Terms Of In-sample Fit Research Articles

Binary Pattern Recognition in the Presence of Correlated Multiple Dependent Variables

In binary spatial pattern recognition, there are many situations where the researcher could be interested in a number of dependent variables that are themselves correlated. For instance, different types of crime often coexist in the same area, or different species could share the same habitat. In cases like these, a natural correlation exists amongst the dependent variables of interest and is informative for spatial probability mapping. Weights of evidence (WE) modelling is a popular Bayesian probability method for binary pattern recognition, but it only deals with one single dependent variable at a time and ignores the correlation between the dependent variables. In this article, a multiple dependent variable weights of evidence (MDVWE) model will be developed. It will be shown that the new MDVWE model can be viewed as a restricted version of the conditional dependence-adjusted weights of evidence (CDAWE) model of Deng (Nat Resour Res 18(4):249–258, 2009). The MDVWE model is easy to program and implement. By means of a simulation study, it will be shown that the MDVWE model outperforms the traditional WE model both in terms of in-sample fit and out-of-sample prediction accuracy.

An Ordered Weights of Evidence Model for Ordered Discrete Variables

The standard Weights of Evidence (WE) model produces probability estimates for the presence of binary events. However, in many empirical studies the discrete event of interest can take on ordered values. For instance, the presence of mineral deposits may be classified further into different grades. In this paper, a new Ordered Weights of Evidence (OWE) model will be developed. Borrowing the conceptual framework of the latent variable interpretation of the standard Ordered Logistic Regression (OLR) model, the OWE can produce probability estimates for the presence of ordered discrete events. It will be shown that the OWE is computationally less intensive than the OLR. Through a simulation study, it will be shown that the OWE is comparable to the OLR both in terms of in-sample fit and out-of-sample forecasts.

A Spatially Autocorrelated Weights of Evidence Model

One of the most important features of spatial datasets is that they often exhibit spatial autocorrelation, where locational similarities are observed jointly with similarities in values. Both logistic regression (LR) modelling and weights of evidence (WE) modelling are methods commonly applied in binary pattern recognition. While a spatially autocorrelated variant of the LR model, the so-called autologistic regression (ALR) model, exists in the literature, a spatially autocorrelated variant of the WE model does not exist. In this paper, a spatially autocorrelated weights of evidence (SACWE) model will be proposed. It will be demonstrated that the new model contains the same amount of spatial information as does an ALR model, and it is easy to program and implement. Via a simulation study, it will be shown that, in the presence of spatial autocorrelation, both in terms of in-sample fit and out-of-sample predictions the SACWE model is on par with the ALR model, while significantly outperforming the conventional WE model.

Measuring U.S. International Relative Prices: A WARP View of the World

In this paper we construct a new measure of U.S. prices relative to those of its trading partners and use it to reexamine the behavior of U.S. net exports. Our measure differs from existing measures of the dollar's real effective exchange rate (REER) in that it explicitly incorporates both the difference in price levels between the United States and developing economies and the growing importance of these developing economies in world trade. Unlike existing REERs, our measure shows that relative U.S. prices have increased significantly over the past 15 years. In terms of simple correlations, the relationship between our measure of relative prices and U.S. net exports is much more coherent than that between existing REERs and net exports. To explore this relationship further, we use our measure to construct an index of foreign prices relevant for U.S. export volumes and reexamine several export equations. We find that export equations with the new index dominate those with previous measures in terms of in-sample fit, out-of-sample fit, and parameter constancy. In addition, we find that with the new index of foreign prices the estimated elasticity of U.S. exports with respect to foreign income is a good bit higher than the unitary elasticity found in previous studies using other price measures. This has implications for U.S. current account adjustment.

Measuring U.S. International Relative Prices: A Warp View of the World

In this paper we construct a new measure of U.S. prices relative to those of its trading partners and use it to reexamine the behavior of U.S. net exports. Our measure differs from existing measures of the dollar's real effective exchange rate (REER) in that it explicitly incorporates both the difference in price levels between the United States and developing economies and the growing importance of these developing economies in world trade. Unlike existing REERs, our measure shows that relative U.S. prices have increased significantly over the past 15 years. In terms of simple correlations, the relationship between our measure of relative prices and U.S. net exports is much more coherent than that between existing REERs and net exports. To explore this relationship further, we use our measure to construct an index of foreign prices relevant for U.S. export volumes and reexamine several export equations. We find that export equations with the new index dominate those with previous measures in terms of in-sample fit, out-of-sample fit, and parameter constancy. In addition, we find that with the new index of foreign prices the estimated elasticity of U.S. exports with respect to foreign income is a good bit higher than the unitary elasticity found in previous studies using other price measures. This has implications for U.S. current account adjustment.

Predicting returns on Canadian exchange rates with artificial neural networks and EGARCH-M models

This paper investigates the problem of predicting daily returns based on five Canadian exchange rates using artificial neural networks and EGARCH-M models. First, the statistical properties of five daily exchange rate series (US Dollar, German Mark, French Franc, Japanese Yen and British Pound) are analysed. EGARCH-M models on the Generalised Error Distribution (GED) are fitted to the return series, and serve as comparison standards, along with random walk models. Second, backpropagation networks (BPN) using lagged returns as inputs are trained and tested. Estimated volatilities from the EGARCH-M models are used also as inputs to see if performance is affected. The question of spillovers in interrelated markets is investigated with networks of multiple inputs and outputs. In addition, Elman-type recurrent networks are also trained and tested. Comparison of the various methods suggests that, despite their simplicity, neural networks are similar to the EGARCH-M class of nonlinear models, but superior to random walk models, in terms of insample fit and out-of-sample prediction performance.