- February 4, 2020 at 10:37 am #7453
Hello Crypto-ML team,
I am new to this page and subscribed just a week ago.
I am a little confused about the signals that the AI is giving.
I understand that the price has gone down a little bit but today a “close long” signal was issued with a Bull point of 9.972. So is it a bear or bull signal in aggregation?
Also, I wonder if the halvening in April/May is taken into account. That alone should be a bullish sign.
The algorithm also seemed to have missed the LTC price rise and only recently issued a “open long” signal after the price rose 50% during the last month. Ironically, after the “open long” signal, the price declined.
Until now, I am not convinced that this tool is making better choices than I would.
To me, the signals seem to be a reaction to the price instead of a prediction.
I hope you can prove me wrong.
- This topic was modified 3 months, 3 weeks ago by myoungho.
- February 4, 2020 at 12:02 pm #7455JustinModerator
Hi @myoungho, I’m glad you’ve joined us and thanks for the recommendation!
Your approach to this is right–check out the signals to see how they move and see if they make sense for your style or not.
Starting with your BTC question, that trade was opened at $8,547 and just now closed out for a 7.75% gain. Two “bearish anomalies” were issued back to back, so it’s definitely seeing some pattern that is of concern. We can’t say the exact details behind the closure, but I imagine if we see the market drop below $9K, it’ll drop very quickly.
So I’m good with this capture. If we see price hold here, we could certainly get another trade opening.
Regarding LTC, yes–that was an issue here recently. The model effectively learned to be too conservative and was just choosing to sit out.
That said, the models for ETH, LTC, and BCH were all just upgraded this week and won’t face that issue going forward.
This describes how the new models work: https://crypto-ml.com/bitcoin-trading-with-machine-learning-anomaly-detection/
And the upgrade notice: https://crypto-ml.com/machine-learning-models-upgraded-to-version-4/
Some good items/discussions to catch up on:
- February 5, 2020 at 10:17 am #7479
Thanks for the quick reply. I really appreciate the customer service and openness of this site.
I think I need to keep watching the signals from this machine learning tool.
To me it seems that the halvening is not included. Also the “close long” signal for ETH yesterday seemed weird regarding the general tendency of a bull market. Today, ETH rose 6%.
So again, it looks like this AI is reacting to price changes instead of predicting them.
But I will give it a chance for another month or two.
Is there any comparison between Crypto-ML performance and dollar-cost averaging?
That would be an interesting comparison as dollar-cost averaging seems to be the easiest method without AI.
Once again thanks for your excellent communication.
- February 6, 2020 at 5:39 am #7485martyfParticipant
i definitely support doing your own analysis and observation of the crypto-ml platform, and I admit that at first I was skeptical. However, I’ve been using it for a while and I must say I’m very satisfied.
Granted, it doesn’t get all the moves perfectly, but no technical analysis edge is going to do that. And there are some moves that its like “what the hell bot, thats clearly a top. The stochs turned around. Why didn’t you sell?” What I really like about the crypto-ML AI / bot, (especially the auto trader), is that it does the trading for you. Because consistency is key when it comes to any trading strategy.
Compared to trying to manually do this – I always hem and haw about a particular pattern, signal, etc., even though I know the technical analysis edge I have says a certain thing. This happens on either entry or exit.
So, thats my two cents. But I obviously encourage healthy skepticism. After all, any machine learning model is effectively black magic. Well, not really.
- February 6, 2020 at 6:46 pm #7487JustinModerator
@myoungho, thanks so much for note about the openness. That’s a value I’m really pushing more than anything. Both crypto and trading systems have tough reputations to overcome…I want us to be leaders in changing that perception and we can only do it by being open. I think as we grow, we can become more and more open, even ultimately exposing our systems.
In terms of reacting, that’s actually a keen observation. Let me give this explanation-
Our system ingests a large set of data. In the case of ETH, it then runs that data through 1,617 lines of ML-generated code, each of which represents a scenario it has encountered before. It does this rapidly and continuously.
Its goal is to predict large price changes about 24 hours before they happen.
As it turns out, one of the first things it considers in all of that processing is recent price change. It effectively wants to “ride the wave.” That means it figures the best approach is to hop on to existing movements. It senses a swell and tries to surf it.
This is opposed to trying to find reversal points. Instead of riding the wave, it could trade based off of reversal patterns (in traditional technical analysis, this would be overbought and oversold). For a while, our Litecoin model was doing this. It behaved differently than the others.
It turns out, “reversal trading” is a risky approach in crypto trading. In the 2017 run up, indicators like RSI were screaming “overbought” way early. If you listened, you would have sold and missed out big time. On the way down, they were saying “oversold” but the markets kept crashing. If you tried to buy in then, you’d get burned.
We have a post on this here: https://crypto-ml.com/is-rsi-a-good-technical-indicator-for-bitcoin/
The point is, our models do react. But they *ideally* capture most of the move up and avoid most of the move down. Over time, this appears to be the optimal approach. I got most of the 2017 rise and missed most of the following drop…invaluable.
In terms of comparing to DCA, you could do some quick spreadsheet math, but you have the same problem as HODL. When do you start DCA? When do you stop and take profit? Neither DCA nor HODL have an answer for that.
With trading, on the other hand, you should be able to start and stop at any time. You don’t have to try to time the market. You just start trading when you’re ready and stop when you want to move on to something else. Hopefully, you make good money in between. That doesn’t work with DCA or HODL.
@martyf thanks so much for sharing your perspective and experience. Your “hem and haw” statement concisely summarizes my many years of trading. 🙂
And not quite black magic, but ML is a black box for sure. Not being able to know the exact mechanics creates a layer of separation that you don’t have when it comes to technical or fundamental analysis.
Great discussion guys.
- This reply was modified 3 months, 3 weeks ago by admin.
- February 27, 2020 at 11:15 am #7783
Thank you martyf and Justin for your replies.
After a month of trial, I am still a little skeptical regarding the performance of the Crypto-ML algorithm. In my feeling, it is following the trend and reacted a little too late. Your explanation helped me to confirm my initial impression.
At this point I am not totally convinced of the benefits of ML since most signals were triggered too late. But I am willing to give it a shot and hope that it will perform better over the long run.
While I am still not 100% sure, you might have won another subscriber. I will have to make my decision tomorrow when the free trial time expires.
What I definitely liked was the fast and competent communication.
- You must be logged in to reply to this topic.