Forecasts are important to decision-making and risk assessment in many domains. There has been recent interest in integrating forecast queries inside a DBMS. Answering a forecast query requires the creation of forecast models. Creating a forecast model is an expensive process and may require several scans over the base data as well as expensive operations to estimate model parameters. However, if forecast queries are issued repeatedly, answer times can be reduced significantly if forecast models are reused. Due to the possibly high number of forecast queries, existing models need to be found quickly. Therefore, we propose a model index that efficiently stores forecast models and allows for the efficient reuse of existing ones. Our experiments illustrate that the model index shows a negligible overhead for update transactions, but it yields significant improvements during query execution.

I explore whether participation on Estimize.com, a crowdsourced earnings-forecasting platform aimed primarily at novices, improves professional analysts’ forecast accuracy and career outcomes. Estimize provides its contributors with frequent and timely feedback on their forecast performance and offers them a new channel for disseminating their forecasts to a wider public, features that could help analysts improve their forecast accuracy and raise their online visibility. Using proprietary data obtained from Estimize and a difference-in-differences research design, I find that IBES analysts who are active on Estimize improve their EPS forecast accuracy by 13% relative to the sample-mean forecast error, as well as reduce forecast bias. These improvements in performance vary predictably in ways consistent with learning through feedback. Additionally, I find increased market reaction to the positive earnings-forecasts revisions issued by analysts who are active on Estimize. I also find that analysts active on Estimize enjoy incremental positive career outcomes after controlling for forecast accuracy. My results suggest that professional analysts can learn to become better forecasters through online feedback and consequently garner more attention from the market. My results also suggest analysts can improve their career outcomes by gaining additional online visibility. / Business Administration/Accounting

Wirtschaftsprognosen sollen die Unsicherheit bezüglich der zukünftigen wirtschaftlichen Entwicklung mindern und Planungsprozesse von Regierungen und Unternehmen unterstützen. Empirische Studien bescheinigen ihnen jedoch in aller Regel ein unbefriedigendes Qualitätsniveau. Auf der Suche nach den Ursachen hat sich in Form der rationalen Erwartungsbildung eine zentrale Grundforderung an  die Prognostiker herausgebildet. So müssten offensichtliche und systematische Fehler, wie bspw. regelmäßige Überschätzungen, mit der Zeit erkannt und abgestellt werden. Die erste Studie der Dissertation übt Kritik am vorherrschenden Verständnis der Rationalität. Dieses ist zu weitreichend, weshalb den Prognostikern die Rationalität voreilig abgesprochen wird. Anhand einer neuen empirischen Herangehensweise wird deutlich, dass die Prognosen aus einem anderen Blickwinkel heraus durchaus als rational angesehen werden können. Der zweite Aufsatz zeigt auf, dass in Form von Befragungsergebnissen öffentlich verfügbare Informationen bestehen, die bei geeigneter Verwendung zu einer Verbesserung der Qualität von Konjunkturprognosen beitragen würden. Die Rationalität dieser Prognosen ist daher stark eingeschränkt. Im dritten Papier erfolgt eine Analyse von Prognoserevisionen und deren Ursachen. Dabei zeigt sich, dass es keinen Zusammenhang zwischen der Rationalität und der Qualität der untersuchten Prognosezeitreihen gibt. Die vierte Studie dient der Präsentation der Ergebnisse eines Prognoseplanspiels, welches den Vergleich der Prognosen von Amateuren und Experten zum Ziel hatte. Es stellt sich heraus, dass die Prognosefehler erhebliche Übereinstimmungen aufweisen.

