CPF Exam Prep Free practice test →

Free CPF Practice Questions

10 free, exam-style Certified Professional Forecaster (CPF) practice questions with answers and explanations. No signup required. Work through them below, then take the full free CPF practice test to study every exam domain.

Question 1

Forecast error in IBF/Jain convention is computed as:

  1. e_t = A_t / F_t − 1, expressed as a unitless ratio of actual demand divided by forecast minus one
  2. e_t = F_t − A_t (Forecast minus Actual), so positive errors indicate that the forecast exceeded actual demand
  3. e_t = |A_t − F_t| / A_t, the absolute error expressed as a percentage of actual demand for the period
  4. e_t = A_t − F_t (Actual minus Forecast), so positive errors indicate under-forecasting by the model
Show answer & explanation

Correct answer: D - e_t = A_t − F_t (Actual minus Forecast), so positive errors indicate under-forecasting by the model

Question 2

A common surprising finding when companies first implement FVA analysis is that:

  1. Executive overrides always produce better accuracy than the statistical baseline
  2. Statistical models always outperform every form of human judgment by wide margins overall
  3. All analyst overrides consistently improve accuracy versus the baseline forecast
  4. Many manual override steps actually reduce accuracy compared to the statistical baseline
Show answer & explanation

Correct answer: D - Many manual override steps actually reduce accuracy compared to the statistical baseline

Question 3

MAPE is BIASED in favor of forecasts that under-forecast because:

  1. MAPE is bounded above by 100% for over-forecasts but unbounded for under-forecasts
  2. MAPE is bounded above by 100% for under-forecasts but unbounded for over-forecasts
  3. MAPE penalizes over-forecasts and under-forecasts symmetrically across all periods
  4. MAPE eliminates all bias by definition under standard accuracy measurement principles
Show answer & explanation

Correct answer: B - MAPE is bounded above by 100% for under-forecasts but unbounded for over-forecasts

Question 4

The Bass Model adoption equation in continuous time is approximately:

  1. n(t) = γ·(A_t − L_{t−1} − b_{t−1}) + (1−γ)·S_{t−m} identical to the seasonal recursion form
  2. n(t) = α·A(t) + (1−α)·F(t) identical to the SES recursion equation form for the adoption
  3. n(t) = β·(L_t − L_{t−1}) + (1−β)·b_{t−1} identical to the Holt's trend recursion form
  4. n(t) = p[M − N(t)] + (q/M)·N(t)·[M − N(t)], where p, q, and M are the model parameters
Show answer & explanation

Correct answer: D - n(t) = p[M − N(t)] + (q/M)·N(t)·[M − N(t)], where p, q, and M are the model parameters

Question 5

If a new product is forecast to sell 1,000 units/month with 30% expected cannibalization of an existing SKU currently selling 2,000 units/month, the existing SKU's adjusted forecast becomes:

  1. 2,300 units/month, computed by adding 30% of the new product's 1,000 units to the existing SKU
  2. 1,700 units/month, computed as 2,000 minus 30% of 1,000 in the cannibalization adjustment
  3. 1,400 units/month, computed as 2,000 minus 30% of 2,000 across the cannibalization period
  4. 2,600 units/month, computed by adding 30% of the existing SKU's 2,000 units to its own forecast
Show answer & explanation

Correct answer: B - 1,700 units/month, computed as 2,000 minus 30% of 1,000 in the cannibalization adjustment

Question 6

An AR(p) process is identified from ACF and PACF by which pattern?

  1. Both ACF and PACF cut off sharply at the same lag p across the time series shown
  2. PACF decays gradually (often exponentially); ACF cuts off sharply after lag p in the data series
  3. ACF decays gradually (often exponentially); PACF cuts off sharply after lag p in the data
  4. Both ACF and PACF decay gradually with no clear cutoff at any lag in the time series
Show answer & explanation

Correct answer: C - ACF decays gradually (often exponentially); PACF cuts off sharply after lag p in the data

Question 7

A planner fits ARIMA(0, 1, 1) and finds the model is mathematically related to which exponential smoothing method?

  1. Single Exponential Smoothing - ARIMA(0,1,1) is approximately equivalent to SES with optimal smoothing
  2. Holt's Linear ES - ARIMA(0,1,1) is approximately equivalent to Holt's method with two smoothing constants
  3. Holt-Winters - ARIMA(0,1,1) is approximately equivalent to triple exponential smoothing
  4. Naïve forecasting - ARIMA(0,1,1) is approximately equivalent to last-period-equals-next-period
Show answer & explanation

Correct answer: A - Single Exponential Smoothing - ARIMA(0,1,1) is approximately equivalent to SES with optimal smoothing

Question 8

A planner builds a multiple regression model and finds the F-statistic is highly significant but no individual t-statistic is significant. This pattern OFTEN indicates:

  1. Autocorrelation - the residuals are serially correlated across observations in the time-series data
  2. Underfitting - the model is too simple to capture the signal even in the training observations of the data
  3. Heteroscedasticity - the residuals do not have constant variance across the fitted values of the model
  4. Multicollinearity - the variables jointly explain Y but their individual contributions are confounded
Show answer & explanation

Correct answer: D - Multicollinearity - the variables jointly explain Y but their individual contributions are confounded

Question 9

Why is using SALES TARGETS as the demand forecast considered a worst practice?

  1. Targets are mathematically identical to statistical baselines under any market conditions
  2. Targets reflect aspiration not unbiased expectation, embedding optimism bias into supply plans
  3. Targets are required by IBF certification standards as the official basis for the demand forecast
  4. Targets always exactly match historical demand patterns across every category of products
Show answer & explanation

Correct answer: B - Targets reflect aspiration not unbiased expectation, embedding optimism bias into supply plans

Question 10

The bias-variance trade-off in ML refers to:

  1. The trade-off between underfitting (high bias) and overfitting (high variance) in model selection
  2. The trade-off between forecast bias (mean error) and forecast accuracy (mean absolute error)
  3. The trade-off between training time and prediction time across different ML algorithms
  4. The trade-off between model interpretability and model accuracy across the choice of different ML algorithms
Show answer & explanation

Correct answer: A - The trade-off between underfitting (high bias) and overfitting (high variance) in model selection

Ready for the real thing?

Practice hundreds more CPF questions with instant scoring, weak-area drills, and full exam simulations.

Start the free practice test See pricing