Poll: Automated or Discretionary, which has been more profitable for you?

ok so we're talking more an aptitude for CS in the realm of machine learning rather than pure maths. Probably more engineering logic than the actual algorithms themselves. Especially at high speeds.

I just don't like it when people try to compare market modeling to solving poincare conjecture or something like that.

 

Apparently, you do not know anything about computer. I have a graduate degree in electrical engineer and engineer in computer science MIT

 

right. phd from MIT is trolling the ET boards.

im working on my undergrad, but if im on elittrader with a doctorate in the future, then remind me to kill myself.

 

I forgot to add something as I just got done looking at one of my automated trading systems. This system produces some dam good trades. But I cannot trade it 60% of the time. It is a volatility based system. The problem with this system is if volatility gets too high it produces long strings of small consecutive losses as high a 7 or 8 straight; which I have never been able to stomach. So the system sits idle most of the time because I won’t risk having to stomach a 8 straight (small) loss draw down.

This is the other intangible in building an automated system. Traders all make the assumption that if your can program it to make profits you can trade it. Sorry that is just not the case. Automated systems can produce trades that just don’t fit a trader’s goals and objectives.

For example I have another system that produces a high winning percentage (from 55% to 63%) but a low W/L around 1.40. The system works but has the problem of the equity curve going flat for 5 or 15 trades before resuming the climb. This does not fit with my goals and objectives.

The point is whether you trade automated or discretionary a big part of the process is finding an edge that you feel comfortable with (that you can sit through the systems quirks).

 

you are just pretending to be a computer genius. It is all written from your elite history.

 

I was a DBA in a steel mill where we converted the chemical formulas and mechanical processes into a Oracle Rule based system. They used Smalltalk on PC’s hooked to our MODCOMPs to plug in actual steel making results and modify our rule base. This is not as hard as people think it is. It is just time consuming to build a working system.

Anyone who trades profitably is already using a simple form of AI. You are the expert system with rules to handle any trading situation. A discretionary set up is a manual AI rule. All you have to do is start a journal and you can build a base that tells you what you can and cannot do with your object (trade) that worked or did not.

There is always going to be room to squeeze in a make a profit no matter what they throw at us. We can adapt far quicker than a room full of programmers.

 

Interesting replies so far, very good.

As far as automation, I agree with lindq. You don't have to be a complete rocket scientist to have an automated system using colocation , hft, and millions in hardware. You know there are "edges" for the little guy to exploit too. I can get in and out of a trade in a flash. A big hedge fund can't without adverse effects.


I think that the best edges are not programmable. Thoughts?

 

I have no idea what they're doing at renaissance, but I know for a fact that there are plenty of trading opportunities with good return which are algorithmic/automatable in nature which are not HFT or difficult to program. Really that stuff only comes into play when you're trying to do automated market making and REALLY only when you have an obligation to be on NBBO some percentage of the time.

 

It's not a matter of language - C is good enough for nearly anything. Your OS and library facilities will be more of an issue.

The real difficulty is understanding that style of programming where the are a large number of things going on at once asynchronously, and those things may be in error, and there are odd delays (in this case due to network latency). Despite being done on big hardware, it's more like embedded or OS programming than application programming. Not hard for someone with the right background, but not trivial.

My point is simply that unless you just have to do automated market making, there are better opportunities elsewhere that don't require that level of sophistication.