• No results found

Emperical behavioral model of inflation expectations

N/A
N/A
Protected

Academic year: 2021

Share "Emperical behavioral model of inflation expectations"

Copied!
34
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

Emperical Behavioral Model of Inflation

Expectations

Pim Burgers (10351736)

Faculty of Economics and Business, University of Amsterdam

Supervisor dr. Tomasz A. Makarewicz

June 28, 2016

Abstract

This paper investigates the fit of the New Keynesian Phillips curve to inflation data from France, Greece, The Netherlands, Italy, Denmark, Australia and Canada. Instead of rational expectations the Phillips curve has a behavioral part, which uses the Heuristic Switching Model. I found that the Phillips Curve based on the Heuristic Switching Model is a good estimator for inflation data from the chosen countries.

Keywords: Phillips curve, Heuristic Switching model, actual inflation, real marginal costs.

Acknowledgment: I would like to thank my supervisor, dr. Tomasz A. Makarewicz, for his invaluable help and guidance throughout this project. Also I would like to thank Bert ter Woerds for correcting my English.

(2)

Statement of Originality

This document is written by Student Pim Burgers who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervi-sion of completion of the work, not for the contents.

(3)

Contents

1 Introduction 1 2 Research design 4 2.1 Data description . . . 5 2.2 Model . . . 5 2.3 Estimation . . . 7

3 Results and Analysis 7

3.1 Overview of Results . . . 8 3.2 Inflation Predicted by the Model . . . 9

4 Other Countries 12

4.1 Countries . . . 13 4.2 Overview of Results for other countries . . . 13 4.3 Inflation Predicted by the Heuristic Switching Model vs

Al-ternative Fundamentalist Only Model . . . 15

5 Conclusions 20

References 22

Appendix I 25

Appendix II: Greece 26

Appendix III: The Netherlands 27

Appendix IV: Denmark 28

Appendix V: Italy 29

Appendix VI: Australia 30

(4)

1

Introduction

”Inflation is when you pay fifteen dollars for the ten dollar haircut you used to get for five dollars when you had hair.” Sam Ewing Every household has to buy groceries and other needed products. Every year those households see that the prices of these products are rising, whereas the household income stays the same, due to nowadays inflation. These days in-flation is a big problem, because no one can estimate or predict it accuratly enough. It is also a problem because inflation of some European countries al-most equals to zero or even lower (European Central Bank, 2016). Deflation is also a problem because it increases the real value of debt, and may aggra-vate a depression which in turn may lead to a deflationary spiral. The latter may be difficult to escape (Fisher, 1933). At this moment inflation in the European Union is 0.0%. In Greece, Spain, Romania, France, Italy, Cyprus and others the inflation rate is below zero, becoming a major macroeconomic problem (European Central Bank, 2016).

It is important to investigate empirical models about inflation dynamics because first of all there is only a small amount of empirical research avail-able and secondly most empirical models are based on data from the United States. Empirical models are important for any future estimations. In this research data from a European countries will be used, instead of United States data.

This research follows the methodology set by Cornea et al. (2012). The authors estimate a model for behavioral heterogeneity in US inflation dy-namics. The authors make a empirical behavioral model for estimating the inflation based on the real marginal costs and inflation forecast by the agents. Inflation forecast by the agents is based on inflation predicted by fundamen-talist and naive agents. The authors added a section about robustness to alternative measures of marginal costs. This research replicates the paper by Cornea et al. (2012), this research will take an alternative behavioral model

(5)

of inflation and show that it fits well to non-United States data. How to estimate a behavioral model of inflation dynamics?

Other articles also investigate a behavioral model of inflation dynamics. Fuhrer and Moore (1995) demonstrate that the behavior of the conventional Phelps-Taylor model of overlapping wage contracts stands in stark contrast with important features of U. S. macro data for inflation and output.

Furthermore, there is an abundant literature documenting heterogene-ity in inflation expectations. Madeira and Zafar (2015) propose a model in which agents provide inflation forecast based on observable information, such as the previous inflation rates, and unobservable information. Their model estimation reveals that longer term inflation expectations of agents matter very little for near term forecasts. Pfajfar and Santoro (2010) investigate the distribution of households inflation expectations using methodologies like adaptive learning and information stickiness.

Carroll (2003) and Mankiw et al. (2003) show that inflation expectation evolve over time in response to economic volatility. Carroll (2003) shows that a very simple model in which the typical household’s expectations are updated probabilistically with regard to the views of professional forecasters capturing much of the variation in the Michigan Survey’s measures of infla-tion and unemployment expectainfla-tions. Mankiw et al. (2003) state that the existing literature lack in explaining the time relevance of estimated inflation for the expected inflation. In their paper they explain the different methods for estimating inflation and their relevance. They conclude that not every-one has the same expectations and that the existing models are capable of explaining inflation expectations.

