n?u=RePEc:hhs:nhheco:2016_022&r=mac

INSTITUTT FOR SAMFUNNSØKONOMI

DEPARTMENT OF ECONOMICS

SAM 22 **2016**

ISSN: 0804-6824

December **2016**

Discussion paper

Intermittent Price Changes in

Production Plants: Empirical Evidence

using Monthly Data

BY

Øivind A. Nilsen AND Magne Vange

This series consists of papers with limited circulation, intended to stimulate discussion

.

Intermittent Price Changes in Production Plants:

Empirical Evidence using Monthly Data ∗

Øivind A. Nilsen †

(Norwegian School of Economics, CESifo-Munchen, IZA-Bonn)

Magne Vange ‡

(Norwegian School of Economics)

December **2016**

Abstract

The price-setting behaviour of manufacturing plants is examined using a

large panel of monthly surveyed plant- and product-specific prices. The sample

shows a high frequency of zero changes, relatively small price changes, and

a strong seasonal price change pattern. The intermittent feature of price

changes is modelled with thresholds which are lower in January, and a quadratic

loss–function associated with the distance from target price. The findings show

statistically significant pricing thresholds, which are only two thirds in January,

and partial adjustment parameters implying that 60% of the deviation between

the target price and current price is closed each month.

Keywords: Price Setting, Micro Data, Simulated Method of Moments

JEL classification: E30, E31, E37.

Number of words: 10296

∗ Acknowledgement: The paper has benefited greatly from valuable comments by Wilko Letterie,

and from thorough discussions and assistance of Eivind Bøe when testing out the ideas using annual

data. We are also grateful to Magne K. Asphjell for providing help with preparing the dataset. Part

of this work was done while Nilsen visited the Institute for Macroeconomics and Econometrics -

University of Bonn, whose hospitality is gratefully acknowledged. The usual disclaimer applies.

† (Corresponding author); Department of Economics, Norwegian School of Economics, NO-5045

Bergen, Norway; Phone: +47 55 959 281; E-mail address: oivind.nilsen@nhh.no

‡ Department of Economics, Norwegian School of Economics, NO-5045 Bergen, Norway; E-mail

address: magnevange@gmail.com

1 Introduction

Most modern macroeconomic models assume price stickiness, i.e. that price setters are

faced with frictions, without sufficient knowledge about the underlying microeconomic

implications. This calls for a further empirical assessment of the theoretical premises

in the macroeconomic models we use today.

A common method for analysing price stickiness is to investigate the role of

thresholds in the pricing patterns of individual firms. In this literature, the (S, s)

rule proposed by Sheshinski and Weiss (1977) plays an important role. Sheshinski

and Weiss (1977) argued that firms kept the price fixed within the certain bounds,

denoted (S, s). As a result, prices exhibit a pattern of inaction followed by large price

changes, so called “zeros and lumps”. The authors argue that this pattern is caused

by the fact that changing the price induces a fixed cost for the firm, which is referred

to as the menu cost. The (S, s) methodology has later been adopted and further

extended by many, and thereby represents a large share of the current price stickiness

literature (see e.g. Caballero and Engel, 1993; Ratfai, 2006; Alvarez et al., 2011;

Dhyne et al., 2011; Honoré et al., 2012). An essential assumption in these models is

that adjustment costs are independent of the size of the price change (Zbaracki et al.,

2004).

One aspect to consider when searching for thresholds in pricing patterns is whether

the thresholds are symmetric, i.e. if the magnitude of the thresholds are the same

upwards and downwards. A study on microeconometric evidence from Switzerland

by Honoré et al. (2012), finds a smaller upper than lower threshold. According to

this study, price changes are more likely to be positive than negative, ceteris paribus.

The study ignores, however, the magnitude of price changes, as only the frequency

and the duration of inaction are accounted for. Loupias and Sevestre (2012), on the

other hand, include the magnitude of price changes, and find that when firms face

cost variations, they appear to adjust their prices more often and more rapid upwards

1

than downwards.

The counterpart of the (S,s) methodolgy in the price stickiness literature assume

that the adjustment cost is a convex function of the size of the price change, i.e. that

larger changes lead to higher costs (Rotemberg, 1982). While the assumption of fixed

costs implies that one should observe large and infrequent price changes, the convex

cost assumption implies the opposite: frequent changes of small size. As emphasized

by Zbaracki et al. (2004), most of the literature finds evidence supporting the former.

However, if there are only fixed and not convex price adjustments costs, we fail to see

why the pricing data shows a relatively high proportion of small price changes. 1 Earlier

research with (S, s) pricing rules has in part failed to include small price changes.

As highlighted by Klenow and Malin (2011), access to good microeconomic data

is crucial, and is a common problem in all empirical research related to pricing. The

basis of our analysis is monthly collected micro price data for Norwegian manufacturers.

Although consumer prices are relevant for the monitoring of inflation by

central banks, it is the prices on producer level that are most often modelled into

the macroeconomic policy models (Vermeulen et al., 2012). Accordingly, knowledge

about producer price adjustments is essential to improve macroeconomic modelling

and central bank policies.

In this paper we propose a model where the adjustment towards the new price is

conditional on both thresholds and partial adjustments. Our model therefore allows

for both inaction and inertia in pricing.

The hypothesis is that the firm is faced

with a fixed cost when setting a new target price and that there are costs associated

with deviating from the frictionless price in addition to convex costs associated with

adjusting to this price. For example, as in Zbaracki et al. (2004), the convexity of

1 The study of Eichenbaum et al. (2014) on CPI data suggests that the observance of small

price changes is largely due to measurement errors and quality adjustments, and should therefore

be neglected. However, the study is opposed by a vast majority of empirical research suggesting

that small price changes are relatively common (Klenow and Kryvtsov, 2008; Barros et al., 2009;

Bhattarai and Schoenle, 2014; Midrigan, 2011; Wulfsberg, **2016**).

2

managerial-, customer- and negotiation costs makes the firm favour slow adjustments

and small price changes. Thus, our model sets out to explain both the occurrence

price adjustments of different sizes and inaction. The model is tested on a dataset

based on survey data behind the commodity price index for the Norwegian industrial

sector (PPI). These data include monthly prices quotations for a representative

sample of Norwegian plants. The data show a high frequency price change inaction,

by relatively small price changes when changed, and by a much higher occurrence

of price changes in the beginning compared to the end of a year.

To analyse the

intermittent nature of the model, a simulated method of moments is used. The estimations

reveal thresholds such that prices are changed only if the deviations from

the underlying frictionless prices are approximately 15 percent. When changed, the

prices are changed rather quickly with only 10 percent of the initial gap existent after

three months. The asymmetry between upward- and downward rigidities are minor

but statistically significant. Finally, the thresholds in January are approximately two

thirds compared to the other months.

The remainder of the paper is organized as follows. Section 2 describes the data,

while the model, method and moments are presented in Section 3. Section 4 reports

and discusses the results and Section 5 gives some concluding remarks.

2 Data

The basis for our empirical analysis is the survey data behind the commodity price

index for the Norwegian manufacturing industry (PPI) obtained from Statistics Norway

(SSB). 2 The data are collected on a monthly basis for a selection of Norwegian

plants. 3 Plants with more than 100 employees are included in the sample at all times,

and the selection of producers is updated continuously, securing a high level of rel-

2 See SSB (2015) for more information about the PPI.

3 In the remainder of the paper we use the terms plant, firm, producer and establishment interchangeably.

3

evance (SSB, 2015). Plants are repeatedly surveyed, participation is compulsory and

Statistics Norway revise the data regularly to detect measurement errors and nonconformity.

4

Considering this, and that the PPI is an important tool for governing

bodies, it is fair to assume that the data is representative for Norwegian producers

and of high quality.

The initial dataset contains monthly price observations for Norwegian producers

ranging from year 2002 until 2009. In the construction of the final dataset for this

study, plants with observations for less than 24 months have been omitted, as well

as plants with less than 10 employees. Furthermore, only years with observations

for all months in a given year are included. Due to the implementation of a new

sampling procedure at Statistics Norway, there was a clear shift in the reported price

change frequency in 2004. We therefore choose to discard the data prior to January

2004. Furthermore, plants related to the energy sector (oil, gas, electricity, etc.), and

mining and quarrying have been left out of the sample as they are known to have

an abnormally high adjustment frequency. The original dataset also contains prices

for both domestic and export markets, but to prevent interference by exchange rate

movements and international competition, export market prices are omitted. Additionally,

as very large price changes are likely to reflect changes to design or quality of

the product rather than common pricing decisions, price growth observations outside

the [0.01, 0.99] interval we consider to be new products. Finally, we focus on single

plant firms only. 5 This leaves us with a final sample of 76 804 observations for 1676

products over the years 2004-2009 covering 21 2-digit SIC2002 industry codes.

4 One plant might be recorded with one or multiple products. It should be noted that for data

collection purposes firms may be targeted for certain, but not all of the products they manufacture.

If Statistics Norway regards a subset of the products to be important to obtain an accurate estimate

of the price index, data will be requested for these ones only.

5 With this choice, we are sure that the price decisions are not made beyond the plant level.

4

2.1 Descriptives

[Figure 1 "Distribution of Price Change Rates" about here]

Figure 1 shows the proportion of observations in different price change intervals,

both for the actual data and for the later preferred simulated model (black, and grey

columns, respectively). As seen, observations with price changes with absolute value

less than 0.005 represent the majority of the dataset. We will later on refer to this as

the “zero price-change” interval. In other words, most observations are characterized

with inaction. This indicates the existence of fixed price adjustment costs. At the

same time, we observe that a substantial proportion of the observations are small price

changes in the intervals between 0.5% and 5.0% (in absolute value). If there is only

a fixed cost, which is independent of the magnitude of the price change, one would

not expect to see these small price changes. This observation could, however, be an

indication of convex adjustment costs, which put a penalty on large adjustments and

thereby force the producers to adjust gradually. The observation of several periods of

inaction combined with series of small price changes may tell a story of firms being

faced with both fixed and convex price adjustment costs.

