Can Triple-Digit Annual Returns Last?

Unveiling Crypto Centaur’s Maiden Live Trading Results

Published in

Coinmonks

11 min readMay 9, 2024

In December 2023, we announced the first live trading beta-run of our Crypto Centaur Long/Short Momentum Strategy, which started on November 17, 2023.

In today’s update, we’ll cover the results of our first live trading beta-run from its start on November 17, 2023, up to April 30, 2024 — a period of five and a half months. We’ll also look at backtested paper trading results to shed some light on potential long-term performance. Finally, we’ll delve into the sources of our returns — “If you don’t know why your strategy is making money, it’s probably just luck….”

(All data below reflect net performance, accounting for all trading costs, management and performance fees.)

1. Actual Performance of Live Trading

The simplest and safest cryptocurrency investment strategy is often considered to be buying and holding Bitcoin. Our goal is to exceed the performance of this baseline strategy over time, thus our performance is benchmarked against Bitcoin’s returns.

For reporting purposes, we will track and analyze the performance of two leverage levels of our Crypto Centaur strategy:

Savage Surge (SS) — Medium leverage, and
Insane Gain (IG) — High leverage.

The chart and table below (Figures 1 & 2) summarize the performance of the strategy’s two leverage levels against the performance of spot Bitcoin/USD (buy and hold).

During the 165-day beta-run with live trading, both of our strategies outperformed Bitcoin in terms of total returns. Although Bitcoin exhibited lower volatility, our strategies provided superior risk-adjusted returns.

While we’re excited about the strong performance so far, it’s crucial to remember that these results, particularly the annualized ones, may not be realistic since they’re based on such a short period. For instance, the CAGR of 1327% for IG and 408% for SS assumes that the impressive returns we’ve seen over 165 days will continue compounding over an entire year, which is unlikely. Similarly, claiming a Sortino Ratio of nearly six for both leverage levels might be over optimistic, suggesting these results might not accurately reflect the strategy’s long-term performance potential.

Perhaps a more meaningful comparison would be between our live trading performance and the back-tested results, which spans a much longer period. We’ll delve deeper into this in section #3 below, but first, let’s examine how the back-test performance stacks up against Bitcoin.

2. Back-tested Results: 14 Feb 2018–30 Apr 2024

The earliest date that offered sufficient liquidity in Altcoin futures for our strategy was February 14, 2018. Therefore, we’ll begin our back-test paper trading from this date. To maximize our analysis time-period, we’ll combine the back-test period with the live trading period’s data. Readers familiar with financial modeling and strategy development might be rolling their eyes now 🙄, understanding the critical need to separate in-sample and out-of-sample data. However, since the strategy was developed adhering to this principle, merging these datasets for performance analysis now, doesn’t present an issue. At this stage, we’re simply comparing actual to theoretical results to assess the validity of the live trading performance up to this point. Concerns like overfitting, look-ahead bias, and data snooping will be addressed later in this article, in section #4.3.

In the chart and table below (Figures 3 & 4), we compare the performance of the two available risk levels of Crypto Centaur against that of Bitcoin over the combined period (14 Feb 2018–30 April 2024) which includes live trading of 165 days (5.5 months).

Over the combined period, both leverage levels of Crypto Centaur surpassed Bitcoin in terms of total returns and risk-adjusted performance metrics. Savage Surge exceeded Bitcoin across all three risk metrics, while Insane Gain outperformed in terms of maximum drawdown, but showed higher volatility with respect to standard deviation and downside deviation. In our view, underperforming in terms of standard deviation is irrelevant, as the bulk of the volatility observed was to the upside. — both positive and negative swings are captured by this measure. We prefer using downside deviation to gauge volatility since, like most investors and traders, we like upside volatility.

The correlation of 7-day returns between both strategies and Bitcoin returns is notably low, registering at only 0.3. This is important to note as it indicates that the strategies can have positive returns when Bitcoin is in a downtrend and vice versa. The low correlation can significantly enhance the performance of a long-only crypto portfolio.