Groen, Paap and Ravazzolo (2013) discuss the use of time series based variable with one lag to forecast inflation. They discuss which variables should be used for estimating tomorrow’s inflation, for example they use GDP in volume terms, the unemployment rate and the real spot price of oil. The authors use a very different model to estimate the inflation; they try to estimate inflation on the basis of expected variables, e.g. the price of oil or the unemployment rate.

(6)

(2012) use in their article the output gap, unit labor costs, the labor share of income and consumption-output ratio for the real marginal costs variable. However, there are other papers which have a different view, these papers will be discussed here. Adam and Padula (2011) state that the estimated model will not be optimal if the output gap is used as a measure of real marginal costs. They also state that it is crucial how one measures the real marginal costs to determine the inflation. Adam and Padula’s (2011) main findings are that both measures, the unit labor costs and the output gap, perform well for the estimated inflation. Gali and Gertler (1999) and Sbordone (2002) show that average unit labor costs perform well as an approximation for the real marginal costs. Gali and Gertler (1999) and Gali, Gertler and Lopez-Salido (2001) investigate the importance of marginal costs for inflation dynamics. They also investigate the main components of marginal costs, in their inves-tigation the labor productivity and real wages. They conclude that the real marginal costs with these components have a high correlation with inflation in the Euro area in 1970-1998. They also compare characteristics of Euro-pean inflation dynamics with those observed in the United States.

Atkeson and Ohanian (2001) show that the Phillips curve is a good me-assure to forecast inflation. The importance of the Phillips curve is shown in their article. However, they do this with other variables, e.g. they use the unemployment rate.

There is an abundance of research on inflation dynamics; however, there is a limited number of literature about estimating a behavioral model of inflation dynamics. There is also quite a bit of literature about inflation expectations. Some authors (Gali and Gertler (1999)) state that it is most necessary in order to use at least the unit labor costs and the output gap in real marginal costs.

The paper is organized as follows. Section 2 proposes a model and de-scribes the research plan. Section 3 gives an analytic review of the results for France. Section 4 does this for other countries. Finally section 5 concludes.

(7)

2

Research design

Inflation is a sustained increase in the price level of goods and services over a period of time. Economists use the term inflation to refer to a rise in the price level.

These days inflation is almost equal to zero. After the recession in 2008 the inflation dropped rapidly, even below zero (European Central Bank, in-flation rate, 2016). From 2008 until 2012 the inin-flation was rising. However, after 2012 it decreased again. Mishkin (2007) proves that inflation, over time, becomes less responsible from other shocks in the economy.

Figure 1: Inflation France, source: data from the World Bank of data, infla-tion rate, 2016

The French inflation (Figure 1) shows that recessions are visible in the infla-tion dynamics, for example the rapid decrease in 2008, the 1981 high inflainfla-tion crisis and the first 1973 oil crisis. Furthermore figure 1 shows that inflation is quite low these days, which is a problem. If prices fall (deflation) people avoid buying products. Spending decreases, causing companies to sell less. If they do sell less, they have to fire staff. Then unemployment rates rise. Thus people have less to spend. The decreased demand for products forces companies to lower their prices, thus lowering their sales and so on. Deflation is dangerous as the economy can fall in a downward spiral that is almost im-possible to escape. The result is a deep recession (Bemanke & James, 1991; Fisher, 1933; Kehoe & Prescott 2007). Akerlof et al. (1996) investigate this

(8)

problem in a different way. They investigate what should be done about it. This discussion shows the importance of a good empirical model of inflation, since it is an important macroeconomic variable.

In this section first the data are described. Then the model is described and explained. Also the variables or parameters are described (this is also summarized in appendix I).

2.1

Data description

This study starts with annual French data on the inflation rate in the 1980-2010 period. After France the same research will be done for other countries, the Netherlands, Greece, Denmark, Italy, Australia and Canada.

The real data used have been obtained from the World Bank of Data or Organisation for Economic Co-Operation and Development (OECD). For the real marginal costs variable the unit labor costs (supported by Cornea et al., (2012)) are used. The output gap is obtained from the International Monetary Fund (IMF).

2.2

Model

In this thesis an estimation for a behavioral model of inflation dynamics will be made. The general Phillips Curve is:

πtPhillips= δ ˆEtπt+1+ γmct (+ξt)

The inflation is calculated on the basis of inflation forecast by the agents of the next period and the real marginal costs ’mct’. Also an error term (ξt) is

added, which allows for a margin.

