Volatility and correlation aren't fixed numbers โ they move, and there's a Nobel Prize behind the math that tracks it.
Every other correlation or covariance estimate on this hub โ RMT, Ledoit-Wolf, plain sample correlation โ produces one static number for your whole loaded history. DCC-GARCH asks a different, more honest question: how does volatility, and how do assets' relationships with each other, actually change day by day? Markets don't have one fixed "risk level" โ calm stretches and stressed stretches genuinely behave differently, and correlations themselves tend to rise exactly when markets get scary, which is precisely when diversification matters most.
Stage one: each asset gets its own univariate GARCH(1,1) model โ a way of estimating today's volatility as a blend of yesterday's volatility and yesterday's surprise return, so volatility can cluster (a big move tends to be followed by more big moves) instead of being treated as constant.
Stage two โ the "DCC" part: take each asset's standardized surprises (today's return, scaled by its own just-estimated volatility) and let the correlation between them drift over time too, governed by its own two-parameter recursion. The result is a full covariance matrix that actually breathes with the market, not a single frozen snapshot.
Top: GARCH(1,1) โ today's variance blends a long-run baseline (ฯ), yesterday's squared surprise (ฮฑ), and yesterday's own variance (ฮฒ). Bottom: the DCC recursion doing the same kind of blending, one level up, for the correlation matrix itself.
MFL-PORTFOLIO-SCIENCE's DCC-GARCH tab fits both stages via Quasi-Maximum Likelihood โ a hand-rolled Nelder-Mead simplex search does the numerical optimization for each stage (no closed form exists for either), the same honest, dependency-free approach this app uses elsewhere rather than reaching for a black-box solver. What comes out is a real, time-varying view of how your own basket's risk has actually been moving โ not an assumption that it's been constant the whole time.
Robert Engle built the foundational idea โ ARCH, "Autoregressive Conditional Heteroskedasticity" โ in 1982, formalizing the simple observation that financial volatility clusters instead of staying constant. Tim Bollerslev, one of Engle's own students, generalized it into GARCH in 1986, the version actually used almost everywhere today, including here.
Engle then returned to the problem himself in 2002 with the Dynamic Conditional Correlation extension โ letting the correlation between assets move the same way volatility does, the second stage this app's own DCC-GARCH tab runs. In 2003, Robert Engle won the Nobel Memorial Prize in Economics specifically "for methods of analyzing economic time series with time-varying volatility" โ this exact line of work, recognized at the highest level economics offers.
DCC-GARCH describes how volatility and correlation have moved so far โ it doesn't predict the next genuine shock before it happens. A true regime break (a crisis nobody in the fitted history resembles) can still catch it flat-footed, same as every model on this hub. Its real value is refusing to pretend risk is constant when it demonstrably isn't, not a promise of foresight.
General education only. Nothing here is personal financial advice or takes into account your objectives, financial situation or needs. Past performance is not a guarantee of future results.