Pretty cool seeing those Vann Diagrams. Kamala would be proud. Have you tested to see if the trades (mean returns) in the intersection of the Venn Diagram are different from the mean returns in non-intersecting circles? One would think intersection trades should have a higher win rate, since they have 2 successful signals supporting them.
Pretty cool seeing those Vann Diagrams. Kamala would be proud. Have you tested to see if the trades (mean returns) in the intersection of the Venn Diagram are different from the mean returns in non-intersecting circles? One would think intersection trades should have a higher win rate, since they have 2 successful signals supporting them.
Excellent point! I tested it. Win rate is slightly better (56%), but the mean returns are worse.
Backtesting it produces a 15.1% annual return with a 25% max drawdown, which is slightly worse than what I shared. I believe that this happens because, once we do the intersection, we remove the trades on the tails of the distribution, which are important to the performance (especially in times when the market is bouncing back... my intuition comes from observing a weaker performance during these times by looking at the equity curve and comparing it with the previous exercises)
Interesting. It's also less data + more fitted, so higher expected future volatility.
Out of curiosity, what's the difference between a strategy that you test like this one & a strategy that you trade live? When you run it live, do you add stop losses & etc? Or just trade it as is
Never as is. At minimum, I add risk management measures (filters, different position sizing, etc), but usually not stop losses in mean reversion strats as they degrade the performance
Pretty cool seeing those Vann Diagrams. Kamala would be proud. Have you tested to see if the trades (mean returns) in the intersection of the Venn Diagram are different from the mean returns in non-intersecting circles? One would think intersection trades should have a higher win rate, since they have 2 successful signals supporting them.
Excellent point! I tested it. Win rate is slightly better (56%), but the mean returns are worse.
Backtesting it produces a 15.1% annual return with a 25% max drawdown, which is slightly worse than what I shared. I believe that this happens because, once we do the intersection, we remove the trades on the tails of the distribution, which are important to the performance (especially in times when the market is bouncing back... my intuition comes from observing a weaker performance during these times by looking at the equity curve and comparing it with the previous exercises)
Interesting. It's also less data + more fitted, so higher expected future volatility.
Out of curiosity, what's the difference between a strategy that you test like this one & a strategy that you trade live? When you run it live, do you add stop losses & etc? Or just trade it as is
Never as is. At minimum, I add risk management measures (filters, different position sizing, etc), but usually not stop losses in mean reversion strats as they degrade the performance