Pub Date : 2025-10-01Epub Date: 2025-01-22DOI: 10.1016/j.ijforecast.2024.12.006
Jose M.G. Vilar
This paper presents the winning method that achieved fifth place overall in the M6 financial forecasting competition. The method is based on the idea that, under the efficient market hypothesis, it is often more effective to predict values close to the expected averages of categories and trends than to try to make precise predictions. By leveraging low-variability prediction methods, we forecast both the relative performance of multiple assets and their optimal investment positions. We demonstrate that combining asset-class and temporal averages yields modest but consistent advantages over reference indices. The results highlight the challenges of achieving above-average returns in efficient markets and the potential benefits of low-variability prediction methods in such contexts.
{"title":"Quasi-average predictions and regression to the trend: An application to the M6 financial forecasting competition","authors":"Jose M.G. Vilar","doi":"10.1016/j.ijforecast.2024.12.006","DOIUrl":"10.1016/j.ijforecast.2024.12.006","url":null,"abstract":"<div><div>This paper presents the winning method that achieved fifth place overall in the M6 financial forecasting competition. The method is based on the idea that, under the efficient market hypothesis, it is often more effective to predict values close to the expected averages of categories and trends than to try to make precise predictions. By leveraging low-variability prediction methods, we forecast both the relative performance of multiple assets and their optimal investment positions. We demonstrate that combining asset-class and temporal averages yields modest but consistent advantages over reference indices. The results highlight the challenges of achieving above-average returns in efficient markets and the potential benefits of low-variability prediction methods in such contexts.</div></div>","PeriodicalId":14061,"journal":{"name":"International Journal of Forecasting","volume":"41 4","pages":"Pages 1505-1513"},"PeriodicalIF":7.1,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145020566","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-01Epub Date: 2025-03-18DOI: 10.1016/j.ijforecast.2025.02.006
Michael S. Lewis-Beck , John Kenny , Debra Leiter , Andreas Erwin Murr , Onyinye B. Ogili , Mary Stegmaier , Charles Tien
We draw globally on a major election forecasting tool, political economy models. Vote intention polls in pre-election public surveys are a widely known approach; however, the lesser-known political economy models take a different scientific tack, relying on regression analysis and voting theory, particularly the force of “fundamentals.” We begin our discussion with two advanced industrial democracies, the US and UK. We then examine two less frequently forecasted cases, Mexico and Ghana, to highlight the potential for political-economic forecasting and the challenges faced. In evaluating the performance of political economy models, we argue for their accuracy but do not neglect lead time, parsimony, and transparency. Furthermore, we suggest how the political economic approach can be adapted to the changing landscape that democratic electorates face.
{"title":"Election forecasting: Political economy models","authors":"Michael S. Lewis-Beck , John Kenny , Debra Leiter , Andreas Erwin Murr , Onyinye B. Ogili , Mary Stegmaier , Charles Tien","doi":"10.1016/j.ijforecast.2025.02.006","DOIUrl":"10.1016/j.ijforecast.2025.02.006","url":null,"abstract":"<div><div><span>We draw globally on a major election forecasting tool, political economy models. Vote intention polls in pre-election public surveys are a widely known approach; however, the lesser-known political economy models take a different scientific tack, relying on regression analysis and voting theory, particularly the force of “fundamentals.” We begin our discussion with two advanced industrial democracies, the US and UK. We then examine two less frequently forecasted cases, Mexico and Ghana, to highlight the potential for political-economic forecasting and the challenges faced. In evaluating the performance of political economy models, we argue for their accuracy but do not neglect lead time, parsimony, and transparency. Furthermore, we suggest how the </span>political economic approach can be adapted to the changing landscape that democratic electorates face.</div></div>","PeriodicalId":14061,"journal":{"name":"International Journal of Forecasting","volume":"41 4","pages":"Pages 1655-1665"},"PeriodicalIF":7.1,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145019324","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-01Epub Date: 2025-04-11DOI: 10.1016/j.ijforecast.2025.03.002
Kenwin Maung, Norman R. Swanson
The Makridakis M6 Financial Duathalon competition builds on prior M-competitions that focus on the properties of point and probabilistic forecasts of random variables by also evaluating investment decisions in financial markets. In particular, the M6 competition evaluates both forecasts and investment outcomes associated with the analysis of a large group of financial time series variables. Given the importance of return and risk forecasting when making investment decisions, a natural question in this context concerns what sorts of methods and models are available for said forecasting and were used by participants of the competition. In this survey, we discuss such methods and models, with a specific focus on the construction of financial time series forecasts using approaches designed for both discrete and continuous time setups and using both small and large (high dimensional and/or high frequency) datasets. Examples covered range from simple random walk-type models of returns to parametric GARCH and nonparametric integrated volatility methods for forecasting volatility (risk). We also present the results of a novel empirical illustration that underscores the difficulty in forecasting financial returns, even when using so-called big data.
