| This dissertation studies density estimation and portfolio selection problems using the maximum entropy (ME) principle. Since an entropy measure turns out to be a distance measure between two distributions, it can be used to estimate unknown density function. Entropy can be also interpreted as a measure of the degree of diversification and thus provides an useful way to construct optimal portfolio weights. In this dissertation three subjects are studied extensively. First, we propose ME autoregressive conditional heteroskedasticity model with demonstrating how we can extract informative functional from the data in the form of moment function. Second, the portfolio selection problem is considered using ME principle. We propose to use cross entropy measure as the objective function (to minimize) with side conditions coming from the mean and variance-covariance matrix of the resampled asset returns. Finally, using ME principle, we provided characterization of some well-known income distributions and flexible parametric income distributions which satisfy certain stylized facts of personal income data. Empirical results showed that maximum entropy principle is quite useful for analyzing economic and financial data. |
