Intro
The point of this journal will be to share my real money *fully* automated trading results throughout the process of iterative refinement. Backtesting results, while interesting, provide an incomplete picture. The only way to really know how a strategy will perform is to run it for real.
Running a strategy against a live account can be very costly, as such I am using a $500 Forex account trading 1k lots. This setup will remove the pressure and fear of losing money, allowing me to objectively gauge the success of the strategy as I continue to improve it. It also provides an easy mechanism to quickly scale up from 1k lots with "predictable" results.
Design Methods
The strategy is far from complete. In-fact 1 week ago it did not even exist. The strategy is designed by data-mining for patterns in an automated fashion. I have two computers set up to perform the data mining 24/7. One computer is looking for long patterns, while the other is looking for short patterns.
The patterns must contain at least 1000 trades during the in-sample data to even be considered. The in-sample period is done in 4 year increments. For example 2007-2011, or 2002-2006, etc.
The out of sample period contains the years left out. The years 2011-present is considered SACRED. NO data mining occurs during this period. It is only used as a sanity check against the "live" strategy.
Each night, I take the patterns discovered and look for how those patterns can be used to improve the live strategy. The strategy changes will not make it until the live strategy until the weekend unless the changes are important.
Trading Setup/Markets
I am using NinjaTrader running on an Amazon EC2 instance 24/7. This instance is left untouched except during contract rollover each day. I will login briefly just to ensure no NT crashes.
The current strategy is only being trained and developed against EURUSD. However, the strategy is LIVE on the following currencies:
EURUSD
AUDUSD
GBPUSD
NZDUSD
I believe a good strategy should be able to perform well on multiple currency pairs. This also "should" have smoothing effects on draw down.
Going Live
The first iteration of the strategy went live on July 17,2012 at 9:00 AM EST. I will refer to this version of the strategy as v1.
In the first 30 hours, the strategy has performed 15 trades resulting in a gain of $3 leaving current account balance at $503
*Note All results are real money account balances (ie after all commissions etc applied)
I'm not an FX trader, but I'd be interested in seeing a sort of metric of pips per 10k or pips per 100k metric, if that even makes sense.
In the stock world, I am usually interested in profit per share (after fees) to figure out how much wiggle room I have for innovation on a trade. And profit per share also tends to be a function of the time horizon, so, if you'd be willing to share, I'd like to look at hold time in seconds (or minutes) also on these trades. (Maybe a histogram of hold times?)
So on those 15 trades, I'm assuming they were 10k lots. You made 3 pips on 15 lots, so .2 pips per lot profit (after fees)? Is that good in the FX world for a short-term trading strategy?
So on those 15 trades, I'm assuming they were 10k lots. You made 3 pips on 15 lots, so .2 pips per lot profit (after fees)? Is that good in the FX world for a short-term trading strategy?
I am trading 1k lots. So it was 30 pips or 2 pips per trade. This does not give a lot of wiggle room. However, it is after all slippage/commission etc in a real account. So I will take it as a starting point.
The strategy is far from complete. In-fact 1 week ago it did not even exist. The strategy is designed by data-mining for patterns in an automated fashion. I have two computers set up to perform the data mining 24/7. One computer is looking for long patterns, while the other is looking for short patterns.
