Forecast Metrics Research Articles

Heat Waves in India during MAM 2019: Verification of ensemble based probabilistic forecasts and impact of bias correction

This paper deals with the analysis of the ability of the operational global ensemble prediction system (EPS) at the National Centre for Medium Range Weather Forecasting (NCMRWF, NEPS in predicting the probabilities of the maximum 2 m temperature (TMAX). The mean of the TMAX forecasts from NEPS are bias corrected (1st moment) using the technique of Decaying Average. This method does not lead to any correction in the spread of the EPS. Therefore the method of Variation Inflation is used to correct the spread (2nd moment) of NEPS. The forecasts of TMAX from the raw and the bias corrected data are then compared for March to May 2019 using standard verification metrics for probabilistic forecasts like Brier Score (BS), Brier Skill Score (BSS), ROC and Reliability Diagrams as well as the Value score. A case by case comparison of several heat wave cases during March to May 2019 is also performed in order to assess the day-to-day performance of the model.The forecasts after bias correction in the mean (1st moment) show a large improvement in terms of reduced false alarms and increased hit rate. However these forecasts show a reduced reliability in terms of under forecasting events with low probabilities and over forecasting events with high probabilities. The analysis of the forecast reliability after the correction of both the mean and the spread (1st and 2nd moments) shows an improvement in reliability. The forecasts with correction in the spread show a slight decrease in the ROC (i.e., increased false alarms) as compared to forecasts with just the corrected mean but the improvement in the reliability is appreciable. Thus the study demonstrates improved skill in probabilistic forecasts of TMAX over India during MAM 2019.

AR-GARCH-EVT-Copula for securitised real estate: an approach to improving risk forecasts?

ABSTRACT This study presents a quantitative analysis of the so-called AR-GARCH-EVT-Copula model aimed at forecasting risk metrics for multi-asset portfolios, including securitised real estate positions. The model incorporates a non-linear dependence structure and time-varying volatility in asset returns. Accordingly, an empirical study using data from six major global markets is carried out. The approach is applied to forecast risk metrics, in comparison to classical methods like historical simulation and variance-covariance models. Forecasts are then compared with realised returns, to calculate hit sequences and conduct statistical interference on the respective models. It is empirically shown that, the AR-GARCH-EVT-Copula model provides a superior forecast concerning risk metrics. This is mainly due to the usage of copulas, allowing us to individually model the dependence structure of random variables. Back testing and test results confirm the superiority of our model in comparison with classic methods such as historical simulation and Variance-Covariance approach. The decomposition of the univariate and multivariate models of the target model reveal the necessity to allow for high order and thus long-lasting autoregressive modelling as well as asymmetric tail dependence and rotated copulae across different portfolios.

Forecast combination approach in the loss given default estimation

ABSTRACT This paper examines a novel method of including macroeconomic variables into Loss Given Default models. The approach is transparent, and it easily translates changes in the overall credit environment into Expected Loss estimates, which is one of the crucial points that was recently introduced in the International Financial Reporting Standard 9. We propose a forecast combination procedure that separates the contract-based variables from the macroeconomic indicators. Two models are prepared and benchmarked to a single ordinary least-squares (OLS) model. To combine the forecasts we use three approaches: simple average, the Granger–Ramanathan Method, and Mallows Model Averaging. We tested our predictions on out-of-time data and found that the forecast combination outperforms the single OLS model in terms of the selected forecast quality metrics.

A Gray‐Box Model for a Probabilistic Estimate of Regional Ground Magnetic Perturbations: Enhancing the NOAA Operational Geospace Model With Machine Learning

AbstractWe present a novel algorithm that predicts the probability that the time derivative of the horizontal component of the ground magnetic field dB/dt exceeds a specified threshold at a given location. This quantity provides important information that is physically relevant to geomagnetically induced currents (GICs), which are electric currents associated with sudden changes in the Earth's magnetic field due to space weather events. The model follows a “gray‐box” approach by combining the output of a physics‐based model with machine learning. Specifically, we combine the University of Michigan's Geospace model that is operational at the National Oceanic and Atmospheric Administration (NOAA) Space Weather Prediction Center, with a boosted ensemble of classification trees. We discuss the problem of recalibrating the output of the decision tree to obtain reliable probabilities. The performance of the model is assessed by typical metrics for probabilistic forecasts: Probability of Detection and False Detection, True Skill Statistic, Heidke Skill Score, and Receiver Operating Characteristic curve. We show that the ML‐enhanced algorithm consistently improves all the metrics considered.