明天股票的價格是會漲還是會跌呢? 明天到底會不會下雨? 下期樂透開獎會是哪些號碼呢? 未來不知道會發生哪些事情? 大家總是希望能夠未卜先知、洞悉未來! 可是我們要如何進行預測呢? 本文比較了不同時間序列模型的預測績效, 而且測試預測的結合是否能夠改進預測的準確度? 時間序列模型的研究在近年來非常蓬勃地發展, 所以本文簡單介紹了時間序列模型(Time series models)當中的線性AR模型、非線性TAR模型、非線性STAR模型, 以及這些模型該如何來進行在樣本外的預測。 同時本文說明了預測的結合(Combined forecast)該如何進行? 預測結合的目的是希望能夠達到截長補短的效果! 除了傳統迴歸(Regression-based)方法和變動係數(Time-varying coefficients)方法外, 本文提出了兩種非迴歸類型的預測結合方法, 績效權數(Fitness weight)和支向機(Support Vector Machine)。 其中主要的焦點放在支向機, 因為迴歸方法可能會有共線性的問題, 支向機則是沒有這個問題。 本文實證的結果顯示, 在時間序列模型方面, 非線性模型的預測能力, 在大多數的情形底下, 都不如簡單的線性AR模型; 在預測結合的方面, 支向機的績效是和迴歸方法的績效是差不多的, 這兩者都比變動係數方法的績效來得穩固, 可是如果基底模型的預測值存在共線性的問題或樣本數目過少的問題, 那麼支向機的績效是優於迴歸方法的績效。 最後, 時間序列模型的預測績效會受到資料性質的影響, 而有極大的改變, 或許我們可以考慮使用比較保險的預測策略-預測結合, 因為預測結合的預測誤差範圍是小於時間序列模型的預測誤差範圍!

本研究以中國2007年至2013年盈餘預測的A股上市公司為主體。探討管理當局發佈強制性盈餘預測的預測形態、預測誤差以及市場反應與管理當局能力之關聯性，並進一步檢測管理當局能力是否影響到其對於強制性盈餘預告門檻的規避以及對於來年發佈自願性業績預測的意願。 實證的結果顯示管理當局能力越好其提供的強制性盈餘預測的形式越精準其預告資訊含量越多；再者，管理當局能力越好，其盈餘預測的誤差越低；實證結果也證明了市場對於能力較佳之管理當局所發佈的強制性盈餘預測的反應程度也較高。增額測試的結果顯示管理當局能力佳者盈餘品質較佳，具體表現為：相較於能力差的管理者，管理當局能力較好時不會通過盈餘管理去避免導因於產生增長50%或是下降50%而強制發佈盈餘預告的門檻。另外，管理當局能力越好，其在隔年度發佈自願性盈餘預測的幾率也越高。 關鍵詞：管理當局能力、強制性盈餘預測、預測形態、盈餘預測誤差、市場反應 / This thesis focuses on mandatory forecast issued from 2007 to 2013 in China and investigate whether managerial ability is related to mandatory forecasts types, forecast error and market reaction. Additionally, this thesis also examines whether managerial ability decrease the likelihood to avoid mandatory forecast thresholds. Finally, whether the managerial ability will increase the probability of issuing voluntary forecasts in the following year is an interesting but unsolved issue; I will fill the gap. Empirical results show that managers with superior ability tend to issue mandatory forecasts in the more precise type. As for the accuracy, the mandatory forecasts issued by better managers tend to have less error. I also find that managerial ability can promotes the informativeness of management earnings forecasts for the public. Additionally, high ability managers are less likely to avoid the thresholds of mandatory forecasts. Furthermore, better managers are more likely to issue voluntary forecasts in the following year of mandatory forecasts. Key Words: Managerial Ability, Mandatory Earnings Forecast, Forecast Format, Forecast Error, Market Reaction.

This thesis consists of three papers that makes independent contributions to the fields of forecast evaluation and financial econometrics. As such, the papers, chapter 1-3, can be read independently of each other. In Chapter 1, “Inferring an agent’s loss function based on a term structure of forecasts”, we provide conditions for identification, estimation and inference of an agent’s loss function based on an observed term structure of point forecasts. The loss function specification is flexible as we allow the preferences to be both asymmetric and to vary non-linearly across the forecast horizon. In addition, we introduce a novel forecast rationality test based on the estimated loss function. We employ the approach to analyse the U.S. Government’s preferences over budget surplus forecast errors. Interestingly, we find that it is relatively more costly for the government to underestimate the budget surplus and that this asymmetry is stronger at long forecast horizons. In Chapter 2, “Monitoring Systemic Risk”, we define systemic risk as the conditional probability of a systemic banking crisis. This conditional probability is modelled in a fixed effect binary response panel-model framework that allows for cross-sectional dependence (e.g. due to contagion effects). In the empirical application we identify several risk factors and it is shown that the level of systemic risk contains a predictable component which varies through time. Furthermore, we illustrate how the forecasts of systemic risk map into dynamic policy thresholds in this framework. Finally, by conducting a pseudo out-of-sample exercise we find that the systemic risk estimates provided reliable early-warning signals ahead of the recent financial crisis for several economies. Finally, in Chapter 3, “Equity Premium Predictability”, we reassess the evidence of out-of- sample equity premium predictability. The empirical finance literature has identified several financial variables that appear to predict the equity premium in-sample. However, Welch & Goyal (2008) find that none of these variables have any predictive power out-of-sample. We show that the equity premium is predictable out-of-sample once you impose certain shrinkage restrictions on the model parameters. The approach is motivated by the observation that many of the proposed financial variables can be characterised as ’weak predictors’ and this suggest that a James-Stein type estimator will provide a substantial risk reduction. The out-of-sample explanatory power is small, but we show that it is, in fact, economically meaningful to an investor with time-invariant risk aversion. Using a shrinkage decomposition we also show that standard combination forecast techniques tends to ’overshrink’ the model parameters leading to suboptimal model forecasts.

