Successful System Traders

[Slope Trader]

Hi Dom,

How much do you focus on the magnitude of the Profit Factor when you analyze your backtest results?

Have you drawn conclusions about what minimum PF value a system should have in order to work well in live trading?

The method I'm currently using repeatedly gave PFs from 1.6-1.9 in backtesting the four different markets that I trade, but live trading results are giving numbers of 1.3-1.5. I'm quite happy with this, but I keep testing modifications, especially to the exit strategy, to try to increase it .

Here are the performance figures for my live trading of CL AlwaysIn since I switched to v46 on October 15:

- 137 trades
- win% : 59%
- avg W/avg L: 0.92
- avg net/trade : $86
- P/F : 1.38

Now, the global view since I started trading that system, all versions combined, is the following:

- 560 trades
- win% : 53%
- avg W/avg L: 0.93
- avg net/trade : $19
- P/F : 1.08

 

[dom993]

Hi Dom,

How much do you focus on the magnitude of the Profit Factor when you analyze your backtest results?

Have you drawn conclusions about what minimum PF value a system should have in order to work well in live trading?

The method I'm currently using repeatedly gave PFs from 1.6-1.9 in backtesting the four different markets that I trade, but live trading results are giving numbers of 1.3-1.5. I'm quite happy with this, but I keep testing modifications, especially to the exit strategy, to try to increase it .

It depends on the intent behind the strategy.

For CL AlwaysIn, I wanted to get 750+ trades / year, the largest statistical base as possible, at least in excess of 2000 for a couple patterns (I have 4000+ & 2800+ for the 2 strongest patterns, with overlaps), and I was prepared to settle on a "low" P/F to achieve this. But really low it is, ranging from 1.17 in 2007 to 1.57 in 2008, the 2007..2013 global P/F being 1.36 (1.29 excluding 2008). But the average net per trade is decent at $75 (excluding 2008, which own avg net / trade is $192 for 1178 trades) & gives enough room for the inevitable lower live performance to remain positive.

For CL Selective, I wanted to find the best patterns, in terms of avg. net / trade, P/F & win%. The corresponding patterns have a much lower base (a couple hundred to a thousand), 4 out of 7 are positive very year on the 10.5 years backtest, the other 3 are positive or BE 9(.5) years out of 10.5, with a number of trades / year pretty stable, which gives me some confidence to balance the low footprint of these patterns. The resulting performance figures are much higher (avg net/trade > $200, win% ~60%, P/F 1.85) but I think this is much needed given the smaller base of these patterns.

What I focus on no matter what, is max.DD, backtest vs MonteCarlo simulations (using the trade distribution out of backtest). I would be very worried if the backtest DD was significantly higher than the MC mean for the same number of trades, in the case of both systems the backtest DD is either at or below the mean.

Last, I try to get at least a 2:1 reward:risk ratio, using the 1-year average ending P&L (MC sim) vs my system-stop (which is at least, twice the max historical DD, sometimes more, depending on the MC mean & stdev for the 1-year DD).

 

[Slope Trader]

It depends on the intent behind the strategy....

Dom,

Thank you.

Very interesting approach.

 

[ehsmama]

Are there any successful system traders out there?

I would like to hear from automated traders who have been running profitable systems for more than a couple of years.

Maybe they can advise on some of the pitfalls they encountered?

I look at my strategies and constantly think either they are too simple to work or over-fitted and so forth. I have been running live for few months now and so far small profit but there have been many times where i wanted to override a trade.

Sometimes i think maybe this is all a big waste of time and unless you have real edge like HFT or arbitrage then odds are this is random walk.

I still have my doubts about my systems. Maybe someone successful can share their story. Would love to hear some success stories for inspiration.

I have been trading Systems for a while. The thoughts that come to your mind used to come to my mind too...Actually I analyzed the reasons behind my doubts about my system arise when one of these three things happen ---
1. The TIME between 2 new Equity highs is higher than what I hope or want to be
2. Drawdowns are larger than what I really hope or wish for
3. Returns are smaller than what I wish for.

One these three criteria are met by your system. You will follow the system without question

 

[giatabaza]

i'm really hooked on www.aiforecasts.com, since it narrows down the search-for-trade quest through its algorithm by creating stock charts and assesing them in order to provide a ranking. It also virtually buys based on those ranks showing its performance and saying why. pretty cool.

