DASHBOARD.md

In hughshuwang/isocyanate: isocyanate (R-N=C=O)

Dashboard

• Topics:
• [x] Non-para Inf & Stat Learning
  • [x] Bootstrap resampling & sampling noise measurement
  • [x] Average shifted histogram (ASH)
  • [x] Adaptive Gaussian kernel smoothing density (KernSmooth)
  • [x] PENDING: Bootstrap-based/Density-based adaptive radius, Learning algorithm
  • [x] Fitting CRRA for returns (asymmetric)
  • [x] Define distances (Measure for Modeling Risks and differentiating power)
  • [x] Hierarchical clustering (diff methods)
• [ ] Predictive Signals (Forward Looking)
  • [ ] Momentum/Mean Reverting (systematic and individual/sector)
  • [ ] Volatility clustering (general volatility modeling, systematic and individual/sector)
  • [ ] Market turbulence (both side moderate fluctuation modeling, systematic)
  • [ ] Tail risk modeling and forecasts (extreme downside collapse modeling, systematic)
  • [ ] Factor loadings/Alpha dynamics (individual/sector, Kalman Filter, Regime Identification)
  • [ ] Other Market variable (VIX, IG & HY OAS, etc.)
• [ ] Numerical Simulations
  • [x] Multi-asset T-Copulas for better tail risk fits (for systematic signals)
  • [ ] Conditional correlation structure measurement
  • [ ] Empirical Measure Change (for individual/sector signals)

• [ ] Single-period Convex Stochastic Control

  • [ ] Convex risk-adjusted objective functions
  • [ ] Convex constraint sets
  • [ ] Self financing constraints
  • [ ] Beta constraints for alpha signals

• Signal Generation and Optimization Pipelines:

• NumericSignal(df::1dxts/2dxts, f::ScoringMethod, period::Int64, dimdf::Number)::1dxts{Number} `

• BoolSignal(numsignal::1dxts{Number/Binary}, ngroup::Number=10, f::DiffMethod, cuts::Vector{Number} = collect(0:0.1:1))::Vector{1dxts{Bool}}

  • Default using cuts to cut empirical quantiles (for continuous signal)

• DensityBound(ret::1dxts{Number}, qs::Vector{2, Number})

  • Generate bounds for density estimation, unify the breaks series obtained by density estimation

• EstDensity(idx::1dxts{Bool}, ret::1dxts{Number}, bounds::Vector{2, Number}; rkernel::Number, nbreaks::Number)::Vector/Dict{Number} (more adaptive/resampling methods required, index not lagged)

• BootBand(idx::1dxts{Bool}, ret::1dxts{Number}, bounds::Vector{2, Number})

• NoiseBB(idx::1dxts{Bool}, ret::1dxts{Number}, bounds::Vector{2, Number})

• DensityDist(d1::Vector{Number}, d2::Vector{Number}, measure::Vector{Number})::Number

• DistMatrix(boolsignal::Vector{1dxts{Bool}}, est = EstDensity, dis = DensityDist)::Array{2, Number}

• HieClustering(dist::Array{2, Number}, ngroup::Int64)::Vector{Vector{Int64}; dthreshold::Number}::Number

  • Output a number measuring differentiating power, can be the distance, distance should be comparable for all density functions, then to decide whether this signal can be used or not

• Select signal sets (systematic and individual signals) and obtain conditional distributions, copulas, individual measure change for each asset every day.

hughshuwang/isocyanate documentation built on May 30, 2019, 7:17 a.m.