Quantifying Risks of DeFi Options Vaults
By Andrei Anisimov, Lukas Kiss, Adam Feather, Robert Koschig, and IV League DAO contributors
Abstract
What average yield can you expect as an LP in DeFi Options Vault (DOV)? Here we provide in-depth historical backtests and simulation of DOV performance.
We also introduce the IV League DAO — a quantitative research DAO created to design better yield strategies for options and other derivatives.
With recent decline in yields across lending and yield farming protocols, LPs are seeking alternative ways to generate income on their crypto — either via off-chain lending (e.g. Maple Finance, Goldfinch) or selling risk on-chain via options.
DeFi options vaults (DOVs) like Ribbon Finance, Dopex, Siren and others experienced massive growth in TVL over the past year due to their unparalleled ability to generate high yields on a variety of crypto assets. Currently, more than $270M across assets like ETH, USDC, AVAX and others are being used to underwrite options via systematic selling strategies.
As TVL continues to grow, it is important to ask what are the risks of these strategies? In this research post we’ll do a deep dive into backtesting popular Call- and Put- selling strategies as well as simulating returns across hypothetical scenarios. The goal is to lay the foundation for quantifying risks of DOVs to help foster transparency for healthy and sustainable growth.
Strategies
There are several slight variations to the DOVs strategies, but ultimately it comes down to selling short-term out-of-the-money (OTM) covered call and cash-secured put options. The strike is systematically selected to minimize the risk of exercise while earning high yields via option premiums (typically 20%+). The duration of the option is typically limited to 1 week (some protocols up to 1 month).
The ideal scenario: the option expires OTM and LPs collect the premium that gets reinvested into the next selling cycle, compounding the returns. However, there is a non-zero probability of ITM expiration, in which case LPs are exposed to theoretically unlimited losses — our goal is to try to quantify those risks.
Here we will focus on a strategy that sells weekly options with strike selection based on 0.1 delta. This strategy was pioneered by Ribbon Finance and currently accounts for the larger portion of TVL. In the future, the quantitative framework proposed here can be used to evaluate and design other strategies.
Weekly call selling strategy seeks to sell OTM covered calls with delta close to 0.1 every Friday. If the option expires OTM one week later, LPs pocket the premium (20%+ APY). If the option expires ITM, LPs lose some of their asset (depending on how deep in the money). Typical assets: ETH, WBTC, SOL, etc
Weekly put selling strategy seeks to sell OTM cash-secured puts with delta close to -0.1 every Friday. Same rules apply for calls. Typical asset: USDC
Most DOVs are cash-settled, which means in case of ITM expiration the collateral asset (e.g. ETH for calls and USDC for puts) gets distributed between the buyer and the seller according to their net profit or loss. For example, as a seller of ETH calls you don’t actually sell your ETH for USDC, but simply lose some of your ETH in case the option expires ITM.
Backtest
We analyze the performance of the weekly 0.1 delta calls and puts strategy using ETH price data from 2017–06–01 to 2022–05–25. To calculate options pricing, we use Deribit DVOL index where available. For the earlier period, we calculate implied volatility based on the realized volatility (see Appendix: Methodology for details).
ETH price experienced a dramatic rise during the backtest period, which might not be representative of the future. However, it is still valuable to take a look at DOV returns for calls and puts.
Also, note that unlike most DOV products that only measure performance weekly at expiration, we measure it daily, showing mark-to-market value of the vaults. This leads to multiple downward “spikes” on the pool value chart. In our view, it is important to account for that intra-expiry volatility when evaluating risk of these products.
ETH Calls selling strategy
The ETH Calls selling strategy didn’t perform well due to significant increase in underlying price. Interestingly, even during the period of bear/flat market of 2018–2020 the pool value stayed flat due to temporary price spikes that quickly reversed gains accrued during downward trends.
The return for the entire period was -2% (-0.4% APY). Overall, the results are inline with the intuition that if options are priced correctly, then the expected value (EV) of systematically buying or selling them should be about 0.
In USD terms, the strategy did quite well, although underperforming buy-and-hold by -2% with significant volatility in-between.
ETH Puts selling strategy
Puts selling strategy generated 13.7% return (2.6% APY). It also performed relatively flat, inline with the Call selling strategy.
Overall, the backtest empirically validates the notion that systematically selling fairly-priced options should result in EV of 0. Although, anyone with the ability to correctly time the market or intelligently hedge their exposure can generate significant returns.
Another interesting observation is that long-term direction of the market has less impact than the short-term market volatility. For example, despite significant increase in ETH price during the backtest period, the annualized performance is relatively flat due to large losses caused by short-term volatility.