Ecological Forecasts Reveal Limitations of Common Model Selection Methods: Predicting Changes in Beaver Colony Densities

Over the past two decades, there have been numerous calls to make ecology a more predictive science through direct empirical assessments of ecological models and predictions. While the widespread use of model selection using information criteria has pushed ecology toward placing a higher emphasis on prediction, few attempts have been made to validate the ability of information criteria to correctly identify the most parsimonious model with the greatest predictive accuracy. Here, we used an ecological forecasting framework to test the ability of information criteria to accurately predict the relative contribution of density dependence and density-independent factors (forage availability, harvest, weather, wolf [Canis lupus] density) on inter-annual fluctuations in beaver (Castor canadensis) colony densities. We modeled changes in colony densities using a discrete-time Gompertz model, and assessed the performance of four models using information criteria values: density-independent models with (1) and without (2) environmental covariates; and density-dependent models with (3) and without (4) environmental covariates. We then evaluated the forecasting accuracy of each model by withholding the final one-third of observations from each population and compared observed vs. predicted densities. Information criteria and our forecasting accuracy metrics both provided strong evidence of compensatory density dependence in the annual dynamics of beaver colony densities. However, despite strong within-sample performance by the most complex model (density-dependent with covariates) as determined using information criteria, hindcasts of colony densities revealed that the much simpler density-dependent model without covariates performed nearly as well predicting out-of-sample colony densities. The hindcast results indicated that the complex model over-fit our data, suggesting that parameters identified by information criteria as important predictor variables are only marginally valuable for predicting landscape-scale beaver colony dynamics. Our study demonstrates the importance of evaluating ecological models and predictions with long-term data and revealed how a known limitation of information criteria (over-fitting of complex models) can affect our interpretation of ecological dynamics. While incorporating knowledge of the factors that influence animal population dynamics can improve population forecasts, we suggest that comparing forecast performance metrics can likewise improve our knowledge of the factors driving population dynamics.

Systematic Evaluation of the Impact of Assimilating a Network of Ground-Based Remote Sensing Profilers for Forecasts of Nocturnal Convection Initiation during PECAN

AbstractNocturnal convection is often initiated by mechanisms that cannot be easily observed within the large gaps between rawinsondes or by conventional surface networks. To improve forecasts of such events, we evaluate the systematic impact of assimilating a collocated network of high-frequency, ground-based thermodynamic and kinematic profilers collected as part of the 2015 Plains Elevated Convection At Night (PECAN) experiment. For 13 nocturnal convection initiation (CI) events, we find small but consistent improvements when assimilating thermodynamic observations collected by Atmospheric Emitted Radiance Interferometers (AERIs). Through midlevel cooling and moistening, assimilating the AERIs increases the fractions skill score (FSS) for both nocturnal CI and precipitation forecasts. The AERIs also improve various contingency metrics for CI forecasts. Assimilating composite kinematic datasets collected by Doppler lidars and radar wind profilers (RWPs) results in slight degradations to the forecast quality, including decreases in the FSS and traditional contingency metrics. The impacts from assimilating thermodynamic and kinematic profilers often counteract each other, such that we find little impact on the detection of CI when both are assimilated. However, assimilating both datasets improves various properties of the CI events that are successfully detected (timing, distance, shape, etc.). We also find large variability in the impact of assimilating these remote sensing profilers, likely due to the number of observing sites and the strength of the synoptic forcing for each case. We hypothesize that the lack of flow-dependent methods to diagnose observation errors likely contributes to degradations in forecast skill for many cases, especially when assimilating kinematic profilers.

Beyond univariate calibration: verifying spatial structure in ensembles of forecast fields

Abstract. Most available verification metrics for ensemble forecasts focus on univariate quantities. That is, they assess whether the ensemble provides an adequate representation of the forecast uncertainty about the quantity of interest at a particular location and time. For spatially indexed ensemble forecasts, however, it is also important that forecast fields reproduce the spatial structure of the observed field and represent the uncertainty about spatial properties such as the size of the area for which heavy precipitation, high winds, critical fire weather conditions, etc., are expected. In this article we study the properties of the fraction of threshold exceedance (FTE) histogram, a new diagnostic tool designed for spatially indexed ensemble forecast fields. Defined as the fraction of grid points where a prescribed threshold is exceeded, the FTE is calculated for the verification field and separately for each ensemble member. It yields a projection of a – possibly high-dimensional – multivariate quantity onto a univariate quantity that can be studied with standard tools like verification rank histograms. This projection is appealing since it reflects a spatial property that is intuitive and directly relevant in applications, though it is not obvious whether the FTE is sufficiently sensitive to misrepresentation of spatial structure in the ensemble. In a comprehensive simulation study we find that departures from uniformity of the FTE histograms can indeed be related to forecast ensembles with biased spatial variability and that these histograms detect shortcomings in the spatial structure of ensemble forecast fields that are not obvious by eye. For demonstration, FTE histograms are applied in the context of spatially downscaled ensemble precipitation forecast fields from NOAA's Global Ensemble Forecast System.

