Redistributional effects of monetary policy in the Netherlands : real interest rate exposures of Dutch households

(1)

Redistributional Effects of Monetary Policy in the Netherlands:

Real Interest Rate Exposures of Dutch Households

Master Thesis

Faculty of Economics and Business University of Amsterdam Supervisor: Dr. Christian A. Stoltenberg Submitted by: Paul Ruth Student No. 11355743 MSc Economics

Track: Monetary Policy and Banking

(2)

(3)

iii

Abstract

In this thesis, I empirically investigate how changes in the real interest rate affect aggregate spending of Dutch households via wealth redistribution. I analyse how a fall in policy rates affects household balance sheets and aggregate consumption. As my main result, I find that households that benefit from a fall in the real interest rate exhibit a large consumption response relative to households that are hurt. This redistribution of wealth to households with a high marginal propensity to consume increases aggregate consumption. In comparison to Italy and the United States, my findings suggest that there exists a moderate redistribution channel in the Netherlands.

(4)

iv

List of Abbreviations

CE Consumer Expenditure Survey of the United States CPI Consumer Price Index

DNB De Nederlandsche Bank (Dutch Central Bank) EIS Elasticity of Intertemporal Substitution

LISS Longitudinal Internet Studies for the Social Sciences MPC Marginal Propensity to Consume

NNP Net Nominal Position

PSID Panel Study of Income Dynamics of the United States SHIW Survey of Household Income and Wealth

UMP Unconventional Monetary Policy URE Unhedged Real Interest Rate Exposure

(6)

vi

List of Tables

Table 1: Components of the Unhedged Real Interest Rate Exposure…………...………….15

Table 2: Summary Statistic using LISS Panel data…………...………18

Table 3: MPCs out of Income………...………19

Table 4: Correlations of URE with demographic variables……….……….20 Table 5: Estimates of 𝜀̂_𝑟 and 𝜎̂ using LISS Panel data………...……….22 _𝑟

(7)

vii

List of Figures

Figures 1: Household final consumption expenditure (in % of GDP)……...……….……...….7 Figure 2: Auclert’s transmission channels of monetary policy and their effect on aggregate consumption………...………...…….….8 Figure 3: Marginal propensity to consume and the redistribution channel……..……….21 Figure 4: The redistribution elasticity and the equivalent EIS in the Netherlands, Italy and the United States………..………...…23

(8)

1

Chapter I: Introduction

Traditionally, central banks are not charged with the task of addressing inequalities in the distribution of income and wealth. Instead, their primary goal is to maintain price stability in order to facilitate economic growth and employment. However, recent studies such as Coibion, Gorodnichenko, Kueng and Silvia (2012) and Auclert (2017) point out that monetary policy can have an effect on income and wealth inequality. Especially at times of exceptionally low interest rates and non-standard monetary policy measures, it is important to be aware of collateral effects of monetary policy, including redistributional effects.

In this thesis I seek to analyse the impact of monetary policy on the distribution of income and wealth in the Netherlands. Specifically, I examine how changes in the real interest rate affect aggregate spending of Dutch households via wealth redistribution. I do so by measuring the extent to which households’ balance sheets are exposed to rising or falling policy rates, and analysing how households’ wealth is consequently affected. I focus on the effects of falling interest rates, due to the European Central Bank having implemented a set of expansionary monetary policy measures since the global financial crisis.

Auclert (2017) evaluates the effect of changes in the policy rate on the aggregate consumption in Italy and the United States. He argues that aggregate spending should increase since those households that benefit from a fall in the real interest rate tend to increase their consumption expenditure more after an increase in income than households that lose from such a fall. I apply a methodology that is similar to Auclert’s in order to measure how Dutch households’ balance sheets are affected by changes in the real interest rate. As far as I am aware, this contribution to research is novel and complements Auclert’s studies on Italian and U.S. households, and contributes to the knowledge on the redistributional effects of monetary policy.

In order to capture the response of aggregate spending of Dutch households, I quantify the marginal propensity to consume (MPC): the increase in consumption after an unexpected rise in income. Knowledge of how consumption responds to income shocks is crucial for understanding the impact of labour market and tax reforms as well as for designing stabilization and income maintenance policies. I seek to analyse the relation between households’ exposures to changes in the real interest rate and their consumption responses. Generally, aggregate consumption should increase if those who benefit from real interest rate fall have a higher MPC than households that are hurt by such a fall.

