Trades: 30 P/L: -$8.18 Pips: -81 Account: $510.71 Since Starting: Trades: 121 Pips: +107 Today was interesting on several fronts. First I had a trade lose $14. My stop loss prevents losing more than $10. For some reason, my stop did not work. So I am looking into the glitch to determine how to prevent in the future. Second issue. I set up a second account and put it running this afternoon. When I came back to check, I realized two orders were canceled due to some TIF error. After further inspection, I realized the new account is not accepting my short orders for some reason. Through trial and error determining the cause of the trades not going through, I lost a few pennies putting my total account value at: $1,010.05 (account #1 + account #2) For tonight shorting on account #2 is off and the strategies are long only. I am looking forward to the upcoming week to see how the dual strategy method works out.
<b>Today's Results:</b> P/L: $2.33 Pips: +23 Account: 1012.38 <b>Weekly Results:</b> Week 1 results: +164 pips Account value: $516.40 Week 2 results: -40 pips Account value: $1012.38
Hello frostengine, I've started writing a trading system myself, though I am good with C++ but this is the first time I am writing trading software hence I'd like to know if I've made the right architectural decisions. My main source of information are books by Richard Stevens on programming in Unix environment. I am using C++ on Ubuntu. Could you recommend me a book, papers, or point me to some source of information on writing such system using C++, preferably on Linux/UNIX platform. I've looked so far at source code of QuickFix and Marketcetera, both seems bloated. Marketcetera is also written in Java but I am looking for C++ code. Briefly my system consists of two executables: system1 is a multitreaded app that: thread1: monitors socket to download ticks in real time. thread2: processes ticks to generate buy/sell signals, places new_order_message to FIFO (named pipe) to be read by system2, saves orders in order_book. thread3: monitors FIFO for order confirmation messages from system2. system2 is pretty much a FIX engine that is a single threaded app that: monitors socket for a new FIX messages from a broker, if order confirmation arrived place it into FIFO to be read by system1. monitors FIFO for new orders from system1, if a new_order_message arrived convert it into FIX format and send to broker. I am also using MB Trading but I place orders over their FIX server. Thank you.
The biggest hurdle I am currently facing is trying to determine how "robust" a strategy needs to be. For instance, if I train a strategy on the EUR/USD. Does it need to also at least be profitable on GBP/USD,NZD/USD,and AUD/USD? Or am I filtering out the more "optimal" strategies by imposing this restriction. Anyone have any experiences in this area?
Based on my experience, strategies that only work on certain markets and not others trend to fall apart over time. You should keep a watchful eye on those types of strategies...in fact, you should expect them to blow up....i'll go even further that say that, in fact, you should expect EVERY strategy to blowup. So how do you trade those types of strategies? If the strategy continues to make money under simulated conditions for a few weeks or months (your choice), then jump in with very small position sizes with real money....Once you get into the profit with the "houses" money, stop trading the strategy if a sizable drawdown hits before you've made enough money and then either start the process over under simulated conditions or move on to a new strategy. If you are able to generate significant profits, then increase your position size using the house money and not your money. This is a low-risk way of trading any strategy without losing much money. Also, perform some statistical analyses on the average drawdowns. A 3 Standard Deviation drawdown is a signal to stand-aside and reassess.
The system went live with 8 total strategies last night, 4 strategies trading on a 10 min chart, and 4 strategies trading on a 4 range chart, all for the following pairs. EUR/USD GBP/USD NZD/USD AUD/USD Each pair has a strategy trained "roughly" for the pair. However, due to not having enough time to train properly, the strategies are FAR from ideal... But good enough for now. I also discovered last night a simple filter I could apply at the time of training that increase profitability of the trained strategy by 50%. I am now in the process of training new strategies including this filter.
It looks like a very simple volatility filter increases average trade by another 25% on the 4 range strategy. I will need to retrain all of the 4 range strategies to utilize these two new filters. It should result in drastically higher per trade expectancy once in place.
Frostengine, Great thread. Hey do you think you could tell us a bit more about how you designed your pattern matching software and how you integrated it with ninjatrader? How long it took? Any problems you encountered? Without giving away specifics of course. Do you use your custom program to search for patterns in real time then execute through ninja or did you back test to find viable patterns with your program then hard code your strategy directly in ninja? Also, purely from an executional standpoint, can you tell us how well ninjatrader performs. I have read reports y some auto traders that when running for long periods, ninja can lose sync will what it thinks its positions are and what they really are. Can you let us know if you're experienced anything like this. thanks, Runningbear