nep-for New Economics Papers
on Forecasting
Issue of 2013‒02‒03
nine papers chosen by
Rob J Hyndman
Monash University

  1. The out-of-sample forecasting performance of non-linear models of real exchange rate behaviour: The case of the South African Rand By Goodness C. Aye; Mehmet Balcilar; Adel Bosch; Rangan Gupta; Francois Stofberg
  3. Using forecasts to uncover the loss function of FOMC members By Christian Pierdzioch; Jan-Christoph Rülke; Peter Tillmann
  4. Predicting BRICS Stock Returns Using ARFIMA Models By Goodness C. Aye; Mehmet Balcilar; Rangan Gupta; Nicholas Kilimani; Amandine Nakumuryango; Siobhan Redford
  5. On the relation between forecast precision and trading profitability of financial analysts By Carlo Marinelli; Alex Weissensteiner
  6. Optimal Weights and Stress Banking Indexes By Stefano Puddu
  7. Predicting Customer Profitability During Acquisition: Finding the Optimal Combination of Data Source and Data Mining Technique By J. D’HAEN; D. VAN DEN POEL; D. THORLEUCHTER
  8. Kernel Factory: An Ensemble of Kernel Machines By M. BALLINGS; D. VAN DEN POEL
  9. Improving Customer Acquisition Models by Incorporating Spatial Autocorrelation at Different Levels of Granularity By P. BAECKE; D. VAN DEN POEL

  1. By: Goodness C. Aye (Department of Economics, University of Pretoria); Mehmet Balcilar (Department of Economics, Eastern Mediterranean University, Famagusta, North Cyprus,via Mersin 10, Turkey); Adel Bosch (Department of Economics, University of Pretoria); Rangan Gupta (Department of Economics, University of Pretoria); Francois Stofberg (Department of Economics, University of Pretoria)
    Abstract: This paper analyses the out-of-sample forecasting performance of non-linear vs. linear models for the South African Rand against the United States dollar and the British Pound, in real terms. We compare the forecasting performance of point, interval and density forecasts for non-linear Band- TAR and ESTAR models to linear autoregressive models. Our data spans from 1970:01 to 2012:07, and we found that there are no significant gains from using either the Band-TAR or ESTAR non-linear models, compared to the linear AR model in terms of out-of-sample forecasting performance, especially at short horizons. We draw similar conclusions to other literature, and find that for the South African rand against the United States dollar and British pound, non-linearities are too weak for Band-TAR and ESTAR models to estimate.
    Keywords: Real exchange rate; Transaction costs; Band-threshold autoregressive model; Exponential smooth transition autoregressive model; Point forecast; Interval forecast; Density forecast; South Africa
    JEL: C22 C52 C53 F31 F47
    Date: 2013–01
  2. By: Riane de Bruyn (Department of Economics, University of Pretoria); Rangan Gupta (Department of Economics, University of Pretoria); Renee van Eyden (Department of Economics, University of Pretoria)
    Abstract: Traditionally, the literature on forecasting exchange rates with many potential predictors have primarily only accounted for parameter uncertainty using Bayesian Model Averaging (BMA). Though BMA-based models of exchange rates tend to outperform the random walk model, we show that when accounting for model uncertainty over and above parameter uncertainty through the use of Dynamic model Averaging (DMA), the gains relative to the random walk model are even bigger. That is, DMA models outperform not only the random walk model, but also the BMA model of exchange rates. We obtain these results based on fifteen potential predictors used to forecast two South African Rand-based exchange rates. In the process, we also unveil variables, which tends to vary over time, that are good predictors of the Rand-Dollar and Rand-Pound exchange rates at different forecasting horizons.
