Non‐linear forecasts of stock returns

Following recent non‐linear extensions of the present‐value model, this paper examines the out‐of‐sample forecast performance of two parametric and two non‐parametric nonlinear models of stock returns. The parametric models include the standard regime switching and the Markov regime switching, whereas the non‐parametric are the nearest‐neighbour and the artificial neural network models. We focused on the US stock market using annual observations spanning the period 1872–1999. Evaluation of forecasts was based on two criteria, namely forecast accuracy and forecast encompassing. In terms of accuracy, the Markov and the artificial neural network models produce at least as accurate forecasts as the other models. In terms of encompassing, the Markov model outperforms all the others. Overall, both criteria suggest that the Markov regime switching model is the most preferable non‐linear empirical extension of the present‐value model for out‐of‐sample stock return forecasting. Copyright © 2003 John Wiley & Sons, Ltd.     

Quantile estimators with orthogonal pinball loss function

To guarantee stable quantile estimations even for noisy data, a novel loss function and novel quantile estimators are developed, by introducing the effective concept of orthogonal loss considering the noise in both response and explanatory variables. In particular, the pinball loss used in classical quantile estimators is improved into novel orthogonal pinball loss (OPL) by replacing vertical loss by orthogonal loss. Accordingly, linear quantile regression (QR) and support vector machine quantile regression (SVMQR) can be respectively extended into novel OPL‐based QR and OPL‐based SVMQR models. The empirical study on 10 publicly available datasets statistically verifies the superiority of the two OPL‐based models over their respective original forms in terms of prediction accuracy and quantile property, especially for extreme quantiles. Furthermore, the novel OPL‐based SVMQR model with both OPL and artificial intelligence (AI) outperforms all benchmark models, which can be used as a promising quantile estimator, especially for noisy data.     

Forecasting volatility with support vector machine‐based GARCH model

Recently, support vector machine (SVM), a novel artificial neural network (ANN), has been successfully used for financial forecasting. This paper deals with the application of SVM in volatility forecasting under the GARCH framework, the performance of which is compared with simple moving average, standard GARCH, nonlinear EGARCH and traditional ANN‐GARCH models by using two evaluation measures and robust Diebold–Mariano tests. The real data used in this study are daily GBP exchange rates and NYSE composite index. Empirical results from both simulation and real data reveal that, under a recursive forecasting scheme, SVM‐GARCH models significantly outperform the competing models in most situations of one‐period‐ahead volatility forecasting, which confirms the theoretical advantage of SVM. The standard GARCH model also performs well in the case of normality and large sample size, while EGARCH model is good at forecasting volatility under the high skewed distribution. The sensitivity analysis to choose SVM parameters and cross‐validation to determine the stopping point of the recurrent SVM procedure are also examined in this study. Copyright © 2009 John Wiley & Sons, Ltd.     

Unemployment variation over the business cycles: a comparison of forecasting models

Asymmetry has been well documented in the business cycle literature. The asymmetric business cycle suggests that major macroeconomic series, such as a country's unemployment rate, are non‐linear and, therefore, the use of linear models to explain their behaviour and forecast their future values may not be appropriate. Many researchers have focused on providing evidence for the non‐linearity in the unemployment series. Only recently have there been some developments in applying non‐linear models to estimate and forecast unemployment rates. A major concern of non‐linear modelling is the model specification problem; it is very hard to test all possible non‐linear specifications, and to select the most appropriate specification for a particular model. Artificial neural network (ANN) models provide a solution to the difficulty of forecasting unemployment over the asymmetric business cycle. ANN models are non‐linear, do not rely upon the classical regression assumptions, are capable of learning the structure of all kinds of patterns in a data set with a specified degree of accuracy, and can then use this structure to forecast future values of the data. In this paper, we apply two ANN models, a back‐propagation model and a generalized regression neural network model to estimate and forecast post‐war aggregate unemployment rates in the USA, Canada, UK, France and Japan. We compare the out‐of‐sample forecast results obtained by the ANN models with those obtained by several linear and non‐linear times series models currently used in the literature. It is shown that the artificial neural network models are able to forecast the unemployment series as well as, and in some cases better than, the other univariate econometrics time series models in our test. Copyright © 2004 John Wiley & Sons, Ltd.     

Economic factors and the stock market: a new perspective

It has been widely accepted that many financial and economic variables are non‐linear, and neural networks can model flexible linear or non‐linear relationships among variables. The present paper deals with an important issue: Can the many studies in the finance literature evidencing predictability of stock returns by means of linear regression be improved by a neural network? We show that the predictive accuracy can be improved by a neural network, and the results largely hold out‐of‐sample. Both the neural network and linear forecasts show significant market timing ability. While the switching portfolio based on the linear forecasts outperforms the buy‐and‐hold market portfolio under all three transaction cost scenarios, the switching portfolio based on the neural network forecasts beats the market only if there is no transaction cost. Copyright © 1999 John Wiley & Sons, Ltd.     

