This research investigates three approaches to new product sales forecasting: statistical, judgmental and the integration of these two approaches. The aim of the research is to find a simple, easy-to-use, low cost and accurate tool which can be used by managers to forecast the sales of new products. A review of the literature suggested that the Bass diffusion model was an appropriate statistical method for new product sales forecasting. For the judgmental approach, after considering different methods and constraints, such as bias, complexity, lack of accuracy, high cost and time involvement, the Delphi method was identified from the literature as a method, which has the potential to mitigate bias and produces accurate predictions at a low cost in a relatively short time. However, the literature also revealed that neither of the methods: statistical or judgmental, can be guaranteed to give the best forecasts independently, and a combination of them is the often best approach to obtaining the most accurate predictions. The study aims to compare these three approaches by applying them to actual sales data. To forecast the sales of new products, the Bass diffusion model was fitted to the sales history of similar (analogous) products that had been launched in the past and the resulting model was used to produce forecasts for the new products at the time of their launch. These forecasts were compared with forecasts produced through the Delphi method and also through a combination of statistical and judgmental methods. All results were also compared to the benchmark levels of accuracy, based on previous research and forecasts based on various combinations of the analogous products’ historic sales data. Although no statistically significant difference was found in the accuracy of forecasts, produced by the three approaches, the results were more accurate than those obtained using parameters suggested by previous researchers. The limitations of the research are discussed at the end of the thesis, together with suggestions for future research.

Yang, Haiqin. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2003. / Includes bibliographical references (leaves 98-109). / Abstracts in English and Chinese. / Abstract --- p.ii / Acknowledgement --- p.v / Chapter 1 --- Introduction --- p.1 / Chapter 1.1 --- Time Series Prediction and Its Problems --- p.1 / Chapter 1.2 --- Major Contributions --- p.2 / Chapter 1.3 --- Thesis Organization --- p.3 / Chapter 1.4 --- Notation --- p.4 / Chapter 2 --- Literature Review --- p.5 / Chapter 2.1 --- Framework --- p.6 / Chapter 2.1.1 --- Data Processing --- p.8 / Chapter 2.1.2 --- Model Building --- p.10 / Chapter 2.1.3 --- Forecasting Procedure --- p.12 / Chapter 2.2 --- Model Descriptions --- p.13 / Chapter 2.2.1 --- Linear Models --- p.15 / Chapter 2.2.2 --- Non-linear Models --- p.17 / Chapter 2.2.3 --- ARMA Models --- p.21 / Chapter 2.2.4 --- Support Vector Machines --- p.23 / Chapter 3 --- Support Vector Regression --- p.27 / Chapter 3.1 --- Regression Problem --- p.27 / Chapter 3.2 --- Loss Function --- p.29 / Chapter 3.3 --- Kernel Function --- p.34 / Chapter 3.4 --- Relation to Other Models --- p.36 / Chapter 3.4.1 --- Relation to Support Vector Classification --- p.36 / Chapter 3.4.2 --- Relation to Ridge Regression --- p.38 / Chapter 3.4.3 --- Relation to Radial Basis Function --- p.40 / Chapter 3.5 --- Implemented Algorithms --- p.40 / Chapter 4 --- Margins in Support Vector Regression --- p.46 / Chapter 4.1 --- Problem --- p.47 / Chapter 4.2 --- General ε-insensitive Loss Function --- p.48 / Chapter 4.3 --- Accuracy Metrics and Risk Measures --- p.52 / Chapter 5 --- Margin Variation --- p.55 / Chapter 5.1 --- Non-fixed Margin Cases --- p.55 / Chapter 5.1.1 --- Momentum --- p.55 / Chapter 5.1.2 --- GARCH --- p.57 / Chapter 5.2 --- Experiments --- p.58 / Chapter 5.2.1 --- Momentum --- p.58 / Chapter 5.2.2 --- GARCH --- p.65 / Chapter 5.3 --- Discussions --- p.72 / Chapter 6 --- Relation between Downside Risk and Asymmetrical Margin Settings --- p.77 / Chapter 6.1 --- Mathematical Derivation --- p.77 / Chapter 6.2 --- Algorithm --- p.81 / Chapter 6.3 --- Experiments --- p.83 / Chapter 6.4 --- Discussions --- p.86 / Chapter 7 --- Conclusion --- p.92 / Chapter A --- Basic Results for Solving SVR --- p.94 / Chapter A.1 --- Dual Theory --- p.94 / Chapter A.2 --- Standard Method to Solve SVR --- p.96 / Bibliography --- p.98

No description available.

