rt ES ATS journal

In the OP, you said that the system is fully automated. Does that mean that it could run unattended if you chose? Is any part discretionary? Is there anything other than net profit ($5k delta) that you use to decide trade size?

You said the system is stop-and-reverse and always in the market, but I noticed that it goes flat (no positions) for a while, and sometimes takes several round turns in the same direction (e.g. long, flat, long, flat, etc). The system looks a bit more complex than it first appeared unless you've been tweaking it on the fly.

[rwk]
 
Quote from rwk:

In the OP, you said that the system is fully automated. Does that mean that it could run unattended if you chose? Is any part discretionary? Is there anything other than net profit ($5k delta) that you use to decide trade size?

You said the system is stop-and-reverse and always in the market, but I noticed that it goes flat (no positions) for a while, and sometimes takes several round turns in the same direction (e.g. long, flat, long, flat, etc). The system looks a bit more complex than it first appeared unless you've been tweaking it on the fly.

[rwk]
More...

It's completely unattended, and I don't do anything except reoptimize every weekend for each instrument I trade.

The system's objective generally is stop-and-reverse. However, I employ profit targets for each position. There are a total of 4, and the position size is broken up as equally as possible among the targets. The system also has the option of "Always in Market", which means that the system must always be in the market with at least 1 contract. If the option is turned off, then the system can take the account flat if all targets are hit. I decided to turn it off, but the performance statistics are fairly insensitive to whether it's on or off. Again, a matter of preference.

For the ES I choose a comfortable 6 points (24 ticks) per target, such as I scale out at 6, 12, 18, and 24 points.
Interestingly, I have found that long-term profitability doesn't improve when using targets, but the volatility of the equity curve somewhat improves. For me, this makes it easier to not get discouraged during times of less-than-desirable performance.

In any case, the essence of the system is that it has no bias for the session. If price moves up by a certain amount, the setup is long. If it moves down by a certain amount, the setup is short. The vast majority of trades end by closing an existing position and opening a new one at the same price in the opposite direction.

rt
 
Two things I forgot to add:

I have done a statistical simulation of limit order fills. In terms of stopping and filling, the fill distribution is largely bimodal. I.e. the order is filled or it is not. From the fill samples I have observed in my live trading, I have found that about 40% of "touch but not pierce" limit orders are filled for less than 20 contracts.

A good way to reshape the fill statistics is to "stack" the orders. Take the example of entering a long position at the ideal price "x". If a short position exists, exit at price "x + tick". Then split the long entry into 3 levels, "x + tick", "x", "x - tick". This reframes the bimodal fill distribution into a more triangular distribution, and overall helps to keep deviation from "filled if touched" simulation orders to a minimum.

The second point of observation concerns target orders, which are also limit orders. If the price touches but does not fill the order by the end of the bar, I have placed a failsafe mechanism to convert the order to market at the close. I have had experiences where price touches but does not fill, and the position reverses in the future without yielding profit. This causes the simulation profit to become significantly discrepant from the live trading profit.

The entire point of all these little tricks is to minimize deviation from the ideal simulated performance. I have always felt that the most difficult aspect of designing robust day-trade systems is not designing setups, entries, and exits, but rather making sure that order execution on an intraday timeframe is feasible.

rt
 
Another rather random, directionless session.

Thursday, 04/23/2009
Security: ESM9
Session PNL: $816.50 USD
To-Date PNL: $9,824.54 USD
Account Balance: $95,201.68 USD
 

Attachments

You must log in or register to reply here.
Back
Top