Scaling out (or in)

I've been wanting to wrap up an article I had on scaling for awhile now, and some recent questions I've gotten, and posts I've seen on this topic, got me moving and I finally finished it. I thought I would post this here because this forum has a lot of strong proponents of the concept that scaling is statistically not a good idea. I have also seen a few posts discussing some of the psychological aspects of scaling. As I have done here many times, I like to show the other side of an issue, and so I thought some of the readers might like to read the article. It definitely is a wholly different look at this topic.

It's posted here:

http://www.kanetrading.com/free_art/scaling.html

Good article Jim, thanks for sharing it.

Towards the end you ask "Well, what if a blended, scaled approach yielded a little less return, but had a much smoother equity curve, with less draw-down, and a person decided to choose that because the lesser return was a great trade-off with the much higher comfort level?"

A question, which I think summarizes it all very well.

I've never thought of looking at scaling out of a trade as multiple separate trades but always as a single strategy. My trade record keeping is such that multi-contract trades are recorded as a single trade with an average entry and average exit price. As such it never occurred to me that that each "leg" of a strategy could be viewed as a strategy by itself.

What's more important is that if I had each scale-out separated out I would be able to post-examine my results and discover which of the targets was most optimized to (1) my system (if trading mechanically) or (2) my style (if trading discretionary).

DT,

What you discuss at the end of your post has been my 'bone of contention' all along with the studies. Here's something I posted at another forum:

"I have spent a fair amount of time studying statistical approaches, and I have come to conclude (my background is not strong enough in formal statistics that I would say I am a true 'expert' in the field by any stretch) that this is not a cut and dried field (as it seems), and the value of any study is dependent on how the study is constructed. If you alter the study you may reach a different conclusion, or if you construct it differently the conclusion may be 'more valid'.

Here's what I mean. Did anyone test various single exit approaches, say ten of them, and rank them by performance. Then take the best three or four and run them simultaneously, my 'blended scale approach', and see how that series compares to the ten individual results? Or try several different combinations of the top four? Perhaps that is exactly what the studies you refer to do, and if I read the specific references I wouldn't have to ask the question (hence I ask what studies, so I can save time by knowing exactly what was done). If this is not what was done, then I don't see how any conclusions, for or against scaling, could be drawn.

It would be amazing to me that under all market conditions over a long period of time one strategy, as the market varies, would outperform a blend of strategies. This is not what we see as far as trading various instruments in a given account, or various strategies in a given account. Yet one 'sub-strategy' (single exit) in a given account will always outperform all other management strategies. It just seems so unlikely. Hence, I suspect it is an anomaly of the structure of the studies."

and:

"I just wanted to add one thing. In the studies I had seen they were comparing a scaling strategy to various single exit strategies. This is really testing the comparative robustness of the strategies, not the scaling vs. single exit concept. I contend, in the cases I saw, that it only showed that the scaling strategy tested was not that good. By taking a set group of strategies, and then blending the top few (say trying various combinations of the top three or four, and running them and when they hit, taking off 33.3% as each of the blended three trigger) and comparing them directly, one could see if a single strategy will always outperform a blend. As I said, perhaps that is exactly what the studies Tim is referring to did, but the ones I looked at didn't.

Also, this idea still says that the particular strategies are good for scaling, and they may well not be. They are single exit strategies, forced to act like scaling strategies to ensure a better statistical comparison. If one works for years on specific scaling strategies, and supposedly comes up with something viable (as I feel I have, for my own 'Trading Plan'), it would be more informative for me if I could compare that strategy directly with various single exit strategies. And the fact that the strategies I use are discretionary at every step only adds to the difficulty. However, I'd want to feel pretty comfortable with the above points before I would draw conclusions."

Just some thoughts, of which I have many. I'm not saying scaling will outperform, or that it is better. My whole point is that it is not as clear as 'they' say, and that the studies themselves many times are not constructed to discover the truth, they are constructed to prove a pre-existing belief, just like most of what I have seen in the academic world of science (after all, to get a grant (I call this the 'grant phenomenon') or do your dissertation, you must state your belief beforehand and how you intend to prove it, not state you are seeking to discover what the truth is, regardless). Hence, I hope people will study this and not take what they hear and read as gospel.

A couple of comments immediately came to mind when reading your post. I'm not sure how valid they are.

You talk about a particular strategy and the desire to be able to compare blended, scaling and single exit. What about isolating just the exit methods and using a random entry and back testing that.