Machine Learning and Deep Learning Methods in Mining Operations: a Data-Driven SAG Mill Energy Consumption Prediction Application

Semi-autogenous grinding mills play a critical role in the processing stage of many mining operations. They are also one of the most intensive energy consumers of the entire process. Current forecasting techniques of energy consumption base their inferences on feeding ore mineralogical features, SAG dimensions, and operational variables. Experts recognize their capabilities to provide adequate guidelines but also their lack of accuracy when real-time forecasting is desired. As an alternative, we propose the use of real-time operational variables (feed tonnage, bearing pressure, and spindle speed) to forecast the upcoming energy consumption via machine learning and deep learning techniques. Several predictive methods were studied: polynomial regression, k-nearest neighbor, support vector machine, multilayer perceptron, long short-term memory, and gated recurrent units. A step-by-step workflow on how to deal with real datasets, and how to find optimum models and final model selection is presented. In particular, recurrent neural networks achieved the best forecasting metrics in the energy consumption prediction task. The workflow has the potential of being extended to any other temporal and multivariate mineral processing datasets.

An Evaluation of a Hybrid, Terrain-Following Vertical Coordinate in the WRF-Based RAP and HRRR Models

Abstract A new hybrid, sigma-pressure vertical coordinate was recently added to the Weather Research and Forecasting (WRF) Model in an effort to reduce numerical noise in the model equations near complex terrain. Testing of this hybrid, terrain-following coordinate was undertaken in the WRF-based Rapid Refresh (RAP) and High-Resolution Rapid Refresh (HRRR) models to assess impacts on retrospective and real-time simulations. Initial cold-start simulations indicated that the majority of differences between the hybrid and traditional sigma coordinate were confined to regions downstream of mountainous terrain and focused in the upper levels. Week-long retrospective simulations generally resulted in small improvements for the RAP, and a neutral impact in the HRRR when the hybrid coordinate was used. However, one possibility is that the inclusion of data assimilation in the experiments may have minimized differences between the vertical coordinates. Finally, analysis of turbulence forecasts with the new hybrid coordinate indicate a significant reduction in spurious vertical motion over the full length of the Rocky Mountains. Overall, the results indicate a potential to improve forecast metrics through implementation of the hybrid coordinate, particularly at upper levels, and downstream of complex terrain.

Ensemble-Based Targeted Observation Method Applied to Radar Radial Velocity Observations on Idealized Supercell Low-Level Rotation Forecasts: A Proof of Concept

Abstract The potential future installation of a multifunction phased-array radar (MPAR) network will provide capabilities of case-specific adaptive scanning. Knowing the impacts adaptive scanning may have on short-term forecasts will influence scanning strategy decision-making in hopes to produce the most optimal ensemble forecast while also benefiting human severe weather warning decision-making. An ensemble-based targeted observation algorithm is applied to an observing system simulation experiment (OSSE) where the impacts of synthetic idealized supercell radial velocity observations are estimated before the observations are “collected” and assimilated. The forecast metric of interest is the low-level rotation forecast metric (0–1-km updraft helicity), a surrogate for tornado prediction. It is found that the ensemble-based targeted observation approach can reasonably estimate the true error variance reduction when an effective method that treats sampling error is applied, the period of model forecast is associated with less degrees of nonlinearity, and the observation information content relative to the background forecast is larger. In some scenarios, a subset of a full-volume scan assimilation produces better forecasts than all observations within the full volume. Assimilating the full-volume scan increases the number of potential spurious correlations arising between the forecast metric and radial velocity observation induced state perturbations, which may degrade the forecast metric accuracy.

