Price Level Dispersion versus Ination Rate

Dispersion: Evidence from Three Countries

David Fielding

Christopher Hajzler

James MacGee

Abstract
Aggregate ination potentially affects both the dispersion of price levels across lo-
cations (relative price variability, RPV) and the dispersion of ination rates (relative in-
ation variability, RIV). Some theory suggests that the RIV-ination relationship could
differ markedly from the RPV-ination relationship. However, most empirical studies
deal with RIV alone, and there is little evidence about how RIV-ination patterns differ
from RPV-ination patterns. Using price data from three countries, we show that the
patterns are very different. The effect of ination on social welfare therefore depends on
the relative importance of RPV and RIV in the social welfare function.
JEL classication: E31
Keywords: Relative price variability, Ination
This paper provides new evidence on the relationship between ination and dispersion in
relative prices across locations within a country. The evidence is based on data from different
historical periods in Canada, Japan and Nigeria. We use this data to explore whether the
relationship depends on the measure of dispersion, focusing on two alternative measures.
The rst measure, which is used more frequently in this large and growing literature, is the
variation across goods in the rate of price ination; we term this measure relative ination

Corresponding author. Address for correspondence: Department of Economics, University of Otago, PO

Box 56, Dunedin 9054, New Zealand. E-mail david.elding@otago.ac.nz; telephone +6434798653.

Department of Economics, University of Otago.

Department of Economics, University of Western Ontario.

The authors are grateful for helpful comments and suggestions from Nicolas Groshenny, Martin Berka, and
Steffen Lippert, as well as seminar participants at the Southern Workshop in Macroeconomics (Auckland, 2012)
and the University of Otago, New Zealand.
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 2
variability (RIV). The second measure, which has received less attention in the literature, is
the variation in relative price levels; we term this measure relative price variability (RPV).
1
The empirical focus on RIV was originally motivated by theories emphasising the impor-
tance of signal extraction problems (Lucas Jr, 1973; Barro, 1976). Using a signal extraction
model, it can be shown that positive or negative ination shocks increase both RIV and RPV
(Parks, 1978). However, other theories imply that the RIV-ination relationship will be very
different from the RPV-ination relationship. This is the case with menu cost models and
with several models of costly consumer search: for example, if there is a sudden increase
in ination following a period of stable and and homogeneous ination rates across loca-
tions, consumers may be motivated to search more intensely for lower prices. This can cause
relative prices to converge and relative ination rates to diverge.
The number of empirical papers that investigate the effects of ination on RPV is rel-
atively small. As noted by Parsley (1996), this is partly because the construction of RPV
gures requires detailed price level data for different products and locations. In the absence
of such detailed data, many authors use price index series instead, which limits the analy-
sis to measures of RIV.
2
The few papers that have compared the impact of ination on RIV
with its impact on RPV have reached seemingly different conclusions. Parsley (1996) and
Tommasi (1993), using US and Argentinean data, both nd some evidence that RPV and
RIV are positively related to aggregate ination. However Reinsdorf (1994), using monthly
prices for food in 9 US cities between 1980 and 1982, nds a negative average relationship
between RPV and unanticipated ination. By contrast, the estimated RIV-ination relation-
ship is V-shaped.
3
As Reinsdorf (1994) admits, the differences between his results and those
1
Many papers refer to RIV as relative price variability or relative price-change variability, but our terminol-
ogy follows that of Parsley (1996).
2
In other instances the choice appears to be motivated by the desire to compare the results with the existing
literature. This desire for comparability has also led many authors to focus on inter-market price dispersion
that is, dispersion across products in a particular location rather than intra-market price dispersion that is,
dispersion across locations for a particular product.
3
However, it is difcult to compare the ndings of Tommasi (1993) and Parsley (1996) with those of authors
such as Reinsdorf (1994) who examine the separate effects of anticipated and unanticipated ination.
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 3
of previous authors might be driven by the short and atypical disination period that he con-
siders, and it remains to be seen whether his results hold over longer time horizons and in
different economic conditions.
The aim in this paper is to investigate the heterogeneity in the RIV and RPV results by
comparing RIV and RPV models within the same dataset, allowing for a variety of functional
forms in both cases. Our price dispersion measures are constructed from detailed, homoge-
nous product prices in different cities, as in Reinsdorf (1994), and we compare the effects of
ination (decomposed into anticipated and unanticipated components). Our data are drawn
from three country datasets spanning a range of historical and inationary periods: (i) Canada
between 1922 and 1940, which includes a sustained deationary period during 1931-33, fol-
lowed by a rapid recovery in prices,
4
(ii) near-zero ination in Japan between 2000 and 2006,
and (iii) moderate ination in Nigeria between 2001 and 2006.
5
Our results show that the RPV-ination relationship differs signicantly from the RIV-
ination relationship in all three countries. More specically, RPV in Canada and Japan is
monotonically decreasing in unanticipated ination (there is no signicant effect in Nigeria),
but in all three countries RIV is increasing in the absolute value of unanticipated ination
(in other words, there is a V-shaped relationship). There is even more heterogeneity in the
effects of anticipated ination. In Canada the RPV-ination relationship is U-shaped, but
there is no signicant effect of anticipated ination on RIV. In Japan and Nigeria there is no
signicant effect of ination on RPV, but in Japan RIV is decreasing in anticipated ination
while in Nigeria it is increasing. Many of these results are consistent with existing papers
that focus on a subset of the dispersion and ination measures that we consider. What we are
able to show is that the heterogeneity in existing results is not simply a consequence of using
4
Hickey and Jacks (2011) examine examine retail price dispersion in Canada using 10 of the 44 products
that we consider, and nd that the infrequency of price adjustments is negatively related to intra-market price
dispersion.
5
The unweighted average ination rate in Canada over the sample period is -0.14% with a standard deviation
of 1.1%. The average ination rate in Japan is -0.08% with a standard deviation of 1.1%. For Nigeria, the
average ination rate is 0.9% with a standard deviation of 3.0%.
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 4
different samples in different countries: we nd substantial heterogeneity between RPV and
RIV functions within individual datasets.
Our empirical ndings also inform a growing theoretical literature on price dispersion and
ination. The monotonic negative relationship between RPV and ination shocks, combined
with a V-shaped RIV-ination relationship, is consistent with consumer search models in the
style of Reinganum (1979) and B enabou and Gertner (1993). The U-shaped relationship be-
tween anticipated ination and RPV identied in our Canadian data is also consistent with
the dynamic search models of van Hoomissen (1988) and Head and Kumar (2005), as well
as with menu cost theories. We believe these theoretical perspectives are worthy of further
empirical research. If these mechanisms are the main drivers of the ination-RPV relation-
ships observed, this implies that that a stable ination rate close to zero offers welfare gains
by facilitating lower price dispersion and fewer market inefciencies. It also suggests that the
efciency costs associated with negative ination shocks may be larger than those associated
with positive ones. More generally, our ndings imply that if policymakers are primarily
concerned with RPV because it reects market inefciencies associated with price disparities
across homogenous goods (and less concerned about diverging rates of ination, if this cor-
responds to price convergence), then understanding the RPV-ination relationship will be at
least as important as understanding the RIV-ination relationship. However, a comprehensive
assessment of the welfare effects of aggregate ination depends on clarity about the relative
importance of RPV and RIV in the social welfare function.
1 Background
Our empirical strategy builds on three key insights from the theoretical and empirical lit-
erature. First, theory suggests that the effects of ination on RPV and RIV need not be
the same, and which measure matters most will depend on the social welfare objectives of
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 5
the policymaker. Second, different theories imply a role for either anticipated ination, or
unanticipated ination, or both, and this suggests that one should estimate the effects of an-
ticipated and unanticipated ination on price dispersion separately. Finally, different authors
have chosen to impose a variety of different functional forms on the data, and in some cases
these choices have been shown to have a substantial impact on the results: our empirical
strategy should not be overly restrictive with respect to functional form. We summarize the
salient features of this literature that inform our empirical approach.
We begin by dening more precisely the two measures of dispersion that we study. (This
paper focuses on measures of intra-market price dispersion, although the models described in
this paper could also be applied to inter-market price dispersion. Some notation will clarify
the distinction.) Intra-market RPV is measured as the coefcient of variation across locations
in the price level of product i in location j in period t, which we denote p
ijt
:
v
it
=

1
N

j
_
p
ijt
p
it
1
_
2
(1)
Here, p
it
is the regional or national average price. Inter-market RPV is dened in a similar
way, by reassigning the subscripts i to products and the subscripts j to locations.
6
Denoting
the rate of change of the price of product i over period t in location j as
ijt
= ln(p
ijt
),
and average product-specic ination as
it
= ln( p
it
), intra-market RIV is measured as
the standard deviation across locations of the rate of change of prices:
w
it
=
_
1
N

j
(
ijt

it
)
2
. (2)
Inter-market RIV is dened in a similar way, by again reassigning the subscripts i and j. The
6
Some studies use weighted averages reecting the relative size of locations or the relative value of trade in
individual commodities. Our analysis will use unweighted averages.
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 6
relationship between RIV and RPV is non-linear,
7
and the two measures of dispersion will
not necessarily respond in the same way to mean aggregate ination.
The existing literature focuses mainly on RIV instead of RPV. Parks (1978) shows that
a simple log-linear signal extraction model implies a positive effect of the absolute value
of unanticipated ination on both RIV and RPV. Using US data for 1927-1975 to exam-
ine the impact of ination shocks on RIV, the estimated relationship is consistent with this
prediction, a nding that has been conrmed by numerous other studies from different coun-
tries using either time-series data (van Hoomissen, 1988; Lach and Tsiddon, 1992; Tommasi,
1993; Jaramillo, 1999; Aarstol, 1999; Becker and Nautz, 2009, 2012) or cross-sectional data
(Debelle and Lamont, 1997; Beaulieu and Mattey, 1999).
8
Although this theory also predicts
a V-shaped relationship for RPV, a reliance on CPI index data (instead of detailed data on
individual product prices) precludes the extension of his empirical model to RPV.
The few papers that have compared RPV and RIV models have reached a range of dif-
ferent conclusions. (see Reinsdorf, 1994; Parsley, 1996; Tommasi, 1993; C aglayan and Fil-
iztekin, 2003). One explanation for the heterogeneity in these results is that different pro-
cesses are at work in different countries and at different times. Another explanation is that
several different processes are at work in the same data. The different theories implying a re-
lationship between price dispersion and anticipated ination make contrasting predictions, as
do the theories implying a relationship between price dispersion and unanticipated ination,
and the patterns in the data might result from a combination of different theoretical effects.
7
More specically, intra-market RIV for each commodity is
w
it

