Index | Recent Threads | Who's Online | Search

Posts: 82    Pages: 9    Prev 1 2 3 4 5 6 7 8 9 Next
Last Post
New Thread
This topic has been viewed 3797 times and has 81 replies
sgmd01
Re: Bitcoin (PlanB : Stock-to-Flow model and Ben Cowen : Algorithmic Regression Rainbow)

James,

You can backtest gbtc on tradestation using their drag & drop technical indicators. These indicators can also be modified by opening up the code box. It looks like you can also backtest short indicators but I haven't played around with it long enough to understand how to do this. It's not as good as P123 (less user friendly & less options) but it's a start until the grayscale data sets are loaded into P123. Note that ytd bitcoin is up 65% vs gbtc 15%

Scott

Dec 15, 2021 1:30:14 PM       
Edit 2 times, last edit by sgmd01 at Dec 15, 2021 1:32:11 PM
ustonapc
Re: Bitcoin (PlanB : Stock-to-Flow model and Ben Cowen : Algorithmic Regression Rainbow)

Scott,

Thanks for the update.

I was hoping that there is also bitcoin (and ethereum) price data in tradestation for backtesting.

As you mentioned, there is a large discrepancy in price between gbtc and bitcoin and it is risky to assume that a GBTC backtest will also works for bitcoin. However, it seems that is what P123 and tradestation can do at best.

Regards
James

Dec 15, 2021 1:51:07 PM       
sgmd01
Re: Bitcoin (PlanB : Stock-to-Flow model and Ben Cowen : Algorithmic Regression Rainbow)

Hi James,

I think we can use GBTC and ETHE to approximate Bitcoin and Ether if we pay attention to the discount/premium. This hopefully will be included when the increased history is added to P123. GBTC and Bitcoin were tracking pretty closely until about a year ago when GBTC went from a 40% Premium to a 17% discount or a change of 57%.

Scott

Dec 15, 2021 9:09:48 PM       
ustonapc
Re: Bitcoin (PlanB : Stock-to-Flow model and Ben Cowen : Algorithmic Regression Rainbow)

Scott,

Alternatively, you can trade BLOK instead of GBTC/ETHE and not worry about the premium/discount issue.

If you look at the two charts below, the performance of cryptocurrency index and blockchain companies index are almost in sync in the past few years

I think I will continue to use my manual backtesting results with Binance and use P123 to backtest BLOK.

The Premium/Discount issue can be a huge problem (when trading spot bitcoin/ethereum based on backtesting of Grayscale trusts) if you look at the 5 year history of Premium/Discount to NAV (especially for ETHE).

Regards
James

Attachment HFR Blockchain Composite Index Performance History.png (83356 bytes) (Download count: 75)


Attachment HFR Cryptocurrency Index Performance History.png (83295 bytes) (Download count: 71)


Attachment ETHE Discount-Premium.png (54091 bytes) (Download count: 68)


Attachment GBTC Discount-Premium.png (63217 bytes) (Download count: 70)


Dec 15, 2021 11:18:23 PM       
Edit 2 times, last edit by ustonapc at Dec 16, 2021 1:59:49 AM
sgmd01
Re: Bitcoin (PlanB : Stock-to-Flow model and Ben Cowen : Algorithmic Regression Rainbow)

James,

Good point on the correlation between bitcoin and blok (I've noticed that too). That premium erosion on ethe is brutal. Bitcoin has diverged (taken off) relative to Blok in returns which started a little over a year ago (The divergence between gbtc and bitcoin occurred around the same time). Sticking with Blok and Bitcoin/Ether is a good idea. I'd love to have Bitcoin and Ether here rather than the Grayscale funds.

Scott

Attachment BitcoinVsBlok.png (119733 bytes) (Download count: 70) (Bitcoin vs Blok)


Dec 16, 2021 7:49:23 AM       
ustonapc
Re: Bitcoin (PlanB : Stock-to-Flow model and Ben Cowen : Algorithmic Regression Rainbow)

Scott,

I have a screen for BLOK and currently using it. (3 year annualized return 92% max drawdown 13%)

If you want it, I can share it with you. (just give me your email address so that I can send you the formulas or we can arrange a time via email to make it "public" so that you can download it from P123).

