ES 1/09 backtest results better without stops. Your experience?

[ju1234]

For ES 1-09, I have backtested several strategies with reversal technique, over teh last 10 days (of course a volatile period) with protective stops of 10, 20, 40 ticks. With each strategy, it seems that the results are much much better without any stops at all, although the draw downs were at times really high. If people in this forum don't mind sharing their experince, I would like to hear them. Thanks.

[ctrlbrk]

I would suggest heading over to Elite Trader forums. The ninja forums aren't really about this kind of thing, but the Elite Trader forums are. Tons of great info there.

Mike

[TCT-Peter]

For ES 1-09, I have backtested several strategies with reversal technique, over teh last 10 days (of course a volatile period) with protective stops of 10, 20, 40 ticks. With each strategy, it seems that the results are much much better without any stops at all, although the draw downs were at times really high. If people in this forum don't mind sharing their experince, I would like to hear them. Thanks.

You have come close to answering your own question. This style of trading creates large draw-downs which is not a good idea in itself, but also creates logistical difficulties in deciding the position size for the trade. That is, if you don't know your maximum loss per contract beforehand, then you don't know what your position size should be to achieve a desired maximum potential loss for the trade.

As a general rule, the win/loss statistics will usually be worse when using an initial stop, however this is only one of the statistics that you should check during your back-testing. Another important statistic is the Av. Profit divided by initial risk, which will influence the overall profits that you will make with decent money management applied. A large stop loss will create a very poor result for this statistic.

Regards
Peter

[cunparis]

I would suggest heading over to Elite Trader forums. The ninja forums aren't really about this kind of thing, but the Elite Trader forums are. Tons of great info there.
Mike

I posted a message there and one of the first replies was a guy who called me an idiot. I find ET very childish, full of people with multiple aliases and such. It seems much more civilized here. Post a question about backtesting there and you'll get people telling you backtesting is useless!

To answer the OP's question: I have found in almost every one of my strategies that they work better without stops. So I try to pick a stop that is only hit few very times. It's my "catastrophic stop". It gives me piece of mind knowing that if the market crashes I won't lose more than $x. This peace of mind, like any kind of insurance, has a cost and the cost is slightly less performance.

It's a paradox because everyone says to use stops but if you do the backtest it's easy to see that it's better not too. Larry Williams has an S&P system and he doesn't use stops. So I asked him, why he said in his book to "always use stops" and he doesn't use them. And he replied that he had been trying to find a stop that improves performance but none of them do so he doesn't use them.

PS: FWIW - I lost $22k one day trading his strategy and quit trading it.

[ctrlbrk]

I have to agree about ET's childish underpinnings, but there is also a lot of very intelligent people who are willing to answer questions and explain themselves, and a lot can be learned. At least to me!

I would suggest to stop using a certain 'tick' count, or certain $$ threshold for stops (and profits for that matter). It is better to set these targets based on market conditions, for instance using the average true range (ATR).

When I started changing my strategies to use a calculation like below, they became more profitable. I can't afford to hit even one 'disaster stop', or it would be a disaster!

SetStopLoss("long 1 es", CalculationMode.Price, Close[0] - (ATR(BarsArray[2], 4)[0] * Long1smaatr1), false);

SetProfitTarget("long 1 es", CalculationMode.Price, Close[0] + (ATR(BarsArray[2], 4)[0]) * Long1smaatr2);

BarsArray 2 is a 14m bar, the strategy runs on 1 min bars.

Long1smaatr 1 & 2 are variables that let me control risk/reward ratio. Typically I keep it around 0.8:1.5. It may hurt my overall percentage, but net profit is better. Cut losses quick, let profits ride...

Happy turkey day!

[cunparis]

Thanks for sharing your idea & code. Up until now I found my optimum stop value was this:

SetStopLoss(CalculationMode.Price, Close[0] - 7*ATR(20)[0]);

This is traded multi-timeframe but I use the ATR of the current timeframe. Your idea to use the ATR of a larger timeframe makes sense. And I hadn't thought of using the ATR for profit as well. I'm going to try that.

What's cool here is that everyone uses the same tool so sharing ideas, methods, even code is easy. If there were more activity here it could be better than ET by a mile. I agree there are some great traders there, but all it takes is a few idiots to drive them away and your post goes unnoticed.

[ctrlbrk]

Also, stops don't necessarily have to be about a pre-defined price/limit. Not in my opinion at least.