Forecasting U.S. stock returns

We forecast quarterly US stock returns using a breadth of forecast variables, methods and metrics, including linear and non-linear regressions, rolling and recursive techniques, forecast combinations and statistical and economic evaluation. Thus, extending research in terms of the range of predictor series and the scope of analysis. Consistent with much of literature, a broad view over the full set of predictor variables indicates that such models are unable to beat the historical mean model. However, nuances reveal forecast success varies according to how the forecasts are evaluated and over time. Results reveal that the term structure of interest rates consistently provides the preferred forecast performance, especially when evaluated using the Sharpe ratio. The purchasing managers index also consistently provides a strong forecast performance. Further results reveal that forecast combinations over the full set of variables do not outperform the preferred single variable forecasts, while an interest rate forecast combination subset does perform well. The success of the term structure and the purchasing managers index highlights the importance of, respectively, investor and firm expectations of future economic performance in providing valuable stock return forecasts and is consistent with asset pricing models that indicate movements in returns are conditioned by such expectations.

Systematic Comparison of Convection-Allowing Models during the 2017 NOAA HWT Spring Forecasting Experiment

Abstract The 2016–18 NOAA Hazardous Weather Testbed (HWT) Spring Forecasting Experiments (SFE) featured the Community Leveraged Unified Ensemble (CLUE), a coordinated convection-allowing model (CAM) ensemble framework designed to provide empirical guidance for development of operational CAM systems. The 2017 CLUE included 81 members that all used 3-km horizontal grid spacing over the CONUS, enabling direct comparison of forecasts generated using different dynamical cores, physics schemes, and initialization procedures. This study uses forecasts from several of the 2017 CLUE members and one operational model to evaluate and compare CAM representation and next-day prediction of thunderstorms. The analysis utilizes existing techniques and novel, object-based techniques that distill important information about modeled and observed storms from many cases. The National Severe Storms Laboratory Multi-Radar Multi-Sensor product suite is used to verify model forecasts and climatologies of observed variables. Unobserved model fields are also examined to further illuminate important intermodel differences in storms and near-storm environments. No single model performed better than the others in all respects. However, there were many systematic intermodel and intercore differences in specific forecast metrics and model fields. Some of these differences can be confidently attributed to particular differences in model design. Model intercomparison studies similar to the one presented here are important to better understand the impacts of model and ensemble configurations on storm forecasts and to help optimize future operational CAM systems.

Can wave coupling improve operational regional ocean forecasts for the north-west European Shelf?

Abstract. Operational ocean forecasts are typically produced by modelling systems run using a forced mode approach. The evolution of the ocean state is not directly influenced by surface waves, and the ocean dynamics are driven by an external source of meteorological data which are independent of the ocean state. Model coupling provides one approach to increase the extent to which ocean forecast systems can represent the interactions and feedbacks between ocean, waves, and the atmosphere seen in nature. This paper demonstrates the impact of improving how the effect of waves on the momentum exchange across the ocean–atmosphere interface is represented through ocean–wave coupling on the performance of an operational regional ocean prediction system. This study focuses on the eddy-resolving (1.5 km resolution) Atlantic Margin Model (AMM15) ocean model configuration for the north-west European Shelf (NWS) region. A series of 2-year duration forecast trials of the Copernicus Marine Environment Monitoring Service (CMEMS) north-west European Shelf regional ocean prediction system are analysed. The impact of including ocean–wave feedbacks via dynamic coupling on the simulated ocean is discussed. The main interactions included are the modification of surface stress by wave growth and dissipation, Stokes–Coriolis forcing, and wave-height-dependent ocean surface roughness. Given the relevance to operational forecasting, trials with and without ocean data assimilation are considered. Summary forecast metrics demonstrate that the ocean–wave coupled system is a viable evolution for future operational implementation. When results are considered in more depth, wave coupling was found to result in an annual cycle of relatively warmer winter and cooler summer sea surface temperatures for seasonally stratified regions of the NWS. This is driven by enhanced mixing due to waves, and a deepening of the ocean mixed layer during summer. The impact of wave coupling is shown to be reduced within the mixed layer with assimilation of ocean observations. Evaluation of salinity and ocean currents against profile measurements in the German Bight demonstrates improved simulation with wave coupling relative to control simulations. Further, evidence is provided of improvement to simulation of extremes of sea surface height anomalies relative to coastal tide gauges.

