Discussion about Shaigan_'s bot

Entelechy

Stadivm Owner
Staff member
Joined
Jul 8, 2018
Messages
130
Reaction score
14
Location
Stadivm.com
** This is a continuation of a discussion started on the 4th page of this (https://stadivm.com/threads/welcome-to-any-new-crypto-algo-traders.178/page-4) thread.

Sure thing @shaigan_, I've built a few trading bots myself (non-ml/dl-based though) and it threw me a bit when you said the indicators are obfuscated until I realised that you might be using a neural net or more likely a combination of typical ml functions which of course would be more like a black box than with standard MACD, RSI etc.

Do you mind me asking roughly how long production took to build this bot?
 

shaigan_

Freshman
Joined
May 1, 2019
Messages
12
Reaction score
1
Location
Paris
Yep, you guessed right: I am using a pure ML black box approach, in order not to stay stucked with the classic technical analysis indicators that everyone is already using.

I took me about two months full time to write the current version of my bot from scratch. But as I told you I had already made a lot of attempts before so I had a very clear idea of where I was going when I started over again, and I already knew how to avoid a lot of mistakes. Actually the most time-consuming task, as usual with machine learning, has been data exploration, model tuning, feature selection etc. With existing python libraries and APIs the purely algorithmic trading part (I mean data processing, strategy backtesting, order execution etc.) was really fast (like just a few days).
I also spent a bit of time understanding margin trading because it was my first time using it and I didn't want to make costly mistakes which happen so fast when trading on margin.
 

Entelechy

Stadivm Owner
Staff member
Joined
Jul 8, 2018
Messages
130
Reaction score
14
Location
Stadivm.com
Sure I saw that you were using leveraging in the reports of the trades completed, definitely a good way to go if you're consistently getting positive trades.

How long have you been doing ML for? Academic learning or self-teaching with Coursera solely? I think ML crypto algo trading presents a lot more opportunities overall compared to typical strategies, just based on the barrier to entry for anyone who won't take the time to learn ML. There's a higher level of complexity (the math, the concepts) compared to anything in traditional trading.. being able to program + understanding trading well + ML knowledge is a strong combination.
 

shaigan_

Freshman
Joined
May 1, 2019
Messages
12
Reaction score
1
Location
Paris
Yeah I agree with you about leverage. When my algorithm first began to show good returns in backtesting, I spent a lot of time wondering how I could most benefit from it while keeping a good risk/reward ratio and keeping operations simple: I thought about just putting more money into it, selling signals or investing other people's money and taking a commission, but in the end leverage was the clear winner (constant risk/reward ratio, same opertional complexity, it just basically scales my ROI by a constant factor).

I've been doing ML since 2017 only, but I had a solid background in computer science and applied mathematics so I learned really fast. At the time I was finishing my MBA and I just thought ML was something I had to master if I wanted to be a real 21st century engineer. So yeah I did it while finishing my MBA, learning mainly on Coursera and with Youtube tutorials.

I agree with you about the barrier to entry. And personnaly I think that even a lot of people who are good at ML are really bad at implementing trading strategies. That's kind of what we're seeing with the people from this community: ex-traders with little coding experience on one side, geeks with little understanding of the markets on the other side. One really has to master both sides of this complex problem to be succesful, and that's what I've been trying to do so far.

I am quite shocked at the state of algotrading people are mostly talking about on crypto trading forums. From an AI/ML perspective it looks like medieval times. I don't understand why people keep using hand-written trading signals while even simple ML algorithms can craft much more relevant ones.
 

Entelechy

Stadivm Owner
Staff member
Joined
Jul 8, 2018
Messages
130
Reaction score
14
Location
Stadivm.com
Leverage is a good option compared to the effort of building, distributing and marketing signals for sure.

The combination of trading knowledge and programming skills opens a range of options for generating revenue within the crypto space. The beauty of ML/DL as you would know well is having the model optimise for you instead of manually testing over and over again. I've spoken with quite a few traders that use ML and they're seeing solid returns, and more importantly, very reliable and robust strategies.

Do you have any suggestions for things you'd like to see on Stadivm, or tools that would help you to network or generate more income? I know a lot of the newer guys that signed up are really interested in learning from others who have experience, but for guys with experience my goal is to develop ways to provide value to them on Stadivm as well.
 

shaigan_

Freshman
Joined
May 1, 2019
Messages
12
Reaction score
1
Location
Paris
I'm not here to make more profit. What I'd really like is to share experience with people following similar approaches as me in order to improve my bot and help them improve theirs. I'm also open to helping less advanced people but obviously I wouldn't share my full code, or explain exactly how it works. I can't just give away the result of months of hard personal work and I don't think that would be the best way to help beginners either. It's hard to organize sharing of information and experience when one has to maintain a certain level of information asymetry in order to keep an advantage in the market. I love open source but it is quite contradictory with trading.
 

