX

Here \(Y\) is total cost for the period, \(a\) is the fixed part, \(b\) is the variable rate per unit of activity, and \(X\) is the level of activity measured by the cost driver. On a graph, \(a\) is where the line meets the cost axis and \(b\) is its slope. Once \(a\) and \(b\) are known, a manager can predict cost at any planned activity inside the relevant range, which is the basis of the budgets and break-even analysis in later sections.

Building it from known costs

When every cost's behaviour is already known, the equation is built by adding. Thrushwick Laundry, a fictional commercial laundry serving hotels, pays rent of $5,200 and a supervisor's salary of $3,800 a month, so \(a = 9{,}000\). Detergent costs $0.30 per kilogram washed and water and drying $0.12 per kilogram, so \(b = 0.42\) per kilogram. The equation is \(Y = 9{,}000 + 0.42X\). At 20,000 kilograms, predicted cost is \(9{,}000 + 0.42 \times 20{,}000 = 17{,}400\) dollars. Often, though, a cost arrives as a single bill with no split, such as Thrushwick's utility bill. Then the parts must be estimated from past records.

Choosing the driver and the data

Start by choosing an activity that plausibly causes the cost. Thrushwick's machines draw most of its electricity and gas, so machine hours are a sensible driver for the utility bill. Then gather pairs of observations, here monthly hours and the matching bill, over periods of equal length. Clean the data first: remove or adjust months distorted by one-off events, such as a burst pipe or a billing error, and be wary of mixing years if prices have changed. Eight months of Thrushwick's records:

\begin{array}{lrr} & \text{Hours} & \text{Cost} \\ \hline \text{Jan} & 1{,}200 & 7{,}150 \\ \text{Feb} & 1{,}550 & 7{,}900 \\ \text{Mar} & 980 & 6{,}400 \\ \text{Apr} & 1{,}400 & 7{,}450 \\ \text{May} & 1{,}700 & 8{,}200 \\ \text{Jun} & 1{,}100 & 6{,}850 \\ \text{Jul} & 1{,}300 & 7{,}300 \\ \text{Aug} & 1{,}620 & 8{,}300 \end{array}

The scatter graph

A scatter graph plots each observation as a point, with activity on the horizontal axis and cost on the vertical axis. It is a diagnostic tool, used before any calculation. If the points lie roughly along a straight, upward-sloping band, a linear cost equation is reasonable. If they show no pattern, the chosen driver probably does not cause the cost, and another should be tried. If they curve, a straight line will mislead at the extremes. A point far from the others, an outlier, should be investigated before it is used. Thrushwick's eight points rise in a tight band with no outliers, so a straight line suits them.

The high-low method

The high-low method estimates the line from just two observations: the periods with the highest and lowest activity. Note that it is activity, not cost, that picks the two points.

b = \frac{\text{cost at high} - \text{cost at low}}{\text{high activity} - \text{low activity}}

The change in cost between the two points can only come from the variable part, because the fixed part is the same in both. Then find the fixed part by substituting either point: \(a = Y - bX\). The method is quick and needs no software, but it ignores every other observation, so if either extreme month was unusual, the whole equation inherits the distortion.

Worked example: high-low at Thrushwick

1. Choose the points by activity: the high is May, 1,700 hours and $8,200; the low is March, 980 hours and $6,400. August had the highest cost, $8,300, but not the highest activity, so it is not used. 2. Variable rate, in dollars per machine hour:

b = \frac{8{,}200 - 6{,}400}{1{,}700 - 980} = \frac{1{,}800}{720} = 2.50

3. Fixed part, from May: \(a = 8{,}200 - 2.50 \times 1{,}700 = 3{,}950\) dollars. 4. Check with March: \(3{,}950 + 2.50 \times 980 = 6{,}400\), as recorded. 5. Equation: \(Y = 3{,}950 + 2.50X\). 6. Prediction for a month of 1,500 hours: \(3{,}950 + 2.50 \times 1{,}500 = 7{,}700\) dollars.

Least-squares regression

Least-squares regression fits the straight line that makes the sum of the squared vertical distances from every point to the line as small as possible. Unlike high-low, it uses all the observations, so one odd month has less influence. Spreadsheets do the arithmetic: in Excel, for example, the SLOPE and INTERCEPT functions return \(b\) and \(a\). Regression also reports \(R^2\), the coefficient of determination, which gives the share of the variation in cost that the activity explains. It runs from 0 to 1: values near 1 mean the points cluster closely around the line, and low values mean the driver explains little and predictions will be rough.

Worked example: regression at Thrushwick

Running a regression on all eight months gives, after rounding, \(a = 3{,}999\) dollars, \(b = 2.54\) dollars per hour and \(R^2 = 0.97\). 1. Interpret \(b\): each extra machine hour adds about $2.54 to the utility bill. 2. Interpret \(a\): about $3,999 a month is the fixed part within the range observed. 3. Interpret \(R^2\): machine hours explain about 97% of the month-to-month variation in the bill, a strong fit. 4. Predict 1,500 hours: \(3{,}999 + 2.54 \times 1{,}500 = 7{,}809\) dollars. 5. Compare: high-low predicted $7,700, about $109 less, because its line was drawn through only two points. Regression is preferred when the data are available; high-low is a quick first estimate.

Using the equation wisely

Every estimate rests on assumptions. The equation describes the past, so adjust it for known changes, such as a new energy tariff. It holds only within the relevant range observed, here roughly 980 to 1,700 hours; predicting 3,000 hours would assume the same equipment and prices far beyond any month in the data. The intercept is not necessarily the cost at zero activity: Thrushwick never closed, so $3,999 is simply where the fitted line meets the axis. And correlation is not causation, so the driver should make operational sense, not merely fit the numbers.

Common mistakes to avoid

First, choosing the high and low points by cost instead of by activity; using August's $8,300 would give the wrong slope. Second, computing the variable rate correctly but subtracting it from the wrong point, or forgetting to multiply it by activity, when finding the fixed part. Third, using months distorted by one-off events. Fourth, predicting far outside the relevant range. Fifth, treating a high \(R^2\) as proof that the driver causes the cost, or a low one as harmless.

Key takeaways

A cost equation, Y = a + bX, splits total cost into a fixed part a and a variable rate b per unit of the cost driver X. Build it by adding known costs, or estimate it from past data. A scatter graph checks first that the relationship looks linear and reveals outliers. The high-low method takes the highest and lowest activity periods: b is the change in cost divided by the change in activity, and a = Y − bX. Least-squares regression uses every point and reports R², the share of cost variation explained. Use the equation only within the relevant range and adjust it for known changes.

Flashcards

Flashcard

Y = a + bX

Total cost = fixed part + variable rate × activity

a is the intercept, b the slope.

Test your understanding

Question 1 of 5

In its lowest-activity month a firm made 2,000 units at a cost of $15,000; in its highest-activity month it made 5,000 units at $24,000. Using the high-low method, what is the cost equation?

Question 2 of 5

Using Y = 9,000 + 3.00X, what cost is predicted for 4,000 units, within the relevant range?

Question 3 of 5

When using the high-low method, how should the two observations be chosen?

Question 4 of 5

A regression of maintenance cost on machine hours gives Y = 12,400 + 1.85X with R² = 0.91. What do these results say?

Question 5 of 5

Why is least-squares regression usually preferred to the high-low method?

← Previous topicNext topic →