100% Mech Stock Day Trade System

[markies21]

I've spent the last month building a mech day trading system for stocks.

The Long system was as follows (the Short is the reverse).

a) take the midpoint of the opening 5 minute range as the central pivot point
b) look for a failed long break out
c) if the long breakout fails, place a buy limit order in below the central pivot point by about double that amount (exact amount is determined by previous 9 day trading ranges)
d) exit after 1400hrs by trailing the highest high of the last 15 minutes (limit orders)
e) there is also some logic when the trade runs against you; which to make things simple I won't cover right now

I then backtested this strategy on the Top 500 traded stocks by turnover, and then selected all that returned PF > 2.0 with a Coefficient of Variation less than 300% for the last 12 months. That gets about 20 or 30 stocks.

I then walk forward these 20-30 stocks over the next month to see what they return.

The first walkforward month - August 2008 returned 30% on my account.
But that turned out to by a folly as June & January returned on average about 0-5%.
I stopped testing after that as I was somewhat disheartened - each monthly iteration takes about 2 days (I've automated as much as I can).

I think the system is 75% there, it is a case of making it profitable all months which was my original goal (I define that as >10%/pcm in backtesting, after slippage and commissions).

I would like to collaborate with someone that is using a similar technique or interested in developing a mechincal day trading strategy, if there is anyone out there interested please get in touch.

Thanks in advance.

 

[tomorton]

Good grief but you're thorough.

I can't daytrade as its against my nature and screen-time availability so I'm only an armchair commentator but I don't believe daytrading stocks in the UK is viable via spreadbetting, the spreads are way too high and the price bias will kill you. While your work is thorough, are you aiming at the right target? And if you are, you should put a little money in as a test, because you might be surprised how the game changes when you put real money in.

You deserve to do well, I hope you can report god news soon.

 

[markies21]

What I'm talking about

Thanks tomorton, your reply made me revisit what I've done so far . BTW - it trades US stocks that have high volume and low or no spread.

Here's some pics of where I'm at. I've this time included a short example (reverse for long trades).

The good_fade.jpg attachment shows two fades that are successful. The stock was identified in backtesting to have strong mean reversion and this statistical inference was valid in the following month. These trades make about $2000 after 1c slippage each side and $30 commission. Great.

Some stocks don't behave the same way - which I've attached in bad_fade.jpg. There is a failed breakout down, and we thus set ourself up to sell again, but the stock breaks out strongly and we got in too early.

I can increase the amount the failed breakout must occur at, but this removes profitable as well as unprofitable trades. I can raise the limit order at which I sell at - but this has the same effect.

For example: By increasing the amount the failed break out must occur by and reducing the profit target by the same amount, I can turn a loss for the Short trades in January from -$5286 (113 trades), to a profit of $15,881 (63 trades) - not bad for a $100,000 account.

On the other hand, the same change causes June $18,045 (123 trades) to post a loss of -$119 (75 trades). By filtering out the bad trades in one month, we are also filtering out the good trades in another.

All my pivot points are calculated off the opening 5 minutes using the average of the last 9 days daily ranges.

So my question to all is: Is there a way of either filtering the bad trades (when there is too much strength in the buying/selling action and we should not be fading), or dynamically working out where we should be entering during the day instead of at the start of the day?

 

[markies21]

finished

Okay, finally got this to work (*finally*). I needed to change to 1 minute bars (was getting to much lag off the 5 min bars), add in a couple more conditions to the trade. Also needed to add multiple timeframes (daily bars) to the calculation to determine when to (and when not to) trade.

Returns 3.32PF out of sample (400 trades) across 15 high vol stocks for the year to August.

The system goes into test mode today for the next month, although I've added 1c slippage & $14 comms each way in the backtesting, concerned about how this would affect me in RL and whether or not the fills are achievable.

 

[dnof]

Are you still using this system - I am attempting something similar....

I've spent the last month building a mech day trading system for stocks.

The Long system was as follows (the Short is the reverse).

a) take the midpoint of the opening 5 minute range as the central pivot point
b) look for a failed long break out
c) if the long breakout fails, place a buy limit order in below the central pivot point by about double that amount (exact amount is determined by previous 9 day trading ranges)
d) exit after 1400hrs by trailing the highest high of the last 15 minutes (limit orders)
e) there is also some logic when the trade runs against you; which to make things simple I won't cover right now

I then backtested this strategy on the Top 500 traded stocks by turnover, and then selected all that returned PF > 2.0 with a Coefficient of Variation less than 300% for the last 12 months. That gets about 20 or 30 stocks.

I then walk forward these 20-30 stocks over the next month to see what they return.

The first walkforward month - August 2008 returned 30% on my account.
But that turned out to by a folly as June & January returned on average about 0-5%.
I stopped testing after that as I was somewhat disheartened - each monthly iteration takes about 2 days (I've automated as much as I can).

I think the system is 75% there, it is a case of making it profitable all months which was my original goal (I define that as >10%/pcm in backtesting, after slippage and commissions).

I would like to collaborate with someone that is using a similar technique or interested in developing a mechincal day trading strategy, if there is anyone out there interested please get in touch.

Thanks in advance.

 

[markies21]

Yes, it's running on one of my trade servers since October.

I never live traded any Long positions, trading shorts only.

Hows it going with your trading / system development?