Entelechy

Stadivm Owner
Staff member
Joined
Jul 8, 2018
Messages
130
Reaction score
14
Location
Stadivm.com
Yeah I mean this is the paradox I've thought about and had some other mention to me, is that by it's nature algo trading is typically not something where people can share the exact details of strategies they're using. But my reply to that has been that there's more ways to help people and work communally than just telling people how you are doing it or giving them your code.

There's plenty of guys who have solid experience with manual trading but new coded a line of Python, and providing advice on which libraries are good to use to build bots and methods of building backtesting engines are very helpful to new guys - I know I would've loved to have had a group to bounce questions and ideas of when I was teaching myself.

I've got some ideas for where to take Stadivm in order to provide as much value as possible - the real thing I'd like to do is start building a bit more cohesion here. It's easy to have people come and join in with a conversation, but hard to have people coming back week after week, and I think it just needs to get to a tipping point of enough people coming here daily for the glue to form.

Once there's a more regular community here I think we can really do a lot in the way of sharing knowledge surrounding crypto algo trading + other areas
 

shaigan_

Freshman
Joined
May 1, 2019
Messages
12
Reaction score
1
Location
Paris
Yep there's so much to be shared, like good data sources etc. I hope more people join the boat ! I'll keep checking the feed and see how I can contribute.
 

Entelechy

Stadivm Owner
Staff member
Joined
Jul 8, 2018
Messages
130
Reaction score
14
Location
Stadivm.com
Sure thing, good talking with you and good to see your algo is working well! Drop me a line any time
 
Reactions: shaigan_

daybyter

Freshman
Joined
Mar 15, 2019
Messages
17
Reaction score
2
Location
Germany
Hi!

Did you try your bot with forex markets? That's a very different situation. A lot more competitive, but also a lot more fun. At least to me...
 

shaigan_

Freshman
Joined
May 1, 2019
Messages
12
Reaction score
1
Location
Paris
Well I’m doing it for the money first, not for fun, but it’s a nice plus. I’m still optimising my strategy on crypto right now, I don’t see why I should bother with less predictable and less volatile markets for now. I may consider it for liquidity reasons in the future.
 

daybyter

Freshman
Joined
Mar 15, 2019
Messages
17
Reaction score
2
Location
Germany
In a way you are right. But then I'm so fed up with those immature crypto exchanges. Most of the time, you just code around API shortcomings or bugs, instead of focussing on your strategy.
 

Entelechy

Stadivm Owner
Staff member
Joined
Jul 8, 2018
Messages
130
Reaction score
14
Location
Stadivm.com
In a way you are right. But then I'm so fed up with those immature crypto exchanges. Most of the time, you just code around API shortcomings or bugs, instead of focussing on your strategy.
Sounds like you've programmed algos/bots before? Are you trading within crypto right now or forex as you mentioned?
 

daybyter

Freshman
Joined
Mar 15, 2019
Messages
17
Reaction score
2
Location
Germany
Yeah, I coded crypto bots since 2012 and forex bots for the last few years. Returned to crypto for the last few months and still struggle with crappy APIs... :-(
 
Reactions: Entelechy

Entelechy

Stadivm Owner
Staff member
Joined
Jul 8, 2018
Messages
130
Reaction score
14
Location
Stadivm.com
Nice. Which ones do you think are crappy? I've worked with most of the major exchange APIs and depending on what you're doing with them a lot are good to work with and a few are definitely a pain in the ass
 

daybyter

Freshman
Joined
Mar 15, 2019
Messages
17
Reaction score
2
Location
Germany
Have you ever worked with a professional exchange like LMax, or so? Even with a crappy protocol like FIX, thinks are soooo much better. You get up to 10000 prices per second and no request limits and such nonsense. Spend some more money and you can get ITCH or SBE protocol with a lot more prices. Very easy and quick to parse.
 

Entelechy

Stadivm Owner
Staff member
Joined
Jul 8, 2018
Messages
130
Reaction score
14
Location
Stadivm.com
I haven't seen LMax before, but the exchange we're working with currently is trading $100 million+ 24 hour volume right now, so pretty big. I've heard of FIX and SBE but never worked with them, when you say that ITCH and SBE cost more, why is that?
 

daybyter

Freshman
Joined
Mar 15, 2019
Messages
17
Reaction score
2
Location
Germany
Because you have lower latency as your advantage, and if you want an advantage, it costs money.

Well, you can say, that you create more traffic, but the real reason is, that your chances for profit are higher, so people are willing to spend more money.
 

Entelechy

Stadivm Owner
Staff member
Joined
Jul 8, 2018
Messages
130
Reaction score
14
Location
Stadivm.com
Do you purchase the different protocols and ITCH + SBE are more expensive, or is the a different way that they are more expensive? Are they open-source?