International Journal of Forecasting最新文献

The purpose of this paper is to show that the effect of the zero-correlation assumption in combining forecasts can be huge, and that ignoring (positive) correlation can lead to confidence bands around the forecast combination that are much too narrow. In the typical case where three or more forecasts are combined, the estimated variance increases without bound when correlation increases. Intuitively, this is because similar forecasts provide little information if we know that they are highly correlated. Although we concentrate on forecast combinations and confidence bands, our theory applies to any statistic where the observations are linearly combined. We apply our theoretical results to explain why forecasts by central banks (in our case, the Bank of Japan and the European Central Bank) are so frequently misleadingly precise. In most cases ignoring correlation is harmful, and an estimated historical correlation or an imposed fixed correlation larger than 0.7 is required to produce credible confidence bands.

Harry Markowitz passed on June 22, 2023; some four years short of reaching 100 years old. Dr. Markowitz was not a traditional economist. That fact was well- established and documented from his thesis defense at the University of Chicago. When Milton Friedman uttered lines to the effect that Harry’s thesis has nothing wrong with it, but is not an economics dissertation, Dr. Friedman applied a very narrow definition of economics. Harry is acknowledged as a (the) creator of Portfolio Theory. His dissertation was its genesis.

In this paper we consider the value of Google Trends search data for nowcasting (and forecasting) GDP growth for a developed economy (the U.S.) and an emerging-market economy (Brazil). Our focus is on the marginal contribution of big data in the form of Google Trends data over and above that of traditional predictors, and we use a dynamic factor model to handle the large number of potential predictors and the “ragged-edge” problem. We find that factor models based on economic indicators and Google “categories” data provide gains compared to models that exclude this information. The benefits of using Google Trends data appear to be broadly similar for Brazil and the U.S., and depend on the factor model variable-selection strategy. Using more disaggregated Google Trends data than its “categories” is not beneficial.

Demand forecasting is becoming increasingly important as firms launch new products with short life cycles more frequently. This paper provides a framework based on state-of-the-art techniques that enables firms to use quantitative methods to forecast sales of newly launched, short-lived products that are similar to previous products when there is limited availability of historical sales data for the new product. In addition to exploiting historical data using time-series clustering, we perform data augmentation to generate sufficient sales data and consider two quantitative cluster assignment methods. We apply one traditional statistical (ARIMAX) and three machine learning methods based on deep neural networks (DNNs) – long short-term memory, gated recurrent units, and convolutional neural networks. Using two large data sets, we investigate the forecasting methods’ comparative performance and, for the larger data set, show that clustering generally results in substantially lower forecast errors. Our key empirical finding is that simple ARIMAX considerably outperforms the more advanced DNNs, with mean absolute errors up to 21%–24% lower. However, when adding Gaussian white noise in our robustness analysis, we find that ARIMAX’s performance deteriorates dramatically, whereas the considered DNNs display robust performance. Our results provide insights for practitioners on when to use advanced deep learning methods and when to use traditional methods.

Transition probabilities between delinquency states play a key role in determining the risk profile of a lending portfolio. Stress testing and IFRS9 are topics widely discussed by academics and practitioners. In this paper, we combine dynamic multi-state models and macroeconomic scenarios to estimate a stress testing model that forecasts delinquency states and transition probabilities at the borrower level for a mortgage portfolio. For the first time, a delinquency multi-state model is estimated for residential mortgages. We explicitly analyse and control for repeated events, an aspect previously not considered in credit risk multi-state models. Furthermore, we enhance the existing methodology by estimating scenario-specific forecasts beyond the lag of time-dependent covariates. We find that the number of previous transitions have a significant impact on the level of the transition probabilities, that severe economic conditions affect younger vintages the most, and that the relative impact of the stress scenario differs by attributes observed at origination.