In order to identify lumpy adjustment behaviour, we rank for each plant, for each

year, the 12 monthly price changes from lowest to highest. 6 Rank 1 thereby represents

the average largest monthly price change, Rank 2 the average second largest price

change, and so on. The intuition is that if there is a large gap between the largest

(smallest) and the second largest (second smallest) price change compared to the

other ranks, this indicates that producers are faced with fixed costs of adjustment

and therefore change the price quite substantially when first changing it. Opposite,

with normally distributed shocks to the fundamentals, and no adjustment costs, one

would expect the mean investment rate of adjacent observations to be rather similar,

6 Such a measure has been used in the investment- and labour demand literature (see for instance

Doms and Dunne (1998), Nilsen and Schiantarelli (2003) and Varejão and Portugal (2007)).

5

and therefore that there is a downward sloping linear relationship between the ranks

(for more details, see Doms and Dunne 1998).

[Figure 2 "Ranked Price Change Rates" about here]

Figure 2 shows the ranking of the monthly price changes of each plant by month

for every year, and then taking the average of each rank across plants and years.

As seen from Figure 2, there is a gap of approximately 3 percentage points between

both the first and second rank and the eleventh and twelfth rank. In contrast, the

differences between the intermediate ranks are modest. The importance of episodes

characterized by relatively large prices changes is consistent with nonconvexities in

adjustment costs. Note however, even though the difference between the gaps on the

edges and the gaps between the middle ranks is existent, the potential co-existence

of both fixed and convex adjustment costs can not be excluded. That means, even

if fixed adjustment costs are preventing the firms to adjust continuously, when they

actually do change their price, convex costs are forcing them to do so gradually. 7

[Figure 3 "The Occurrence of Price Changes by Months" about here]

Figure 3 shows the average share of price change quotations within each month.

The shares are given as the number of price changes larger than |0.005| within each

month divided by the total number of price quotations within the same month. As

seen, there is a relatively high price change frequency in the beginning of the year

compared to the remaining months. This seasonality could of course be explained by

producers’ economic environment, for instance seasonal demand effects. Furthermore,

it may be explained by staggered contracts with for instance price contracts starting in

January with a duration of one year (see for instance Taylor (1980, 1999)). It should

be noted that the higher incidence of price changes in January, is also consistent

7 In addition, we observe that all the ranks are shifted to the left, as only rank five is below

zero. This observation is expected as inflation will cause the producers to have more positive price

changes than negative.

6

with theories focusing on the costs of information acquisition and processing (see

Maćkowiak and Wiederholt (2009) and Mankiw and Reis (2002)).

3 Model, Method and Moments

As already discussed, there are several theories that could explain the intermittent

price adjustment patterns observed in many datasets at both consumer level and at

producer level.

Here we suggest a simple reduced form model that describes production

plants’ price adjustment behaviour with the following three features; plants

adjust prices infrequently as only 20 percent of the price observations change from one

month to another, there is a lot of small price changes, and there is a seasonal pattern

in the incidence of price changes with most price changes taking place in January.

3.1 Model Specification and Predefined Parameters

As firms require a degree of monopoly power to be able to set prices, we assume that

producers operate in monopolistic competitive markets. Furthermore, it is assumed

that each firm is able to continuously observe and monitor its frictionless price without

any costs.

We start from the observations of high frequencies of zero price adjustments. This

would be observed if there would be some menu costs, and that it would be costly to

continuously adjust the product prices. The firm operates with a target price (in logs)

for product i at time t denoted p # it , and leave this unchanged unless the distance to

the frictionless price p ∗ it (also in logs) is becoming too large. The latter mentioned p ∗ it

represents the frictionless equilibrium price decided by the market conditions. The

costs associated with setting a new target price is F · I ( p # it ≠ p # it−1)

, where F is the

actual costs and I ( p # it ≠ p # it−1)

is an indicator function. The formation of the target

7

price is determined by:

⎧

⎪⎨

p # it =

⎪⎩

p ∗ it

p # it−1

if | p ∗ it − p # it−1 |> τ

otherwise

(1)

where τ denotes a threshold.

Thus, if the shock to the frictionless price is large

enough, in absolute value, relative to its target value, the firm finds it profitable to

set a new target price and to start to adjust its price in response to the shock. The

formulation in eq. (1) states that the threshold is symmetric, i.e. that the “band

of inaction” is the same whether the price shock is positive or negative. We relax

this restriction and allow the thresholds for price increase and price decreases to be

different. With this modification, the formation of the target price is determined by:

⎧

⎪⎨ p ∗

p # it if p ∗ it − p # it−1 > U or p ∗ it − p # it−1 < L,

it =

⎪⎩ otherwise

p # it−1

where U denotes the upper threshold and L denotes the lower threshold, i.e. L ≤

0 ≤ U. It means that the target price is changed only if the frictionless price moves

(2)

outside the interval determined by L and U.

Opposite, if the frictionless price is

larger than L and lower than U, the producer leaves its target price p # it unchanged.

Following Alvarez et al. (2011), Dias et al. (2015), and others, we let the logarithm

of the frictionless nominal price for product i at time t , denoted by p ∗ it, follow a random

walk with drift:

p ∗ it = α + p ∗ it−1 + ε it , where ε it ∼ N(0, σ 2 ε) (3)

Here α denotes the deterministic drift, and ε it

denotes idiosyncratic shocks with

variance σ 2 ε.

If α were not included, trend inflation would be embedded into the

threshold parameters and therefore bias the results. If α is set too low (too high)

compared to the actual trend inflation, the estimated threshold parameters L and U

8

would be biased downwards (upwards). The deterministic trend parameter, α, is set

as close to the actual inflation as possible to limit the effect of inflation bias. 8

The

idiosyncratic shock parameter, ε it , is meant to reflect any shocks to either demand,

cost or technology excess of the underlying trend captured by the trend parameter α. 9

In principle, it is possible to allow for serial correlation in ε it , but for computational

ease and in order to simplify the exposition, we assume that the shocks are serially

uncorrelated.

Even though the firm has decided to change its target price, and therefore also to

change the price of the given product, it does not move directly to the new target price.

As discussed by for instance Zbaracki et al. (2004), price change costs also include

convex components such it is costlier to make one larger price change compared to

several smaller ones. Furthermore, the descriptive evidence given in Figure 1 indicates

that smaller prices changes are not that uncommon, while larger price changes are.

Nevertheless, there might also be losses of being too far away from the new target price

p # it . A formulation that encompasses both these elements, given that the deviation

between the new frictionless price, p ∗ it, and the target price p # it

is large enough to

initiate price changes, the “out-of-equilibrium costs” could be modelled as follows

AC(p it ) = C ·

{

(1 − θ) ( p it − p # ) }

2

it + θ (pit − p it−1 ) 2

(4)

Thus, the formulation consists of a weighted sum (where 0 ≤ θ ≤ 1) of two

8 We could also include α as a parameter to be estimated. A simpler approach, as adopted here,

is to perform a series of simulations with different values for α. We comment on this further when

testing the robustness of our model.

9 An argument against letting demand-, technology-, and or cost- shocks to be treated identically

is that firms react quicker to positive than to negative cost shocks, but slower to positive than

to negative demand shocks (see for instance found in Dias et al. (2015) and Loupias and Sevestre

(2012)). Differencing between the types of shocks would implicate a more sophisticated derivation

of both the frictionless price and the inertia parameters than presented above. Furthermore, with

the available data it might also be difficult to identify different types of shocks. Considering this,

and that the focus of our paper is the combination of thresholds and inertia, we choose to include

the shocks as an aggregate effect (see Appendix for an analysis how the price is affected by demand-,

technology, and cost- shocks).

9

quadratic terms, which denotes the difference between the new price and the target

price, and the difference between the new price and the previous price, respectively.

A plant seeks to minimise these “out-of-equilibrium costs” AC(.). The first order

condition of equation (4) with respect to new current price p it rearranged is therefore;

(p it − p it−1 ) = (1 − θ) ( p # it − p it−1

)

(5)

Thus, we have the traditional partial adjustment model where the “out-of-equilibrium

costs” AC(.) prevent the producer to adjust immediately to its target price, only that

the “usual” frictionless seen in partial adjustment models is exchanged with the target

price p # it . An implication is that the producer will close (1 − θ) of the deviation

between the target price and the old actual price. For example, θ = 0.10 will implicate

that the producer closes 90 percent of its desired price change in the first period. If

the target price remains unchanged in the subsequent period, the producer will close

90 percent of the remaining price gap. This will keep on until the producer decides

to set a new target price or when the target price is reached.

To avoid the restriction that the weights in equation (4), (1 − θ) and θ, are common

to price increases and price decreases, we allow the model specification for asymmetric

inertia in addition to the already discussed asymmetric thresholds. This reflects

asymmetric adjustment costs discussed and analysed in the microeconomic literature

(e.g. Peltzman, 2000; Yang and Ye, 2008; Xia and Li, 2010; Lewis, 2011; Loy

et al., **2016**). Thus, we allow p it to have three outcomes, depending on whether the

price is either equal to the target price, heading upwards or heading downwards. If

the price is heading upwards, θ up is supposed to capture upward inertia. Conversely,

if the price is heading downwards, θ down is supposed to capture downward inertia.

