•
Hansen, Huang & Shek (2012) jointly model daily returns and realized measures: RV is a noisy observation of latent variance plus a leverage polynomial. High-frequency data into the GARCH workhorse. Plain-language intro. Not investment advice.
Takeaway. Classic GARCH only sees squared close-to-close returns. Realized GARCH adds a second equation: today’s realized measure (e.g. 5-minute RV) is a noisy reading of latent variance, not variance itself—plus a leverage function of the return shock.
Realized GARCH: A Joint Model for Returns and Realized Measures of Volatility (Peter R. Hansen, Zhuo Huang, Howard Howan Shek, Journal of Applied Econometrics, 2012). Open working-paper PDF; published DOI 10.1002/jae.1234.
Fit for learners: Joint return + realized-measure system: a measurement equation turns RV into a noisy reading of latent σ², with leverage in the residual.
This Hub note is a plain-language intro. Not investment advice.
Daily GARCH is slow after a jump: one squared return is a noisy variance proxy. High-frequency RV is faster but not equal to the true conditional variance (microstructure, overnight gap). Two-step “plug RV into GARCH” ignores measurement error and the joint likelihood of returns and RV.
The ancestor of later “LSTM inside RealGARCH” papers—read this first.
You get the state-space hygiene: two noises, one latent h_t. Before Keras, a weekend loop that updates h from yesterday’s RV already beats “annualise the last 20 daily squares.”
Tiny evening exercise: given a list of 5-minute log-returns for one day, rv = sum(x*x for x in bars); print rv vs the squared overnight+day close-to-close. Learning only.
Landing: Journal of Applied Econometrics (DOI)
Not investment advice. This Hub note is educational only. Hansen–Huang–Shek is academic volatility research—not a recommendation to trade variance or set limits from RealGARCH.
Single course CTA: SoloQuant Academy’s Introduction to Python Programming—realized variance is a running sum of squares; start with accumulators.