Autoregressive Model (AR)

What Is an Autoregressive Model?

An autoregressive model is a statistical method that predicts a value in a time series using a weighted combination of its own earlier values. The word means "regressing on itself." Instead of relying on outside variables, the model looks at how the series behaved in the recent past and uses that pattern to estimate what comes next. Analysts apply it to asset prices, returns, volatility, and economic indicators.

How It Works

The simplest version, called an AR(1) model, assumes that today's value depends on yesterday's value plus a random shock:

y(t) = c + φ × y(t-1) + e(t)

Here y(t) is the value at time t, c is a constant, φ (phi) is a coefficient that measures how strongly the previous value carries forward, and e(t) is an unpredictable error term. Models with more lags, such as AR(2) or AR(p), include several earlier values in the same way. The coefficients are usually estimated from historical data with ordinary least squares, the same method used in basic regression.

The model works best when the series is stationarity|stationary, meaning its average and variability stay roughly stable over time. For an AR(1) model, the coefficient must have an absolute value below 1 for the series to remain stable. If it does not, the series can drift or grow without limit.

Uses in Finance

In investing, autoregressive models help describe persistence in data. For example, volatility tends to cluster, so calm periods often follow calm periods and turbulent periods often follow turbulent ones. The GARCH model extends the autoregressive idea to variance, allowing today's volatility to depend on past squared shocks and past volatility. Similar approaches are used to forecast interest rates.

Advantages and Limitations

The approach is simple, transparent, and relatively easy to estimate. Its main limitation is the assumption that past patterns will continue. Financial series can shift abruptly during market crises or policy changes, which can make estimated coefficients unreliable. Forecasts from a stationary autoregressive model also tend to move toward the series' long-run average as the forecast horizon grows. Because of these limits, the models are usually used alongside other tools rather than on their own.