Maximum Loss Stops: Do you really need them?

We hear it all the time. “You must use stops.” And most of us use them. But do you know how they change your strategy results? Are they improving your results by giving you higher CAR or lower maximum drawdown? Recently I was speaking with a reader about this topic and he insisted that it you had to have stops to trade. Well, does one?

Early in my time while working with Larry Connors, he had me do a test on a mean reversion strategy and of course he said put stops in. We tried 5%, 7.5% and 10% maximum loss stops. I ran the test and noticed that higher stops had better results. I then got curious and tested larger and larger stops all the to a 50% stop, which at the level really is not a stop. A general pattern emerged that the larger the stop the better the results. I which point I decided to completely remove stops and this gave the best results. We could not believe it but tests on other mean reversion strategies showed the same thing. From that point on we traded our mean reversion strategies with no stops.

I trade two stock mean reversion strategies with no stops. But it has been 10 years since that original stop research. Does not using maximum loss stops on a mean reversion strategy still produce the best results in terms of Compounded Annual Growth or Maximum Draw Down?

The Strategy

I will be adding stops to one of the most popular posts, Simple Ideas for a Mean Reversion Strategy with Good Results. I will have to make minor changes to test it.

Testing timeframe is from 1/1/2005 to 6/30/2015.

Setup

  • Close greater than 100-day moving average
  • Close less than the 5-day moving average
  • 3 lower lows
  • Member of the Russell 1000
  • 21-day moving average of dollar-volume greater than $10 million
  • Price as trade greater than 1

Buy

  • Rank stocks from high to low using 100 day Historical Volatility
  • Place orders for the top ranked stocks such that if they all get filled will be 100% invested. This means orders will often not get filled even though lower ranked stocks may have
  • Set a limit buy order for the next day if price falls another .5 times 10-day average true range.

Sell

  • Close is greater than the previous day’s close
  • Sell on the next open

Misc.

  • Maximum of 15 positions [Corrected on 10/7/2015 ]
  • 6.7% allocation per position [Corrected on 10/7/2015 ]
  • No margin is used

 

General Info

The big change from the original test is only placing orders for the top ranked stocks. I did this to avoid having to run lots of Monte Carlo simulations which would have taken a very long time to run.

All stops tested will be end of day stops. If the closing price is below our stop point, then exit on the open on the next day.

I am not testing other stops because it greatly complicates the testing as each type affects the other. I am keeping position size constant because it also can affect the results. We are trying to isolate the results of only a max loss stop.

 

Base case: No Stops

150805a

These are good results considering no optimizing has been done. Surely a max loss stop will help. We will focus on Compounded Annual Return(CAR), Maximum Drawdown (MDD), the average of the 5 worst drawdowns and the average % p/l of the worst 50 trades (Avg % loss 50 worst).

 

Percent Max Loss Stop

The first test is using a simple percent max loss stop. These tests generated a lot of data which you can get in the spreadsheet.

150805b

On the far left we have the no stop case. Using no stop produces the best CAR of 14.1. The worst CAR is 12.0 at a 12% stop. OK, maybe we are willing to give up some CAR if the drawdowns and worst trades come down. The MDD goes from 20.7 to as 17.2 low as at a 3% stop. A 15% improvement which is good but not great. The average of the top 5 drawdowns also improves from 13.8 to as low as 13.4. A minimal improvement. The worst trades drop from 17.2 to 14.4 again an improvement but still a big losses. What is interesting here is that at around 15% stop, you are now making your worst trades worse than not even having stop. If I had to pick a stop to trade it would be at 4%. I like keeping things simple. If was evaluating this rule to add or not, I would not do it.

Using ATR for Maximum Stop Loss

Next we look at using a factor of the 10 period ATR.

150805c

The same general pattern holds. Using a .1*ATR10 really has significant impact. A common value for the factor that I read about is using 2.5 or 3. At that level, the results are very similar to not using a stop. Most of the bang comes with a factor under 1. Again, I probably would trade none of these values. Adding complexity but not getting enough reward for it.

Bar Stop

This is not a maximum loss in the traditional sense. If you are still in the stock after N bars, you exit.

150805d

Now this is interesting. The exit after 1 bar, meaning if you entered on Monday, you exit on the open on Tuesday, really improves the negative numbers but a significant hit to CAR. This stop would warrant more investigation but looks promising.

Spreadsheet

Fill the form below to get the spreadsheet with lots of additional information. This includes yearly breakdown and top 5 drawdowns.

Final Thoughts

The results for using stops vs not are not as clear cut as when I did the tests ten years ago. One factor that may influence the results is the exit. The exit of one up day is not a very traditional exit as opposed to a Close above a short moving average.

A question often asked, if you use no stop how do you determine position size? One could use the average losing trade or the median losing trade. Or some factor of those values.

Backtesting platform used: AmiBroker. Data provider: Norgate Data

 

Guide the research

Do you want to see how intraday stops would do? What about checking the exit from up-day to crossing above the five day moving average? What about profit targets?

I am trying something new. This survey is open until 8/19/2015. Fill in the survey if you want to see more on this area of research. Or if you have a different idea on the stop research, enter it in the comment section below.

Backtesting platform used: AmiBroker. Data provider:Norgate Data (referral link)

Good quant trading,


Results

See test results from this survey at Stop Losses and Profit Targets. Plus Happy Birthday Excel!