Ensembles are today routinely applied to estimate uncertainty in numerical predictions of complex systems such as the weather. Instead of initializing a single numerical forecast, using only the best guess of the present state as initial conditions, a collection (an ensemble) of forecasts whose members start from slightly different initial conditions is calculated. By varying the initial conditions within their error bars, the sensitivity of the resulting forecasts to these measurement errors can be accounted for. The ensemble approach can also be applied to estimate forecast errors that are due to insufficiently known model parameters by varying these parameters between ensemble members. An important (and difficult) question in ensemble weather forecasting is how well does an ensemble of forecasts reproduce the actual forecast uncertainty. A widely used criterion to assess the quality of forecast ensembles is statistical consistency which demands that the ensemble members and the corresponding measurement (the ``verification\'\') behave like random independent draws from the same underlying probability distribution. Since this forecast distribution is generally unknown, such an analysis is nontrivial. An established criterion to assess statistical consistency of a historical archive of scalar ensembles and verifications is uniformity of the verification rank: If the verification falls between the (k-1)-st and k-th largest ensemble member it is said to have rank k. Statistical consistency implies that the average frequency of occurrence should be the same for each rank. A central result of the present thesis is that, in a statistically consistent K-member ensemble, the (K+1)-dimensional vector of rank probabilities is a random vector that is uniformly distributed on the K-dimensional probability simplex. This behavior is universal for all possible forecast distributions. It thus provides a way to describe forecast ensembles in a nonparametric way, without making any assumptions about the statistical behavior of the ensemble data. The physical details of the forecast model are eliminated, and the notion of statistical consistency is captured in an elementary way. Two applications of this result to ensemble analysis are presented. Ensemble stratification, the partitioning of an archive of ensemble forecasts into subsets using a discriminating criterion, is considered in the light of the above result. It is shown that certain stratification criteria can make the individual subsets of ensembles appear statistically inconsistent, even though the unstratified ensemble is statistically consistent. This effect is explained by considering statistical fluctuations of rank probabilities. A new hypothesis test is developed to assess statistical consistency of stratified ensembles while taking these potentially misleading stratification effects into account. The distribution of rank probabilities is further used to study the predictability of outliers, which are defined as events where the verification falls outside the range of the ensemble, being either smaller than the smallest, or larger than the largest ensemble member. It is shown that these events are better predictable than by a naive benchmark prediction, which unconditionally issues the average outlier frequency of 2/(K+1) as a forecast. Predictability of outlier events, quantified in terms of probabilistic skill scores and receiver operating characteristics (ROC), is shown to be universal in a hypothetical forecast ensemble. An empirical study shows that in an operational temperature forecast ensemble, outliers are likewise predictable, and that the corresponding predictability measures agree with the analytically calculated ones.

