NuclearPhynance Archive

Hi,



This is my first (proper) post here. So hi to everyone.



I would greatly appreciate if someone could point me to any research, papers, experiences in the area of scaling into and out of positions. Here I refer to something like spot FX, shares, or Futures (i.e. as opposed to derivatives hedge or augmenting an outright position – long or short the underlying instrument – with derivatives).



Scaling out is not recommended by Van Tharp (Tharp, Van K (2007) Trade your way to financial freedom (2nd Edition) – McGraw Hill Professional), as he correctly shows that you are holding all the possible risk at the start of the trade then not “letting your winners run” as you are taking off position size as the position runs in your direction. Again, Van Tharp’s views are not supported by empirical or researched evidence. Whilst I would recommend developing an understanding of Tharp’s position what about the other side the coin? Doesn’t it depended on the pay-off profile of the strategy a method of scaling out of a position may be preferable?



My rough and ready research into the topic is this:



1. Using scaling out for trend following strategy takes up the percentage of profitable trades (this is good thing for standard trend following systems usually aren’t good with 30 – 40% winners) but takes down the profit per trade

2. Scaling into trend following strategies takes the % of profitable trades down



And what about scaling stops? For example, if you had a stop that is 3 * ATR (or some other measure of volatility GARCH), what about if we took 1/3 of the position off at 1 ATR to see if the position turned around. If the position moved against us 2 * ATR we could take off another 1/3. If the position moved 3 ATR then we take all the remaining position off. My thinking goes that it would allow for positions that move against us initially then take off in the desired direction. I call this idea graduated stops. Though, it is exactly that, just an idea (it might be a rubbish idea). I haven’t done any significant testing on it.



I’ve seen and discussed on another forum a guy suggesting the use of scaling out, but trading a larger position size. But doesn’t that increase the risk of ruins Ralph Vince quotes in his book (The Mathematics of Money Management - 1992):



“if you play a game with unlimited liability, you will go broke with a probability that approaches certainty as the length of the game approaches infinity.”



I’ve included here a calculation sheet that models some basic trades (+ a screen shot). That is:



1. Assume that $10,000 funded account

2. We risk 1% of that account per trade

3. Cell D5 is the number of trades

4. The percentage of winners is in column G (with the loser percentage being 1-column G)

5. Row 1 is the actual pay-off in dollars for each trade

6. Row 2 is the percentage increase in the winner pay-off above the stop loss pay-off

7. Below is a surface plot of the outcomes



[img]/User%20Files/8715/Screen%20shot%20-%20scaling%20calucation%20spreadsheet%20v2.JPG[/img]



If we increase our position size then we increase our probably of ruin (surely?). But if we scale out earlier in a trade don’t we also increase our probably of a winner as we are scaling out when the trade moves with us? From the spreadsheet we can see that if we increase our percentage of winners than we increase our probably of landing in a profitable zone. Which could lead one to assume if by increasing our percentage of winners, this is good. But also, as per Van Tharp’s point, don’t we reduce our pay-off per winner if we scale out? Doesn’t that drag us back towards the left hand side of the table in the spreadsheet as we are reducing the size of the difference between the winners and losers.



Any help, thoughts or comments would be appreciated.



Thanks,



drolles



[url=/User%20Files/8715/Scaling%20calcuations.xls]Attached File: Scaling calcuations.xls[/url]

There are guys here who did alot more work than I did on this...



But based on my experience in sizing/scaling in and out, it ultimately depends on the underlying process and how you attempt to trade it. Then it depends on your own risk tolerance to different sizing strategy. As you said, if you scale out too early, sometimes you lose out on profits per trade but increase hit ratio.



At the end of the day it is more of which trading statistic matters to you the most and what you are willing to give up to for it. For instance, if you think you have a static non-noisy edge and all you care about is growing your money, then you would choose KC. However that is not very practical for trend following and maybe some form of adjusted kelly/optimal f is what you would use.



The problem gets a lot more complicated when you add different types of underlying process and strategy to your portfolio as one statistic decrease/increase in the strategy, it can change the simulated portfolio characteristics when pieced together.