{"title":"A survey of models and methods used for forecasting when investing in financial markets","authors":"Kenwin Maung, Norman R. Swanson","doi":"10.1016/j.ijforecast.2025.03.002","DOIUrl":"10.1016/j.ijforecast.2025.03.002","url":null,"abstract":"<div><div>The <em>Makridakis M6 Financial Duathalon</em><span><span> competition builds on prior M-competitions that focus on the properties of point and probabilistic forecasts of random variables<span> by also evaluating investment decisions in financial markets. In particular, the M6 competition evaluates both forecasts and investment outcomes associated with the analysis of a large group of financial time series<span> variables. Given the importance of return and risk forecasting when making investment decisions, a natural question in this context concerns what sorts of methods and models are available for said forecasting and were used by participants of the competition. In this survey, we discuss such methods and models, with a specific focus on the construction of financial time series forecasts using approaches designed for both discrete and continuous time setups and using both small and large (high dimensional and/or high frequency) datasets. Examples covered range from simple random walk-type models of returns to parametric </span></span></span>GARCH<span> and nonparametric integrated volatility methods for forecasting volatility (risk). We also present the results of a novel empirical illustration that underscores the difficulty in forecasting financial returns, even when using so-called big data.</span></span></div></div>","PeriodicalId":14061,"journal":{"name":"International Journal of Forecasting","volume":"41 4","pages":"Pages 1355-1382"},"PeriodicalIF":7.1,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145020557","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-01Epub Date: 2025-01-21DOI: 10.1016/j.ijforecast.2024.12.007
Jianqiu Wang , Zhuo Wang , Ke Wu
We empirically investigate the relation between anomaly portfolio returns and market return predictability in the Chinese stock market. Using 132 long-leg, short-leg, and long-short anomaly portfolio returns, we employ various shrinkage-based statistical learning methods to capture predictive signals of the anomalies in a high-dimensional setting. Our analysis reveals statistically and economically significant return predictability using long- and short-leg anomaly portfolio returns. Moreover, high arbitrage risk enhances forecasting performance, supporting that the predictability stems from mispricing correction persistence. Contrary to findings in the US stock market, we find little evidence that the long-short anomaly portfolios contribute to market return predictability in China, due to the low persistence of asymmetric mispricing corrections. We provide simulation evidence to justify the distinct prediction patterns for the US and Chinese stock markets.
