Time Series Interview Questions (2026)
Practice 45 Time Series interview questions on trend, seasonality, stationarity, lags, rolling windows, ARIMA, forecasting horizons, backtesting, metrics, and drift.
45 questions with answers- What Is Time Series Forecasting?
- All Questions on This Page
- Time Series Fundamentals
- Time Series Practical Interview Questions
- Time Series Advanced Scenarios
- ARIMA vs Machine Learning vs Deep Learning Forecasts
- How to Prepare for a Time Series Interview
- How Strong Time Series Answers Sound
- Time Series Quiz
- Frequently Asked Questions
What Is Time Series Forecasting?
Key Takeaways
- ✓Time series forecasting predicts future values from ordered observations over time.
- ✓Interviews test trend, seasonality, stationarity, autocorrelation, lags, rolling windows, horizons, backtesting, metrics, and drift.
- ✓Strong answers use time-based validation. Random split is usually wrong for forecasting.
- ✓Forecasting quality depends on the prediction horizon, data frequency, external events, and how errors affect decisions.
Time series forecasting predicts future values from time-ordered data such as sales, traffic, demand, inventory, temperature, or sensor readings. In interviews, time series questions test whether you understand trend, seasonality, stationarity, autocorrelation, lag features, rolling windows, forecasting horizons, backtesting, model choice, metrics, and production monitoring.
Watch: Time Series Forecasting in Python
Video: Time Series Forecasting in Python (freeCodeCamp.org, YouTube)
Test yourself and earn a certificate
6 quick questions. Score 70%+ to download your Time Series certificate.
All Questions on This Page
45 questions- 1. What is trend in time series?
- 2. What is seasonality?
- 3. What is stationarity?
- 4. What is autocorrelation?
- 5. What is a lag feature?
- 6. What is a rolling window feature?
- 7. What is a forecast horizon?
- 8. What is backtesting?
- 9. What is walk-forward validation?
- 10. What is ARIMA?
- 11. What is SARIMA?
- 12. What is an exogenous variable?
- 13. What is a prediction interval?
- 14. What is MAPE?
- 15. What is WAPE?
- 16. Design a sales forecasting model.
- 17. How do you create lag features safely?
- 18. How do you use rolling averages?
- 19. How do you choose a forecast metric?
- 20. How do you handle missing timestamps?
- 21. How do you model holidays?
- 22. Why do stockouts matter in demand forecasting?
- 23. How do you forecast multiple steps ahead?
- 24. Which baselines should you try?
- 25. How do you split time series data?
- 26. How do you forecast many related series?
- 27. How do you use external features?
- 28. How do you produce uncertainty ranges?
- 29. How do you monitor a forecast model?
- 30. How do you decide retraining frequency?
- 31. A candidate uses random train-test split. What is wrong?
- 32. One-step forecasts work but 12-step forecasts fail. Why?
- 33. MAPE is huge for low-volume items. What happened?
- 34. Forecast misses a holiday spike. What is missing?
- 35. The model underforecasts after stockouts. Why?
- 36. How do you forecast a new product?
- 37. A promotion changes demand pattern. What do you do?
- 38. A sensor series has missing timestamps. What do you inspect?
- 39. Weekly seasonality changes after product launch. What happens?
- 40. A useful weather feature is not known at forecast time. What is the issue?
- 41. Store-level forecasts do not sum to region forecast. What do you consider?
- 42. Intervals miss too many actuals. What is wrong?
- 43. Recent data arrives late. How does it affect forecasting?
- 44. A deep model beats baseline in training but not backtest. What do you choose?
- 45. What do you ask before approving a forecast model?
Time Series Fundamentals
Foundational15 questionsStart here. These are the definitions and first-principle checks that open most rounds.
Q1. What is trend in time series?
Trend is the long-term upward or downward movement in a series.
A sales series can grow over years even while weekly values fluctuate.
Trend changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Watch a deeper explanation
Video: Time Series Forecasting in Python (freeCodeCamp.org, YouTube)
Q2. What is seasonality?
Seasonality is a repeating pattern at a known frequency such as daily, weekly, monthly, or yearly.
Retail and traffic series often have strong seasonality.
Seasonality changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q3. What is stationarity?
A stationary series has statistical properties such as mean and variance that stay roughly stable over time.
Many statistical models assume stationarity.
Stationarity changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q4. What is autocorrelation?
Autocorrelation measures how a series relates to its past values.
It helps identify useful lags.
Autocorrelation changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
| Answer part | What to say | Evidence to mention |
|---|---|---|
| Definition | Autocorrelation in one direct sentence. | Official docs or course material |
| Use case | The work where it changes a decision. | Dataset, model, query, dashboard, or pipeline |
| Risk | What breaks when it is misunderstood. | Metric, log, test result, or review note |
Q5. What is a lag feature?
A lag feature uses a past value of the target or another variable as model input.
Lag features must use only past data.
Lag feature changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Forecasting path
Never train on future values.
Watch a deeper explanation
Video: Time-series prediction with TensorFlow (TensorFlow tutorial, YouTube)
Q6. What is a rolling window feature?