 

[dom993]

i'm really hooked on www.aiforecasts.com, since it narrows down the search-for-trade quest through its algorithm by creating stock charts and assesing them in order to provide a ranking. It also virtually buys based on those ranks showing its performance and saying why. pretty cool.

Thank you Mr. Shill

 

[giatabaza]

Thank you Mr. Shill

thats not cool. a friend sent me a link from his mobile and i have not seen one similar service like it. in any case its a free site, so if you think i have some benefit from sharing this you are absolutely wrong.

 

[Emil]

I’ve been trading systematically since start of 2011, swing trading stocks and ETFs with holding periods of a 1-5 days and more recently intraday minutes to hours. My returns are nothing fancy, CAGR of 18% and a Sharpe of just above 1 with a max dd of 16%. Since the end of 2012 everything has been automated. Most big leaps have come during the summer breaks when I have time to focus purely on trading, this summer I spent researching and programming the day trading-part of my trading system as well as adding lots of new strategies to my lineup. I used to only trade around five instruments since it was easier to monitor, now a trade around a hundred. I found the good trading is mostly boring repetitive work, you have a small edge and it takes time for the expected value to pan out. I’ve made 290 trades so far this year; thankfully I haven’t had to watch the computer screen from 08:45 to 17:30 every day to make them.

 

[mushinseeker]

Hi . Am very new to system trading and just completed some backtests in Amibroker. Clicked on report and other than obvious ones like %W/L , saw some stats (ie last 7 numbers) which are foreign to me. Could the pros here offer an opinion on whether these backtest numbers are great, ok , or a trainwreck that needs to be reworked? Thanks. This covers 5 spot fx pairs from 8.2013 to yesterday. 10,000 units


Trade # 854
avg pnl $14
avg bars held 204 10m bars
win% 63%
Recovery factor 11
CAR/Maxdd 21
profit factor 1.72
rr ratio 33
ulcer index .03
sharpe 2.75
K Ratio .0374

 

[dom993]

Hi . Am very new to system trading and just completed some backtests in Amibroker. Clicked on report and other than obvious ones like %W/L , saw some stats (ie last 7 numbers) which are foreign to me. Could the pros here offer an opinion on whether these backtest numbers are great, ok , or a trainwreck that needs to be reworked? Thanks. This covers 5 spot fx pairs from 8.2013 to yesterday. 10,000 units


Trade # 854
avg pnl $14
avg bars held 204 10m bars
win% 63%
Recovery factor 11
CAR/Maxdd 21
profit factor 1.72
rr ratio 33
ulcer index .03
sharpe 2.75
K Ratio .0374

I am certainly not a "pro" when it comes to FX, but in my experience, the first aspects to look at with a new strategy (and especially for someone new to this), are revolving around "how close/far are the backtest results from what I would have gotten trading it live":

1. Are the fills realistics ?

• 
  For LMT orders, the correct side of the bid/ask has to go through the limit price for you to have a chance of getting filled - Not sure if AmiBroker does backtest for FX using both bid & ask, and of course you need to have both in your historical data

• 
  For MKT & STP orders, you must assume a standard level of slippage for each order executed - of course, off the correct side of the bid/ask. Most likely, actual slippage will be considerably more than your worse assumption (actual slippage can't be calibrated by any other way than live-trading at a decent size, and carefully recording & comparing actual fills vs after-the-fact "backtest" fills for the same execution)

• 
  If your platform only supports backtesting with either bid or ask but not both, you are in the dark regardless of the assumed size of that spread, because FX spreads do vary quite considerably, in particular in the minutes before/after planned news release

• 
  My recommendation would be to backtest using FX futures historical data - much more reliable backtest off a "reliable" transactions history

2. Is the historical data reliable?

• 
  I buy my historical data (futures) from TickData. I doubt any FX historical data can be trusted, and certainly not coming from your broker.

3. Is the backtest reliable ?

• 
  There is no way around this - spot-check at least 100 trades, and verify the fill price makes sense. There are many ways to mis-use a backtest engine.

As for performance metrics ... there is no point discussing these figures before you drill down on the above set of questions.

Cheers
D.