Comprehensive Testing of the Present Value Formulation

[jrhea]

The idea here is to implement a methodical way of comparing the Present Value formulation to the various ways a set of trades could close. To do this, we can think of each open trade as having certain attributes: size (as a percent of liquidity), open/close, % mature, etc. We then close them in different orderings based on how they are sorted. For example, the following matrix would be a full factorial set of orderings to test against against the Present Value:

• size (high to low), % mature ( high to low), long/short
• size (high to low), % mature ( high to low), short/long
• size (high to low), % mature ( low to high), long/short
• size (high to low), % mature ( low to high), short/long
• size (low to high), % mature ( high to low), long/short
• size (low to high), % mature ( high to low), short/long
• size (low to high), % mature ( low to high), long/short
• size (low to high), % mature ( low to high), short/long

Each trial would contain a decent sample size (probably at least 30) of trades and test the full factorial set of orderings against the Present Value. After running a few hundred trials, we should be able to determine what (if any) attributes cause the largest deviation from the Present Value

[wakamex]

what if we... close them all at the same time? 🤯

[dpaiton]

PV is expected to be fairly invariant to these changes now, so we will add PV checks with a tolerance to the path independence interactive test notebook instead of writing up a new analysis notebook to do Jonny's original experiment. Issue here: delvtech/agent0#1123