The Heuristic Switching Model is a model where agents use simple rules of thumb (in forecasting) and in learning agents switch between different forecasting rules on the basis of their performances.

(9)

Inflation forecast by the agents of the next period is calculated by adding up the estimated inflation by naive agents and the estimated inflation by fundamentalist agents. ˆ πt+1 = nftπˆ f t+1+ n n tπˆ n t+1

Meaning that agents switch between naive or fundamentalists forecast. Both the inflation by naive and the fundamentalist agents are multiplied by a fixed value, both values adding up to 1.

s.t nnt + nft = 1

The Phillips curve is now:

πHSMt = δˆπt+1+ γmct (+ξt).

Tomorrow’s expected inflation by naive agents is calculated by taking yes-terday’s inflation from the data set.

ˆ

πt+1n = πt−1

The fundamentalists believe that there is a relation between inflation and real marginal costs.

ˆ

πt+1f = γe01(I − δA)AZt

Where e01 is a unit vector, I is an identity matrix, the variables δ and γ are from the expected inflation equation, and where A is calculated by a VAR model, more specific a VAR(1) model, a model with one lag, more specification about this model can be found in the article from Cornea et al. (2012).

Zt = AZt−1+ εt

Where the Zt matrix consists of:

(10)

The nft variable will be calculated for every t, this variable is not fixed, but changes with relative performances; it is also used in the heuristic switching model. nft = 1 1 + exp  β  F Et−1f −F En t−1 F Et−1f +F En t−1  With F Et−1i = K X k=1 |ˆπt−ki − πt−k| with i = f, n.

2.3

Estimation

To make the Phillips curve most ideally fitting to the actual inflation, the following mean squared error (MSE) can be minimalized:

MSE(δ, γ, β) = ∞ X t=1 (πHSMt − πt)2 → min δ,γ,β MSE.

The mean squared error is a function of three (δ, γ and β) model parameters, all three parameters will be estimated.

This model will be made in MATLAB. Furthermore the Heuristic Switch-ing Model is compared to a model without the Heuristic SwitchSwitch-ing Model. So a difference is made between these two: one is that agents are fully funda-mentalists (alternative, nft = 1), the other is that agents use a combination of estimated inflation by naive and fundamentalist agents (HSM).

3

Results and Analysis

The model specified in the previous chapter provides a Phillips curve. The question is whether this Phillips curve is in any way similar to the actual inflation dynamic.

This chapter provides the results and analysis of the Phillips curve for France, the inflation forecasted by agents, the weight variable and an overview

(11)

of results.

3.1

Overview of Results

The general statistics for the quality of the estimators for the HSM and alternative fundamentalist only model are reported in Table 1.

Table 1: MSE and R2 of France, in both cases.

MSE R2

HSM 34.5006 0.9112

Alternative (nft = 1) 34.9143 0.9083

The fit of the estimated model, as seen in its Mean Squared Error over the period of thirty years is not high, for both models (HSM and alternative fun-damentalist only model) it can be concluded that the New Keynesian Phillips curve for France is a good estimation for the actual inflation. The coefficient of determination (R2) is in both cases high, meaning that much is explained

by the used variables; however, the combination of both estimated infla-tion by naive and fundamentalist agents has a higher R2 and a lower mean

squared error. The results are close to each other, but the HSM is preferred1.

Table 2: Descriptive statistics of actual inflation, inflation predicted by HSM and inflation predicted by fundamentalist only for France.

π πHSM πnft=1 Mean 3.6668 3.6508 3.5852 Minimum 0.5334 0.4338 0.4246 Maximum 13.5391 14.2320 14.7910 Std. dev. 3.7067 3.6659 3.6261 Skewness 1.7254 1.8541 6.1197 Kurtosis 4.7139 5.2747 6.1197

1The process of the HSM is stationary. Tested on stationarity by Augmented Dickey Fuller test

(12)

From the descriptive statistics (Table 2) of actual inflation, HSM inflation and inflation predicted by the fundamentalist only model it can be concluded that over thirty years the mean and standard deviation of the HSM and fun-damentalist only model are almost equal to actual inflation, so over time inflation predicted by the HSM gives a reliable estimation (average and stan-dard deviation) of actual inflation. However, in both cases the minimum and maximum value are different, which means that high and low shocks are difficult to predict. When comparing the actual inflation, HSM inflation and fundamentalist only inflation it can be concluded that high and low shocks are more difficult to predict for inflation predicted by only the fundamental-ist. This means that this is the second proof that fundamentalist only model estimates actual inflation worse than inflation predicted by the HSM.

3.2

Inflation Predicted by the Model

First the behavioral model will be scrutinized and second it will be compared to the alternative.