[Truncated abstract] Given the theoretical and historical evidence that support the benefit of investing internationally. there is Iittle knowledge available of proper international portfolio construction in terms of how much should be invested in foreign countries, which countries should be targeted, and types of assets to be included in the portfolio. The prospects of these benefits depend on the market volatilities, cross-country correlations, and currency risks to change in the future. Another important issue in international portfolio diversification is the growth of newly emerging markets which have different characteristics from the developed ones. Addressing the issues, the thesis intends to investigate the nature of volatility, conditional correlations, and the impact of currency risks in international portfolio, both in developed and emerging markets. Chapter 2 provides literature review on volatility spillovers, conditional correlations, and forecasting both VaR and conditional correlations using GARCH-type models. Attention is made on the estimated models, type of assets, regions of markets, and tests of forecasts. Chapter 3 investigates the nature of volatility spillovers across intemational assets, which is important in determining the nature of portfolio's volatility when most assets are seems to be connected. ... The impacts of incorporating volatility spillovers and asymmetric effect on the forecast performance of conditional correlation will also be examined in this thesis. The VARMA-AGARCH of McAleer, Hoti and Chan (2008) and the VARMA-GARCH model of Ling and McAleer (2003) will be estimated to accommodate volatility spillovers and asymmetric effect. The CCC model of Bollerslev (1990) will also be estimated as benchmark as the model does not incorporate both volatility spillovers and asymmetric effects. Given the information about the nature of conditional correlations resulted from the forecasts using a rolling window technique, Section 2 of Chapter 4 investigates the nature of conditional correlations by estimating two multivariate GARCH models allowing for time-varying conditional correlations, namely the DCC model of Engle (2002) and the GARCC model of McAleer et al. (2008). Chapter 5 conducts VaR forecast considering the important role of VaR as a standard tool for risk management. Especially, the chapter investigates whether volatility spillovers and time-varying conditional correlations discussed in the previous two chapters are of helps in providing better VaR forecasts. The BEKK model of Engle and Kroner (1995) and the DCC model of Engle (2002) will be estimated to incorporate volatility spillovers and conditional correlations, respectively. The DVEC model of Bollerslev et al. (1998) and the CCC model of Bollerslev (1990) will be estimated to serve benchmarks, as both models do not incorporate both volatility spillovers and timevarying conditional correlations. Chapter 6 concludes the thesis and lists somc possible future research.

Diversas atividades humanas são fortemente dependentes do clima e da sua variabilidade, especialmente aquelas relacionadas ao uso da água. A operação integrada de reservatórios com múltiplos usos requer uma série de decisões que definem quanta água deve ser alocada, ao longo do tempo para cada um dos usos, e quais os volumes dos reservatórios a serem mantidos. O conhecimento antecipado das condições climáticas resulta de vital importância para os operadores de reservatórios, pois o insumo dos reservatórios é a vazão dos rios, que por sua vez é dependente de condições atmosféricas e hidrológicas em diferentes escalas de tempo e espaço. A pesquisa trata sobre três importantes elementos de subsídio à tomada de decisão na operação de reservatórios baseada na previsão de variáveis hidrológicas: (a) as previsões de vazão de curto prazo; (b) as previsões de precipitação de longo prazo e (c) as medidas de desempenho das previsões. O reservatório de Furnas, localizado na bacia do Rio Grande, em Minas Gerais, foi selecionado como estudo de caso devido, principalmente, à disponibilidade de previsões quantitativas de chuva e pela importância desse reservatório na região analisada. A previsão de curto prazo de vazão com base na precipitação foi estimada com um modelo empírico (rede neural artificial) e a previsão de precipitação foi obtida pelo modelo regional ETA. Uma metodologia de treinamento e validação da rede neural artificial foi desenvolvida utilizando previsões perfeitas de chuva (considerando a chuva observada como previsão) e utilizando o maior número de dados disponíveis, favorecendo a representatividade dos resultados obtidos. A metodologia empírica alcançou os desempenhos obtidos com um modelo hidrológico conceitual, mostrando-se menos sensitiva aos erros na previsão quantitativa de precipitação nessa bacia. Os resultados obtidos mostraram que as previsões de vazão utilizando modelos empíricos e conceituais e incorporando previsões quantitativas de precipitação são melhores que a metodologia utilizada pelo ONS no local de estudo. A redução dos erros de previsão relativos à metodologia empregada pelo ONS foi em torno de 20% quando usadas previsões quantitativas de precipitação definidas pelo modelo regional ETA e superiores a 50% quando usadas previsões perfeitas de precipitação. Embora essas últimas previsões nunca possam ser obtidas na prática, os resultados sugerem o quanto o incremento do desempenho das previsões quantitativas de chuva melhoraria as previsões de vazão. A previsão de precipitação de longo prazo para a bacia analisada foi também estimada com um modelo empírico de redes neurais artificiais e utilizando índices climáticos como variáveis de entrada. Nesse sentido, foram estimadas previsões de precipitação acumulada no período mais chuvoso (DJF) utilizando índices climáticos associados a fenômenos climáticos, como o El Niño - Oscilação Sul e a