BAR · Business Analysis · 6 practice questions
Calculate Predicted Cost Using Simple Linear Regression
In Y = a + bX, predicted cost equals estimated fixed cost plus the variable rate times activity. Below: a worked calculation showing how to turn regression coefficients into a monthly cost forecast.
Try one first
Hint
Linear regression Y = a + bX for cost behavior: a (intercept) = FIXED cost component; b (slope) = VARIABLE cost per unit of activity. Decomposes mixed cost into fixed + variable. Valid within relevant range; check R² for fit quality.
Answer B. LINEAR REGRESSION FOR COST BEHAVIOR: the equation Y = a + bX decomposes a MIXED COST into FIXED and VARIABLE components. INTERCEPT (a) = $5,000: the FIXED COST that is incurred even at ZERO machine hours. Represents costs that don't scale with activity within the relevant range, examples for maintenance: routine inspections, supervisor salary, basic facility maintenance, equipment-monitoring contracts. SLOPE (b) = $12 per machine hour: the VARIABLE COST per UNIT OF ACTIVITY (machine hour). Represents costs that scale linearly with activity, examples for maintenance: lubricants, replacement parts proportional to running hours, hourly maintenance technician charges. TOTAL COST FORMULA: Y = $5,000 + $12X. At X = 1,000 hours: Y = $5,000 + $12,000 = $17,000. At X = 2,000 hours: Y = $5,000 + $24,000 = $29,000. The fixed component stays constant; only the variable component changes. INCREMENTAL COST from a change in activity: ΔY = b × ΔX = $12 × ΔX. The intercept doesn't 'add' to incremental cost (already absorbed in baseline). USES OF COST DECOMPOSITION: (1) BUDGETING, separate fixed budget from activity-based budget. (2) FORECASTING, adjust budget for actual activity volume. (3) DECISION ANALYSIS, variable cost matters for marginal decisions (order acceptance, make-or-buy); fixed cost is sunk for short-term decisions. (4) COST BEHAVIOR ANALYSIS, flexible budgeting, contribution margin, breakeven. KEY ASSUMPTIONS: (1) LINEARITY, the relationship is linear within the relevant range. (2) RELEVANT RANGE, the analysis is valid only within the historical activity range. Outside that range, fixed cost may step up (capacity additions) and variable cost may shift. (3) STEADY STATE, the cost structure isn't undergoing fundamental change. ASSESSING FIT: R² should be reasonably high (e.g., 0.75+) for the decomposition to be reliable. Lower R² means significant unexplained variation, weakening the fixed/variable distinction. Answer A is incorrect: the intercept ($5,000) IS economically meaningful, it represents the fixed cost component. Answer C is incorrect: the labels are reversed. The intercept (constant) is fixed; the slope (per-unit) is variable. Answer D is incorrect: the equation describes cost behavior at a point in time, not year-over-year inflation.
Worked example
Cedar Co. models monthly maintenance cost as Y = $6,000 + $8X, where X is monthly machine hours. R² is 0.80. Calculate predicted maintenance cost for 2,000 machine hours.
| 1 | Interpret the fixed componentAt X = 0, the equation gives Y = $6,000 + ($8 × 0) = $6,000. This is the model's intercept, interpreted as estimated monthly fixed cost, not evidence of actual cost at zero activity. | $6,000 per month |
| 2 | Calculate the variable component$8 per machine hour × 2,000 machine hours. The $8 slope estimates added cost per hour, not average total cost per hour. | $16,000 |
| 3 | Add fixed and variable costs$6,000 + $16,000 | $22,000 |
| 4 | Keep R² out of the dollar calculationR² = 0.80 means 80% of observed cost variation is explained. No R² adjustment enters Y = a + bX. | $22,000 unchanged |
Predicted monthly maintenance cost at 2,000 machine hours is $22,000.
Check: Work backward: ($22,000 − $6,000) ÷ 2,000 hours = $8 per hour, matching the regression slope.
Key points
- The slope estimates cost added per activity unit, not average total cost per unit.
- The intercept estimates cost at zero activity; it does not prove actual zero-activity cost.
