Forum Replies Created
Hi, thanks for the question.
We used to have short-selling signals, but we currently do long trades only. Our methodology has refined quite a bit over time so you should see any losing periods be managed much more tightly.
In terms of how the strategy plays out over time:
– During bull runs, we want to capture most of the gain.
– During bear runs, we want to avoid most of the loss.
So there still will be losing times, but over a few cycles, we come out as the winner.
Practically speaking, when our first public trade was issued, BTC was at $8,200. It has gone way up and way down since then. It has crossed $8,200 multiple times.
In fact, about 1.5 years later, it landed back at $8,200. For buy-and-hold folks, they had a 0% gain over that 1.5 years. But our system was grabbing most of the run ups and avoiding most of the drops. That resulted in us having realized gains.
We then hit $8,200 again in April of this year (over 2 years after the initial trade). At that point, we’re up 235% while buy-and-hold is again flat.
Hopefully that makes sense. Great question!Community AdvocateMay 26, 2020 at 9:52 pm in reply to: BCH Model Upgraded to 5.2 – Dynamic TSL and Profit Target #8957
Yes–I agree.Community Advocate
Hello–sorry for the delay.
As of now, we’re not seeing liquidity issues with Binance.us, but it’s something to monitor. But the most volume (best liquidity) is in BTC/USDT there.
In terms of fees, if you’re not trading a large account, the Coinbase Pro fees can be a killer for active traders. As you move up their volume stack, you’ll get more favorable rates.
We do also have the calculator here which can give some good approximations: https://crypto-ml.com/trading-bot-roi-calculator/Community Advocate
Yes–the signals have been very bearish. Over the last few weeks, volatility has been off the charts–even for the crypto markets. It was punctuated by a 15% drop in 7 minutes and Coinbase Pro subsequently going dark for 45 minutes. Activity around this much-anticipated halving is difficult for any human or machine to predict.
I agree, the signals didn’t call this last run properly. I’m going to copy in some info from the Telegram chat to hopefully shed some light on why we may see behavior like this.
From the Telegram discussion:
“It’s best to think about the predictions in terms of probability. Basically, we’ve had huge volatility over the last few weeks. Therefore its predictions will tend to be extreme. It is choosing bearish posture despite the recent move up. The bearish stance probably must be slightly more “likely” from its perspective than a bullish stance. (unfortunately).
For example, it could be thinking it’s 45% likely there will be a strong bear move, 40% likely there will be a strong bull move, and 15% likely there will be something in between. That’s what happens in high volatility.
Last night, these probabilities were likely very close must have flipped for a couple of hours. During that flip, we go an extremely bullish prediction. It was bullish enough to trigger both BTC-A and BTC-C (for the first time).
The action from our end is to train this recent volatility in. We, of course, would have preferred the ML captured this run.”
We have been running additional training all day today to follow-up on this action.
- This reply was modified 2 weeks, 3 days ago by Justin.
There have only been two halvings in the past. That means ML cannot come to a statistically interesting prediction based on that information alone. Rather, it will need to rely on the variety of other inputs that help it make decisions.
A few points to consider:
1) Many retail traders/investors seem to be of the mindset that we’ve had rallies after the last two halvings, so therefore we must have a rally again.
2) Institutional investors have likely already priced the halving in and see it as a potential selling point.
3) There are economic reasons for miners to collectively take actions to drive up price, thereby offsetting profit erosion.
4) It’s possible #3 has largely already happened.
With all of that said, the one guarantee is this is an unusual event with a lot of anticipation.
Personally, I believe the likelihood of an extreme move either way is likely. That is risky. When there is risk, it’s good to reduce position size.
I and others on the team will be following the signals for guidance.
What are others thinking?Community Advocate
Hey, very astute question.
And yes, you’re exactly right…if we ran the evaluations against new data, we could see a difference in precision (either + or -).
There are two ways to look at it:
– In terms of overall precision, it works out to needing around 15 straight days of 100% inaccurate predictions to slip overall precision by 1%.
– But if you looked at precision of that 15 days only, you would have a very low value.
We run training and evaluations throughout the month in order to feed in new scenarios and hopefully avoid this sort of thing. Practically speaking, we see the precision hold fairly constant.
