Predictive Inference Based on Markov Chain Monte Carlo Output

Krüger, Fabian; Lerch, Sebastian; Thorarinsdottir, Thordis L.; Gneiting, Tilmann

doi:10.1111/insr.12405

Cited by 58 publications

(41 citation statements)

References 88 publications

(74 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We compute the CRPS with an empirical CDF estimated from samples of f i ( x ). For details on the numerical implementation of the CRPS for simulated forecasts, we refer to the corresponding literature (Jordan et al, 2019; Krüger et al, 2016). For the proportional response variable, relative loss, the CRPS is defined on the interval [0, 1] with the optimum at 0.…”

Section: Methodsmentioning

confidence: 99%

Probabilistic Flood Loss Models for Companies

Schoppa

Sieg

Vogel

et al. 2020

Water Resources Research

View full text Add to dashboard Cite

Flood loss modeling is a central component of flood risk analysis. Conventionally, this involves univariable and deterministic stage-damage functions. Recent advancements in the field promote the use of multivariable and probabilistic loss models, which consider variables beyond inundation depth and account for prediction uncertainty. Although companies contribute significantly to total loss figures, novel modeling approaches for companies are lacking. Scarce data and the heterogeneity among companies impede the development of company flood loss models. We present three multivariable flood loss models for companies from the manufacturing, commercial, financial, and service sector that intrinsically quantify prediction uncertainty. Based on object-level loss data (n = 1,306), we comparatively evaluate the predictive capacity of Bayesian networks, Bayesian regression, and random forest in relation to deterministic and probabilistic stage-damage functions, serving as benchmarks. The company loss data stem from four postevent surveys in Germany between 2002 and 2013 and include information on flood intensity, company characteristics, emergency response, private precaution, and resulting loss to building, equipment, and goods and stock. We find that the multivariable probabilistic models successfully identify and reproduce essential relationships of flood damage processes in the data. The assessment of model skill focuses on the precision of the probabilistic predictions and reveals that the candidate models outperform the stage-damage functions, while differences among the proposed models are negligible. Although the combination of multivariable and probabilistic loss estimation improves predictive accuracy over the entire data set, wide predictive distributions stress the necessity for the quantification of uncertainty. Plain Language Summary River floods are among the costliest natural disasters. The appraisal of financial flood loss is integral to flood risk analysis. Scientists and practitioners use stage-damage functions to assess flood loss from the water depth at an inundated building. However, flood loss is also controlled by other factors (e.g., building characteristics and private flood precaution), and stage-damage functions only infrequently provide information on their reliability. Therefore, researchers developed more complex flood loss models, which consider multiple variables and provide information on associated uncertainties. While these novel flood loss models exist for private households, they are lacking for companies. In this study, we present three complex flood loss models for companies. We employ company loss data from flood events in Germany to compare the skill of the proposed flood loss models to each other and to two stage-damage functions. Our results show that the complex models estimate company flood loss more accurately than the stage-damage functions. The three complex models work equally well. Yet their loss estimates remain fairly uncertain, underlining the importance of reliabilit...

show abstract

Section: Methodsmentioning

confidence: 99%

Probabilistic Flood Loss Models for Companies

Schoppa

Sieg

Vogel

et al. 2020

Water Resources Research

View full text Add to dashboard Cite

show abstract

“…30 Generally, across variables and forecast origins, point forecasts generated by SVO and SV-t are fairly close, as seen also in our comparison of forecast performance pre-COVID-19 in Table 2. 31 However, among our heteroskedastic VARs, stark differences emerge considering the uncertainty around forecasts made with and without outlier adjustments. As discussed in forecasts.…”

Section: Forecasts Made In 2020mentioning

confidence: 95%

“…Overall, both forms of outlier-adjusted SV generate broadly similar forecast densities over the course of 2020, with a few instances of sizable differences, such as with forecasts of real income made 30 These forecasts made in April jump off a reading for the unemployment rate of just under 15 percent. 31 For better readability, forecasts generated by SV are displayed on different scales in the top and bottom rows of panels shown in Figures 11-14. Similarly, the SVO forecast densities shown in the top-row panels of these figures are also shown in the middle-row panels of each figure.…”

Section: Forecasts Made In 2020mentioning

confidence: 99%

“…The forecast horizons considered extend from 1 to 24 months. We evaluate point and density forecasts based on root-mean-squared errors (RMSE) and continuous ranked probability scores (CRPS), respectively, as described in, among others,Clark and Ravazzolo (2015) andKrüger et al (2020). Statistical significance of differences in loss functions is evaluated using theDiebold and Mariano (1995) andWest (1996) test.Tables 2 and 3compare point and density forecasts generated by BVARs with SV and SVO specifications relative to those resulting from a homoskedastic BVAR.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Addressing COVID-19 Outliers in BVARs with Stochastic Volatility

et al. 2021

View full text Add to dashboard Cite

Incoming data in 2020 posed sizable challenges for the use of VARs in economic analysis: Enormous movements in a number of series have had strong effects on parameters and forecasts constructed with standard VAR methods. We propose the use of VAR models with time-varying volatility that include a treatment of the COVID extremes as outlier observations. Typical VARs with time-varying volatility assume changes in uncertainty to be highly persistent. Instead, we adopt an outlier-adjusted stochastic volatility (SV) model for VAR residuals that combines transitory and persistent changes in volatility. In addition, we consider the treatment of outliers as missing data. Evaluating forecast performance over the last few decades in quasi-real time, we find that the outlier-augmented SV scheme does at least as well as a conventional SV model, while both outperform standard homoskedastic VARs. Point forecasts made in 2020 from heteroskedastic VARs are much less sensitive to outliers in the data, and the outlier-adjusted SV model generates more reasonable gauges of forecast uncertainty than a standard SV model. At least pre-COVID, a close alternative to the outlier-adjusted model is an SV model with t-distributed shocks. Treating outliers as missing data also generates better-behaved forecasts than the conventional SV model. However, since uncertainty about the incidence of outliers is ignored in that approach, it leads to strikingly tight predictive densities.

show abstract

“…We use a Gaussian approximation for g to compute the LogS and an empirical CDF-based approximation for G to compute the CRPS (see Krüger et al (2019) for details).…”

Section: Scoring Rulesmentioning

confidence: 99%

Forecasting of Cohort Fertility Under a Hierarchical Bayesian Approach

Ellison

Dodd

Forster

2020

Journal of the Royal Statistical Society Series A: Statistics in Society

View full text Add to dashboard Cite

Fertility projections are a key determinant of population forecasts, which are widely used by government policymakers and planners. In keeping with the recent literature, we propose an intuitive and transparent hierarchical Bayesian model to forecast cohort fertility. Using Hamiltonian Monte Carlo methods and a dataset from the Human Fertility Database, we obtain fertility forecasts for 30 countries. We use scoring rules to quantitatively assess the predictive accuracy of the forecasts; these indicate that our model predicts with an accuracy comparable to that of the best-performing models in the current literature overall, with stronger performance for countries without a recent structural shift. Our findings support the position of hierarchical Bayesian modelling at the forefront of population forecasting methods.

show abstract

Predictive Inference Based on Markov Chain Monte Carlo Output

Cited by 58 publications

References 88 publications

Probabilistic Flood Loss Models for Companies

Probabilistic Flood Loss Models for Companies

Addressing COVID-19 Outliers in BVARs with Stochastic Volatility

Forecasting of Cohort Fertility Under a Hierarchical Bayesian Approach

Contact Info

Product

Resources

About