A full asset allocation strategy could make use of the following sources of information:

• stochastic representation of asset returns and their future evolution
• asset prices: converting fractional weights to integers
• investment history
  • turnover minimization
  • historic portfolio performance and risks taken
  • taxable gains / losses for tax optimization
• cash-flow data
  • initial investment
  • deposit in / out
• client data: tax allowance, tax classes
• partial execution
  • live trading with feedback loop
• unknown unknowns?

Different strategies require different granularity of overall information. All strategies are subtypes of abstract type FullStrategy.

Fix weights regardless of market data.

Optimizes portfolio with respect to single period. Turnover problematic becomes irrelevant: only asset return model as additional information required.

optimizeWgts(strat::InitialStrategy, univ::UniverseModel)

Multi-period setting: portfolio rebalancing with turnover maximization. Specifies initial investment strategy, rebalancing strategy and turnover filter. Doesn't deal with weight discreteness, cash-flows or taxes.

optimizeWgts(strat::??, univ::UniverseModel, invHistory::Investments)

Estimate model for each date in order to get insights into variation in asset return distributions. If possible, array of estimated models is simplified to more concise output.

applyMuSigmaModelEstimator

Output: array of mus, sigmas and correlations, with values only for those days with enough data for model estimation.

Recursive.

Output: investments, expected portfolio properties

• estimator: way of getting concrete specification of asset return model
• model: complete specification of asset return model
• universe: complete specification of asset return model together with information to redo estimation process (for resampling techniques to deal with estimation uncertainty)
  • model
  • estimator
  • data
• strategy: investment rules of different granularity
• investments: series of resulting portfolio weights