{"title":"Forecasting stock market return with anomalies: Evidence from China","authors":"Jianqiu Wang , Zhuo Wang , Ke Wu","doi":"10.1016/j.ijforecast.2024.12.007","DOIUrl":"10.1016/j.ijforecast.2024.12.007","url":null,"abstract":"<div><div>We empirically investigate the relation between anomaly portfolio returns and market return predictability in the Chinese stock market. Using 132 long-leg, short-leg, and long-short anomaly portfolio returns, we employ various shrinkage-based statistical learning methods to capture predictive signals of the anomalies in a high-dimensional setting. Our analysis reveals statistically and economically significant return predictability using long- and short-leg anomaly portfolio returns. Moreover, high arbitrage risk enhances forecasting performance, supporting that the predictability stems from mispricing correction persistence. Contrary to findings in the US stock market, we find little evidence that the long-short anomaly portfolios contribute to market return predictability in China, due to the low persistence of asymmetric mispricing corrections. We provide simulation evidence to justify the distinct prediction patterns for the US and Chinese stock markets.</div></div>","PeriodicalId":14061,"journal":{"name":"International Journal of Forecasting","volume":"41 3","pages":"Pages 1278-1295"},"PeriodicalIF":6.9,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144211944","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-01Epub Date: 2025-01-27DOI: 10.1016/j.ijforecast.2024.12.005
E.J. Whitehouse , D.I. Harvey , S.J. Leybourne
Asset price bubbles and crashes can have severe consequences for the stability of financial and economic systems. Policymakers require timely identification of such bubbles in order to respond to their emergence. In this paper we propose new econometric procedures that improve the speed of detection for an emerging asset price bubble in real time. Our new monitoring procedures make use of alternative variance standardisations that are better able to capture the behaviour of the underlying process during a bubble phase. We derive asymptotic results to show that using these alternative variance standardisations does not increase the probability of false detection under the no-bubble (unit root) null hypothesis relative to existing procedures. However, Monte Carlo simulations demonstrate that much earlier detection becomes possible with our new procedures under the bubble (explosive autoregressive) alternative. Empirical applications to OECD housing markets and bitcoin prices show the value in terms of earlier detection of bubbles that our new procedures can achieve. In particular, we show that the United States housing bubble that preceded the global financial crisis could have been detected as early as 1999:Q1 by our new procedures.
{"title":"Real-time monitoring procedures for early detection of bubbles","authors":"E.J. Whitehouse , D.I. Harvey , S.J. Leybourne","doi":"10.1016/j.ijforecast.2024.12.005","DOIUrl":"10.1016/j.ijforecast.2024.12.005","url":null,"abstract":"<div><div>Asset price bubbles and crashes can have severe consequences for the stability of financial and economic systems. Policymakers require timely identification of such bubbles in order to respond to their emergence. In this paper we propose new econometric procedures that improve the speed of detection for an emerging asset price bubble in real time. Our new monitoring procedures make use of alternative variance standardisations that are better able to capture the behaviour of the underlying process during a bubble phase. We derive asymptotic results to show that using these alternative variance standardisations does not increase the probability of false detection under the no-bubble (unit root) null hypothesis relative to existing procedures. However, Monte Carlo simulations demonstrate that much earlier detection becomes possible with our new procedures under the bubble (explosive autoregressive) alternative. Empirical applications to OECD housing markets and bitcoin prices show the value in terms of earlier detection of bubbles that our new procedures can achieve. In particular, we show that the United States housing bubble that preceded the global financial crisis could have been detected as early as 1999:Q1 by our new procedures.</div></div>","PeriodicalId":14061,"journal":{"name":"International Journal of Forecasting","volume":"41 3","pages":"Pages 1260-1277"},"PeriodicalIF":6.9,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144211943","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Forecasting consumer price index (CPI) inflation is of paramount importance for both academics and policymakers at central banks. This study introduces the filtered ensemble wavelet neural network (FEWNet) to forecast CPI inflation, tested in BRIC countries. FEWNet decomposes inflation data into high- and low-frequency components using wavelet transforms, and incorporates additional economic factors, such as economic policy uncertainty and geopolitical risk, to enhance forecast accuracy. These wavelet-transformed series and filtered exogenous variables are input into downstream autoregressive neural networks, producing the final ensemble forecast. Theoretically, we demonstrate that FEWNet reduces empirical risk compared to fully connected autoregressive neural networks. Empirically, FEWNet outperforms other forecasting methods and effectively estimates prediction uncertainty, due to its ability to capture non-linearities and long-range dependencies through its adaptable architecture. Consequently, FEWNet emerges as a valuable tool for central banks to manage inflation and enhance monetary policy decisions.