- R² measures explained cost variation, not a percentage adjustment to predicted cost.
How the exam traps you
- Reporting $16,000 by using only the variable portion. Add the $6,000 fixed component to get $22,000.
- Multiplying the $22,000 forecast by R² to report $17,600. Do not multiply by 0.80. R² measures fit, not forecast dollars.
Question 2
Hint
TIME SERIES (one entity over time): preferred for cost-BEHAVIOR modeling, captures the activity-cost relationship for that process. Cross-sectional conflates product differences with behavior. Address time-series concerns with diagnostics; panel data combines both.
Answer C. TIME SERIES data is generally preferred for cost-BEHAVIOR modeling because it tracks how costs respond to activity changes for a SINGLE process over time, precisely what cost-behavior analysis seeks to measure. Cross-sectional data conflates inter-product cost differences with cost-behavior responses, making it less informative about how a specific product's costs vary with its specific activity. KEY CONCERNS WITH TIME SERIES (and remedies): (1) Economic conditions changing over time (inflation, technology, scale), diagnostic via trend analysis or segment the data. (2) Seasonality, control via seasonal dummies. (3) Operational changes (new equipment, process redesign), segment data around the change. (4) Autocorrelation in residuals, diagnostic via Durbin-Watson; address with appropriate models. KEY CONCERNS WITH CROSS-SECTIONAL: (1) Product mix differences (different drivers, complexity levels) dominate cost variation, masking the activity-cost relationship. (2) Heteroscedasticity (variance differs across products of different sizes). (3) Selection bias (which products are sampled). HYBRID APPROACH (PANEL DATA): combines time series and cross-sectional dimensions, multiple products tracked over time. Provides both within-product behavior and between-product variation; statistically powerful but more complex. Cross-sectional data IS useful for benchmarking analyses (comparing one product's metrics to others). Answer A is incorrect: cross-sectional data conflates inter-product differences with cost behavior. Holding 'conditions constant' is illusory when the products themselves differ. Answer B is incorrect: the two structures have substantively different analytical implications. Answer D is incorrect: time series is not 'unreliable', known concerns have established diagnostics and remedies. Cross-sectional is not 'only valid'; it has its own concerns.
Question 3
Hint
Plug the given X value into intercept plus slope times X, and keep the slope's sign.
Answer A. Substitute 12 for X in the regression equation. The prediction is 840 - 15(12) = 840 - 180 = 660. A predicted value comes from combining the intercept with the slope times the given X value.
Why not B: This comes from treating the slope as positive instead of negative. In this model, more training lowers predicted returns, so the product is subtracted, not added.
Why not C: This comes from subtracting the slope once instead of multiplying it by 12. The coefficient must be applied to the full X value in the equation.
Why not D: This comes from multiplying 15 by 12 but forgetting the intercept. The model does not start at zero; it starts at 840 when X equals 0.
Question 4
Hint
Identify which variable changes by one unit and which variable is expected to change in response.
Answer A. The slope coefficient estimates the change in the dependent variable for a one-unit increase in the independent variable. Here, delivery cost is the dependent variable and packages delivered is the independent variable. Therefore, 2.75 means estimated delivery cost increases by $2.75 for each additional package delivered.
Why not B: This mistake treats the slope as the intercept. The estimated cost when packages delivered equals zero is 4,800, not 2.75.
Why not C: This mistake reverses the dependent and independent variables. The coefficient shows the estimated change in cost for one more package, not the change in packages for one more dollar.
Why not D: This mistake treats the slope as an overall average. A slope is a marginal estimate, showing the expected change in cost for one additional package.
Question 5
Hint
Regression validity = linear range. When data is linear within historical range but suspected non-linear above: restrict regression to linear range, document range of validity, supplement with engineering-based estimates for forecasts beyond. Don't blindly extrapolate or impose non-linear form.
