Assessing Forecast Uncertainty : An Information Bayesian Approach
Regardless of the field, forecasts are widely used and yet assessments of the embedded uncertainty—the magnitude of the downside and upside risks of the prediction itself—are often missing. Particularly in policy-making and investment, accounting f...
| Main Author: | |
|---|---|
| Format: | Working Paper |
| Language: | English en_US |
| Published: |
World Bank, Washington, DC
2017
|
| Subjects: | |
| Online Access: | http://documents.worldbank.org/curated/en/802551502718519493/Assessing-forecast-uncertainty-an-information-Bayesian-approach http://hdl.handle.net/10986/27972 |
| Summary: | Regardless of the field, forecasts are
widely used and yet assessments of the embedded
uncertainty—the magnitude of the downside and upside risks
of the prediction itself—are often missing. Particularly in
policy-making and investment, accounting for these risks
around baseline predictions is of outstanding importance for
making better and more informed decisions. This paper
introduces a procedure to assess risks associated with a
random phenomenon. The methodology assigns probability
distributions to baseline-projections of an economic or
social random variable—for example gross domestic product
growth, inflation, population growth, poverty headcount, and
so forth—combining ex-post and ex-ante market information.
The generated asymmetric density forecasts use information
derived from surveys on expectations and implied statistics
of predictive models. The methodology also decomposes the
variance and skewness of the predictive distribution
accounting for the shares of selected risk factors. The
procedure relies on a Bayesian information-theoretical
approach, which allows the inclusion of judgment and
forecaster expertise. For reliability purposes and
transparency, the paper also evaluates the constructed
density forecasts assigning a score. The continuous ranked
probability score is used to assess the prediction accuracy
of elicited density forecasts. The selected score
incentivizes the forecaster to provide its true and best
predictive distribution. An empirical application to
forecast world gross domestic product growth is used to test
the Bayesian entropy methodology. Predictive variance and
skewness of world gross domestic product growth are
associated with ex-ante information of four risk factors:
term spreads, absolute deviations of headline inflation
targets, energy prices, and the Standard and Poor's 500
index prices. The Bayesian entropy technique is benchmarked
with naïve-generated density forecasts that utilize
information from historical forecast errors. The results
show that the Bayesian density forecasts outperform the
naïve-generated benchmark predictions, illustrating the
value added of the introduced methodology. |
|---|