Ex-Millennium Veteran Shares Crypto Insights
In last week’s update we teased the announcement of a major addition to the Valmar Team, and this week we are pleased to shatter the suspense: Steve Armatas has joined Valmar as Senior Portfolio Manager.
Mr. Armatas boasts loads of relevant experience and notably spent 16-years at Millennium Management managing a high-performance statistical arbitrage strategy, designing and developing risk management, portfolio management, and research frameworks, and interviewing/hiring portfolio managers. He deeply shares Valmar’s multi-strategy, multi-manager vision, and his experience is crucial to establishing Valmar in that Millennium model. To say we are excited to work with and learn from him is an understatement. Mr. Armatas has already brought sophisticated and sober thoughts to the table, and today we will use this blog to share insights from an individual who helped build the big thing in tradfi and is now up to it again in crypto. Who better to listen to?
Manager Sourcing and Support
At Valmar, we firmly believe the multi-strategy, multi-manager model will provide a necessary alternative to single strategy funds and fund of funds (for a myriad of reasons, including diversification, capital efficiency, active risk management, consolidated technology, etc.). And a successful multi-manager platform really serves two client bases: first and foremost, its investors, but also its managers. To have and maintain the best portfolio managers in the world, Valmar must differentiate itself in terms of service and support for those managers. What does this mean to Mr. Armatas?
Focus on the needs of the trader. Provide tools, interfaces, data that make it easy for him/her to connect and trade through Valmar’s platform. This sounds simple, but few firms truly do this well. Many firms provide no type of support whatsoever. Other firms may have a low-latency architecture that works well, but the trader is left to their own connection to it, and it often contains errors, bugs, design flaws, etc. Just avoiding the errors of other firms is a competitive advantage.
It’s good to have flexibility in how traders can connect to the platform. We should adapt to the traders’ needs and not force them to adapt to us. As much as possible, it should be designed so that managers need to do little or no development to start trading. For example, some traders may want to just send in target positions, some may want to send orders, some may want to send files, some may want to connect to through an API, etc.
Platforms should expect a certain amount of manager turnover. There will be a continual need to onboard new managers so the onboarding process should be completely automated and as cost-free as possible.
Over time it is crucial to build an extensive reporting capability. More information leads to better decisions and processes. Reporting should include detailed reports and plots that can compare return series from any source: live, simulated, track record, and paper trading. Also, the same should be done with portfolio exposures and position sizing. Over time, this will create a huge knowledge base and provide a detailed understanding of the characteristics of different types of models. This will help in evaluating new candidates and for managing existing traders.
The same reporting system can support manager monitoring, checking for style shift and live monitoring. The philosophy should be to have one way of doing calculations and charts, and repeating that where necessary, not having separate ways of running reports for different portfolio sources. As appropriate, reports are shared with traders.
Crypto native traders will especially benefit from more support. Even with a high level of support and tools there may be traders that will have difficulty being able to independently maintain a trading platform on their own. They may have discovered an alpha and built a trading bot, but it is a very difficult task to be able to independently maintain a database, connect to a live price feed, maintain a web-server, keep accurate positions, validate prices and other inputs, and gather social media data.
Manager Monitoring
Of course, sourcing great managers is only part of the battle with an active multi-strategy model. In crypto, markets are 24/7/365, and so strategies must continuously be monitored. How does Mr. Armatas view monitoring?
Start basic. What are the exposures? How do they vary across time? How sensitive are they to intraday volatility? Does it match what’s expected? Does it make sense? Does it match the DDQ? Does performance match the previous track record or simulation? Reports should be run daily and in real time. The best way to learn about a model is watching live trading in different market conditions. How does trading volume vary with changes in market volume? Typically, they are correlated. Same with volatility. What is the average slippage? Does it match the slippage used in backtesting?
Understand the baseline behavior of a strategy. Do slippage and average fill percentage get worse after increasing investment? What do the portfolio characteristics look like? We would expect to have more concentration in large cap and less volatility.
Model performance and drawdowns must be looked at in a relative sense. How do they relate to market conditions? If returns are lower than expected what has the overall market volatility been? Is the drawdown a quick sharp drawdown that follows an extended period of large gains? Have all similar models had drawdowns as well? Is the drawdown tied to an economic event? During covid many quant books got killed early in the year, but the gains after were tremendous. Having good contacts in the industry is very useful.
Poor model performance doesn’t always lead to a large drawdown. Often, a model will capture alpha but not enough to offset trading costs and the net P&L will resultingly underperform. Tracking error is probably one of the most important metrics, though it may not always be available. Similar strategies should be compared with each other.
Risk Management
One of the most important lessons learned in this year’s past washout has been the lack of robust and institutional grade risk management. Providing an extra layer of detailed risk management is a foundational element to a multi-strategy platform. Risk management in crypto can be tricky because it requires a deep knowledge of traditional risk management frameworks that then must be applied to the unique curvature of crypto risk. We’ve talked about this before. And Mr. Armatas has quite a few insights here as well.