Phillips curve

(13)

Figure 2 shows that the inflation predicted by the HSM does not differ much from the actual inflation, which means that there are not very big outliers. However, at the end in 2010 there is a very big one, this is after the 2008 financial crisis. This economic shock is difficult to predict. Furthermore fig-ure 2 shows that after year 2000 the Heuristic Switching models’ inflation is continuously above the actual inflation, so it can be concluded that inflation predicted by the HSM is more optimistic than actual inflation. For the case of the HSM, the estimated parameters in the Phillips curve are: δ = 0.5327, γ = 0.7042 and β = 1.2219. The inflation predicted by the HSM is on the basis of 0.53 times the inflation forecast of the agents and 0.70 times the real marginal costs. Whereas the beta is the intensity of choice in the forecast error.

(a) Expected fundamentalists naive and agents forecast inflation over time.

(b) The weight of the forward looking variable nft over time.

Figure 3: Plots of agents, fundamentalist, naive inflation and weight variable for the estimated HSM for France.

When comparing Figure 3a to Figure 3b, it seems that when things settle down people become more naive in this case the weight of the fundamental-ist is lower. So there is a correlation between economic instability and the weight variable of fundamentalists.

Over time the inflation forecasted by the agents stabilizes, both the in-flation predicted by fundamentalist and naive agents are equal to each other

(14)

around 2010 this is due to the decreasing of economic shocks.

The weight (nft) variable differs over time. The figure of the weight variable shows that over time it does not become stable, due to relative performances.

Table 3: Descriptive statistics of weight nf.

Mean 0.4663 Minimum 0.2735 Maximum 0.7384 Std. dev. 0.1327 Skewness 0.4338 Kurtosis 2.4242

Table 3 shows that inflation forecast of the agents are more dependend on inflation predicted by naive agents than fundamentalist agents. This could indicate that choosing the alternative fundamentalist only model (nft = 1) is worse and even the maximum of the weight variable never actually is one: max nft < 1.

Phillips curve of the fundamentalist only model.

Figure 4: Inflation predicted by the fundamentalist only model for France. Figure 4 shows that there is one big difference with the HSM, the beginning is predicted with more accuracy.

(15)

The following arguments show that HSM outperforms the alternative fundamentalist model:

• The alternative fundamentalist model has a wider gap between the Phillips curve and the actual inflation, due through the less accurate prediction. Furthermore the coefficient of prediction (R2) is lower in

al-ternative fundamentalist model, this means that the chosen variables in the alternative fundamentalist model explain less of the Phillips curve. • The alternative fundamentalist model has a higher maximum and a lower minimum than inflation predicted by the HSM, which means that the alternative fundamentalist model has more difficulty to predict abnormalities.

• Descriptive statistics of weight nf show that the inflation forecast of

agents is not only based on inflation predicted by fundamentalist agents. It can be concluded that making inflation forecast of the agents only depend-able on fundamentalists is not preferred.

4

Other Countries

How do the Phillips curve (inflation predicted by the HSM), the weight (nft) variable and the alternative fundamentalist only model look for other coun-tries than France? This section gives an analysis of Greece, the Netherlands, Denmark, Italy and Australia. Figures of these countries can be find in ap-pendices II-VI. First of all a small introduction of each country is given, second a look is taken at some of the results (mean squared error and R2).

Third the graphs of inflation predicted by the HSM are analysed. Finally there will be a discussion which model is best, the Heuristic Switching Model or the alternative fundamentalist only model.

(16)

4.1

Countries

France is taken as a starting point, but other countries could be investigated as well. Greece (appendix II) and Italy (appendix V) serve as examples of southern European countries. Greece does not have a stable economy in 2007-2008 the financial crisis struck the country very hard. Greece is chosen because it is economically speaking one of the worst performing countries of Europe. In terms of economic performance Italy is a country between the northern European countries and Greece.

The Netherlands (appendix III) and Denmark (appendix IV) are exam-ples of northern European countries. Both countries have a stable economy. There is a difference made between a country from the mainland and a Scan-danavian country.

This research also consider two non-European economies, namely Aus-tralia (appendix VI) and Canada (appendix VII). AusAus-tralia has other influ-ences on their economy than European countries, e.g. there economy depend very much on export and trade. Canada has been chosen due to its geo-graphical location: north of the United States of America, most empirical macroeconomic models are based on data from the USA, does the model works for a country north of the USA? Non-European countries are chosen to verify if the model can be applied outside Europe.

