Real Money Automated Trading Journal

Quote from frostengine:

I do not believe OandA works with NinjaTrader? I also believe, there are fundamental problems with Backtesting. It can help give you an"idea" how a strategy will perform, but is not much more than that.
More...

You are correct. OandA doesn't work with NT. Sorry for that suggestion.

Also, I agree with your assertion that there are "fundamental problems with backtesting". Having said that, there is no way you can test a strategy and can have confidence in it if the historical data used to develop the strategy is different from the live trading data (in the sense that live data for today is different from the data if you query it tomorrow from the historical server of MB Trading.

This difference between historical and live data (because they come from different servers) is the bigger problem here. This is easily solvable if you ditch your currrent broker and rather move to a broker whose historical and real time data are similarly generated. Do you see what I mean? IB and CME futures are the obvious choices that will work with NT.
 
Frost,

would you mind also putting your total cumulative Pips gained/lost since starting this strategy along with the other stats?
 
Since starting this journal:
Trades: 91
Pips: +188

Its worth noting almost ALL of the profit has come from the EUR/USD pair. Which makes sense as the strategy up to this point has been trained against that pair.

over the next few days, I will be introducing strategies specific to the other pairs as well.
 
Quote from frostengine:

Deciding whether or not to stay with V1 or move to V2 is proving very difficult. The same reason I MUST trade live to evaluate a strategy, is what is preventing me from adapting quickly.

For example, I have data from 4 total sources, 2 different static sources, MB Trading loading tick data from Ninjatrader, and MB Trading downloading tick data then importing.

With all 4 sources I get VASTLY different backtesting results. These inconsistencies is why I term backtesting as just "interesting".

A good way to illustrate this is as follows:
1. Live account for Thurs and Friday +$6.16 on EURO
2. Backesting loading data THROUGH MBTrading -.10 on EURO
3. Market Replay, +5.50 on EURO

As you can see the backtesting results are not even REMOTELY close to what actually occured in both replay and live. This further proves why you can not trust backtesting.
More...

Hi,

Good journal, i like to follow

However, do you have a demo account running side by side with your live trading experiment?

Also, you have said you are expecting big DD around the corner, what is your measures to avoid/deal with that?

Could you include some more stat for your automated strategy? Just to enrich your experiment for further development

Good luck
 
I am making a fairly substantial change to the system starting tonight. I have created a second $500 live account. I will have a 4 range bar strategy running on account #1 and a 10 min bar strategy running on account #2.

The goal is to produce a smoother equity curve, by having two unrelated strategies running independent of one another, I should see a smoothing affect in the overall account equity.

At start the strategy breakup will be as follows:
Account #1
EUR/USD Range 4 Strategy

Account #2
EUR/USD 10 min
GBP/USD 10 min
NZD/USD 10 min
AUD/USD 10 min

For account #2, all 4 strategies were trained specifically for the given currency pair. Development efforts will now switch to designing strategies for the other 3 currency pairs on account #1.

Any thoughts or comments about this approach?
 
Quote from frostengine:


For account #2, all 4 strategies were trained specifically for the given currency pair.
More...

"Trained" - Are you using Neural networks? Have you found them useful in terms of consistent out of sample performance on unseen data in trading?
 
You can think of it in terms of neural networks, its actually slightly different. Its really a way of data mining for patterns. Little different than a traditional neural net.

Yes, it appears to perform very well out of sample. However, this live experiment will be the true test
 
You must log in or register to reply here.
Back
Top