- 02/18/2020 at 8:44 am #77724delphionParticipant
What would you consider the best period in the optimizer to set to reflect the best or most realistic current conditions? Another way to ask the question is a certain period setting more ideal for optimizing cagr, volatility, or max drawdown?02/19/2020 at 8:43 pm #77738Mark VincentParticipant
I am not sure what you mean by period but I presume you mean History Range and not the Lookback(LB)Per.(Days). The current conditions in a 12 year bull market the settings in most strategies are set to hedge more. This sacrifices a bit of Alpha but I would agree it’s the right choice. There is no one History Range that is best in my opinion. If you take a very simple strategy like UIS SPY-hedged and run it across multiple time frames like 2, 3, 5, 10, 20 years you will see that the optimizer is very stable across all Volatility Attenuator options and lookback periods. Which means it is more reliable and more likely to have the same future results but nothing is guaranteed. If you have the opposite when you run the optimizer the strategy is probably not that stable and is very sensitive to any changes. You probably don’t want this. The optimizer settings and how to use it are one of the hardest things to figure out what is best. I am hoping LI or other members can add to the discussion. For instance I have never used the advanced settings I am not sure how or when I would use them.
Mark Vincent02/22/2020 at 2:25 am #77760StefanMParticipant
I agree that the choice of which Lookback period to run the Optimiser is one of the most challenging and somewhat opaque aspects of QT.
I agree that finding a strategy with stable results over 2, 3, 4, 5, 10, 20 year periods (not 1 because that is too much of a curve fit), is ideal.
Having said that, I would suggest a Lookback of 2 years for the more aggressive strategies as they may have a limited shelf-life because market conditions keep changing.02/22/2020 at 4:26 am #77764VangelisKeymaster
It’s a challenge to pick an optimization period. Mark and Stefan are pointing you in the right direction. What I can add is this: Our most robust strategy, out-of-sample. has been the Universal Investment strategy (UIS). If you open UIS in QT and optimise different periods you always get fairly good results no matter how much you change lookback periods and other parameters. This is because the base strategy, the 60/40 portfolio, is robust on a fundamental level. So if you have some free time, an exercise is to run different optimisations on your strategy at 2 to 4 year lookback periods from different points in time. For example you could optimize from 2013-2016, write down the parameters and then see how they would perform from 2016-2018.
All the above refers to QuantTrader. For the online portfolio optimizer a lookback period should at minimum include a bad year (2015 or 2018) and a good one. Because the underlying strategies are being optimized themselves, there is an argument for not optimizing too much or too far back on a meta-strategy level. Some studies argue that equal weight portfolios perform better than most other combinations. There is nothing wrong in running the online optimizer, adjusting the weights towards equal-weight while still using common sense, i.e. putting less weight on risky ones. In other words optimisation on the meta-strategy level is not critical and could be subjectively corrected.
- You must be logged in to reply to this topic.