Non‐linear,non‐parametric,non‐fundamental exchange rate forecasting

This paper employs a non‐parametric method to forecast high‐frequency Canadian/US dollar exchange rate. The introduction of a microstructure variable, order flow, substantially improves the predictive power of both linear and non‐linear models. The non‐linear models outperform random walk and linear models based on a number of recursive out‐of‐sample forecasts. Two main criteria that are applied to evaluate model performance are root mean squared error (RMSE) and the ability to predict the direction of exchange rate moves. The artificial neural network (ANN) model is consistently better in RMSE to random walk and linear models for the various out‐of‐sample set sizes. Moreover, ANN performs better than other models in terms of percentage of correctly predicted exchange rate changes. The empirical results suggest that optimal ANN architecture is superior to random walk and any linear competing model for high‐frequency exchange rate forecasting. Copyright © 2006 John Wiley & Sons, Ltd.     

Forecasting Australia's real house price index: A comparison of time series and machine learning methods

We employ 47 different algorithms to forecast Australian log real house prices and growth rates, and compare their ability to produce accurate out-of-sample predictions. The algorithms, which are specified in both single- and multi-equation frameworks, consist of traditional time series models, machine learning (ML) procedures, and deep learning neural networks. A method is adopted to compute iterated multistep forecasts from nonlinear ML specifications. While the rankings of forecast accuracy depend on the length of the forecast horizon, as well as on the choice of the dependent variable (log price or growth rate), a few generalizations can be made. For one- and two-quarter-ahead forecasts we find a large number of algorithms that outperform the random walk with drift benchmark. We also report several such outperformances at longer horizons of four and eight quarters, although these are not statistically significant at any conventional level. Six of the eight top forecasts (4 horizons × 2 dependent variables) are generated by the same algorithm, namely a linear support vector regressor (SVR). The other two highest ranked forecasts are produced as simple mean forecast combinations. Linear autoregressive moving average and vector autoregression models produce accurate olne-quarter-ahead predictions, while forecasts generated by deep learning nets rank well across medium and long forecast horizons.     

The Informational Content of the Term Spread in Forecasting the US Inflation Rate: A Nonlinear Approach

The difficulty in modelling inflation and the significance in discovering the underlying data‐generating process of inflation is expressed in an extensive literature regarding inflation forecasting. In this paper we evaluate nonlinear machine learning and econometric methodologies in forecasting US inflation based on autoregressive and structural models of the term structure. We employ two nonlinear methodologies: the econometric least absolute shrinkage and selection operator (LASSO) and the machine‐learning support vector regression (SVR) method. The SVR has never been used before in inflation forecasting considering the term spread as a regressor. In doing so, we use a long monthly dataset spanning the period 1871:1–2015:3 that covers the entire history of inflation in the US economy. For comparison purposes we also use ordinary least squares regression models as a benchmark. In order to evaluate the contribution of the term spread in inflation forecasting in different time periods, we measure the out‐of‐sample forecasting performance of all models using rolling window regressions. Considering various forecasting horizons, the empirical evidence suggests that the structural models do not outperform the autoregressive ones, regardless of the model's method. Thus we conclude that the term spread models are not more accurate than autoregressive models in inflation forecasting. Copyright © 2016 John Wiley & Sons, Ltd.     

A hybrid model considering cointegration for interval-valued pork price forecasting in China

Compared with point forecasting, interval forecasting is believed to be more effective and helpful in decision making, as it provides more information about the data generation process. Based on the well-established “linear and nonlinear” modeling framework, a hybrid model is proposed by coupling the vector error correction model (VECM) with artificial intelligence models which consider the cointegration relationship between the lower and upper bounds (Coin-AIs). VECM is first employed to fit the original time series with the residual error series modeled by Coin-AIs. Using pork price as a research sample, the empirical results statistically confirm the superiority of the proposed VECM-CoinAIs over other competing models, which include six single models and six hybrid models. This result suggests that considering the cointegration relationship is a workable direction for improving the forecast performance of the interval-valued time series. Moreover, with a reasonable data transformation process, interval forecasting is proven to be more accurate than point forecasting.     

