Most profitable Algorithms on Quantopian?

Hey guys,

I want to make a list of the most profitable algorithms here on this great site.

Feel free to help me to collect algos;)

6 responses

Hi Max,

Welcome to Quantopian! I'm glad you like the site.

I think one of my early algos is still the leader a 200,000% return. That really just serves to point out how complex your question actually is. The algo has to be profitable going forward, and it has to have an acceptable risk level. Figuring out those details is really where the problem gets interesting.


Hello Dan,

One thing to think about is how Quantopian can provide a framework and tools for assessing the risk-reward profile of an algorithm. In a simplistic view, it is a matter of plotting reward versus risk over algorithm "settings." For example:

Max's question is complex, but it would seem that there should be a systematic way of addressing it. More later...gotta run.


Grant, I think you're absolutely right. Quantopian is uniquely positioned to evaluate other people's algorithms. We're not trading ourselves, and we're not trying to get people to invest in our algorithms - that permits us to be relatively unbiased. We have an open-source backtester that permits anyone to scrutinize our methods of measurement - that makes us a lot more trustworthy than just providing a black box and claiming the box is backtester.

It starts getting really interesting when you start trying to define risk and reward. There are some basic ideas of risk that most people accept, but the details get very contentious. We're going to be able to permit anyone to choose the risk/valuations methods that they prefer. That's going to be fun.


Hello Dan,

My thinking is that some sort of batch mode would be useful. It is perhaps a bit down the road for you guys, but there needs to be some way to iterate a backtest algorithm over parameter space (e.g. buy/sell thresholds, moving average windows, time frames, etc.). If you tabulate the results (e.g returns, risk metrics, etc.), users could cut-and-paste (or download) them into plotting/optimization software for analysis. Let the computer do the work, rather than manually adjusting parameters and then running the backtest, repeatedly. This manual approach carries the risk that the user won't explore the entire parameter space because of the tedium and time it takes to execute.


Grant, have you read Thomas's bit on parameter optimization? Does that sound like what you're looking for?

Thomas has done some work on the feature, but it got stuck behind our other work, like set_universe.

Thanks Dan,

I read through parameter optimization. Sorta makes sense. I'll ask Thomas how it might be applied to the OLMAR algorithm as a test case.