We present a hybrid method for computing volatility forecasts that can be used to implement a risk-controlled strategy for a multi-asset portfolio consisting of both US and international equities. Recent years have been characterized by extremely low yields, with 2022 marked by rising interest rates and an increasing inflation rate. These factors produced new challenges for both private and institutional investors, including the need for robust forecast methods for financial assets' volatilities. Addressing such task, our research focuses on a hybrid solution that combines classical statistical models with specific classes of Recurrent Neural Networks (RNNs). In particular, we first use the Generalized Autoregressive Conditional Heteroscedasticity (GARCH) approach within the preprocessing phase to capture volatility clustering, striking an efficient balance between computational effort and accuracy, to then apply RNN architectures, namely GRU, LSTM, and a mixed model with both units, as to maximize performances of volatility forecasts later used as input factors for risk-controlled investment strategies. In terms of portfolio allocation, we focus on a simplified version of the Risk Parity method that was first proposed by the Research division of S&P Global. This version ignores the contribution of cross-correlations among assets, nevertheless providing encouraging results. Indeed, we show the effectiveness of the chosen approach by providing forward-looking risk parity portfolio strategies that outperform standard risk/return portfolio structures.

Volatility forecasting with hybrid neural networks methods for Risk Parity investment strategies

Luca Di Persio;Matteo Garbelli;
2023-01-01

Abstract

We present a hybrid method for computing volatility forecasts that can be used to implement a risk-controlled strategy for a multi-asset portfolio consisting of both US and international equities. Recent years have been characterized by extremely low yields, with 2022 marked by rising interest rates and an increasing inflation rate. These factors produced new challenges for both private and institutional investors, including the need for robust forecast methods for financial assets' volatilities. Addressing such task, our research focuses on a hybrid solution that combines classical statistical models with specific classes of Recurrent Neural Networks (RNNs). In particular, we first use the Generalized Autoregressive Conditional Heteroscedasticity (GARCH) approach within the preprocessing phase to capture volatility clustering, striking an efficient balance between computational effort and accuracy, to then apply RNN architectures, namely GRU, LSTM, and a mixed model with both units, as to maximize performances of volatility forecasts later used as input factors for risk-controlled investment strategies. In terms of portfolio allocation, we focus on a simplified version of the Risk Parity method that was first proposed by the Research division of S&P Global. This version ignores the contribution of cross-correlations among assets, nevertheless providing encouraging results. Indeed, we show the effectiveness of the chosen approach by providing forward-looking risk parity portfolio strategies that outperform standard risk/return portfolio structures.
2023
Volatility
Forecast
Time series
Neural Networks
Risk Parity
Finance
Neural Networks
File in questo prodotto:
File Dimensione Formato  
1-s2.0-S095741742300920X-main.pdf

solo utenti autorizzati

Licenza: Copyright dell'editore
Dimensione 4.71 MB
Formato Adobe PDF
4.71 MB Adobe PDF   Visualizza/Apri   Richiedi una copia

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11562/1105826
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 1
  • ???jsp.display-item.citation.isi??? 1
social impact