Reasons for failur in automatic trading systems

[traderdragon]

I drool when I think about what it would have made with full access to my funds and 2X leverage Its drawdown behavior could easily handle 2X leverage without giving me a heart attack

 

[BertH]

Quote from traderdragon:

Too late, ive already found several in OHLC data. Backtests great for 10 years through all market conditions, and has been running live and exceeding backest simulations for over 3 years. (Exceeding because I use large slippage numbers as padding). Beats 99%+ of random testing.

The edges are out there. But the OHLC data is the most data mined data there is and therefore the most efficient. If you can find a remaining edge in OHLC data, you are doing very well. They are very tough to find and get thinner every year.

I have to assume he meant from the daily, because inevitably isn't all trading based on OHLC on some timeframe?

 

[traderdragon]

No, tick data has no OHLC, its just a constant stream, instead of candles.

Quote from BertH:

I have to assume he meant from the daily, because inevitably isn't all trading based on OHLC on some timeframe?

 

[Cheesy]

I have a system I just launched live on the first of June in a contest at InterbankFX, I am currently in the lead, if I win I get 2000 bucks... the competetion is stiff, I see a lot of good working systems come and go here.. But it's a great place to showcase if your system can produce a profit.

http://66.114.120.20/contest/top.php

The name of my system is Luna Tick

 

[BertH]

Quote from traderdragon:

No, tick data has no OHLC, its just a constant stream, instead of candles.

Thought of that after I posted. I'm dubious he meant tick data, but he might have.
Thanks.

 

[jstox]

Quote from traderdragon:

I sorted the profit from winners to losers.

67 trades todays. 65% winners.

Schweeet!!!

Mine did good also ($1000.30). 27 trades, a little more than normal. Had some initial slippage issues and quickly changed to LMT orders in the program.
Definitely see room for improvement.

I'm not getting excited. My system has faults and fully expect to get bitten in the ass on a series of choppy days, today was good.

Congrats, again.
--jeff

 

[jstox]

Back OT. One of the questions is reasons for failure in AT. I've attached a 3-D graph of my current system. Part of the system has 2 variables for optimization. 1st is a max GetTheHellOut Stop and the 2nd is a ProfitTarget.

If I were to optimize the Stop for best case profit the value would be 11 ticks. If you look at the 3-D graph, this is right on the edge of the cliff. A better setting would be at the center of the plateau region. In this case, I chose 26 ticks.

As for ProfitTarget, I can get better Sharpe and Win Ratios by fixing the number. Instead, I did optimization and chose a value of 0 (i.e., no Profit Target). The net profit associated with no ProfitTarget was possibly worth the gamble. Maybe this is a mistake and I'm gonna re-evaluate. Not really what I wanted to discuss, but it shows the temptations to make wrong decisions.

The point is that a lot of systems have awesome results. But, if you look at their 3-D graph they optimize at the very top of a very steep mountain. No room for margin and they fall off the cliffs. Try to get as much guardband (plateau) around the chosen parameters as possible.

--jeff

 

[WmWaster]

Quote from traderdragon:

No, tick data has no OHLC, its just a constant stream, instead of candles.

Yes, tick data has OHLC, but O = H = L = C for every tick.
Hehe...

Anyway, it still makes use of "C"!

 

[WmWaster]

Quote from WmWaster:

Yes, tick data has OHLC, but O = H = L = C for every tick.
Hehe...

Anyway, it still makes use of "C"!

Correction:
Yes, tick data has OHLC, but O = L // C = H for every tick.
Hehe...

Anyway, it still makes use of "O & C"!

 

[traderdragon]

Good post. I used to use similar charts, but realized I got a better feel just looking at the numbers in a spreadsheet.

I only use systems with what I call "a fat edge". Rules that work with a very wide set of parameters and have no "cliffs" to fall off of. In a 3D picture, they form very large fat islands in the middles, with no abrupt spikes anywhere.

Quote from jstox:

Back OT. One of the questions is reasons for failure in AT. I've attached a 3-D graph of my current system. Part of the system has 2 variables for optimization. 1st is a max GetTheHellOut Stop and the 2nd is a ProfitTarget.

If I were to optimize the Stop for best case profit the value would be 11 ticks. If you look at the 3-D graph, this is right on the edge of the cliff. A better setting would be at the center of the plateau region. In this case, I chose 26 ticks.

As for ProfitTarget, I can get better Sharpe and Win Ratios by fixing the number. Instead, I did optimization and chose a value of 0 (i.e., no Profit Target). The net profit associated with no ProfitTarget was possibly worth the gamble. Maybe this is a mistake and I'm gonna re-evaluate. Not really what I wanted to discuss, but it shows the temptations to make wrong decisions.

The point is that a lot of systems have awesome results. But, if you look at their 3-D graph they optimize at the very top of a very steep mountain. No room for margin and they fall off the cliffs. Try to get as much guardband (plateau) around the chosen parameters as possible.

--jeff