10

Thus, we let the logarithm of the nominal price of product i at time t be given by:

⎧

(1 − θ up )(p # it − p it−1 ) if p # it − p it−1 > 0.005,

⎪⎨

(p it − p it−1 ) = 0 if |p # it − p it−1 | ≤ 0.005, (6)

⎪⎩ (1 − θ down )(p # it − p it−1 ) if p # it − p it−1 < −0.005,

As seen from (6), we consider price differences within the following bounds [−0.005, 0.005]

as zeros. Such small variations are likely to have little economic meaning. In addition,

the numerical simulations described in subsequent section, makes it necessary

to allow for small deviations from a mathematical true zero. Thus, our definition of

zero corresponds to the definition used in later applied moments. The formulation in

(6) also implies that the price does not adjust further when the deviation from the

target price gets small enough.

It should be mentioned that if θ up ≠ 0, θ down ≠ 0, U = L = 0, the model specification

reduces to a partial adjustment model (since then p ∗ it = p # it and eq.s (1) and (2)

would be irrelevant). Conversely, if θ up = θ down = 0, U ≠ 0, L ≠ 0, the model reduces

to a (S, s) pricing model. 10 Note also, if the target price p # it would not have been

introduced explicitly, the (symmetric) threshold specification would be as follows;

⎧

⎪⎨

p it =

⎪⎩

p ∗ it

p it−1

if | p ∗ it − p it−1 |> τ

otherwise

(7)

Furthermore, the (symmetric) partial adjustment expression would then be:

⎧

⎪⎨ (1 − θ)(p ∗ it − p it−1 ), if |p ∗ it − p it−1 | > 0.005,

(p it − p it−1 ) =

⎪⎩ 0 if |p ∗ it − p it−1 | ≤ 0.005,

(8)

Note however, then the price-adjustment process would stop when p ∗ it − p it−1 would

10 Eq. (5) states that conditional on changing, one immediately goes to the new target price.

11

each the threshold τ. Thus, without the target price p # it in the model we would not

observe that many small price changes and p ∗ it would never will be fully reached. 11

It is assumed that the friction parameters U, L, θ up and θ down are all independent

of product characteristics. There is a broad agreement in the literature that price

setting is heterogeneous across sectors, firms and products (Álvarez et al., 2006; Nakamura

and Steinsson, 2008; Dhyne et al., 2011; Fougère et al., 2007; Dias et al.,

2015)). Instead of controlling for such heterogeneity by introducing product- or firm

specific friction parameters, we will estimate the model for different product groups

and therefore allowing all parameters to take different values when checking the robustness

of our model.

[Figure 4: Illustration of Price Change Process - about here]

In Figure 4 we illustrate how our model is working. Starting with the evolvement

of the frictionless price, p ∗ , we see clearly the upward trend, but an interim period

with sudden price decreases. The thresholds have a constant distance relative to the

actual price (bold line). We see in period t = t A , the frictionless price has evolved

such that is larger than the upper threshold U, and consequently is the target price

and the actual price both changed. We see however, that the actual price is moving

slowly towards the new target price. This is caused by the inertia parameter(s) θ.

In period t = t B a sudden negative shock comes, pushing the target price below the

lower threshold L. Subsequently the price is reaching this new target price. Thus,

we see that price changes can be caused by accumulated small shocks, or one large

shock to the underlying frictionless price. We also see intermittence, and small price

changes consistent with the descriptive statistics. Finally, the figure shows that the

thresholds change across time.

Finally, the numbers from the descriptive statistics shows that the incidence of

11 Dhyne et al. (2011) have such a model, but with asymmetries. To be able to incorporate the

existence of small price changes, they let the threshold be stochastic.

12

price changes is 31% in January, while the average over the other eleven months is

19%. To control for this seasonal effect, we include a January specific parameter,

defined as 0 ≤ y ≤ 1, which is multiplied with the threshold parameters U and L

if the current month is January. This enables the model to lower the thresholds in

the beginning of each year and thereby increase the probability of a price change.

Furthermore, this might also reflect the potential existence of staggered contracts

starting in January and with 12 months duration. 12

This leaves us with the following parameters to be estimated:

Standard deviation of idiosyncratic shocks: σ ε

Upper threshold: U

Lower threshold: L

Inertia upwards: θ up

Inertia downwards: θ down

January specific scalar: y

In our main estimates, we set α =0.0025, which gives an annual inflation equal

to 0.03, close to the average annual inflation rate of the producer price index (PPI)

between the years 2004 and 2009.

3.2 Estimation Method

Given that the empirical model does include the thresholds, the model does not have

an analytical closed form solution.

This again prevent us from using “standard”

regression techniques.

Therefore, we estimate our specification using a Simulated

Method of Moments (SMM). This estimation technique allows some of the parameters

to be estimated, in our case σ 2 ε, U, L, θ up , θ down , and y. Other parameters are

12 Nilsen et al. (**2016**) show, using the same rawdata as used in this paper, a flat price change

hazard with a peak after 12 months.

13

predefined to reduce the computational burden. In short, SMM seeks to minimize

the distance between two sets of moments, the moment vector generated conditional

on a vector of parameters to be estimated β, and the corresponding moment vector

in the actual data, i.e. to find the vector of unknown parameters β that minimizes

the following quadratic form J(β):

J(β) = [Φ A − 1 κ∑

Φ S (β)] ′ W [Φ A − 1 κ

j=1

κ

κ∑

Φ S (β)] (9)

j=1

where Φ A and Φ S (β) denote the vector of m actual moments and the vector of m

simulated counterparts respectively. W denotes an optimal weighting matrix, while

κ denotes the number of panels with the same size as the actual data. The measure

of the distance between two sets of moments, J(β), has a χ 2 distribution with m − l

degrees of freedom, where l is the number of unknown parameters. 13

When searching for values of β that minimize the criterion function, an annealing

cooling algorithm is used. On the basis of starting values for the estimated parameters,

this routine takes random jumps in a predefined parameter space. The routine accepts

worse solutions with a decreasing probability, which ensures that the global optimum

is found. As the final solution is somewhat sensitive to the initial values, we do several

computations with different starting values.

3.3 Selection of moments

The model should explain both inaction and small price changes (defined as less than

5 percent in absolute value and strictly positive) at the same time, the proportion of

13 See the appendix for more details about the Simulated Method of Moments approach.

14

observations within the following intervals are included 14 :

−0.050 ≤ p it − p it−1 < −0.025

−0.025 ≤ p it − p it−1 < −0.005

−0.005 ≤ p it − p it−1 ≤ 0.005 (10)

0.005 < p it − p it−1 ≤ 0.025

0.025 < p it − p it−1 ≤ 0.050

These moments should contribute in identifying all the parameters, especially the

threshold parameters and inertia parameters: Non-zero U and L will cause zeroinflated

price changes, and positive θ up and θ down will cause small price changes.

Larger inertia parameters, θ up and θ down , will make plants smooth their adjustments

over time, which implicates that there will be several consecutive periods of

small price changes.

A consequence of this gradual adjustment is increased serial

correlation in price changes. To identify θ up and θ down , we therefore choose to include

the following correlation coefficient moments:

Corr[p it − p it−1 , p it−1 − p it−2 ] if p it − p it−1 > 0.005 (11)

Corr[p it − p it−1 , p it−1 − p it−2 ] if p it − p it−1 < −0.005 (12)

On the other hand, the threshold parameters U and L will also be affected by these

moments, as larger |U| and larger |L| will lead to more inaction and lower serial

correlation. The asymmetry is such that the moment in (11) should identify θ up ,

while the moment in (12) should identify θ down .

The standard deviation of the shocks to frictionless price, σ ε , is likely to be directly

related to the standard deviation of price changes, sd(p it − p it−1 ).

We therefore

choose to include the standard deviation of price changes as a moment. The standard

14 Our definition of small price changes (less than five percent) is consistent with the assumptions

of Klenow and Kryvtsov (2008) and Eichenbaum et al. (2014).

15

deviation or variance of price changes is also likely to be affected by the friction

parameters: As already pointed out, larger |U| and |L| leads to more inaction and

thereby lower variance of the observed price changes. The aforementioned equation

(5) shows that larger values of θ leads to price changes of more similar size which

again will reduce the variance of price changes. Thus, the standard deviation of price

changes will not only identify σ ε , but it will also contribute to the identification of

U, L, θ up and θ down .

The January specific scalar, y, is supposed to capture the abnormally high adjustment

frequency in the beginning of the year. As a primary identifier, the following

moment is therefore included:

Number of price quotations with |p it − p it−1 | > 0.005 in January

Total number of price quotations in January

In order to replicate the initial high share of price changes observed in the data, the

model will therefore select a value of y such that the thresholds in January are lower

than in the rest of the year.

As previously mentioned, ranked price changes can be a good indicator of lumpy

adjustment behaviour. Four of the ranks presented are therefore used as moments.

More specifically, we include the two first and the two last ranks. These are meant

to be the primary identifiers of the threshold parameters U and L. The ranks are

likely to be affected by σ ε and the inertia parameters as well: More variation in the

frictionless price will cause more variation in the ranks and higher inertia parameters

will bring the ranks closer to each other. Hence, the rank moments will also affect

θ up , θ down and σ ε .

One might think that the use of ranks is just another way of describing the seasonal

effects, and thus that there is not much gain in adding ranks for identification.

When holding for each month the share of the highest ranked price changes (rank 1

observations), the evolvement of these shares mimic very much the pattern of price

16

changes by months. Formal testing shows a corr. coeff. of 0.99 (and z-value = 21.7).

That means that within each plant (and year), the pattern of largest price changes

follows to a large extent the same pattern as the frequencies of price changes seen in

Figure 1. On the other hand, the correlation of the share of the lowest ranked price

changes, and the frequency of price changes by month, is small (0.11) and statistically

insignificant. Thus, there are likely benefits with regard to identification from

including both the Jan-effect, and information about the highest and lowest ranks.

