Here is the performance of Calvin (my wife’s crypto tutor/crypto hedge fund manager) two fully automated algo trading accounts (personal) at Binance for the past day.
Based on my understanding, they are momentum based and trading at no leverage. All the trading instructions are fully automated initiated by a python program and linked to Binance through API.
It successfully capture the downward movement of the crypto market in the past trading day.
My wife is trying to finish her python script (the best 3 year backtesting is SNX (10 min data) with a 81% AR and 3.3 Sharpe and 30.7% MDD.-less than ideal.)
I have offer to buy Calvin’s program but he refused and only willing to share a few basic non-price data (he used over 50+) despite a very high all cash offer. He is concerned about capacity constraint as he used almost the same algo (at a higher frequency) for his crypto hedge fund at Amber Group.
Here is the list of price and non-price data at Glassnode.
I have an active subscription (Enterprise T-3) and glad to share the Glassnode data if there is anyone interested to try and build a python algo program and share results, pls send me an email.
From the web site: " * Up to 10-minute resolution"
Interesting use of the word “resolution” BTW/
I would just be interested in what time-frames you or your wife are trading in (e.g., end of the day or frequently during the day). And whether that is fully automated. Whether your wife is at a terminal most of the day or if the computer is executing some (or all) of the trades without her.
Thank you for sharing. And thank you for your response below!!! Very interesting.
In any case, not a lot of latency in your program. 10-minutes to a day with some interest in moving completely to 10-minutes, if I understand.
As you already know, we have been trading crypto for a while at Binance using limit orders with some success (mainly shorting cryptos with time series momentum during bear market).
We are still backtesting/experimenting fully automated algo trade which is what Calvin is doing for his own accounts and his hedge fund. However, without knowing the exact non-price data that he is using and how he applies them, we are unable to replicate his results. Calvin suggests we explore with the non-price data at Glassnode which provides upto 10 mins real time price data and EOD non-price data which is also one of the data sources he is using. He currently has a remote PC (cloud) linked to Binance through API trading non-stop 24 hours a day at mid-frequency (1 min to 15 min timeframe). Most of the algo traders at Binance (my best guess is at least 25% of Binance total daily trading volume) are connecting to Binance API this way and trade automatically (to and prevent windows update, PC hangs and other potential problems with PC …etc)
I have been using for a while https://charts.coinmetrics.io/.
You can download to csv for free many factors, prices and calculate backtest.
After doing some (basic) research, I found that in crypto most dominant factor is momentum.
Value factors does not work as expected (for example NVT variations). Crypto with high NVT outperform crypto with lower NVT possibly due momentum factor. But maybe other factors will work.
Thanks for the info. I have checked the link you provided.
The only thing better about Glassnode is that it has over 200 non-price data (factors) to use for T3 subscription (up-to 10 min) provided continously throughout the day through API for 24 hour trading that can be linked to any large exchange plaform. (even T1 free subscription contains more than 20+ non price factors - EOD data.)
Maybe Marco can consider adding those free EOD blockchain data to P123 by contacting Glassnode.
Here is the link again for comparison T1 (up to daily), T2 (upto hourly) and T3 (upto 10 mins) and the range of non-price data (factors).
I don’t have access to the Binance trading bots so it’s unclear to me what advantage buying Calvin’s bot would provide. Are there limitations to using the bot on Binance?
Binance has trading bots (like DMs) that can trade for you designed by Binance. From my observation, they only offer very average performance.
Calvin is not selling his bot or “follow trade” service. In fact, he is only providing the basic python tools for my wife to conduct backtest and run a sciprt through API at Binance with Glassnode data (seperate subscription).
He is not sharing anything in details except “high level” advice, like using momentum and factors like “exchange flow” from data at Glassnode.
We have been experimenting/backtesting with the basic script he provided based on moving averagemomentum for a few months now but can’t replicate his results so far since he is not disclosing the non-price data that are used. (a big trade secret/secret sauce for his hedge fund I guess)
Thanks James. I now see what you want to do. I assumed that Calvin’s model was followable. Are Calvin’s Binance results just to humble brag? Feed clients to his hedge fund?
The leaderboard at Binance are updated daily for all their users. I don’t think they are biased. (I have seen his account daily ROI down more than 5-6% in one day.)
From what I learnt from Calvin, part of the reason he maintain these “personal” accounts is to show that his algo works and leaves a track record for future job opportunities.
He is not running his own hedge fund, he is a Quant Portfolio Manager running one of the books for a hedge fund at Amber Group.
