Examine get_backtest() format and make quantrisk compat

[twiecki]

get_backtest() on the quantopian research platform (QRP) returns backtest results, positions and transactions. For quantrisk to be useful on the QRP, it has to work with that format.

I imagine this could go 2 ways. Either we switch internally to always use that format if it's convenient, or we transform whatever get_backtest() returns.

[twiecki]

The relevant inputs we currently need are:

• daily returns
• positions
• transactions

get_backtest() returns an object that is a simple collection of DataFrames that contain the relevant information. Returns are very easy to get and have the same format we require.

There is one major difference between the formats used for positions and transactions. While we have one index dt per day for positions and then store the positions as a dict, this uses duplicated daily dts as an index, one for each position. The format in QRP is probably better to work with in the long term so I propose we switch internally to use the prior format established by @ssanderson.

[twiecki]

@ssanderson Can you point me to the code where you are building up these data structures? Do you think it's a good idea to borrow that same code here?

[twiecki]

From @ssanderson: "qexec/research/backtest.py, in the constructor of the BacktestResult class" https://github.com/quantopian/qexec/blob/master/qexec/research/backtest.py#L424

[justinlent]

@ssanderson @twiecki do either of you know whether the result from get_backtest() contains an attribute that might contain the date which the algo started live zipline papertrading (if it has)? This could easily allow us to compute both in-sample and out-of-sample stats for algos that both have a backtest+papertrading

[twiecki]

Good point, we'd definitely need that, I don't know the answer though. Maybe a backtest and live algo are different entities for research? So we'd have to stitch them together manually.

[ssanderson]

@justinlent @twiecki the BacktestResult object has start_date and end_date attributes, which should have the correct values for both backtesting and live results.

Note that Q research doesn't currently support accessing live algo results at all. There's a PR from @seongsean that implemented the functionality here: https://github.com/quantopian/qexec/pull/5252. At the time, it was sidelined due to security concerns because we didn't have user_id validation.

[gusgordon]

I believe this does not fully support minute backtest results.

[twiecki]

I checked and I think it should. The positions are daily to begin with and the transaction code also aggregates over full days.