{"title":"Forecasting CPI inflation under economic policy and geopolitical uncertainties","authors":"Shovon Sengupta , Tanujit Chakraborty , Sunny Kumar Singh","doi":"10.1016/j.ijforecast.2024.08.005","DOIUrl":"10.1016/j.ijforecast.2024.08.005","url":null,"abstract":"<div><div>Forecasting consumer price index (CPI) inflation is of paramount importance for both academics and policymakers at central banks. This study introduces the filtered ensemble wavelet neural network (FEWNet) to forecast CPI inflation, tested in BRIC countries. FEWNet decomposes inflation data into high- and low-frequency components using wavelet transforms, and incorporates additional economic factors, such as economic policy uncertainty and geopolitical risk, to enhance forecast accuracy. These wavelet-transformed series and filtered exogenous variables are input into downstream autoregressive neural networks, producing the final ensemble forecast. Theoretically, we demonstrate that FEWNet reduces empirical risk compared to fully connected autoregressive neural networks. Empirically, FEWNet outperforms other forecasting methods and effectively estimates prediction uncertainty, due to its ability to capture non-linearities and long-range dependencies through its adaptable architecture. Consequently, FEWNet emerges as a valuable tool for central banks to manage inflation and enhance monetary policy decisions.</div></div>","PeriodicalId":14061,"journal":{"name":"International Journal of Forecasting","volume":"41 3","pages":"Pages 953-981"},"PeriodicalIF":6.9,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144211932","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-01Epub Date: 2024-12-20DOI: 10.1016/j.ijforecast.2024.11.010
Luc Bauwens , Yongdeng Xu
Realized variance–covariance models define the conditional expectation of a realized variance–covariance matrix as a function of past matrices using a GARCH-type structure. We evaluate the forecasting performance of various models in terms of economic value, measured through economic loss functions, across two datasets. Our empirical findings reveal that models incorporating realized volatilities offer significant economic value and outperform GARCH models relying solely on daily returns for daily and weekly horizons. Among two realized variance–covariance measures, using a directly rescaled intraday measure for full-day estimation provides more economic value than employing overnight returns, which tends to produce noisier estimators of overnight covariance, diminishing their predictive effectiveness. The HEAVY-H model for the variance–covariance matrix of the daily return demonstrates superior or comparable performance to the best-performing realized variance–covariance models, making it a preferred choice for empirical analysis.
{"title":"The contribution of realized variance–covariance models to the economic value of volatility timing","authors":"Luc Bauwens , Yongdeng Xu","doi":"10.1016/j.ijforecast.2024.11.010","DOIUrl":"10.1016/j.ijforecast.2024.11.010","url":null,"abstract":"<div><div>Realized variance–covariance models define the conditional expectation of a realized variance–covariance matrix as a function of past matrices using a GARCH-type structure. We evaluate the forecasting performance of various models in terms of economic value, measured through economic loss functions, across two datasets. Our empirical findings reveal that models incorporating realized volatilities offer significant economic value and outperform GARCH models relying solely on daily returns for daily and weekly horizons. Among two realized variance–covariance measures, using a directly rescaled intraday measure for full-day estimation provides more economic value than employing overnight returns, which tends to produce noisier estimators of overnight covariance, diminishing their predictive effectiveness. The HEAVY-H model for the variance–covariance matrix of the daily return demonstrates superior or comparable performance to the best-performing realized variance–covariance models, making it a preferred choice for empirical analysis.</div></div>","PeriodicalId":14061,"journal":{"name":"International Journal of Forecasting","volume":"41 3","pages":"Pages 1165-1183"},"PeriodicalIF":6.9,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144211938","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-01Epub Date: 2024-11-13DOI: 10.1016/j.ijforecast.2024.10.002
Bohan Zhang , Anastasios Panagiotelis , Han Li
Forecast reconciliation has attracted significant research interest in recent years, with most studies taking the hierarchy of time series as given. We extend existing work that uses time series clustering to construct hierarchies to improve forecast accuracy in three ways. First, we investigate multiple approaches to clustering, including different clustering algorithms, how time series are represented, and how the distance between time series is defined. We find that cluster-based hierarchies improve forecast accuracy relative to two-level hierarchies. Second, we devise an approach based on random permutation of hierarchies, keeping the hierarchy structure fixed while time series are randomly allocated to clusters. In doing so, we find that improvements in forecast accuracy that accrue from using clustering do not arise from grouping similar series but from the structure of the hierarchy. Third, we propose an approach based on averaging forecasts across hierarchies constructed using different clustering methods that is shown to outperform any single clustering method. All analysis is carried out on two benchmark datasets and a simulated dataset. Our findings provide new insights into the role of hierarchy construction in forecast reconciliation and offer valuable guidance on forecasting practice.
