# Umami V2 Vault Backtesting Preliminary Results

For the last few months at Umami Labs, we’ve been working hard to develop and refine the strategy for our upcoming “V2” vault - a product that will enable investors to gain exposure to the impressive levels of yield generated by GMX’s liquidity provider token (GLP) without being exposed to unwanted delta. Today we are excited to share some of the key findings and results from our simulation and backtesting phase in this preliminary report.

### Simulation purpose and methodology

The purpose of the simulation step was to optimize the vault strategy, test the strategy's performance over real historic data, and stress test it with likely scenarios and edge cases. To achieve this, we pulled _all_ GMX and GLP data directly from the blockchain with a time granularity of 1 second and recreated the GMX contracts and our vault contracts in Python. The simulation accounted for every price, allocation, fee, and yield change, with careful consideration of accurately accounting for our vault's own effects in the simulation. The result was a powerful and invaluable tool that gives deep insight into the challenges and payoffs of potential strategies built on top of GMX.

### Key GLP findings

#### GLP performance

GLP benefits from receiving 70% of GMX's platform fees. It also acts as a counterparty to margin traders, and as such will benefit from traders losing and will suffer when they lose. GMX hosts a [comprehensive stats page](https://stats.gmx.io) which makes it easy for us to analyze its past performance both in terms of fees generated, and trader proft and loss (pnl).

<span style="background-color:yellow">[graphs of trader pnl & fees taken for the simulation period]</span>

During the simulation period, we saw traders take heavy losses, with the cumulative loss at the end of the period reaching approximately $30M, with daily fees relatively volatile and averaging around $200k.

#### GLP composition

The performance figures above show that GLP is a very strong product, however, building a vault on top of it is not free from challenges. When hedging out GLP's delta exposure, it is important to understand GLP's asset allocation changes along with the volatile asset price movements. With swaps, glp mint and burns, as well as trader activity, the asset weightings inside GLP can change rapidly and unpredictibly. Strategies based on passive weighting or ignoring these sharp asset changes performed very poorly as the delta exposure coupled with price movements could quickly wipe out gains.
<span style="background-color:yellow">Update this graph with area plot w/ stables combined - what period?</span>

External hedging was identified very early on as the only viable way to build to achieve our delta neutral/tracking goals and maintaining proper delta exposure is critical to a successful long term strategy. A proper rebalancing algorithm analyzing the correct conditions and timing are key to driving performance, but this hedging can accrue a high cost if not done in the correct way. GMX fees in leverage positions of mint/burn/open/close/funding are all important to balance.

<span style="background-color:yellow">[Need Delta Tracking Image Here]</span>

### Creating a Baseline Strategy

A baseline strategy was created and optimized initially (Strategy A) of a stand alone USDC vault that fronts the full cost of hedging the volatile assets in GLP. This initial strategy was run to give a baseline to compare our Final Netting strategy results, and to ensure that even in the most extreme vault TVL imbalances, (User deposit and withdrawal fluctuations in the Umami vaults) this strategy can perform well.


##### Strategy A key metrics

- Overall gain for the period of 18.55%, (27.8% annualized)
- Sharpe ratio of 1.4
- Max peak-to-trough drawdown of 4.37%

GMX/GLP had great performance in this period from trader pnl, which helped the overall performance of the strategy, with $29.7m worth of assets being added to GLP aum ending in $272m for the period. After isolating the performance and removing any trader pnl effect the baseline Strategy A was still positive, which is exactly the result we were looking for.

### Final Strategy Design

It was then time to finalize our master strategy; a series of vaults that cover each asset in GLP with advanced internal accounting and an allocation algorithm that can ensure extremely close delta exposure coverage with greatly reduced reliance on external hedging.

<span style="background-color:yellow">[chart or graphic here]</span>

The Netting Strategy (Strategy B) designed by Umami utilizes 5 vaults to reduce fees substantially and increase glp ownership to boost yield generated from the GMX platform fees. On average, the netting strategy saw a reduced hedging cost of 65-95%, based on vault balance and glp asset allocation volatility. It was able to boost GLP allocation, on average, by 11%. This allows for an extremely strong performance through the 8 months.

##### Strategy B key metrics
- Overall gain for the period of 26.35%, (39.52% annualized)
- Sharpe ratio of <span style="background-color:yellow">XXX</span>
- Max peak-to-trough drawdown of <span style="background-color:yellow">X.X%</span>

It's worth emphasising again that trader losses translating to GLP asset gains were a signficiant contributor to the overall performance during the backtesting period, yet even with these gains removed, the vault performance still comes in at an extremely strong 16.2% APR annualized, considering only GLP's WETH yield minus the fees incurred by the strategy.

### The Next Steps

Whilst our focus is now on finalizing the smart contract code for the vault and taking through audits, we still have some extremely valuable information we can obtain from our simulation including optimizing strategy variables for increased performance, testing the effects of hedging open interest, and ensuring protection in edge cases and extremely rapid, dramatic changes in allocation, price or TVL.

The nature of our vault's design and implementation means we can continue to optimize these parameters right up to, and even beyond the launch of the vault, and can exercise a level of responsiveness to changes in market conditions and trends.

Whilst we hope that this report has been informantive and encouraging, we will also be working on a more comprehensive whitepaper that will be released at around the time that our new vaults launch. That document will go through the backtesting process and strategy in greater depth.

Until then be sure to keep up to speed with the latest developments by following us on twitter, joining our discord, and tuning in to our weekly AMAs!