Discrete choice models (DCMs) are used to forecast demand in a variety of engineering, marketing, and policy contexts, and understanding the uncertainty associated with model forecasts is crucial to inform decision-making. This thesis evaluates the suitability of DCMs for forecasting automotive demand. The entire scope of this investigation is too broad to be covered here, but I explore several elements with a focus on three themes: defining how to measure forecast accuracy, comparing model specifications and forecasting methods in terms of prediction accuracy, and comparing the implications of model specifications and forecasting methods on vehicle design. Specifically I address several questions regarding the accuracy and uncertainty of market share predictions resulting from choice of utility function and structural specification, estimation method, and data structure assumptions. I1 compare more than 9,000 models based on those used in peer-reviewed literature and academic and government studies. Firstly, I find that including more model covariates generally improves predictive accuracy, but that the form those covariates take in the utility function is less important. Secondly, better model fit correlates well with better predictive accuracy; however, the models I construct— representative of those in extant literature— exhibit substantial prediction error stemming largely from limited model fit due to unobserved attributes. Lastly, accuracy of predictions in existing markets is neither a necessary nor sufficient condition for use in design. Much of the econometrics literature on vehicle market modeling has presumed that biased coefficients make for bad models. For purely predictive purposes, the drawbacks of potentially mitigating bias using generalized method of moments estimation coupled with instrumental variables outweigh the expected benefits in the experiments conducted in this dissertation. The risk of specifying invalid instruments is high, and my results suggest that the instruments frequently used in the automotive demand literature are likely invalid. Furthermore, biased coefficients are not necessarily bad for maximizing the predictive power of the model. Bias can even aid predictions by implicitly capturing persistent unobserved effects in some circumstances. Including alternative specific constants (ASCs) in DCM utility functions improves model fit but not necessarily forecast accuracy. For frequentist estimated models all tested methods of forecasting ASCs improved share predictions of the whole midsize sedan market over excluding ASC in predictions, but only one method results in improved long term new vehicle, or entrant, forecasts. As seen in a synthetic data study, assuming an incorrect relationship between observed attributes and the ASC for forecasting risks making worse forecasts than would be made by a model that excludes ASCs entirely. Treating the ASCs as model parameters with full distributions of uncertainty via Bayesian estimation is more robust to selection of ASC forecasting method and less reliant on persistent market structures, however it comes at increased computational cost. Additionally, the best long term forecasts are made by the frequentist model that treats ASCs as calibration constants fit to the model post estimation of other parameters.

Aim: The purpose of this study is to find new variables that can help explain the variation in the prices of apartments in the county of Stockholm. Method: By using recreated variables from an existing model created by Claussen, Jonsson and Lagerwall (2011) on behalf of Sweden’s central bank, and adding new variables obtained by the realtor Erik Olsson and the housing ad website Hemnet, the goal is to find a model that explains as much of the price variation as possible. The accuracy of the model is tested by an out-of- sample forecast. The study is based on monthly data for the years 2008 up until February 2013 and is written in cooperation with Valueguard. Findings: By combining the two data sets the adjusted coefficient of determination was 0,703, which means that 70,3% of the variation in the prices can be explained by the model.

In times of demand shocks, when quantitative forecasting based on historical time series becomes obsolete, the only information about future demand is "advance demand information", i.e. interpreting early customer bookings as an indicator of not yet known demand. This paper deals with a forecasting method which selects the optimal forecasting model type and the level of integration of advance demand information, depending on the patterns of the particular time series. This constitutes the applicability of the procedure within an industrial application where a large number of time series is automatically forecasted in a flexible and data-driven way. The architecture of such a planning system is explained and using real-world data from a make-to-order industry it is shown that the system is flexible enough to cover different demand patterns and is well-suited to forecast demand shocks. (authors' abstract)