_
v
2
it
+ v
2
it1
2COV(ln(p
ijt
), ln(p
ijt1
)).
8
Most empirical papers (including Parks (1978)) measure RIV as the variation in rates of ination across
fairly broad commodity or industry groups. This is consistent with theories that emphasize the distortions caused
by shifts in relative prices across different goods when suppliers face different demand and/or supply elastici-
ties (see Hercowitz, 1981; Cukierman, 1983). However, menu cost and consumer search theories suggest that
relative price changes across suppliers within industries are also important. Domberger (1987), van Hoomissen
(1988), Lach and Tsiddon (1992), Tommasi (1993) and Parsley (1996) explore this issue using commodity-level
RIV measures for individual stores or locations.
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 7
For example, Danziger (1987) shows that menu cost models in the style of Rotemberg (1983)
imply a U- or V-shaped relationship between anticipated ination and RPV, but an RIV func-
tion that could be increasing, decreasing, or V-shaped depending on the ination rate. These
predictions contrast with those of the dynamic models of consumer search proposed by van
Hoomissen (1988) and Head and Kumar (2005), which imply that higher anticipated ina-
tion or deation will increase both RPV and RIV.
9
Among theories that predict a relationship
between price dispersion and ination shocks, Fielding and Hajzler (2013) show that in con-
sumer search models such as Reinganum (1979) and B enabou and Gertner (1993), where
buyers search sequentially for costly price quotes from heterogenous sellers, RPV is mono-
tonically decreasing in unanticipated ination but RIV can be a U- or V-shaped function of
unanticipated ination, if sellers relative cost differences are persistent. These predictions
are at odds with those of signal-extraction models. The Reinganum-type search cost model
would explain the differences that Reinsdorf (1994) nds between the RIV and RPV func-
tions for unanticipated ination, and if the effect on RPV and RIV of anticipated ination (via
menu cost or van Hoomissen-type search cost effects) differs from that of unanticipated ina-
tion (via Reinganum-type search cost effects), it is not surprising that papers using aggregate
ination (Parsley, 1996; Tommasi, 1993) nd effects that are at odds with those from papers
that decompose ination into its anticipated and unanticipated components (C aglayan and
Filiztekin, 2003). Investigating the sources of heterogeneity in the RIV and RPV results re-
quires a comparison of RIV and RPV models within the same data set, allowing for a variety
of functional forms and distinguishing between anticipated and unanticipated ination.
Our paper builds on the work of Reinsdorf (1994) and others by adopting a modeling
9
In Head and Kumar (2005), the relationship between RPVand ination is driven by the tradeoff between the
marginal benets of additional search due to a wider range of posted prices (primarily through a rise in prices at
the upper end of the distribution), which lowers RPV, and an increase in the opportunity cost of holding money,
which increases monopoly power and price dispersion. At higher ination rates the latter effect dominates.
In van Hoomissens (1988) ination erodes the informational value of obtaining additional price quotes due
to the asynchronous price adjustments of rms, provided this ination is persistent, resulting in higher price
dispersion.
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 8
framework that allows for different possible functional forms. As in Reinsdorf (1994) and
C aglayan, Filiztekin, and Rauh (2008) for RPV, and Lach and Tsiddon (1992), Fielding and
Mizen (2008) and Becker and Nautz (2012) for RIV, we report the effects of anticipated
and unanticipated ination separately. Previous papers have employed a variety of different
functional forms,
10
and we are mindful of the theories which suggest that some functional
form restrictions, including the assumption of symmetry between the effects of positive and
negative ination, may bias parameter estimates. For example, Jaramillo (1999) shows that
relaxing the Parks (1978) assumption of linear demand and supply curves results in a break-
down of the symmetric relationship between RIV and unanticipated ination, and Bomberger
and Makinen (1993) argue that differences in the extent of downward price stickiness across
producers results in negative ination shocks having a more pronounced effect on RIV than
do positive shocks. Similarly, Choi and Kim (2010) argue that imposing a symmetric V-shape
function will be misleading if the true function is U-shaped with a (variable) turning point
close to the expected ination rate. Becker and Nautz (2009, 2012) also emphasize that the
search model of Head and Kumar (2005) predicts an asymmetric relationship between aver-
age ination and price dispersion. Our empirical model addresses these concerns by allowing
for asymmetric effects of positive and negative ination shocks and of positive and negative
anticipated ination. The next section describes our modeling strategy in more detail.
2 The Econometric Model
Our model is designed to identify the form of the relationship between intra-market RPV
(or RIV) and anticipated and unanticipated ination, using the monthly data described in
the next section. The two alternative dependent variables will be the deseasonalized values
10
Examples include models that are quadratic in both anticipated and unanticipated ination (Parks, 1978;
Aarstol, 1999; Becker and Nautz, 2009), models that are linear in both anticipated and unanticipated ina-
tion(Parsley, 1996), and models that are a mixture of the two approaches (Lach and Tsiddon, 1992).
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 9
of v
it
as dened in equation (1) and w
it
as dened in equation (2); these are designated
v
D
it
and w
D
it
respectively. Similarly, deseasonalized monthly ination for each commodity is
designated
D
it
. This ination rate is decomposed into an anticipated component (
A
it
) and an
unanticipated component (
U
it
), as in papers such as Fielding and Mizen (2008) and Becker
and Nautz (2009). The decomposition is based on an ARCH model of aggregate ination:

D
it
=
A
it
+
U
it
(3)

A
it
=
0i
+
1i

D
it1
+
2i

D
it2
+
3i
t (4)

U
it
N
_
0, h
2
it
_
(5)
h
2
it
=
0i
+
1i

_

U
it1
_
2
. (6)
Here, h
2
it
is the conditional variance of the ination forecast, capturing ination uncertainty.
Note that the and parameters are specic to each item i; in other words, the dynamics of
ination are allowed to vary from one item to another.
Then we t a number of alternative RPV regression equations, each having the following
general form:
ln(v
D
it
) =
v
0i
+
v
1i
ln(v
D
it1
) +
v
2i

UP
it
+
v
3i

UN
it
+
v
4i
t +
v
i
_

A
it
_
+
v
i
(h
it
) +u
v
it
(7)
ln(w
D
it
) =
w
0i
+
w
1i
ln(w
D
it1
)+
w
2i

UP
it
+
w
3i

UN
it
+
w
4i
t +
w
i
_

A
it
_
+
w
i
(h
it
)+u
w
it
(8)
Here, the u
it
terms are regression residuals, the s are xed parameters estimated separately
for each commodity, the () and () terms stand for commodity-specic non-linear func-
tions described below,
UP
it
= max
_
0,
U
it
_
, and
UN
it
= min
_
0,
U
it
_
. The inclusion of
these two unanticipated ination terms in the regression equations allows RPV or RIV to be
a monotonic function of unanticipated ination (as in some theoretical search models) or a
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 10
non-monotonic one (as in theoretical signal-extraction models). However, if the function is
non-monotonic, then the turning point is constrained to be equal to zero (as in theoretical
signal-extraction models).
Different theories suggest a wide range of functional forms for the relationship between
RPV and anticipated ination, and this range is reected in the variety of functional forms
in existing empirical studies: the relationship can be U-shaped or V-shaped, and the turning
point is not necessarily at
A
= 0. For this reason, we t alternative versions of equations
(7)-(8) with different parameterizations of the -function. The rst of these is a quadratic
function:

x
i
=
x
5i

A
it
+
x
6i

_

A
it
_
2
, x {v, w} (9)
We will compare this quadratic parameterization with a piecewise-linear parameterization
that has been used in some other papers:

x
i
=
x
5i

AP
it
+
x
6i

AN
it
, x {v, w} (10)
Here,
AP
it
= max
_
0,
A
it
_
and
N
it
= min
_
0, ,
A
it
_
, and equation (10) allows for a V-shaped
curve with a turning point at zero. In Appendix A, we also explore the possibility of tting
a non-parametric -function. Finally, we allow for the possibility that RPV and RIV depend
either on the standard deviation of the ination forecast (with
x
i
=
x
7i
h
it
) or on the variance
(with
x
i
=
x
7i
h
2
it
).
3 The Data
The model in Section 2 will be applied to three different datasets: one from Canada, one from
Japan and one from Nigeria. In this section we describe the construction of the RPV, RIV
and ination variables in each dataset, and present some descriptive statistics.
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 11
3.1 Canada
Following Hajzler and MacGee (2012), our data are taken from monthly issues of the Canada
Labour Gazette, which are available for the period November 1922 November 1940. This
publication lists the monthly prices of a variety of grocery items in a number of Canadian
cities. These prices are averages over a number of stores in each city, reported in tenths of
cents. Not all prices are available for all cities, but the prices of 42 items are reported for
69 cities over the whole period with just a few missing observations; it is these prices that
form our data set.
11
The cities and grocery items are listed in Appendix B. For each of the
42 items, we construct the variables v
it
, w
it
and
it
according to the above denitions. The
corresponding deseasonalized series v
D
it
, w
D
it
and
D
it
are constructed from regressions of v
it
,
w
it
and
it
on a set of dummy variables for each month (February December).
3.2 Japan
The Japanese price series are taken from the dataset published by the Center for International
Price Research
12
and documented by Crucini, Shintani, and Tsuruga (2010). This monthly
dataset spans the period January 2000 December 2006; the prices of 146 household grocery
items and 163 other household goods are reported for 70 cities over this period. These prices
are averages over a number of stores in each city, reported in yen. The cities and commodities
are listed in Appendix B. For each of the 309 commodities, we construct the variables v
it
,
w
it
and
it
in the same way as for Canada, and then deseaonalize each series using the same
method.
11
Newfoundland did not become part of Canada until 1949, so there are no Newfoundland cities in the data
set.
12
These data are available at: www.vanderbilt.edu/econ/cipr/japan.html
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 12
3.3 Nigeria
The Nigerian price series are taken from the dataset published by the Nigerian National
Bureau of Statistics (www.nigerianstat.gov.ng) and documented by Fielding (2010). This
monthly dataset spans the period January 2001 December 2006; the prices of 22 house-
hold grocery items and 16 other household goods and services are reported for each of the
36 state capitals, plus the federal capital, Abuja.
13
These prices are averages over a num-
ber of stores and markets in each city, reported in kobo.
14
The cities and commodities are
listed in Appendix B. For each of the 38 items, we construct the variables v
it
, w
it
and
it
and
deseaonalize each series using the same method as for Canada.
3.4 Descriptive statistics
Our three datasets are drawn from three very different economies pre-war Canada, mod-
ern Japan, and modern Nigeria and encompass different ranges of consumer goods. The
Japanese data are the most comprehensive; the Canadian and Nigerian datasets are limited
to items that would typically be available in a wide range of small local stores (Canada) or
traditional markets (Nigeria), the typical consumer not having access to large stores or su-
permarkets on a regular basis.
15
Many of the items in the datasets reect spending patterns
that reveal the cultural idiosyncrasies of the society concerned: for example, lard in pre-
war Canada, salted sh guts in Japan, and kola nuts in Nigeria. Therefore, we should not
necessarily expect the parameters in the RPV and RIV equations to be identical across the
three datasets. Nevertheless, common patterns in the parameter values across countries could
13
These 38 items are a subset of the items included in the National Bureau of Statistics dataset. Excluded
from our sample are (i) alcoholic beverages, the prices of which are not recorded in states with a Muslim
majority population, and (ii) a range of packaged and branded food and other household items (for example, a
tin of Andrews liver salts; a packet of 20 Benson and Hedges cigarettes; a Bic biro). These items are mostly sold
only in large stores, not in traditional markets, and for many the average value of v
it
is extremely low. There is
reason to suspect that the prices of some of these items are set centrally, and are not controlled by local retailers,
so it is doubtful whether the theories discussed in Section 1 would be applicable to them.
14
100 kobo = one naira one US cent in 2001.
15
The Nigerian dataset also includes some locally provided services, including accommodation and taxis.
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 13
reveal some of the underlying fundamentals driving dispersion.
Some of the basic characteristics of the data are presented in Table 1 and in Figures 1-9.
Figures 1-3 illustrate average price ination across all the items in each sample. Canadas av-
erage price ination during the interwar period looks very similar to modern Japans average
price ination in the early 21st century, except for a deation and subsequent ination be-
tween 1931 and 1934, and a spike at the beginning of World War Two. Ination volatility in
Canada is a little higher than in Japan. Average ination in Nigeria over 2001-2006 is a per-
centage point higher than the Canadian and Japanese averages, and exhibits higher volatility.
Figures 4-6 show that the individual ination series
D
it
are not normally distributed: there is
excess kurtosis in all three countries. This means that it will be important to ascertain whether
any of our regression results is affected by outliers in the ination distribution, and for this
reason two versions of each regression will be tted: one with the original ination series
(
A
it
,
U
it
), and another with the series trimmed at 10% per month.
Means and standard deviations over time for both trimmed and untrimmed
A
it
and
U
it
series are included in Table 1. These series are constructed by applying the GARCH model in
equations (3)-(6) to each of the
D
it
series in each country. It can be seen that Nigeria is again
somewhat different from the other two countries, with a standard deviation of anticipated and
unanticipated ination (both trimmed and untrimmed) that is about twice as high as that in
Canada and Japan. The within-commodity standard deviations are also similar in Canada
and Japan, but much larger in Nigeria. This difference is not surprising: like many other
developing countries, Nigeria faces macroeconomic shocks that are larger than those typical
of developed countries in most eras.
Table 1 shows that the mean values of ln(v
D
it
) and ln(w
D
it
) are very similar in Canada and
Japan, and slightly lower in Nigeria. Figures 7-9 show that in all three countries the price
dispersion variables are approximately normally distributed.
Table 1 and Figures 1-9 about here
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 14
4 Results
Equations (7)-(8) are time-series regression equations to be tted for each commodity i. In
Canada, where T = 217 and M = 42, it is possible to t the regression equations simultane-
ously using SUR and estimate a variance-covariance matrix for all of the parameters in all of
the commodity-specic equations. This matrix can then be used to compute standard errors
on the average values of the parameters across all of the commodities (
1
M