I remember reading one of Woodies followers saying that you could money manage your way to success and use a random entry strategy. I back tested this to the end of the world and not a single back test came out profitable. So I would expect that a random entry strategy would produce a loss but I am thinking that maybe one of the exit methods may prove to be superior over a number of tests. This would be a way to isolate which method produced better results if you took out the strategy or entry method.

The other thought that came to mind was how do you select which entry to match with which exit (when measuring this scale out comparison) when you have a multiple entries at different prices and multiple exits at different prices. If you traded 5 ES contracts and entered longs at 5 different price levels (i.e. scaled in) and then exited at 5 different price levels (i.e. scaled out) then how do you match an entry with an exit when splitting this into single trades? You could do a first in first out (FIFO) approach or a last in first out approach (LIFO) but which is it?

Just some random thoughts generated by you posting.

Good questions. Since everything I do is so discretionary, I have never 'tested' anything I have come up with. I have always been very clear about that, because it couldn't be 'coded', no way no how. I have also not ever done any 'stats' on anything because I feel I can 'tell' by 'feel' how well something works, and to take trades that, as you gather, would each be unique, and try to group them and evaluate them compared to each other (in one way or another, that's what it would amount to), it makes no sense to me.

Now, this approach seems very unscientific to many, and many can't accept it as allowable in their approach at all. To that I say fine. I recently heard a comment from a very rigid technical analyst who said something to the effect of if you can't statistically verify it, it doesn't exist. My immediate thought was 'Even if I'm making money with it?'. The answer to that, I'm sure, would be that sure, you may be making money now, but it won't last, or something of that sort. I can't spend any of my time addressing that approach, so I let those that want to spend time verifying things do what they do, and I'll do things my way.

I know what I do well enough to see what effect a small change here or there has on the methodology. If I'm going to be discretionary,and hence have a methodology that can't be 'coded' or 'back-tested', then I better be able to 'feel' how a change affects the process. If I can accept a discretionary approach to the methodology, then it follows for me that I can accept a discretioary approach to the evaluation. I just can't spend a lot of time trying to break things down and to try to come up with tests to evaluate what seems obvious to me as I 'tweak' the system. But this is my approach...

As for your second point, I don't do much with scaling in, so that isn't an issue, but for statistical purposes, I don't think it matters if you chose the first entry, any given entry, or an average (which is what I would do) because the idea is to look for relative results, not absolute. Even if you chose the last entry and that gave you all net negative results for both methods, it would still be valid, in my opinion, for doing a comparative analysis.

Excellent article Jim, thank you for sharing it !

I did want to contribute a few random thoughts on the topic(s)...

For point of reference, I have used a variety of scaled exit strategies for many years. I have also used single exit strategies (so called All-Out strategies), primarily when using scalping techniques.

I do agree with your astute observation that we cannot predict the future, and thus will not know today what exit strategy will be the ideal one to apply tomorrow. Your analogy and observations on this point reminded me of the volatility contraction phase we experienced following the July 2006 low and subsequent slow steady price uptrend. In our method we had to make a number of adjustments to holding time frames and price targets in response to the phase of diminishing volatility. We had no way to predict this development, but did adapt "on the fly" in response over time. This phase of diminishing volatility tended to flatten out the equity curve. Now we appear to be returning to a more normal volatility phase, and have already begun the process of returning to more traditional parameters.

On the point of viewing each exit point of the scaled exit strategy as a separate trade, doing this does help psychologically, breaking the larger position down into smaller pieces to manage independently.

One of the hardest things I have found to teach new traders is to get their mind around the concept that it really helps things along to capture a big win every once in a while, to participate in that 20 or 30 point move in the ES (preferably with more than 1 contract on). In a practical sense it is very difficult to make much headway on the equity curve for the trader scalping 1 or 2 contracts for 10 or 15 ticks in the YM, especially when your giving back 5 or 10 ticks on every loss. This becomes more problematic when the trader scales out half the position size at 5 ticks. I see this as a form of churning an account over time. This approach of scaling out is my primary contention with teaching the method. New traders tend to reduce position size prematurely, and in so doing significantly lower their probability of long term success. Trading involves risk, and risk equates to position size in my view of things. To make money you have to be able (psychologically) to trade some size, to hold open a meaningful position for a period of time. New traders tend to struggle on this point.

