@backtrader My sketch might have been a poor illustration for what I'm trying to achieve.

In backtrader, let's say I want to take that standard RSI function and run an exponential moving average over the values to create a sort of RSIEMA (or, perhaps a fit it against a polynomial approximation with ARIMA forecast weighting.) It would be convenient if the code for the indicator was there in my notebook, ready for experimentation. Then, rather than run the algorithm with the display turned off for months / years at a time, I'd like to watch the algorithm's behavior and indicator signals.

In this sense, it would be really cool to decouple the algorithm from the portfolio performance testing environment, and see more in the way of strategy / indicator debugging. I'd like to be able to test a strategy with multiple backtesting frameworks and ultimately see the same results for confirmation.

I was just curious if there was an example for working in this fashion.

@backtrader My apologies. What I'm looking for is a general strategy which lives outside of Backtrader for easy debugging, tweaking, and visualization. Eg., perhaps I'd like something that works like an RSI, but works differently, and I'd like to use threshold values with this strategy after seeing what the general shape of the output wave looks like.

From here, a small Backtrader wrapper, or Backtesting.py wrapper, or QuantConnect wrapper might be able to interact with the script. I imported Backtesting, but I also have the other two libraries installed.

I'd like to abstract my technical indicator processing, trade decision logic, tradebar data streams, and final algorithm performance metrics inside some sort of notebook. Conceptually, I'd like to have a collection of rolling windows with a historical data warmup to fill queues, ready the indicators & smoothing functions, and train models.

With this sort of modular design, it seems like there would be a nice balance between fast research prototyping, debugging, and easy tradebar-accurate porting to other vendor configurations.

As a sketch, I set this up in Python with Jupyter Notebook, which more or less performs the task. It's surprisingly slow, however. https://github.com/iRyanBell/chart_exploration_notebook/blob/master/Chart Exploration.ipynb

Are there any better examples for performing this sort of exploratory research work in Python?

On QuantConnect, trade performance is largely evaluated by a probabilistic Sharpe ratio which updates on a per-trade basis. In backtrader, there's a Sharpe Ratio https://www.backtrader.com/docu/analyzers-reference/#sharperatio which can be retrieved after a run.

Has anyone explored getting a continuous probabilistic metric similar to the one from QC?