Answer D. REGRESSION VALIDITY IS LIMITED TO THE OBSERVED LINEAR RANGE. When data shows linearity within historical activity but suspected non-linearity at higher volumes, the appropriate approach is: (1) RESTRICT THE REGRESSION to the linear range (1,500-3,500 hours where the relationship is well-behaved). (2) DOCUMENT THE 'RANGE OF VALIDITY' for the regression, communicate to forecast users that predictions are only reliable within this range. (3) FOR PREDICTIONS BEYOND THE LINEAR RANGE: don't blindly extrapolate. Supplement with engineering-based estimates (peak-time energy rates × hours), step-function adjustments (e.g., 'add 10% per hour above 3,500'), or judgment overlays based on operational knowledge. (4) IF NEW HIGH-VOLUME DATA BECOMES AVAILABLE: re-fit the regression including the new data, possibly using a non-linear form or two-piece (segmented) regression. WHY 'SINGLE LINEAR REGRESSION OVER FULL RANGE' IS WRONG: forces linear assumption on data that's known to be non-linear at high volumes, produces biased estimates and unreliable predictions in BOTH the linear and non-linear regions (the linear estimates get pulled by the non-linear high-volume observations). WHY 'HIGH-LOW METHOD' IS WRONG: high-low uses only two extreme observations. If those happen to fall in the non-linear region, the entire estimate is distorted. High-low has no conceptual advantage for non-linearity; in fact it's MORE vulnerable. WHY 'NONLINEAR REGRESSION FOR ENTIRE DATASET' IS WRONG: imposes a non-linear form (quadratic, log, etc.) on data that's known to be linear within most of the historical range. The non-linear fit may distort the linear region. Better to use linear regression where linearity holds and supplement at the extremes. PROFESSIONAL JUDGMENT: regression is a tool with KNOWN ASSUMPTIONS (linearity, range of validity). When facts suggest those assumptions break, the analyst's job is to apply the model where it's valid and explicitly handle the cases where it isn't. Answer A is incorrect: extrapolation through known non-linearity produces unreliable predictions. Answer B is incorrect: high-low method has the SAME range-of-validity issue and is even more sensitive to outliers. Answer C is incorrect: blindly applying non-linear regression to data that's mostly linear distorts the model.
Question 6
Hint
Compare each shipment level to the X-values used to build the regression model.
Answer C. Regression forecasts are usually most dependable for X-values within the data used to fit the model. The 2,700-shipment forecast lies above the observed 1,400 to 2,300 range, so it is an extrapolation and should be viewed most cautiously. A strong R squared does not remove that limitation.
Why not A: This choice assumes any single forecast from the equation is equally weak. But 1,900 shipments is well within the sample range, so the specific extrapolation concern does not apply.
Why not B: This choice focuses on being close to the upper boundary rather than beyond it. But 2,250 shipments remains inside the observed range, so it is still an interpolation.
Why not D: This choice treats a value near the lower end as if it were outside the sample. But 1,500 shipments is still within the observed 1,400 to 2,300 range, so the model is interpolating, not extrapolating.
Common questions
How do you calculate predicted cost from a regression equation?
Multiply the activity level by its coefficient, then add the intercept. Here, $6,000 + ($8 × 2,000) = $22,000.
What does the intercept mean in a cost regression?
It is the model's predicted cost when activity equals zero. Here, the $6,000 intercept represents estimated monthly fixed cost, not evidence of actual cost at zero hours.
Do you multiply a regression cost forecast by R-squared?
No. R² = 0.80 means the model explains 80% of the observed cost variation. It does not reduce the $22,000 forecast.
Practice BAR like the real exam
The free ChatCPA simulator: real exam layout, timed testlets, starting with a question on this topic. No account needed to start.
Open the free simulator →More on Data analytics, regression analysis, and quantitative methods
- Adjusted R-Squared When Adding Regression Predictors
- Correlation vs. Causation: Confounding and Selection Bias
- Time-Series Forecasting: Match the Method to the Pattern
- Sensitivity Analysis vs Scenario Analysis: One Input or Many
All Data analytics, regression analysis, and quantitative methods practice →
Questions from the ChatCPA bank of 17,658 CPA exam questions, each with a written reason for every wrong answer. ChatCPA is built by Nicholas Miller, CPA (Oregon #14907). How these pages are made. Spot an error? Tell us.