There must be risks to investing. However, we always want to eliminate it as much as possible. How to trasform the tail risks into our return is this report want to solve. We test different risk measures mentioned in Turning Tail Risks into Tailwinds to control the tail risks of our portfolios. We can avoid unreasonable risk and ensure expected return. In the experiments, their maximum Sharpe can reach almost one stably.
We use Equally Risk Contribution and spectral risk measures to construct portfolios. First, each asset has the same level of risk rather than the same amount of capital. Second, estimated volatility through the cumulative volatility of the tail risks and spectral risk measures.
The assets we used in the experimental process are sector ETFs, including XLE, XLF, XLK, XLV, XLI, XLY, XLP, XLU, and XLB. The data source is from Dec, 26th, 1998 to Dec, 23th, 2022. We construct portfolios by naive measures and correlated measures. Also, we examine that the latter has better performance than the former. Table 1 and Table 2 display the performances. There are eight different risk measures for naive and correlated, respectively. Vol is the abbreviation of volatility; D.Vol is the abbreviation of downside volatility; CVaR is the cumulative value at risk; PSM is the power spectral measure. MB means moment-based, including both kurtosis and skewness. ER means extreme risks, using simulation data to estimate risks.
We tested different timeframes for our portfolio. The best and most intuitive one is rolling one year. We can observe the performance in Figure 1 to Figure 3. Figure 1 shows that correlated D.Vol performs better than SPY on cumulative return. Figure 2 describes the difference of rolling 252 days Sharpe Ratio between D.Vol and SPY. The mean of the difference is more significant than zero, implying that the Sharpe of D.Vol is greater than SPY on average. Figure 3 shows the rolling 252 days return of D.Vol and SPY, and D.Vol trim the extreme gain and loss to robust gain.
In this report, there are some critical points we have examined. First, the correlated portfolios can reach better performances than the naive ones. Second, we can change the original setting of the paper into 3-year and 1-year rolling data; correlated portfolios still could work excellently. Third, even when we split the backtesting timeframe, the D.Vol correlated portfolio could maintain a high standard. Thus, we affirm that the D.Vol correlated portfolio is an excellent method for constructing a portfolio under sector assets. Taking more different-market assets into account will be future research direction.
Contact us at firstname.lastname@example.org for the full report.
Quantitative Research Intern, Quantitative Finance, Gamma Paradigm