Traders' Roundtable

Aside from total, group, trade, etc risk, does anybody employ a method to reduce positions if volatility of the portfolio simply gets "too high". I'm thinking this could happen if positions become too correlated.

I'm exploring this right now using standard deviation of the daily returns as the volatility measure...

The idea appears in numerous books and articles. Here are some representative Google results:

#1: Winton's investment technique consists of trading a portfolio of around 60 contracts on major commodity exchanges and forward markets worldwide, employing a totally computerized, technical, and broadly trend-following trading system developed by its principals. This system tracks the daily price movements from these markets around the world, and carries out certain computations to determine each day how long or short the portfolio should be to maximize profit within a certain range of risk. If rising prices are anticipated, a long position will be established; a short position will be established if prices are expected to fall.

#2: According to Asness and Berger, quant managers often target constant volatility not constant dollar exposure. As a result, they lever up in times of low market volatility and lever down in times of higher volatility. The problem in August 2007, they say, was that many quant managers were using higher leverage than usual at that point in the cycle.

I wonder if standard deviation is an appropriate measure...

Run a bake-off; code up several different volatility measures all within the same Risk Manager blok, and have a Selector parameter that chooses among them. Then simulate using Step All Values and see which one you like most. Maybe it'll be a tie, you like choices B and D equally well. You could calculate your sizes & resizes twice: once with B, again with D, and take the average. Or just have a two system suite: sys_with_B, sys_with_D, and give each of them a 50% allocation.

Another concern that I'm dealing with is that the rules I'm playing with dont come into effect very often. Looking at my chart here

viewtopic.php?t=7386&highlight=

The std dev is only spiking in more recent history. Say i test a function that says peel off x% of positions if std dev > y%, and it only comes into effect 5 times in the test, thats not very statistically sound.

On the other hand, I'm only counting on this function to reduce volatility/risk during extremes, so maybe its ok that it only happens a few times...

Slightly off topic, but I know of a few discretionary traders who actually cut all their positions back - even if they are longer term holds - when their PL gets excessive. It works for them even if just as a fudge.

I also saw another systematic money manager in the UK that seemed to run a portfolio whereby they effectively had a constant portfolio of positions that showed the maximum exposure per instrument based on historical and recent volatility - then the model ran through the market prices and gave the optimal portfolio to match and they then readjusted their positions.
I dont really know much more than that - but it seemed interesting in that they appeared to almost have a preset portfolio that they could select from and then they took the positions they wanted from that rather than building from the ground up so to speak by pluggin in positions as they came along.
They mentioned that for example - say they were long 6 units the optimal portfolio based on the model showed they should be long 4 units, they just sold two out. the next few days they might have to buy those 4 back.