Just for anecdotal evidence on something I worked on 5 years ago, if you define a mean reverting process, and as the process moves forward in time, I think Michael Boguslavsky had a paper about its optimal sizing. However, based on the risk tolerance of the fund, simulated return on capital of the strategy, actual size constaints, transaction cost, expected time to convergence, and other parameters I can't quite remember at the moment, we needed to size with a scale in and scale out function that is away from optimal. In the end, we were willing to sacrifice terminal payoff to increase hit ratio/profitably per trade and decrease daily risk profiles.



Hope that makes sense.

what he said^ .

When you do the homework a lot of the turtles and van tharp scaling in stuff appears to be just dogma that has perpetuated well out of proportion to it's usefulness. Who woulda thunk it, coming from a salesman who uses "Dr" in his title?



I have spent an undue amount of time on this issue in the past, here is my best wisdom for you:



scaling in, in whatever form, typically boils down to increasing your variance and skew of trade results. In the right application, it may also bump up your mean return, sometimes significantly if the market/strategy is in a phase of high autocorrelation of either returns and/or volatility.



The key thing i have found is that in all but a vanishingly small number of cases, by scaling in, you raise variance by more than you raise average return, in both backtests and real trading. So your risk adjusted return is usually damaged.

The dark flipside to this, is that scaling out often appears to improve risk.adj.ret. in bactest....not so much in real trading(though it is still measurably there). Again I say this with over a decade of practical efforts trying to squeeze some mileage out of this in typical directional trading.



Historically, think on this: lets say a lot of people try and make money with trendfollowing and scaling in over the last 20 years. They do this, because in the prior 20 years markets were apparently less efficient(displayed huge exploitable autocorrelations) and this approach worked really well for those that stumbled on it. Most of our group fail, but some succeed and make fanatstic wealth/reputations.

My conjecture is that this is what you expect from a random trial of a process with extremely high variance but perhaps nonexistant average returns. i.e. survivorship bias of the most highprofile successful few perpetuates the myth of scaling in as a good all round approach.

In reality it may reduce your expected returns to negative. So you have a choice: play the lottery for glory or bankruptcy*, or grind out a more moderate living.



* quite a few of the scaling in characters went bust a few times before succceeding with this approach. Hey if you can rely on acquiring startup capital repeatedly, then this approach might not be so bad heheh.



When you look at the ongoing performance of the funds/superstar trader scaling-in trendfollowers of the last 20 years, you start to see the following disturbing signs: high variance and drawdowns, only moderate returns. I would hypothesize that the fact that they are still around is most likely due to them adopting a more neutral approach to positioning once their algorithm designers and risk managers got together with the marketing brains and decided that perhaps good things never last forever(w.r.t. scaling in) and that now they had enough AuM, might as well tone it down and enjoy the fee income.



bottom line: it is a lot of extra complication and heartache, and it is very hard to really add value with bog-standard scaling in/out in the long-run. the short-run, as ever with markets, may persist for a little longer than one might expect.



EDIT:after all that negative rant, I ought to disclose that I personally do occasionally still increase positions after positive move from initial entry. However I do it very quickly, and usually as a discretionary response to some instinctive feeling that I have a bunch of participants wrongsided and cornered and screaming for the exit i.e. Suddenly out of the ether I get a big leap in confidence of a short-term improvement in the price.Furthermore, I exit the additional size much more rapidly than the core position, usually trying to offer out liquidity to those in pain. So really what i am doing is adding a short-term scalp to my typical directional trade, where it is most likely to pay-off. whilst the outcome of the individual bonus scalp is correlated to that actual core trade, the bonus scalp strat as a whole may augment my core strategy in a less correlated manner. This leads me to my final point:



Best thing is to view the additional scaling in entries/exits as an entirely separate system. Evaluate it based on how they work together synergistically. Usually you can do better at creating less correlated enrty/exits starting from scratch. I just do the bonus scalp thing because it is more convenient than running multiple disparate strategies mentally, but if you are a algo trader there is no reason to not try something better.

"I’ve seen and discussed on another forum a guy suggesting the use of scaling out, but trading a larger position size. But doesn’t that increase the risk of ruin..."



IMHO No, because the prob. of ruin decreases expontially with increasing win rate. As a matter of fact, any strategy with win rate < 50% is guaranteed to blow up if there is no terminal equity growth target.



