Lag Features for Time Series Forecasting Quiz

Answers and explanations appear as you take the quiz.

1. To forecast y_t before time t, which lag feature is temporally valid?
   • y_(t-1)
   • y_(t+1)
   • The realized value y_t
   • The average of all observations through the test period
2. Where should a trailing seven-day mean end when predicting the next day's value?
   • At the target day including its actual value
   • At the last fully observed day before the target
   • At the end of the full evaluation dataset
   • At the future forecast horizon
3. Why can a random split give an unrealistic time-series evaluation?
   • It removes all seasonal structure
   • It makes every lag feature missing
   • Future rows may train a model evaluated on earlier periods
   • It guarantees that test rows precede training rows
4. When is a planned holiday indicator a valid forecast feature?
   • Whenever the model accuracy improves
   • Only after the holiday has occurred
   • When it is calculated using test outcomes
   • When the calendar information is known at forecast time
5. Why might a validation split include a gap?
   • To make the forecast horizon random
   • Delayed labels or overlapping windows can otherwise cross the boundary
   • To ensure all features are standardized twice
   • To increase training data with future observations
6. How should scaling or imputation be fit during time-series cross-validation?
   • Only on the final test set
   • Once on all dates before splitting
   • Within each training fold only
   • Separately on each individual feature after prediction
7. Which procedure recursively generates a multi-step forecast?
   • Remove all lag features after the first forecast
   • Use only future actual values at every step
   • Train a separate model for each target class
   • Feed earlier forecast predictions into later forecast features
8. What should a backtest reproduce for reliable assessment?
   • Information available at each forecast origin and the intended retraining cadence
   • Only the final feature matrix after all data are collected
   • A random order of historical targets
   • The model's training accuracy alone