Currently https://tradenotebook-alpha.herokuapp.com/ points to develop.
Secrets in environment variable - source .env or add to config vars in production.
The purpose of this app is to simulate actors in financial markets, deriving from the concept of a zero intelligence agent-based model on ETFs. The finished model should present an educational visualization of the effects of primary actors in financial markets.
Lightweight code refactor:
- Remove H5 output and associated code
- Implement dataframe logic for delta over time and document output
- Consider implementing a wrapper for broader time periods - higher computational work, less json output
- Update schema design and add constraints
- Implement fuzzy logic prediction, perhaps with
fbprohet
The main priority is to expose some of the internal endpoints to be able to POST data such as number of market makers, erratic actors, etc...
The current marketmaker class is a bit too arbitrary in pulling/pushing quotes. There should be some sort of fuzzy logic there that's not just a random guassian distribution.
Implement drift with Geometric brownian motion - needs to be quantifiable.
What economic model do you want to "enforce"? Merton?
What do you actually need to return? List those out; what "steps" do you actually need to return to make a nicer time series? I don't think it will end up being "live". At least not in this implementation. But can still graph the marketmaker's profitability on the same scale.
Need a way to set the starting and stop prices. Also no need for pennyjumpers?
Need "Trader" instances to be very one-directional at times. Execute at a certain time...
In order to simulate a trading day, from 9:30 AM (timestamp 1) to 4pm there are 6 and a half hours. That makes 390 minutes. That makes 23400 seconds. So you should parse out the "randomness" to group into certain periods according to a bell curve.