Volatility Adjusted Stops

Last night I was reading the book 'Volatility Trading', in the second chapter it was enumerating the different available ways of measuring volatility. The very first one was simply standard deviation, no real surprise there, however this got me thinking...Whenever people talk of volatility adjusted stops they almost always link it to a multiple of ATR (as do I). I thought, would it not make more sense to measure volatility via standard deviation (not the weird MT4 deviation indicator, but a proper calculation of the standard deviation of bar sizes). One could then set stops at say mean + 2 * standard deviations, this way one has an intuitive sense of what the measure means (i.e. 85% of previous bar sizes will not exceed this measure), as opposed to ATR * some random number I have pulled out of my ass.
Am I missing something?

I think what you're looking for is a bollinger band... it does exactly what you said, create a band which height is a multiple of the standard deviation of bar size over the moving average period.

Close, but not quite...here is a demo calculation I whipped up. As you can see the dynamics are quite different to bollinger bands but similar to ATR.
Taking an actual deviation of daily range does not suffer from the 'blow out' dynamics of bollinger bands, but it can be somewhat discontinuous.
Edit: I uploaded a clearer pic.

Hello Craig

I agree with your statement. ATR is the simple average of the range of all previous bars within the period set to calculate.

But the entire question evolves around the probability of a bar-range being larger than all other previous bars. A median does a better job in measuring the probability in this case. The median means: 50% of numbers measured are bigger and 50% of numbers are smaller than the median.

There are of course other parameters in the field of statistics. One can also calculate a figure that represents lets say: 30% of all ranges are bigger and 70% of all ranges are smaller. The meaning would be: there is a 70% probability that the present range is smaller than the number given in this 30/70 median.

I have tried for long to find an indicator similar to ATR that does this - especially for daily ranges. Currently, I calculate on a excel spreadsheet the daily range that is likely to develop on a present day.

Aren't you basically talking about keltner channels vs bollinger bands? That would be the easy way to compare the exit points visually.

Very good question however. I have to do some studies myself too.

This is interesting link related to this (if this works longer term, it's google books page) especially the correlation part
http://books.google.com/books?id=MK1...age&q=&f=false

I had a look at 'keltner channel', it's not quiet the same as there is no MA involved in the pic above. The calculation is simply mean + 2 * standard dev of the (high - low) of the last n bars.

Good link, however I am not taking 'close to close' volatility, I am taking 'high to low' which made more intuitive sense to me. I'm sure which ever way it is done will have some bias, but if one could dispense with a degree of freedom in a system (i.e. the ATR multiplier) then one would think this is a good thing.

Is it a band you want? The problem is, if you want to add standard deviation to current bar high/low to make a band, the price will NEVER touch the band unless the standard deviation becomes 0. That's why a moving average is used... it lags behind to it is possible for the price to go ahead of the average and touch the average + STD.

Craig this is very good idea and worth testing. Sadly my PC has a problem this moment so I can't test it immediately. But I am not sure you can completely eliminate the degree of freedom, I believe you might still want to have a stddev multiplier, especially since prices are not exactly normally distributed.

Correct, I guess what meant was that instead of saying 'well 2 * ATR looks good', one could say 'only 15% of bars will exceed 2 * StdDev', which while we are still making a choice (hence a degree of freedom) at least it has a mathematical basis beyond curve fitting. Also you are correct that prices have 'fat tails', but I'm just looking for rules of thumb.

we need an indicator for it to visually check it first at least.

Here is the one I cannibalized, maybe somebody can double check it as I'm not an mql expert.

yes well I did canibalize one too - here is the original based on ATR and the other based on StdDev, they are about the same. Note that I use different calculation for stddev case, half (mean + stddev) each side of open, so it comes a bit tighter than the ATR one.

This is the actual value, not the trailing stop BTW.

I expect they would be somewhat the same given the original picture I posted, my point was that perhaps it is a more intuitive way to parameterize ones stop sizes.

Hmm I don't know. To me they just seem scaled version of each other. In the end of the day I would leave it to the optimiser to find the best distance, and either of the 2 choices would do. If there was some nonlinear relationship between the 2 then perhaps it would be more useful.

That is exactly what I am trying to avoid.

Here is my version from last night, it's written from scratch (I had my own bollinger bands implementation so I just took the stdev calculation there). They definitely are not the same, there are some clear differences.

Weird, the calculations look much the same, yet there are obvious differences.

One thing might be that I always ignore sunday bars.

The same differences occur on smaller time-frames, anyway, I'll be coding up the C++ version plus unit tests later today so I'll soon know.