Oscilação Decadal do Pacífico, e a modos de variabilidade climática, como a Oscilação do Atlântico Norte e o Modo Anular do Hemisfério Sul. Apesar das redes neurais artificiais terem sido aplicadas em diversos problemas relacionados a hidrometeorologia, a aplicação dessas técnicas na previsão de precipitação de longo prazo é ainda rara. Os resultados obtidos nesse trabalho mostraram que consideráveis reduções dos erros da previsão relativos ao uso apenas da média climatológica como previsão podem ser obtidos com a metodologia utilizada. Foram obtidas reduções dos erros de, no mínimo 50%, e chegando até um valor próximo a 75% nos diferentes testes efetuados no estudo de caso. Uma medida de desempenho da previsão foi desenvolvida baseada no uso de tabelas de contingência e levando em conta a utilidade da previsão. Essa medida de desempenho foi calculada com base nos resultados do uso das previsões por um modelo de operação de reservatório, e não apenas na comparação de vazões previstas e observadas. Nos testes realizados durante essa pesquisa, ficou evidente que não existe uma relação unívoca entre qualidade das previsões e utilidade das previsões. No entanto, em função de comportamentos particulares das previsões, tendências foram encontradas, como por exemplo nos modelos cuja previsão apresenta apenas defasagem. Nesses modelos, a utilidade das previsões tende a crescer na medida que a qualidade das mesmas aumenta. Por fim, uma das grandes virtudes da medida de desempenho desenvolvida nesse trabalho foi sua capacidade de distinguir o desempenho de modelos que apresentaram a mesma qualidade. / Several human activities are strongly dependent on climate and its variability, especially those related to water use. The operation of multi-purpose reservoirs systems defines how much water should be allocated and the reservoir storage volumes to be maintained, over time. Knowing in advance the weather conditions helps the decision making process, as the major inputs to reservoirs are the streamflows, which are dependent on atmospheric and hydrological conditions at different time-space scales. This research deals with three important aspects towards the decision making process of multi-purpose reservoir operation based on forecast of hydrological variables: (a) short-term streamflow forecast, (b) long-range precipitation forecast and (c) performance measures. The Furnas reservoir on the Rio Grande basin was selected as the case study, primarily because of the availability of quantitative precipitation forecasts from the Brazilian Center for Weather Prediction and Climate Studies and due to its importance in the Brazilian hydropower generation system. Short-term streamflow forecasts were estimated by an empirical model (artificial neural network – ANN) and incorporating forecast of rainfall. Quantitative precipitation forecasts (QPFs), defined by the ETA regional model, were used as inputs to the ANN models. A methodology for training and validating the ANN models was developed using perfect precipitation forecasts (i.e., using the observed precipitation as if it was a forecast) and considering the largest number of available samples, in order to increase the representativeness of the results. The empirical methodology achieved the performance obtained with a conceptual hydrological model and seemed to be less sensitive to precipitation forecast error relative to the conceptual hydrological model. Although limited to one reservoir, the results obtained show that streamflow forecasting using empirical and conceptual models and incorporating QPFs performs better than the methodology used by ONS. Reduction in the forecast errors relative to the ONS method was about 20% when using QPFs provided by ETA model, and greater than 50% when using the perfect precipitation forecast. Although the latter can never be achieved in practice, these results suggest that improving QPFs would lead to better forecasts of reservoir inflows. Long-range precipitation forecast was also estimated by an empirical model based on artificial neural networks and using climate indices as input variables. The output variable is the summer (DJF) precipitation over the Furnas watershed. It was estimated using climate indices related to climatic phenomena such as El Niño - Southern Oscillation and the Pacific Decadal Oscillation and modes of climate variability, such as the North Atlantic Oscillation and the Southern Annular Mode. Despite of ANN has been applied in several problems of hydrometeorological areas, the application of such technique for long-range precipitation forecast is still rare. The results obtained demonstrate how the methodology for seasonal precipitation forecast based on ANN can be particularly helpful, with the use of available time series of climate indices. Reductions in the forecast errors achieved by using only the climatological mean as forecast were considerable, being at least of 50% and reaching values close to 75% in several tests. A performance measure based on the use of contingency tables was developed taking into account the utility of the forecast. This performance measure was calculated based on the results of the use of the forecasts by a reservoir operation model, and not only by comparing streamflow observed and forecast. The performed tests show that there is no unequivocal relationship between quality and utility of the forecasts. However, when the forecast has a particular behavior, trends were found in the relationship between utility and quality of the forecast, such as models that generate streamflow forecast with lags in comparison to the observed values. In these models, the utility of the forecasts tends to enhance as the quality increases. Finally, the ability to distinguish the performance of forecast models having similar quality was one of the main merits of the performance measure developed in this research.