Time series forecasting is an important area of forecasting in which past observations of the same variable are collected and analyzed to develop a model describing the underlying relationship. The model is then used to extrapolate the time series into the future. This modeling approach is particularly useful when little knowledge is available on the underlying data generating process or when there is no satisfactory explanatory model that relates the prediction variable to other explanatory variables. Time series can be modeled in a variety of ways e.g. using exponential smoothing techniques, regression models, autoregressive (AR) techniques, moving averages (MA) etc. Recent research activities in forecasting also suggested that artificial neural networks can be used as an alternative to traditional linear forecasting models. This study will, along the lines of an existing study in the literature, investigate the use of a hybrid approach to time series forecasting using both linear and neural network models. The proposed methodology consists of two basic steps. In the first step, a linear model is used to analyze the linear part of the problem and in the second step a neural network model is developed to model the residuals from the linear model. The results from the neural network can then be used to predict the error terms for the linear model. This means that the combined forecast of the time series will depend on both models. Following an overview of the models, empirical tests on real world data will be performed to determine the forecasting performance of such a hybrid model. Results have indicated that depending on the forecasting period, it might be worthwhile to consider the use of a hybrid model. / Thesis (M.Sc. (Computer Science))--North-West University, Vaal Triangle Campus, 2010.

Time series forecasting is an important area of forecasting in which past observations of the same variable are collected and analyzed to develop a model describing the underlying relationship. The model is then used to extrapolate the time series into the future. This modeling approach is particularly useful when little knowledge is available on the underlying data generating process or when there is no satisfactory explanatory model that relates the prediction variable to other explanatory variables. Time series can be modeled in a variety of ways e.g. using exponential smoothing techniques, regression models, autoregressive (AR) techniques, moving averages (MA) etc. Recent research activities in forecasting also suggested that artificial neural networks can be used as an alternative to traditional linear forecasting models. This study will, along the lines of an existing study in the literature, investigate the use of a hybrid approach to time series forecasting using both linear and neural network models. The proposed methodology consists of two basic steps. In the first step, a linear model is used to analyze the linear part of the problem and in the second step a neural network model is developed to model the residuals from the linear model. The results from the neural network can then be used to predict the error terms for the linear model. This means that the combined forecast of the time series will depend on both models. Following an overview of the models, empirical tests on real world data will be performed to determine the forecasting performance of such a hybrid model. Results have indicated that depending on the forecasting period, it might be worthwhile to consider the use of a hybrid model. / Thesis (M.Sc. (Computer Science))--North-West University, Vaal Triangle Campus, 2010.

Abstract   Purpose - The purpose with this thesis is to investigate how to increase sales forecast accuracy.   Methodology – To fulfil the purpose a case study was conducted. To collect data from the case study the authors performed interviews and gathered documents. The empirical data was then analysed and compared with the theoretical framework.   Result – The result shows that inaccuracies in forecasts are not necessarily because of the forecasting technique but can be a result from an unorganized forecasting process and having an inefficient information flow. The result further shows that it is not only important to review the information flow within the company but in the supply chain as whole to improve a forecast’s accuracy. The result also shows that time series can generate more accurate sales forecasts compared to only using qualitative techniques. It is, however, necessary to use a qualitative technique when creating time series. Time series only take time and sales history into account when forecasting, expertise regarding consumer behaviour, promotion activity, and so on, is therefore needed. It is also crucial to use qualitative techniques when selecting time series technique to achieve higher sales forecast accuracy. Personal expertise and experience are needed to identify if there is enough sales history, how much the sales are fluctuating, and if there will be any seasonality in the forecast. If companies gain knowledge about the benefits from each technique the combination can improve the forecasting process and increase the accuracy of the sales forecast.   Conclusions – This thesis, with support from a case study, shows how time series and qualitative techniques can be combined to achieve higher accuracy. Companies that want to achieve higher accuracy need to know how the different techniques work and what is needed to take into account when creating a sales forecast. It is also important to have knowledge about the benefits of a well-designed forecasting process, and to do that, improving the information flow both within the company and the supply chain is a necessity.      Research limitations – Because there are several different techniques to apply when creating a sales forecast, the authors could have involved more techniques in the investigation. The thesis work could also have used multiple case study objects to increase the external validity of the thesis.