i

x
ni
) using the
Delta Method. The focus of our discussion will be on these averages, which indicate the
pattern of the RPV and RIV relationships for a typical commodity. In Japan T < M, so it
is not possible to t the regression equations simultaneously, and we assume an orthogonal
variance-covariance matrix when computing the standard errors on the average parameter
values. In Nigeria Mis almost as large as T: it is possible to t the regression equations si-
multaneously, but estimates of the individual elements of the variance-covariance matrix will
be very imprecise. Therefore, we also assume orthogonality in the Nigerian case.
Tables 2-7 report estimates of the average parameter values in equations (7)-(8). For each
country there are two tables: one for RPV (measured as ln(v
D
it
)) and one for RIV (measured
as ln(w
D
it
)). In each table there are four sets of parameter estimates using trimmed ina-
tion and four using untrimmed ination; these four sets of estimates correspond to the two
alternative parameterizations of the -function and the two alternative parameterizations of
the -function described in Section 2. (The parameters of the -function are statistically in-
signicant, except in the case of Canadian RPV, where a higher variance of ination shocks
reduces dispersion. The results do not vary signicantly across the alternative parameter-
izations of this function.) T-ratios are reported underneath the parameter estimates, with
parameters signicant at the 5% level highlighted in bold. First of all we discuss the esti-
mated effects of unanticipated ination, as captured by the parameters on
UP
it
and
UN
it
; then
we discuss the estimated effects of anticipated ination, as captured by the parameterizations
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 15
of the -function.
4.1 The effects of unanticipated ination
Overall, the form of the relationship between RPV / RIV and unanticipated ination in the
three countries can be summarized as follows:
Canada Japan Nigeria
RPV equation negative monotonic negative monotonic insignicant
RIV equation V-shaped V-shaped V-shaped
The V-shaped relationship for RIV is consistent with many of the inter-market RIV studies.
The negative monotonic relationship for RPV is consistent with Reinsdorf (1994), and with
C aglayan and Filiztekins nding that the imposition of a non-monotonic functional form in
the RPV equation produces an insignicant unanticipated ination effect. It is also consistent
with search theories of the type introduced by Reinganum (1979) and B enabou and Gertner
(1993), but not with signal extraction models. If policymakers care primarily about RPV
rather than RIV, then the results imply an asymmetric optimal policy response: negative
ination shocks raise RPV with potential welfare losses, but positive shocks do not. If on the
other hand policymakers are concerned primarily with RIV, then both positive and negative
ination shocks are costly.
In both the Canadian and Japanese RPV results (Tables 2 and 4), the curve is signicantly
steeper for negative shocks than it is for positive ones. In Canada, a one percentage point
positive ination shocks reduces RPV (as measured by v
D
it
) by about 0.5%; a one percentage
point negative ination shock raises RPV by about 1%. In Japan, a one percentage point
positive ination shocks reduces RPV by about 1%; a one percentage point negative ination
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 16
shock raises RPV by about 1.5%. The within-commodity standard deviation of unanticipated
ination is about two percentage points in both countries, whereas the standard deviation of
RPV is about 15%, so the order of magnitude of the RPV response to a typical shock to
ination is in the region of 10-20% of one standard deviation of the dependent variable.
In all three countries, the absolute size of the estimated effect of ination shocks on RIV
is greater than the absolute size of the estimated effect on RPV. In Japan this difference is
particularly large: a one percentage point ination shock (positive or negative) raises RIV by
20-30%. The within-commodity standard deviation of RIV in Japan is just under 40%, so a
typical ination shock (about two percentage points) raises RIV by more than one standard
deviation. In Canada and Nigeria the unanticipated ination coefcients are much smaller: a
one percentage point ination shock raises RIV by about 3%, or if it is a negative shock in
Nigeria, by about half this much. (Using trimmed ination, the effect of the negative shock in
Nigeria is not quite statistically signicant; this is the only substantial difference between the
trimmed and untrimmed ination results in the tables.) Comparing Canada with Japan, very
similar RPV results do not entail very similar RIV results: the Japanese RIV coefcients are
much larger, and this difference warrants further research. Nevertheless, there is a common
pattern in the results across the countries, monotonic RPV functions contrasting with V-
shaped RIV functions.
Tables 2-7 about here
4.2 The effects of anticipated ination
The anticipated ination effects show more cross-country heterogeneity than the unantici-
pated ination effects. In Canada, there is a signicant coefcient on the quadratic term in the
RPV equation, with a turning-point insignicantly different from zero. The standard devia-
tion of anticipated ination is very close to one percentage point. If anticipated ination devi-
ates by one percentage point from its sample mean (which is very close to zero) then RPV can
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 17
be expected to rise by about 2%. This signicance of the quadratic term is consistent with the
piecewise-linear regression estimates, insofar as ln
_
v
D
it
_
/
AP
it
> 0 > ln
_
v
D
it
_
/
AN
it
;
however, the coefcient on
AP
it
is insignicantly different from zero. The non-parametric
estimates in Appendix A suggest that the function is indeed quadratic, with ln
_
v
D
it
_
/
AP
it
increasing in
AP
it
, which may explain why the coefcient on
AP
it
is very imprecisely es-
timated. In the RIV equation there are no signicant coefcients on any of the anticipated
ination terms, although the non-parametric estimates in Appendix A do suggest a signicant
and approximately quadratic relationship, a rare case of similarity between RPV results and
RIV results.
In Japan, the results are rather different. The quadratic RPV model does not produce any
signicant results, but the piecewise-linear model produces a positive coefcient on
AP
it
that
is just signicant at the 5% level, so there is some weak evidence that RPV is increasing
in anticipated ination, at least when the ination rate is greater than zero. The standard
deviation of anticipated ination in Japan is about 0.6 percentage points. The estimated
parameter on
AP
it
indicates that a one standard deviation rise in anticipated ination, when
positive, can be expected to raise RPV by around 8%. By contrast, the piecewise-linear
RIV model does not produce any signicant results for anticipated ination, whereas the
quadratic model produces a signicant negative effect of
A
it
, though without any signicant
non-linearity. A one standard deviation rise in anticipated ination can be expected to reduce
RIV by around 30%. If RPV is increasing in anticipated ination and RIV is decreasing in
anticipated ination, then the choice of an optimal ination target will certainly depend on
whether RPV or RIV matters more to policymakers.
The results for Nigeria are different again. There are no signicant effects of anticipated
ination in the RPV equation, and no signicant effects in the quadratic version of the RIV
equation. However, the piecewise-linear version of the RIV equation suggests a positive
monotonic function; the effect is marginally signicant at the 5% level. The standard devia-
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 18
tion of trimmed anticipated ination in Nigeria is about 1.8 percentage points, and a standard
deviation increase in
A
it
can be expected to raise RIV by around 9%.
The heterogeneity of the anticipated ination effects across countries does have a the-
oretical interpretation. As shown by Danziger (1987), in a menu cost model the shape of
the RPV-ination function will depend on the range of trend ination, and on the shape of
a typical rms cost function, which will determine the value of the menu cost parameter.
Table 1 shows that the distribution of trend ination in Nigeria is rather different from the
distributions in the other two countries, and there is no reason to suppose that the typical rm
in pre-war Canada faced a cost curve similar to that of a typical rm in modern Japan. But
note that again there is substantial within-country heterogeneity between the RPV effects and
the RIV effects. A signicant anticipated ination effect for one does not entail a signicant
effect for the other.
4.3 A note on parameter stability
A question yet to be addressed fully (either in this paper or in the literature) is whether
the effects of ination on RPV and RIV are stable over time.
16
The Japanese and Nigerian
sample periods are quite short (2000 2006 and 2001 2006 respectively), so addressing
parameter stability issues in these two cases is not feasible. However, the Canadian dataset
encompasses a much longer sample period (1922 1940), so it is possible to investigate how
stable the parameters of equations (7)-(8) are in Canada.
In order to do this, we t the two equations to eight-year sub-samples, the rst ending
in December 1930, the second in December 1931, and so on to the last sub-sample, ending
16
Important exceptions to this comment are Choi (2010) and C aglayan and Filiztekin (2003). Choi (2010)
studies parameter stability in the ination RIV relationship using CPI data for the United States (1978-2007) and
for Japan (1970-2006). He nds a positive relationship during the high-ination periods of the 1970s and 1980s
for both countries, whereas the U-shape is prevalent during recent decades of low ination. Using disaggregated
annual price data in Turkey, C aglayan and Filiztekin (2003) nd that the effect of ination on RPV and RIV
are signicant during the relatively low-ination 1948-1975 period, but are mainly insignicantly during the
1976-1997 rising ination period.
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 19
in November 1940. Each of sub-sample has 96 observations except the rst one (missing
January 1923) and the last one (missing December 1940), which have 95 observations. These
rst and last subsamples both exclude the trough of the Great Depression (1931-1932); other
subsamples include the trough. If the Depression affects the relationship between RPV and
ination, this should be apparent in differences in parameter estimates across subsamples.
The charts in Figure 10 illustrate the
A
and
U
parameter estimates in equation (4) the
RPV model using untrimmed ination and quadratic forms for the - and -functions. The
sub-sample results in this gure correspond to the whole-sample results in the rst column
in Table 2. The stylized facts discussed here also apply to the parameter estimates in the
other versions of equation (7), which are not shown. The charts in Figure 11 illustrate the
equivalent estimates for equation (8) the RIV model. The sub-sample results in this gure
correspond to the whole-sample results in the rst column in Table 3. In both gures, the
parameter estimates are indicated by the black lines, with the 95% condence interval in gray.
In each chart, the horizontal axis indicates the last year in the sub-sample corresponding to
the parameter estimate measured on the vertical axis.
Overall, there does seem to be some change in the relationship between RPV and antici-
pated ination, as shown in Figure 10. The (
A
)
2
parameter is signicantly greater than zero
in subsamples ending in 1935 or earlier, but its value falls over time, and is insignicantly dif-
ferent from zero in later subsamples. By the nal subsample, neither the mean
A
parameter
nor the mean (
A
)
2
parameter is signicantly different from zero. The relationship between
RPV and unanticipated ination is somewhat more stable. The
UP
and
UN
parameter es-
timates are signicantly below zero in all subsamples, with little change in the value of the
parameters over time. Also, Figure 11 shows that there is very little change in any of the RIV
parameter values over time. Estimates of the
A
parameters remain insignicantly different
from zero throughout the sample period. The
UP
parameter estimate is signicantly greater
than zero for the whole sample period, and the
UN
parameter estimate signicantly less than
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 20
zero, as in Table 3.
The contrast between the stability and signicance of the
U
parameters here with the
insignicance or instability of the
A
parameters reinforces the impression that there is more
potential for heterogeneity in the effects of anticipated ination than there is in the effects of
unanticipated ination. RPV can be expected to be lower with large positive ination shocks,
and higher with large negative shocks; RIV can be expected to be higher with large positive
or negative shocks. The effects of trend ination, however, seem to vary somewhat across
countries and over time.
Figures 10-11 about here
5 Summary and Conclusion
Economic theory suggests the possibility of a wide range of different relationships between
the dispersion of commodity- or region-specic relative price levels and the aggregate ina-
tion rate (either anticipated or unanticipated). The same is true of the dispersion of ination
rates. Existing evidence has produced an equally wide range of different results, although
methodological heterogeneity limits the extent to which different sets of results can be com-
pared. One key question that needs to be answered is whether the impact of aggregate ina-
tion on price level dispersion resembles its impact on ination rate dispersion. This matters
if, for example, monetary policymakers care about dispersion of a particular kind, or indeed
of both kinds.
In this paper, we t the same set of models to datasets from three different countries
(Canada, Japan and Nigeria) in order to establish the form of the relationship between price
level dispersion and aggregate ination, and measure the extent to which it resembles the
relationship between ination rate dispersion and aggregate ination. With regard to the
effects of unanticipated ination, we nd similar results across all three countries. Large
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 21
negative ination shocks tend to increase both price level and ination rate dispersion, if they
have any affect at all; large positive ination shocks tend to increase ination rate dispersion
but reduce price level dispersion. These effects are consistent with some of the relevant
economic theory based on search costs; they mean that a monetary policy maker who cares
about price level dispersion might respond very differently to an aggregate ination shock
than one who cares about ination rate dispersion.
With regard to the effects of anticipated ination, there is evidence of substantial hetero-
geneity across the two measures of dispersion, across countries, and (when the sample period
is long enough to test this) over time. This heterogeneity is consistent with some of the rele-
vant theory based on menu costs. It means that any generalisations about the effect of trend
ination on dispersion are likely to be highly misleading.
It follows that if we wish to assign a welfare level to each aggregate ination rate, then
we need to know exactly how much to value reductions in price level dispersion relative to
reductions in ination rate dispersion. Such an exercise will require the modeling of the
sources of welfare loss associated with dispersion, both insofar as an increase in dispersion
represents a loss of economic efciency and insofar as it represents an increase in inequality
between regions. This will be the subject of future research.
References
Aarstol, Michael. (1999). Ination, Ination Uncertainty, and Relative Price Variability.
Southern Economic Journal 66 (2), 414423.
Barro, Robert J.. (1976). Rational Expectations and the Role of Monetary Policy. Journal
of Monetary Economics 2 (1), 132.
Beaulieu, Joe and Joe Mattey. (1999). The Effects of General Ination and Idiosyncratic
Cost Shocks on Within-Commodity Price Dispersion: Evidence from Microdata. Review
of Economics and Statistics 81 (2), 205216.
Becker, Sascha S. and Dieter Nautz. (2009). Ination and Relative Price Variability: New
Evidence for the United States. Southern Economic Journal 76 (1), 146164.
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 22
Becker, Sascha S and Dieter Nautz. (2012). Ination, price dispersion and market integration
through the lens of a monetary search model. European Economic Review , 624634.
B enabou, Roland and Robert Gertner. (1993). Search with Learning from Prices: Does
Increased Inationary Uncertainty Lead to Higher Markups? The Review of Economic
Studies 60 (1), 6993.
Bomberger, William A. and Gail E. Makinen. (1993). Ination and Relative Price Variabil-
ity: Parks Study Reexamined. Journal of Money, Credit and Banking 25 (4), 854861.
C aglayan, Mustafa and Alpay Filiztekin. (2003). Nonlinear Impact of Ination on Relative
Price Variability. Economics Letters 79 (2), 213218.
C aglayan, Mustafa, Alpay Filiztekin, and Michael T. Rauh. (2008). Ination, Price Disper-
sion, and Market Structure. European Economic Review 52 (7), 11871208.
Choi, Chi-Young. (2010). Reconsidering the Relationship between Ination and Relative
Price Variability. Journal of Money, Credit and Banking 42 (5), 769798.
Choi, Chi-Young and Young Se Kim. (2010). Is There Any Asymmetry in the Effect of
Ination on Relative Price Variability? Economics Letters 108 (2), 233236.
Crucini, Mario J, Mototsugu Shintani, and Takayuki Tsuruga. (2010). The Law of One Price
without the Border: The Role of Distance versus Sticky Prices*. The Economic Journal
120 (544), 462480.
Cukierman, Alex. (1983). Relative Price Variability and Ination: a Survey and Further
Results. In Variability in Employment, Prices, and Money, edited by Karl Brunner and
Allan H. Meltzer. Amsterdam: North-Holland.
Danziger, Leif. (1987). Ination, Fixed Cost of Price Adjustment, and Measurement of
Relative-Price Variability: Theory and Evidence. The American Economic Review 77 (4),
704713.
Deaton, Angus and Christina Paxson. (1998). Economies of Scale, Household Size, and the
Demand for Food. Journal of Political Economy 106 (5), 897930.
Debelle, Guy and Owen Lamont. (1997). Relative Price Variability and Ination: Evidence
from US Cities. Journal of Political Economy 105 (1), 132152.
Domberger, Simon. (1987). Relative Price Variability and Ination: A Disaggregated Anal-
ysis. The Journal of Political Economy 95 (3), 547566.
Fan, Jianqing. (1992). Design-Adaptive Nonparametric Regression. Journal of the Ameri-
can statistical Association 87 (420), 9981004.
Fan, Jianqing. (1993). Local Linear Regression Smoothers and their Minimax Efciencies.
The Annals of Statistics 21 (1), 196216.
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 23
Fielding, David. (2010). Non-monetary Determinants of Ination Volatility: Evidence from
Nigeria. Journal of African Economies 19 (1), 111139.
Fielding, David and Christopher Hajzler. (2013). Comment on Relative Price Variability
and Ination in Reinganums Consumer Search Model. University of Otago Discussion
Paper in Economics 1305.
Fielding, David and Paul Mizen. (2008). Evidence on the Functional Relationship between
Relative Price Variability and Ination with Implications for Monetary Policy. Economica
75 (300), 683699.
Hajzler, Christopher and James MacGee. (2012). Retail price Differences across U.S. and
Canadian Cities during the Interwar Period. Paper presented to the Society of Economic
Dynamics, Cyprus .
H ardle, Wolfgang. (1992). Applied Nonparametric Regression, vol. 19. Cambridge, MA:
Cambridge University Press.
Head, Allen and Alok Kumar. (2005). Price Dispersion, Ination, and Welfare. Interna-
tional Economic Review 46 (2), 533572.
Hercowitz, Zvi. (1981). Money and the Dispersion of Relative Prices. The Journal of
Political Economy 89 (2), 328356.
Hickey, Ross D and David S Jacks. (2011). Nominal rigidities and retail price dispersion
in Canada over the twentieth century. Canadian Journal of Economics/Revue canadienne
d economique 44 (3), 749780.
Jaramillo, Carlos Felipe. (1999). Ination and Relative Price Variability: Reinstating Parks
Results. Journal of Money Credit and Banking 31 (3), 375385.
Lach, Saul and Daniel Tsiddon. (1992). The Behavior of Prices and Ination: An Empirical
Analysis of Disaggregate Price Data. Journal of Political Economy 100 (2), 349389.
Lucas Jr, Robert E.. (1973). Some International Evidence on Output-Ination Tradeoffs.
The American Economic Review 63 (3), 326334.
Parks, Richard W.. (1978). Ination and Relative Price Variability. The Journal of Political
Economy 86 (1), 7995.
Parsley, David C.. (1996). Ination and Relative Price Variability in the Short and Long
Run: New Evidence from the United States. Journal of Money, Credit and Banking
28 (3), 323341.
Reinganum, Jennifer F. (1979). A simple model of equilibrium price dispersion. The
Journal of Political Economy 87 (4), 851858.
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 24
Reinsdorf, Marshall. (1994). New Evidence on the Relation between Ination and Price
Dispersion. The American Economic Review 84 (3), 720731.
Robinson, P.M.. (1988). Root-N-consistent Semiparametric Regression. Econometrica:
Journal of the Econometric Society 56 (4), 931954.
Rotemberg, Julio J.. (1983). Aggregate Consequences of Fixed Costs of Price Adjustment.
The American Economic Review 73 (3), 433436.
Tommasi, Mariano. (1993). Ination and Relative Prices: Evidence from Argentina. Un-
published manuscript .
van Hoomissen, Theresa. (1988). Price Dispersion and Ination: Evidence from Israel.
Journal of Political Economy 96 (6), 13031314.
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 25
Appendices
A A Semi-parametric Model of RPV and RIV
As noted in the literature review, there is some diversity in the way that existing papers
parameterize the relationship between RPV/RIV and anticipated ination, and the quadratic
and piecewise-linear functions in equations (9)-(10) do not encompass all of them. (For
example, these equations do not allow for a V-shaped function with a turning point at a
positive ination rate.) However, we can also t a semi-parametric model similar to the ones
used by Fielding and Mizen (2008) and Choi (2010). In this model, the parameterizations of
the -function in equations (9)-(10) are replaced by a non-parametric estimate of the function,
using the method described by Robinson (1988) and H ardle (1992, Chapter 9.1). Here, we
present estimates of a semi-parametric model applied to the Canadian data which are relevant
to the discussion in the main text; results for the other countries are available on request.
Robust estimation of a semi-parametric model requires a large number of observations,
so now the data are pooled across all grocery items, and the following regression equation is
tted to the panel:
ln
_
x
D
it
_
=
0
+
1
ln
_
x
D
it1
_
+
2