Answer %
Change exit to Close > MA5 39%
Profit Targets (intra-day) 25%
Intraday stops 20%
Profit Targets (using close) 13%
No more research on this topic 3%

Fill in for free spreadsheet:

spreadsheeticon

 
Visited 36 times, 1 visit(s) today

Click Here to Leave a Comment Below

matt haines - August 5, 2015 Reply

Hi Cesar!

I’ve found roughly the same thing. When people say “always use stops” they are only thinking inside the momentum/trend-following world. Mean reversion is in many ways the opposite.

Where trend-following says cut your losers early, mean reversion says hang onto your hat with losses, because things will (hopefully) get better.

Where trend following says let your winner run, mean reversion says get out while you have a profit, because it might be short-lived.

When I develop a new mean-reversion strategy, stops are the last thing I experiment with. First I set a fixed duration exit. Usually it’s small duration, because any technical signal I have will dissipate over too long a period of time. I’ll start with maybe a 5 day duration. If my entry can’t make money over a fixed duration, it’s probably a bad concept to begin with.

Once it’s showing an overall positive expectancy, then I’ll see if setting a profit target helps. Sometimes it does, but sometimes it doesn’t.

And only then do I throw in a stop loss test. For mean reversion systems, usually it shows no benefit. Occasionally a really wide stop (e.g. 30%) will show a minor benefit. But that’s close to not bothering with a stop at all. My duration is usually my only stop.

The unfortunate nature of equities is the overnight gap. No stop loss can prevent it. Trading individual stocks will always be subject to this.

As a side note, one way the overnight gap problem might be reduced is to enter a trade *after* the overnight gap happens. While I haven’t tested it, I think lighting doesn’t strike twice, i.e. when you’ve had one big gap downward, you’re less likely to have another one in the next few days. If one develops a system that incorporates the gap, perhaps it reduces the chances of a second gap? I will have to go test that (and post the results on my blog).

Keep up the good work!

-Matt

Does Lightning Strike Twice? | Throwing Good Money After Bad - August 5, 2015 Reply

[…] was reading a fine blog post by Cesar Alvarez, where he discussed the use of maximum-loss stops with mean reversion systems, and how they usually […]

Quantocracy's Daily Wrap for 08/05/2015 | Quantocracy - August 5, 2015 Reply

[…] Maximum Loss Stops: Do you really need them? [Alvarez Quant Trading] We hear it all the time. You must use stops. And most of us use them. But do you know how they change your strategy results? Are they improving your results by giving you higher CAR or lower maximum drawdown? Recently I was speaking with a reader about this topic and he insisted that it you had to have stops to trade. Well, does one? Early in my time while working with La […]

Simple Ideas for a Mean Reversion Strategy with Good Results | Alvarez Quant Trading - August 7, 2015 Reply

[…] Added on 8/5/2015: Want to see how a maximum loss stop changes the results, read Maximum Loss Stops: Do you really need them? […]

Mark Schmid - August 11, 2015 Reply

Hi Cesar,

Another great post, the results are consistent with my research into using stops with a mean reversion system, although trading without them has its own implications as you tend to have the occasional heavy loss to deal with.

The one thing i’m wondering is to what extent is it important that the exit is paired with the entry and whether this would change the results. For example, the simple system in your research is buy an oversold stock as it becomes more oversold intraday. There’s no magic to the 0.5x ATR entry level and therefore there is no knowing where the stock will make a reversal. As such, the concept of a stop doesn’t pair well with the entry as whatever level you choose, the stock may not have bottomed yet (in a short term sense) until it becomes so large as to be worthless.

Have you looked at testing a daily oversold set-up with a simple intraday trend following entry e.g. n-bar breakout and whether (a) Such an entry improves results on its own and (b) Stops have value with such an entry?

Thanks for the post!

Mark

    Cesar Alvarez - August 11, 2015 Reply

    Mark, I don’t have intraday data to test n-bar breakout. It is a good thought.

Maximum Loss Stops: Do you really need them? | SAMUELSSONS RAPPORT - August 21, 2015 Reply

[…] Källa: Maximum Loss Stops: Do you really need them? […]

Tristan Rhodes - September 9, 2015 Reply

Cesar, if you want to backtest your strategies using intraday data, have you considered Quantopian.com? It is free to use.

Cheers,

Tristan

    Cesar Alvarez - September 10, 2015 Reply

    I did not know that. I will have to take a look. My biggest concern is dealing with survivorship bias and having index constituent data.

Stop Losses and Profit Targets. Plus Happy Birthday Excel! | Alvarez Quant Trading - September 30, 2015 Reply

[…] the post, Maximum Loss Stops: Do you really need them?, we looked at how maximum loss stops changed the results of a mean reversion strategy. At the end […]

Maximum Loss Stops: Do you really need them? | XLNtrading - October 29, 2015 Reply

[…] Source: Maximum Loss Stops: Do you really need them? | Alvarez Quant Trading […]

Stopplosser, är det verkligen något man alltid behöver? | SAMUELSSONS RAPPORT - October 30, 2015 Reply

[…] Källa: Maximum Loss Stops: Do you really need them? | Alvarez Quant Trading […]

Leave a Reply: