Moving Average (MA) Models
What you'll learn: You'll understand how MA models use past forecast errors—not past values—to predict what comes next in a time series.
The Core Idea
Unlike Autoregressive (AR) models that use previous actual values to predict the future, Moving Average (MA) models use previous forecast errors (also called residuals or shocks). Think of it like this: if you're navigating and you notice you've been veering left consistently, you adjust right based on those past mistakes—not just your past positions.
In an MA model, today's value is the average (weighted sum) of past random shocks plus some baseline level. The model "remembers" recent surprises and incorporates them into the current prediction.
The MA(q) Model
An MA(q) model uses the last q forecast errors. The notation works like this:
- MA(1): Uses only the most recent error
- MA(2): Uses the two most recent errors
- And so on...
The mathematical form looks like:
y_t = μ + ε_t + θ₁ε_{t-1} + θ₂ε_{t-2} + ... + θ_qε_{t-q}
Where:
y_tis the current valueμis the mean/baselineε_tis the current random errorθ₁, θ₂, ...are the MA coefficients (weights on past errors)ε_{t-1}, ε_{t-2}, ...are past errors
Why This Matters
MA models capture temporary shocks that affect a series for a limited time. If a sudden event (weather, news, supply disruption) creates an error spike, the MA component models how that shock ripples through subsequent periods before fading away.
Key Takeaway: Moving Average models predict current values using a weighted combination of past forecast errors, making them ideal for capturing short-term shock effects in time series data.