A rolling window feature aggregates values over a moving past period such as 7-day average.
Window boundaries must exclude the forecast target period.
Rolling window changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q7. What is a forecast horizon?
The forecast horizon is how far into the future predictions are required.
The horizon changes feature design and model choice.
Forecast horizon changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q8. What is backtesting?
Backtesting evaluates forecasts on historical periods by simulating how the model would have predicted the future.
It is the core validation pattern for forecasting.
Backtesting changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q9. What is walk-forward validation?
Walk-forward validation repeatedly trains on earlier data and tests on the next time window.
It mimics live forecasting.
Walk-forward validation changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q10. What is ARIMA?
ARIMA is a statistical forecasting model using autoregression, differencing, and moving-average terms.
It is a useful baseline for many univariate series.
ARIMA changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q11. What is SARIMA?
SARIMA extends ARIMA with seasonal terms.
It fits series with repeating seasonal patterns.
SARIMA changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q12. What is an exogenous variable?
An exogenous variable is an external input that may help forecast the target, such as price, holiday, weather, or promotion.
It must be known or forecastable at prediction time.
Exogenous variable changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q13. What is a prediction interval?
A prediction interval gives a likely range for a future value rather than only a point forecast.
It helps planning under uncertainty.
Prediction interval changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Watch a deeper explanation
Video: Time Series Analysis with StatsModels (PyData, YouTube)
Q14. What is MAPE?
MAPE measures average absolute percentage error.
It breaks or behaves poorly when actual values are zero or near zero.
MAPE changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q15. What is WAPE?
WAPE measures total absolute error divided by total actual value.
It is often more stable than MAPE across many series.
WAPE changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Time Series Practical Interview Questions
Intermediate15 questionsThese questions test whether you can apply the topic to real data, real code, and messy constraints.
Q16. Design a sales forecasting model.
Define product, store, frequency, horizon, target, known future features, baselines, backtest windows, and error metric.
Promotions and stockouts need special handling.
sales forecast changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
# Interview check: split data, run baseline, inspect metric
from sklearn.metrics import accuracy_score
print(accuracy_score(y_true, y_pred))Q17. How do you create lag features safely?
Sort by time, group by entity, shift past values, and verify no future target leaks into features.
Group boundaries matter.
create lag features changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q18. How do you use rolling averages?
Compute rolling mean, sum, min, max, or volatility over past windows before the prediction time.
Use closed-left windows when predicting the next period.
rolling averages changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q19. How do you choose a forecast metric?
Use MAE for readable absolute error, RMSE for large-error penalty, WAPE for aggregate demand, and MAPE only when zeros are not a problem.
Metric should match planning cost.
choose metric changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q20. How do you handle missing timestamps?
Reindex the time range, distinguish zero from missing, impute carefully, and flag missingness where useful.
Missing and zero mean different things.
handle missing periods changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q21. How do you model holidays?
Add holiday flags, pre-holiday and post-holiday windows, region calendars, and event-specific features.
Holiday impact can shift by category.
holiday effects changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q22. Why do stockouts matter in demand forecasting?
Observed sales may be lower than true demand when inventory is unavailable.
Use inventory signals or censored-demand logic.
stockout bias changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q23. How do you forecast multiple steps ahead?
Use recursive, direct, or multi-output forecasting depending on horizon, error accumulation, and model type.
Recursive forecasts can compound error.
multi-step forecast changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q24. Which baselines should you try?
Try naive last value, seasonal naive, moving average, and simple trend baselines.
A complex model should beat simple baselines.
baseline forecast changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q25. How do you split time series data?
Train on earlier periods and validate on later periods using one or more backtest windows.
Do not shuffle rows for forecasting.
time split changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Watch a deeper explanation
Video: LSTM time series forecasting (Python tutorial, YouTube)
Q26. How do you forecast many related series?
Use global models with entity features, hierarchical rules, or per-series models depending on data volume and similarity.
Sparse series may need pooling.
many series changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q27. How do you use external features?
Include only variables known at forecast time or forecasted separately, such as holidays or planned price.
Unknown future weather may not be available.
external regressors changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q28. How do you produce uncertainty ranges?
Use quantile models, bootstrap, conformal methods, or model-specific intervals.
Planning often needs ranges, not only points.
prediction intervals changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q29. How do you monitor a forecast model?
Track error by horizon, segment, season, input drift, missing data, and business impact.
Aggregate metrics can hide segment failures.
monitor forecast changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q30. How do you decide retraining frequency?
Base it on drift, new data volume, seasonality, model decay, and operational cost.
Retraining too often can add noise.
retrain schedule changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Time Series Advanced Scenarios
Advanced15 questionsAdvanced rounds test trade-offs, failure modes, and whether the decision can hold up under production pressure.
Q31. A candidate uses random train-test split. What is wrong?
Random split leaks future patterns into training for forecasting.
Use time-based validation.
random split mistake changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q32. One-step forecasts work but 12-step forecasts fail. Why?
