Overview:
In traditional finance, volatility selling has garnered a somewhat questionable reputation due to its periodic propensity for implosion. Much of this perceived risk can be attributed to the excessive use of leverage when engaging in volatility selling. Nevertheless, there remains an opportunity to capture the variance risk premium, akin to the equity risk premium, by participating in option selling. On its own, systematic volatility selling doesn’t inherently generate alpha; instead, it can be likened to beta exposure, much like taking a long position in the broader equity index. In this piece, we will evaluate the performance of systematic BTC volatility selling while integrating filtering mechanisms with the aim of enhancing the risk-return profile of a basic strategy.
Baseline Straddle Performance:
For the purpose of this research piece, our focus will be on selling weekly straddles, where a new position will be initiated upon expiration. All option trades will execute at the top of the book bid price, and the net delta of the straddle will be hedged on a daily basis using BTC perpetuals on Deribit. Additionally, this analysis incorporates all fees and funding costs associated with maintaining the straddle position. Below, we present the baseline performance of selling BTC weekly straddles and rolling positions after each expiration. Overall, the results are commendable, albeit with moderate drawdowns during the tumultuous events of 2020 and 2021.
At this point, we can consider including certain features to filter out low-quality trades. First, we explore the variance risk premium (VRP) ratio, defined as ATM IV / RV. In theory, the most profitable vol selling should occur when ATM IV surpasses RV. We can sort the vanilla straddle PNL by this VRP ratio to identify any trends in the cumulative PNL of this strategy. Notice how the cumulative PNL surges as the VRP crosses the 1.00 level, indicating that the strategy accumulates the most PNL when IV exceeds RV, aligning with our expectations.
Additionally, vol selling thrives in tranquil market conditions, typically expressed by a term structure in contango (back-end vols > front-end vols). Hence, we can sort our PNL based on the 60DTE/30DTE ATM IV ratio. Similar to our prior analysis, notice that the most significant PNL is amassed when the ratio surpasses 1, signifying contango markets.
Lastly, we can also normalize these two ratios by employing a simple rolling 30-day z-score transformation. As will be demonstrated below, the purpose of normalization is to further refine trade selection by pinpointing opportunities at market extremes.
Filtered Straddle Performance:
Based on the insights gained from the previous analysis, we can now proceed to re-run our backtests, incorporating the filtering criteria outlined above. In these revised backtests, trades will only be initiated if a signal is activated, and they will be held until expiration, irrespective of intratrade signal activity. This approach allows us to assess whether there is any substantial improvement in performance using these filters.
Starting with the VRP strategy, it becomes evident that adopting the straightforward approach of selling straddles exclusively when the VRP ratio exceeds one consistently yields the highest absolute returns across all combinations related to this filter. However, as we introduce the z-score entry threshold criteria, overall returns do decrease, albeit with the benefit of reduced drawdowns. In the broader context, it’s noteworthy that none of these filters surpass the total return associated with the simple vanilla straddle strategy.
A similar narrative can be inferred from the term structure filter, where selling straddles during contango yields the highest total returns. Similar to the z-score filter for VRP, the use of various thresholds to select trades results in reduced overall returns and volatility.
Below is a table detailing the investment statistics for each of the various strategies. Remarkably, the plain vanilla straddle selling strategy without filters delivers the highest total return and maintains a respectable Sharpe ratio. However, when viewed through a risk-adjusted lens, the strategy of selling straddles when the 60/30 term structure ratio exceeds one offers the most favorable risk-adjusted reward, despite its lower total return. Strategies employing the VRP z-score signal also exhibit promise, with commendable risk-adjusted ratios (ie: VRP 7DTE z-score > 2); however, it’s essential to note that such signal opportunities tend to be sporadic, limiting the frequency of trade occurrences.
Conclusion:
In summary, the initial goal of outperforming our naive approach of selling vol proved more challenging than anticipated. Surprisingly, most of the features used to filter for higher-quality trades failed to enhance overall risk-adjusted returns. With the exception of several filters such as the 60/30 contango indicator and the 7DTE VRP z-score, in many cases, harvesting vol beta offered the best risk-adjusted return with minimal risk of backtest overfitting. As is often the case in finance, simplicity frequently outperforms complexity.