To quantify how exposed households’ balance sheets are to changes in the real interest rate, I measure each household’s unhedged real interest rate exposure (URE). This entails

(9)

2

measuring the difference between all maturing assets (including income) and all maturing liabilities (including consumption) at a certain point in time. According to Auclert (2017), households with a high amount of short-term assets generally tend to have a positive URE, while households whose financial wealth is mainly invested in long-term bonds or adjustable-rate mortgage liabilities usually have a negative URE.

I use data from the Dutch Longitudinal Internet Studies for the Social Sciences (LISS) panel to measure the covariance between the MPC and the URE in order to analyse the redistributive component of monetary policy. Aggregate consumption responds to changes in the real interest rate through redistribution and intertemporal substitution. In order to analyse the redistribution component, I estimate the redistribution elasticity and the elasticity of intertemporal substitution (EIS) which captures the total response of aggregate consumption including the redistribution component.

As my main result, I find a negative correlation between the MPC and the URE. This is in line with Auclert’s (2017) findings that a decrease in the real interest rate leads to a redistribution away from low-MPC households with a positive URE to high-MPC households with a negative URE. Compared to Auclert’s findings, I find evidence that the redistribution channel in the Netherlands is larger than in Italy but smaller than in the United States. Hence, Dutch households are moderately exposed to changes in the real interest rate when compared to other countries. Therefore, the conduction of monetary policy should take redistributional effects into consideration in the Netherlands.

My thesis is structured as follows. Chapter 2 provides an overview of the relevant literature regarding the distributional effects of monetary policy. Next, in chapter 3 I discuss Auclert’s (2017) empirical approach and my own methodology. Then, chapter 4 contains my empirical findings. In chapter 5 I discuss the limitations and potential measurement errors of my research. Finally, chapter 6 concludes.

(10)

3

Chapter II: Literature review

This chapter provides an overview of the relevant literature regarding the distributional effects of monetary policy. I describe different channels through which monetary policy can have an effect on income and wealth inequality, and discuss several studies that seek to quantify these channels. I also discuss the response of consumption to income shocks, and why my research on redistribution effects of interest rate changes in the Netherlands is relevant and complements existing research.

2.1 Transmissions Channels of Monetary Policy

This section describes different channels through which monetary policy can have an effect on income and wealth inequality. This effect remains ambiguous as the quantitative importance of different transmission channels can result in its increase or decrease. Coibion et al. (2012) classify five channels through which monetary policy can have an impact on income inequality, which are also used by other authors. These channels are discussed below.

(1) The income composition channel refers to the heterogeneity in income sources across households. Many households depend mainly on wages while others primarily gain their income out of financial and business gains. If expansionary monetary policy leads to a higher increase in profits than it does in labour earnings, households owning assets and firms benefit relatively more than those households earning their income primarily out of wages. Since households with a large share of financial assets generally tend to be more wealthy, expansionary monetary policy may lead to higher income inequality through this channel.

(2) The financial segmentation channel implies that households that are highly connected to financial markets benefit more from expansionary monetary policy. These households are more affected by the money supply and therefore acquire a greater amount of income through these transactions. Since these households generally tend to earn a higher amount of income than households not as connected to financial markets, expansionary monetary policy contributes to an increase in income inequality via this channel.

(11)

4

(3) The portfolio channel represents the redistribution of income based on the structure of owned assets. Low income households tend to have mainly cash and deposits as assets whereas high income households usually have more varied portfolios. As such, inflation and financial market booms tend to hurt households with less accumulated wealth while wealthier households benefit. Therefore, expansionary monetary policy also leads to higher income inequality via this channel.

(4) The savings redistribution channel expresses the impact of unexpected inflation on nominal contracts. An unexpected increase in inflation benefits borrowers due to lower debt repayments while it hurts savers because of a lower real value of their savings. Thus, a decrease in the long-term real interest rate redistributes from savers to borrowers. Since savers are usually wealthier than borrowers, expansionary monetary policy here leads to a decrease in income inequality through this channel.

(5) The labour earnings heterogeneity channel concerns the fact that low income households are more affected by fluctuations in the business cycle as their income decreases more strongly due to unemployment. When expansionary monetary policy seeks to restore the financial market after a recession, unemployment rates tend to go down and labour demand tends to rise. This generally benefits low income households more as they are more affected by changes in unemployment and receive a bigger share of income from government transfers than other households. Therefore, expansionary monetary policy may decrease income inequality through this channel.

