From 959b8a670d5025f10a318d99e2b6ff69982eafd6 Mon Sep 17 00:00:00 2001 From: Berend Bouvy Date: Fri, 29 Nov 2024 10:33:17 +0100 Subject: [PATCH] forecast update --- book/time_series/forecasting.md | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/book/time_series/forecasting.md b/book/time_series/forecasting.md index 595c8c6..a8c8341 100644 --- a/book/time_series/forecasting.md +++ b/book/time_series/forecasting.md @@ -24,8 +24,11 @@ In summary, given a time series $Y=\mathrm{Ax}+\epsilon$, the workflow is as fol 3. Predict the signal-of-interest: $\hat{Y}_{signal}=\mathrm{A}_p\hat{X}$, where $\mathrm{A}_p$ is the design matrix describing the functional relationship between the future values $Y_p$ and $\mathrm{x}$. -4. Predict the noise $\hat{\epsilon}_p$ based on the AR model. +4. Predict the noise $\hat{\epsilon}_p$ based on the AR model. $\hat{\epsilon}_p = \Sigma_{Y_pY}\Sigma_Y^{-1}\hat{\epsilon}$, where $\Sigma_{Y_pY}$ is the covariance matrix between the future values $Y_p$ and the observed values $Y$. 5. Predict future values of the time series: $\hat{Y}_p=\mathrm{A}_p\hat{X}+\hat{\epsilon}_p$. +```(note) +This procedure is a general approach to forecasting time series data. It resembles the process of stochastic inter- and extrapolation, which is used in many fields of science and engineering. +```