model0.m

Go to the documentation of this file.

%% @file model0.m
%% @author weaves
%% The first model used for GARCH systems.
%
%  Revision 1.0, Mar-19-2010
%    - File created

% MODEL0
%   obj = model0(spec)
%    creates a container for an MA-GARCH model.
%    This follows the strategy pattern.
%
% Example:
%    spec = garchset('M',1,'P',1,'Q',1);
%    spec = garchset(spec, 'Display','off', 'Distribution', 'T');
%
%    model0_ = model0(spec);
%
%  As new data arrives and a calibrate is requested.
%    model0_.update( collector0_ );
%    disp(model0_.coeff);
%

classdef model0 < handle
  properties (Constant)
    samples = 10;
    epsilon = 1e-10;
    UnityThreshold = 0.999;
  end

  properties (SetAccess = protected, GetAccess = public)
    Description = 'GARCH model container';
    Date = date;   % When started

    Type = 'model0';
    name = 'unassigned';
    
    expiry_;                        % 30 sec ticks
    issane;
    logger_;
    logReturn_;
    volatility0_;
  end

  properties
    spec; % container for the GARCH model
    coeff; % GARCH co-efficients
    returns;
    volatility0;
    logReturn;
    meanForecast0;
    InterpFactor0;
    maxTrend = 1e-5;            % Maximum trend.

    % An indication of the regime.
    volRegime = -1;

    % Indexes into conversion vectors
    idxes = [];
  end

  methods

    %% GARCH model
    % spec_ is the result of a call to garchset().
    function obj = model0(spec_)
      obj.spec = spec_.model;
      obj.maxTrend = spec_.maxTrend;
      obj.name = spec_.name;
      obj.issane = -1;
      obj.logger_ = settings0.instance();
    end

    function delete(obj)
    % delete(obj.spec);
    end
    
    function p0_ = p0(obj, expected_, actual_, volatility0, volf)
      p0_ = [];
      if isempty(obj.coeff)
        return
      end
      % Estimated probability using standard cdf from probability
      % theory and the degrees of freedom of the t-test.
      at_ = (expected_ - actual_ - obj.logReturn) / ...
            volf;
      at_ = (expected_ - actual_) / ...
            volf;
      p0_ = tcdf( at_ / volatility0, obj.coeff.DoF);
      if ~isreal(p0_) || p0_ < 0
        obj.issane = 4;
      end
    end
    
    function update(obj, returner) 
      obj.issane = -1;
      returns = returner.returns0;
      if length(returns) <= obj.samples || ...
            rtolerance(sum(abs(returns)), 0)
        return;
      end
      
      obj.returns = returns;
      
      [coeff] = garchfit(obj.spec, returns);

      if coeff.K <= 0, coeff.K = obj.epsilon; end
      if coeff.GARCH < 0, coeff.GARCH = 0; end
      if coeff.ARCH < 0, coeff.ARCH = 0; end

      % Sum of ARCH + GARCH + 0.5 * Leverage coefficients must be < 1 for
      % GJR(P,Q) models.
      % The current model of choice is MA-GARCH (not GJR) because it
      % scored best among all competing models.
      Sum = coeff.ARCH + coeff.GARCH;
      if ~(Sum < 1)
        obj.UnityThreshold = 0.999;
        coeff.ARCH = obj.UnityThreshold * coeff.ARCH/Sum;
        coeff.GARCH = obj.UnityThreshold * coeff.GARCH/Sum;
      end

      obj.coeff = coeff;
    end
    
  end


end

---