Metaphors are tricky things to play with. They serve a purpose to illuminate relationships in terms we are familiar with, but for those who insist on taking every little thing literally, they are deadly. Every reader of literature understands that metaphors break down at some point, but they are fine with that. Suspend your disbelief for a moment, and you may gain some useful insights. With this forewarning, I present the System Trading Game metaphor. You can think of a game as chess, golf or the 100-yard dash. We are interested in how well a player plays a game. We'd like to know a few things about their natural ability, their ability to improve their performance and their actual game-time record. We are interested in these factors because we, as system traders, decide what players to play in what game and how much playing time they get.
Our player is our trade system. It has a genetic make-up that is fixed (a set of trading rules). Think of the genetic make-up of an athlete in a typical sport. Tallness is good for basketball players, but shortness doesn't necessarily limit someone from going pro. Eye-hand coordination is important for tennis players, so it gives someone with two good eyes and at least one hand a distinct advantage over someone who doesn't have these traits. Some player traits, such as tallness and physical strength give them an advantage at some games, but not necessarily others. Think of the results you would get if Shaquille O'Neal and Bobby Fisher played one-on-one basketball versus if they engaged in a chess match. Keep this in the back of your mind.
Our game is our market. The Corn market is different from the SPX market is different from the Gold market. Some markets (like sporting games) are similar, but that doesn't mean our player is suited to playing both. Remember when perhaps the greatest basketball player of our time, Michael Jordan, decided to start playing baseball? You get the idea.
Our game plan is our trade plan. We are the head coach, the manager or what have you. If we are playing a chess tournament, we may decide to bench Jordan and O'Neal and instead play Fisher and Garry Kasparov. If we are in a violin recital, we probably give the instrument to Jascha Heifetz and Yehudi Menuhin, and let Lester Flatts and Earl Scruggs wait for the next folk music tournament.
As game managers, we need a standardized metric to judge each player's natural ability to play a particular game. This is only the first of three metrics for player evaluation though, as we all know that natural ability does not always translate into a winner. Our natural ability test is the backtest of our system on a particular market over a prescribed period. Corn, GOOG, Palladium, US Bonds, what you will. How did our player perform? Net Profit? Max Drawdown? Correlation to Perfect Profit? These metrics can be enumerated later, but for now we painting with broad strokes.
Our player is then judged on its ability to improve its performance. Can our naturally gifted long-distance runner improve her times if we gave her better shoes? This improvement test is the optimization of our system across the same data as the original backtest. And again, we compile information about how they did.
Finally, our player is tested in real game situations. Just to make sure our initial excitement does not get quashed by someone who simply chokes under pressure. Hey, it happens. Remember Tin Cup? The game test is our walk-forward. It is conducted on fresh data, previously not trained on. We once again compile information about the results. It is these game test results that allow us to start ranking players. The previous tests (backtest, optimization) gave us insight, but game time is the real thing.
This methodology is most eloquently elaborated in Bob Pardo's The Evaluation and Optimization of Trading Strategies. A must-read for system traders.
There are many ranking algorithms already in use that can be applied to our trade systems. The method of determining the top-ranked tennis player in the nation or the top-ranked collegiate basketball team may be a place to start.
Now that we have a player rank, we as managers need to develop the ability to get the most out of our players in the games most favorable to their abilities. This art goes to the question of what markets to play, and what trade-size allotments to make. Position sizing is really deciding how much playing time to give a player.
This metaphor is not over though. Players and coaches always seek to gain an extra edge. They may put their best players on a specific meal-plan, vitamin supplement regime or (and this is cheating) performance-enhancement drugs.
Our performance-enhancing efforts will be focused on intermarket algorithms. The basic idea behind this neat little drug is that it can boost performance of a player, but only if the player is already good or excellent. This area has had many false starts, but so did steroids. The future of intermarket algorithms is not some naive correlation plot, but something a bit more complex, to match the reality of the markets. It involves factor graphs and machine learning approaches based on Bayesian probability distributions. And that clearly is the topic of another post.
3 comments: