Content area
Full text
Because it is difficult to estimate expected stock returns, the global minimum variance portfolio has been widely used in practice (e.g., Basak, Jagannathan, and Ma 2009). Under some simplifying assumptions, the portfolio essentially puts smaller weights on stocks with greater volatility and larger weights on those with less volatility. Early studies by Fleming, Kirby, and Ostdiek (2001, 2003) considered daily asset allocation across stocks and found supportive evidence for the economic value of volatility timing—that is, using volatility information to improve portfolio performance. Recent studies, such as by Barroso and Santa-Clara (2015) and Han, Huang, and Zhou (2019), confirmed the usefulness of volatility timing at the stock level. Related to these studies, Moreira and Muir (2017) provided a market volatility-timing strategy that exploits the well-known property of volatility persistence. The volatility-managed portfolio of Moreira and Muir (2017) is a leverage of the market, with a greater weight assigned to the market when recent volatility is low and a lower weight assigned when recent volatility is high. They showed that their strategy beats the market, yielding a positive alpha, a greater Sharpe ratio (SR), and sizable utility gains for mean–variance investors who follow their volatility-timing strategy versus buy-and-hold. They also showed that their strategy works for factor portfolios that are sorted on value, momentum, profitability, return on equity investment, and so on, as well as for currency carry trade.
In this article, we identify a look-ahead bias in the strategy of Moreira and Muir (2017). In their empirical results, they calibrated the weight parameter based on the unconditional volatility over the entire sample period. This appears innocuous at first, but a careful examination reveals that the results are highly sensitive to the calibrated parameter. To avoid the look-ahead bias and estimate the weight parameter at each time with only the data available at that time, we adopt two popular estimation windows: a fixed-window approach and a rolling-window approach. With both approaches and with the window size ranging from 5 to 20 years, we find that the maximum drawdown (MDD) of the volatility-managed portfolio is 68%–93% in most cases and thus likely infeasible in practice. In contrast, without correcting the look-ahead bias, the MDD is only 56%, comparable to 50% of the market.
Assuming the strategy is...