4 Results

Table 1 shows the parameter estimates for the various model specifications by columns.

Standard errors are presented in parentheses. The parameter estimates and standard

errors are presented in the upper part of the table. The corresponding moments of the

various model specifications are presented in the lower part of the table. We estimate

all specifications against the 13 moments already described, i.e. distribution of △p/p,

Jan-effect, serial correlations, st.error of △p/p, and rank-moments. 15

[Table 1 "Parameter Estimates and Moments" about here]

Starting with a broad look at this table, we see that all the estimated parameters

are statistically significant (with the exception of the ones in Column (5)). In Column

(1), we report the results of the full model which include both thresholds and partial

adjustment parameters, and that their magnitude depends on whether prices are

increasing/decreasing relative to previous month.

The U = 0.1398 states that the distance between the frictionless price and the

existing target price, p ∗ it −p # it−1, has to be almost 14 percent before a price-adjustment

process is initiated. Then the actual price changes are decided by the partial adjustment

model. The value of θ up = 0.3698 is to be interpreted as the producer will close

15 All the estimation results are robust to initiating the estimation algorithm from different sets

of starting values. Thus, the parameter estimates reported seem to correspond to global maxima.

17

63 percent (= 1−0.3698) of its desired price change, p # it − p it−1 (= p ∗ it − p it−1 ), in the

same period as it decides to reset the target price and start a price increase process.

The parameter estimates of U = 0.1398 and θ up = 0.3698 together state that the initial

price increase will be at least 0.0881 of the current price. 16 Thus, the new prices

are reached quite quickly. The lower threshold, L = −0.1669 states that the (absolute)

distance between the frictionless price and the existing target price, ∣ ∣p

∗

it − p # ∣

it−1∣,

has to be almost 17 percent before a price decrease process takes place. And the

producer will again quite quickly adjust to the new lower price, seen from the θ down

= 0.4094. 17 The values of θ up = 0.3694 and θ down = 0.3964 state that the producers

put more weight on ( p it − p # )

it than the (pit − p it−1 ) in the out-of-equilibrium costs

function (eq. (4)). Said differently, it seems to be more important to close the gap

relative to the new target price than reducing the implied convex adjustment costs

associated with the period to period adjustment. Furthermore, the findings suggest

that adjustments are faced with two different forms of frictions. Firstly, the effect of

the threshold is that it must be desired to change the price by at least the size of the

threshold before the firm decides to adjust. Secondly, the effect of the inertia is that

the initial price change will be equal to (1−θ) of the target price gap p ∗ it −p it−1 , while

subsequent adjustments will be smaller. The January effect, meant to capture the

fact that the incidence of price changes is higher in January compared to the other

months, states that the thresholds U and L are two thirds in January compared to

the other months. This is consistent with theories focusing on the costs of information

acquisition and processing but might also be consistent with the existence of

staggered contracts with duration of one year.

The model performs relatively well, as seen from the J-statistic in the last row.

16 The initial price increase is found by multiplying U with (1−θ up ): 0.1398×(1−0.3698)≈ 0.0881.

17 The initial price decrease will be at least 0.0986 of the current price, and there will subsequently

be several smaller adjustments downwards until the firm reaches the target price or decides to set

a new one. Thus one may think that price changes of 8.5% and 10.0% (initial price changes for

positive and negative price adjustments, respectively) are definitely not ignorable.

18

This reflects that the empirical moments, reported in the lower part of the table,

are matched quite well with the empirical ones reported in last column of the table,

Column (8). 18

In Column (1) are the magnitude of the parameter estimates for the pairs U and

L, and θ up and θ down , very similar, even though their significance clearly states that

they are statistically different (|L| ̸= U, and θ up ≠ θ down ). Still, when forcing the

parameters within each pair to be the same (in absolute values), reported in Column

(2), the J-statistic states a worse model fit. 19

We have also tested a frictionless model, reported in Column (3). What we see is a

very bad model fit, both measured by the J-statistic, and by comparing the individual

moments of the simulated model with the empirical ones. Thus, a model without

any price adjustment frictions is inferior compared to the two former models. The

frictionless model in Column (3) states that the estimated standard error ̂σ frictionless

ε =

0.0012 and that the frequency of zero price changes is 0.9814. In our model we have

△p ∗ it = p ∗ it − p ∗ it−1 = 0.0025 + ε it where ε it ∼ N(0, σ 2 ε). Note also, in a frictionless

model will p it = p ∗ it, i.e. the new price is equal to the frictionless price. Simple

calculation shows that σ ε = 0.0029 is necessary in a frictionless model to get 80% of

the observations of △p ∗ it (and therefore also △p it ) within the interval [-0.005, 0.005],

the “zero price changes”-interval moment. When our frictionless model gives the best

fit with a frequency of zero price changes of 0.9814, it is due to the other moments

which force the standard error ̂σ frictionless

ε

= 0.0012 for best possible weighted match

with all of the empirical moments. 20

We have also tested specifications where only

18 The J-statistic has a χ 2 distribution where the degrees of freedom are determined by the number

of moments being matched minus the number of parameters estimated. Though compared to similar

studies, the J-statistics reported in Table 1 are low and thus satisfactory, the numbers imply that

all specifications are rejected.

19 Clearly, the difference in J-statistic = 59.7 (=235.5-175.8), df = 2, indicates that this restriction

largely distorts the performance of the model.

20 The J-statistic is 2546.3 for the estimated frictionless model, while it is J = 16805.0 when we

set σ ε = 0.0029. It turns out that when we force the magnitude of σ ε going from the ̂σ ε frictionless =

0.0012 up to σ ε = 0.0029, we get too many small positive price adjustments 〈0.5%, 2.5%] very

quickly. This indicates that the moments describing the distribution is good for identification of the

19

parts of the friction parameters are present.

In Column (4) we include only the

threshold parameters U and L, while in Column (5) we have estimated a pure partial

adjustment model. The results of the latter one is identical to the ones reported in

Column (3). 21 The overall finding based on the results reported in Columns (1)-(5),

is that a full model with both asymmetric price thresholds and asymmetric partial

adjustment process, is the preferred specification.

As an additional way to show the model fit, in addition to the J statistics and the

individual moments compared to their empirical counterpart, we report the whole

distribution of price changes, the share of price changes in given month, and the

mean price-change for all ranks. Going back to Figures 1-3, we have there reported

the simulated moments for the full model (grey bars) reported in Table 1, Column (1).

Starting with the predictions of monthly changes (Figure 1), show that the seasonal

pattern is reasonably good. Especially for the share of price changes, and the ranks

(Figure 2, and Figure 3, respectively) the fit is also very good for the moments not

used when estimating the model. We interpret this as supporting evidence for our

model formulation.

4.1 Robustness checks and discussion

In our underlying frictionless price, modelled as a random walk with a deterministic

trend, the predetermined trend parameter α = 0.0025 which corresponds to an annual

price increase of 3 percent. Two alternative simulations are done where α is set equal

to 0.0016 and 0.0035. These denoted α low and α high , corresponds to an underlying

annual price increase of 2% and 4%. The results of these two alternative trends, and

using the full specification, are reported in Table 1, Columns (6)-(7). The model fit,

variance of the underlying process of the frictionless nominal prices.

21 The similarity of the two sets of results is as expected. The two serial correlation moments

meant to capture inertia are based on the autocovariance of △p ∗ it = p∗ it − p∗ it−1 . When the process

of p ∗ it is a random walk with a drift, the two autocovariances - upwards and downwards - should

both be zero. Thus, when U = L = 0, there is neither anything in the model nor moments that help

identifying the frictionless model and the partial adjustment model from each other. The results

should therefore be the same for the two model specifications, as we do get.

20

measured by the J-statistic, is worse when using the alpha low. Still comparing the

individual coefficients, U and L are both somewhat lower compared to the full model

when α = 0.0025. The largest difference, moderate though, is for the θ up (0.3698 and

0.2979, Column (6) and Column (1) respectively). Turning to the α high results, we

see to overall picture is that the J statistic is very close to the initial one reported

in Column (1). Again the overall picture is that the coefficients are more or less the

same. The conclusion we draw from this robustness check, is that the initial guess of

α = 0.0025 is not too wrong, and if any changes should be made, the α should be

set somewhat higher rather than lower compared to 0.0025. 22

Admittedly, we have

considered to estimate α, instead of pre-determining it as now.

Nevertheless, the

small deviation between the estimates based on different values of α tells us that the

gain from doing this is limited.

[Table 2: Estimation Results by Product Groups - about here]

As already mentioned, earlier research has shown that there is significant heterogeneity

in pricing patterns, both between firms and between products. To address

this in our study, we estimate the model for five different product groups.

These

results are reported in Table 2. What we see is a very good model fit, seen from the

low J statistics, much better than the J statistics found when estimating the model

with the sample of all product types (Table 1, Column 1). This shows that also in

our data there is difference in the pricing patterns between products, and that these

differences should be taken into account. Starting with “Capital goods”, we see much

larger thresholds compared to our previous results. Furthermore, the difference in

thresholds in January and the other months is substantial. The product “Durables”,

“Intermediate goods”, and “Non-durables, food”, seem to have a quite similar pricing

pattern with an upper threshold U in the interval 0.12-0.15, and with the lower

22 A regression model where the dependent variable is log-transformed product prices, and where

a time trend together with product-specific dummies, month-specific dummies and year-specific

dummies are included, gives a time-trend α = 0.0029, which corresponds to a 3.5% annual increase.

