3. Hardware and software realization My automated trading system operates from home. I have a good full-time job, so I do not see my trade screen during the day, all monitors are tuned off. Strong discipline includes my intervention only if the system did any absolutely wrong action or computer lost the market connection. The computer has nothing special: one Pentium 4 CPU and 512 MB of RAM, one good monitor, a sound (the system can talk as the reward, when it did something superior), direct cable internet connection having about 5 megabits per second, all others routers for home network are connected to the second Ethernet PCI card. As I said, the system includes the custom designed data processing and trading platform interacting through API with IBâs stand along Trader Workstation 4.0. It cost me 0.00 $/month to collect the big market data files from around the world. May be this data are not super quality, but if I will trade something from the collected data set, I will adjust another system using the same type of data, which I will have during real operations. I can agree with $4.80 for 1 ES roundtrip, because even in critical situation the system has the stop order execution time or picking up bid (ask) offer << 1 second. Other components are Windows XP Professional with automatic update, Norton AntiVirus, MS AntiSpyware. I like my simple hardware and software configuration, for not going into millisecond range, it is more than enough. I could invest 5 times more money into hardware or software, but it will not change the core system characteristics. One point I think has to be modified only, system needs the second independence internet connection. 6 trading days were lost for 9 months due to the temporally short internet data interruptions. ---------------------------------------
vlad, have you thought about co-locating your machine, so you'd have a fibre optic connection to the internet backbone. Connection problems would then be almost eliminated. You could then log in from anywhere and will have no problems. Runningbear
Yes, I have thought about co-locating the reliable machine close to the internet backbone, but hadnât enough time to try it. I would like to test it in the next year, when I will be ready to launch automated trading system fighting for money on diverse market segments simultaneously. Runningbear, could you recommend good place or could you share any experience with such kind of system exploitation? Thank you. -----------------------------------
Vlad, I haven't gone to co-location as yet, so I cannot recommend anyone. Maybe down the track I might. It makes a lot of sense if you're tradig decent size. Runningbear
Man, I am moving forward through this thread according the plan shown in the first post. Sorry for the delay, I do not have much free time to prepare my data for representation. Roughly speaking, I have only 20% in real trades relative to back testing results. All details will follow. -------------------------------------
4a. Backtesting system results Figure 2, attached below, illustrates the results of system backtesting before the automated system trading has been began. This system, STM 18, composed of 4 different trading systems having approximately the same results for 301 analyzed days. These selected 4 systems do not interfere with each other. Each system has max dropdown < 10 ES pts. Optimization algorithm worked out parametric set independently for each system. Optimization criteria were not simple maximum of the net P/L value. Additional requirements were predetermined such as a smoothness of the partial derivatives. -----------------------------------
hmm. i try to summarise (and round). hit ratio 71% profit factor 3.1 trades 279 systems 4 test sounds good to me. but if you just make 20% of that in real time you might ask yourself if it is "true". i would try adding other markets, the more uncorrelated the better. i would check for a longer time period if the system correlates with any feature of the emini. natural guess is that vola shrink has reached a level your system(s) can hardly (=20% of backtest) swallow. though your test already happened in relatively low vola environment. i would recommend that you take daily data of the VIX, run a corr against your equity curve, or take a slow EMA on the VIX and try it out. category I error would be to trade a system that is untrue, category II error would mean you quit though it is fine. analysing it against the market could help distinguish between the two. i dare some advice (really no offense intended). you already showed that you are able to run a reasonable set up in terms of IT, hardware and so forth. now (if i were in your shows) i would turn the subject to the main issue: bringing up your 20% of real in relation to test. some guys better than myself might be able to more profoundly judge whether your result is due to overfit oder market change or whatever else ... peace
I'm watching this with great interest, vlad. I have been thinking about building/deploying an automated system as well. I am in the infancy stage, having backtested a couple of systems and am cutting my teeth on the IB API. My main concern at this juncture is... RISK ABATEMENT and I don't mean risk of running a crappy system. I write code professionally, and know all too well that the code I write is never perfect, bugfree, flawless. The idea of letting a program manage an account with large sums of money while I work my dayjob, unable to monitor or intervene, scares the daylights out of me. I think a very carefully thought out plan for risk control and fault tolerance is a huge part of the work in a set and forget system that can operate while one works his/her day job. Have you thought much about this?