This outcome asks you to judge when regression, expected values and time-series analysis give unreliable budget figures. The examiner likes it because it tests whether you can question a number instead of just calculating it.
The rule
Regression and time-series analysis both assume past patterns continue. Regression is only reliable within the range of data used, so forecasting outside it (extrapolation) is risky, and a high r² only shows fit within that range. Time-series analysis needs enough history and cannot predict random variations or suit new products. Expected value is a long-run average, so in a one-off event it may never occur and it hides the spread of outcomes and the downside risk.
Worked example
A firm's regression line is based on 10,000 to 50,000 machine hours, with r² = 0.96. The budget uses 85,000 hours.
Step 1: 85,000 is outside the 10,000 to 50,000 range, so this is extrapolation.
Step 2: r² = 0.96 shows the line fits the observed data well. It says nothing about behavior at 85,000 hours, where step costs or capacity limits may apply.
Step 3: Conclusion: the forecast is unreliable.
Expected value check: a launch has a 60% chance of $200,000 profit and a 40% chance of a $50,000 loss.
EV = (0.6 × $200,000) + (0.4 × –$50,000) = $120,000 – $20,000 = $100,000.
A one-off launch will give either $200,000 or –$50,000, never $100,000, so budgeting $100,000 hides the risk of loss.
Remember: a high r² only proves a good fit within the observed data, and an expected value is a long-run average that a single one-off event will never actually produce.
Where candidates lose marks
Treating a high r² as proof that a forecast outside the data range is reliable.
Calling EV useless because probabilities are subjective, when the real point is that a one-off outcome cannot equal the EV and risk is hidden.
Mistaking a strength of time-series analysis (separating trend from seasonal variations) for a limitation.
Try it: 4 questions
Sunbeam Co, a solar panel installer, used regression analysis on data covering 10,000 to 50,000 machine hours, giving r² = 0.96. The budget assumes 85,000 machine hours next year. Which TWO of the following conclusions are consistent with the limitations of regression analysis?
Select the two that apply (each shows on its own).
Answer: B, C. The regression line is based on 10,000 to 50,000 hours. Using it at 85,000 hours is extrapolation outside the observed range. Fixed costs may step up, capacity limits may bite or the cost per hour may change, so the linear relationship may not hold. A high r² measures how well the line fits the existing data, not whether it applies beyond that range. B and C are correct. A is the tempting error of relying on r² alone. D is wrong because a positive intercept says nothing about validity outside the range. E is wrong because a high r² does not make data unrealistic. It simply shows a good fit, and the problem is the extrapolation.
Weave Co, a textile mill, is considering a one-off product launch. There is a 60% chance of a $200,000 profit and a 40% chance of a $50,000 loss. The expected value is $100,000 and the finance director wants to include $100,000 profit in the budget. Which limitation of expected values does this best illustrate?
Answer: B. EV = (0.6 × $200,000) + (0.4 × –$50,000) = $120,000 – $20,000 = $100,000. EV is a long-run average over many repetitions. In a one-off event the outcome will be either $200,000 or –$50,000, never $100,000. EV also gives no indication of the spread of outcomes or the downside risk, so it suits a risk-neutral decision maker. B is correct. A overstates a weaker point. Subjectivity is a limitation, but it does not make EV useless and it is not what the scenario illustrates. C is false, because EV can be calculated for any number of outcomes. D is not a feature of this scenario, since no timing of cash flows is given.
Pixel Co, a mobile game studio, uses time-series analysis (trend and seasonal variations) to prepare its sales budget. Which of the following statements is NOT a valid limitation of the technique?
Answer: D. Statement D describes a strength of time-series analysis, not a limitation. It splits data into trend and seasonal components so that each can be projected into the budget. A is a genuine limitation because the technique extrapolates the past and assumes it will continue. B is a genuine limitation because random variations are unforeseeable. C is a genuine limitation because the technique needs enough historical data, so it cannot be used for new products. The common errors are rejecting C or A as non-limitations, or choosing B because it sounds unusual.
Press Co, a printing firm, budgets labour cost for a new product using an 80% learning curve. The workforce has high staff turnover, production is in small, irregular batches, and the process is largely machine-paced. Which option best identifies the main limitation of the learning-curve-based budget and the likely direction of error?
Answer: A. The learning curve assumes stable labour, continuous production and mainly manual work. Staff turnover means new workers start at the beginning of the curve, so learning is lost. Irregular batches mean skills are forgotten between runs. A machine-paced process leaves little scope for learning at all. The budget assumes cumulative average time keeps falling by 20% per doubling. Actual learning will be weaker, so actual hours will exceed the budget and the budget is too low. A is correct. B wrongly assumes the curve is prudent and that learning continues regardless. C wrongly says machine pacing speeds up learning, when it limits it. D identifies a limitation but wrongly says the direction is indeterminate. Lost learning pushes hours up.
0 more PM questions on this outcome, free for life
The full ACCA PM bank, mock exams and written questions marked against the scheme. Email only, no card, no trial.
What are the limitations of regression analysis in budgeting?
It assumes a linear relationship and that past cost behavior continues. Forecasts outside the observed data range are unreliable because of step costs or capacity limits, and r² only measures fit within the data used.
What are the limitations of expected values?
EV is a long-run average, so in a one-off decision the actual outcome will differ from it. It also gives no indication of risk or the spread of outcomes, and the probabilities used are often subjective.
What are the limitations of time-series analysis?
It assumes past trend and seasonal patterns will continue, it needs a reasonable amount of historical data, and it cannot predict random variations. It is therefore unsuitable for new products with no sales history.