"Scaling out" is inferior behavior
- Thread starter Buy1Sell2
- Start date
Ok,so its clear you have never backtested and you are entitled to your opinion,but without somesort of analysis it is just your gu talking..
You would be better served by saying scaling out may outperform,but on average it offers no advantage to full liquidation...
It is pure fallacy to say that the system that does not scale out will always outperform the one that does.
I agree with this assertion, but probably not for the same reason you do.
If you are using game theoretic computations where you put a trade on, and you have targets and risk as your parameters, what you are saying is correct. This is especially true in the context that these are the parameters you tested with, and therefore this is the probability distribution that you are using to get your expectancy.
However, that does not mean that there aren't superior forms of trading. For example, if your algorithm allowed you to dynamically run [game] theoretical computations on each tick, now everything changes.
A similar problem is the following. Imagine that you had three choices to limit risk on a trade you just entered: create a stop loss, delta-hedge a position, or dynamically delta-hedge a position. You can show mathematically that most of the time, the dynamic delta-hedge is superior to either stop loss or the static delta-hedge. But you are only able to do a dynamic delta-hedge if you had the equations that showed you how to do it at each delta-time step.
What I am trying to say is, you are doing good thinking, but be aware that the greatest amount of sophisticaion (as measured by the ratio of risk/reward) is almost always to be able to update your strategy dynamically, because the market is doing the same thing. Setting targets and taking out a position all at once is almost certainly a special case of scaling out as the dynamics of the markets evolve.
Notice that this discussion comes down to knowledge of the geometry of the manifold, as opposed to the topology of the manifold. Only God can see the topology of a manifold at once (the forrest), the rest of us traders have to deal with little pertubations from local coordinates and at infinitesimal time step dx from now, the trees. That means our decisions have to be updated as we ride the manifold locally. If it weren't for the B/A spread and transaction costs, it would not matter that our strategy is being run on local properties of the manifold. We would do as well as someone that could see the forrest.
For these reasons, IMO, trading is a game for traders that can adjust their positions in "realtime" with close to zero costs. The rest of us are gambling with "stop losses", albeit with maybe a tiny statistical edge.
nitro
If you are using game theoretic computations where you put a trade on, and you have targets and risk as your parameters, what you are saying is correct. This is especially true in the context that these are the parameters you tested with, and therefore this is the probability distribution that you are using to get your expectancy.
However, that does not mean that there aren't superior forms of trading. For example, if your algorithm allowed you to dynamically run [game] theoretical computations on each tick, now everything changes.
A similar problem is the following. Imagine that you had three choices to limit risk on a trade you just entered: create a stop loss, delta-hedge a position, or dynamically delta-hedge a position. You can show mathematically that most of the time, the dynamic delta-hedge is superior to either stop loss or the static delta-hedge. But you are only able to do a dynamic delta-hedge if you had the equations that showed you how to do it at each delta-time step.
What I am trying to say is, you are doing good thinking, but be aware that the greatest amount of sophisticaion (as measured by the ratio of risk/reward) is almost always to be able to update your strategy dynamically, because the market is doing the same thing. Setting targets and taking out a position all at once is almost certainly a special case of scaling out as the dynamics of the markets evolve.
Notice that this discussion comes down to knowledge of the geometry of the manifold, as opposed to the topology of the manifold. Only God can see the topology of a manifold at once (the forrest), the rest of us traders have to deal with little pertubations from local coordinates and at infinitesimal time step dx from now, the trees. That means our decisions have to be updated as we ride the manifold locally. If it weren't for the B/A spread and transaction costs, it would not matter that our strategy is being run on local properties of the manifold. We would do as well as someone that could see the forrest.
For these reasons, IMO, trading is a game for traders that can adjust their positions in "realtime" with close to zero costs. The rest of us are gambling with "stop losses", albeit with maybe a tiny statistical edge.
nitro
You guys are talking way over my simple head. The only thing I am fairly certain of is that when you are dealing with uncertainty (i.e., market price movement), all-or-nothing behavior is unnecessarily risky and, therefore, probably suboptimal in the longer run. I imagine it always looks better on historical data.
Yes. I discussed this issue with Mark some pages back. Each time you recalculate and change the parameters, then those are without question new trades. I personally don't believe in that kind of recalculation and instead use trailing stops to change my trade dynamically. That way, I can have my full poisition reaping the full reward. I strongly believe that that is the proper way, but it is not the subject of this particular thread.