Do Automated Trading Systems Work?

newbunch · Dec 2, 2005

Quote from stephencrowley:

Sometimes, I'm not even sure what I do. My system makes very-very short term 'predictions' about direction and volatility and also generates uncertainty estimates along with the predictions. Volatility is quite easy to predict, direction is another thing. I use the uncertainty estimates to take the opposite side on another highly correlated symbol. I have a strict stop-loss policy where I set the maximum loss based on the volatility prediction so I don't get stopped out by random flucuations and my 'winning' side has a trailing stop loss based on volatility as well. The hedging ensures that my maximum loss is massively reduced and the system shuts itself down if it starts performing badly. I can rest easily this way.

That's some crazy stuff. We are exact opposites. You do "very-very short term" and I do very long term (my system was long the S&P from Oct 02 until Jun 05, though in varying amount but not changing much).

Just curious, what software and broker do you use?

rufus_4000 · Dec 2, 2005

Quote from stephencrowley:

Sometimes, I'm not even sure what I do. My system makes very-very short term 'predictions' about direction and volatility and also generates uncertainty estimates along with the predictions. Volatility is quite easy to predict, direction is another thing. I use the uncertainty estimates to take the opposite side on another highly correlated symbol. I have a strict stop-loss policy where I set the maximum loss based on the volatility prediction so I don't get stopped out by random flucuations and my 'winning' side has a trailing stop loss based on volatility as well. The hedging ensures that my maximum loss is massively reduced and the system shuts itself down if it starts performing badly. I can rest easily this way.

Interesting, so it is a strict pairs trading system then. I haven't played around with equities too much. My system mostly trade futures and options in a pseudo market making (I don't put on two-sided markets, I put on one-sided orders in sequence to simulate the behavior), and when I get on a favorable position, I keep it until either the original price got broken, or the statistical property (I guess you can call it a "trend") disappears.

stephencrowley · Dec 2, 2005

Imho, short term stuff is less risky than long term. In any chaotic system (which I think the markets partly are.. ) the near term predictions are always more accurate than long term. Play around with some chaotic functions and try different starting points for some fun examples to see how small initial value changes blow up into very large changes as the time frame increases.

My software is completely custom.. java based. I do all the modelling/theory testing, etc in matlab and then re-write the final product in java so the design is easy to understand and maintain. Matlab code can get very..very ugly.

I have an account with limebrokerage which has an outstanding java api. Very professional guys there. They are an automated/institutional firm but somehow I managed to get an account with them.

Quote from newbunch:

That's some crazy stuff. We are exact opposites. You do "very-very short term" and I do very long term (my system was long the S&P from Oct 02 until Jun 05, though in varying amount but not changing much).

Just curious, what software and broker do you use?

stephencrowley · Dec 2, 2005

Basically.. yeah. pairs/market making strategy.. same way you described.. continually taking different sides of the market all day.. the average profit is very small so it has to make a lot of trades, it scales in and out as well.

Quote from rufus_4000:

Interesting, so it is a strict pairs trading system then. I haven't played around with equities too much. My system mostly trade futures and options in a pseudo market making (I don't put on two-sided markets, I put on one-sided orders in sequence to simulate the behavior), and when I get on a favorable position, I keep it until either the original price got broken, or the statistical property (I guess you can call it a "trend") disappears.

newbunch · Dec 2, 2005

Quote from stephencrowley:

Imho, short term stuff is less risky than long term.

I think the risks are different. The risk of long-term is consecutive large losses. The risk of short-term is that the system stops working.

My long-term system is backtested to 1952 and I've tested it going forward many many times. The risk of it stopping to work is minimal. Only a fundamental change in the world (ie. closing the market based system) would ruin my system.

But I am well aware of the risk of large losses. You have to have very strict stops (though those don't work if the market gaps) and/or limit exposures.

rufus_4000 · Dec 2, 2005

Quote from stephencrowley:

Basically.. yeah. pairs/market making strategy.. same way you described.. continually taking different sides of the market all day.. the average profit is very small so it has to make a lot of trades, it scales in and out as well.

Similar systems then, you trade with a different symbol opposite direction, I take opposite side of the market. Heh.

mats · Dec 2, 2005

Rufus
you are absolutely right about arb/hedge/spreading etc models. These models have good results because of market inefficiencies. But I have my doubt about the other group; technical analysis models.

LoosenUp · Dec 3, 2005

Quote from rufus_4000:

Interesting, so it is a strict pairs trading system then. I haven't played around with equities too much. My system mostly trade futures and options in a pseudo market making (I don't put on two-sided markets, I put on one-sided orders in sequence to simulate the behavior), and when I get on a favorable position, I keep it until either the original price got broken, or the statistical property (I guess you can call it a "trend") disappears.

Hi rufus,

Can you explain a bit more what you mean by pseudo market making? Put on one-sided orders in sequence to simulate the behaviour? Do you mean selling on strength and buying on weakness (hence, one-sided) looking for some sort of mean reversion?

And for those of you doing so many trades per day, how much are you paying in commission per contract ( all inclusive ) of say, Emini S&P futures or options?

Thanks

rufus_4000 · Dec 3, 2005

Quote from LoosenUp:

Hi rufus,

Can you explain a bit more what you mean by pseudo market making? Put on one-sided orders in sequence to simulate the behaviour? Do you mean selling on strength and buying on weakness (hence, one-sided) looking for some sort of mean reversion?

And for those of you doing so many trades per day, how much are you paying in commission per contract ( all inclusive ) of say, Emini S&P futures or options?

Thanks

It just means that I have two working orders at the same time, that's all. Of course the trick is not just having two orders at the same time, it is how the two working orders in relation to each other, and in what sequence and price they are put on, which is the bulk of the tuning. In a way, the system is looking for a slight mean reversion, which is why I also compute the theoretical values. The system is a little more complex than a standard autospreader which just peppers the book with orders.

In terms of commissions, I am a professional (exchange member), so the concept of an all-in fee is foreign to me, I pay a commission (which is far lower than retail rates, less than .20) plus the exchange and regulatory fees.

momo trader99 · Dec 3, 2005

they do work.i built from scratch a front and back end and wrote directly to an api.i have spent well over 500k developing what i have.have tried 5 differant strategies and so far 2 have worked great.keep the faith