How Do You Build Systems Using Python?
- Thread starter tommo
- Start date
I wouldn't trust an emotional human (me) to control risk. I'd rather trust a system that remains objective in every situation. Market conditions need to be built into it.
By the look of things, that's the trend for successful trading nowadays. Opinions are cheap but the fact is, non-discretionary trading systems have been proven to work well.
https://www.rencap.com/
... and obviously I could post names of another hundred or so hedge funds managing more than $1 billion on a full systematic basis.
GAT
I've never seen any evidence of human decision trading working, even the big players don't use like this, they run systems via rules while a human monitors it
I am still not able to create a long-term profitable objective automated system. I have been trying for many years, and I think market keeps on changing (volatility, etc). I found some profitable systems but they lasted only for a short period of time. It is based on the market behaviour at that particular period. When the period is gone, the system started to lose money.
Any suggestions?
Any suggestions?
The simplest is often to have a database my such as MySQL or MSSQL with OHLC data per timeframe, downloaded for backtesting. If you want realtime it becomes more complex, you have various quant sites that you can code in their infrastructure but you lose flexibility. I have some code for realtime tick FX data which is downloaded and processed from TrueFX, for realtime in other instruments it becomes more of a problem.
You need connectors and you end up with a complete architecture trading platform. At that point you start to need integration experience, there are connectors to desktop platforms but often they are quite expensive, you can also buy feeds but again it's integration. Bringing all the pieces in to a seamless yet stable environment is actually difficult, companies charge a lot of money for that.
You can easily do one, code on backtest manual load data, or code on a platform realtime, the issue is combining the two which is why most go for quant platforms like Quantopian. Don't believe in them personally and have access to R Services which is based on SQL Server with connectors to realtime data, but it's not designed for retail. Speaking of which, R is proven to be the most profitable language when they did trading competitions, you might want to reconsider python.
Of course not, a supercomputer can process 0.04% of the human brain in contextual analysis. At advanced levels you can generate 20%/mth return on a few hours per week, to generate 2%/mth with algos you generally work 80hrs per week as a quant rebuilding them constantly for the contextual changes.
If you catch the wave correctly it will obviously last longer, but with the increasing volatility that is a thing of the past, hence why all the algos are breaking down and causing ripple effects. The most optimal is to algo to 95% and contextually analyze the last 5% yourself on shorter timeframes and just stick with contextual analysis on the higher timeframes suplimented with algos, the timeframe equals the entry window.
Why the hell am I giving this advice, I don't get anything from providing institutional level information, and no-one has a clue what to do with it no matter how much detail you give them. Time to rethink this approach.
Last edited:
Look harder. I myself have never put on a discretionary trade (except currency hedging/converting). While I'm not exactly evidence and you can call me a liar, I know for myself that it works. There are many others, some even on ET like @globalarbtrader, if you're suggesting he's a hoax too then he's one amazingly thought out hoax.
Human beings are superior in coming up with ideas but when it comes to faultless execution, we have no chance.
I am long time programmer/trader and having developed various trading systems (automated) and tested on various currency pairs. I have come to the conclusion that 100% automatic systems are very rare and needs lots of analysis or data or computing power which is possible with large organizations like GS or some big hedge fund with very huge infrastructure. There was a GS trader/developer who stole 500k+ worth of code/trading system from GS and resigned GS and went to Germany and eventually caught. So it is possible to have a system but it takes lots of time/systems/effort. But I think you can add human intervention at execution and take the guesswork from the system (automation for trade decision/logic) and then when the signal comes out of the system, evaluate and you pull the trigger. This is my conclusion as of now. I may be wrong... Goodluck.