False Safe Haven Assets: Evidence From the Target Volatility Strategy Based on Recurrent Neural Network

Abstract

This paper examines which safe haven assets should be used when improving out-of-sample portfolio performance. We define a market state with recurrent neural network (RNN) volatility predictions and construct an investment strategy that dynamically combines equity, cash, and safe havens. The equity is allocated by targeting the volatility, and investing in safe havens depends on the predicted market state. We consider the S&P500 index with 13 safe haven assets, such as long-term government bonds, commodities, gold, and other precious metals. Other indices, NIKKEI225, NIFTY50, and STOXX50, are examined for robustness. With analysis conducted over a 20-year sample period, we find that RNN delivers sound predictions to construct the volatility targeting strategy. Among considered assets, only long-term Treasury bonds act as a safe haven and improve the strategy performance. Other considered assets have no such potential. Our findings are relevant to portfolio managers and investors actively managing portfolio risk.