[SIPS-014 & 015] Upcoming ClearGate ETH-BTC Cellar Proposals

Governance
1

Upcoming ClearGate ETH-BTC Cellar Proposals

Goal

The goal of this forum post is to discuss the upcoming proposals asking Governance to accept the Cellar smart contracts for ClearGate strategies:

  1. ETH-BTC Trend
  2. ETH-BTC Momentum

The two proposals will be submitted simultaneously next week. Both of the two strategies will use the same smart contract source code.

Context

ClearGate was funded by the Community (see SIPS-013) to build a pipeline of Cellars on Sommelier. ClearGate is about to become the second Strategy Provider on the Sommelier platform by proposing two new strategies and has more strategies in the pipeline for the coming weeks and months.

We wanted to take the opportunity to discuss these upcoming proposals with the community before the vote formally takes place next week. We are happy to answer any questions the community may have.

The two strategies will be executed on-chain by 7Seas on behalf of ClearGate as part of 7Seas’ strategy integration services. See the two proposals below for more information.

1. Proposal for cellar authorization for strategy ETH-BTC Trend

This proposal is for the authorization of ETH-BTC Trend Cellar. The strategy for the cellar is provided by ClearGate.

ETH-BTC Trend is a dynamic long-only strategy for BTC-ETH portfolio management. It aims to be a better way to hold ETH and BTC while managing the negative volatility. Detailed information about the strategy, including strategy description and backtesting results can be found here: https://cgcellar.info/strategy-eth-btc-trend/

More information and comparison between the two simultaneously proposed strategies by GlearGate can also be found here: https://www.sommelier.finance/blog/deep-dive-on-trend-and-momentum-strategies

If approved, the chain will accept signed function calls submitted to the cellar contract from the cellar manager.

Name: ETH-BTC Trend

Cellar share token: ETHBTCTrend

Platform fee: 2% (1.5% for strategy provider + 0.5% for protocol)

Performance fee: 10% (7.5% for strategy provider + 2.5% for protocol)

Strategy provider: ClearGate (https://cleargate.capital/)

Cellar managed by: 7Seas https://www.7seas.capital/)

Cellar address: 0x6b7f87279982d919bbf85182ddeab179b366d8f2

Etherscan: https://etherscan.io/address/0x6b7f87279982d919bbf85182ddeab179b366d8f2

Source: cellar-contracts/Cellar.sol at main · PeggyJV/cellar-contracts · GitHub

Audits (Macro): Sommelier A-3 | Macro Audits | The 0xMacro Library

The strategy for the cellar is provided by ClearGate, including strategy development, continuous maintenance, and signal generation. Seven Seas Capital has developed tooling and infrastructure as per Governance proposal SIPS-012 and will manage the proposed Cellar. Management of the Cellar means that Seven Seas will act as an intermediary institution by providing strategy integration services to ClearGate, including forwarding strategy signals generated by ClearGate to be submitted on-chain for execution.

2. Proposal for cellar authorization for strategy ETH-BTC Momentum

This proposal is for the authorization of ETH-BTC Momentum Cellar. The strategy for the cellar is provided by ClearGate.

ETH-BTC Momentum is a dynamic long-only strategy for BTC-ETH portfolio management. It is more responsive to recent market changes and slightly more risky than BTC-ETH Trend strategy. Detailed information about the strategy, including strategy description and backtesting results can be found here: Strategy: ETH-BTC Momentum – ClearGate strategies

More information and comparison between the two simultaneously proposed strategies by GlearGate can also be found here: https://www.sommelier.finance/blog/deep-dive-on-trend-and-momentum-strategies

If approved, the chain will accept signed function calls submitted to the cellar contract from the cellar manager.

Name: ETH-BTC Momentum

Cellar share token: ETHBTCMom

Platform fee: 2% (1.5% for strategy provider + 0.5% for protocol)

Performance fee: 10% (7.5% for strategy provider + 2.5% for protocol)

Strategy provider: ClearGate (https://cleargate.capital/)

Cellar managed by: 7Seas (https://www.7seas.capital/)

Cellar address: 0x6e2dac3b9e9adc0cbbae2d0b9fd81952a8d33872

Etherscan: Cellar | Address 0x6e2dac3b9e9adc0cbbae2d0b9fd81952a8d33872 | Etherscan

Source: cellar-contracts/Cellar.sol at main · PeggyJV/cellar-contracts · GitHub

Audits (Macro): Sommelier A-3 | Macro Audits | The 0xMacro Library

The strategy for the cellar is provided by ClearGate, including strategy development, continuous maintenance, and signal generation. Seven Seas Capital has developed tooling and infrastructure as per Governance proposal SIPS-012 and will manage the proposed Cellar. Management of the Cellar means that Seven Seas will act as an intermediary institution by providing strategy integration services to ClearGate, including forwarding strategy signals generated by ClearGate.

3 Likes
2

I have a few questions regarding the fees and rebalancing of the portfolio:

  1. How were the fees decided? 10% Performance fee might need to be justified (i.e. how does it compare to other portfolios of this similarity)

  2. What are the metrics that determine when the trend is ‘down’? This should be somewhat mentioned given the fact that the whole portfolio gets rebalanced to not hold any ETH-BTC. Is it a Moving Average type situation? RSI? Bollinger?

