Introducing Unified GARCH-RNN Volatility Forecasting for after-work quants

Takeaway. Most “GARCH + LSTM” papers are a pipeline: fit GARCH, dump σ into a net. This paper embeds the GARCH(1,1) recursion inside GRU and LSTM cells, so ω, α, β remain interpretable while gates pick up nonlinear memory. GARCH-GRU is the speed/accuracy workhorse.

Unified GARCH-Recurrent Neural Networks in Financial Volatility Forecasting (Jingyi Wei, Steve Yang, Zhenyu Cui, 2025; arXiv:2504.09380). S&P 500, DJI, NASDAQ, 2010–2020; 2019 calm OOS and 2020 COVID stress.

Fit for learners: Embed GARCH(1,1) inside GRU/LSTM cells so ω, α, β stay readable while gates learn nonlinear persistence—not a two-stage GARCH-then-LSTM pipeline.

This Hub note is a plain-language intro. Not investment advice.

The problem

Linear GARCH captures clustering but not breaks. Pure LSTM/Transformer vol models forget stylised facts (persistence α+β ≈ 0.97, leverage). Two-stage hybrids train the pieces separately, so the net cannot modulate the GARCH state.

The idea / core intuition

• GARCH-GRU: g_t = φ(ω + α ε²_{t-1} + β σ²_{t-1}); add γ g_t into the GRU hidden state (additive fusion).
• GARCH-LSTM: same g_t, multiplicative scale of the LSTM hidden state (Zhao-style gate).
• Reparameterise so ω>0, α,β≥0, α+β<1. Softplus output for positive vol.
• Horse-race vs GARCH, GJR, pipeline GARCH→LSTM, and Transformer on MSE / MAE / SMAPE / OOS R²; 99% VaR pinball in stress.

A volatility lab—not a trading bot.

Why after-work Python learners care

You can implement GARCH(1,1) in a loop this month, then read why people put that loop inside a cell instead of stacking sklearn. Interpretable α, β are a sanity check when the net overfits.

Tiny evening exercise: h = w + a*r2 + b*h in a for loop over fake squared returns; print a+b. Learning only.

One-evening reading checklist

1. Pipeline hybrid vs embedded cell: what is jointly trained.
2. Additive (GRU) vs multiplicative (LSTM) fusion.
3. Why constrain α+β<1.
4. One metric besides MSE (VaR pinball / SMAPE in COVID).

Download PDF

Download PDF

Landing: arXiv abs

Important disclaimer

Not investment advice. This Hub note is educational only. Wei–Yang–Cui is academic volatility research—not a recommendation to trade vol or set VaR from GARCH-GRU.

What this is not

• Not CEF / subsidy framing
• Not Quant or Macro course CTA
• Not a vol-product pitch

Next step

Single course CTA: SoloQuant Academy’s Introduction to Python Programming—a variance recursion is a for loop over a list.