Stops, from Newton Linchen

September 4, 2011 |

I'm reading old posts on Daily Spec, specifically an interesting discussion on the topic of Stops, that took place in 2003.

At the end, Faisal mentioned a paper of Dr. Eric Berger on calculating the probability of hitting the barrier (stop or price of the knock-out option).

Could anyone share this paper?

And, any other improvements on the discussion?

P.S. What I find difficult in those answers about not using stops, "no stops required", "stops will double your probability of loss", etc, is the time horizon. We have to give profit to our clients every month, otherwise they will fly away to another firm. How does a "no-stop" policy deals with this? And… in the event of a year such as 2008, (or 2011, in Brazil), a no-stop approach isn't a sure path to greater loss?

And for those not using stops, what's the time horizon? Do they NEVER take a loss? When will the loss ever be taken? (They put that part of their portfolio in a box, never to look at it again?)

Phil McDonnell writes: 

In introductory Stat there is almost always a section on calculating the probability of a certain sized observation in a normal distribution. That calculation tells you the probability of being at or above a certain level after a fixed period of time. To calculate the probability of having touched that level at some time during the fixed period of time simply double the probability. The reason is the Reflection Principle. For every path which wound up past that level, there is an equal and opposite path that was 'reflected' back from the level.

The same thing applies to stops on the downside. Note that all of this does not assume an infinite time period, but rather a fixed period of time - kind of an investment planning window if you will. The reason for this is that the variance always grows as time is extended. Given large enough variance then almost any price will be hit eventually both up and down.

Newton Linchen replies:


This is the subject that I have read several times in your book, but still can't imagine how this would fit someone who must give monthly performance reports, instead of a well-capitalized investor who can stand some period in the red…

This approach doesn't seem to fit short-term trading (trading with less than 30 days), unless you exit the position until the end of the month, which would be your final "chance" for the position to work.

And in the time horizon of one month, one would have to ensure protection against the ineluctable downside.

Phil McDonnell adds: 

The main point is that stop losses will double the number of losing trades at your 'maximum loss' level. So you will report more losing months. Is that what you want? So if not stops then choosing a prudent position size for a trade is really the way to control risk.

Most of my comments are made on the theoretical basis based on what should happen. But in fact in my empirical tests, reality is worse. Using stops actually hurts expected returns in most cases. Theory says that should not happen. Theoretically expectation should remain unchanged if you use stocks but that is not the case.

Attached is a little table excerpted from Larry Connors and Cesar Alvarez's book Short term Strategies That Work. It shows that for a particular simple system the average P/L was .69 per trade. with 69.81% winners. But when you add a stop at the 1% level the return plummets to .19% and only 26.89% winners. For all stop levels tested up to the 50% level the returns were lower when one added stops to the strategy.

So looking at probability per trade and return per trade stops seems worse. But theory says that they may help by reducing variance. So far as I can see that is the only good thing about them.


Dave G writes: 

Hedging makes more sense than the antiquated and disgraced use of "stops".





Speak your mind

8 Comments so far

  1. Ed on September 4, 2011 9:57 pm

    2 things if not stops:

    1.) Continually reevaluate forward expectation over the course of your trade. If it goes negative or is not a favorable risk, exit regardless of if the trade is a profit or loss. For example a swing from X day low to X day high might be a good time to exit.

    2.) position size, both individual positions and total portfolio limits. If one as using leverage, at times the market might force one to de-leverage during a trade, but it should be avoided. Operational leverage is much better than financial leverage, both in business and trading.

    3.) The approach that I have found works for discretionary trading is the above combined with simple time stops. Exit if X Time period have passed regardless of P@L on trade. This keeps one from becoming dead equity or freezing up. If you buy point has an edge most of these time-based exits should be gains. Certainly monitor them. If they are not mostly winners your edge might be suspect.

  2. Arthur on September 5, 2011 7:13 am


    Thank you for all the info and examples. You just gave me an idea. Looking at the data above it seems that, whatever the instrument was traded, it would make sense instead of a stop to put take profit at 1% level. I’ve tried it in forex, the only problem is you get killed if the swing is too large into the opposite direction. How to overcome that? (a 5% stop would not work) Eventually I learned that the best would be to take lots of small losses, and try not to die from 1000 cuts, and take huge wins. I analysed my trades and your data seems about right, with that strategy 18% are positive, but those that are positive are dramatically positive.

  3. jeff watson on September 5, 2011 8:40 am

    Just my $0.02 If you leave a stop with a broker, it becomes part of the market. I leave a big enough footprint on the main market I trade, and have no need to leave more clues or evidence so someone can beat me over the head.

  4. Kermit on September 5, 2011 1:14 pm

    This is an interesting subject, and one that I have struggled with for many years. One of my trading models that is consistently profitable shows an occasional draw down due to a strongly trending market. Even though the profits greatly outweigh the occasional losses, the draw downs are not psychologically easy to take. I initially trained the model on ten years of data ending in 1999 and used out-of-sample data to verify trading results through 2006. Since then I have been running it in real time with no re-training. The model uses no stop loss rules, and I have for years been working to incorporate stop loss rules into the model to try to limit the occasional loss, always keeping the training set from 1989-1999 and everything since as out-of-sample. I have used the genetic algorithm to try to optimize any stop loss rules I have incorporated into the AI model. As of yet, everything that I have done to attempt to limit draw downs has degraded the trading model so much that it has made the model of very limited value. What initially seemed like it should be so simple has proved to be very difficult if not impossible.

    One final thought - all we need is a model that would generate a simple binary signal. One or zero - trend or no trend. Everything after that seems pretty elementary. Again, this sounds simple, unless you have tried to do it.

  5. Arthur on September 5, 2011 2:30 pm

    @Jeff, we talked about this, you are right, mental stops or algos works, thank god there is no need for broker stops.
    One thing I didn’t mention above is that once the initial stop of 1% is not hit it only works if there is no trailing stop. Take profit target starts to depend on the number of trades it takes to be positive (after all the years still have trouble with it).

  6. duncan on September 5, 2011 7:57 pm

    Fixed stops also give a false sense of security that I think is dangerous. Exit when the prospective foward looking expectation is no longer in your favor. Any other reason to leave a trade takes away from your edge and may eliminate that edge entirely.

  7. vic on September 6, 2011 12:12 pm

    one would recommend that all training models be updted through 2008 at least. vic

  8. Kermit on September 10, 2011 4:17 pm

    “one would recommend that all training models be updted through 2008 at least. vic ”

    IMHO, not necessarily. There are times when, after decades of making a living in a particular market, you just have a feeling that something is not “normal” and it is time to be on the sidelines. There is probably no way that you can teach a model to recognize this, as there are most likely not enough examples of similar conditions to use in the training data set. Including a year like 2008 *in the training set* could very well degrade the model rather than improve it. I do, however, very much like to see how the model handled a year like 2008, but I like to see it in the out-of-sample data set and not the training set.


Resources & Links