Yo!
I’m a sophomore working on an experimental volatility framework based on GARCH, called **GARCH-FX** (GARCH Forecasting eXtension). It’s my attempt to fix the “flatlining” issue in long-term GARCH forecasts and generate more realistic volatility paths, with room for regime switching.
Long story short:
* GARCH long term forecasts decay to the mean -> unrealistic
* I inject Gamma distributed noise to make the paths stochastic and more lifelike
What worked:
* Stochastic Volatility paths look way more natural than GARCH.
* Comparable to Heston model in performance, but simpler (No closed form though).
What didn’t:
* Tried a 3-state Markov chain for regimes… yeah that flopped lol. Still, it’s modular enough to accept better signals.
* The vol-of-vol parameter (theta) is still heuristic. Haven’t cracked a proper calibration method yet.
Here’s the SSRN paper: [https://papers.ssrn.com/sol3/papers.cfm?abstract\_id=5345734](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5345734)
Thoughts and Feedbacks welcome!
Trend Engine: AI-Powered News & Trends 