Dec

19

Most people search or try optimize for highest system return. It is not the most profitable over time. The amount of profit over time is determined by the money management you apply to the system more than by the system itself. This is mind boggling to me.

- James Sogi

H. Humbert counters:

In one of the many money manager podcasts I listen to, one of them used this very assertion as an example of, shall we say politely, a less than optimal belief. But he used stronger language.

Peter Ringel writes:

It is still important to aim for a good naked system (without position sizing applied). The risk/drawdown vs overall return relation comes from the position sizing applied world. A better core system makes more aggressive position sizing possible.

Zubin Al Genubi replies:

A better core system makes more aggressive position sizing possible.

Disagree. According to Ralph Vince bets in excess of optimal f results in lower overall system returns due to larger drawdowns with larger size! Comparing core systems should be by geometric mean, not necessarily w/l, %win, t score, etc. Interestingly Sptiznagel says something very similar. There is something very important going on here that is being missed.

Gyve Bones comments:

Depending on the breaks of course, there is no money management system method that can turn a no-edge “loser” naked trading system into a winner apart from lucky breaks. But a winner with a naked edge can be ruinous with over-sized bets, or smothered by various vig drags if the bets are under-sized. As one guy put it in this article from 2000, the key is to find the sweet spot in between.

But as Ralph has shown, the sweet spot, the “optimal-ƒ”, means that the better the system, the higher the ƒ value, on a scale of 0.0 to 1.0 means that if the largest losing trade used in the sample ever re-occurs, your stake will have a single-trade drawdown equal to ƒ%. That is, if the optimal–ƒ is 0.65, and then you have a re-occurrence of the worst trade from the history of the system, you will have a 65% drawdown of the portfolio. But trading at ƒ is the only way to make sure you’re not over betting or under-betting in order to maximize the potential gains of the trading system, if you accept the premise that the series of trades you feed into the optimal-ƒ algorithm is a reasonable and realistic representative of the trade returns going forward trading that system.

Larry Williams has a definite view:

BETTER CORE SYSTEM ETC IS MEANINGLESS. The past is never the future and it takes only one trade to put a bullet through your skull when you mess up. Past ’good numbers’ from a trading strategy are meaningless.

Peter Ringel responds:

but even the Kamikaze-trader dialed it up to 11 to win championships in a stellar way and endured brutal drawdowns. and the final win, of course, impossible without an underlying strategy.

Larry Williams replies:

Kamikaze man was clueless, mindless and fearless as well as blessed with luck and Mr Vince to plug holes in the dyke.

Zubin Al Genubi gets statistical:

A benefit of using parametric techniques is that empirical data isn't required and we can do what if's as conditions change.

James Goldcamp writes:

When coming up with a position size rule it must be as with the system itself subjected to in and out of sample testing. We used to have a program circa 1998 that would calculate the optimal ("f") amount of capital over first X trades then apply to the rest of history using the optimal method. This led to hypothetical out of sample blow up not infrequently due to the instability of model returns (even for models that were to some degree still profitable on blind data).

My subjective belief is that most edges (perhaps other than those derived for market making ultra, frequent, or arbitrage/structural type trades) are way too unstable to try to extract anything approaching a past optimal bet size. It seems like the 3 questions or dimensions that one deals with are will it still work at all in future, if it does how much will it vary from the past (expectation and path), and how will the aforementioned two work in relation to other methods you have that work. The last point relates to in my observation the most common form of risk management, multiple bets with negative or low correlation, that's perceived to be a better way of managing risk than dialing leverage of any particular return stream. Any of the aspects are subject to the ever changing cycles.

Big Al adds:

Often the tricky part is finding uncorrelated assets that are reasonable trades or investments.

James Goldcamp responds:

I agree totally. For me it's the 3rd uncontrollable variable - if the ideas work, how well they repeat (robustness I guess), and how they continue to relate to other things. Hypothetical modeling of complex portfolios often assumes all of these properties will continue. There are lots of ways for a leg on your table to collapse!

H. Humbert comments:

Since the number of unknown important variables in complex real-world problems as opposed to simple games of chance of even poker can never be fully known, and the influence of even known variables, by themselves and in combination, can only be examined via past data and in no controlled experiments, it seems like any system can experience a catastrophic failure and/or change in being amenable to any strategy at any time. I admire traders who brave these unknowns and prefer to rely on drift that seems to be more robust and stopped only by major wars and revolutions.


Comments

Name

Email

Website

Speak your mind

Archives

Resources & Links

Search