Conditional predictability of daily exchange rates

At what forecast horizon is one time series more predictable than another? This paper applies the Diebold–Kilian conditional predictability measure to assess the out‐of‐sample performance of three alternative models of daily GBP/USD and DEM/USD exchange rate returns. Predictability is defined as a non‐linear statistic of a model's relative expected losses at short and long forecast horizons, allowing flexible choice of both the estimation procedure and loss function. The long horizon is set to 2 weeks and one month ahead and forecasts evaluated according to MSE loss. Bootstrap methodology is used to estimate the data's conditional predictability using GARCH models. This is then compared to predictability under a random walk and a model using the prediction bias in uncovered interest parity (UIP). We find that both exchange rates are less predictable using GARCH than using a random walk, but they are more predictable using UIP than a random walk. Predictability using GARCH is relatively higher for the 2‐weeks‐than for the 1‐month long forecast horizon. Comparing the results using a random walk to that using UIP reveals ‘pockets’ of predictability, that is, particular short horizons for which predictability using the random walk exceeds that using UIP, or vice versa. Overall, GBP/USD returns appear more predictable than DEM/USD returns at short horizons. Copyright © 2002 John Wiley & Sons, Ltd.     

Forecasting daily foreign exchange rates using genetically optimized neural networks

Forecasting currency exchange rates is an important financial problem that has received much attention especially because of its intrinsic difficulty and practical applications. The statistical distribution of foreign exchange rates and their linear unpredictability are recurrent themes in the literature of international finance. Failure of various structural econometric models and models based on linear time series techniques to deliver superior forecasts to the simplest of all models, the simple random walk model, have prompted researchers to use various non‐linear techniques. A number of non‐linear time series models have been proposed in the recent past for obtaining accurate prediction results, in an attempt to ameliorate the performance of simple random walk models. In this paper, we use a hybrid artificial intelligence method, based on neural network and genetic algorithm for modelling daily foreign exchange rates. A detailed comparison of the proposed method with non‐linear statistical models is also performed. The results indicate superior performance of the proposed method as compared to the traditional non‐linear time series techniques and also fixed‐geometry neural network models. Copyright © 2002 John Wiley & Sons, Ltd.     

Technological forecasting with nonlinear models

The S-shaped growth curves such as Gompertz, logistic, normal and Weibuli are widely used for forecasting technological substitutions. A family of data-based transformed (DBT) models, which are linear in the regression parameters, including the above-mentioned four models as special cases has been shown to be quite useful for short-term forecasts. This paper explores modeling the technology penetration data directly with assumed S-shaped growth curves. The resulting models, which are nonlinear in the regression parameters, also incorporate proper dependence structure and power transformation. It appears that the nonlinear modeling is a viable alternative to the DBT and other conventional forecasting models in forecasting technological substitutions. Hence, an appropriate strategy is to consider the nonlinear modeling approaches as possible alternatives and use the data at hand to select, via pseudo-cross-validation, the best model for forecasting purposes.     

Building neural network models for time series: a statistical approach

This paper is concerned with modelling time series by single hidden layer feedforward neural network models. A coherent modelling strategy based on statistical inference is presented. Variable selection is carried out using simple existing techniques. The problem of selecting the number of hidden units is solved by sequentially applying Lagrange multiplier type tests, with the aim of avoiding the estimation of unidentified models. Misspecification tests are derived for evaluating an estimated neural network model. All the tests are entirely based on auxiliary regressions and are easily implemented. A small‐sample simulation experiment is carried out to show how the proposed modelling strategy works and how the misspecification tests behave in small samples. Two applications to real time series, one univariate and the other multivariate, are considered as well. Sets of one‐step‐ahead forecasts are constructed and forecast accuracy is compared with that of other nonlinear models applied to the same series. Copyright © 2006 John Wiley & Sons, Ltd.     

A generalized regression model based on hybrid empirical mode decomposition and support vector regression with back-propagation neural network for mid-short-term load forecasting

Since load forecasting plays a decisive role in the safe and stable operation of power systems, it is particularly important to explore forecasting methods accurately. In this article, the hybrid empirical mode decomposition (EMD) and support vector regression (SVR) with back-propagation neural network (BPNN), namely the EMDHR-SVR-BPNN model, is proposed. Information theory is mainly used to solve the data tendency problem, and the EMD method is used to solve the data volatility problem. There is no interaction between these two methods; thus these two models can complement each other through generalized regression of orthogonal decomposition. Taking the load data from the New South Wales (NSW, Australia) market as an example, the obtained simulation results are compared with other models. It is concluded that the proposed EMDHR-SVR-BPNN model not only improves the forecasting accuracy but also has good fitting ability. It can reflect the changing tendency of data in a timely manner, providing a strong basis for the electricity generation of the power sector in the future, thus reducing electricity waste. The proposed EMDHR-SVR-BPNN model has potential for employment in mid-short term load forecasting.     

An intelligent model selection and forecasting system

In this paper we present an intelligent decision‐support system based on neural network technology for model selection and forecasting. While most of the literature on the application of neural networks in forecasting addresses the use of neural network technology as an alternative forecasting tool, limited research has focused on its use for selection of forecasting methods based on time‐series characteristics. In this research, a neural network‐based decision support system is presented as a method for forecast model selection. The neural network approach provides a framework for directly incorporating time‐series characteristics into the model‐selection phase. Using a neural network, a forecasting group is initially selected for a given data set, based on a set of time‐series characteristics. Then, using an additional neural network, a specific forecasting method is selected from a pool of three candidate methods. The results of training and testing of the networks are presented along with conclusions. Copyright © 1999 John Wiley & Sons, Ltd.     