Diversas atividades humanas são fortemente dependentes do clima e da sua variabilidade, especialmente aquelas relacionadas ao uso da água. A operação integrada de reservatórios com múltiplos usos requer uma série de decisões que definem quanta água deve ser alocada, ao longo do tempo para cada um dos usos, e quais os volumes dos reservatórios a serem mantidos. O conhecimento antecipado das condições climáticas resulta de vital importância para os operadores de reservatórios, pois o insumo dos reservatórios é a vazão dos rios, que por sua vez é dependente de condições atmosféricas e hidrológicas em diferentes escalas de tempo e espaço. A pesquisa trata sobre três importantes elementos de subsídio à tomada de decisão na operação de reservatórios baseada na previsão de variáveis hidrológicas: (a) as previsões de vazão de curto prazo; (b) as previsões de precipitação de longo prazo e (c) as medidas de desempenho das previsões. O reservatório de Furnas, localizado na bacia do Rio Grande, em Minas Gerais, foi selecionado como estudo de caso devido, principalmente, à disponibilidade de previsões quantitativas de chuva e pela importância desse reservatório na região analisada. A previsão de curto prazo de vazão com base na precipitação foi estimada com um modelo empírico (rede neural artificial) e a previsão de precipitação foi obtida pelo modelo regional ETA. Uma metodologia de treinamento e validação da rede neural artificial foi desenvolvida utilizando previsões perfeitas de chuva (considerando a chuva observada como previsão) e utilizando o maior número de dados disponíveis, favorecendo a representatividade dos resultados obtidos. A metodologia empírica alcançou os desempenhos obtidos com um modelo hidrológico conceitual, mostrando-se menos sensitiva aos erros na previsão quantitativa de precipitação nessa bacia. Os resultados obtidos mostraram que as previsões de vazão utilizando modelos empíricos e conceituais e incorporando previsões quantitativas de precipitação são melhores que a metodologia utilizada pelo ONS no local de estudo. A redução dos erros de previsão relativos à metodologia empregada pelo ONS foi em torno de 20% quando usadas previsões quantitativas de precipitação definidas pelo modelo regional ETA e superiores a 50% quando usadas previsões perfeitas de precipitação. Embora essas últimas previsões nunca possam ser obtidas na prática, os resultados sugerem o quanto o incremento do desempenho das previsões quantitativas de chuva melhoraria as previsões de vazão. A previsão de precipitação de longo prazo para a bacia analisada foi também estimada com um modelo empírico de redes neurais artificiais e utilizando índices climáticos como variáveis de entrada. Nesse sentido, foram estimadas previsões de precipitação acumulada no período mais chuvoso (DJF) utilizando índices climáticos associados a fenômenos climáticos, como o El Niño - Oscilação Sul e a Oscilação Decadal do Pacífico, e a modos de variabilidade climática, como a Oscilação do Atlântico Norte e o Modo Anular do Hemisfério Sul. Apesar das redes neurais artificiais terem sido aplicadas em diversos problemas relacionados a hidrometeorologia, a aplicação dessas técnicas na previsão de precipitação de longo prazo é ainda rara. Os resultados obtidos nesse trabalho mostraram que consideráveis reduções dos erros da previsão relativos ao uso apenas da média climatológica como previsão podem ser obtidos com a metodologia utilizada. Foram obtidas reduções dos erros de, no mínimo 50%, e chegando até um valor próximo a 75% nos diferentes testes efetuados no estudo de caso. Uma medida de desempenho da previsão foi desenvolvida baseada no uso de tabelas de contingência e levando em conta a utilidade da previsão. Essa medida de desempenho foi calculada com base nos resultados do uso das previsões por um modelo de operação de reservatório, e não apenas na comparação de vazões previstas e observadas. Nos testes realizados durante essa pesquisa, ficou evidente que não existe uma relação unívoca entre qualidade das previsões e utilidade das previsões. No entanto, em função de comportamentos particulares das previsões, tendências foram encontradas, como por exemplo nos modelos cuja previsão apresenta apenas defasagem. Nesses modelos, a utilidade das previsões tende a crescer na medida que a qualidade das mesmas aumenta. Por fim, uma das grandes virtudes da medida de desempenho desenvolvida