I don't think stop losses are necessarily useless.
It's not hard to imagine a process in which using them adds value (of course the problem is finding something that can be proxied by such process...)
This is a far cry from saying that they "the" way to manage risk (or even "a" way ) - actually it's totally a different story for me.
And for how much consideration I have of you goldorak - no, the example with CHF proves only that they don't always work the way you would think, and that you have to be careful (wohoo ). It doesn't tell you anything about any possible value they add on average in specific cases.
> It's not hard to imagine a process in which using them adds value (of course the problem is finding something that can be proxied by such process...)
Well, you have named it. Easy in theory, and in my opinion worthless in practice.
The concept is very seducing in itself: it looks like you might be able to cut-off tails of your distributions. However they always omit to mention that you have an uncertainty on the stop-loss itself, a P&L distribution of its own. This uncertainty, if you apply your stop-loss long enough, will unfortunately be very much like the one you observe for your trading P&L: fat tails, extremes, jumps (call it whatever you want), non-stationarity, etc.
If you call PL(1) the base distribution of your returns, PL(2) its expected distribution with a strict stop loss and PL(3) the stop-loss distribution, in the long run you will recognize that PL(2)+PL(3) is not all that different than PL(1). However occurrences of stop-losses being less frequent than your observed returns, you will have a false sense of safety.
Then of course if the market follows a certain stochastic process, stop-loss will be great tool, but don't you think that your trading P&L will also be great if you know ex-ante the market follows that particular stochastic process?
If you are not living on the edge you are taking up too much space.
i don't think that stop loss or take profit in any way mean increase safety or reduce risk, it's just market order executed by broker,
if your algorithm need fill market order when some assertion will be met on quotes feed then using stoploss may have a sense because you don't need actually send it and lose time on it,
but then you also don't have possibility of not sending it if your assertion is more complicated
also hard to argue about something when you don't have specific case described guys,
but it clearly have some usability because using it we have possibility of trigger some well known action by third party/computer system/exchange/broker that will able to help our met desired goals in specific cases (avoid sending lagged closing market order over network)
btw. op don't say anything about risk management or increasing safety using that functionality.
Stop loss for a single instrument it pretty much useless and provides a false sense of security. For broad indices with 500 or more constituents it may make some sense. The broader, the more smooth the index will move in any direction in the case of a black swan event.
For instruments look to options, it will come with an attached price tag compared to the free stop loss. An Protective Collar will save you, as long as the clearing house doesn't go belly up.
Goldorak- Forget all the bulls* about cutting your losses short, cutting off a tail of the distrib, managing risk, etc. Of course if you consider the technical analysis "literature" on SL, I can see why you are sceptical.
The idea for me is simply that there might be processes in play for which using certain orders in a certain way makes sense.
At the end of the day, a SL (or any part of a trading model) is just some feature (in a ML sense) that can be good or bad depending on what's applied to and how.
A part for cases where based on your knowledge you can exclude that something could EVER add value, I don't see how can you make such a generic claim about something so generic such as "Stop Loss" (heck, Jim Simons even tested Moon phases! not that this per se shows anything)
"Then of course if the market follows a certain stochastic process, stop-loss will be great tool, but don't you think that your trading P&L will also be great if you know ex-ante the market follows that particular stochastic process?"
Not really, it depends on the process in play no? It might well be a process that can only be (slightly or massively, it depends) exploited using certain features (e.g. SL)
E.g. svisstack example on the possibility to trigger other orders
> Not really, it depends on the process in play no? It might well be a process that can only be (slightly or massively, it depends) exploited using certain features (e.g. SL)
I was obviously referring to a process whose parameters are known with sufficient precision. It is needed for your stop loss to work, and as a consequence your trading will be almost perfect.
If you are not living on the edge you are taking up too much space.
> The ONE SINGLE WAY to limit a loss or a drawdown (call it whatever you want as in the end it just depends on your time frame) to x% is to invest x% of your capital.
Problem is if you manage a lot of money this is not going to work. If there is a crash where you have a margin call and say you owe the broker 20M vs. 10M you had in the account (assuming use of margin and leverage), the broker is going to come after you.
It's going to happen anyway, if you had a Stop Loss or not.
I am afraid there isn't really a solution.
I actually think the use of Stops in combination with certain models is useful, in any case, depending on the strategy it may not be useful at all. Mutual fund managers usually don't have stops, for example, as they manage too large sizes for stops to work and in the end they do not sell a product that promises to go up when the market tanks. If the market tanks, a LONG mutual fund tanks in synch, is that simple - you don't need a Stop Loss in that case.
"I was obviously referring to a process whose parameters are known with sufficient precision. It is needed for your stop loss to work, and as a consequence your trading will be almost perfect."
Well, I was thinking about non-deterministic processes, even if you know exactly all the parameters, there is still some indetermination. So for example: if X happens, then there is a positive expectancy in betting price will move in this or that fashion.
I am assuming your argument was ~
you know exactly the process -> you make gazillions -> you can hardly know exactly the process.
Which probably is just a different way to say the same thing, but I don't see how this says anything about SL or any other feature you might be using - you can still find that something adds value even if you don't know exactly the process in play.
I tend to agree with this:
I actually think the use of Stops in combination with certain models is useful, in any case, depending on the strategy it may not be useful at all.
and I would stop calling them "stop losses" in favour of "stop orders"
Pay attention, I am not talking about reality but about a perfect ideal and sci-fi context.
Take a plain brownian motion with drift. The process is stochastic, but if you know precisely its parameters, average and standard deviation), you can set an optimal stop loss based on your risk attitude. But knowing precisely that the average is large enough for your needs, I guess you do not really need to worry about stop losses. Santa Claus
Take an example. My investment has for sure an average daily 50bps return with a 50bps, and this cannot change in my ideal world. However today I just experienced a -500 bps return. Shall I cut my losses? Of course not as I know my process average return is still 50 bps per day.
It is in the reality, when I cannot be sure of the parameters that I might consider using stop losses, trying to get out of an investment whose parameters (especially its mu) might have changed. But then the problem is exactly the same as trying to identify a trading strategy of its own.
To make a long story short, one could summarize it all with: "stop losses are a strategy, with the same flaws, and do not need to be considered/studied per se."
But there is no way stop losses can be considered out of a strategy development context, as an add-on tool, one would use to "enhance" or "risk manage" a strategy.
If you are not living on the edge you are taking up too much space.
). It doesn't tell you anything about any possible value they add on average in specific cases.
