Acrary is a genius!
- Thread starter greaterreturn
- Start date
- Status
While we have some experts here, I was hoping that I could ask you guys a question or two.
How does one avoid the trap of overoptimizing when combining stratagies?
I have a couple of stratagies that have sharpes in the 1.5-2 range. All stratagies trade daily, I ran the analysis on about 4 years worth of data.
Now I also have stratagies that by themselves have ~1 or just under 1 sharpe, that when combined optimally on historica data increase the Sharpe of the bundle.
I'm worried that I'm overoptimizing on the combining ratios. How does one know if the improvement will carry forward, or if the combo only improves in hind sight?
Thanks for any insight you are willing to share.
How does one avoid the trap of overoptimizing when combining stratagies?
I have a couple of stratagies that have sharpes in the 1.5-2 range. All stratagies trade daily, I ran the analysis on about 4 years worth of data.
Now I also have stratagies that by themselves have ~1 or just under 1 sharpe, that when combined optimally on historica data increase the Sharpe of the bundle.
I'm worried that I'm overoptimizing on the combining ratios. How does one know if the improvement will carry forward, or if the combo only improves in hind sight?
Thanks for any insight you are willing to share.
In predictive modeling what you are doing is called composite modeling, voting models or similar terms. In that domain it does work dues to a process where error cancel out since they are randomly distributed while correct predictions reinforce and amplify each other.
I can't speak to the effect in just combining a strategy which I assume is a linear rule set.
Warning to those that are new: there is alot of misinformation in this thread.
"your combine sharpe is above 1, but your profits are less per capital in use at the same time, so the overall risk/reward ratio is still the same."
Note that Maestro has his own definition of "risk/reward" that may likely be different than the way it is commonly used.
I can tell anyone for a fact that at the institutional/professional level people care about Sharpe and Sortinos, amongst others...
Transatlantic
"your combine sharpe is above 1, but your profits are less per capital in use at the same time, so the overall risk/reward ratio is still the same."
Note that Maestro has his own definition of "risk/reward" that may likely be different than the way it is commonly used.
I can tell anyone for a fact that at the institutional/professional level people care about Sharpe and Sortinos, amongst others...
Transatlantic
Iâll try:
When forming a trading portfolio with multiple trading strategies one has to always check whether those strategies increase or decrease the margin. If a trader simply uses correlated/uncorrelated strategies on a set of underlying securities the margin most often goes up reducing the profitability of the portfolio. However, there are strategies that are capable of the margin reductions. Such strategies are usually formed out of derivatives (options, futures, future options etc.) For example playing Call/Put butterflies could reduce overall margin while increasing the profitability of the portfolio. Also, it is increasingly important to know your risk/reward ratio that includes your margin requirements. There are many techniques that allow traders utilize well-known margin calculating formulas to improve the profitability of their âsyntheticâ portfolios. Again, the risk/reward ratio could be calculated using many different formulas. We use simple ratio of the probabilities multiplied by the profit/margin and loss/margin formulas. I hope it clears it a bit. I really do not want to go into further details as they would require further disclosures.
Cheers,
MAESTRO
- Status