Traditional versus novel forecasting techniques: how much do we gain?

This article applies two novel techniques to forecast the value of US manufacturing shipments over the period 1956–2000: wavelets and support vector machines (SVM). Wavelets have become increasingly popular in the fields of economics and finance in recent years, whereas SVM has emerged as a more user‐friendly alternative to artificial neural networks. These two methodologies are compared with two well‐known time series techniques: multiplicative seasonal autoregressive integrated moving average (ARIMA) and unobserved components (UC). Based on forecasting accuracy and encompassing tests, and forecasting combination, we conclude that UC and ARIMA generally outperform wavelets and SVM. However, in some cases the latter provide valuable forecasting information that it is not contained in the former. Copyright © 2008 John Wiley & Sons, Ltd.     

A deep residual compensation extreme learning machine and applications

The extreme learning machine (ELM) is a type of machine learning algorithm for training a single hidden layer feedforward neural network. Randomly initializing the weight between the input layer and the hidden layer and the threshold of each hidden layer neuron, the weight matrix of the hidden layer can be calculated by the least squares method. The efficient learning ability in ELM makes it widely applicable in classification, regression, and more. However, owing to some unutilized information in the residual, there are relatively huge prediction errors involving ELM. In this paper, a deep residual compensation extreme learning machine model (DRC-ELM) of multilayer structures applied to regression is presented. The first layer is the basic ELM layer, which helps in obtaining an approximation of the objective function by learning the characteristics of the sample. The other layers are the residual compensation layers in which the learned residual is corrected layer by layer to the predicted value obtained in the previous layer by constructing a feature mapping between the input layer and the output of the upper layer. This model is applied to two practical problems: gold price forecasting and airfoil self-noise prediction. We used the DRC-ELM with 50, 100, and 200 residual compensation layers respectively for experiments, which show that DRC-ELM does better in generalization and robustness than classical ELM, improved ELM models such as GA-RELM and OS-ELM, and other traditional machine learning algorithms such as support vector machine (SVM) and back-propagation neural network (BPNN).     

改进果蝇优化算法优化广义回归神经网络的预测问题研究

为提高传统非线性预测模型的预测精度,提出一种基于改进果蝇优化算法优化广义回归神经网络的预测方法,将果蝇群体分两部分分别进行迭代寻优,从而改进了果蝇优化算法的寻优性能,进而避免了在寻优过程中陷入局部最优。该方法利用改进果蝇优化算法优化广义回归神经网络的径向基函数扩展参数,然后用训练好的广义回归神经网络预测模型进行预测,最后通过订单预测算例进行实证研究。实证研究结果显示,该方法在解决订单预测问题中与未改进的果蝇优化算法优化广义回归神经网络和传统的广义回归神经网络方法对比,具有更高的预测精度和更好的非线性拟合能力。     

Forecasting US interest rates and business cycle with a nonlinear regime switching VAR model

This paper introduces a regime switching vector autoregressive model with time‐varying regime probabilities, where the regime switching dynamics is described by an observable binary response variable predicted simultaneously with the variables subject to regime changes. Dependence on the observed binary variable distinguishes the model from various previously proposed multivariate regime switching models, facilitating a handy simulation‐based multistep forecasting method. An empirical application shows a strong bidirectional predictive linkage between US interest rates and NBER business cycle recession and expansion periods. Due to the predictability of the business cycle regimes, the proposed model yields superior out‐of‐sample forecasts of the US short‐term interest rate and the term spread compared with the linear and nonlinear vector autoregressive (VAR) models, including the Markov switching VAR model.     

A Neuro‐wavelet Model for the Short‐Term Forecasting of High‐Frequency Time Series of Stock Returns

We propose a wavelet neural network (neuro‐wavelet) model for the short‐term forecast of stock returns from high‐frequency financial data. The proposed hybrid model combines the capability of wavelets and neural networks to capture non‐stationary nonlinear attributes embedded in financial time series. A comparison study was performed on the predictive power of two econometric models and four recurrent neural network topologies. Several statistical measures were applied to the predictions and standard errors to evaluate the performance of all models. A Jordan net that used as input the coefficients resulting from a non‐decimated wavelet‐based multi‐resolution decomposition of an exogenous signal showed a consistent superior forecasting performance. Reasonable forecasting accuracy for the one‐, three‐ and five step‐ahead horizons was achieved by the proposed model. The procedure used to build the neuro‐wavelet model is reusable and can be applied to any high‐frequency financial series to specify the model characteristics associated with that particular series. Copyright © 2013 John Wiley & Sons, Ltd.