21

threshold L significantly larger (in absolute terms). The reduction of the threshold

in January is smaller for these three product groups than for the other two. For the

last group, “Non-durables, non-food”, we also find the two thresholds U and L to be

very similar as for “Durables”, “Intermediate goods”, and “Non-durables, food”. We

notice, however that the two θs for “Non-durables, non-food” are close to zero, indicating

that the gap between the old and new price is closed momentarily. It should

be mentioned, though, that this latter group of goods, “Non-durables, non-food”, is

quite heterogeneous with production of textiles and footwear, pharmaceuticals, and

sports goods. Nevertheless, the results shown in Table 2 state clearly that there is

huge difference in price change patterns across product groups.

[Table 3: Counterfactual Analyses Results - about here]

An interesting question is the importance of the respective price adjustment parameters

in explaining the main characteristics of observed price changes. To shed some

more light over this, we simulate the preferred asymmetric model under exactly the

same circumstances as the estimated model, but setting different price adjustment

cost parameters to zero, and measure the impact on the set of moments used for

identification. Table 3 shows the result of this exercise. The first thing to notice for

all of three sets of results, is the huge increase in the J-statistic, meaning that overall

model fit is much worse when some of the friction parameters are ignored. A more

detailed look, starting with Column (1) where the two θs are set equal to zero, we see

an increased share of zero price changes, no small price changes and therefore also an

increase in the observations outside the [-5%, 5%] interval. The two inertia parameters,

θs, are therefore very important for creating small price changes. Omitting them

from a model would lead to a conclusion that the dynamics of price changes described

as zeros and lumps driven by a fixed costs model. Turning to Column (2) where the

thresholds are ignored, the model with is extremely bad, especially in producing a

22

large enough share of inaction. Thus, both sets of friction parameters, inertia and

thresholds, are important for understanding the dynamics of price changes. 23 , 24

As already mentioned, each plant may produce one or several products. 25

In our

model we have treated each product independent from each other. That means we

have ignored the strategic complementarity in the adjustments of multiple products

and assume that the various products are sufficiently differentiated in order to abstract

from substitution within the plant’s portfolio of products. 26

We have analysed the

various components of the prices in the dataset by using a multi-level mixed effects

linear regression model (see Baltagi et al., 2001).

(Log-) prices are regressed on

time effects (a linear time trend, month- and year dummies) with random intercepts

at both the plant and the product-within-plant level such that products are nested

within plants. The analysis shows that the price variation is mainly driven by variance

in plant specific effects, and to a smaller degree variance in product specific effects.

Admittedly, this interrelation between the products within each plant is ignored in

our analysis. One modification to control for this effect would be to introduce two

variance components in our idiosyncratic shocks σ 2 ε, such that σ 2 ε = σ 2 p + σ 2 u where

σ 2 p denotes variance of product-specific shocks, and σ 2 u denotes idiosyncratic product

shocks. One could also consider to model frictions at the level of the plant, not at

the level of products. Our simplification of the shock process is likely to overestimate

the importance of the threshold parameters, since they will pick up the effect of both

23 A more complicated shock process of the frictionless prices has been considered. Indeed, ignored

persistency in the shock process is likely to bias the θ parameters. However, our random walk is

of course highly persistent. Furthermore, our estimation results show that the distance between

existing prices and target prices is closed quit quickly. Thus it is not clear, though, what one would

gain from a modification of the shock process of the frictionless prices.

24 We have also estimated a model where we look at annual data, i.e. the price changes from

June one year to June the subsequent year, and of course estimated with a different moment vector.

Unsurprisingly, these results indicate much smaller thresholds, asymmetric but still statistically significant.

The downward inertia parameter θ down is significant, while θ up = 0. Thus, time aggregation

blurs the price changing picture compared to using a model that are able to take advantage of the

monthly frequency. These results are not reported, but available from the authors on request.

25 The mean number of products per producer is 5, while the maximum is 20.

26 See also Woodford (2003) and Gertler and Leahy (2008) for discussions about strategic complementarity

and “real rigidity” but then in relation to firms’ competitors.

23

price change costs, and to some degree shocks that are common to all products of a

producer. The sketched changes would be interesting extensions of our model. Again

it is a matter of identification and tractability of the model. Note however, there

are no other variables than prices in the data with monthly frequency. Neither is

there anything about product specific quantities and costs, only plant level (annual)

information. Thus, we leave these potential extensions for a later paper.

Our estimates show that price adjustment frictions are important for understanding

the intermittent price adjustment pattern seen in the data, and that ignoring

frictions bias our results. Furthermore, our findings of both thresholds and inertia together,

indicate that different forms of rigidities exist in the data, which is only partly

consistent with the assumptions of macroeconomic pricing models. Some menu cost

models include thresholds with (S, s) pricing rules, while others incorporate inertia

by assuming partial adjustments.

However, none of the models incorporates both

thresholds and inertia in price setting.

In general, macro models therefore fail to

include all the evidence provided in this paper.

Moreover, our findings imply the

occurrence of both large and small price changes. While the threshold parameters enable

inaction, the inertia parameters implicate a large initial price change followed by

smaller adjustments. Accordingly, the results imply that our model is able to account

for periods of inaction, as well as both large and small price changes. In contrast,

models such as in Golosov and Lucas (2007) and Gertler and Leahy (2008), explain

patterns of inaction followed by large price changes by assuming thresholds, but these

models seem to neglect small price adjustments. Our findings indicate that prices are

similarly flexible upwards and downwards. 27

The seasonal effect, picked up by our

January parameter, y, may come from uneven staggering of nominal contracts and

could have implications for the effectiveness of monetary policy interventions taking

27 The literature is not conclusive when it comes to whether nominal price rigidities are symmetric

or not. For studies finding asymmetry, see for instance Dias et al. (2015); Laxton et al. (1995, 1999);

Dolado et al. (2005); Dobrynskaya (2008).

24

place in different months of the year (for related findings, see Olivei and Tenreyro

(2007)).

5 Concluding Remarks

In this paper we specify and estimate a model that describes production plants’ price

adjustment behaviour. The model includes thresholds which are lower in January

than in the other months, together with a quadratic loss–function associated with

the distance from a target price. The simplistic reduced form model is tested on a

sample based on repeated monthly plant- and product-specific survey data from the

Norwegian manufacturing industry. The model is meant to reproduce the following

features of the data. First, plants adjust prices infrequently as only 20 percent of

the price observations change from one month to another. Secondly, the is a seasonal

pattern in the incidence of price changes, with most price changes taking place in

January. Thirdly, there is also a lot small price changes.

The simulated method-of-moment estimates reveal thresholds that are such that

prices are only changed if the deviation from the target price to the underlying frictionless

price is larger than approximately 15 percent. However, if the shocks are such

that the prices should be changed, the gap between the current price and the new

target price is reduced quite quickly and only 10 percent of the initial gap exists after

three months. There are statistically significant cost differences whether the prices

should move upwards or downwards. However, the magnitude of these differences

are very moderate. Furthermore, the January-effect, indicating that the thresholds

are only two third this month, is consistent with theories focusing on the costs of

information acquisition and processing, and/or on staggered contracts.

Several checks are applied to test the robustness of the model and findings. First,

the preferred specification outperforms a frictionless model, or models with only some

parts of the price adjustment friction parameters present. A counterfactual analysis

25

where some of the friction parameters are set equal to zero, shows that the moment

fit becomes much worse compared to the preferred model specification. Furthermore,

the model seems to be fairly robust to changes in the underlying deterministic trend,

as our approximation of the trend gives better fit than alternative approximations.

While our evidence implies both large and small price changes, many model contributions

in the literature are only able to account for one of these two characteristics.

Having said this, there are a few of the models assuming thresholds in the price setting

are able to explain small price changes. These models assume either stochastic

thresholds or economies of scope in price setting, and represent an increasingly sophisticated

group of pricing models in which more micro evidence is incorporated. However,

our model is rather simple and transparent, and computationally easy.

There are a set of issues we have not addressed and that need to be explored

in future work. The model is admittedly a reduced form model. A more structural

model would be more informative. This would call for a full dynamic- specification

and optimisation. However, with the current dataset there is no information about

quantities, even though annual revenues and costs are available at plant level. Furthermore,

the only information available at product level with monthly frequency is

prices themselves. Thus, a structural model would partly require non-verifiable assumptions

about inputs and outputs. Still, our findings strongly indicate that such

a model needs to include both convex and non-convex price adjustment costs. Our

goods-specific estimates also point in the direction for taking into account and control

for product (and plant) specific heterogeneity. Note however, the mixed frequency of

price information, and other plant-or firm-specific information (monthly versus annually),

give some econometric challenges. Still, the evidence provided in this paper,

based on a simple and transparent simulation model, shows the importance and potential

fruitfulness of using model formulations and estimation techniques that able

to take into account the non-convexities in price adjustment costs function.

26

References

Alvarez, F. E., Lippi, F., and Paciello, L. (2011). Optimal price setting with observation

and menu costs. The Quarterly Journal of Economics, 126(4):1909–1960.

Álvarez, L. J., Dhyne, E., Hoeberichts, M., Kwapil, C., Bihan, H. L., Lünnemann,

P., Martins, F., Sabbatini, R., Stahl, H., Vermeulen, P., and Vilmunen, J. (2006).

Sticky prices in the Euro area: A summary of new micro-evidence. Journal of the

European Economic Association, 4(2-3):575–584.

Baltagi, B. H., Song, S. H., and Jung, B. C. (2001). The unbalanced nested error

component regression model. Journal of Econometrics, 101:357–381.

Barros, R., Carvalho, C., Bonomo, M., and Matos, S. (2009).

Price setting in a

variable macroeconomic environment: Evidence from Brazilian CPI. Mimeo, pages

1–45.

Bhattarai, S. and Schoenle, R. (2014). Multiproduct firms and price-setting: Theory

and evidence from U.S. producer prices. Journal of Monetary Economics, 66:178–

192.

Bloom, N. (2009). The impact of uncertainty shocks. Econometrica, 77(3):623–685.