{"title":"Constructing hierarchical time series through clustering: Is there an optimal way for forecasting?","authors":"Bohan Zhang , Anastasios Panagiotelis , Han Li","doi":"10.1016/j.ijforecast.2024.10.002","DOIUrl":"10.1016/j.ijforecast.2024.10.002","url":null,"abstract":"<div><div>Forecast reconciliation has attracted significant research interest in recent years, with most studies taking the hierarchy of time series as given. We extend existing work that uses time series clustering to construct hierarchies to improve forecast accuracy in three ways. First, we investigate multiple approaches to clustering, including different clustering algorithms, how time series are represented, and how the distance between time series is defined. We find that cluster-based hierarchies improve forecast accuracy relative to two-level hierarchies. Second, we devise an approach based on random permutation of hierarchies, keeping the hierarchy structure fixed while time series are randomly allocated to clusters. In doing so, we find that improvements in forecast accuracy that accrue from using clustering do not arise from grouping similar series but from the structure of the hierarchy. Third, we propose an approach based on averaging forecasts across hierarchies constructed using different clustering methods that is shown to outperform any single clustering method. All analysis is carried out on two benchmark datasets and a simulated dataset. Our findings provide new insights into the role of hierarchy construction in forecast reconciliation and offer valuable guidance on forecasting practice.</div></div>","PeriodicalId":14061,"journal":{"name":"International Journal of Forecasting","volume":"41 3","pages":"Pages 1022-1036"},"PeriodicalIF":6.9,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144212011","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-01Epub Date: 2024-08-31DOI: 10.1016/j.ijforecast.2024.07.011
Liao Chen , Ning Jia , Zhixian Jiao , Hongke Zhao , Runbang Cui , Huimin Wang
Credit scoring is a popular tool for loan assessment, i.e., deciding whether to accept or reject a loan application. Traditional research into learning for credit scoring has only applied historically accepted samples without rejected applicants whose true repayment performance is absent, thereby causing both sample selection bias and wasting data. Some methods have been proposed for inferring rejected samples but they are still affected by several open problems, especially for medium- and long-term loan applications with a higher rejection rate. In particular, the heterogeneous relationships between accepted and rejected applications have not been well studied. Moreover, the complex repayment behaviors resulting from long repayment terms may lead to poor learning performance. Thus, we propose a reject inference framework with Semi-supervised Hierarchical Heterogeneous Network (S2HN) for credit scoring. We introduce a hierarchical heterogeneous network for revealing the complex connections between accepted and rejected applications, and use prospective heterogeneous repayment patterns as auxiliary information through clustering and a two-layer prediction architecture. Extensive experiments conducted based on real-world data sets demonstrated the effectiveness of our proposed method.
