We consider tests of forecast encompassing for probability forecasts, for both quadratic and logarithmic scoring rules. We propose test statistics for the null of forecast encompassing, present the limiting distributions of the test statistics, and investigate the impact of estimating the forecasting models’ parameters on these distributions. The small-sample performance is investigated, in ter...

We construct an arti cial neural network to act as a system of leading indicators. We focus on radial basis functions as the architecture and forward selection as the method for determining the number of basis functions in the network. A brief review is given of the advantages of this as a strategy. Using common heuristics to determine scaling, radii and centre population, we nd that the result...

The use of log-transformed data has become standard in macroeconomic forecasting with VAR models. However, its appropriateness in the context of out-of-sample forecasts has not yet been exposed to a thorough empirical investigation. With the aim of filling this void, a broad sample of VAR models is employed in a multi-country setup and approximately 16 Mio. pseudo-out-of-sample forecasts are ev...

We propose new forecast combination schemes for predicting turning points of business cycles. The combination schemes deal with the forecasting performance of a given set of models and possibly providing better turning point predictions. We consider turning point predictions generated by autoregressive (AR) and Markov-Switching AR models, which are commonly used for business cycle analysis. In ...

In the exponential smoothing approach to forecasting, restrictions are often imposed on the smoothing parameters which ensure that certain components are exponentially weighted averages. In this paper, a new general restriction is derived on the basis that the one-step ahead prediction error can be decomposed into permanent and transient components. It is found that this general restriction red...

Density forecast combinations are becoming increasingly popular as a means of improving forecast ‘accuracy’, as measured by a scoring rule. In this paper we generalise this literature by letting the combination weights follow more general schemes. Sieve estimation is used to optimise the score of the generalised density combination where the combination weights depend on the variable one is try...

We generalise the Black-Litterman (BL) portfolio management framework to incorporate time-variation in the conditional distribution of returns in the asset allocation process. We evaluate the performance of the dynamic BL model using both standard performance ratios as well as other measures that are designed to capture tail risk in the presence of non-normally distributed asset returns. We fin...

This paper proposes a computationally efficient algorithm for quantifying the impact of interest-rate risk and longevity risk on the distribution of annuity values in the distant future. The algorithm simulates the state variables out to the end of the horizon period and then uses a Taylor series approximation to compute approximate annuity values at the end of that period, thereby avoiding a c...

The methods used in most SAT validity studies cannot be justified by any sample selection assumptions and are uninformative about the source of the SAT’s predictive power. A new omitted variables estimator is proposed; plausibly consistent estimates of the SAT’s contribution to predictions of University of California freshman grade point averages are about 20% smaller than the usual methods imp...