Caballero, R. J. and Engel, E. M. (1993). Microeconomic rigidities and aggregate

price dynamics. European Economic Review, 37(4):697–711.

Dhyne, E., Fuss, C., Pesaran, M. H., and Sevestre, P. (2011). Lumpy price adjustments:

A microeconometric analysis. Journal of Business & Economic Statistics,

29(4):529–540.

Dias, D. A., Marques, C. R., Martins, F., and Santos Silva, J. M. C. (2015). Understanding

price stickiness: Firm-level evidence on price adjustment lags and their

asymmetries. Oxford Bulletin of Economics and Statistics, 77(5):701–718.

27

Dobrynskaya, V. V. (2008). Asymmetric price rigidity and the optimal interest rate

defense of the exchange rate: Some evidence for the US. Journal of Policy Modeling,

30(5):713–724.

Dolado, J. J., Maria-Dolores, R., and Naveira, M. (2005). Are monetary-policy reaction

functions asymmetric? the role of nonlinearity in the phillips curve. European

Economic Review, 49(2):485–503.

Doms, M. and Dunne, T. (1998).

Capital adjustment patterns in manufacturing

plants. Review of Economic Dynamics, 1(2):409–429.

Eichenbaum, M., Jaimovich, N., Rebelo, S., and Smith, J. (2014). How frequent are

small price changes? American Economic Journal: Macroeconomics, 6(2):137–155.

Fougère, D., Le Bihan, H., and Sevestre, P. (2007). Heterogeneity in consumer price

stickiness. Journal of Business & Economic Statistics, 25(3):247–264.

Gertler, M. and Leahy, J. (2008). A Phillips curve with an Ss foundation. Journal of

Political Economy, 116(3):533–572.

Golosov, M. and Lucas, R. E. (2007). Menu costs and Phillips curves. Journal of

Political Economy, 115:171–199.

Honoré, B. E., Kaufmann, D., and Lein, S. (2012).

Asymmetries in price-setting

behavior: New microeconometric evidence from Switzerland. Journal of Money,

Credit and Banking, 44(2):211–236.

Klenow, P. and Malin, B. (2011). Microeconomic evidence on price-setting. Handbook

of Monetary Economics, 3:231–284.

Klenow, P. J. and Kryvtsov, O. (2008).

State-dependent or time-dependent pricing:

Does it matter for recent U.S. inflation?

Quarterly Journal of Economics,

123(3):863–903.

28

Laxton, D., Meredith, G., and Rose, D. (1995).

Asymmetric effects of economic

activity on inflation: Evidence and policy implications. Staff papers, 42(2):344–

374.

Laxton, D., Rose, D., and Tambakis, D. (1999). The U.S. Phillips curve: The case

for asymmetry. Journal of Economic Dynamics and Control, 23(9-10):1459–1485.

Lee, B. S. and Ingram, B. F. (1991). Simulation estimation of time-series models.

Journal of Econometrics, 47(2-3):197–205.

Lewis, M. S. (2011).

Asymmetric price adjustment and consumer search: An examination

of the retail gasoline market. Journal of Economics and Management

Strategy, 20(2):409–449.

Loupias, C. and Sevestre, P. (2012). Costs, demand, and producer price changes.

Review of Economics and Statistics, 95(1):315–327.

Loy, J. P., Weiss, C. R., and Glauben, T. (**2016**). Asymmetric cost pass-through?

Empirical evidence on the role of market power, search and menu costs. Journal of

Economic Behavior and Organization, 123:184–192.

Maćkowiak, B. and Wiederholt, M. (2009).

Optimal sticky prices under rational

inattention. The American Economic Review, 99(3):769–803.

Mankiw, N. G. and Reis, R. (2002). Sticky information versus sticky prices: A proposal

to replace the new keynesian phillips curve. Quarterly Journal of Economics,

117(4):1295–1328.

McFadden, D. (1989). A method of simulated moments for estimation of discrete

response models without numerical integration. Econometrica, 57(5):995–1026.

Midrigan, V. (2011).

Menu costs, multiproduct firms, and aggregate fluctuations.

Econometrica, 79(4):1139–1180.

29

Nakamura, E. and Steinsson, J. (2008). Five facts about prices: A reevaluation of

menu cost models. Quarterly Journal of Economics, 123(4):1415–1464.

Nilsen, . A., Bratlie, J., and Pettersen, P. (**2016**). Time-dependency in producers

price adjustments: Evidence from micro panel data. mimeo, Norwegian School of

Economics.

Nilsen, Ø. A. and Schiantarelli, F. (2003). Zeros and lumps in investment: Empirical

evidence on irreversibilities and nonconvexities. Review of Economics and Statistics,

85(4):1021–1037.

Olivei, G. and Tenreyro, S. (2007). The timing of monetary policy shocks. American

Economic Review, 97:636–663.

Pakes, A. and Pollard, D. (1989). Simulation and the asymptotics of optimization

estimators. Econometrica, 57(5):1027–1057.

Peltzman, S. (2000). Prices rise faster than they fall. Journal of Political Economy,

108(3):466.

Ratfai, A. (2006). Linking individual and aggregate price changes. Journal of Money,

Credit and Banking, 38(8):2199–2224.

Rotemberg, J. (1982). Sticky prices in the United States. Journal of Political Economy,

90(6):1187–1211.

Sheshinski, E. and Weiss, Y. (1977). Inflation and costs of price adjustment. The

Review of Economic Studies, 44(2):287–303.

SSB (2015). Commodity price index for the industrial sector.

http://www.ssb.no/en/vppi.

Taylor, J. B. (1980). Aggregate dynamics and staggered contracts. The Journal of

Political Economy, pages 1–23.

30

Taylor, J. B. (1999). Staggered price and wage setting in macroeconomics. Handbook

of macroeconomics, 1:1009–1050.

Varejão, J. and Portugal, P. (2007). Employment dynamics and the structure of labor

adjustment costs. Journal of Labor Economics, 25(1):137–165.

Vermeulen, P., Dias, D. A., Dossche, M., Gautier, E., Hernando, I., Sabbatini, R., and

Stahl, H. (2012). Price setting in the Euro area: Some stylized facts from individual

producer price data. Journal of Money, Credit and Banking, 44(8):1631–1650.

Woodford, M. (2003). Optimal interest-rate smoothing. Review of Economic Studies,

70:861–886.

Wulfsberg, F. (**2016**). Inflation and price adjustments: Micro evidence from norwegian

consumer prices 1975-2004. American Economic Journal: Macroeconomics,

8(3):175–194.

Xia, T. and Li, X. (2010). Consumption inertia and asymmetric price transmission.

Journal of Agricultural and Resource Economics, 35(2):209–227.

Yang, H. and Ye, L. (2008). Search with learning: Understanding asymmetric price

adjustments. RAND Journal of Economics, 39(2):547–564.

Zbaracki, M. J., Ritson, M., Levy, D., Dutta, S., and Bergen, M. (2004). Managerial

and customer costs of price adjustment: Direct evidence from industrial markets.

Review of Economics and Statistics, 86(October 2015):514–533.

31

Figures

Figure 1: Empirical and Simulated Distribution of Monthly Price Changes

32

Figure 2: Empirical and Simulated Ranks

33

Figure 3: Average Share of Price Changes, by month - empirical and simulated

34

Figure 4: Illustration of the Price Changing Process

35

Tables

Specifications: (1) (2) (3) (4) (5) (6) (7) (8)

σ ε 0.0417 0.0397 0.0012 0.0161 0.0012 0.0410 0.0412 -

(0.0007) (0.0007) (1.98E-05) (0.0010) (1.98E-05) (0.0006) (0.0007) -

U 0.1398 0.1439 - 0.0440 - 0.1182 0.1281 -

(0.0034) (0.0026) - (0.0004) - (0.0019) (0.0029) -

L -0.1669 -0.1439 - -0.1289 - -0.1799 -0.1624 -

(0.0035) (0.0026) - (0.0252) - (0.0033) (0.0044) -

θ up 0.3698 0.3787 - - 0.0000 0.2979 0.3580 -

(0.0030) (0.0027) - - (0.0000) (0.0066) (0.0050) -

θ down 0.4094 0.3787 - - 0.0000 0.3720 0.3861 -

(0.0099) (0.0027) - - (0.0000) (0.0072) (0.0097) -

y 0.7120 0.7319 - 0.6346 - 0.6809 0.6928 -

(0.0166) (0.0161) - (0.0221) - (0.0148) (0.0167) -

Moments

[−5.0%, −2.5%〉 0.0188 0.0226 0.0000 0.0000 0.0000 0.0186 0.0193 0.0171

[−2.5%, −0.5%〉 0.0410 0.0387 0.0000 0.0000 0.0000 0.0384 0.0370 0.0331

[−0.5%, 0.5%] 0.7786 0.7745 0.9814 0.9387 0.9814 0.7900 0.7739 0.8000

〈0.5%, 2.5%] 0.0592 0.0648 0.0186 0.0000 0.0186 0.0504 0.0578 0.0470

〈2.5%, 5.0%] 0.0379 0.0371 0.0000 0.0278 0.0000 0.0349 0.0430 0.0362

Chgs in Jan 0.3325 0.3315 0.0185 0.2001 0.0185 0.3348 0.3478 0.3146

Serial corr (up) -0.3642 -0.3086 0.0106 -0.0590 0.0106 -0.4879 -0.4454 -0.1358

Serial corr(down) -0.2636 -0.3211 0.0000 0.0000 0.0000 -0.2916 -0.2909 -0.3851

sd(p it − p it−1) 0.0275 0.0267 0.0012 0.0142 0.0012 0.0283 0.0276 0.0413

Rank 12 -0.0263 -0.0260 0.0005 -0.0053 0.0005 -0.0302 -0.0257 -0.0286

Rank 11 -0.0108 -0.0102 0.0012 0.0000 0.0012 -0.0113 -0.0100 -0.0097

Rank 2 0.0218 0.0211 0.0039 0.0103 0.0039 0.0202 0.0237 0.0213

Rank 1 0.0429 0.0422 0.0045 0.0271 0.0045 0.0423 0.0460 0.0526

J: 175.8 235.5 2546.3 819.2 2546.3 347.7 188.8 -

The column numbers represents the following specifications: (1): Full model, (2): Full model with symmetric

friction parameters, (3): Frictionless, (4): No inertia parameters, (5): No threshold parameters, (6): Full model

with α = 0.0017, (7): Full model with α = 0.0033, (8): Empirical moments.