UP
it
+
3

UN
it
+
4
t +
_

A
it
_
+
5
h
it
(A1)
+
1
ln
_
x
D
i0
_
+
2

UP
i
+
3

UN
i
+
4

A
i
+
5
h
i
+ u
it
where x {v, w}. The second row of the equation contains a term in the initial value of price
dispersion, and terms in the mean values of the different regressors: y
i
=
1
T

t
y
it
. These
terms are included to control for unobserved heterogeneity across the different grocery items.
The rst step in tting equation (A1) to the data is to create transformed regressors that
are orthogonal to
A
it
. This is achieved by tting a non-parametric regression equation for
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 26
each of the regressors other than
A
it
:
y
it
=
y
_

A
it
_
+ y
it
(A2)
Here, y
it
is a regression residual. The non-parametric function
y
() is tted in the same
way as the function () which is described below. The and parameters in equation (A1)
are then estimated using the following regression equation:
ln
_
x
D
it
_
=
0
+
1
ln
_
x
D
it1
_
+
2

UP
it
+
3

UN
it
+
4


t +
5


h
it
(A3)
+
1
ln
_
x
D
0
_
+
2

UP
i
+
3

UN
i
+ eta
4

A
i
+
5


h
i
+
it
.
Here,
it
is a regression residual. Finally, the shape of () is estimated using the following
non-parametric regression equation:

it
=
_

A
it
_
+ u
it
(A4)
There are several different kernel density estimators that could be used to estimate the shape
of (). The results reported beloware based on one particular kernel density function, but re-
sults using other kernel density functions that are robust to outliers (such as the Epanechnikov
Kernel) produce similar results.
17
First, we choose specic values of anticipated ination at
which the derivative of () is to be estimated. These values are equidistant points within
the observed range of
A
it
. (The estimate at each point is independent of the others; enough
points are chosen for the shape of () to be clear.) At any particular point
0
, the derivative

0
is estimated by tting a linear regression equation using Weighted Least Squares. The
regression equation is:

it
=
0
+
0

A
it
+ u
it
(A5)
17
The kernel density function here is used for example in Deaton and Paxson (1998). See Fan (1992, 1993)
for a discussion of the properties of alternative kernel density functions.
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 27
and the weights W
it
are as follows:
W
it
=
15
16
_
1
_