4.2

Overview of Results for other countries

Before submitting the countries to closer scrutiny, the general results of the models’ estimations across these countries is discussed. These are the mean squared error and the coefficient of determination (R2) as was the case of

France. As in the case of France, there is a difference made between the HSM and the alternative fundamentalist only model.

(17)

Table 4: MSE and R2 of all Countries, HSM and fundamentalist only.

The Netherlands Greece Denmark Italy Australia Canada

MSE 47.3768 149.98 98.717 114.71 72.308 33.294

R2 0.1434 0.9105 0.4488 0.8611 0.6906 0.8626

MSEalt 98.4687 270.88 135.34 76.510 125.10 50.182

R2

alt 0.0443 0.8436 0.2712 0.8987 0.5205 0.8154

The results in Table 4 show that for every country the alternative fundamen-talist only model has a higher mean squared error and a lower coefficient of determination. The exception is Italy where the opposite occurs. When a country already has a low R2 in the HSM the coefficient of determination drops with a higher value in the alternative fundamentalist model than when the country has a higher R2. Proof has been found that the alternative

fun-damentalist model is worse.

The results in Table 4 show that in numbers the mean squared error for the HSM is highest for Greece and lowest for Canada; however, the Nether-lands and France come close to result of Canada. While the mean squared error for the Netherlands is low, the R2 is for example lower than Denmark, whereas Denmark has a higher mean squared error, due to the high inflation for Denmark in the year 1980 to 1985. The coefficient of determination for the Netherlands is very low, which could indicate that there could be bet-ter albet-ternative variables for the real marginal costs and in the VAR model. The high mean squared error and the low coefficient of determination for the Netherlands can be explained by the difference at year 1980 between infla-tion predicted by the HSM and the actual inflainfla-tion, this causes an increase in the mean squared error, removing this value makes the mean squared error lowest of all countries. Furthermore both Greece and Italy have a difference between the inflation predicted by the model and the actual inflation; how-ever these countries have a high R2 despite the also high mean squared error, for these countries much is explained by the chosen variables.2

2The Netherlands, Australia, Canada and Denmark are stationary in the first differ-ences. The other countries (Italy and Greece) are stationary in the process itself (like France). Tested on the basis of the Augmented Dickey-Fuller test.

(18)

4.3

Inflation Predicted by the Heuristic Switching Model

vs Alternative Fundamentalist Only Model

The figures of each country are submitted to closer scrutiny. The figures of the Netherlands (Figure 5), Greece (Figure 6) and Italy (Figure 7) are both shown in the thesis and in the appedices, for all other countries the figures can be found in the appendices.

(a) Inflation predicted by the HSM over time (thirty years).

(b) Fundamentalists, naive and agents forecast inflation.

(c) Inflation predicted by the alternative fundamen-talist model over time.

Figure 5: Predicted inflation by the HSM, agents and alternative fundamen-talist model for the Netherlands.

For the Netherlands the HSM predicts the inflation with higher precision than the alternative fundamentalist only model (Figure 5). Inflation predicted by the HSM is more around the actual inflation than inflation predicted by the alternative fundamentalist only model. Figure 5b also shows that inflation forecasted by the agents (aggregated forecast over the two types of agents) is closer to the curve of inflation predicted by naive agents than fundamentalist agents. The mean squared error from table 4 is not very high; it is close to the mean squared error of France.

(19)

(a) Inflation predicted by the HSM over time (thirty years).

(b) Fundamentalists, naive and agents forecast inflation.

(c) Inflation predicted by the alternative fundamen-talist model over time.

Figure 6: Predicted inflation by the HSM, agents and alternative fundamen-talist model for Greece.

Figures 6a and 6c show that after 1992 inflation predicted by the HSM is more precise to the actual inflation than inflation predicted by the fundamentalist only model, it follows that this is another proof for rejecting the alternative fundamentalist only model. Figure 6b shows that when the economy settles down people become more naive.

For Greece the difference between the inflation predicted by the HSM and actual inflation over a period of thirty years (calculated by the mean squared error) is very high. The mean squared error for Greece is highest of all countries, due to its economic instability over the past thirty years. The Phillips curve (Figure 6a) shows that Greece has some sudden increases and decreases in inflation predicted by the HSM, e.g. the peak at 1985. These peaks in inflation are difficult to predict. On the other hand from year 2000 to 2010 inflation follows a downward trend, there are no sudden increases or decreases. This is easier to predict.

(20)

(a) Inflation predicted by the HSM over time (thirty years).

(b) Fundamentalists, naive and agents forecast inflation.

(c) Inflation predicted by the alternative fundamen-talist model over time.

