If you have such a high sharpe then use all the leverage you can get and compound. Why be scared of 1% DD, that is irrationally risk averse. Now returns are drying up, probably due to increased market efficiency as much as anything, and you are blaming Trump. You should blame yourself for not taking full advantage while the getting was good. Good luck on your next system.
The thing is that the person who is capable of developing such a system and running it for a decade obviously understands the irrationality of staying within 2%DD. One candidate explanation is that allowing 10%DD would not help improve returns but the author replies "maybe i will try to scale up a little bit in 2020." like if the latter was not studied experimentally.
there is no president in history who manipulated market as much as he does. if market goes down for a while, you can expect he will tweet some fake good news to boost the market, e.g., great process made on trade deal with china, etc. I am leveraging a little bit more on my system. there is a limit how much I can leverage before i influence the entry.
>> there is a limit how much I can leverage before i influence the entry. Makes sense. Did you try to run two versions of your system such that entries/trades are split between the two systems so that one can check whether having 2x greater leverage for the second system worth it?
how can i do that test? if i run same system on second account, equivalent to leverage 2 times. hard to know if the entry has been influenced, only time can tell by its performance. anyway, I am testing a second system live on another account which is derived from the first system, they do not compete necessarily for the entry. i made profits on the last two days. wait to see how it goes.
Why do you believe that market going up is bad for your system? Was your strategy making a killing in 2018 when VIX was higher? Maybe post your curve from 2018. I see no correlation with volatility and the equity curve you posted.
when market is up, vix will be low, which is bad for most day traders. Dec 2018 was the best month of the year.
>> how can i do that test? You don't need to have another account to test it. I assume you are good at programming to implement the following. You have an iterator i which counts trade index, every time you plan to enter you set i = i + 1. When i div 2 == 0 you use strategy#1. When i div 2 == 1 you use strategy#2. You make it so that both strategy cannot use more than 50% of your cash and they operate independently of each other meaning that their statistics (gains/losses) are counted independently. Yes, the results whether #1 is better than #2 is stochastic but running both for say a month or two (i.e., after a good number of trades) should already tell you which one delivers somewhat better results. You customize strategies as you wish, e.g., by allowing different leverage levels. The point is to view your only account as two and deal with it. You can then extend it to many strategies and thus perform online tuning of hyperparameters by keeping better strategies and killing worse ones. For instance, I run 48 bots with different hyperparameter values and periodically adjust/select better hyperparameter values.
thanks. I understand what you mean. in your case, the 48 bots are same strategies but with different parameters?
The are essentially the same. They actually compete for trades, i.e., they have the same suggested stock (supplied by a ML model) to trade on but they set orders at a different distance w.r.t. best price and they try to exit at different distances as well. You can see the dynamics that some strategies trade much more often with less profit while some others want to get a bigger gain each time and thus usually are traded less often. Some things are almost impossible to backtest so you have to run them online and see which one is better (e.g., it depends on volatility) for the current market conditions. You may try to pick a hyperparameter which you believe is among the most important ones and deviate its value by some percentage which would be sufficiently large to believe that the difference in performance (e.g., after a few weeks) is not due to noise only but really due to a difference in performance profile. On the other hand, you should not deviate your parameters too much, otherwise one of your strategies might be broken. It is basically noisy optimization. Pick a hyperparameter, set a range of possible values and select a set of candidate solutions, run them long enough, adjust the range, iterate. If it works well, do the same for a set of hyperparameters. Ultimately, you can run some optimization algorithm which will adjust hyperparameter values for you. The key problem is that some parameters which are better this week might be worse next week, so one should do it rather conservatively (e.g., always run your all-time-best choice strategy to define a baseline to compare to)