Simulation
Now let’s take a look at how the Covered Calls selling strategy performs using simulated price data. The price data generator parameterized using average annualized log return mu and annual volatility sigma. We run 3,000 simulations for each combination of these parameters to generate price paths with various properties: upward, downward and sideways trending, with low and high volatility and anything in between (there are other parameters to the simulation that are described in the Appendix).
Here’s an example of a single simulated price path and resulting Calls vault performance.
We generate 9 sets of price paths for each combination of average returns (-50%, 0%, +50%) and volatilities (70%, 100%, 130%). Each simulation set consists of 3K price paths that satisfy given parameters. This allows us to simulate DOV returns under bear, flat and bull as well as more and less volatile markets.
The output is the distribution of pool returns. Left to right is 70%, 100% and 130% price volatility. Top to bottom is -50%, 0% and +50% average price return during the simulated year. For each simulation set, we show average APY and aggregated sharpe ratio of returns.
We can observe that only one simulation set (high price return, low volatility) results in Sharpe Ratio above 1 (strategies with sharpe ratio below 1 are typically not considered attractive). The distributions have longer left tails, reflecting the asymmetric risk profile of options writing — steady, but limited gains from options premiums and rare, but theoretically unlimited losses.
Token holders hungry for yield are eager to deposit funds into strategies promising high returns. However, our research shows that a simplistic passive options selling strategy is generally not very attractive on a risk-adjusted basis.
The market for DeFi structured products involving options and other derivatives is very early. We believe that better strategies can be designed to provide more attractive risk-adjusted returns for LPs. The design space is quite large, we’re excited to create a research collective to work on solving these problems.
Introducing IV League DAO
In order to continue this work, we are creating the IV League DAO, a community of DeFi users focused on better understanding implied volatility, derivatives, and ideating on DeFi products that will support the crypto market at large. We will design and stress-test automated options strategies that offer sustainable returns for liquidity providers. This research, methodology, backtesting, and simulation code is the founding contribution to the IV League DAO.
We are in the early days of automated DeFi structured products strategies. However, it is clear that the future of sustainable yield lies in enabling the flow of risk that creates fluid and efficient markets unlike anything we’ve seen in traditional finance.
The IV League DAO currently includes community members of the Siren project–a composable DeFi option protocol–who were vital in our research and have offered a vault in which the IV League DAO can build option strategies and iterate quickly. Our community will drive innovation in the DeFi options space and invest the DAO holdings into new, cutting edge strategies to create value for our members and DeFi as a whole.
Admissions
For the time being, as we build this community, we’re keeping it small. We believe the greatest value we can add in these early days will come from a team of strong collaborators. So we’re opening up the admissions process to a select few. We have certain roles and values we are looking for from this early community, join the team, applications will be open until Friday, August 5th. We will be reaching out to conduct 1v1s from August 5th-19th.
Appendix I: Backtest Methodology
In order to calculate options pricing and strike selection we use Deribit DVOL Index where available (since 2021–03–24). Prior to that we use 30-day realized volatility plus 5%. When comparing DVOL to RV we found that on average IV is higher than RV by ~6%, we reduced the shift to 5% to avoid overfitting.
We found that 30 day window RV produces overall satisfactory correlation with IV on average, but is not a perfect measure when it comes to options pricing. However, the impact of the RV/IV discrepancy on strike selection is minimal and shouldn’t influence the result in a significant way.
Strikes are selected based on their delta. Strike that is closest to 0.1 delta is used, whether above or below, for example 0.09 delta will be chosen over 0.12. The strike price step is based on the asset price — typically $100, except early periods where step is $10.
Appendix II: Simulation Methodology
We use 1-hour ETH data from Gemini. From this data, we have created a log_normal distribution of log returns for one hour intervals. We used this distribution for block bootstrapping without replacement, because our sample length was ¼ (1 year) from the original period. The block length was randomly picked from 24 to 24*3. (one to three days) to keep autocorrelation patterns present in the original time series.
To change the mean and standard deviation of our log returns distribution, we normalized it to mean=0 and standard deviation = 1 and then we have added our mu and scaled it by sigma.
IV for options pricing and strike selection is calculated using 30 day realized volatility plus 5% — same as for the earlier period of the backtest.
mu, sigma and APY
muis an average value of log returns per hour. If we calculateinitialPrice*exp(mu)this results in the median of last prices (median APY) of generated paths. (Half the runs will have lower final APY and half will have higher final APY). To select appropriatemu, we first determine the median APY for our price paths. Then we scale it to log hourly returnsmuusing formula:log(APY + 1)/(365*24).sigmais the annual volatility.
Below is an example of aggregated 3,000 paths with initial price of $2,000, average APY of 86% and volatility (sigma) of 100%.
