This weekly trade review is part 5 of my progress on rebuilding my trading system to be more systematic and semi-automated.
I am currently working on the auto entry, with discretionary context and discretionary buy/sell areas.
The plan for this week was to test the system live with 1 lots.
Unfortunately, the week was cut short by the trading computer completely shitting the bed on Thursday morning.
So three days of testing was not enough sample period to make any serious deductions about the performance of the auto triggers.
What have I learnt
The auto triggers send the orders to the live execution platform and they match up.
The execution speed is faster than manually entering orders on the trigger setup.
Even with me hovering over the order button on the DOM.
There was one order that was sent off prematurely, this was the fault of my coding, this particular trigger was referencing the wrong level.
That I must remember to back up the system every time I make a change. If it is a major change then this is saved as a new version, but if I do minor edits I must back up the system to the hard drive.
On Wednesday, I had a computer crash, which was precursor to the fucker giving me the finger on Thursday morning. And on rebooting the auto trigger reverted to a previous version, which I did not notice for a couple of trades.
Down the line, I will need to work on having different versions of the trigger, i.e. having settings for low, normal and high volume. Now, they miss entries in low volume, get the majority in normal and signal too much in high volume.
I have discussed this with a good trading friend and have some ideas to take this forward. I will start to work on this next week, but will not be making any adjustments to the core triggers that I am testing now. This phase is to ensure that it works and I have confidence in it.
I do not like making loads of changes at the same time, as it makes it difficult to correlated any performance differences with the changes made. It can be tempting to make more than one change at a time to speed up the testing process but in my opinion this rarely saves time in the long run.
This is something I learnt from racing.
If we made 5 changes to the boat and it went quicker, which change is responsible, and it can even be the case that 2 changes was responsible for the boat going faster, 1 made no difference, and 2 made the boat go slower to the original benchmark, but the cumulative effect was the boat going faster. Trying to work out, what change is responsible for what, can end up in a confusing mess, which takes longer to sort out then just applying one change at a time.
There was once a racing project that took a racing yacht that was an absolute weapon, made a heap of changes, which were all sound decisions on paper, and turned the weapon into a something with the racing abilities of a shit house. They never got the project back to the original benchmark.
BTW, I was not involved in this, I had already learnt that lesson, a long time ago.
One trade at a time is my motto. Whilst this can be slow, I find it easier to keep control of what I am doing.
There is not enough data to make any sort of assumptions about the performance, but the very limited data indicates that I am having similar results on 1 lot as to my 2-lot trading. This is mainly down to the systematic nature and not missing / skipping the triggers.
And with a similar amount of trades, which points to the auto trigger picking better triggers than I do as a discretionary trader.
These are only indications, as there is no way, enough data to support this.
Important to note, that whilst the triggers are performing well, the rest of the system still requires me to manually do the context and buy / sell areas. The triggers will underperform when I under perform with my context read.
To continue testing with one lot, get another week under my belt and then update the risk / trade management part of the trading system to reflect any minor changes that needed to make. And trade the system normally.
To work on the trigger criteria for normal volume and high volume conditions. The first step of this is how do I define normal volume and high volume conditions.
One of the problems with the triggers auto monitoring the conditions that it will always be lagging. If I use a moving average to define the type of condition, then it must wait for the average to catch up to the current conditions.
I am leaning towards a clear defined manual but systematic approach. But this needs to have more thought as to how this will work and how I clearly defined it. What I like about this is that, I can test the principles and rules, which will give me more data, which I can then use down the line to generate better versions of the trigger.