0

it
4z
_
2
_
2
if |
0

it
| < 4z, else W
it
= 0. (A6)
Here, z is a smoothing parameter, and the truncation of the weighting function at
0
4z
ensures that extreme outliers do not inuence the estimates. The standard error of
0
is
estimated using a bootstrap with 100,000 replications.
First of all, we discuss the RPV results. Figures A1-A3 illustrate the derivative of the
-function in the ln(v
D
it
) equation at different anticipated ination rates for alternative values
of z between 1% and 2%,
18
along with the corresponding standard error bars. (The buttery
shape of the error bars arises from the fact that there are fewer observations at more extreme
values of anticipated ination.) Note that the gures are drawn to different scales, so that
each function occupies the whole chart and its shape is clear. As the value of z increases,
the function becomes atter but the error bars become smaller. In principle, it is possible to
select an optimal value of z based on an in-sample forecast error criterion. However, in
our case the mean squared forecast error changes very little within the range of zshown.
19
Nevertheless, all of the gures show a line that is approximately straight, that is, a -function
that is approximately quadratic. The turning point of the -function is indicated by the point
at which the line in the gure crosses the y-axis; this is always at a positive anticipated
ination rate. We can compare Figures A1-A3 with a gure based on the parameters of
the quadratic model reported in Table 2 (panel C) in the main text; this is Figure A4. The
imposition of a quadratic functional form produces a curve similar to the one in the semi-
parametric model with z = 1%. The error bars associated with the semi-parametric model are
smaller, so a turning point at zero can be rejected with more condence.
Table A1 reports estimated values of
2
and
3
(the unanticipated ination parameters)
18
Recall that the ination data are monthly, so typical absolute anticipated ination rates are less than 1%.
19
For smaller values of z, the forecast errors are larger.
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 28
in the RPV equation for the different values of z, along with the corresponding t-ratios.
Other parameter estimates are available on request. The parameter estimates are not very
sensitive to the choice of z; they have the same sign as the estimates reported in Table 2 of
the main text (implying a negative monotonic function), and are signicantly different from
zero. Their absolute value is somewhat smaller than in Table 2, and in the case of
3
this
difference is statistically signicant. However, the overall conclusions regarding the effect of
unanticipated ination on RPV are unchanged.
Next we discuss the anticipated ination effects in the RIV function shown in Figures A5-
A7. Figure A5 shows that with z = 1%there is a smooth and approximately quadratic function
with a signicantly negative slope for ination rates below 0.25% and a signicantly positive
slope for ination rates above 0.75%. Generally, the curve for RIV with z = 1% is quite
similar to the curve for RPV with z = 1%; both indicate that the minimal level of dispersion
is reached when ination is positive. Recall that the parametric models in Table 3 of the main
text do not produce any signicant anticipated ination effect. One possible explanation for
this difference is that the parametric results are confounded by extreme values of ination
(when
A
it
is outside the range shown in the gures) at which the quadratic relationship fails
to hold. This suspicion is reinforced by the observation that the slope of the () function is
insignicantly different from zero at all levels of anticipated ination when we set z 2%,
as shown in Figure A7.
Table A2 reports estimated values of
2
and
3
(the unanticipated ination parameters)
in the RIV equation for the different values of z, along with the corresponding t-ratios. The
parameter estimates are again not very sensitive to the choice of z; they have the same sign
as the estimates reported in Table 3 of the main text (implying a V-shaped function), and
are signicantly different from zero. Their absolute value is again somewhat smaller than
in Table 3, and for both parameters this difference is statistically signicant. However, the
overall conclusions regarding the effect of unanticipated ination on RPV are unchanged.
PRICE LEVEL DISPERSION VS INFLATION RATE DISPERSION 29
Appendix Tables A1-A2 and Figures A1-A7 about here.
B Lists of Cities and Items Included in the Three Samples
Appendix Tables A3-A8 about here.
i
TABLE 1
DESCRIPTIVE STATISTICS FOR THE THREE DIFFERENT SAMPLES


Canada Japan Nigeria

mean
within-
commodity
std. dev.
mean
within-
commodity
std. dev.
mean
within-
commodity
std. dev.

ln
D
it
v
-2.210 0.187 -1.944 0.138 -2.678 0.531

ln
D
it
w
-2.869 0.245 -3.152 0.388 -3.844 1.440
A
it

-0.001 0.010 -0.001 0.006 0.007 0.018

U
it

0.000 0.025 0.000 0.026 0.000 0.061

A
it

(trimmed) -0.001 0.010 -0.001 0.006 0.001 0.018

U
it

(trimmed) 0.000 0.022 0.000 0.016 0.000 0.049

ii
TABLE 2
AVERAGE PARAMETER VALUES IN THE MODELS OF

ln
D
it
v
IN CANADA
T-ratios are reported in italics.
A B C D

untrimmed
inflation
trimmed
inflation
untrimmed
inflation
trimmed
inflation
untrimmed
inflation
trimmed
inflation
untrimmed
inflation
trimmed
inflation

1
ln
D
it
v

0.757 0.758 0.756 0.757 0.758 0.758 0.756 0.757
110.5 111.1 110.4 110.7 110.8 110.9 110.3 110.7
A
it

-0.332 -0.272 -0.324 -0.293

-0.793 -0.649 -0.776 -0.701

2
) ( 100
A
it


2.119 2.150 2.035 2.033

2.964 3.005 2.848 2.845

AP
it

0.127 0.080 0.016 0.076

0.307 0.193 0.037 0.182
AN
it

-2.090 -2.068 -2.018 -2.052

-6.852 -6.754 -6.516 -6.629
UP
it

-0.505 -0.564 -0.513 -0.568 -0.513 -0.568 -0.514 -0.564

-5.581 -6.003 -5.688 -6.069 -5.659 -6.035 -5.694 -6.026
UN
it

-0.922 -0.933 -0.924 -0.925 -0.911 -0.931 -0.915 -0.924

-9.922 -9.901 -9.982 -9.851 -9.788 -9.874 -9.890 -9.842
100 (h
it
)
2
-3.896 -3.681 -2.679 -2.788

-4.686 -4.892 -4.686 -4.400

100 h
it
-0.125 -0.113 -0.083 -0.091

-3.484 -3.200 -4.087 -2.418
iii
TABLE 3
AVERAGE PARAMETER VALUES IN THE MODELS OF

ln
D
it
w
IN CANADA
T-ratios are reported in italics.
A B C D

untrimmed
inflation
trimmed
inflation
untrimmed
inflation
trimmed
inflation
untrimmed
inflation
trimmed
inflation
untrimmed
inflation
trimmed
inflation

1
ln
D
it
w

0.476 0.477 0.474 0.474 0.476 0.476 0.473 0.473
51.31 51.30 50.72 50.74 51.15 51.17 50.543 50.64
A
it

-0.799 -0.820 -0.757 -0.830

-1.134 -1.161

-1.076 -1.178


2
) ( 100
A
it


-1.951 -2.011 -2.025 -2.105

-1.343 -1.384

-1.407 -1.463


AP
it

-9.425 -9.500 -9.608 -9.562



-1.480 -1.484

-1.507 -1.492
AN
it

-0.404 -0.384 -0.351 -0.380



-0.567 -0.537

-0.489 -0.528
UP
it

2.674 2.884 2.671 2.900 2.657 2.868 2.661 2.883

13.07 13.71 13.08 13.81 12.97 13.62 13.02 13.74
UN
it

-2.708 -2.797 -2.721 -2.814 -2.688 -2.784 -2.697 -2.799

-12.56 -12.81 -12.64 -12.90 -12.46 -12.76 -12.54 -12.84
100 (h
it
)
2
0.771 -0.617 1.148 0.582

0.346 -0.239 0.694 0.285




100 h
it
3.847 -8.449 1.629 -1.703


0.385 -0.734 0.236 -0.180
iv
TABLE 4
AVERAGE PARAMETER VALUES IN THE MODELS OF

ln
D
it
v
IN JAPAN
T-ratios are reported in italics.
A B C D

untrimmed
inflation
trimmed
inflation
untrimmed
inflation
trimmed
inflation
untrimmed
inflation
trimmed
inflation
untrimmed
inflation
trimmed
inflation

1
ln
D
it
v

0.738 0.736 0.740 0.737 0.738 0.736 0.739 0.737
168.9 167.9 170.1 169.2 168.9 167.9 170.4 169.4
A
it

-2.089 -1.966 -1.998 -1.875

-1.799 -1.676 -1.725 -1.602

2
) ( 100
A
it


-5.472 -5.431 -5.687 -5.652

-0.946 -0.939 -0.979 -0.973

AP
it

12.59 12.57 13.32 13.30

2.175 2.171 2.255 2.251
AN
it

2.000 1.587 1.991 1.576

1.323 1.023 1.321 1.018
UP
it

-1.044 -1.093 -1.044 -1.093 -1.044 -1.093 -1.043 -1.091

-10.65 -10.91 -10.71 -10.96 -10.61 -10.86 -10.66 -10.90
UN
it

-1.506 -1.495 -1.514 -1.502 -1.500 -1.489 -1.511 -1.499

-15.71 -15.45 -15.95 -15.67 -15.58 -15.32 -15.88 -15.60
100 (h
it
)
2
3.715 3.714 2.959 2.967

1.309 1.308 1.098 1.101

100 h
it
2.195 2.218 1.542 1.676

0.482 0.487 0.342 0.371
v
TABLE 5
AVERAGE PARAMETER VALUES IN THE MODELS OF

ln
D
it
w
IN JAPAN
T-ratios are reported in italics.
A B C D

untrimmed
inflation
trimmed
inflation
untrimmed
inflation
trimmed
inflation
untrimmed
inflation
trimmed
inflation
untrimmed
inflation
trimmed
inflation

1
ln
D
it
w

0.198 0.196 0.201 0.200 0.198 0.197 0.201 0.200
30.64 30.46 31.19 31.10 30.65 30.47 31.24 31.12
A
it

-47.31 -47.34 -46.48 -46.51

-2.445 -2.447 -2.400 -2.403

2
) ( 100
A
it


-32.69 -32.75 -28.54 -28.80

-0.182 -0.182 -0.157 -0.159

AP
it

52.66 51.61 53.57 52.72

0.940 0.922 0.960 0.945
AN
it

40.79 37.75 40.55 37.68

1.526 1.414 1.519 1.412
UP
it

23.70 24.63 23.36 24.29 23.72 24.65 23.39 24.32

16.00 16.59 15.63 16.23 15.93 16.53 15.64 16.24
UN
it

-30.33 -30.95 -30.05 -30.69 -30.36 -31.00 -30.13 -30.76

-19.97 -20.35 -19.80 -20.19 -19.95 -20.34 -19.84 -20.23
100 (h
it
)
2
69.32 69.32 80.51 80.52

