Progress 3 – Machine Learning

This weekly trade review is part 3 of my progress on rebuilding my trading system to be more systematic and semi-automated.

See My Weekly Trade Review Part 1 and Part 2 for previous updates.

Weekly Trade Review

What have I been doing this week?

Context

I have spent time looking at various market conditions and indications that these are changing.

One area was trend days. Not from the purpose of getting in to these days, but from the angle of recognising them as early as possible to manage risk.

I have done some study in the past which did not yield any significant results, but as my experience grows I thought it would be a good idea to revisit this to see if a fresh look produce anything new.

What have I learnt?

Trend days happen about 20-25% of the time. 20% if using a strict definition, and 25% if using a definition that fits the concept of a one direction with larger than typical range and with a close near the extremes.

There is a lot more days that start of as a trend day but either run out of steam or reverse (22%).

There are days where it starts of a normal day and ends with a trend type day.

Before a trend day starts there is very little difference in delta/volume/range when comparing against a non-trend day.

Trend days and generally one direction moves tend to come from balance / range areas. What these have in common that they have typically built decent Gaussian distribution. This seems more important than the time that is taken to build the balance / range.

Weekly trade review
Decent volume build before breakout.

Alongside this I have continued working on the balance context play, looking at the best way to play the outside in and the breakout.

Ideas I have been studying is the skew and kurtosis, but apart from numbers indicating that a normal distribution curve is there, which can be seen by eye, then there seems to be little to predict breakouts. I.e. it can get a number close to .3 and then breakout fail and continue in balance or just continue in balance.

Weekly trade review
Not so decent build before breakout but has filled in the LVN to become more guassian

And even non-prefect balances breakout and trend.

The best indicator so far is that breaks and closes easily a level that should have produced at least a reaction. And this normally involves a loss. So be it.

Triggers

The work on testing auto triggers is progressing well. A couple of bugs, so time spent sorting that.

Basically, how I combined signals into one signal for the auto trigger to work, did not work as expected during heavy than average volume.

Also, assigning variables that I had already assign to other tasks, does not help.

RT Investor tip, always name the V# in the preferences. Don’t do what I did, and name some but not others and end up trying to find which V# is not being used.

I had the triggers on a separate chart so I could monitor them. But found that was hard to keep an eye on, so moved them to main trigger chart.

Problems I found that, it was not easy to see that the triggers where active.

A couple of times, I forgot to turn them off. As these are execution only triggers then there is no stop loss or target set. First one cost me 25 ticks and the second time, I left both buy and sell triggers active and went for lunch.

The little fucker played the balance area very well, and made 50 ticks over 3 trades, whilst I was mediating and stuffing my face.

Whilst this was an error it does show the possibilities of this approach. Decide context, set the triggers and let them get me in when the conditions are met.

Part of my trigger setup is that they will only take a short if price extreme is equal or higher to a set price (and of course that the trigger is active) and vice versa for short.

This is done by a horizontal line in the chart, so I place the buy line a couple of ticks in front of an area of interest I am interested in and then if price trades this line or lower and certain criteria or meet then the trigger sends an order.

This done either aggressively on when the signal happens or conservatively on the close of the bar.

weekly trade reviews
Auto trigger buy and sell lines

This weekend I will make the indicator that indicates the triggers are active more visible.

The next week, I will be taking the trades live manually in my account on the signal. The purpose of this is to compare my entries against where the auto trigger gets in. As a few of the trades this week seem to be out a few ticks, compared to where they are meant to be placing limit orders.

Other things I am considering is setting an auto stop and target set at knuckle settings, to limit any damage if I forget to turn the little fucker off.

Leave a Reply

Your email address will not be published. Required fields are marked *