Table 1: Estimation Results and Empirical Counterparts

36

Specifications:

(1) (2) (3) (4) (5)

Capital

goods

Durables

Intermediate

goods

Non-durables,

food

Non-durables,

non-food

σ e 0.0297 0.0413 0.0512 0.0484 0.0188

(0.0018) (0.0035) (0.0015) (0.0020) (0.0044)

U 0.2820 0.1193 0.1561 0.1527 0.1700

(0.0608 ) (0.0058) (0.0038) (0.0042) (0.0149)

L -0.2647 -0.2397 -0.2119 -0.1621 -0.2061

(0.0699) (0.0314) (0.0081) (0.0056) (0.1609)

θ up 0.3050 0.2994 0.3555 0.4068 0.0262

(0.0226) (0.0254) (0.0039) (0.0177) (0.0024)

θ down 0.2520 0.4433 0.4778 0.3755 0.0026

(0.0337) (0.0201) (0.0138) (0.0171) (0.1714)

y 0.4184 0.8198 0.7116 0.8834 0.6161

(0.0861) (0.0870) (0.0258) (0.0389) (0.0645)

Moments

[−5.0%, −2.5%〉 0.0053 0.0107 0.0218 0.0236 0.0000

[−2.5%, −0.5%〉 0.0108 0.0232 0.0432 0.0516 0.0000

[−0.5%, 0.5%] 0.9133 0.8257 0.7498 0.7305 0.9814

〈0.5%, 2.5%] 0.0240 0.0414 0.0621 0.0791 0.0005

〈2.5%, 5.0%] 0.0172 0.0361 0.0375 0.0361 0.0000

Chgs in Jan 0.3399 0.2300 0.3539 0.3071 0.1776

Serial corr(up) -0.3581 -0.5031 -0.3398 -0.2587 0.0000

Serial corr(down) -0.4010 -0.1192 -0.1057 -0.2772 0.0000

sd(p it − p it−1 ) 0.0222 0.0270 0.0325 0.0319 0.0200

Rank 12 -0.0108 -0.0214 -0.0314 -0.0360 -0.0030

Rank 11 -0.0027 -0.0089 -0.0147 -0.0143 0.0000

Rank 2 0.0103 0.0192 0.0257 0.0258 0.0029

Rank1 0.0329 0.0395 0.0504 0.0481 0.0317

N 243 134 807 359 133

J 93.7 75.9 85.7 40.5 75.7

Note: standard errors in parentheses, N denotes number of products, J denotes the

criterion value.

Table 2: Estimation Results by Product Groups

37

(1) (2) (3)

Specifications: No inertia No thresholds

No friction

parameters

σ ε 0.0417 0.0417 0.0417

(0.0005) (0.0009) (0.0001)

U 0.1398 - -

(0.0033) - -

L -0.1669 - -

(0.0064) - -

θ up - 0.3698 -

- (0.0199) -

θ down - 0.4094 -

- (0.0185) -

y 0.7120 - -

(0.0143) - -

[−5%, −2.5%) 0.0000 0.1327 0.1508

[−2.5%, −0.5%) 0.0000 0.2406 0.1733

[−0.5%, 0.5%] 0.9396 0.1434 0.0958

(0.5%, 2.5%] 0.0000 0.2484 0.1811

(2.5%, 5.0%] 0.0000 0.1625 0.1678

Chgs in Jan 0.1891 0.8571 0.9046

Serial corr (up) -0.0193 0.2326 0.0006

Serial corr (down) 0.0000 0.2344 0.0008

sd(p it − p it−1 ) 0.0413 0.0277 0.0417

Rank 12 -0.0421 -0.0388 -0.0653

Rank 11 -0.0025 -0.0264 -0.0437

Rank 2 0.0175 0.0342 0.0523

Rank 1 0.0659 0.0455 0.0705

J 3 213.8 28 850.6 27 037.6

Table 3: Counterfactual Analysis

38

A

Appendices

Coef. Std. Error Z-value 95% Conf. Interval

[−5%, −2.5%) 0.0171 0.0009 19.0000 0.0154 0.0189

[−2.5%, −0.5%) 0.0331 0.0017 19.0400 0.0297 0.0365

[−0.5%, 0.5%] 0.8000 0.0062 128.6700 0.7878 0.8122

(0.5%, 2.5%] 0.0470 0.0022 21.8100 0.0428 0.0512

(2.5%, 5.0%] 0.0362 0.0010 36.0400 0.0343 0.0382

Chgs in Jan 0.3146 0.0094 33.3700 0.2961 0.3330

Serial corr (up) -0.1358 0.0284 -4.7700 -0.1915 -0.0800

Serial corr (down) -0.3851 0.1341 -2.8700 -0.6479 -0.1222

sd(p it − p it−1 ) 0.0413 0.0027 15.0200 0.0359 0.0466

Rank 12 -0.0286 0.0012 -22.9100 -0.0310 -0.0261

Rank 11 -0.0097 0.0006 -16.7200 -0.0108 -0.0085

Rank 2 0.0213 0.0007 28.6900 0.0199 0.0228

Rank 1 0.0526 0.0014 38.0200 0.0499 0.0553

Note: The first five rows represents the total shares of observations within the

given intervals and the following row represents the share of price changes in

January.

Table A1: Bootstrapped Moments with Std. Errors

39

A.1 A Simple Model for Frictionless Price

Assume a Cobb-Douglas production technology with a flexible input factor, K. The

costs of this factor is exogenous to the plant and denoted r. Assume also that the

plants have some market power and demand is given by an iso-elastic function. Also

assume that goods are sufficiently differentiated in order to abstract from substitution

within a multi-product-firm’s portfolio of products. We abstract from sub-indices for

the plant, product and time for notational convenience. Then production is determined

by Q S (K) = A · K a where 0 < a < 1 and the iso-elastic demand function is

given by Q D (P ) = B · (P/P C) −ɛ

where ɛ > 1. The price of a plant’s product is given

by P , and P C denotes the general price level in the industry. The price level P C is

exogenous to the plant which implies that we employ a partial equilibrium model.

Abstracting from inventory, profit for a single product is then given by

π(A, B, P C , r) = P · B · (

P/P C) ( )

−ɛ B 1/a

− r · · (

P/P C) −ɛ/a

,

A

where A captures supply shocks, and B captures demand shocks. With these assumptions

the first order derivative of profit π(.) with respect to price P can be expressed

as follows;

[

P ∗ ɛ

=

a(ɛ − 1) B 1−a

a

A

− 1 a

] a

ɛ(1−γ)

} {{ }

(1)

a

×r

ɛ(1−γ)

} {{ } × (P C ) 1−a

ɛ(1−γ)

, where γ = a(1 − 1 } {{ }

ɛ )

(2)

(3)

This expression is a non-linear function of the state of supply A, the state of

demand B, the input costs r, and the general price in the industry P C . Given that

a < 1, we see that a positive supply shock, A ↑, will implicate a lower price, as

expression (1) will get a lower value. This could, for example, be that the producer

obtains better technology that increases productivity. We also see that a positive

demand shock, B ↑, will implicate a higher price, as the net effect on expression (1)

will be positive. Furthermore, if producers are faced with a positive cost shock, r ↑,

the frictionless equilibrium price will increase, as expression (2) will be more positive.

Note however that the degree of pass-through, the share of the cost increase that

will be borne by costumers, depends on the parameter values. Higher competitor’s

prices, P C ↑, induce also a price increase. In our model setup presented in the main

text, part of this latter effect - that the general price level increases - is picked up by

the trend parameter α. The remaining supply-, demand-, and cost- shocks, together

with the non-deterministic part of the competitor’s prices, will all be picked up by

the idiosyncratic shocks, ε it , in the model presented in the main text. Finally, we see

that the marginal effects of the various shocks affect the price differently.

40

A.2 Simulated Method of Moments

In the Simulated Method of Moments (SMM) approach, κ simulated datasets are

generated for N panels and 96 + T time periods. N and T are set equal to the

number of panels and time periods in the empirical data. 28 In order to limit the

impact of initial conditions, the first 96 time periods (8 years) are discarded when

calculating the simulated moments, leaving only T time periods. 29

If we let the vector of l unknown parameters be denoted by the vector β, the

optimal vector of unknown parameters, ˆβ, is given by:

ˆβ = argmin [Φ A − 1

β κ

κ∑

Φ S (β)] ′ W [Φ A − 1 κ∑

Φ S (β)] (13)

j=1

κ

j=1

where W denotes the optimal weighting matrix, and Φ A and Φ S (β) denote the vector

of m actual moments and the vector of m simulated counterparts respectively. The

weighting matrix W is given by the inverse of the variance-covariance matrix of [Φ A −

1 ∑ κj=1

Φ S (β)], which is best estimated using the following matrix (see Lee and Ingram

κ

(1991)):

W = [(1 + 1 κ )Ω]−1 (14)

Here, Ω denotes the variance-covariance matrix of the empirical moments, Φ A . Ω is

obtained by a block bootstrap with replacement on empirical data. An implication

of using this weighting matrix is that moments with a large variation are given less

weight than moments with a small variation.