1.148 1.148 1.378 1.378

100 h
it
117.1 117.1 124.9 125.0

1.136 1.136 1.221 1.222
vi
TABLE 6
AVERAGE PARAMETER VALUES IN THE MODELS OF

ln
D
it
v
IN NIGERIA
T-ratios are reported in italics.
A B C D

untrimmed
inflation
trimmed
inflation
untrimmed
inflation
trimmed
inflation
untrimmed
inflation
trimmed
inflation
untrimmed
inflation
trimmed
inflation

1
ln
D
it
v

0.616 0.608 0.617 0.609 0.614 0.608 0.617 0.609
39.12 38.28 39.30 38.23 39.10 38.28 39.22 38.25
A
it

1.148 0.357 1.298 0.369

0.417 0.453 0.472 0.469
2
) ( 100
A
it


0.142 0.302 0.045 0.286

0.105 0.794 0.033 0.754
AP
it

1.369 -0.026 1.236 -0.144

1.283 -0.033 1.151 -0.179
AN
it

-2.094 -1.276 -2.421 -1.481

-1.042 -0.732 -1.201 -0.846
UP
it

0.110 0.426 0.028 0.346 0.136 0.397 0.039 0.342

0.433 1.276 0.112 1.041 0.538 1.193 0.153 1.027
UN
it

-0.034 -0.448 -0.028 -0.444 -0.049 -0.420 -0.035 -0.438

-0.137 -1.298 -0.111 -1.277 -0.198 -1.218 -0.140 -1.259
100 (h
it
)
2
2.628 -1.828 1.137 -1.371


0.786 -0.455 0.420 -0.326

100 h
it


50.38 -28.71 24.15 -19.72


1.085 -0.507 0.626 -0.334
vii
TABLE 7
AVERAGE PARAMETER VALUES IN THE MODELS OF

ln
D
it
w
IN NIGERIA
T-ratios are reported in italics.
A B C D

untrimmed
inflation
trimmed
inflation
untrimmed
inflation
trimmed
inflation
untrimmed
inflation
trimmed
inflation
untrimmed
inflation
trimmed
inflation

1
ln
D
it
w

0.125 0.121 0.124 0.121 0.124 0.121 0.123 0.121
7.715 7.563 7.661 7.575 7.621 7.563 7.612 7.566
A
it

-7.868 -5.147 -7.468 -5.084

-1.041 -0.925 -0.989 -0.914
2
) ( 100
A
it


4.365 1.065 1.065 1.065

1.065 1.124 1.023 1.112
AP
it

5.525 4.300 5.331 4.271

1.978 1.997 1.898 1.979
AN
it

14.69 5.889 13.95 5.650

2.597 1.786 2.462 1.703
UP
it

3.315 3.420 3.172 3.402 3.313 3.396 3.181 3.385

5.034 4.044 4.829 4.040 5.055 4.017 4.858 4.021
UN
it

-1.416 -0.899 -1.408 -1.019 -1.412 -0.886 -1.425 -1.009

-2.036 -1.035 -2.033 -1.171 -2.032 -1.020 -2.059 -1.160
100 (h
it
)
2
2.440 10.25 4.927 10.28

0.426 1.561 1.060 1.458
100 h
it
33.74 143.6 65.38 142.9

0.424 1.576 0.995 1.475
viii
APPENDIX TABLE A1
UNANTICIPATED INFLATION COEFFICIENTS IN THE SEMI-PARAMETRIC MODEL OF ln(
D
it
v
)

T-ratios are in italics.

z = 1.0 z = 1.5 z = 2.0
UP
it

-0.361 -0.314 -0.300

-6.311 -6.035 -6.044
UN
it

-0.503 -0.601 -0.616
-7.684 -10.086 -10.841



APPENDIX TABLE A2
UNANTICIPATED INFLATION COEFFICIENTS IN THE SEMI-PARAMETRIC MODEL OF ln(
D
it
w
)

T-ratios are in italics.

z = 1.0 z = 1.5 z = 2.0
UP
it

1.522 1.558 1.573

13.671 14.567 14.975
UN
it

-0.768 -0.852 -0.850
-6.012 -6.943 -7.059




ix
APPENDIX TABLE A3
ITEMS INCLUDED IN THE CANADIAN DATASET
Fresh food Dry & packaged food
Bacon (unsliced)

Coffee
Bacon (sliced)

Corn (canned)
Butter (creamery)

Corn syrup
Butter solids

Currants
Cheese

Flour
Eggs (cooking)

Peaches (canned)
Eggs (fresh)

Peas (canned)
Finnan haddie

Prunes
Ham (sliced)

Raisins
Lard

Rice
Leg of lamb

Rolled oats
Milk

Salmon (canned)
Mutton leg roast

Sugar (granulated)
Onions

Sugar (yellow)
Potatoes (15lb bag)

Tapioca
Potatoes (100lb bag)

Tea
Rib roast

Tomatoes (canned)
Round steak


Salt cod


Salt mess pork


Shoulder roast


Sirloin steak

Soda biscuits

Stewing beef

Veal shoulder

x
APPENDIX TABLE A4
CITIES INCLUDED IN THE CANADIAN DATASET
City Province City Province
Amherst Nova Scotia Stratford Ontario
Halifax Nova Scotia Sudbury Ontario
New Glasgow Nova Scotia Timmins Ontario
Sydney Nova Scotia Toronto Ontario
Truro Nova Scotia Windsor Ontario
Windsor Nova Scotia Woodstock Ontario
Charlottetown Prince Edward Island Hull Quebec
Bathurst New Brunswick Montreal Quebec
Fredericton New Brunswick Quebec Quebec
Moncton New Brunswick Saint Hyacinthe Quebec
Saint John New Brunswick Saint Johns Quebec
Belleville Ontario Sherbrooke Quebec
Brantford Ontario Sorel Quebec
Brockville Ontario Thetford Mines Quebec
Chatham Ontario Trois-Rivires Quebec
Cobalt Ontario Brandon Manitoba
Fort William Ontario Winnipeg Manitoba
Galt Ontario Moose Jaw Saskatchewan
Guelph Ontario Prince Albert Saskatchewan
Hamilton Ontario Regina Saskatchewan
Kingston Ontario Saskatoon Saskatchewan
Kitchener Ontario Calgary Alberta
London Ontario Drumheller Alberta
Niagara Falls Ontario Edmonton Alberta
North Bay Ontario Lethbridge Alberta
Orillia Ontario Medicine Hat Alberta
Oshawa Ontario Fernie British Columbia
Ottawa Ontario Nanaimo British Columbia
Owen Sound Ontario Nelson British Columbia
Peterborough Ontario New Westminster British Columbia
Port Arthur Ontario Prince Rupert British Columbia
Saint Catharines Ontario Trail British Columbia
Saint Thomas Ontario Vancouver British Columbia
Sarnia Ontario Victoria British Columbia
Sault Sainte Marie Ontario
xi
APPENDIX TABLE A5
ITEMS INCLUDED IN THE JAPANESE DATASET

(i): Food for the Home
Asparagus Clams in soy sauce Furikake seasonings Oranges Sausages
Bacon Cooked curry Grapefruit Peanuts Scallops
Baked fish bars Cream puffs Green peppers Pickled cabbage Sea bream
Bananas Croquettes Gyoza Pickled plums Shiitake mushrooms
Bean curd Cucumbers Hens eggs Pickled radishes Shimeji mushrooms
Bean sprouts Cuttlefish Horse mackerel Pork cutlets Soy sauce
Bean jam buns Deep fried chicken Ice cream Pork loin Soybean paste
Bean jam cakes Devil's tongue jelly Imported beef Pork shoulder Spaghetti
Beef loin Dried bonito fillets Imported cheese Potato chips Spinach
Beef shoulder Dried horse mackerel Instant curry Powdered milk Steamed fish cakes
Biscuits Dried laver Jam Prawns Sugar
Boiled beans Dried sardines Jelly Pudding Sweet bean jelly
Boiled noodles Dried mushrooms Kasutera cakes Pumpkins Sweet potatoes
Boxed lunches Dried small sardines Kidney beans Radishes Tangle in soy sauce
Broccoli Dried tangle Kimuchi Red beans Taros
Broiled eels Dried young sardines Kiwi fruits Rice (not koshihikari) Tomatoes
Burdock Edible oil Lemons Rice (koshihikari) Tuna fish
Butter Eggplants Lettuce Rice balls Uncooked noodles
Cabbage Enokidake mushrooms Liquid seasonings Rice cakes Veg in soy sauce
Cakes Fermented soybeans Liver Rice crackers Vinegar
Candies Fish in soybean paste Lotus roots Roast ham Wakame seaweed
Canned oranges Flavor seasonings Mackerel Salad Welsh onions
Canned peaches Flounder Margarine Salmon Wheat crackers
Capelin Fresh milk (bottled) Mayonnaise Salted cod roe Wheat flour
Carrots Fresh milk (cartons) Mazegohan no moto Salted fish guts White bread
Cheese Fried bean curd Mochi rice-cakes Salted salmon White potatoes
Chicken Fried fish patties Nagaimo Sandwiches Worcester sauce
Chinese cabbage Frozen croquettes Octopus Sardines Yellowtail
Chocolate Frozen pilaf Onions Saury Yogurt
Clams

xii

(ii): Other I tems
Drink Household items Room air conditioner Mini disk player Toothbrushes
100% fruit drinks Alarm clock Rush floor covering Notebook Toothpaste
Black tea Bathtub Scrubbing brush Pants for exercise

