Two Rules Governing the Success of Trading System Automation, 09/05/09
Two Rules Governing the Success of Trading System Automation
The liklihood of success and the degree of degredation of a trading system under automation follow two rules:- The liklihood of success of the automation varies proprtionately with the expected length
of the trade, i.e., a trade with a longer time horizon has a greater chance of being
successfully automated than a trade with a shorter time horizon - all things being equal.
- The degree of degredation of the system varies inversely with the expected length of the trade, i.e., a trade with a shorter time horizon will suffer greater performance degredation than one with a longer time horizon - all things being equal.
What does this mean?
It's easier to automate long term systems than short term systems : )How I derived the Rules
At the beginning of the year I discovered a group of short term anomalies that tested out really well but in practice didn't perform to expectations (even low expectations!)The systems' smaller expectations just couldn't overcome the increased standard deviation from slippage, missed trades, and increased stop losses of the shorter time frame, i.e., when you are trading for a single point of expectation, decreases in performance come in -25% chunks and it doesn't take very many chunks to get to zero.
And it's a short insight from -25% chunks to 'Aha!'
What else does this mean?
Brett helped me flesh the rules out this morning and this is what I think it means:Automated trading systems should outperform discretionary traders in longer time periods and discretionary traders should outperform automated systems in shorter time periods - all things being equal*.
Further that performance differential should follow a stable curve with respect to time horizon of trades.
*All things are _not_ equal at the millisecond time frame so the 'curve' will roll back over when machine performance exceeds human capabilities.