Automated trading system outcomes and real trading results

2. Development history

2003
Beginning development is dated as 08/01/2003. I selected potential list of different indexes and market derivatives and started to experiment with the data characterizing the market state as I said in description to Figure 1. From a very start I have been using only my own designed software interacting with the market through IB’s API. I do not say I like this java-based platform, but a lot of people use it, the functions are accompanied by easy description. I had tested it and had seen that it is possible to make some research using this API. Five months of the year 2003 fast have gone away in data collecting, analyzing and coding the trading platform.

2004
At the January 2004 I practically finished with the selection of system input parameters.
Again, data analysis, system testing and debugging. More 20,000 written code lines. For fast prototyping, only interface includes hundreds element, I used Borland C++ Builder (Interprise Suite). Some parts and ideas I tested before in MATLAB using Financial Toolbox, Financial Derivatives Toolbox, Financial Time Series Toolbox, Fuzzy Logic Toolbox, GARCH Toolbox, Neural Network Toolbox, Signal Processing Toolbox, and
Statistics Toolbox. Professionally, I am not a programmer. For me it is much easier and faster to test any idea by myself than ask about it guru-programmer. Especially, if I have had often only intuitive understanding what I really wanted. So, about one year ago, in June 2004, the system was ready for forward testing. I decided to start real trading in September 2004 with 1 (one) ES contract (E-MINI S&P 500 Index). Why ES? It includes 500 stocks, hard to manipulate, noisy data, an illusion that ES must be very easy to trade, a lot of newbies trade ES, a lot of professional traders trade ES too, acceptable spread/ commissions structure, big leverage, and good working IB’s simulated stop order on GLOBEX in regular trading hours from 9:30 to 16:00. Ok, I didn’t want any overnight position. Day-by-day system trading, analysis, testing and system mainly logic corrections in September, October, November and December 2004.

2005
This system really makes automated daytrading. I only from time to time find new logic errors and correct it. Right now, I still have feeling that 9 months of automated system trading didn’t reveal possible misinterpretations. Other words, the system can make unpredictable actions in uncommon situation. So, I cannot trade yet more than 1 contract for a while. :cool:
--------------------------------------------

:D
 

Attachments

Quote from OddTrader:

It would be interesting to know if the above testing results were carried out concurrently, with all set-ups/ variables/ factors equal, specifically for your particular evaluation purpose.

:confused:
More...

OddTrader,

Sorry, I just noticed your post. Yes, all of the tests were conducted equally, although not exactly concurrently. I tested them consecutively, and 15 to 30 minutes premarket, when trading was very dead so there would be nothing my inherent system delays to slow the results. The only variables were the route (ARCA/BRUT/ISLD) and the broker.

-Eric
 
Quote from EricP:

OddTrader,

Sorry, I just noticed your post. Yes, all of the tests were conducted equally, although not exactly concurrently. I tested them consecutively, and 15 to 30 minutes premarket, when trading was very dead so there would be nothing my inherent system delays to slow the results. The only variables were the route (ARCA/BRUT/ISLD) and the broker.

-Eric
More...

Thanks Eric.

Do you have any rough ideas as how much variance (say % wise) and what's an acceptable level between system signals vs actual execution?

I guess that's what you try to improve performance. :confused:
 
Quote from nononsense:

making money while sleeping
More...


yesyes but nono, this guy teases into this thread, sounding technicalAndSoForth, but never mentions performance in real world. so, does he make it or not. is he now starting to complain, or market himself, or ... what???
 
Quote from OddTrader:

Thanks Eric.

Do you have any rough ideas as how much variance (say % wise) and what's an acceptable level between system signals vs actual execution?

I guess that's what you try to improve performance. :confused:
More...

It sounds like you are asking about the expected slippage in a system. Your slippage will vary depending on the liquidity in the market you trade, as well as the typical size of your order. For non-marketable limit orders, 'slippage' is not really applicable. I suppose that you could consider limit order slippage to be the amount of time that you are not filled, when the market trades at your price and then reverses.

However, for marketable orders, slippage can easily be measured. I measure the slippage on every single marketable order that I place, and record the average cents per share slippage into my tracking spreadsheet every day. This number will vary up and down, depending on a few good or bad executions. However, over the period of a week or month (thousands of executions), my slippage for marketable orders averages about 0.2 cents per share. I do try to focus on ways to minimize this slippage, whenever possible. Obviously, this is can be a significant cost impact to your trading.

The actual impact of slippage on your trading will be heavily influenced by your average trade profit. For example, a longer term swing system that averages profits of 25 cents per share will not be especially affected by an average slippage of 0.5 cps (2% of profits lost to slippage). However, a very active scalping system that averages profits of only a penny per share will give up over half of it's profits to slippage, with a slippage rate of 0.5 cps average.

To summarize, slippage can be measured and the importance of slippage is highly system dependent. Any strategy backtesting should certainly include a reasonable estimate for the slippage that is to be expected for your markets and typical order size.

-Eric
 
Quote from EricP:

It sounds like you are asking about the expected slippage in a system.
More...

Not sure my concepts below would be clear/ correct:

A. Signal $1.00;
B. Automated (quicker timing) Order $1.01, and Automated (shorter queue) Execution $1.02;
C. Manual (slower timing) Order $1.03, and Manual (longer queue) Execution $1.05

The difference between Order and Execution is slippage, whereas between Signal and Order is not.

I think the third one between Signal and Execution is important.

:confused:
 
Quote from OddTrader:

Not sure my concepts below would be clear/ correct:

A. Signal $1.00;
B. Automated (quicker timing) Order $1.01, and Automated (shorter queue) Execution $1.02;
C. Manual (slower timing) Order $1.03, and Manual (longer queue) Execution $1.05

The difference between Order and Execution is slippage, whereas between Signal and Order is not.

I think the third one between Signal and Execution is important.

:confused:
More...

For automated systems, there is zero difference between signal and order. The order is placed during the same millisecond as the signal is generated.

For a manual system, I would suggest that the slippage could still be the difference between the price available at the time of the signal and the execution price. In that case, a portion of the 'slippage' would be due to the time delay required for manual order placement, and the rest of the slippage could be attributed to other delays transmitting the order to the market, as well as liquidity issues.

However, I think these are merely terminology differences and there is no 'correct' answer to how the word slippage must be used IMO. Obviously, we can all agree on the need to get the best price for every trade entry/exit.

-Eric
 
Quote from EricP:

Obviously, we can all agree on the need to get the best price for every trade entry/exit.
More...

Probably we have to work very hard to improve this result marginally. :)
 
You must log in or register to reply here.
Back
Top