    Keywords: Bayesian, state space models, exchange rates, macroeconomic fundamentals, forecasting
    JEL: C11 C53 F37 F47
    Date: 2013–01
  3. By: Christian Pierdzioch (University of Hamburg); Jan-Christoph Rülke (University of Vallendar); Peter Tillmann (University of Giessen)
    Abstract: We revisit the sources of the bias in Federal Reserve forecasts and assess whether a precautionary motive can explain the forecast bias. In contrast to the existing literature, we use forecasts submitted by individual FOMC members to uncover members' implicit loss function. Our key finding is that the loss function of FOMC members is asymmetric: FOMC members incur a higher loss when they underpredict (overpredict) inflation and unemployment (real GDP) as compared to an overprediction (underprediction) of similar size. Our findings add to the recent controversy on the relative quality of FOMC forecasts compared to staff forecasts. Together with Capistran's (2008) finding of similar asymmetries in Federal Reserve staff forecasts our results suggest that differences in predictive ability do not stem from differences in preferences. This is underlined by our second result: forecasts remain biased even after accepting an asymmetric loss function.
    Keywords: Federal Open Market Committee; forecasting; asymmetric loss function;monetary policy
    JEL: E58 E37 E27
    Date: 2013
  4. By: Goodness C. Aye (Department of Economics, University of Pretoria); Mehmet Balcilar (Department of Economics, Eastern Mediterranean University, Famagusta, North Cyprus,via Mersin 10, Turkey); Rangan Gupta (Department of Economics, University of Pretoria); Nicholas Kilimani (Department of Economics, University of Pretoria); Amandine Nakumuryango (Department of Economics, University of Pretoria); Siobhan Redford (Department of Economics, University of Pretoria)
    Abstract: This paper examines the existence of long memory in daily stock market returns from Brazil, Russia, India, China, and South Africa (BRICS) countries and also attempts to shed light on the efficacy of Autoregressive Fractionally Integrated Moving Average (ARFIMA) models in predicting stock returns. We present evidence which suggests that ARFIMA models estimated using a variety of estimation procedures yield better forecasting results than the non-ARFIMA (AR, MA, ARMA and GARCH) models with regard to prediction of stock returns. These findings hold consistently the different countries whose economies differ in size, nature and sophistication.
    Keywords: Fractional integration, long-memory, stock returns, long-horizon prediction, ARFIMA, BRICS
    JEL: C15 C22 C53
    Date: 2012–12
  5. By: Carlo Marinelli; Alex Weissensteiner
    Abstract: We analyze the relation between earning forecast accuracy and expected profitability of financial analysts. Modeling forecast errors with a multivariate Gaussian distribution, a complete characterization of the payoff of each analyst is provided. In particular, closed-form expressions for the probability density function, for the expectation, and, more generally, for moments of all orders are obtained. Our analysis shows that the relationship between forecast precision and trading profitability need not to be monotonic, and that, for any analyst, the impact on his expected payoff of the correlation between his forecasts and those of the other market participants depends on the accuracy of his signals. Furthermore, our model accommodates a unique full-communication equilibrium in the sense of Radner (1979): if all information is reflected in the market price, then the expected payoff of all market participants is equal to zero.
    Date: 2013–01
  6. By: Stefano Puddu (Institute of economic research IRENE, Faculty of Economics, University of Neuchâtel, Switzerland)
    Abstract: The goal of this paper is to provide alternative approaches to generate indexes in order to assess banking distress. Specifically, we focus on two groups of indexes that are based on the signalling approach and on the zero in ated Poisson models. The results show that the indexes based on these approaches perform better than those constructed by using the variance-equal and the factor analysis methods. Specifically, they are better at capturing relevant events, signalling distress episodes and forecasting properties. The importance of this study is two-fold: first, we contribute extra information that can be useful for forecasting banking system soundness in the aim of preventing future financial crises; second we provide alternative methods for measuring banking distress.
