The estimation of the profit and loss distribution of loan portfolios requires the modelling of the marginal and multivariate distributions that describe the individual and joint credit risk of the loans making up a portfolio. Unfortunately, portfolio credit risk measurement suffers from extremely restricted datasets. For many countries, this problem holds at any point in time and through time. It is especially severe when modelling the credit risk of loans given to small and medium size enterprises (SME's) and non-listed firms, which are the focus of my research. Earlier attempts to deal with this problem have resulted in measurement methodologies adopting convenient, but not necessarily properly specified, parametric distributions or simply ignoring the effects of macroeconomic developments in credit risk. In order to improve the measurement of portfolio credit risk at any point in time and through time, I propose new methodologies, i. e. the consistent information density optimising (CIDO), the consistent information multivariate density optimising (CIMDO), and the conditional probability of default (CoPoD) methodologies. CIDO and CIMDO are based on the cross-entropy approach. They recover the marginal and multivariate distributions of the loans that make up a portfolio from the limited information available. Using the Probability Integral Transformation and the Probability Squared Deviations criteria, it is proved that the distributions recovered by CIDO and CIMDO outperform parametric distributions. CoPoD is based on the pure-entropy approach. It allows for the modelling of the probabilities of loan default (PoD's) as functions of macroeconomic variables. The latter represents a challenging task, since the time series of PoD's usually contain few observations; thus making OLS estimation imprecise. CoPoD recovers estimators that show superior robustness to OLS estimators in finite sample settings under the mean square error criterion. CoPoD also incorporates a procedure to select a relevant set of explanatory variables that explain PoD's through time.