This thesis describes a short term water demand forecasting application that is based upon a combination of a neural network forecast generator and a rule based system that modifies the resulting forecasts. Conventionally, short term forecasting of both water consumption and electrical load demand has been based upon mathematical models that aim to either extract the mathematical properties displayed by a time series of historical data, or represent the causal relationships between the level of demand and the key factors that determine that demand. These conventional approaches have been able to achieve acceptable levels of prediction accuracy for those days where distorting, non cyclic influences are not present to a significant degree. However, when such distortions are present, then the resultant decrease in prediction accuracy has a detrimental effect upon the controlling systems that are attempting to optimise the operation of the water or electricity supply network. The abnormal, non cyclic factors can be divided into those which are related to changes in the supply network itself, those that are related to particular dates or times of the year and those which are related to the prevailing meteorological conditions. If a prediction system is to provide consistently accurate forecasts then it has to be able to incorporate the effects of each of the factor types outlined above. The prediction system proposed in this thesis achieves this by the use of a neural network that by the application of appropriately classified example sets, can track the varying relationship between the level of demand and key meteorological variables. The influence of supply network changes and calendar related events are accounted for by the use of a rule base of prediction adjusting rules that are built up with reference to past occurrences of similar events. The resulting system is capable of eliminating a significant proportion of the large prediction errors that can lead to non optimal supply network operation.

This essay investigates three different GARCH-models (GARCH, EGARCH and GJR-GARCH) along with two distributions (Normal and Student’s t), which are used to forecast the Value at Risk (VaR) for different return series. Seven major international equity indices are examined. The purpose of the essay is to answer which of the three models that is better at forecasting the VaR and which distribution is more appropriate.  The results show that the EGARCH(1,1)  is preferred for all indices included in the study.

No description available.

In this dissertation, we propose a set of computationally efficient methods based on approximating/representing nonlinear processes by linear ones, so-called linearization. Firstly, a linearization method is introduced for estimating the multiple frequencies in sinusoidal processes. It utilizes a regularized autoregressive (AR) approximation, which can be regarded as a "large p - small n" approach in a time series context. An appealing property of regularized AR is that it avoids a model selection step and allows for an efficient updating of the frequency estimates whenever new observations are obtained. The theoretical analysis shows that the regularized AR frequency estimates are consistent and asymptotically normally distributed. Secondly, a sieve bootstrap scheme is proposed using the linear representation of generalized autoregressive conditional heteroscedastic (GARCH) models to construct prediction intervals (PIs) for the returns and volatilities. Our method is simple, fast and distribution-free, while providing sharp and well-calibrated PIs. A similar linear bootstrap scheme can also be used for diagnostic testing. Thirdly, we introduce a robust lagrange multiplier (LM) test, which utilizes either the bootstrap or permutation procedure to obtain critical values, for detecting GARCH effects. We justify that both bootstrap and permutation LM tests are consistent. Intensive numerical studies indicate that the proposed resampling algorithms significantly improve the size and power of the LM test in both skewed and heavy-tailed processes. Moreover, fourthly, we introduce a nonparametric trend test in the presence of GARCH effects (NT-GARCH) based on heteroscedastic ANOVA. Our empirical evidence show that NT-GARCH can effectively detect non-monotonic trends under GARCH, especially in the presence of irregular seasonal components. We suggest to apply the bootstrap procedure for both selecting the window length and finding critical values. The newly proposed methods are illustrated by applications to astronomical data, to foreign currency exchange rates as well as to water and air pollution data. Finally, the dissertation is concluded by an outlook on further extensions of linearization methods, e.g., in model order selection and change point detection.