Ensemble System Development is a framework for researching, building, and testing trading strategies. The 'ensemble' in the name is a legacy as originally the packages main purpose was to investigate the performance of groups of strategies. While this is still possible in the framework, it is now more of a general purpose set of research tools.

• Assess weighted vs unweighted positions
• Variable stop sizing (i.e. with position size)
• ETD vs Max Drawdown, Volatility adjusted MAE plot
• Portfolio events for logging
• Review summary_report
• Fix Fuzzer multiprocessing

• Investigate drawdown profile for moves of a certain magnitude
• Neural network trial
• Getting more data
  • Extraction from pdfs
  • Using simplified fundamentals (i.e. CMC summary data)
• Define system testing process
  • In-sample exploration
    • Subset NYSE data
    • Parameter sensitivity
    • Filters
    • Use of stops
    • Position weighting
    • Cross-validation across market constituents
    • Performance report
    • Equity curve
  • Portfolio testing
    • Effect of starting capital
    • Cross-validation across market constituents
    • Different money management strategies
  • Out of sample testing
    • Extended NYSE data
    • ASX data

• DONE - Define process for strategy execution
• DONE - Construct strategy with measure parameter set.
• DONE - Construct strategy with model parameter set
• DONE - Construct strategy with model and measure parameter sets
• DONE - Clean up modules names and associated classes

• DONE - Replace DataFrame and Panel builds with helper method constructions
• DONE - Weight for positions to be renamed to PositionSelector
• DONE - Move ind_timing from Measure to Strategy. Have strategy return required prices.
• DONE - Ensemble Strategy to contain sub-strategies instead of collections of forecasts etc. *Speed up forecast window method. *Maybe change strategy initialisation to require market input. *Abstract out a data container object for use by market, models etc.

• DONE - Indicator class definition
• DONE - Indicator implements getitem method to allow index by ticker
• DONE - Supply indicator data when initiating
• DONE - PositionModel must return Position object
• DONE - Strategy to provide strat returns series.

• DONE - Add ability to plot long only, and short only results
• DONE - Strategy does not redo existing components on initialise *Plot long and short results together *Calculate performance metrics
• DONE - Strategy to produce label for self

*Ensemble forecast mean to handle missing values.

• DONE - Indicator levels are appropriate data type (e.g. string)
• DONE - Testing for IndicatorMeasure
• DONE - Measure object must implement ind_timing parameter
• DONE - Confirm Forecast data is same content (ticker and dates) as input

• DONE - Change BlockForecaster execute method not require ticker
• DONE - Change Crossover execute method not require ticker
• DONE - PositionModel expects Forecast, whereas Strategy provides itself when calling model.
• DONE - Strategy.indicator returns DataFrame: conflict between object and need for indicator lag.
• DONE - Forecast optF method is generating infinities
• DONE - Position model is not able to handle NaN values
• DONE - Indicator.levels doesn't handle NaNs
• DONE - Returns shouldn't be shifted after initial calculation within Market.
• DONE - Indicator date shifts are incorrect.
• DONE - Positions need to be shifted when calculating strat returns.