nesse trabalho foi sua capacidade de distinguir o desempenho de modelos que apresentaram a mesma qualidade. / Several human activities are strongly dependent on climate and its variability, especially those related to water use. The operation of multi-purpose reservoirs systems defines how much water should be allocated and the reservoir storage volumes to be maintained, over time. Knowing in advance the weather conditions helps the decision making process, as the major inputs to reservoirs are the streamflows, which are dependent on atmospheric and hydrological conditions at different time-space scales. This research deals with three important aspects towards the decision making process of multi-purpose reservoir operation based on forecast of hydrological variables: (a) short-term streamflow forecast, (b) long-range precipitation forecast and (c) performance measures. The Furnas reservoir on the Rio Grande basin was selected as the case study, primarily because of the availability of quantitative precipitation forecasts from the Brazilian Center for Weather Prediction and Climate Studies and due to its importance in the Brazilian hydropower generation system. Short-term streamflow forecasts were estimated by an empirical model (artificial neural network – ANN) and incorporating forecast of rainfall. Quantitative precipitation forecasts (QPFs), defined by the ETA regional model, were used as inputs to the ANN models. A methodology for training and validating the ANN models was developed using perfect precipitation forecasts (i.e., using the observed precipitation as if it was a forecast) and considering the largest number of available samples, in order to increase the representativeness of the results. The empirical methodology achieved the performance obtained with a conceptual hydrological model and seemed to be less sensitive to precipitation forecast error relative to the conceptual hydrological model. Although limited to one reservoir, the results obtained show that streamflow forecasting using empirical and conceptual models and incorporating QPFs performs better than the methodology used by ONS. Reduction in the forecast errors relative to the ONS method was about 20% when using QPFs provided by ETA model, and greater than 50% when using the perfect precipitation forecast. Although the latter can never be achieved in practice, these results suggest that improving QPFs would lead to better forecasts of reservoir inflows. Long-range precipitation forecast was also estimated by an empirical model based on artificial neural networks and using climate indices as input variables. The output variable is the summer (DJF) precipitation over the Furnas watershed. It was estimated using climate indices related to climatic phenomena such as El Niño - Southern Oscillation and the Pacific Decadal Oscillation and modes of climate variability, such as the North Atlantic Oscillation and the Southern Annular Mode. Despite of ANN has been applied in several problems of hydrometeorological areas, the application of such technique for long-range precipitation forecast is still rare. The results obtained demonstrate how the methodology for seasonal precipitation forecast based on ANN can be particularly helpful, with the use of available time series of climate indices. Reductions in the forecast errors achieved by using only the climatological mean as forecast were considerable, being at least of 50% and reaching values close to 75% in several tests. A performance measure based on the use of contingency tables was developed taking into account the utility of the forecast. This performance measure was calculated based on the results of the use of the forecasts by a reservoir operation model, and not only by comparing streamflow observed and forecast. The performed tests show that there is no unequivocal relationship between quality and utility of the forecasts. However, when the forecast has a particular behavior, trends were found in the relationship between utility and quality of the forecast, such as models that generate streamflow forecast with lags in comparison to the observed values. In these models, the utility of the forecasts tends to enhance as the quality increases. Finally, the ability to distinguish the performance of forecast models having similar quality was one of the main merits of the performance measure developed in this research.