Short-Term Time-Series Forecasting with Empirical Mode Decomposition
Summary
The article adapts empirical mode decomposition (EMD) for short-horizon forecasting. EMD splits a time series into intrinsic mode functions and a residual by iteratively finding extrema and interpolating them with splines. The proposed forecast extends the decomposition beyond the last observed point, then uses a smoothed reconstruction from selected components as the forecast source. The article also discusses a partial autocorrelation indicator and differencing as tools for examining series behavior and improving stationarity.
A key implementation issue is how future points are handled during decomposition. If the extended points are allowed to contribute extrema, the resulting components can adjust to one another and produce a self-balancing projection that no longer reflects the original series well. The revised method restricts extrema detection to observed data and extrapolates from the final real point. Because spline extrapolation can diverge, the forecast horizon should remain short. The article presents indicators and an EA implementation, but does not establish broad predictive performance; forecasts are best treated as an additional input to trading decisions.
Key ideas
- EMD decomposes a series into intrinsic mode functions and a residual using extrema and spline interpolation.
- A forecast can be formed by extrapolating a smoothed reconstruction built from selected decomposition components.
- Future synthetic points should not be treated as observed extrema during decomposition, as this can distort the forecast.
- Spline extrapolation can diverge, so the method is intended for a short horizon.
- Differencing can increase stationarity, though repeated differencing also has drawbacks.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.