The standard errors of the parameters are given by the square roots of the diagonals

of the variance-covariance matrix for ˆβ, which is given by:

Here, ∂ΦS ( ˆβ)

∂β

Q s (W ) = (1 + 1 ( ˆβ)

′

κ )[∂ΦS W ∂ΦS ( ˆβ) ] −1 (15)

∂β ∂β

is the Jacobian m × l matrix of the moment vector with respect to the

parameter vector β evaluated at ˆβ. In lack of an analytical solution of the components

of this matrix, numerical derivatives are used. More specifically, we use the symmetric

difference quotient which is given by:

f ′ (x) ≈

f(x + h) − f(x − h)

2h

(16)

In expression (16), x denotes the components of ˆβ, f(x) denotes the components of

Φ S ( ˆβ) and h is a small positive number. A problem with this approach is that the

approximate depends on the size of h. We therefore follow Bloom (2009) and calculate

four values of the numerical derivative with steps of 0.1%, 1%, 2.5% and 5% from ˆβ,

and use the median value of these numerical derivatives.

28 See for instance McFadden (1989); Pakes and Pollard (1989) for more details regarding the

approach.

29 In our estimation we use κ = 10, and have N = 1676 and T = 60.

41

Issued in the series Discussion Papers 2015

2015

01/15 January, Antonio Mele, Krisztina Molnár, and Sergio Santoro, “On the perils

of stabilizing prices when agents are learning”.

02/15 March, Liam Brunt, “Weather shocks and English wheat yields, 1690-1871”.

03/15 March, Kjetil Bjorvatn, Alexander W. Cappelen, Linda Helgesson Sekei, Erik

Ø. Sørensen, and Bertil Tungodden, “Teaching through television:

Experimental evidence on entrepreneurship education in Tanzania”.

04/15 March, Kurt R. Brekke, Chiara Canta, Odd Rune Straume, “Reference pricing

with endogenous generic entry”.

05/15 March, Richard Gilbert and Eirik Gaard Kristiansen, “Licensing and

Innovation with Imperfect Contract Enforcement”.

06/15 March, Liam Brunt and Edmund Cannon, “Variations in the price and quality

of English grain, 1750-1914: quantitative evidence and empirical implications”.

07/15 April, Jari Ojala and Stig Tenold, “Sharing Mare Nostrum: An analysis of

Mediterranean maritime history articles in English-language journals”.

08/15 April, Bjørn L. Basberg, “Keynes, Trouton and the Hector Whaling Company.

A personal and professional relationship”.

09/15 April, Nils G. May and Øivind A. Nilsen, “The Local Economic Impact of

Wind Power Deployment”.

10/15 May, Ragnhild Balsvik and Stefanie Haller, “Ownership change and its

implications for the match between the plant and its workers”.

11/15 June, Kurt R. Brekke, Chiara Canta, Odd Rune Straume, “Does Reference

Pricing Drive Out Generic Competition in Pharmaceutical Markets? Evidence

from a Policy Reform”.

12/15 June, Kurt R. Brekke, Tor Helge Holmås, Karin Monstad, and Odd Rune

Straume, “Socioeconomic Status and Physicians’Treatment Decisions”.

13/15 June, Bjørn L. Basberg, “Commercial and Economic Aspects of Antarctic

Exploration ‐ From the Earliest Discoveries into the 19th Century”.

14/15 June, Astrid Kunze and Amalia R. Miller, “Women Helping Women?

Evidence from Private Sector Data on Workplace Hierarchies”

15/15 July, Kurt R. Brekke, Tor Helge Holmås, Karin Monstad, Odd Rune Straume,

«Do Treatment Decisions Depend on Physicians’ Financial Incentives?”

16/15 July, Ola Honningdal Grytten, “Norwegian GDP by industry 1830-1930”.

17/15 August, Alexander W. Cappelen, Roland I. Luttens, Erik Ø. Sørensen, and

Bertil Tungodden, «Fairness in bankruptcy situations: an experimental study».

18/15 August, Ingvild Almås, Alexander W. Cappelen, Erik Ø. Sørensen, and

Bertil Tungodden, “Fairness and the Development of Inequality Acceptance”.

19/15 August, Alexander W. Cappelen, Tom Eichele,Kenneth Hugdah, Karsten

Specht, Erik Ø. Sørensen, and Bertil Tungodden, “Equity theory and fair

inequality: a neuroconomic study”.

20/15 August, Frank Jensen and Linda Nøstbakken, «A Corporate-Crime

Perspective on Fisheries: Liability Rules and Non-Compliance”.

21/15 August, Itziar Lazkano and Linda Nøstbakken, “Quota Enforcement and

Capital Investment in Natural Resource Industries”.

22/15 October, Ole-Petter Moe Hansen and Stefan Legge, “Trading off Welfare and

Immigration in Europe”.

23/15 October, Pedro Carneiro, Italo Lopez Garcia, Kjell G. Salvanes, and Emma

Tominey, “Intergenerational Mobility and the Timing of Parental Income”.

24/15 October, David Figlio, Krzysztof Karbownik, and Kjell G. Salvanes,

“Education Research and Administrative Data”.

25/15 October, Ingvild Almås, Alexander W. Cappelen, Kjell G. Salvanes, Erik Ø.

Sørensen, and Bertil Tungodden: «Fairness and family background».

26/15 November, Lars Ivar Oppedal Berge, Kjetil Bjorvatn, Simon Galle, Edward

Miguel, Daniel Posner, Bertil Tungodden, and Kelly Zhang “How Strong are

Ethnic Preferences?”.

27/15 November, Agnar Sandmo, “The Public Economics of Climate Change”.

28/15 November, Aline Bütikofer and Kjell G. Salvanes, “Disease Control and

Inequality Reduction: Evidence from a Tuberculosis Testing and Vaccination

Campaign”.

29/15 December, Aline Bütikofer, Katrine V. Løken and Kjell G. Salvanes, “Long-

Term Consequences of Access to Well-child Visits”

30/15 December, Roger Bivand, “Revisiting the Boston data set (Harrison and

Rubinfeld, 1978): a case study in the challenges of system articulation”.

**2016**

01/16 January, Ingvild Almås and Anders Kjelsrud, “Pro-poor price trends and

inequality|the case of India”.

02/16 January, Tuomas Pekkarinen, Kjell G. Salvanes, and Matti Sarvimäki, «The

evolution of social mobility: Norway over the 20th century”

03/16 February, Sandra E. Black, Paul J. Devereux, and Kjell G. Salvanes,

“Healthy(?), Wealthy, and Wise. Birth Order and Adult Health”.

04/16 March, Hiroshi Aiura, “The effect of cross-border healthcare on quality, public

health insurance, and income redistribution"

05/16 March, Jan Tore Klovland, “Shipping in dire straits: New evidence on trends

and cycles in coal freights from Britain, 1919-1939”

06/16 April, Branko Bošković and Linda Nøstbakken, “The Cost of Endangered

Species Protection: Evidence from Auctions for Natural Resources”

07/16 April, Cheti Nicoletti, Kjell G. Salvanes, and Emma Tominey, “The Family

Peer Effect on Mothers’ Labour Supply”

08/16 April, Hugh Gravelle and Fred Schroyen, “Optimal hospital payment rules

under rationing by random waiting”

09/16 May, Branko Bošković and Linda Nøstbakken “Do land markets anticipate

regulatory change? Evidence from Canadian conservation policy”

10/16 May, Ingvild Almås, Eleonora Freddi, and Øystein Thøgersen, “Saving and

Bequest in China: An Analysis of Intergenerational Exchange”

11/16 May, Itziar Lazkano, Linda Nøstbakken, and Martino Pelli, “From Fossil

Fuels to Renewables: The Role of Electricity Storage”

12/16 June, Jari Ojala and Stig Tenold, «Maritime trade and merchant shipping: The

shipping/trade-ratio from the 1870s until today”

13/16 September, Elias Braunfels, “Further Unbundling Institutions”

14/16 September, Alexander W. Cappelen, Cornelius Cappelen, and Bertil

Tungodden», “Second-best fairness: false positives and false negatives in

distributive choices”

15/16 October, Aline Bütikofer, Eirin Mølland, and Kjell G. Salvanes, “Childhood

Nutrition and Labor Market Outcomes: Evidence from a School Breakfast

Program”

16/16 October, Itziar Lazkano and Linh Pham, “Do Fossil-Fuel Taxes Promote

Innovation in Renewable Electricity Generation?

17/16 October, Kristiina Huttunen, Jarle Møen, and Kjell G. Salvanes, «Job Loss and

Regional Mobility».

18/16 November, Ingvild Almås, Alexander Cappelen, Bertil Tungodden,

«Cutthroat capitalism versus cuddly socialism: Are Americans more

meritocratic and efficiency-seeking than Scandinavians?”

19/16 December, Gernot Doppelhofer, Ole-Petter Moe Hansen, Melwyn Weeks,

“Determinants of long-term economic growth redux: A Measurement Error

Model Averaging (MEMA) approach “

20/16 December, Leroy Andersland and Øivind A. Nilsen, “Households’ responses

to price changes of formal childcare”

21/16 December, Wilko Letterie and Øivind A. Nilsen, “Price Changes - Stickiness

and Internal Coordination in Multiproduct Firms”

22/16 December, Øivind A. Nilsen and Magne Vange, “Intermittent Price Changes

in Production Plants: Empirical Evidence using Monthly Data”

Norges

Handelshøyskole

Norwegian School of Economics

NHH

Helleveien 30

NO-5045 Bergen

Norway

Tlf/Tel: +47 55 95 90 00

Faks/Fax: +47 55 95 91 00

nhh.postmottak@nhh.no

www.nhh.no