CTR Forecasting Using SARIMA

A time-series analytics project that models and forecasts daily Click-Through Rate (CTR) behavior, identifies seasonal engagement patterns, and produces a 30-day forecast to support marketing pacing, budget planning, and performance optimization.

Project Overview

Click-Through Rate (CTR) is a critical KPI in digital advertising, reflecting how effectively users engage with ads. This project models CTR as a time-series problem, leveraging statistical forecasting techniques to answer key business questions:

• How does CTR behave over time?
• Are there recurring weekly patterns in user engagement?
• Is CTR predictable enough to support short-term planning?
• Can forecasts be used to guide spend and campaign scheduling?

Using SARIMA, the project delivers a stable and interpretable forecasting framework suitable for real-world marketing decision-making.

Approach

• Exploratory data analysis & weekly seasonality detection
• Stationarity testing (ADF, KPSS) and differencing
• ARIMA baseline vs SARIMA (weekly seasonality)
• Train–test evaluation with MAE, RMSE, MAPE
• Residual diagnostics and final 30-day forecast

Key Insights

• CTR shows strong weekly seasonality, with mid-week (Tue–Thu) consistently outperforming weekends.
• A gradual decline from late 2022 stabilizes in recent months, indicating predictable performance.
• CTR oscillates within a narrow band (~3.7%–4.1%) in the forecast horizon.
• Residuals indicate acceptable model fit with expected real-world volatility.

Business Impact

This analysis enables improved decision-making around:

• Budget allocation and pacing
• Campaign scheduling
• Performance expectation setting
• Creative testing and optimization

The forecast indicates a stable and controllable CTR environment, making it suitable for confident short-term scaling and planning.

Tech Stack

Python · pandas · numpy · statsmodels · matplotlib · seaborn

Future Improvements

• Incorporate exogenous variables (spend, impressions, creatives)
• Add anomaly detection for sudden CTR drops
• Automate monthly model retraining
• Deploy forecasting pipeline with scheduled updates

Conclusion

This project demonstrates how CTR can be effectively modeled and forecasted using statistical time-series techniques. With clear seasonality, strong predictive performance, and interpretable outputs, the SARIMA framework provides a practical foundation for data-driven marketing optimization.