Standard Deviation (SD) gets a bad rep because of people like Taleb and etc. But I'm starting to think SD makes a huge amount of sense when it comes to stop placement. Usually when you place a stop, you want to put it away from the "noise", so you only get hit if the price does something that is usual and it goes quite a bit against you. So lets say you are trading Crude Oil and you have a bullish bias. You can put a stop at the extreme of 2 SDs away from the current price (and the SD is the 60 day SD). So your stop will be in a area that in the last 60 days, ~95% of the time it would not have been hit. If prices starts to come down and it reaches that 2 SD limit band and you get stopped out, the SD worked as expected. It alerted you about an unusual price movement and you were able to get out. Does that mean that one levers up and think "There is a 95% chance I won't be stopped out", no, that would be misusing the SD. Its not telling you that at all, its telling you that in the last 60 days, 95% of the time, your stop would not have been hit. Using SD based stops tell you nothing about what kind of leverage one should use, position sizing, etc. The very things that anti-SD fundamentalists rage against Why would a SD based stop be better than say an ATR based stop or some other volatility metric? Because SD are a lot more intuitive than ATRs, at least for people with finance backgrounds. ATRs or other vol metrics don't feel as natural. When something trades at more than 2 SDs than your entry, you know something is up. You know that in the past, that would only have happened 5% of the time, so therefore, something usual is going on. That makes it a lot easier for the trader to respect his stop, get out and live to fight another day. But if you get hit on a different type of stop, you might not follow the stop, or even if you do, you might reenter the trade not much after because you still feel that you are right etc. Its hard for a trader to still feel that he is right if something moved 2 6-month SDs against him Thoughts?
1. Stop placement is an art, and usually arbitrary. 2. I prefer "technical" stops. That is, "stop when some parameter (and perhaps its overshoot/buffer) you were playing to hold, breaks."
There really aren't many choices for stop placement. Fixed stops like 20 pips or whatever. You see these a lot on Forex message boards. Completely arbitrary. Support/resistance levels. Justifiable based on chart patterns but somewhat subjective. Volatility stops based on statistical or implied volatility. The most objective at least mathematically. Van Tharp was a big proponent of volatility stops and used them to size positions. Wider stops have smaller size, narrower stops have larger size. The goal is to have a fixed risk (e.g. 1% of capital) so if the stop is hit you lose a predictable fraction of your capital.
This is essentially the idea of Bollinger bands; moving average and standard deviations. Read Mandelbrot's book on fractals in economics. We all know this in practice, but there is plenty of theory describing how Standard Deviations, Gaussian curves, are insufficient for the markets. "Working most of the time" is not very comforting when catastrophic loss happens. Kurtosis, fat tails, are not to be ignored on any time frame. 2 Sigma and Up moves happen way more frequently than described in a Standard Deviation.
I instictively disagree with most of what you say, there, but all that shows is that my instincts are very different from yours, and that doesn't matter at all to me, to you, or to anyone else. What matters is evidence. You should be using an SD-based stop-loss if you have evidence that it works better, overall, with your own trading, than an ATR-based stop-loss (and other kinds), and vice-versa. For me, positioning the stop-loss just above/below the most recently formed swing-high/low happens to work better, overall, than either of the above (and I do have evidence of that), so that's what I normally use. Yes, I do agree with that (and there are some reasons for that discussed by Taleb, but I can't remember at the moment in which book/paper).
Stops should be based on risk. They are a way of managing risk. If you're basing them on technical levels then they're not a stop, they're something else. If I used explicit stops I'd use SD based stops. You have to have some model of risk. A model is a simplified form of the world. You can read Taleb and still think SD is the best one parameter risk model that exists. I don't really understand ATR, but I guess given some assumptions about autocorrelations we can translate between ATR and SD measures. So in practice it shouldn't matter too much but I agree for someone with some quant finance training there is no reason to use ATR. Trivial point: The nice intutition of 2 SD meaning a 5% tail event only works if you're measuring SD at the same frequency as your expected holding period. If your expected holding period is one week then it should be 2 SD of weekly returns (or square root 5 times daily return SD). GAT
I dont understand it either. I know how its calculated but it has no meaning to me. It doesn't 'feel' like anything that matters. I'm aware that you can set it to match the SD but still, psychologically it doesn't speak to me. I bet a lot of people feel that way.
I guess ATR was supposed to be more accurate because it uses high, low and close. Volatility just uses the close. I'd much rather use familiar formulas from my old stats book like SD.
I use both, and find both speak intuitively. FWIW, if you put up a graph of volatility, you can tune an ATR to match it. It is, for me, an instructive process.
I guess the nice thing about SD is that its an 'auto backtest'. It immediately tells how it fits the past data, which gives you confidence that the stop is wide enough to stay out of the noise. With ATR and other metrics, you got to do the backtest yourself. But even if you do it, if volatility changes, then you got to run the backtest again, otherwise you will have no confidence in it and might end up secong guessing it