Understanding ARIMA, SARIMA, and SARIMAX — and How They Help in PPC Forecasting

1. ARIMA — The Core Forecasting Model

ARIMA (AutoRegressive Integrated Moving Average) is the backbone of classical time-series forecasting. It models how a metric (for example, daily spend or clicks) depends on its past values and the structure of recent errors.

Example in PPC: use ARIMA to forecast daily ad spend or impressions when the account is relatively stable and has no strong seasonality.

2. SARIMA — Adding Seasonality

When your campaigns show weekly or monthly cycles (which most do), ARIMA alone misses repeating patterns. SARIMA adds seasonal components so the model learns recurring peaks and dips.

Example in PPC:

• Forecasting weekly conversions when performance drops on weekends.
• Estimating monthly spend if bids increase at quarter ends.

Understanding SARIMA parameters

SARIMA is written as ARIMA(p, d, q)(P, D, Q, m) where:

• p: autoregressive order — how many past observations the model uses to predict the current value.
• d: differencing order — how many times we subtract consecutive values to make the series stationary (removing trends).
• q: moving average order — how many past forecast errors are included to correct the prediction.
• P: seasonal autoregressive order — how many past seasonal periods influence the current value.
• D: seasonal differencing order — how many seasonal periods we subtract to remove seasonal trends.
• Q: seasonal moving average order — how many past seasonal errors are used to correct predictions.
• m: the length of the seasonal cycle — e.g., 7 for weekly, 12 for monthly.

Tip: correctly tuning (P, D, Q, m) is key for accurate seasonal forecasts.

3. SARIMAX — When External Factors Matter

SARIMAX (Seasonal ARIMA with eXogenous variables) allows external regressors — things that actually drive PPC metrics beyond their history.

Example in PPC:

• Include budget or campaign impressions as regressors to forecast conversions.
• Add search interest or CTR changes to predict revenue.
• Combine data from multiple channels (Google Ads + Meta spend) to model total lead volume.

4. Which One Should You Use?

5. Key Takeaway

These models quantify the patterns you already sense: slow Mondays, promo spikes, end-of-period pushes. They don't replace intuition — they turn it into measurable forecasts. Automate a simple SARIMAX with spend and conversions, and you'll trade guessing for optimization.