Figure 7: Predicted inflation by the HSM, agents and alternative fundamen-talist model for Italy

Figure 7b shows that inflation predicted by fundamentalist agents is very contiuously around zero which is unexpected, due to stability of inflation pre-dicted by fundamentalist agents the alternative fundamentalist only model is much better. The alternative fundamentalist model (Figure 7c) shows less sudden increases and decreases than inflation predicted by the HSM (Figure 7a). It follows that this is the first proof for using the alternative model; however, Italy was an exception. For Italy the mean squared error (Table 4) is lower than for Greece, which means that inflation predicted by the HSM is more precise to actual inflation for Italy than for Greece. This could also be concluded from Figure 7a. After 2005 actual inflation has become more stable, so it is easier to predict.

For Denmark, Australia and Canada exactly the same occurs as for the Netherlands, Greece and France (Figures are in the appendices). Inflation predicted by the alternative fundamentalist only model is worse than infla-tion predicted by the HSM, meaning that there is again (like for France) overwhelming evidence to suggest that the alternative fundamentalist only model is not compatible. Also the figures of inflation predicted by the HSM (Phillips curve) show that the Phillips curve in year 1980 and actual infla-tion in year 1980 are very different from each other, it could be considered to remove the first value.

(21)

Table 5: Inflation predicted by the HSM, actual inflation and inflation pre-dicted by fundamentalist only.

Net Gre Den

π πHSM πnf=1 π πHSM πnf=1 π πHSM πnf=1

Mean 2.48 2.24 1.71 11.3 11.2 10.79 3.71 3.62 3.36

Min -0.70 -0.48 -1.76 1.21 1.50 0.87 1.16 -0.66 -1.99

Max 6.74 6.22 4.46 24.8 23.7 24.91 12.3 10.5 8.48

Std. dev. 1.65 1.39 1.64 7.95 7.92 8.03 2.97 2.50 2.60

Ita Aus Can

π πHSM πnf=1 π πHSM πnf=1 π πHSM πnf=1 Mean 5.86 5.42 5.71 4.69 4.72 4.35 3.61 3.49 3.28 Min 0.75 -1.12 -0.13 4.69 4.72 1.26 3.61 3.49 -1.41 Max 21.3 21.2 20.70 11.14 11.48 13.59 12.46 12.42 12.05 Std. dev. 5.24 5.36 5.16 3.21 2.88 3.04 2.96 2.89 3.06

Table 5 shows that the average inflation predicted by the HSM over thirty years is for every country a good indicator to the average actual inflation. When comparing the actual inflation and the inflation predicted by the al-ternative fundamentalist only model the results have a larger difference with actual inflation (except for Italy) than inflation predicted by the HSM. The results from inflation predicted by the alternative fundamentalist only model suggest that predicting peaks in inflation is more difficult with the funda-mentalist only model. It follows that this is the third proof for not using the alternative fundamentalist only model. Furthermore the standard deviation of inflation predicted by both models is very high for Greece, due to the economic instability and the high average (models) inflation. The minimum value of inflation for some countries is negative, this is favourable if inflations minimum for the actual inflation is also negative. The best models predicted inflations minimum can be found in the Netherlands, wereas inflation pre-dicted by the HSM and actual inflation minimum are closest to each others, for the maximum value of inflation this occurs for Canada.

(22)

Table 6: Estimated δ, γ and β for all Countries.

Netherlands Greece Denmark Italy Australia Canada

δ 0.8522 1.1638 0.7661 0.1444 0.8746 0.5042

γ 0.4363 -0.0237 0.5984 0.8354 0.3801 0.7451

β 2.6651 2.7812 0.5538 134.12 0.2519 0.4261

The δ is the coefficient which determines the weight of the inflation forecast by the agents. The γ determines the the weight of the real marginal cost in predicting the inflation. The β is the coefficient in the weight variable (nft) and determines the weigth of the forecast errors.

Table 6 shows that δ is highest for the Greece, which means that inflation predicted by the HSM in Greece depends most on the inflation forecast by the agents. Also it occurs that there is a negative relation between inflation predicted by the model and real marginal costs for Greece. This is perhaps due to the loans Greece has with European countries and so the real marginal costs (unit labor costs) become extremly low. The γ is highest for Italy, which means that inflation predicted by the HSM in Italy depends most on the real marginal costs. The γ for Italy is high, due to the low δ, the low δ could be explained by the inflation predicted by fundamentalist agents, which is around zero. This has an influence on the aggregate inflation forecast by the agents. What is remarkable is that β is very high for Italy. The β is a variable for the weight variable (nft), which influences the fraction of the forecast error from inflation predicted by fundamentalist and naive agents. Which means that when the β is very high (in the case of Italy) the weight variable is more often low, it follows that the inflation forecast by the agents depends more on inflation forecast by naive agents. The exception (Italy) can be explained, the inflation forecast by naive agents is less stable than inflation predicted by fundamentalist agents, meaning that when inflation predicted by only fundamentalist agents, the difference between the estimated inflation by the fundamentalist model and actual is inflation is lower (due to its stability) than the inflation predicted by the HSM and actual inflation.