I code several 'stops' which are 'exit signals'. In other words, if there is a new LL or LH or a 3BR for instance, or if MACD blah blah... or if SMA crosses blah blah... then get out of the trade.

In fact, my trade management stops/exit signals usually fire before the stop loss, so about half the time I don't even hit a stop when I am exiting.

I recently switched my strategies to multi-time frame to accomplish this. Most of my strategies are in trades for 80-120 minutes, but I use 2m, 5m and 14m bars for entry signals. However, I wanted to manage the trade every 1 minute for exits.

I also find it helpful to have two sets of conditions when entering a trade... the first set fires, then the second set refines the entry position. So for instance, the first condition set might be for an SMA cross, and then the second condition set will try for xx bars (ie 10 minutes) to look for the best entry point for that signal, like when MACD diff nears 0, etc.

Ninja is so powerful it is easy to accomplish these things with some basic programming knowledge!

[ctrlbrk]

I agree on the openness of sharing here, and in that spirit...

Example of above concept:

// condition grouping 1
if (CrossAbove(this, that, here, period)

{ nextbarmaybe = true; }

// we've found an entry signal, now lets refine our exact timing

if (nextbarmaybe && nextbarmaybebars < 10)
{
nextbarmaybebars++; // lets only try this for 10 bars, otherwise we wont chase any more

if ((LeaderOfMACD(12, 26, 9).Diff[0] < 1.2) && (Rising(LeaderOfMACD(12, 26, 9).Diff)))
{ EnterLong; }
else
{ nextbarmaybe = false; nextbarmaybebars = 0; }


Just remember to reset all your parms when flat. Basically I like this approach because I find a basic entry, then I refine exact timing from there. I just typed this code so there may be errors.

Enjoy!

[cunparis]

Thanks for sharing this algorithm. This is a great idea. It'll take me a while to try it out though.

I'm currently trying your ATR profit & stop targets. I'm curious if you use the optimizer to find the optimum values? I've been having a love/hate relationship with the optimizer. I often feel I optimize too much.

[ctrlbrk]

I do use the optimizer, but I try to use as long of a period as possible (no less than 1 year, hopefully 3 years).

I also have market replay data that I test manually (and want to shoot myself) because I find many times the backtesting shows trades that don't happen in real life (market replay) or vice versa.

Add to that using it in sim mode for a few days... This helps me feel like the optimized results are not curve fitted.

My general rule of thumb is not to optimize any single param beyond 10%, in other words the difference between 10 and 10+10% (11) is fine, but using 5% (10+5% = 10.5) is too much of a curve fitted solution.

I also try to give more weight to more recent results, so if an optimized pattern is performing better in the last 3 months (what isn't!!) then I tend to use that one.

[cunparis]

I do use the optimizer, but I try to use as long of a period as possible (no less than 1 year, hopefully 3 years).

I also have market replay data that I test manually (and want to shoot myself) because I find many times the backtesting shows trades that don't happen in real life (market replay) or vice versa.

Add to that using it in sim mode for a few days... This helps me feel like the optimized results are not curve fitted.

My general rule of thumb is not to optimize any single param beyond 10%, in other words the difference between 10 and 10+10% (11) is fine, but using 5% (10+5% = 10.5) is too much of a curve fitted solution.

I also try to give more weight to more recent results, so if an optimized pattern is performing better in the last 3 months (what isn't!!) then I tend to use that one.

Let's say you use Bollinger Bands. and you find that the optimal setting for the BB is different for each market and for each timeframe. Do you think that's too much optimization? What I'm currently working on I just use 10,15,20 & 1.0,1.5,2.0 for my choices. I'm not interested to know that 17 1.4 is the optimal value, like you said that's too strict. But maybe 10,15,20 isn't enough?

I'm still learning all this. I remember when I first programmed my first strategy.. I had discovered that if the stoch was in certain ranges the S&P would go up or down. So I had a complicated thing like if stoch between 20-30 it'd go up and 30-35 down, etc. That was pretty funny. So now I'm a bit paranoid about optimizing too much.

[ctrlbrk]

Some people on ET would say that if your strategy is truly going to work, then it should seamlessly work on all markets. I think that is crap, and so I have different strategies for ES than NQ, for instance. I've tried to get working strategies for YM and TF and EMD, but haven't been able to make them work yet. So just that alone tells me either I am way off base with what I think a good strategy is, or, each market is certainly different.

The core principles are the same, but the values for each param are not. I think you should feel free to optimize per symbol (futures) and per time frame.

