Haha, I appreciate the input. If you can believe, the last thing I did before going live was put together a unified logging system rather than just outputting everything to System.out like a dummy. And yeah, I have to work on everything you mentioned except backups. Backups up the wazoo!
my point is that this game is pretty straight(unlike any other business,forget about working for someone else)-for every mistake you will pay right on spot. you are lucky that time,when trade was triggered by unfiltered data,but it wont' happen again just make sure you know that. i might let it fly once,but that was it
undefined, I hope everything is going well with your system. Do you have any updates that you would like to share? I think my strategy is similar in style to yours. What is your average position holding time? How many trades per day? I am still trying to prove that my strategy is credible, which I think it is especially after hearing about your success. I have an average of 45 roundtrip trades per day with a max of 165 trades and a min of 2 trades. So my position hold time is not very long. The minimum amount of time that I can hold a position is 1 bar. How do these numbers compare to your strategy? Thanks!
wfeagin, my main strategy's average hold time is 2 hours and 10 minutes with approx 0.3 trades/day per equity. Another number you might be interested in is commissions/profit which is about 4.5%. I've always struggled with strategies that had low hold times. I could make them work in backtesting but in the real world I couldn't get them to perform. If you can make it work for you, I'm jealous. Having that many trades really lets the stats work in your favour.
Your strategy has a trading frequency which is on the same order of magnitude as mine. On average my system generates 38 round trip trades per instrument per day. It is my experience that at these frequencies, the quality of trade execution becomes important and can be a decisive factor for shifting the profit-loss distribution into a territory that is attractive risk-adjusted. Quality is defined by the average execution price relative to the price at which the trading signal was generated. For this reason I have invested a lot of time in the development of an algorithmic execution algorithm. Basically, this algorithm kicks in immediately after a trading signal is produced. Within a time horizon of 5 minutes, it selects a set of limit orders which can be changed dynamically at one minute time intervals. Any number of shares that have not been executed at the end of the 5 minute time horizon will be executed using a market order. The core of the execution algorithm is designed using approximate dynamic programming principles. Simulations have shown that it can produce additional alpha, leading to a significant improvement in the profit-loss distribution. How important is the quality of execution in your case?
Your use of Dynamic Programming has peaked my interest I use it quite a bit in my day job (aerospace control systems). Would you mind sharing some of the details?
I have tick data that I then compile into 133 tick bars. From the OHLC of those bars, I backtest and optimize the algo. Is this not recommended?
It is very important. I have not done any dynamic programming to account for this. My thinking is that if I have small orders in a very liquid instrument, then I will just use market orders. To give you an example, my average daily volume is around 10,000 shares per instrument. The instruments daily volume is almost 11 million. So my market impact on the daily volume for this one instrument is < 0.1%. With this type of ratio, I am expecting my orders to get filled very close to my order execution price. I do expect some slippage, but not enough slippage to kill this type of strategy. I still have to verify all of my assumptions at this point. I hope its not a bunch of wishful thinking. What are your thoughts? Thanks!