Again it can be concluded that making inflation forecast of the agents only dependable on fundamentalists is not preferred, on the other hand there

(23)

is an exception (Italy).

5

Conclusions

This paper investigated whether it is possible to estimate a behavioral model of inflation dynamics. It is important to investigate a empirical macroe-conomic model with European data, due to the already existing empirical models with USA data but the lack of empirical European models existing. The fit of the Phillips curve towards the actual inflation is optimized by min-imizing the difference between the inflation predicted by the HSM and actual inflation. This thesis estimated the inflation on the basis of real marginal costs and the inflation forecast by the agents. Inflation forecasted by the agents is based on expected inflation by fundamentalist agents and naive agents. Fundamentalists believe that there is a relation between inflation and real marginal costs, the naive type of people believe that tomorrow’s inflation is equal to yesterday’s. Both have a weight variable that adapts over time, based on their relative performances. The alternative fundamen-talist only model states that inflation forecast by the agents only depends on fundamentalists agents. The R2 of both models is compared.

The main results of this paper are summarised as follows. First, the New Keynesian Phillips curve for the countries (France, the Netherlands and Canada) gives a good fit for actual inflation over the past thirty years. Sec-ond, there is proof that the inflation predicted by the alternative model is less optimal than the inflation predicted by the behavioral model. Third, there are countries (e.g. Greece, Italy) that have unstable economies and therefore suffer from many economic shocks, which ensures that the differ-ence between the actual inflation and inflation predicted by both models is higher and inflation predicted by both models is less than optimal. Finally, the process is tested on stationarity.

As for future work, it is possible to investigate different real marginal costs. It is possible to add more terms in the VAR model, e.g. labor share of income or maybe even the price of oil over the past thirty years. This model

(24)

can aslo be used for other countries.

To summarize, it seems that during periods of economic shocks, infla-tion is more difficult to predict than when the economy is more stable. I have found evidence to support this when investigating different countries. It is possible to make a behavioral model of inflation dynamics, a behavioral model is more precise than a non behavioral model.

(25)

References

Adam, K., & Padula, M. (2011). Inflation dynamics and subjective expecta-tions in the United States. Economic Inquiry, 49(1), 13-25.

Akerlof, G. A., Dickens, W. T., Perry, G. L., Gordon, R. J., & Mankiw, N. G.

(1996). The macroeconomics of low inflation. Brookings papers on economic activity, 1996(1), 1-76.

Atkeson, A., & Ohanian, L. E. (2001). Are Phillips curves useful for

forecast-ing inflation?. Federal Reserve Bank of Minneapolis. Quarterly Review-Federal Reserve Bank of Minneapolis, 25(1), 2.

Bemanke, B., & James, H. (1991). The gold standard, deflation, and fi-nancial crisis in the Great Depression: An international comparison. In Financial markets and financial crises (pp. 33-68). University of Chicago Press.

Carroll, C. D. (2003). Macroeconomic expectations of households and pro-fessional forecasters. the Quarterly Journal of economics, 269-298.

Cornea, A., Hommes, C., & Massaro, D. (2012). Behavioral heterogene-ity in US inflation dynamics. Universheterogene-ity of Amsterdam.

European Central Bank (2016). Inflation rate.

Retrieved from: https://www.ecb.europa.eu/stats/prices/hicp/html/inflation.en.html on April 22, 2016

Fisher, I. (1933). The debt-deflation theory of great depressions. Econometrica: Journal of the Econometric Society, 337-357.

Fuhrer, J., & Moore, G. (1995). Inflation persistence. The Quarterly Journal of Economics, 127-159.

(26)

Gali, J., & Gertler, M. (1999). Inflation dynamics: A structural econometric analysis. Journal of monetary Economics, 44(2), 195-222.

Gali, J., Gertler, M., & Lopez-Salido, J. D. (2001). European inflation dy-namics. European Economic Review, 45(7), 1237-1270.

Groen, J. J., Paap, R., & Ravazzolo, F. (2013). Real-time inflation forecast-ing in a changforecast-ing world. Journal of Business & Economic Statistics, 31(1), 29-44.

International Monetary Fund (IMF) (2016). Output gap. Retrieved from IMF: https://www.imf.org/external/pubs/ft/weo/2016/01/weodata/weoselgr.aspx

Kehoe, T. J., & Prescott, E. C. (2007). Great depressions of the twentieth century. Research Department, Federal Reserve Bank of Minneapolis.

