systematic test

[metals_trader]

With all the talk of algorithms and systematic trading being soon to take over i have decided to fully test just how different the results of a systematic trading program is compared to execution from myself.

I have a friend who has created an algorithm to look for the exact same characteristics and pre defined entry and exits that i have deemed that i look for when trading myself, i shall be trading against this system and shall be seeing how the results differ compared to my own execution.

the trading will be done on dma in real cash no sims and the market traded for now will just be copper on both comex and lme.

 

[metals_trader]

Presumably trading in size manageable enough to make sure slippage should, in theory be negligible?

exactly, right now it is just with a 1 lot the basic clip size and eventually it will be scaled up to test how it reacts at different sizes.

 

[NetTecture]

Well, problem is - make sure to include some slippage. First, you may not get optimal fills, second, rather check how good you possibly can scale up to start with Better be conservative in testing than to loose money.

 

[Adamus]

There's been talk about systematic trading taking over from humans for years. If it was going to happen, it would have happened by now. To survive, a computer would need to be able to recognise when it's algorithm had died and needed replacing, i.e. adapt itself.

 

[NetTecture]

That is interesting enough not so hard.

After all, a real trader has to do the same.

Just start tracking trades. If the execution derives too much from acceptable parameters, stop trading and call admin

Real traders can not do otherwise anyway - remember, one has to trust ones system, and follow it. Only when trade histories get whacky can one start thinking whether the system has died or needs fundamental adjustment.

That said, most environments used for automated trading seem to ignore that - there is no link between historical & backtesting and real testing (i.e. comparing the last 50 trades, for example, with the 50 trades before etc. to grab whether one runs out of sync). One is expected to run backtests, then run the system, and hope for the best, but there is no check done by the framework on the validity of the backtest or "historical forward results" (i.e. results the system did collect in the past compared to what it collects now). Call that an oversight by framework people if you want

 

[metals_trader]

There's been talk about systematic trading taking over from humans for years. If it was going to happen, it would have happened by now. To survive, a computer would need to be able to recognise when it's algorithm had died and needed replacing, i.e. adapt itself.

if you look at the volume now of trades done by algoirthms on exchanges such as liffe and eurex compared to self executed by humans you will see the vast majority is now algorithmic.
You have to remember although these algorithms are computer operated they still have researchers, developers and programmers constantly changing the specifications and perameters.

 

[MrGecko]

I think I was going to mention a similar thing - even though there is an algorithm working the order, someone, somewhere, had to make the buy/sell decision. The Algorithms are just there to reduce transaction costs (in terms of moving the market against you).

Take Iceberg orders for example. Dead simple, run by a computer - but you need a trader at the other end to make the decisions. In the equities markets, these things (algorithmic execution) are very popular I believe, with the Banks / Brokers all trying to create new algo's to sell on to their clients (at a higher commision, obviously, but under the guise that it's better to use an algo than just smash the market for 100,000 shares). It is a better use of time to get the comp to accumulate the position at, say, better than VWAP, than have the trader / broker staring at the L2 for a few hours like their playing on the xBox - let the computer (and the computer guys) get you your order filled as best they can, and move on to generate more trade ideas / call more clients / whatever.

Of the trade decisions that are actually generated solely by a computer, I would have tought alot of it is stat. arb or something similar - i.e. very high frequancy but small clip size. I would think that not alot, if any, uses MACD.

Given that your strategy has been automated, would you say it is a mechanical, indcator based strategy? Or are you looking more on the stat arb (e.g. pairs trading metals) side of things?

 

[metals_trader]

I think I was going to mention a similar thing - even though there is an algorithm working the order, someone, somewhere, had to make the buy/sell decision. The Algorithms are just there to reduce transaction costs (in terms of moving the market against you).

Take Iceberg orders for example. Dead simple, run by a computer - but you need a trader at the other end to make the decisions. In the equities markets, these things (algorithmic execution) are very popular I believe, with the Banks / Brokers all trying to create new algo's to sell on to their clients (at a higher commision, obviously, but under the guise that it's better to use an algo than just smash the market for 100,000 shares). It is a better use of time to get the comp to accumulate the position at, say, better than VWAP, than have the trader / broker staring at the L2 for a few hours like their playing on the xBox - let the computer (and the computer guys) get you your order filled as best they can, and move on to generate more trade ideas / call more clients / whatever.

Of the trade decisions that are actually generated solely by a computer, I would have tought alot of it is stat. arb or something similar - i.e. very high frequancy but small clip size. I would think that not alot, if any, uses MACD.

Given that your strategy has been automated, would you say it is a mechanical, indcator based strategy? Or are you looking more on the stat arb (e.g. pairs trading metals) side of things?

Right now it is going to be more a indcator based strategy that i will be following basically pre set parameters that i myself trade on and the algorithm will be mimicking as if having a double of myself, eventually it will involve more complex ranging from arbitrage against cross exchange traded products to against other commodities, but to start with i want to test it at a basic level

 

[metals_trader]

FYI by no means all of the high frequency stuff is what you guys would call technically driven at all. A lot of it is based on data mining and pattern recognition (which, if we're really being pedantic here, could be something that someone could pick out of a chart somewhere / somehow (especially if it had volume). But what I'm trying to say is that it isn't all just automation on a grand scale of something the average high end retail trader might look at from home.

I totally agree but then you have just pure mathematical formula approaches which takes no decision or market opinions, infact several hedge funds who operate in the systematic approach will not be able to tell you views on the markets as they have none and just purely follow mathermatical outcomes.

The main important thing here is i am not trying to point out how or what an algorithm, systematic trading, mechanical trading or whatever you want to call it trades, merely just how the algorithm i will be using is going to trade on

 

[metals_trader]

Sorry - fair point. Apologies for kinda hijacking the thread. Not my intention. Just got a bit carried away.

GJ

No problem, im always happy for people to give their thoughts on it. Its one of those things like anything in trading where its an endless conversation as there are so many different variables.