Next to backtesting of trading algorithms, backtesting of prognoses is the other big use case of GG's future backtesting framework. We are looking for a way to generally extend zipline to support prognoses instead of applying a quick fix. For instance, TradingAlgorithm and Prognoses could extend from a common base class Model.
Level of Ready
rolling optimization of "intrinsic" model parameters
global (walk-forward, out-of-sample testing) optimisation of "extrinsic" model parameters
Tasks
Explain how to backtest prognoses using zipline, e.g. "Prognosis" and "TradingAlgorithm" could be extensions of a new common base class "Model"
Discuss the design
store prognosis so algo has access
Implement the new functionality
only pass data with newest archive date
overwrite history with updated data
archive updated prognoses
use target values as benchmark
Level of Done
A prognosis can be optimised according to
an objective function, e.g. prognosis error
an optimization strategy
a selected set of parameters and corresponding parameter spaces
There is a rolling optimisation of "intrinsic" prognosis parameters
There is a global (walk-forward, out-of-sample testing) optimisation of "extrinsic" prognosis parameters
All relevant figures are monitored
A new prognosis action can be triggered by a data event and by a time event
:moneybag: :moneybag: :moneybag: :moneybag: :moneybag: :moneybag: :moneybag: :moneybag:
Purpose
Next to backtesting of trading algorithms, backtesting of prognoses is the other big use case of GG's future backtesting framework. We are looking for a way to generally extend zipline to support prognoses instead of applying a quick fix. For instance, TradingAlgorithm and Prognoses could extend from a common base class Model.
Level of Ready
Tasks
Level of Done