INTRADAY MARKET RISK
Sudden and extreme price changes occur in crypto. It’s important to anticipate these and have plans in place for how to respond. For consideration: In what situations should trading be paused? Most models are not prepared to trade through extreme intraday volatility. It’s unlikely that they have been tested or traded through such periods. If the research for a model was conducted on daily data or 1-hour bar data or sometimes even 5-minute bar data, extreme events were probably missed.
What types of price changes and drawdown are possible (use lowest frequency data)? Mean-reversion type models typically do poorly during market crashes.
It may be best not to trade during those volatile periods. It is more prudent to sit out than chance liquidating a book intraday during a crash.
MODEL RISK, CONSTRAINTS and STOP LOSSES
Each model should have specific constraints and stop-losses tailored specifically for it. It won’t be possible to apply the same limits to every model, even if the strategies are similar. There are many considerations that should be taken into account when setting risk limits on a strategy. A lower frequency model will have larger drawdowns than a high frequency model. A high frequency model should have tighter liquidity constraints.
Managing model risk is more nuanced than just having a specific stop-loss. Stop-losses may be necessary for disaster prevention, but they should not be the only tool in managing model specific risk.
Three approaches to evaluating and managing risk on a specific model/strategy include: stop-losses, managing drawdowns, and determining if a model is not working properly and should be shut down.
Stop losses are necessary because they allow us to define the maximum possible loss on the book. They’re useful because they remove the need for any judgement call or discretion, and they may be the only way to manage risk if there is no information to make a judgement call, such as allocating to an external fund or for a new manager whose returns have not been audited. However, there are many bad characteristics about stop-losses. It’s quite difficult to set a good limit. Past performance is not very useful because there’s most likely a limited simulation window, and crypto is constantly cycling through different market regimes. Real-world drawdowns are often greater than expected. Stop-losses also don’t consider the current situation. How are other similar models doing? Is the drawdown due to some economic event? During covid, quant equity models had extreme drawdowns and many books were liquidated and then quant equity came back to have an excellent year.
Stop-losses should be set at a level that is expected to be unlikely to be hit. The number should be determined by many factors. Largest observed drawdown. Is the historical data simulated or live? How long is the live track record or simulation? Again, it’s a hard number to define. A good starting point would be to set it at 50% greater than the max. observed drawdown, especially if the historical track record is short.
Determining when to shut down a book is one of the most important decisions a manager can make. Risk control needs to be balanced with other concerns. Shutting down books, for whatever reason, will lead to a higher manager turnover and will also in some cases lead to shutting down good models. Higher turnover will cause some managers to feel mistreated, which could lead to reputation issues and increase the need to find new managers.
Managing drawdowns is difficult. There is not much that can be done. It’s always important to try to understand why. Are market conditions different? Which type of models are affected?
Setting the investment allocation correctly is a key element. Setting too tight a stop-loss on a model simply guarantees failure. It’s much better to set a lower allocation and a looser stop-loss. Having a single drawdown limit for all models of a similar style will result in rejecting good models because they have too high a drawdown. Why reject a model that makes 100% annually with an expected max. drawdown of 15%? Again, it’s better to set a lower allocation.
RISK - INDIVIDUAL COINS
One of the largest sources of risk is individual coin risk. Even large market cap coins can go to zero in a day or two. It’s even worse for short positions. 100-300% return in a day is completely possible. There should be a stop-loss on individual coin positions enforced at the firm level. Once a level is exceeded the position should be liquidated and put on a restricted list for a certain period. The stop level should be set relative to the average coin P&L or daily return.
Research should be conducted to group coins into different categories of risk, which would have a different stop-loss threshold. Historical volatility or beta could be used, but other metrics could be better.
RISK - PERFORMANCE and P&L MONITORING
There should be real-time monitoring and alerting of all exposures and P&L. This can be supported by the reporting framework and there should be policies and ways of implementing emergency responses. Market circuit breakers could be used. It should be possible to pause or stop all trading for a single book or for single exchange or for all accounts within seconds (“kill” button). Policies on when to do that should be made or considered ahead of time. Also, automated liquidation of a book should be built in.
Conclusion
As we’ve said before, it is easy to say multi-strategy and multi-manager. At the end of the day, the business is sourcing great talent, monitoring the strategies, and providing a fulsome risk management program. But what does that really mean? The devil is truly in the details, and there are quite a few details with this paradigm.
The vision is big, the work is big, but crypto needs this done in the correct manner. We are proud to help lead this charge, and now substantially stronger with Mr. Armatas as part of the Team. It’s a bit easier to build the “Millennium Management of crypto” when you are working with one of the individuals who helped build Millennium Management. We hope his insights are as luminary for you as they are for us.