Regards
James

Dec 16, 2021 8:22:10 AM       
Edit 1 times, last edit by ustonapc at Dec 16, 2021 11:46:23 AM
ustonapc
Re: Bitcoin (PlanB : Stock-to-Flow model and Ben Cowen : Algorithmic Regression Rainbow)

Scott,

I have replied to your email.

The screen is for daily rebalancing and two of the formulas contained open(-1).

Those means sell BLOK at the next day market open if IEF or HYG drops more than 0.75% from previous close.

Regards
James

Dec 16, 2021 7:10:15 PM       
Edit 2 times, last edit by ustonapc at Dec 16, 2021 7:21:43 PM
ustonapc
Re: Bitcoin (PlanB : Stock-to-Flow model and Ben Cowen : Algorithmic Regression Rainbow)

Scott,

I am replying to your email here so that other P123 member can also benefit from our discussion.

But it looks like 20 - 50 day MA are effective for risk control on all of Grayscales funds. Using them does decrease the yearly return to around 50% but it also decreases the drawdown to about the same.


50% annualized return and 50% drawdown is suboptimal. You can easily achieve that with a high risk stock portfolio.
You should aim for at least 75% annualized return and 33% drawdown for a successful bitcoin strategy.

In addition, I don't think you should be over concerned about curve fitting when you are only dealing with one asset and only with TA. Overfitting is more a concern when you are picking a portfolio of stocks using lots of different factors.

Regards
James

Dec 18, 2021 10:18:03 PM       
sgmd01
Re: Bitcoin (PlanB : Stock-to-Flow model and Ben Cowen : Algorithmic Regression Rainbow)

James,

This paper discusses some of the risks in back test over fitting but is not comprehensive

https://www.ams.org/notices/201405/rnoti-p458.pdf

Pseudo-Mathematics and Financial Charlatanism: The Effects of Back test Over fitting on Out-of-Sample Performance

From the paper, "For instance, if only five years of data are available, no more than forty-five independent model configurations should be tried. For that number of trials, the expected maximum SR IS is 1, whereas the expected SR OOS is 0. "

It is very easy (for me) to over fit when back testing a security (i.e. Bitcoin with 6+ yrs & Ether with 3+ yrs.) with a short price history, particularly when there are very few transactions and many configurations are tested. The greater the number of transactions (i.e. number of securities X frequency of re-balancing X length of history ) and the fewer the number of trials then the lower the chance of over fitting. I tested several moving averages on several cryptocurrencies as a sensitivity analysis to determine whether the concept had any merit. It does for controlling draw downs at the expense of return.

You're much better than me as I can't come close to making 50% returns per year out of sample on a consistent basis with a stock system :)

Scott

Dec 18, 2021 11:25:29 PM       
ustonapc
Re: Bitcoin (PlanB : Stock-to-Flow model and Ben Cowen : Algorithmic Regression Rainbow)

Scott,

You are more concern about overfitting than me. (I don't know whether it is a good or bad thing). The paper that you quoted from Marcos López de Prado is from a while back and I have already seen it. You can find more of his recent papers on the subject here :

https://www.quantresearch.org/

I have made 50%+ annualized return out of sample for 3 years in a row, it is more about risk-off to GLD/EDV/TLT when necessary (I do daily rebalacing and may risk-off even for 1 day) AND add some volatility trades (VIXY) when market is in FEAR mode to boost return. For risk- on, I buy QQQ/XLK and occasionally BLOK (based on the screen that was sent to you earlier).

I will send you another message with more info. Pls check your email.

Regards
James

Dec 19, 2021 3:57:31 AM       
Edit 3 times, last edit by ustonapc at Dec 19, 2021 7:46:17 AM
Posts: 82    Pages: 9    Prev 1 2 3 4 5 6 7 8 9 Next
 Last Post