I am amazed at how many possibilities the program has, but I think I am missing many of them. Also, the market is very unstable now, for intraday trading at least, so many simple strategies from the past are not performing well. I would like to discuss the best practices of 2023 to get a stable profit. What timeframes are better to use, what tools help more, what type of strategies work better in your opinion and so on. And what's not worth spending time on, of course.
As for me, I have recently started to study Neurolab. I took the best indicators from Indicator Profiler and added them there. So far I have not been able to achieve stable results, maybe I have a gap somewhere.
As for me, I have recently started to study Neurolab. I took the best indicators from Indicator Profiler and added them there. So far I have not been able to achieve stable results, maybe I have a gap somewhere.
Rename
What I use is a combination of daily strategies in a MetaStrategy. For me this has smoothed out the equity curve and stabilized profits.
QUOTE:
... the best practices of 2023 to get a stable profit.
It's a good question. When building any empirical model, you want to have as many significant orthogonal inputs as possible. For stock trading, the only "daily" raw-orthogonal inputs available are price and volume, so I would choose adaptive indicators that take both of these into account. That is, indicators that accept BarHistory input instead of TimeSeries input. And yes, that includes all the Accumulation/Distribution indicators (which is a big family). For starters, below is a summary from Fidelity Investments' Technical Indicator Guide of a few common indicators fitting this goal. Some of the names might be different than what WL calls them.
Some of these are "Indexes", not indicators, in the sense you should combine a group of like stocks to compute them. For example, the Arms Index (TRIN) should be computed with WL's IndexLab extension to compute the CompTRIN index. Install the IndexLab extension, then read the docs about it.
How feasible is it to test strategies over the long term? For example, does it make sense to test for 15+ years on daily bars?
QUOTE:My simulation runs cover 5 years of Daily bars. I can't really say if that's good or bad because I haven't tried more years. But market conditions change, so a strategy that works well one year may work poorly the next. So simulating too far back maybe a disadvantage.
does it make sense to test for 15+ years on daily bars?
On the down side, if you're simulating with less years back, then you need to optimize more often because your parameters are going to be more of a function of immediate market dynamics. For example, parameter settings that worked well when interest rates were low may need to be totally different (re-optimized) when interest rates are high.
The same is true of strategies. A buy-high strategy may perform better than a buy-low strategy when interest rate are low. But the converse is true when interest rates are going up; buy-high strategies become very problematic in a volatile market.
---
The best solution is to choose timeless parameters that can remain unchanged in all kinds of markets. For example, parameters that set your risk tolerance such as a probability of failure. With that set, you can let your statistical analysis routines employ those probability settings to formulate your ideal trailing stop price. (This example may require "robust" statistics instead of classical statistics [i.e. moment-based statistics], but that's another topic.)
Your Response
Post
Edit Post
Login is required