Calpis Bed Sealed kitchen ware Pencil cases Apparel
Canned coffee Boards Sewing machine Pencils Adults' canvas shoes
Coffee beans Carpet Sheets Personal computer Baby clothes
Foaming liquors Chests of drawers Sitting table Roses Baseball cap
Green tea (Bancha) Curtains Telephone set Soccer ball Belt
Green tea (Sencha) Dining set Toilet seat Swimming suit Boy's short pants
Imported beer Dishes Towel Toy car Child's canvas shoes
Imported whisky Electric iron Vacuum cleaner TV set (CRT) Child's shoes
Imported wine Electric pot Wardrobes

Child's undershirt
Instant coffee Electric rice cooker Washing machine Pharmacy items Handkerchief
Local beer Fabric softener Water purifier Chinese medicine Imported handbag
Local whisky (40%+) Facial tissue Wine glass Cold medicine Imported necktie
Local whisky (43%+) Fluorescent fittings

Contact lens cleaner Imported watch
Mineral water Fluorescent lamp Sports / leisure goods Dermal medicine Local handbag
Sake (grade A) Food wrap Baseball gloves Disposable diapers Local necktie
Sake (grade B) Fragrance Bicycle Eyewashes Local watch
Sports drinks Gas cooking table Building blocks Face cream Men's briefs
Vegetable juice Gasoline Camera Face lotion Men's business shirt

Glasses Carnations Foundation Men's shoes
Restaurant food Hot water equipment Chrysanthemums Stomach medicine Men's suit materials
Chicken & rice Imported pan Computer game Hair dye Men's umbrella
Chinese noodles Insecticide Copy paper Hair liquid Men's undershirt
Coffee Kitchen cabinet Doll Hair rinse Panty hose
Curry & rice Kitchen detergent Dry electric battery Health drinks Slips
Gyudon beef on rice Laundry detergent Film Imported shaver Suitcase
Hamburger steaks Local pan Fishing rod Lipstick Women's blue jeans
Hamburgers Microwave oven Gardening soil Local shavers Women's sandals
Hand rolled sushi Moth balls Golf clubs Plasters Women's shoes
Japanese noodles Quilt Imported tennis racket Sanitary napkins Women's socks
Shrimp & rice Refrigerator Local tennis racket Shampoo Women's zori sandals
Sushi rolled in laver Rice bowl Marking pens Spectacles Woollen yarn

Rolled toilet paper Mini disk media Toilet soap

xiii
APPENDIX TABLE A6
CITIES INCLUDED IN THE JAPANESE DATASET
Akita Kobe Otsu
Aomori Kochi Saga
Asahikawa Kofu Saitama
Atsugi Koriyama Sakura
Chiba Kumamoto Sapporo
Fuchu Kyoto Sasebo
Fukui Maebashi Sendai
Fukuoka Matsue Shizuoka
Fukushima Matsumoto Tachikawa
Fukuyama Matsuyama Takamatsu
Gifu Mito Tokorozawa
Hakodate Miyazaki Tokushima
Hamamatsu Morioka Tokyo
Higashi-Osaka Nagano Tottori
Himeji Nagaoka Toyama
Hirakata Nagasaki Tsu
Hiroshima Nagoya Ube
Itami Naha Utsunomiya
Kagoshima Nara Wakayama
Kanazawa Niigata Yamagata
Kasugai Nishinomiya Yamaguchi
Kawaguchi Oita Yokohama
Kawasaki Okayama Yokosuka
Kitakyushu Osaka

xiv
APPENDIX TABLE A7
ITEMS INCLUDED IN THE NIGERIAN DATASET
Fresh food Apparel
Bananas

Embroidery lace (per metre)
Beans (brown)

Guinea brocade (per metre)
Beans (white)

Khaki drill (per metre)
Beef

Mattress
Carrots

Mens shoes
Chicken (agricultural)

Pillow
Chicken (locally produced)

Poplin (per metre)
Gari (white)

Singlet
Gari (yellow)

Womens shoes
Guinea corn


Irish potatoes

Services
Kola nuts

Blood test
Maize (white)

Rent for a flat
Maize (yellow)

Rent for a bungalow
Okra

Rent for a room with parlour
Onions

Rent for a room
Oranges

Room in a hotel
Rice (locally produced)

Taxi fare (per kilometre)
Salt

Sweet potatoes

Tomatoes

Yams


xv
APPENDIX TABLE A8
CITIES INCLUDED IN THE NIGERIAN DATASET
City State City State

Abakaliki Ebonyi Jalingo Taraba
Abeokuta Ogun

Jos Plateau
Abuja Federal Capital Territory

Kaduna Kaduna
Ado-Ekiti Ekiti

Kano Kano
Akure Ondo Katsina Katsina
Asaba Delta Lafia Nasarawa
Awka Anambra Lokoja Kogi
Bauchi Bauchi Maiduguri Borno
Benin City Edo Makurdi Benue
Birnin Kebbi Kebbi Minna Niger
Calabar Cross River Oshogbo Osun
Damaturu Yobe Owerri Imo
Dutse Jigawa Port Harcourt Rivers
Enugu Enugu Sokoto Sokoto
Gombe Gombe Umuahia Abia
Gusau Zamfara Uyo Akwa Ibom
Ibadan Oyo Yenagoa Bayelsa
Ikeja Lagos Yola Adamawa
Kano Kano

xvi

FIG. 1. Time series of aggregate average inflation (
t
) in Canada.


FIG. 2. Time series of aggregate average inflation (
t
) in Japan.


FIG. 3. Time series of aggregate average inflation (
t
) in Nigeria.
1925 1930 1935 1940
-0.04
-0.02
0.00
0.02
0.04
0.06


t
2001 2002 2003 2004 2005 2006
-0.010
-0.005
0.000
0.005
0.010


t
2002 2003 2004 2005 2006
-0.10
-0.05
0.00
0.05


t
xvii
-0.10 -0.08 -0.06 -0.04 -0.02 0.00 0.02 0.04 0.06 0.08 0.10
200
400
600
800
1000
1200
requency f

FIG. 4. Distribution of inflation (
it
) in Canada trimmed at 10%.
-0.10 -0.08 -0.06 -0.04 -0.02 0.00 0.02 0.04 0.06 0.08 0.10
1000
2000
3000
4000
5000
6000
7000 f requency

FIG. 5. Distribution of inflation (
it
) in Japan trimmed at 10%.
xviii
-0.10 -0.08 -0.06 -0.04 -0.02 0.00 0.02 0.04 0.06 0.08 0.10
50
100
150
200
f requency

FIG. 6. Distribution of inflation (
it
) in Nigeria trimmed at 10%.
-3.5 -3.0 -2.5 -2.0 -1.5 -1.0
100
200
300
400
500
600
requency f

FIG. 7. Distribution of relative price variability (ln(v
it
)) in Canada.
xix
-4.5 -4.0 -3.5 -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 0.0
500
1000
1500
2000
requency f

FIG. 8. Distribution of relative price variability (ln(v
it
)) in Japan.
-6 -5.5 -5 -4.5 -4 -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5
50
100
150
200
250
requency f

FIG. 9. Distribution of relative price variability (ln(v
it
)) in Nigeria.
xx

FIG. 10. Recursive estimates of the Canadian RPV-inflation parameters two standard errors.

FIG. 11. Recursive estimates of the Canadian RIV-inflation parameters two standard errors.
1930 1935 1940
-2
-1
0
1
2

A
p a r a m e t e r
1930 1935 1940
-2
0
2
4
6
8
(
A
)
2
p a r a m e t e r
1930 1935 1940
-0.75
-0.50
-0.25
0.00

U P
p a r a m e t e r
1930 1935 1940
-1.25
-1.00
-0.75

U N
p a r a m e t e r
1930 1935 1940
-4
-3
-2
-1
0
1
2
3
A
p a r a m e t e r
1930 1935 1940
-10
-5
0
5
10
15
(
A
)
2
p a r a m e t e r
1930 1935 1940
2.0
2.5
3.0
3.5
4.0
4.5

U P
p a r a m e t e r
1930 1935 1940
-4.0
-3.5
-3.0
-2.5
-2.0
-1.5

U N
p a r a m e t e r
xxi

FIG. A1. Values of

dln d
D A
it it
v with a 95% confidence interval (semi-parametric model, z = 1%).

FIG. A2. Values of

dln d
D A
it it
v with a 95% confidence interval (semi-parametric model, z = 1.5%).
-2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5
-9
-8
-7
-6
-5
-4
-3
-2
-1
0
1
2
3
4
5
6 dln ( v
D
i t
) / d
A
i t

A
i t
-2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5
-5
-4
-3
-2
-1
0
1
2
3 dln( v
D
i t

) / d
A
i t

A
i t
xxii

FIG. A3. Values of

dln d
D A
it it
v with a 95% confidence interval (semi-parametric model, z = 2%).

FIG. A4. Values of

dln d
D A
it it
v with a 95% confidence interval (quadratic model).
-2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5
-3
-2
-1
0
1
dln( v
D
i t
)

/ d
A
i t

A
i t
-2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5
-16
-14
-12
-10
-8
-6
-4
-2
0
2
4
6
8
10
12
14
16
18
dln( v
D
i t
)

/ d
A
i t

A
i t
xxiii

FIG. A5 Values of

dln d
D A
it it
w with a 95% confidence interval (semi-parametric model, z = 1%).

FIG. A6. Values of

dln d
D A
it it
w with a 95% confidence interval (semi-parametric model, z = 1.5%).
-2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5
-8
-6
-4
-2
0
2
4
6
8
dln( w
D
i t

) / d
A
i t

A
i t
-2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5
-4
-3
-2
-1
0
1
2
3
4
5
dln( w
D
i t

) / d
A
i t

A
i t
xxiv

FIG. A7. Values of

dln d
D A
it it
w with a 95% confidence interval (semi-parametric model, z = 2%)

-2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5
-2
-1
0
1
2
dln( w
D
i t

) / d
A
i t

A
i t