The autoregressive-integrated moving average (AMIRA) process plays an essential role in time series models. Classical method of finding the maximum likelihood (ML) estimate of the parameters in an ARIMA(p; d; q) model consists of evaluating the likelihood function through the Box-Jenkins approach or the Innovations Algorithm and optimizing it by numerical methods such as the quasi-Newton algorithms. However, these approaches have several drawbacks. The quasi-Newton methods tend to be unstable when the likelihood function is highly nonlinear. In this paper, we consider a state space representation of the ARIMA(p; d; q) process. The likelihood function can be easily expressed by the Kalman filter and the ML estimates can be obtained through a combination of Kalman smoother and the EM Algorithm. The updating equations in the EM algorithm possess a simple analytical form. A quasi-Newton scheme has also been implemented to accelerate the convergence of the EM Algorithm. The simulations studies show that the EM algorithm is more robust to starting values and the number of parameters, and the quasi-Newton acceleration scheme significantly improves the convergence rate of the EM algorithm. / 差分自回歸移動平均(AMIRA)模型在時間序列模型中有著重要地位。ARIMA模型的傳統極大似然估計方法通過Box-Jenkins方法或者新息算法(Innovations Algorithm)計算出似然函數，再通過擬牛頓(quasi-Newton)法等數值方法將之極大化，從而得到參數的極大似然估計。然而，此類方法在一定條件下存在缺陷。例如，當似然函數高度非線性時，擬牛頓法表現出不穩定的現象。本文考慮ARIMA模型的一種狀態空間(state-space)模型表示。在此表示下，參數的似然函數可以通過卡爾曼濾波算法計算，而參數的極大似然估計可以通過卡爾曼平滑和EM算法簡單得出。本問題中EM算法的迭代公式有簡潔的解析形式。同時，我們進一步考慮了一個擬牛頓加速算法來加快EM算法的收斂速度。通過模擬實驗我們發現，對於不同的初始值和參數個數，EM算法比擬牛頓法更為穩健。同時，擬牛頓的加速算法可以顯著加快EM算法的收斂速度。 / Huang, Rui. / Thesis M.Phil. Chinese University of Hong Kong 2015. / Includes bibliographical references (leaves 57-58). / Abstracts also in Chinese. / Title from PDF title page (viewed on 06, October, 2016). / Detailed summary in vernacular field only.

Dissertation Submitted for the degree of Master of Science in Geography at the University of the Witwatersrand FEBRUARY 08, 2016 / A flash flood is a flood which occurs within 6 hours from the start of a particular rainfall event. The ability to accurately evaluate and forecast flash floods could help in mitigating their harmful effects by helping communities plan their settlements outside of high risk areas and by providing information for the formulation and implementation of early warning systems. The overall aim of the study is to evaluate the use of RADAR data and hydrological models for flash flood evaluation and prediction. This is done by initialising both a lumped hydrological model (NAM) and a distributed hydrological model (MikeSHE) with both RADAR and raingauge derived precipitation estimates for the Jukskei river catchment located in Gauteng South Africa. The results of the model simulations are compared with each other and with actual streamflow data using various statistical techniques. The hydrometeorological characteristics of flash floods in the study catchment are also evaluated on a case by case basis. A fast response time and short duration are noted as the resounding characteristics of floods in the study catchment. All the model runs failed to correlate with streamflow (with any significant statistical certainty). The models also failed to significantly predict streamflow when using the pair sampled t-test. This highlights the difficulty in using rainfall estimates and hydrological models for discharge prediction. Although it is expected that the more advanced distributed model would fare better when predicting the variables associated with high flow events, it was only marginally better when simulating event timing. The lumped model did, however, fare better when correlating with stream flow, number of high flow events, peak flow, as well as total duration and volume / MT2016

Iowa has long been affected by periods characterized by extreme drought and flood. In 2008, Cedar Rapids, Iowa was devastated by a record flood with damages around $3 billion. Several years later, Iowa was affected by severe drought in 2012, causing upwards of $30 billion in damages and losses across the United States. These climatic regimes can quickly transition from one regime to another, as was observed in the June 2013 major floods to the late summer 2013 severe drought across eastern Iowa. Though it is not possible to prevent a natural disaster from occurring, we explore how predictable these events are by using forecast models and analogs. Iowa's climate records are analyzed from 1950 to 2012 to determine if there are specific surface and upper-air pressure patterns linked to climate regimes (i.e., cold/hot and dry/wet conditions for a given month). We found that opposing climate regimes in Iowa have reversed anomalies in certain geographical regions of the northern hemisphere. These defined patterns and waves suggested to us that it could be possible to forecast extreme temperature and precipitation periods over Iowa if given a skillful forecast system. We examined the CMC, COLA, and GFDL models within the National Multi-Model Ensemble suite to create analog forecasts based on either surface or upper-air pressure forecasts. The verification results show that some analogs have predictability skill at the 0.5-month lead time exceeding random chance, but our overall confidence in the analog forecasts is not high enough to allow us to issue statewide categorical temperature and precipitation climate forecasts.