    Keywords: Stress-banking indexes, Signalling approach, Limited dependent variable methods
    JEL: C16 C25 G21 G33 G34
    Date: 2013–01
    Abstract: The customer acquisition process is generally a stressful undertaking for sales representatives. Luckily there are models that assist them in selecting the ‘right’ leads to pursue. Two factors play a role in this process: the probability of converting into a customer and the profitability once the lead is in fact a customer. This paper focuses on the latter. It makes two main contributions to the existing literature. Firstly, it investigates the predictive performance of two types of data: web data and commercially available data. The aim is to find out which of these two have the highest accuracy as input predictor for profitability and to research if they improve accuracy even more when combined. Secondly, the predictive performance of different data mining techniques is investigated. Results show that bagged decision trees are consistently higher in accuracy. Web data is better in predicting profitability than commercial data, but combining both is even better. The added value of commercial data is, although statistically significant, fairly limited.
    Keywords: marketing analytics; predictive analytics, data source; b2b; web mining; web crawling; bagging; profitability; customer acquisition; external commercial data
    Date: 2012–10
    Abstract: We propose an ensemble method for kernel machines. The training data is randomly split into a number of mutually exclusive partitions defined by a row and column parameter. Each partition forms an input space and is transformed by a kernel function into a kernel matrix K. Subsequently, each K is used as training data for a base binary classifier (Random Forest). This results in a number of predictions equal to the number of partitions. A weighted average combines the predictions into one final prediction. To optimize the weights, a genetic algorithm is used. This approach has the advantage of simultaneously promoting (1) diversity, (2) accuracy, and (3) computational speed. (1) Diversity is fostered because the individual K’s are based on a subset of features and observations, (2) accuracy is sought by optimizing the weights with the genetic algorithm, and (3) computational speed is obtained because the computation of each K can be parallelized. Using five times two-fold cross validation we benchmark the classification performance of Kernel Factory against Random Forest and Kernel-Induced Random Forest (KIRF). We find that Kernel Factory has significantly better performance than Kernel-Induced Random Forest. When the right kernel is specified Kernel Factory is also significantly better than Random Forest. In addition, an open-source Rsoftware package of the algorithm (kernelFactory) is available from CRAN.
    Date: 2012–12
    Abstract: Traditional CRM models often ignore the correlation that could exist among the purchasing behavior of surrounding prospects. Hence, a generalized linear autologistic regression model can be used to capture this interdependence and improve the predictive performance of the model. In particular, customer acquisition models can benefit from this. These models often suffer from a lack of data quality due to the limited amount of information available about potential new customers. Based on a customer acquisition model of a Japanese automobile brand, this study shows that the extra value resulting from incorporating neighborhood effects can vary significantly depending on the granularity level on which the neighborhoods are composed. A model based on a granularity level that is too coarse or too fine will incorporate too much or too little interdependence resulting in a less than optimal predictive improvement. Since neighborhood effects can have several sources (i.e. social influence, homophily and exogeneous shocks), this study suggests that the autocorrelation can be divided into several parts, each optimally measured at a different level of granularity. Therefore, a model is introduced that simultaneously incorporates multiple levels of granularity resulting in even more accurate predictions. Further, the effect of the sample size is examined. This showed that including spatial interdependence using finer levels of granularity is only useful when enough data is available to construct reliable spatial lag effects. As a result, extending a spatial model with multiple granularity levels becomes increasingly valuable when the data sample becomes larger.
    Keywords: Customer Relationship Management (CRM); Predictive Analytics; Customer Intelligence; Marketing; Data Augmentation; Autoregressive Model; Automobile Industry
    Date: 2012–10

This nep-for issue is ©2013 by Rob J Hyndman. It is provided as is without any express or implied warranty. It may be freely redistributed in whole or in part for any purpose. If distributed in part, please include this notice.
General information on the NEP project can be found at For comments please write to the director of NEP, Marco Novarese at <>. Put “NEP” in the subject, otherwise your mail may be rejected.
NEP’s infrastructure is sponsored by the School of Economics and Finance of Massey University in New Zealand.