the other idea, of multiple systems on one market, is mathematically sound, but what do you do in practice, when one system calls for a long and the other calls for a short?
Although these systems were designed around observations I made while trading NQ, the methods might work on stocks. These systems require intraday level I trades and quotes to generate the indicators. I only have clean data for ES and NQ. Iâve tried esignal, Realtick and qcharts for end-of-day stock and Emini data but the data quality is not sufficient (low time resolution, missing ticks, out of order trades), and this impacts performance of my indicators. Iâm not happy that my indicators are sensitive to data quality. If I could make them immune Iâd do it. Currently Iâm using real time logged IB data, which isnât a final solution, but is clean enough after being processed properly. If I can obtain access to clean historical multi-market Level I tick data, Iâll test and optimize these systems for every market, and bring in inter-market indicators. Iâve found success with NQ <-> ES correlation based indicators, and obviously want to explore these further.
I use trained and validated networks (neural nets, constrained least squares, etc.) to determine which subsystem signals to trade. It's a vast improvement over analyzing recent performance with fixed rules. Every system has a character (length of winning/losing streaks, trending nature, counter-trending nature), and a network can learn this character, like a discretionary trader would. Networks seem much better at forecasting semi-profitable system performance than trying to forecast market prices directly.
Cool! Maybe you could tell us about some "methods used to choose systems". Sounds very very interesting.
Yeah. I think good systems need to be âlateralâ in design, as Perry Kaufman writes, having the ability to capture a range of market action, and thus the ability to generalize and adapt. This is in contrast to âverticalâ systems that capture fewer types market action, are less likely to adapt, and have extra analysis, or excessive risk management to squeeze every bit of profit from a narrow range of patterns. Of course every system is a combination of both. I just try to make mine as lateral as possible, and like with an engine, make sure all cylinders are running.
As much as I would love to discuss this stuff in an academic, collaborative way, I donât want to reveal too many details. Iâve spent months developing these systems and methods, and want to insure I can profit from them, and not have others developing more powerful systems, combining their ideas and mine, and making great profits with nothing returned to me, and possibly moving the market against me. I donât mean to sound selfish, but itâs often necessary for system traders to protect their ideas. I donât mean to imply you arenât trustworthy. We can still discuss general concepts.
Some very nice ideas I must say I know not much about. But what you described above is called a bad system. Also, by "system" I see we might be talking about very different animals. The "system" I am talking about cannot adapt to jack shit. It is more like an ambush, it sits around and waits for a certain setup, using very simple rules. After entry, if the price action is not immediately favorable, fitting the "winners profile", it exits. Otherwise it makes money. The only drawback is that it is mostly out of the market. Some might say that's not a drawback.
Not asking for your trading method. Just wanted to learn about mathematical ways to choose which system to trade. p.s. Unless you have a few billion you can't move the market.
1st system: Low Sharpe... wouldn't trade it. These kinds of system is more vulnerable to the changing markets that there are higher risk of the system deteriorating (I've statistically tested it so I'm talking from experience). 2nd and 3rd system: More preferable. The frequency is low but I'll trade those. Using discretion? No... what's the point of having a system. I'd rather have a separate account, one of system and one of discretion. Unless you lack in discipline that you need secondary source of rules to tell you what not to do. Do them all. Do them all and put it in other markets. Finally, the data is too short to analyze that the system is good.
Sure, no problem. I have a colleague with access to about $200M to trade futures trading systems like this, provided I improve the return/risk characteristics and establish a longer track record (simulated and traded). I believe these systems have a liquidity advantage as they can predict and trade fairly long term, large price moves. Iâm not sure how much capital you'd need to trade to affect these systems. It depends on the situation, and how aggressively one trades the signals. Also the buy/sell pressure might mess with my indicators even if price doesn't move too much. I donât know the answer, and it may not matter right now.