1 Like
3

Thank you for the questions! Please find the answers below:

  1. It is up to the strategy provider to propose the performance fee. We feel that 10% is in the lower end of the spectrum or at least at par for similar actively managed strategies.

  2. The exact metrics or methods which will generate signals are proprietary. What I can say is that ClearGate strategies get the signals from data science models which usually combine multiple methods, including machine learning models. Neither weights of the inputs to the final signal nor the models itself are static, and can dynamically change depending on market conditions or risk management principles.

The aim of the strategies is to provide an attractive risk-return tradeoff and the strategies are thoroughly backtested.

1 Like
4

Thanks for the explanation!

5

Hi Cleargate,

What was the training, validation and testing periods for these strategies. Is there any live performance you guys did at your end to check overfitting of the strategy. Most of the time when u start using ensemble approach to predict signals there is high risk of overfitting. Can i get metrics like Worst month, drawdown, annualised sharpe, total return, sortino for training, validation and testing separately?

6

I will provide a short answer and a long answer.

The short answer is that ETH-BTC Trend strategy (the core version aka version 1.0) has been running live from July 10 and has produced a total return of 12.4% till Oct 16. ETH-BTC Momentum strategy (core version) is down 0.3% for the same period. At the same time spot ETH is up 11.7% and BTC is down 8.5%, both with much higher volatility than the strategies. The enhanced versions of our strategies have done better but have not been live.

Past performance is not indicative of future results, regardless of whether past performance is based on backtests or live results.

The long answer:

The overfitting problem is always present when developing quant strategies. It was one of our main priorities not to take the road of getting and presenting the best possible backtesting results, but to develop strategies that provide certain risk-return characteristics and follow a predefined general theme (like momentum or trend) which has proven itself to be worthy in various markets.

One example to illustrate that we’ve taken clear measures not to overfit our models is the issue of periods when prices appreciate very quickly. This has happened in the past and will happen in the future. The only question is how often that will happen. Data shows that it might not be that regular phenomena and we face a question whether we should build a separate model into our strategies so that the strategy would be aggressive when it happens. We deliberately chose not to. This is because it would more likely lead to overfitting and, would introduce unnecessary noise and risk for the type of strategies we are currently presenting.

We know that e.g. some of our Trend strategy users might feel slightly bad when ETH happens to go up 20% one month and the strategy makes only half of that. This is the unfortunate trade-off. You may get slightly better return in such a situation with our Momentum strategies but may lose some edge in the longer run. Or you can combine the strategies if you wish. And we will bring some more aggressive and shorter-term strategies to Sommelier when the time is right in the future.

We’ve observed how the strategies function in real time for some time now. So far, they behave as expected for long-only strategies in the bear market.

The following is a geek speak, which can be skipped as it is just meant to clarify that we have considered various technical and non-technical factors when implementing the strategies and why things are a little bit more complicated than posting here additional validation and testing results compared to what are already available:

The general background about our strategies is that we can see two major market cycles in crypto (this is of course subjective view because we did not consider the early days of crypto). One peaked in Jan 2018 and the other in Nov 2021. And during our 20+ years of being in the business, we’ve seen more market cycle in various TradFi markets as well. So, we took the lessons from previous cycles and implemented a general model and backtested how it works for a cycle starting from 2019 for our crypto strategies. We saw that it worked quite well. The only problem, it discards recent period which already has more data about various crypto assets. So, we implemented “another layer” that considers new data as well. The overall model has other “layers” that have to make decisions about portfolio asset allocation proportions etc. So the models become “slightly more complex”.

Since we are dealing with market cycles, we have the problem that generally markets tend to behave differently in various market phases (e.g. google for: volatility asymmetry around the world). When using a traditional training/validation/testing process we are either throwing away data or getting unexpected results due to changes in market conditions. To (somewhat) overcome that, we used a cross validation type of approach; or iteratively train a new model. The problem is of course that the simplest type of cross validation may not work that well because we are dealing with time series data with autoregressive properties. And one can always say that we can introduce look-ahead bias with that. So we carefully considered all of that when implementing the models and did our best to avoid any biases or overfitting, and still make as much use of recent data as necessary. It will not be that difficult or complex in the end but those are just some factors that need to be considered. And it does not make implementing the models just as straight forward as choosing a training and a validation period, implementing a model, and we are good to go.

Another factor to keep in mind is that we are not predicting the market with those strategies but reacting to market signals and assessing the probability whether it is worth keeping or dropping market exposure. Thus, any traditional data science/machine learning model evaluation metric such as RMSE or MAPE is not that relevant for the overall objective. Even with ensemble approach, we also executed human judgement during the development phase, what to include and what not to include in the final model. Some like that approach, some may dislike it, but it’s our philosophy that human judgement during model development phase can help to improve the results, but any human biases will be taken out during the strategy execution phase.

End of geek speak.

7

Thanks, the strategy which is live. Can i see the trades etc on-chain?

8

Yes - Here’s a link to one of the first trades from the ETH-BTC Trend strategy. Have you been able to participate in a strategy yet?

1 Like
9

thanks, I want to wait for 2-3 months to see live result then i would put money. Anyways, all the best.