I am building a machine learning backtesting platform (C# .NET Core) and it currently supports EOD data. My next step is to add intra-day (1 min) futures data support. I have seen two commercially available products similar to the platform I am building and they range in price from $5000 to $20000 for an annual license. I am offering a free license for my product (once complete) in exchange for continuous 1 minute historical (all) for the following futures symbols: ES, NQ, YM, ZN, CL, GC, SIL I really need all historical data because success of the platform will be in its ability to discover robust systems that perform well over many market cycles. I would be happy to do a Skype call or Google hangout for those interested to demonstrate my platform so you can see what you get for the trade. thanks!
Why not reach some like DTNIQ who may supply this data and may stream it into your platform. Just a thought.
You can get 1 minute prices but continuous requires some sort of stitching. each stitching algo is slightly different and has its tradeoffs. I think you need to find out more about this. Also your system will not be accurate without accounting for futures rolls .
Yes...at some point is will need a proper subscription that can be at least updated daily (my platform is not a live trading platform so realtime is not necessary).
I understand. Ideally the data would be pre-stitched with the knowing of the limitations that entails. I am not looking for a perfect solution, just a enough to prove my technology.
I believe they will supply you EOD data along with historicals as well. If you PM me I can give you the person that handles our account so you can reach out directly and see if they have what you need.
Minimum order is $1000 at www.tickdata.com. If I get no takers on my offer I will buy data as needed.
IQFeed has 1 minute OHLCV data back to 2007. That's likely to be your best route. If you go too far back it may actually be harmful. The market is not the same today as it was 15 years ago. Even something like "noise" is going to be different, the way orders are broken up, executed, how market makers and HFTs move their spreads, all of this is drastically different. Is it worth training your system in an environment that is too different from what it's going to encounter when run live?
I agree. Fan27, you really should build the ability to handle individual contracts and various roll protocols into your backtesting/model discovery software.