But then we get curve balls like COVID-19. So it’s always an adventure. 🙂Community Advocate
The ETH model issued 4 trades in April locking in +36%, which is pretty exceptional. It has sat out for this last upswing, but I assume it’s looking for confirmation to break above these levels. We want to get most of the upswings and minimize losses on the downswings.
In terms of precision, that is referencing R-Squared, which measures the amount of error removed by the model. We use that measure because it correlates very well with how our models do once their output is handed off to the trading system.
It will never hit 1. And each incremental improvement is going to be much more challenging to achieve.
Here’s a link to read more: https://crypto-ml.com/blog/machine-learning-upgrade-to-5-0-deep-neural-networks/#R-Squared
And an image that shows why we use it as one of our main measures:
From the post…
By doubling the R-Squared value, we were able to achieve results 79 times better.
The following chart shows the difference between the 4.x and 5.0 models. By using Deep Neural Networks, we are making a large jump in precision.
- This reply was modified 4 weeks, 1 day ago by Justin.
Yes, the last two ETH trades were great snipes.
In terms of the stop losses, those levels are set by machine learning. It’s really challenging to find the right level–you hate to get bumped out on a minor dip only then to miss out on an upswing. But you also need to control the risk exposure. So we let the ML handle the best levels.
What I can say is as the predictions (Neural Net step) get better, the optimizer step gets more comfortable with tightening the stop losses. Especially the trailing stop.
We expect really good things to come out of the new BTC Neural Net updates. Based on those updates, it did pull in the stop losses. But it has yet to prove itself in the wild.
Assuming all goes to plan, ETH and BCH will take advantage of those new volatility vectors and will likely see tighter stops as a result.Community Advocate
Yes, we have been monitoring BCH closely as it was providing that extreme reading for a fairly long time. While it will go through additional training, readings like that can happen simply because a unique set of conditions happened to match some prior pattern.
Even though the reading was higher than reality, it still resulted in a great trade that closed out at a very good time.
Both ETH and BCH closed out trades for nice gains.Community AdvocateApril 16, 2020 at 1:19 pm in reply to: Upcoming Enhancements: Portfolio, Stop Losses, and More #8601
Great, thanks guys. And #2 really just means better trade performance.Community Advocate
Hi @juliandwillett, yes–the stop loss logic is being redone now. The complexity that’s causing these issues is that it is considering the prediction. That’s why there is a delay.
However, to better avoid these drops, we need it to work faster and act more like a traditional stop loss. This goes for the trailing loss too. In our testing.
So that part is being rewritten and we should have that out ideally this week.Community Advocate
Hello, yes–the stop performance needs to be improved on our end. We’re actively working this.
In some prior cases, it wouldn’t have mattered. The markets moved just too quickly and blew through even standard stop losses. But this BCH one could have performed better. We’ve completed the RCA and are working on it. Generally our stop losses do what they’re supposed to, but they can be improved. Hope to have an update next week.
Regarding setting your own stop–no, that is totally fine and Auto Trade will not interfere. If your exchange stop is hit, BCH will sell. Later, you’ll get an Auto Trade order to sell, but it won’t have any BCH to sell, so nothing happens.
There’s no conflict if you set your own stops.Community Advocate
Hi @Cirke, thanks for the question! Yes, that is a common ask.
In general, we want to do fewer, better models. Effectively managing machine learning takes highly-skilled resources. It’s not something that can be truly done in a fully automated manner, at least yet. This means we need data science, statistical modeling, and software engineering skills fully invested in making a few great things rather than many good things.
With that said, we have recently rolled out major changes. If those settle well and become fairly straight-forward to maintain, we will certainly entertain expanding the crypto offered.Community AdvocateApril 6, 2020 at 3:54 pm in reply to: Dual Momentum, Machine Learning, and Market Cap Index Bots #8278
@Banner, thanks so much for doing this tracking. It’s hugely valuable and drives much-needed transparency into this space.
You started at a very challenging time–but it’s good to see performance in both bull and bear market (not to mention sideways).
We can also take learnings from these other systems and feed them back into our data set. We do some of our own internal monitoring and benchmarking of similar solutions for this exact purpose.
It was a tough drop right when we implemented a whole bunch of changes. But we now have some great trades open and will await to see how they close out.Community AdvocateApril 6, 2020 at 3:49 pm in reply to: Change Announcements: Litecoin, Conservative Strategy, and More #8277