Forecasting cost-per-click of keywords in Google’s competitive paid search advertising market: a time-series clustering approach

Abstract

Accurately forecasting Cost-per-Click of paid search advertising is essential for performance marketers to allocate budgets that optimize marketing campaign returns. In this study, we perform a comprehensive analysis using various time-series forecasting methods to predict daily average CPC of keywords in the car rental sector. Our results show the power of statistical models on noisy keyword-level CPC time-series on short to medium horizons, only being outperformed by more complex neural networks on longer horizons. Advanced forecasting approaches leveraging competition did not yield significant accuracy improvements. Additional experiments with fine tuned foundational models for time-series showed promising results, optimizing practicality and accuracy.