Probabilistic forecasting of high-resolution clear-sky index time-series using a Markov-chain mixture distribution model

This study presents a Markov-chain mixture (MCM) distribution model for forecasting the clear-sky index—normalized global horizontal irradiance. The model is presented in general, but applied to, and tested on, minute resolution clear-sky index data for the two different climatic regions of Norrköping, Sweden, and Hawaii, USA. Model robustness is evaluated based on a cross-validation procedure and on that basis a reference configuration of parameter settings for evaluating the model performance is obtained. Simulation results are compared with persistence ensemble (PeEn) and quantile regression (QR) model simulations for both data sets and for D=1,…,5 steps ahead forecasting scenarios. The results are evaluated by a set of probabilistic forecasting metrics: reliability mean absolute error (reliability MAE), prediction interval normalized average width (PINAW), continuous ranked probability score (CRPS) and continuous ranked probability skill score (skill). Both in terms of reliability MAE and CRPS, the MCM model outperforms PeEn for all simulated scenarios. In terms of reliability MAE, the QR model outperforms the MCM model for most simulated scenarios. However, in terms of mean CRPS, the MCM model outperforms the QR model in most simulated scenarios. A point forecasting estimate is also provided. The MCM model is concluded to be a computationally inexpensive, accurate and parameter insensitive probabilistic model. Based on this, it is suggested as a candidate benchmark model in probabilistic forecasting, in particular for solar irradiance forecasting. For applicability, a Python script of the MCM model is available as SheperoMah/MCM-distribution-forecasting at GitHub.

Are Fund Alphas Robust Proxies for Funds Managers' Managerial Skill?

This study provides formal theoretical evidence that constructions of fund alpha that are implemented using robust specifications of asset pricing models generate alpha estimates that are well defined. Regardless, the formal theoretical model shows fund alphas that are constructed with the market portfolio as benchmark (`global alphas') cannot be construed to be robust metrics for either of stability of fund performance, or expected aggregate future fund performance. Suppose benchmarks for distinct realizations (echelons) of risk tolerance are substituted for the market portfolio. Formal predictions show alphas constructed in relation to such benchmarks (`echelon-specific alphas') have exactly the same functional form as global alphas, as such while they are well defined, they are not robust metrics for global forecasts of funds' expected future performance. This is evident in the formally derived theoretical prediction that long positions in all zero alpha funds have feasibility of superior performance in relation to long positions in all positive alpha funds. Suppose, however, that investors restrict their sample of funds to a specific echelon of risk tolerance. In presence of managerial capacity for taking advantage of stock market imperfections, echelon-specific fund alphas either vary stochastically and asymptotically around zero, or are positive. In presence stated restriction and caveat, echelon-specific alphas are robust metrics for expected future performance of managed funds.

Progress in Forecast Skill at Three Leading Global Operational NWP Centers during 2015–17 as Seen in Summary Assessment Metrics (SAMs)

Abstract The summary assessment metric (SAM) method is applied to an array of primary assessment metrics (PAMs) for the deterministic forecasts of three leading numerical weather prediction (NWP) centers for the years 2015–17. The PAMs include anomaly correlation, RMSE, and absolute mean error (i.e., the absolute value of bias) for different forecast times, vertical levels, geographic domains, and variables. SAMs indicate that in terms of forecast skill ECMWF is better than NCEP, which is better than but approximately the same as UKMO. The use of SAMs allows a number of interesting features of the evolution of forecast skill to be observed. All three centers improve over the 3-yr period. NCEP short-term forecast skill substantially increases during the period. Quantitatively, the effect of the 11 May 2016 NCEP upgrade to the four-dimensional ensemble variational data assimilation (4DEnVar) system is a 7.37% increase in the probability of improved skill relative to a randomly chosen forecast metric from 2015 to 2017. This is the largest SAM impact during the study period. However, the observed impacts are within the context of slowly improving forecast skill for operational global NWP as compared to earlier years. Clearly, the systems lagging ECMWF can improve, and there is evidence from SAMs in addition to the 4DEnVar example that improvements in forecast and data assimilation systems are still leading to forecast skill improvements.

Long Short Term Memory Recurrent Network for Standard and Poor’s 500 Index Modelling

Forecasting time-series data are imperative especially when planning is required through modelling using uncertain knowledge of future events. Recurrent neural network models have been applied in the industry and outperform standard artificial neural networks in forecasting, but fail in long term time-series forecasting due to the vanishing gradient problem. This study offers a robust solution that can be implemented for long-term forecasting using a special architecture of recurrent neural network known as Long Short Term Memory (LSTM) model to overcome the vanishing gradient problem. LSTM is specially designed to avoid the long-term dependency problem as their default behavior. Empirical analysis is performed using quantitative forecasting metrics and comparative model performance on the forecasted outputs. An evaluation analysis is performed to validate that the LSTM model provides better forecasted outputs on Standard & Poor’s 500 Index (S&P 500) in terms of error metrics as compared to other forecasting models.

