Crypto Trading Model Development

Table of contents

Introduction

The trading algorithms which are utilized for the crypto model were taken from the Stock Trading Model, discussed in further detail here.  The primary difference in approaches is that the crypto model attempts to remain trading in Bitcoin and Ethereum practically non-stop whereas the stock trading model selectively chooses rare price occurrences for trade entries.  The crypto model therefore requires the additional usage of trending algorithms in addition to the reversion algorithms which are exclusively used for the stock model.

In a perfect scenario the model would utilize the trending algorithms to measure/determine the trend direction, place a trade in the attempt to ride and profit off the trend, and then once the price becomes materially overextended the reversion algorithms would tell the model to exit the position. In reality however, trends stop randomly and change directions, or price goes through periods of whipsaw which results in a series of losing trades in both directions. There's no holy grail, but if you select products that have significant price swings even a terrible trading system can look magnificent during some periods.

Backtested Results

The historically modeled returns for the crypto trading model described here are, to put simply, astronomical. This is to be expected with a price and time series of the magnitude of volatility and performance of Bitcoin and Ethereum. Although it should be noted that the modeled returns of this systematic trading model exceed that of a buy and hold strategy, which by itself is impressive given the returns of these cryptocurrency products. Think about that again, if you used this trading model instead of simply buying Bitcoin and/or Ethereum, you would be ahead today. I personally think that speaks volumes as to the robustness of the model, but I am bias.

The Buy and Hold returns for Bitcoin as of March 1, 2021 is on the order of magnitude of 9,700%, and Ethereum is 26,300%. The backtested Crypto Trading Model returns going Long and/or Short were 36,400% and 140,600% for Bitcoin and Ethereum, respectively. This model is not curve fit; it simply utilizes algorithms which I had already developed over the last 10 years. I simply plugged in the cryptocurrency prices into the processing software and hit go.

The Bitcoin price series begins on 10/1/2013, and the Ethereum price series begins on 8/9/2015. We used daily price data sourced from Coindesk.com.

Obviously, it should not be assumed that strategies which were profitable in the past will be profitable in the future. By closely following this Crypto Trading Model you and I are assuming that the price action of Bitcoin and Ethereum which made it so successful historically will continue into the future and in a manner in which the algorithms utilized can capture enough of the price variation to be successful. We cannot know the future however and so we cannot say that it will turn out that way.

Returns in the modeling do not include trading costs, slippage or any fees, which can be significant in the case of cryptocurrencies and could cause enough drag (on what otherwise would be a profitable trading strategy) to make it not worthwhile.

The main issue I see with this model will be with users ability to follow the strategy. For one thing, the win rate is barely over 50% and the other concern is that it requires the user to monitor daily and rebalance often. As to the first point, refer to the chart below:

As you can see from the histogram above, almost half the trades will generate losses. The winning trades are simply much larger in magnitude than the losers, which provides the model with trading suitability. Some additional statistics are provided below:

Kelly Criterion

If you apply the Kelly Criterion math individually to the Long and Short statistics, each holds up to being a winning strategy and each performs well enough to be suggestive of adding leverage. We cannot count on things in the future turning out like they did in the past, especially with something as novel as Bitcoin and/or Ethereum, therefore we will not be employing any leverage above 100% of the modeled capital in our account.

Last point of evidence of model robustness: if you remove the top decile of winning trades on the Long side, which is like assuming those wins were anomalous, this system would still be profitable and worth betting, according to the Kelly Criterion math. Unfortunately, the Short side does not pass this same test. It is not easy to short Bitcoin and Ethereum and I'm assuming most visitors of this website will not be duplicating those bets. However, we will continue to track them anyway.

Betting Process

As with our Stock Trading Model, we are taking a unique approach to money management as it's being applied to the Crypto Trading Model. The goals are fairly straightforward:

  1. Maintain exposure to Bitcoin & Ethereum in roughly equal portions
  2. Limit Drawdowns
  3. Bet Long and Short when warranted

In order to accomplish the first goal, we will bet equals portions to each Bitcoin and Ethereum as long as the algorithms dictate approximately equal probabilities of success. This however brings up circumstances when one has a strong "signal" (or probability) and the other doesn't; what should we do with the excess cash which would normally be allotted to the weak signaled cryptocurrency? We decided that in these cases, as long as as the other cryptocurrency does indeed have a strong signal, the excess cash should be bet on the high probability outcome. The result is that on somewhat rare occasions, maybe a couple times a month, one of the model weightings will vastly exceed the other.

Regarding the limiting of drawdowns, the second goal, we are going to apply an Anti-Martingale styled-betting approach. We use a similar approach in the Stock Trading System and we generally believe this approach serves us well when the positioning is broadly "incorrect" and exposed to numerous consecutive losses. There are times when the Anti-Martingale approach does not serve us well, as can occur during a one day selloff followed by a reversal however, our research shows that limiting drawdowns makes up for it all in the long run.

Goal number three is fairly straightforward but we are going to limit our exposure to the Short side somewhat simply due to the nature of these products and the tough nature of shorting markets in general; markets like to take the stairs up and elevator down. We do this by limiting some of the slower algorithms' impact on the Short position weightings, these slow signals do not perform as well on the short side but we would rather cut exposures rather than muting their signal. You never know when these markets can change and we may be able to take better advantage of the short side, if that occurs I want the broad range of signals which have been proven to work in aggregate.

Summary

That's it in a nutshell! If you've read this far congratulations, you may be the analytical type who could/should be designing your own trading models. If you have any questions feel free to reach out to me on Twitter, our handles are below; I realize there is a lot I left out but I'm not sure how far into the weeds to go on something which is honestly not that digestible. Thanks for reading!


Follow @FuseTrader and @Dyer440 on Twitter for any suggestions or inquiries.

All trading and investing strategies come with the risk of loss, including this one. These trades may not be appropriate for your investment goals and requirements, and it is not investment advice.  It should not be assumed that strategies which were profitable in the past will be profitable in the future or will equal the performance of the securities on this page.

Spread the word