Madeira, C., & Zafar, B. (2015). Heterogeneous Inflation Expectations and Learning. Journal of Money, Credit and Banking, 47(5), 867-896.

Mankiw, N. G., Reis, R., & Wolfers, J. (2004). Disagreement about inflation expectations. In NBER Macroeconomics Annual 2003, Volume 18 (pp. 209-270). The MIT Press.

Mishkin, F. S. (2007). Inflation Dynamics*. International Finance, 10(3), 317-334.

Organisation for Economic Co-Operation and Development (OECD)(2016). Unit Labor Costs.

Retrieved from OECD: https://stats.oecd.org/Index.aspx?DataSetCode=ULC ANN

Pfajfar, D., & Santoro, E. (2010). Heterogeneity, learning and information stickiness in inflation expectations. Journal of Economic Behavior & Organization,

(27)

75(3), 426-444.

Sbordone, A. M. (2002). Prices and unit labor costs: a new test of price stickiness. Journal of Monetary economics, 49(2), 265-292.

World Bank of data. (2016). Inflation ratio. Retrieved from world bank of data: http://data.worldbank.org/indicator/FP.CPI.TOTL.ZG

(28)

Appendix I

The variables in one list of appearance.

πt = actual inflation, from the databank at time t.

πtPhillips = inflation predicted by the model without the use of the HSM at time t.

ˆ

Etπt+1= expected inflation for tomorrow t + 1.

πtHSM = inflation predicted by the model at time t. δ = parameter for inflation forecast by the agents. ˆ

πt+1 = inflation forecast by the agents of tomorrow (t + 1).

γ = parameter for real marginal costs. mct = real marginal costs at time t.

ξt= error term at time t.

nft = parameter for the weight of inflation predicted by fundamentalist agents at time t.

ˆ

πt+1f = inflation predicted by fundamentalist agents of tomorrow (t + 1). nnt = parameter for the weight of inflation predicted by naive agents at time t.

ˆ πn

t+1 = estimated inflation predicted by naive agents of tomorrow (t + 1).

πt−1 = actual inflation of yesterday (t − 1).

I = Identity matrix. e01 = unit vector

Zt = Matrix consistent of real marginal costs, inflation, output gap and a

vector with ones at time t.

β = parameter for the forecast errors.

(29)

Appendix II: Greece

(a) nft. (b) Fundamentalist, Naive and Agents expectations inflation.

(c) HSM inflation. (d) Alternative fundamentalists models inflation.

(30)

Appendix III: The Netherlands

(a) nft. (b) Fundamentalist, Naive and Agents expectations inflation.

(c) HSM inflation. (d) Alternative fundamentalists models inflation.

(31)

Appendix IV: Denmark

(a) nft. (b) Fundamentalist, Naive and Agents expectations inflation.

(c) HSM inflation. (d) Alternative fundamentalists models inflation.

(32)

Appendix V: Italy

(a) nft. (b) Fundamentalist, Naive and Agents expectations inflation.

(c) HSM inflation. (d) Alternative fundamentalists models inflation.

(33)

Appendix VI: Australia

(a) nft. (b) Fundamentalist, Naive and Agents expectations inflation.

(c) HSM inflation. (d) Alternative fundamentalists models inflation.

(34)

Appendix VII: Canada

(a) nft. (b) Fundamentalist, Naive and Agents expectations inflation.

(c) HSM inflation. (d) Alternative fundamentalists models inflation.

Referenties

GERELATEERDE DOCUMENTEN

Discussing the work of Walter Segal and describing the Lewisham experience will only be worthwhile when the climate is felt, since one could easily translate

From the behaviour of the reflectivity, both in time and with energy-density, it is inferred that this explosive crystallization is ignited by crystalline silicon

Solutions for the scalar MIMO case, within scaling and permutation ambiguities, have been proposed in the past, based on the canonical decomposition of tensors constructed

In this paper, we frame our analysis around one particular set of regional development actors, university senior managers (rectors, vice-rectors, etc.), and

Da ein Protokollieren dieser Werte (beispielsweise in eine Datei) die Registerinhal- te aber verändert würde, müssen die Zeiger vor dem Aufruf der Protokollierungsroutine 18

In this paper we use a particle filter algorithm to perform a data fusion of several location-related data sources in order to check mobility data plausibility of single-hop

The main question that the paper deals with is how FabLabs allow room for messy improvisation, who the real life users of FabLabs are and what the empirical

As both operations and data elements are represented by transactions in models generated with algorithm Delta, deleting a data element, will result in removing the