3. Comparing Actual Live Trading To Back-tested Results

The table below (Figure 5) compares all performance metrics of SS, IG and Bitcoin during the live trading period (165 days), with the combined back-test paper trading performance results (2268 days).

3.1. Comparing Return Metrics

Over the combined period from 14 Feb 2018 to 30 April 2024, the Compounded Annual Growth Rate (CAGR) of IG drops from 1327% in the live trading period, down to 386%, which makes a lot more sense, albeit still very high (We’re not complaining). For SS, the CAGR drops from 408% down to 177%. We left out total return metrics as comparing these over different timeframes is senseless.

It is useful to look at the CAGR of Bitcoin over these two time-splits as these results are free from any “strategy development risks” because all the returns are actual results and not contingent on any complex algorithm to work. Noting that Bitcoin also outperformed over the shorter time-split as opposed to the combined period, shows that the live trading period was a period of outperformance for the crypto market overall.

3.2. Comparing Risk (volatility) Metrics.

The volatility measures of the live trading period is well within the volatility range of the combined back-testing period. This indicates that the strategy performed within the expected bounds of volatility as implied by the back-test data. This is very encouraging to see and suggests that the algorithm is doing what it is designed to do — especially as far as the risk control overlays are concerned.

Once again, the abovementioned does not apply to the standard deviation (STD) measure as the upside volatility is included in that metric. The higher STD for the live trading period compared to the back-test, as before, emphasizes the outperformance to the upside over this shorter time-split.

3.3. Comparing Risk Adjusted Returns

The Sharpe and Sortino ratios also come down in line with more realistic levels when comparing the live trading period to the back-test. We see a similar pattern for Bitcoin which further points to the live trading time-split being a period of outperformance for the crypto market in general — which is what you’d expect in a bull market with relatively shallow drawdowns as observed.

4. Can the Back-test Results Prior to the Live Trading Period be Trusted?

4.1. Live trading results build confidence

When we started trading live and had only the back-test data to rely on, we were much less confident about the validity of the results. However, after this first beta-run, we have much more conviction — even though it is still a relatively short period of live trading. We think the analysis above speaks for itself — although it can’t prove anything, it does build substantial confidence.

4.2. Tracking

An important question to ask is: How closely does the actual results track the theoretical returns of the trading model? If there is a material deviation between the actual trading and the theoretical paper trading returns during the live testing period, then there might be a problem — even if the actual results are better than the theoretical returns. The chart below (Figure 6) shows the actual capital balances of both leverage levels of Crypto Centaur (indexed to 1000) as compared to the theoretical paper trading results for the same period.

4.3. Strategy development pitfalls

We are well aware of the risks associated with overfitting, data-snooping, lookahead-bias and fine-tuning of a systematic strategy. We’ve learnt this the hard way over many years of research, testing and tinkering. We’ve applied system development techniques to mitigate these risks in building the algorithm, but only time will tell if we were ultimately successful.

We have not added any parameters or features without any mechanical purpose in the system. In other words, each feature and each parameter has a logical reason for being included and no arbitrary elements were added for the simple reason that it improved the back-testing results.

Lastly, all features demonstrated a narrow range of positive outcomes when tested over a wide range of parameter input values. This means that each feature is robust enough to work without relying on finetuned parameters. Said differently, the features are insensitive to specific parameter values. This is, in our view, the definition of a robust system feature.

4.4. Liquidity and Market Impact

Live testing that yields promising results is important, but how will the strategy perform when scaled up with significant assets under management (AUM)? An Important consideration for any new trading strategy is estimating at what AUM size the trading activity will start impacting the market, thereby increasing slippage. This unseen cost could render a strategy unprofitable, so it is essential to estimate this level accurately.

According to our calculations, the average combined volume of the top 40 perpetual futures traded on the two exchanges we use (Bybit and Binance) is approximately $36.5 billion per day. During our live testing period, the maximum volume we traded in one day was $2.92 million per $1 million of AUM, with an average daily volume of $250K per $1 million of AUM.

