ARIMA Forecaster

Fit an ARIMA model to financial time series, auto-select p/d/q via AIC, plot in-sample fit and out-of-sample forecast with confidence intervals.

Feb 23, 2026, Eric

The Core Idea

ARIMA (AutoRegressive Integrated Moving Average) combines three ideas: AR (past values predict future), I (differencing to achieve stationarity), and MA (past forecast errors predict future). It is the workhorse of financial time series forecasting — used for volatility forecasting, spread prediction, and volume modeling. This tool fits an ARIMA(p,d,0) model, auto-selects parameters via AIC, and validates with walk-forward testing.

Data & Parameters

Upload CSV

Train/test split:80/20

Auto-select (p,d) via AIC

300 observations | Train: 240 | Test: 60

Model Diagnostics

Model

ARIMA(2,0,0)

AIC

436.1

In-sample RMSE

2.4685

Out-of-sample RMSE

2.5086

Intercept

9.9494

AR(1)

0.2562

AR(2)

0.2585

Forecast

Residuals

Walk-Forward Validation (5 folds)

Fold	Train size	Test size	MAE	RMSE
1	120	36	1.990	2.311
2	156	36	2.463	3.057
3	192	36	2.612	3.121
4	228	36	1.794	2.351
5	264	36	2.176	2.809
Average			2.207	2.730

What This Means

Best model: ARIMA(2,0,0) with AIC = 436.1. The AR(2) structure means the series has memory extending 2 periods back. No differencing was needed — the series is already stationary.

Overfitting check: In-sample RMSE: 2.468, Out-of-sample RMSE: 2.509 (+2% degradation). Minimal overfitting — the model generalizes well to unseen data.