This leads to my question, what is the real reason the trader is scaling out ? If the trader is scaling out to relieve a little of the pressure (fear) of holding the position open, that is not really a very good habit to form. In this case, I propose the likely problem is the trader is not properly funding the position and the trader lacks staying power in the position, giving it the time it needs to run to fruition. If on the other hand, the traders plan is to exit a portion of the position at the next support or resistance price level (ie. market data, not emotion) and the market reaches that objective price target, then I agree with you, it is hard to argue conclusively in the real world that executing the pre-defined objective strategy of taking the partial profit is a bad idea. One time the market will blow through the target and the trader should have (in hindsight) held the full position, and the next day with equal probability the market will hold the level and reverse proving the scale-out was the best strategy to use on that day.

In position trading I find it helpful to use a scale-in approach, easing slowly into a new position over time as the trade idea proves profitable. This brings up a point which you did not address, that being, the concept of adding to winners. Thus, the idea being, as the trade performs as hoped (becomes profitable), add more to the position. I refer to this as rewarding my winners. To be fair in the discussion, one should really consider this approach, because it does make a significant difference in the overall profitability of winning positions. So if we compare an add to winners approach to a scale out of winners approach, across say 100 trades, how does this influence the equity curve ?

With respect to the equity curve, it has been my experience and is my personal preference to minimize draw down, leading to my objective in trading to create an equity curve that forms more of a stair step structure made up of a flat section followed by a step up, followed by a flat then another step and so on. The reward the winners and all-out exit strategies are employed to maximize the size of the "step-up" to the next level on the curve. Thus in a practical sense I am not concerned about the slope of the curve, but rather focus on the step size.

pt_emini: I know that you first put me on to the Phantom of the Pits book and what you are saying in the last couple of paras above is the same as Phantom's Rule Two.

Upon further consideration of the concepts being developed in this thread, I want to expand the discussion on one point of consideration with respect to the scaling out profit taking strategy.

The basic concept is that the best or optimal profit taking strategy is dependent upon the trading style being used. In trading style I am referring to the traders time frame of reference and the duration the trade is held open.

Let's take a look at some of the typical or traditional trading styles with respect to hold time, risk:reward, and exit strategy:

1. Scalping: High frequency short hold duration trading. Scalpers trade off the tape (Time and Sales) or using the fastest 1 to 5 minute time frame charts. Hold time is measured in minutes, and profits are measured in ticks. Given the extremely tight risk:reward profile (1:1 to 1:2) combined with the associated high commission overhead a high win probability is required to ensure profitability. The ideal entry & exit strategy is to go "all in" and "all out" with size. Given the extremely short hold time scaling in or scaling out is not an ideal or practical strategy for scalping.

2. Intra-day trend/position: Positions are held open for a period of hours with a profit measured in points, however positions are usually closed within the trading session they were initiated and are rarely held overnight. Risk:Reward is expanded out to 3:1 to 4:1 and as high as 10:1. Frequency of trading is reduced with 1 to 4 trades per day initiated, subsequently reducing commission costs. These two factors combine to reduce the associated win probability needed to sustain profitability. Given the longer hold time for a larger final profit target, the scale-out profit taking strategy is applicable and practical. All-in entry and scale-out exit is a very typical position management strategy employed in this trading style. Since this is a trend following style an alternative to the scale-out strategy is to employ a trailing stop on the full position size. The trailing stop technique enables the trader to catch the full profit potential available from so called "trend days". We have all experienced the frustration associated with taking a quick 4 tick profit on what develops into a 20 point trend day.

3. Swing: Hold time is measured in days, with 1 to 5 day hold time typical. Since the swing trader is focusing on a this longer hold time and corresponding wider stop loss levels and larger profit target(s), entry price precision is less essential, thus the scale-in strategy begins to make sense. The scale-in strategy permits the swing trader to slowly ease into a larger position by developing an average entry price over a period of 1 to 2 days. The scale-out exit strategy can be employed, however based on my experience with how the markets typically work on this time frame, I personally am not convinced scale-out is the optimal strategy for swing trading, preferring instead the scale-in entry and all-out exit strategy.

4. Position: Hold time is measured in weeks or even months, thus this is a longer term trend trading style. In this style the scale-in entry and scale-out exit strategy combination is an acceptable choice. In this way the trader is relieved from the need for a perfect entry price obtaining instead a reasonable average entry price. The trader gives up some profit potential on the exit but given the extended hold time is given the opportunity to recycle working capital as the trade develops. As an alternative, the position trader can avoid taking profits prematurely (leaving money on the table) by using a trailing stop on the full position. Thus it is my proposition the optimal strategy for position (trend) trading may be scale-in entry and trailing stop exit.

Interesting thoughts on scaling types related to trading style and objective. Thanks.