In trading, the rule of thumb is that trades below 0.1% of the total exchange volume typically incur negligible slippage. Trades between 0.1% and 1% can start having a moderate impact, and above 1%, slippage could become a significant issue.

The calculations below show that on extreme days, where we might trade 1% of the total exchange volumes for the top 40 instruments, we would only reach this 1% threshold with AUM of $125 million. On average trading days, we could manage capital of $1.46 billion before reaching this threshold. This calculation does not account for the fact that total exchange volumes typically increase substantially during volatile market days.

In summary, we are currently very far from these theoretical AUM limits, ensuring that slippage remains a manageable factor in our trading strategy.

5. So where is the edge?

If you cannot explain your edge, you probably don’t have one.”

We believe the strategy’s source of alpha is derived from:

The general inefficiencies present in a nascent market like crypto. This new asset class is widely misunderstood by many (some very smart) market analysts (Think Peter Zeihan, Nouriel Roubini and Nassim Nicholas Taleb). Therefore, inefficiencies are plentiful.
Crypto futures traders are by and large short-term traders who look at single trades at a time while we have a portfolio approach with a longer time horizon.
We track cross-sectional momentum dynamics within the market as opposed to only looking at crypto assets relative to USD. This enables our system to pick up on many relative trade opportunities within the market which most observants might miss.
Our system is designed to pick up on Altcoin trends early in their cycle. Considering that in the past, some of these coins exploded higher tens of thousands of percentage points, the ‘too-good-to-be-true’ results in the tables above suddenly doesn’t seem that far-fetched. The key is to monitor as many coins as possible and to ensure the strategy has exposure to these trends from the get-go, while controlling downside risk.
We disregard all market narratives and news in our strategy which mitigates the risk of making biased trade decisions. The crypto space is a very noisy environment with wild swings between bullish and bearish narratives which can be extremely confusing. Having a systematic approach to making trade decisions, while disregarding the noise, is a trading superpower.
Our risk measurements are dynamic and adapts to changes in market regimes. This controls risk and optimises performance in different market environments.
Being able to monitor and actively trade 120 positions in real-time, is only possible through automated algorithmic systems and thus, this strategy is impossible to be executed by a human trader.
Lastly, we believe that human creativity and imagination, combined with the power of automated computation (In Chess, referred to as a Centaur), is far more powerful than either one on its own. The human role in our system is to constantly monitor performance relative to what could be expected in a given market regime to identify risks, but also to search for improvements and adaptions to drive the evolution of the system.

6. Conclusion

The Crypto Centaur Strategy relies on two principal factors:

The necessity for individual crypto assets to exhibit clearly defined trends, regardless of whether these trends are upward or downward.
The availability of at least 40 crypto assets with sufficient market liquidity to minimize slippage when entering or exiting positions.

As long as these critical factors remain intact, we are confident in the ongoing profitability of our strategy. At present, there is no indication that either of these conditions is at risk.

The outcomes from our initial live trading beta-run, as detailed above, are quite promising. Although the timeframe has been relatively brief and it’s premature to make definitive conclusions (as the unpredictable nature of markets always carries inherent unknowns), the results to date are decidedly positive.

So far so good…. and with some luck, we might see those triple-digit annual returns stick around!

Reminders

Who can invest in Crypto Centaur?

Currently, we are managing our own capital and have initiated a beta test for a select group of family and friends who are participating in our trading program. At this stage, we are not extending this investment opportunity to the general public.

Why publish results and write about our work?

We’ve launched this Medium blog as a platform to share our learnings and insights. Additionally, we’ll be publishing analyses and performance data of our strategies. Our aim is to enhance our profile and engage with our readers, hoping to attract insightful comments and questions that can further our learning.

For more information about Numeta Strategies and our approach, you can explore these two blog posts:

Catching Waves

An Introduction to Numeta Strategies

medium.com

Trading Turds or Transformative Technology?

Solving the Crypto Investment Problem