The longer horizon has harder uncertainty and may need different features or model strategy.
Evaluate by horizon.
great short forecast bad long forecast changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q33. MAPE is huge for low-volume items. What happened?
Actual values near zero make percentage error unstable.
Use WAPE, MAE, or segment-aware metrics.
MAPE explodes changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q34. Forecast misses a holiday spike. What is missing?
Calendar and event features or enough historical holiday examples may be missing.
Add region-specific event features.
holiday miss changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q35. The model underforecasts after stockouts. Why?
Observed sales during stockouts hide true demand.
Inventory availability should be modeled.
stockout underforecast changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q36. How do you forecast a new product?
Use similar products, hierarchy, launch plan, attributes, and wider uncertainty intervals.
No history means less certainty.
new product cold start changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q37. A promotion changes demand pattern. What do you do?
Model promotions explicitly and monitor post-event behavior.
Promotion effects can linger.
drift after promotion changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q38. A sensor series has missing timestamps. What do you inspect?
Check device outage, ingestion delay, zero readings, and imputation rules.
Do not treat missing as zero by default.
missing timestamp changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q39. Weekly seasonality changes after product launch. What happens?
The model may need recent weighting, retraining, or new features.
Seasonality is not always stable.
seasonality change changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q41. Store-level forecasts do not sum to region forecast. What do you consider?
Use hierarchical reconciliation or choose the level that matches the decision.
Forecast hierarchy needs consistency.
hierarchy mismatch changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q42. Intervals miss too many actuals. What is wrong?
Uncertainty is underestimated or validation distribution shifted.
Calibrate intervals.
prediction interval too narrow changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q43. Recent data arrives late. How does it affect forecasting?
Lagged features and rolling windows may be wrong unless ingestion lag is handled.
Monitor data freshness.
late data changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q44. A deep model beats baseline in training but not backtest. What do you choose?
Prefer the model that wins backtests and is maintainable.
Complexity needs evidence.
overfit complex model changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
Q45. What do you ask before approving a forecast model?
Ask about horizon, frequency, baseline, validation, leakage, events, metrics, uncertainty, drift, and retraining.
Those questions decide whether the forecast is useful.
senior review changes Time Series decisions by affecting the chosen method, the failure mode, or the validation step. Reliable evidence comes from a metric, test result, query output, log, or review note.
ARIMA vs Machine Learning vs Deep Learning Forecasts
Forecasting interviews often ask which model to use. The right answer depends on data size, seasonality, external features, number of series, and explainability needs.
| Approach | Use when | Strength | Risk |
|---|---|---|---|
| ARIMA/SARIMA | Single or few structured series | Interpretable statistical baseline | Needs stationarity and careful diagnostics |
| Tree or linear ML | External features and many signals | Handles tabular lag features | Can leak future data |
| Deep learning | Many related series or complex patterns | Learns shared patterns | Needs more data and monitoring |
| Naive baseline | First comparison | Hard to beat in some cases | Too simple for changing patterns |
Time series answer signals
Validation design is the highest-value signal.
- Time split: no future leakage
- Horizon: what time ahead
- Seasonality: daily, weekly, yearly
- Metrics: MAE, RMSE, MAPE, WAPE
How to Prepare for a Time Series Interview
Prepare by defining the forecast target, frequency, horizon, decision use, validation window, baseline, and production retraining plan.
- Review trend, seasonality, stationarity, autocorrelation, lag features, rolling windows, and exogenous variables.
- Know time-based splits, walk-forward validation, backtesting, leakage, and forecast horizons.
- Practice baselines, ARIMA-style models, tree models with lag features, and neural sequence models at a high level.
- One story about forecasting failure caused by holiday effects, stockouts, drift, or bad validation is useful.
Time series interview prep flow
Most rounds reward clear reasoning more than memorized phrasing.
How Strong Time Series Answers Sound
A strong time series answer starts by defining the forecast horizon and decision. Say whether the forecast is one-step or multi-step, which time frequency is used, what seasonality exists, how validation avoids future leakage, which baseline is used, and which metric matches the business cost. production-ready answers include stockouts, holidays, missing periods, concept drift, prediction intervals, and backtesting design.
Define the horizon first
A one-day forecast and a twelve-month forecast are different problems.
Use time-safe validation
Train on the past and validate on later periods. Random split leaks time structure.
Compare against simple baselines
Naive, seasonal naive, and moving average baselines catch overcomplicated models.
Monitor after launch
Forecast models drift when demand, pricing, calendar effects, or product behavior changes.
Test Yourself: Time Series Quiz
Ready to test your Time Series knowledge?
6 questions, about 4 minutes. Score 70% or higher to earn a shareable certificate.
Frequently Asked Questions
What do Time Series interviews usually ask?
Is ARIMA still useful for interviews?
What Time Series project should I discuss?
What separates experienced Time Series candidates?
Practice forecasting answers before data interviews
Use this Time Series bank for forecasting and validation answers, then practice Python and data tasks with Hyring's AI Coding Interviewer.
See Hyring AI Coding InterviewerSources
Written by Adithyan RK