{"title":"A semi-supervised reject inference framework with hierarchical heterogeneous networks for credit scoring","authors":"Liao Chen , Ning Jia , Zhixian Jiao , Hongke Zhao , Runbang Cui , Huimin Wang","doi":"10.1016/j.ijforecast.2024.07.011","DOIUrl":"10.1016/j.ijforecast.2024.07.011","url":null,"abstract":"<div><div>Credit scoring is a popular tool for loan assessment, i.e., deciding whether to accept or reject a loan application. Traditional research into learning for credit scoring has only applied historically accepted samples without rejected applicants whose true repayment performance is absent, thereby causing both sample selection bias and wasting data. Some methods have been proposed for inferring rejected samples but they are still affected by several open problems, especially for medium- and long-term loan applications with a higher rejection rate. In particular, the heterogeneous relationships between accepted and rejected applications have not been well studied. Moreover, the complex repayment behaviors resulting from long repayment terms may lead to poor learning performance. Thus, we propose a reject inference framework with <strong>S</strong>emi-supervised <strong>H</strong>ierarchical <strong>H</strong>eterogeneous <strong>N</strong>etwork (S2HN) for credit scoring. We introduce a hierarchical heterogeneous network for revealing the complex connections between accepted and rejected applications, and use prospective heterogeneous repayment patterns as auxiliary information through clustering and a two-layer prediction architecture. Extensive experiments conducted based on real-world data sets demonstrated the effectiveness of our proposed method.</div></div>","PeriodicalId":14061,"journal":{"name":"International Journal of Forecasting","volume":"41 3","pages":"Pages 920-939"},"PeriodicalIF":6.9,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144211930","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-01Epub Date: 2024-11-30DOI: 10.1016/j.ijforecast.2024.11.006
Akylas Stratigakos , Salvador Pineda , Juan Miguel Morales
In real-world settings, decision-makers often have access to multiple forecasts for the same unknown quantity. Combining different forecasts has long been known to improve forecast quality, as measured by scoring rules in the case of probabilistic forecasting. However, improved forecast quality does not always translate into better decisions in a downstream problem that utilizes the resultant combined forecast as input. To this end, this work proposes a novel probabilistic forecast combination approach that accounts for the downstream stochastic optimization problem by which the decisions will be made. We propose a linear pool of probabilistic forecasts where the respective weights are learned by minimizing the expected decision cost of the induced combination, which we formulate as a nested optimization problem. Two methods are proposed for its solution: a gradient-based method that utilizes differential optimization layers, and a performance-based weighting method. The proposed decision-focused combination approach is validated in two integral problems associated with renewable energy integration in low-carbon power systems and compared against well-established combination methods. Namely, we examine an electricity market trading problem under stochastic solar production and a grid scheduling problem under stochastic wind production. The results illustrate that the proposed approach leads to lower expected downstream costs, while optimizing for forecast quality when estimating linear pool weights does not always translate into better decisions. Notably, optimizing for a combination of downstream cost and an accuracy-oriented scoring rule consistently leads to better decisions while also improving forecast quality.
{"title":"Decision-focused linear pooling for probabilistic forecast combination","authors":"Akylas Stratigakos , Salvador Pineda , Juan Miguel Morales","doi":"10.1016/j.ijforecast.2024.11.006","DOIUrl":"10.1016/j.ijforecast.2024.11.006","url":null,"abstract":"<div><div>In real-world settings, decision-makers often have access to multiple forecasts for the same unknown quantity. Combining different forecasts has long been known to improve forecast quality, as measured by scoring rules in the case of probabilistic forecasting. However, improved forecast quality does not always translate into better decisions in a downstream problem that utilizes the resultant combined forecast as input. To this end, this work proposes a novel probabilistic forecast combination approach that accounts for the downstream stochastic optimization problem by which the decisions will be made. We propose a linear pool of probabilistic forecasts where the respective weights are learned by minimizing the expected decision cost of the induced combination, which we formulate as a nested optimization problem. Two methods are proposed for its solution: a gradient-based method that utilizes differential optimization layers, and a performance-based weighting method. The proposed decision-focused combination approach is validated in two integral problems associated with renewable energy integration in low-carbon power systems and compared against well-established combination methods. Namely, we examine an electricity market trading problem under stochastic solar production and a grid scheduling problem under stochastic wind production. The results illustrate that the proposed approach leads to lower expected downstream costs, while optimizing for forecast quality when estimating linear pool weights does not always translate into better decisions. Notably, optimizing for a combination of downstream cost and an accuracy-oriented scoring rule consistently leads to better decisions while also improving forecast quality.</div></div>","PeriodicalId":14061,"journal":{"name":"International Journal of Forecasting","volume":"41 3","pages":"Pages 1112-1125"},"PeriodicalIF":6.9,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144211935","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号