This study develops a shrinkage method, LASSO with a Markov regime-switching model (MRS-LASSO), to predict US stock market volatility. A set of 17 well-known macroeconomic and financial factors are used. The out-of-sample results reveal that the MRS-LASSO model yields statistically and economically significant volatility predictions. We further investigate the predictability of MRS-LASSO with respect to different market conditions, business cycles, and variable selection. Three factors (equity market returns, a short-term reversal factor, and a consumer sentiment index) are the most frequent predictors. To investigate the practical implications, we construct the expected variance risk premium (VRP) by using volatility forecasts generated from the LASSO and MRS-LASSO models to forecast future stock returns and find that those models are also powerful.

Standard methods for forecasting electricity loads are not robust to cyberattacks on electricity demand data, potentially leading to severe consequences such as major economic loss or a system blackout. Methods are required that can handle forecasting under these conditions and detect outliers that would otherwise go unnoticed. The key challenge is to remove as many outliers as possible while maintaining enough clean data to use in the regression. In this paper we investigate robust approaches with data-driven tuning parameters, and in particular present an adaptive trimmed regression method that can better detect outliers and provide improved forecasts. In general, data-driven approaches perform much better than their fixed tuning parameter counterparts. Recommendations for future work are provided.

During the Covid-19 pandemic, economists have struggled to obtain reliable economic predictions, with standard models becoming outdated and their forecasting performance deteriorating rapidly. This paper presents two novelties that could be adopted by forecasting institutions in unconventional times. The first innovation is the construction of an extensive data set for macroeconomic forecasting in Europe. We collect more than a thousand time series from conventional and unconventional sources, complementing traditional macroeconomic variables with timely big data indicators and assessing their added value at nowcasting. The second novelty consists of a methodology to merge an enormous amount of non-encompassing data with a large battery of classical and more sophisticated forecasting methods in a seamlessly dynamic Bayesian framework. Specifically, we introduce an innovative “selection prior” that is used not as a way to influence model outcomes, but as a selection device among competing models. By applying this methodology to the Covid-19 crisis, we show which variables are good predictors for nowcasting gross domestic product and draw lessons for dealing with possible future crises.

Credit scoring is an important tool to guard against commercial risks for banks and lending companies and provides good conditions for the construction of individual personal credit. Ensemble algorithms have shown appealing progress for the improvement of credit scoring. In this study, to meet the challenge of large-scale credit scoring, we propose a heterogeneous deep forest model (Heter-DF), which is established based on considerations ranging from base learner selection, encouragement of the diversity of base learners, and ensemble strategies, for credit scoring. Heter-DF is designed as a scalable cascading framework that can increase its complexity with the scale of the credit dataset. Moreover, each level of Heter-DF is built by multiple heterogeneous tree-based ensembled base learners, avoiding the homogeneous prediction of the ensemble framework. In addition, a weighted voting mechanism is introduced to highlight important information and suppress irrelevant features, making Heter-DF a robust model for credit scoring. Experimental results on four credit scoring datasets and six evaluation metrics show that the cascading framework a good choice for the ensemble of tree-based base learners. A comparison among homogeneous ensembles and heterogeneous ensembles further demonstrates the effectiveness of Heter-DF. Experiments on different training sets indicate that Heter-DF is a scalable framework which not only deals with large-scale credit scoring but also satisfies the condition where small-scale credit scoring is desirable. Finally, based on the good interpretability of a tree-based structure, the global interpretation of Heter-DF is preliminarily explored.

We examine the internal consistency of US households’ inflation expectations reported as point and density forecasts by the New York Fed’s Survey of Consumer Expectations. We find that the majority of the households report well-defined histograms, with their central tendencies close to the corresponding point forecasts. We observe higher levels of consistency in forecasts reported by survey respondents with higher levels of income, education, and financial literacy. Furthermore, our results suggest that both the point forecasts directly reported and those derived from the histograms are more accurate when they are from respondents who are more likely to report consistent forecasts. In addition, we find that the consensus derived using only the consistent forecasts is as accurate as the consensus derived using all forecasts.