Object-Based Verification of a Prototype Warn-on-Forecast System

An object-based verification methodology for the NSSL Experimental Warn-on-Forecast System for ensembles (NEWS-e) has been developed and applied to 32 cases between December 2015 and June 2017. NEWS-e forecast objects of composite reflectivity and 30-minute rotation tracks of updraft helicity are matched to corresponding objects in Multi-Radar Multi-Sensor data on space and time scales typical of a National Weather Service warning. Object matching allows contingency table-based verification statistics to be used to establish baseline performance metrics for NEWS-e thunderstorm and mesocyclone forecasts. NEWS-e critical Success Index (CSI) scores of reflectivity (updraft helicity) forecasts decrease from approximately 0.7 (0.4) to 0.4 (0.2) over 3 hours of forecast time. CSI scores decrease through the forecast period, indicating that errors have not saturated and skill is retained at 3 hours of forecast time. Lower verification scores for rotation track forecasts are primarily a result of a high frequency bias. Comparison of different system configurations used in 2016 and 2017 show an increase in skill for 2017 reflectivity forecasts, attributable mainly to improvements in the forecast initial condition. A small decrease in skill in 2017 rotation track forecasts is likely a result of sample differences between 2016 and 2017. Although large case-to-case variation is present, evidence is found that NEWS-e forecast skill improves with increasing object size and intensity, as well as in mesoscale environments in which an enhanced or higher risk of severe thunderstorms was forecast.

Benefits of assimilating SAPHIR observations on analysis and forecasts of tropical fields in the Met Office global model

The Sondeur Atmosphérique du Profil d'Humidité Intertropicale par Radiométrie (SAPHIR) instrument provides improved sampling of tropical atmospheric moisture vertically, horizontally and temporally. The impact of these unique characteristics is investigated through: an idealised study of retrieved humidity profiles; single‐observation experiments; assimilation experiments in a global NWP system; and an investigation into the spin‐down of precipitation in the early phase of the forecast. SAPHIR offers improved performance over similar satellite instruments, and beneficial impacts were found in all investigations. When assimilated in conjunction with observations from the Advanced Microwave Scanning Radiometer imager AMSR‐2 the impact is further improved. Retrieval studies showed the errors in retrievals from SAPHIR were lower than those obtained from MHS or ATMS at all levels above 600 hPa. Single‐observation experiments showed that, when assimilated together with AMSR‐2, AMSR‐2‐driven humidity increments were modified to give more realistic vertical structure. In assimilation experiments employing a near‐operational configuration of the Met Office global model, the assimilation of clear‐sky SAPHIR data improved the root‐mean‐square errors of a number of forecast metrics, most notably temperature at 250 hPa (improved by 2%), relative humidity at 500 hPa (2%) and wind at 500 hPa (1%) at forecast lead times of 12 and 24 h. The results of this work form a clear recommendation for future remote‐sensing missions including both SAPHIR and AMSR‐2 channel configurations.

Confucius confusion: analyst forecast dispersion and business cycles

PurposeFinancial analysts have been found to be overconfident. The purpose of this paper is to study the ramifications of that overconfidence on the dispersion of earnings estimates as a predictor of the US business cycle.Design/methodology/approachWhether aggregate analyst forecast dispersion contains information about turning points in business cycles, especially downturns, is examined by utilizing the analyst earnings forecast dispersion metric. The primary analysis derives from logit regression and Markov switching models. The analysis controls for sentiment (consumer confidence), output (industrial production), and financial indicators (stock returns and turnover). Analyst data come from Institutional Brokers Estimate System, while the economic data are available at the Federal Reserve Bank of St Louis Economic Data site.FindingsA rise in the dispersion of analyst forecasts is a significant predictor of turning points in the US business cycle. Financial analyst uncertainty of earnings estimate contains crucial information about the risks of US business cycle turning points. The results are consistent with some analysts becoming overconfident during the expansion period and misjudging the precision of their information, thus over or under weighting various sources of information. This causes the disagreement among analysts measured as dispersion.Originality/valueThis is the first study to show that analyst forecast dispersion contributions valuable information to predictions of economic downturns. In addition, that dispersion can be attributed to analyst overconfidence.