I've played with trying to write some swing trade strategies for stocks and I failed miserably. I think my mind set with Ninja is about day trading and I haven't figured out how to do swings with Ninja properly. So, I just use Blocks (StockFinder) to identify swings, and Ninja to execute and manage.

Using your BB example, I think you can be a little more strict. I don't use BB's but did previously have a strategy using Keltner which I believe is similar. I remember that one of the values could cause a few point increase/decrease in the area size of the bands, so definitely a tweak from 1.0 to 1.2 might be in order (that is 20%, after all). If it adjusts the outer edge of a price by multiple points then it's significant I would say, depending on the type of strategy you are using.

I think at the end of the day if you can pull up a chart and your strategy makes sense with its executions then job well done. If it's going against the trend and is acting like a mindless bot, then it needs work. And the trick is to make it consistent without having so many if statements that you're curve fitting.

I digress -- but the point is, each market is definitely worthy of optimization.

I smell TURKEY!!!!!!! and dressing... oh.. I love me some dressin'!!!!

[ctrlbrk]

Thanks for sharing this algorithm. This is a great idea. It'll take me a while to try it out though.

I'm currently trying your ATR profit & stop targets. I'm curious if you use the optimizer to find the optimum values? I've been having a love/hate relationship with the optimizer. I often feel I optimize too much.

Good luck -- hope it makes you some extra cash!

[cunparis]

I tried my strategy, written for ES, against NQ and results were the same. Then I tried YM, it was negative. But then after running the optimizer I found that making some of the parameters more strict (less trades) gave me positive results. I now find it works against all the indexes, oil, & gold. But the caveat is that it's not that great because there is 1 trade/week and the avg trade is $100 (this is for ES). So the only way to make any money would be to trade size, which will take a while for me to get comfortable with it. I'll start with 1 contract on a paper account first.

We share similar views, at least we agree and are against the majority opinion of ET. In the Way of the Turtle book (http://www.amazon.com/gp/product/007148664X?ie=UTF8&tag=advanwebamazocom&linkCode=x m2&camp=1789&creativeASIN=007148664X) he talks about overoptimizing. He says that the more you optimize the more likely future results won't match. But he also says that using the optimized values is better than just making them up. So he's got a point. Have to start someone and refine.

I currently have my strategy coded as multi-timeframe. That lets me run it once per symbol. If I optimize each timeframe, which I now think is perfectly valid, then I'll have to run the strategy several times for each symbol. That could get out of hand. So I guess I'll code up the parameters for each timeframe.. like instead of bbPeriod I'll have SixtyMinuteBbPeriod or something like that.

It's funny you mention your stock screener, I got started with stockfetcher but once I learned ninjatrader I haven't used Stockfetcher since. I still have some strategies there that I need to code with NT, but I'm not trading them now due to the increased volatility.

We couldn't find a turkey here in France so we're having a chicken tonight. Happy thanksgiving to you.

[ctrlbrk]

Wow people on Amazon either love or hate that book I am several books behind as-is but just finished High Probability Trading which I found very good. It made several bells go off and I found myself running into the computer room to make changes to the strategy after I read a chapter.

http://www.amazon.com/High-Probability-trading-Marcel-Link/dp/0071381562/ref=pd_bbs_sr_1?ie=UTF8&s=books&qid=1227834436&sr=8-1

Regarding your 1-trade-a-week strategy... I would be concerned because it sounds like it is curve fitted, but hey if it works then awesome! I've yet to be able to make a strategy work with success against currencies or commodities.

As for multi-time frame optimization, yeah I just spend a few minutes up front in the strategy adding all the variables/params then I can optimize from there. Of course I only do this once I find a concept works by manually changing numbers around a bit, then I decide ok this works lets see if we can improve.

Back to the original topic of this particular thread --- stops --- I just read something in the book I mentioned above about using Std Dev for stops. Currently I am using ATR, so I ran into computer room to try StdDev's (told you!) and it's currently running an optimization sequence to see how it works out. I did notice on a chart that for the same period, StdDev is not as averaged out as ATR --- hence the name "Average" True Range! So we'll find out if that is a good or bad thing for my strategy in about 30 minutes...

I am pretty new to Blocks aka StockFinder. But I am really impressed with it so far. I just wish it used C# so I could easily write code in it, I find it's "RealCode" to be junk unfortunately. But it's probably not, I just don't want to learn it. I don't do too many swings because my IB account is not above the 25k SEC requirement yet...

Turkey was good but stuffing is my favorite!

Mike