This study uses a very high resolution numerical weather prediction (NWP) model to investigate the complex structure and behaviour of cold fronts along the New South Wales coast during the warmer months of the year, the complex interaction between the wind flow and coastlines and elevated areas as well as the lee-trough effect, particularly the way it affects waters off the east coast of Tasmania, The study also investigates the utility of the higher resolution NWP model to better predict wind fields compared to a lower resolution model. The University of New South Wales very high resolution model (HIRES), nested in the Australian Bureau of Meteorology's coarse NWP model (GASP), was run at various horizontal resolutions (from 15 to 25km) in order to investigate the above-mentioned features. It was found to bave very good skill in resolving the features and was also found to be very accurate in the prediction of surface wind fields for various yacht race events out to at least four days ahead. It can be concluded that there is considerable skill in the ability of high-resolution NWP models such as HIRES, to predict the major features of the wind fields over the ocean out to several days ahead. Moreover, it was also able to more accurately simulate the complex structure of the summer-time cool change as it progressed along the NSW coast than the lower resolution model runs. The influence of coastlines, particularly ones with complex topographical features, on the wind flow was demonstrated to a limited extent throughout the study. Finally the following concepts were also verified as a result of the study: - air flow takes the path of least resistance - the shape of topography can help generate local turbulence - the orientation of the wind flow to a mountain range is important in determining turbulent effects. - under certain airflow and stability situations, standing wave activity and a lee trough can be observed in the lee of mountains, hills or even high coastal cliffs.

Approximately 60% of New Zealand's electricity is produced from hydro generation. The Waitaki River catchment is located in the centre of the South Island of New Zealand, and produces 35-40% of New Zealand's electricity. Low inflow years in 1992 and 2001 resulted in the threat of power blackouts, and a national demand for electricity that is currently growing at 2 to 5% a year gives strong justification for better management of the hydro resource. Improved seasonal rainfall and inflow forecasts will result in the better management of the water used in hydro generation on a seasonal basis. Seasonal rainfall forecasting has been the focus of much international research in recent years, but seasonal inflow forecasting is in its relative infancy. Researchers have stated that key directions for both fields are to decrease the spatial scale of forecast products, and to tailor forecast products to end user needs, so as to provide more relevant and targeted forecasts, which will hopefully decrease the enormous socio-economic costs of climate fluctuations. This study calibrated several season ahead lake inflow and rainfall forecast models for the Waitaki river catchment, using statistical techniques to quantify relationships between land-ocean-atmosphere state variables and seasonally lagged inflows and rainfall. Techniques included principal components analysis and multiple linear regression, with cross-validation techniques applied to estimate model error. Many of both the continuous and discrete format models calibrated in this study predict anomalously wet and dry seasons better than random chance, and better than the long term mean as a predictor. 95% confidence limits around most model predictions in this study offer significant skill when compared with the range of all probable inflows (based on the 80 year recording history in the catchment). Models predicting winter Lake Pukaki inflows are those with the strongest predictive relationships in this study. Spring and summer predictions were generally less skilful than those for winter and autumn. Inflows could be predicted with some skill in winter and summer, but not rainfall, and rainfall could be predicted with some skill in autumn and spring, but not inflows. Models predicting inflows and rainfall for different seasons in this study use very different sets of predictor variables to accomplish their seasonal predictability. This may be related to the significant seasonal snow storage in the catchment, so that other factors such as temperature and the number of north-westerly storms may have a large part to play in the magnitude of inflows. Similarly, predicting the same dependent variable but for different seasons led to different contributing variables, leading to the conclusion that different wider physical causative mechanisms are behind the predictability in different seasons, and that they too should be studied separately in any future research. SST5 (sea surface temperature to the north of New Zealand) was found to have more relevance than any other predictor in predicting Waitaki river inflows and rainfall in any season. The models calibrated with SOI and IPO included as predictor variables were almost invariably worse in their predictive skill than those without, and the list of the most important predictor variables in all models did not include equatorial sea surface temperatures, sea level pressures, or 700hpa geopotential height variables. The conclusion from these findings is that equatorial ocean-atmosphere state variables do not have significant relationships with season ahead inflows and rainfall in the South Island of New Zealand. Seasonal climate forecasting on single catchment scale, and focussed to end user needs, is possible with some skill, at least in the South Island of New Zealand.

The development of some ability for forecasting low rainfalls would be helpful in Tuvalu as rainwater is the only source of fresh water in the country. The subsurface water is brackish and saline so the entire country depends totally on rainwater for daily domestic supplies, agricultural and farming activities. More importantly, these atolls are often influenced by droughts which consequently make inadequate drinking water an issue. A simple graph-based forecasting scheme is developed and presented in this thesis for forecasting below average mean rainfall in Funafuti over the next n-month period. The approach uses precursor ocean surface temperature data to make predictions of below average rainfall for n = 1, 2 12. The simplicity of the approach makes it a suitable method for the country and thus for the Tuvalu Meteorological Service to use as an operational forecasting tool in the climate forecasting desk. The graphical method was derived from standardised monthly rainfalls from the Funafuti manual raingauge for the period January 1945 to July 2007. The method uses lag-1 and-lag 2 NINO4 sea surface temperatures to define whether prediction conditions hold. The persistence of predictability tends to be maintained when the observed NINO4 ocean surface temperatures fall below 26.0oC. Although the developed method has a high success probability of up to 80 percent, this can only be achieved when conditions are within the predictable field. A considerable number of below average rainfall periods are not within the predictable field and therefore cannot be forecast by this method. However, the graphical approach has particular value in warning when an existing drought is likely to continue.

Includes bibliography.