Decoding Markets and Marwood Research
Download 1.87 Mb. Pdf ko'rish
|
101 Lessons For Aspiring Traders PDF
...on trading systems
101 LESSONS FOR ASPIRING TRADERS Select trading parameters (rules) based on robustness rather than outperformance. For example, neighbouring parameters. An outlier result is often due to luck whereas strong performing parameters found in clusters are a sign of robustness (sensitivity analysis). 11 12 Try to get hold of additional data sets which you can backtest your trading strategy on. The more datasets and markets that your strategy works on, the more robust it likely is. 13 14 Trading strategies don’t always have to make economic sense but you should be able to speculate where the edge comes from. For example, a human bias or market efficiency. If you can’t form a narrative as to why your strategy works, there’s a chance it is over-optimized or data mined. A strategy that has been data mined can work but it requires more stringent validation than a strategy that is based on economic logic or common sense. It takes more work to prove it’s not random, in other words. 15 You are better off trading one robust trading strategy than combining 100 over fitted, losing strategies. 16 17 18 19 Quant traders sometimes struggle by focussing only on the numbers. Don’t be afraid to combine your trading system ideas with fundamental analysis, economics, human judgement etc. This is a fertile area with which to gain ground on pure algo traders. If your backtest results appear too good to be true, be vary careful and look for mistakes in the code. Beware of forward leaks or execution errors where your trading system enters positions that could not have occured in real trading. Use some margin of safety in your backtesting as a way to find more robust strategies. For example, use slippage and commissions slightly higher than what is realistic. If you can make money with high slippage, you could get a pleasant surprise in real trading. It’s entirely possible to come up with profitable trading strategies based on very small sample sizes. But you need a higher degree of conviction that the edge is real. 20 Never assume that a backtest is a realistic representation of reality. Market microstructure, future events and data errors can make a dramatic impression on a strategy once it is taken live. One way to tell a system is not working is if you get a longer losing streak or higher DD than in backtesting. 101 LESSONS FOR ASPIRING TRADERS |
Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2024
ma'muriyatiga murojaat qiling
ma'muriyatiga murojaat qiling