> ## Documentation Index
> Fetch the complete documentation index at: https://sourcetable.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Forecasting

> Predict future values with ARIMA, exponential smoothing, and ML-based time series forecasting.

Sourcetable supports multiple forecasting methods through StatsModels and scikit-learn. Describe your forecasting goal in plain English and the AI selects the appropriate method.

## ARIMA / SARIMAX

```
"Forecast monthly revenue for the next 12 months using ARIMA"
```

ARIMA models handle trend and noise. SARIMAX adds seasonal components and external variables.

| Model   | Use case                                                                  |
| ------- | ------------------------------------------------------------------------- |
| ARIMA   | Non-seasonal time series with trend                                       |
| SARIMA  | Seasonal time series (monthly, quarterly, weekly patterns)                |
| SARIMAX | Seasonal time series with external predictors (marketing spend, holidays) |

The AI automatically:

* Tests for stationarity (ADF test)
* Determines optimal (p, d, q) parameters using AIC/BIC
* Adds seasonal orders when periodicity is detected
* Generates forecasts with confidence intervals

## Exponential smoothing

```
"Use Holt-Winters to forecast daily sales for the next 30 days"
```

| Method                        | Use case                              |
| ----------------------------- | ------------------------------------- |
| Simple Exponential Smoothing  | No trend, no seasonality              |
| Holt's Linear                 | Trend, no seasonality                 |
| Holt-Winters (additive)       | Trend + consistent seasonal amplitude |
| Holt-Winters (multiplicative) | Trend + growing seasonal amplitude    |

## ML-based forecasting

```
"Build a gradient boosting forecast using lag features and rolling averages"
```

The AI creates time series features (lags, rolling means, day-of-week, month) and trains a regression model. This approach works well when:

* You have many external predictors
* The relationship is non-linear
* Traditional time series models underperform

## Seasonal decomposition

```
"Decompose the sales data into trend, seasonal, and residual components"
```

Breaks your time series into:

* **Trend** — long-term direction
* **Seasonal** — repeating patterns at fixed intervals
* **Residual** — unexplained variation

Supports both additive and multiplicative decomposition.

## Confidence intervals

All forecasts include confidence intervals (default 95%). The AI visualizes:

* Historical data as a solid line
* Forecast as a dashed line
* Confidence bands as shaded regions

```
"Forecast quarterly revenue for 2025 with 90% confidence intervals"
```