I have re-read this thread and have some final comments. To me position/risk management can be grouped into: 1. Growth optimal position sizing under risk constraints and the assumption that your trading systems work (i.e. have an edge) and that future returns will be drawn from a distribution similar to the historical. 2. Scaling out of systems that stop working, i.e. where the edge disappears. The assumption is that the system fails slowly compared to the time it takes to scale down leverage (the reaction time of leverage control). 3. Worst case/catastrophe scenarios. These are not really related to if the trading system works or not, but more to having bad luck. Even trading systems that works well can take on positions that result in big losses (news events, disasters, etc) Whereas LSP covers the first two it does not address the third (at least not directly). This was discussed in the beginning of the thread but no systematic approach was presented. We have a test, called âthe unlucky bastard testâ, that we run before deciding on limits on leverage levels. It tries to estimate tail risks for our stock trading systems. Input data is the historical daily close-to-close (or close-to-open) returns and the leverage/position sizes (these are kept unchanged during the period). From these the result of the most adverse selection of stocks is calculated for each day. Moreover it calculates the proportion of the possible combinations of stocks that would violate a daily âruinâ condition. In the end we get a value for the largest daily loss during the period (e.g. 10 years) and the probability of avoiding ruin during the period given our leverage. This type of test will reduce the risk of using too high leverage due to instability of the returns distributions. Hugin
Thanks Hugin, What do you think about manually altering the LSP joint probability table to include the disaster scenarios? For example, say historically you are trading 3 systems and your systems don't have a historical probability associated with all three being down 10% in one day due to some unknown unknown... One can easily add this type of dynamic into the optimal growth algo and see the results. Obviously the end result given such an event will always be the utlization of lower leverage, but depending on how conservative/aggressive one wants to be, the hard numbers will be available for analysis.
Mike, I think it is possible but there are some things to keep in mind. Even if you change the information in the joint prob tables LSP will, in its drawdown simulation, combine "disaster" returns with the unaltered returns so the end result will be hard to interpret. We did some analysis regarding this and in the end we prefer to keep things separated. This means that we will adjust outcomes and probabilities in the tables if we feel uncertain about the returns series from the trading system. This will affect the leverage and reduce geometric return estimates. But we think it does not describe the consequence of catastrophic events, since these are, by definition, outside normal market behaviour. So we still use the catastrophic event test. The result from this test will provide a warning if the leverage suggested by LSP gives too high probability of reaching "ruin" before we have a chance to bring down the leverage. Hugin
As noted on the closed thread, I contributed by pointing out an edge based strategy on the reasons for buying SNT. Since I can't report the results of this trade on that thread, I wanted to post here that SNT has now moved up just today 35% from the price where I noted I was long the stock for the reasons mentioned in the other thread, and this is only the start, since you don't want to sell your winners too soon, I see much more room to the upside.