Comparing Density Forecasts Using Threshold- and Quantile-Weighted Scoring Rules

By Tilmann Gneiting, Roopesh Ranjan. 2011

515 citations recorded by the source corpus.

Subfields: Economics and Econometrics, Finance, Global and Planetary Change

View Paper: https://www.semanticscholar.org/paper/07b14f69099d1fdab8f09ff4e4155f379a5fcd8d

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Economics and Econometrics

In Finance

In Global and Planetary Change

Citations in the corpus

Bayesian structured additive distributional regression with an application to regional income inequality in Germany (2015, methodology): “the whole integral) but also to perform a quantile decomposition and to plot the mean quantile scores versus α in order to compare fits of specific quantiles (Gneiting and Ranjan, 2011).”
Generalised Density Forecast Combinations (2015, unknown): “As discussed by Diks et al. (2011), the weighted logarithmic scoring rule, w t ( y t +1 ) log q it ( y t +1 ), used in (36) is not proper (see also Gneiting & Ranjan (2011)); i.e., there can exist incorrect density forecasts that would receive a higher average score than the actual (true)…”
Copula Methods for Forecasting Multivariate Time Series (2013, background): “H0 : E h log c1 Ût; ̂1t log c2 Ût; ̂2t i = 0 (75) vs H1 : E h log c1 Ût; ̂1t log c2 Ût; ̂2t i > 0”
Forecasting risk with Markov-switching GARCH models:A large-scale performance study (2018, methodology): “…significance of the differences in the forecasting performances of two competing models, say models i and 12 We follow the implementation of Gneiting and Ranjan (2011) and compute wCRPS with the following approximation: wCRPSt ≈ zu − zl M − 1 M∑ m=1 w(zm)(F (zm |It−1) − I{yt ≤ zm})2, where…”
Nonlife Ratemaking and Risk Management with Bayesian Additive Models for Location, Scale and Shape (2013, background): “In such a case, it can be helpful to follow Gneiting and Ranjan (2011) who suggest a quantile decomposition of the quantile score given by −2 ∫ 1”
Ensemble size: How suboptimal is less than infinity? (2018, methodology): “Lerch et al. (2017) recommend using proper scores that emphasize the events of interest, such as the threshold-weighted CRPS (twCRPS: Gneiting and Ranjan, 2011).”
Using Entropic Tilting to Combine BVAR Forecasts With External Nowcasts (2014, unknown): “Studies such as Gneiting and Raftery (2007) and Gneiting and Ranjan (2011) discuss the advantages of the CRPS over other measures.”
COHERENCE AND ELICITABILITY (2013, methodology): “Gneiting and Ranjan (2011) propose a method to compare density forecasts with emphasis on different regions of interest, such as the center or the tails of the distributions.”
Comparing Possibly Misspecified Forecasts (2020, background): “A “scoring rule” (see, e.g., Gneiting and Ranjan 2011), is a loss function mapping the density or distribution forecast and the realization to a measure of gain/loss.”
Conditional transformation models (2012, methodology): “The quantile score representation of the continuous ranked probability score (see Gneiting and Ranjan 2011) might be a basis to develop a boosting technique similar to the one described in this paper for the estimation of full conditional quantile functions.”
Real-Time Inflation Forecasting in a Changing World (2009, unknown): “Gneiting and Ranjan (2011) showed that it is invalid to use weighted log scores to emphasize certain areas of the distribution (such as the tails) in density forecast evaluation.”
Bayesian Structured Additive Distributional Regression (2013, background): “In many applications, e.g. insurances [Klein et al., 2013a, Heller et al., 2006] or weather forecasts [Gneiting and Ranjan, 2011], distributions with point masses at zero are of great interest.”
Spatiotemporal wildfire modeling through point processes with moderate and extreme marks (2021, methodology): “We also considered the threshold-weighted CRPS (twCRPS) (Gneiting & Ranjan 2011), with weight given by the indicator 1y≥100ha, in order to disregard prediction performance below 100 ha.”
Probabilistic Forecasting with Spline Quantile Function RNNs (2019, methodology): “This can easily be incorporated by using a quantile-weighted version of CRPS, as proposed by Gneiting and Ranjan (2011).”