Another way to look at this problem in within the context of %Kelly optimal bet sizing. Just to be sure, not for actually doing %Kelly sizing but for analyzing the problem. See this for example: http://tinyurl.com/6saxntz. I do not agree with everything in there but it is a good starting point. I agree that win rate is everything in trading. Very high win rate makes path dependence irrelevant.



Trend following funds with win rate < 50% are doomed to get ruined if they trade long enough unless they are either lucky in terms of future path or extremely conservative in terms of risk percent.

Thank you all for your posts; all of them thought provoking and helpful.



To pick out a couple of themes that you have all raised:



Increasing win rate – reducing probability of ruin

Intraday Bill’s point on increasing win rate exponentially reducing your probability of ruin. Interesting comment. Of course, this makes sense intuitively, but I wonder is there any work illustrating that? I suspect that there is an assumption in there that is around increasing the position size to maintain the profitability per trade? As discussed below in my experience if you start to scaling out of positions you drop the profitability per trade.



The profitability per trade is important here. As we can see from the sheet it also is a key factor in the risk of ruin calculation of the strategy. Most of my modelling is of trend following strategies, I’ve done only a little bit of work with mean revision models. Again, in my experience if you scale out you drop the profitability per trade as you cut some of the long tails off (which are the meaty bits of trend / profit engines of trend following strategies).



My vision had always been that ideally you could scale-in after that start of the trade then scale-out before the end of the trade. Thereby, ideally missing the larger loss as you didn’t have as large position on if the trade moves against you at the start of the trade, given the unknowable outcome of the trade at its starting point. But my modelling to date has not shown a method of how to do that. I’ve attempted to identify those trades that are moving in the form of likely winners by using Stridsman’s trade profile method from his book – Trading systems that work [Stridsman, Thomas (2000) Trading systems that work – McGraw Hill Professional]. From my backtesting, it is a trading system that doesn’t work. J



Dogma

Yes, I love it. Exactly what I’ve found. I’ve not found any robust research on this topic at all; only views on how it might be done and statements of opinion. The only other good reference on how it might be done that I’ve found is in Kaufman’s book [Kaufman, Perry (2004) New Trading Systems and Methods (4th Edition) – John Wiley & Sons]. But again there isn’t any empirical research on the results of the scaling in/out approaches.



Reference to Michael Boguslavsky – gamma profiles of trading strategies

Tabris’s reference to the above is a good one. I’ve Googled Michael and come up with the following reference: Mike Boguslavsky (follow the link and take a look at “Optimal Trading Rules”) This is really interesting descriptions of the behaviour of different types of trading models – e.g. trend following versus mean revision and that nature of their payoff profiles. It is a good reference that helps explain the difference between trend following strategies and mean revision strategies. Unfortunately, nothing that provides us with what we are missing, a solid body of work from which to draw some robust conclusions about scaling.



Position sizing approach and scaling

Do we think we can align and link the method used for sizing positions from the outset and scaling position sizing algorithm? For instance, if we are using a Kelly approach, then if we try to apply scaling we are seriously changing the nature of the trade outcomes to the portfolio. Is this why Rabid and others say that we should be treading scaled trades as separate trades? However, if we using a fixed fractional approach, are we changing the profile of the payoff of the trade to the portfolio as much? IMHO I don’t think so if you make some basic assumptions. One key assumption would be that you risk that same percentage of the account on the scale trade as you do with a standard entry trade. You will obviously need to use stop losses appropriately (moving them to the previous entry) to assume the same risk. Just a thought / idea – again it might be rubbish.



A little anecdote

I was once looking a mid-term trend following strategy for Forex. A combined idea I got from a colleague and a book [Katz, Jeffrey and McCormick, Donna (2000) The encyclopaedia of trading strategies – McGraw-Hill]. I made a mistake in my assumptions and code; I accidently had coded the scale in points a lower prices than the initial entry (on the long side and higher on the short side). I then “fixed” the strategy, it performed worse than previously with the logic error. Any retail literature / courses you do will tell you not to “average down” i.e. add to a losing position. Of course my one off strategy analysis is hardly a robust test, but food for thought.



Runs theory

What we haven’t linked in at all and make me think after reading the replies was runs theory. The clustering / auto correlation of payoffs, particularly in standard trend following systems, these systems can take a number of losses in a row. If we are trading bigger position size from the outset surely we are increasing the risk of ruin?



I’ll have a further think and will do some further testing.



Cheers,



drolles