Ask HN: Review my startup idea, Algonaut | Hacker News Reader

Ask HN: Review my startup idea, Algonaut

22 pointsabhijitr11 comments

http://www.algonaut.com - a hosted platform for algorithmic trading

We are still circling in on what the MVP should be, would love to hear feedback from any quant-types out there.

I do algorithmic trading myself since 2006, with Interactive Broker, and C++/C# implementation of my own heuristics. As others said you are not providing enough information. I don't want to have to sign up to discover what you offer. At the minimum, you do need to (1) provide some pricing information, (2) put more emphasis on security (how do I know that my crown jewel is secure with your service, do I need to give my password to you), and (3) provide some idea what the workflow is like. For example, give people test account and canned algorithm just so I can easily get a feel of what this is about.

I probably are not interested myself. A big part of my algorithmic trading is many different types of analysis. I need to be able to easily change something and debug and run other tools (e.g. various statistics packages). Having it hosted elsewhere rather than by myself seems to complicate the problem. And what's the benefit, really? Do you have a better feed than IB? Do I get dedicated trading machines or do I get a VM and risk slow execution? What is the cost? Is there enough saving to justify the risk and inconvenience? All these require more than a teaser MVP page.

You do need to define things a bit more. Daily data or tick? Stock, futures, etc. Just US data or more.

How are you going to allow any languages to run on your server? That seems pretty dangerous.

Cool idea, though.

Thanks for your feedback! We will clarify it on the page, but the short answer is both daily and tick data. We plan on starting with US equities unless we get strong feedback to the contrary.

As for security, we plan on running each user's code in its own sandbox, with restricted permissions.

If you want a MVP, then daily US equity is probably the place to start. I'd also wonder if your service could run tick strategies in practice. I've never done it, but it seems like tick strategies are all about having expensive co-located servers.

I'd also suggest getting fundamental data for the equities, so your customers have more than just hi/lo/open/close/volume for their strategies. You'll find that making sure none of this fundamental data is snooped is hard. (i.e. making sure the data isn't revised back in time, so it is as it would appear on a given date). That messiness is your good value add if you can get it right. Also, please encourage out of sample testing using some form of cross validation.

Agreed, we believe there's a huge amount of value in aggregating and cleaning fundamental data.

To your point about back testing, would you want guidance as to what the in-sample and out-sample time ranges should be? Tools to help you make sure you're not overfitting? Or something else?

The user's algorithms are going to have try to control for overfitting on their own. Try to make it hard for the users to evaluate the model on the same data they trained it on. Also educate them that in-sample results are meaningless.

I guess I'd want an easy way to get data in chunks for testing. You can either do it as a rolling window or regular cross validation. (Rolling would be train on N years of data, test on the next year. Then drop the last year and add another year and repeat. CV would be just divide in 10 chunks, train on all combination of 9 and test on the 10th.) It would be good to support both.

It's an interesting idea, but you need to provide more information in order to get feedback. Right now it seems like I would have to sign up to find out more about the system. Could you provide more examples of what a coded strategy would look like, maybe some screenshots of the working interface, and what the business model (pricing etc) is?

Thanks! We're not quite ready to share screenshots, but I see your point about providing more technical details. Some code samples would definitely help. As for the business model, we're still iterating on that, but the current thinking is a straight SaaS play.

Link: http://www.algonaut.com/

I know nil about the subject space, but that seems like a great domain name!

Reply on news.ycombinator.com