RESEARCH ARTICLE
Forecasting the duration of short‐term deflation episodes
Wojciech Charemza
1
| Svetlana Makarova
1,2
| Yinkai Wu
3
1
VCAS, Vistula University, Warsaw,
Poland
2
SSES, University College London,
London, UK
3
Anhui Provincial Academy of Social
Sciences, Hefei, China
Correspondence
Svetlana Makarova, School of Slavonic and
East European Studies, University College
London, 16 Taviton Street, London WC1H
0BW, UK.
Email: s.makarova@ucl.ac.uk
Funding information
Opus 8, Grant/Award Number: 2014/15/
B/HS4/04263; ESRC/ORA, Grant/Award
Number: RES‐360‐25‐0003
Abstract
The paper proposes a simulation‐based approach to multistep probabilistic
forecasting, applied for predicting the probability and duration of negative
inflation. The essence of this approach is in counting runs simulated from a
multivariate distribution representing the probabilistic forecasts, which enters
the negative inflation regime. The marginal distributions of forecasts are
estimated using the series of past forecast errors, and the joint distribution is
obtained by a multivariate copula approach. This technique is applied for esti-
mating the probability of negative inflation in China and its expected duration,
with the marginal distributions computed by fitting weighted skew‐normal
and two‐piece normal distributions to autoregressive moving average ex post
forecast errors and using the multivariate Student t copula.
KEYWORDS
duration forecast, inflation forecasting, multivariate copula analysis, negative inflation, probabilistic
forecasting, simulation
1 | INTRODUCTION
The paper proposes a simple, albeit computationally
intensive, way of analyzing the joint distributions of
multihorizon probabilistic forecasts. The focus is on evalu-
ating the out‐of‐sample duration forecast. The concept is
illustrated by estimating the duration of negative inflation
in China.
There is already a huge body of literature on forecasting
the duration of events. In economics, the most popular
approaches are grounded within extreme value theory
(e.g., Gilli & Këllezi, 2006) and proportional hazard dura-
tion modeling. The latter is a development from survival
modeling, and is often used for analyzing the duration of
unemployment (e.g., Bover, Arellano, & Bentolila, 2002;
Lancaster, 1979). For a discussion on other approaches
see Men, Kolankiewicz, and Wirjanto (2015).
However, these methods do not seem to be fully
appropriate for forecasting the duration of short‐term
deflation episodes, defined by negative inflation. In this
paper, we regard deflation in a purely statistical sense,
as a decline in the average (weighted) level of consumers’
prices, without addressing its possible relation to aggre-
gate demand (see, e.g., Atkeson & Kehoe, 2004; Benhabib
& Spigel, 2009; for another approach to estimation of the
probability of deflation, based on data from inflation sur-
veys and inflation swap rates, see Fleckenstein, Longstaff,
& Lustig, 2017).
Firstly, historical episodes of negative inflation are
infrequent or even nonexistent for most countries, which
makes estimating from its historical appearance ineffi-
cient or actually impossible. Secondly, most of the propor-
tional hazard methods rely implicitly or explicitly on the
assumption of normality of the forecast distribution. With
inflation, this is evidently not a realistic assumption
(for the most recent evidence see Chaudhuri, Kim, &
Shin, 2016). The extreme value models are free of the
assumption of normality, but they usually rely on tight
and not easily testable assumptions (see, e.g., Kotz &
Nadarajah, 2000).
The approach we propose allows the probability and
duration of events (negative inflation in our case) to be
estimated, even if the events did not occur in the past. It
is grounded within multihorizon probability density
Received: 22 October 2016 Revised: 25 September 2017 Accepted: 2 December 2017
DOI: 10.1002/for.2514
Journal of Forecasting. 2018;37:475–488. Copyright © 2018 John Wiley & Sons, Ltd.wileyonlinelibrary.com/journal/for 475
@Phil_Robichaud
@deepthiw
@JoseServera