mcmcSolver
Class gibbsLTEGeneralizedQuiet

java.lang.Object
  extended by mcmcSolver.gibbsLTEGeneralizedQuiet

public class gibbsLTEGeneralizedQuiet
extends java.lang.Object

Implementation of the LTE optimization/estimation method proposed in Chernozhukov and Hong (2003). This approach is very useful for optimizing badly-behaved objective functions, and for simultaneously computing the solution and posterior distributions in well-behaved problems. The regularity condition is that the objective function have a limited chi-squared distribution, such as a log-likelihood or a properly weighted GMM objective function.

There are three steps in the algorithm:

  1. Pre-burn
  2. Variance tuning
  3. Final burn

The pre-burn takes a default sampling variance and executes the algorithm. This typically results in a run toward the (local) minimum. Once the minimum is found, the algorithm switches to a variance tuning stage, where the acceptance/rejection ratio is set to somewhere around 0.234 by adjusting the variance of the proposal density. Finally, the actual values are generated in the last stage. There are many options which are largely self-explanatory when cross-referenced against the code. This version ("Quiet") does not contain any mechanisms for reporting results to a GUI or the command line.


Constructor Summary
gibbsLTEGeneralizedQuiet(mcmcFunction function, int draws, int burnPeriod, double[] guess)
          Constructor for LTE class.
 
Method Summary
 Jama.Matrix getChain()
          Returns the MCMC chain generated by the LTE procedure.
 double[] getLowestPoint()
          Returns the lowest point found during the LTE procedure.
 
Methods inherited from class java.lang.Object
clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait
 

Constructor Detail

gibbsLTEGeneralizedQuiet

public gibbsLTEGeneralizedQuiet(mcmcFunction function,
                                int draws,
                                int burnPeriod,
                                double[] guess)
Constructor for LTE class.

Parameters:
function - An object which implements the mcmcFunction interface as the source of the objective function.
draws - Number of final draws to take in the MCMC chain.
burnPeriod - Number of initial draws in the burn-in period.
guess - Initial guess for the unknown parameters.
Method Detail

getChain

public Jama.Matrix getChain()
Returns the MCMC chain generated by the LTE procedure.

Returns:
LTE chain.

getLowestPoint

public double[] getLowestPoint()
Returns the lowest point found during the LTE procedure.

Returns:
Lowest point found in the LTE procedure.