2.2 Quantification of Transmission Channels

After section 2.1 described the different channels through which monetary policy can have an effect on income and wealth inequality, this section discusses several studies that seek to quantify these channels.

Coibion et al. (2012) study the effects and historical contribution of monetary policy shocks to consumption and income inequality in the United States since 1980. Using micro-data on income and consumption, they analyse how contractionary monetary policy affects inequality through the different channels. They find that through the labour earnings heterogeneity channel (5), contractionary monetary policy has increased inequality in labour earnings, total income, consumption, and total expenditure. Due to a lack of data on household balance sheets, they are not able to quantify the financial segmentation channel (2), portfolio

(12)

5

channel (3), and savings redistribution channel (4), but they argue that these effects should be relatively small.

Domanski, Scatigna and Zabai (2016) analyse the impact of unconventional monetary policy on interest rates and asset prices in a selected number of countries. They conclude that while low interest rates and rising bond prices have had little effect on inequality, a rise in equity prices has majorly contributed to a rise in inequality. This effect is only partly offset by a recovery in housing prices. However, their analysis is limited in the sense that their measure of wealth is incomplete due to data limitations.

Saiki and Frost (2014) use Japanese semi-aggregate data on pre-tax income and argue that inequality measures decrease after an expansionary monetary policy shock. Specifically, they look at the impact of unconventional monetary policy (UMP) and find that inequality increases via the portfolio channel (3). Monetary policy seems to mainly benefit wealthy households that actively participate in financial markets and invest their savings in equities, yielding a higher business income. As such, they argue that inequality measures increase after an expansionary monetary policy via this channel.

Inui, Sudo and Yamada (2017) also study the redistributional effects of monetary policy on Japanese households. This study differs from Saiki and Frost in the sense that they use micro-level household data rather than semi-aggregate data, and do not only focus on pre-tax income, but also on other inequality measures such as earnings, disposable income, consumption, expenditure and financial positions. They argue that the labour earnings heterogeneity channel (5) appears as the central to the distributional effects of monetary policy in Japan. They find evidence that before the 2000s, an expansionary monetary policy shock increased income inequality through a rise in earnings inequality. However, such procyclical responses are scarcely observed when current data is included in the sample period, or when earnings inequality across all households is considered. Estimated effects of expansionary monetary policy shocks on the difference across households in terms of the value of net assets are also not found to be significantly different from zero, suggesting that both the portfolio channel (3) and the savings redistribution channel (4) are not quantitatively important.

Doepke and Schneider (2006) analyse the savings redistribution channel (4) by evaluating the impact of higher inflation through changes in the value of nominal assets. They specifically look at nominal asset positions of households in the United States. They find that higher inflation and lower interest rates leads to a wealth redistribution away from rich, old householders to young, middle-class households. The former households are the main bondholders in the economy and therefore net lenders, while the latter usually hold fixed-rate

(13)

6

mortgage debt and are therefore net debtors.

Auclert (2017) analyses the effect of monetary policy on overall spending. While the transmission channels classified by Coibion et al. (2012) capture the effect of monetary policy on inequality, Auclert defines different channels that distinguish between aggregate and redistributive components. Chapter 3 discusses more in detail how Coibion’s channels and Auclert’s channels compare. Like Doepke and Schneider, Auclert also analyses the savings redistribution channel (4). He specifically focuses on how exposed households’ balance sheets are to changes in the real interest rate and finds that a fall in the real interest rate redistributes wealth away from rich, net lending households to young, middle-class households.

2.3 Response of Consumption to Income Shocks

Each of the channels that Auclert identifies depends on the marginal propensity to consume (MPC). While these channels are discussed more in depth in section 3.1, this section briefly reviews relevant literature on the MPC.

The MPC measures the increase in consumption after an unexpected rise in income. For example, if a household were to earn one extra euro of disposable income, and the household’s MPC is 0.40, then the household would spend 40 cents of that euro on consumption and save 60 cents. Without borrowing, the household can obviously not spend more than the extra euro, and the MPC can therefore never be greater than 1. (Johnson, Parker and Souleles, 2004)

Empirical literature, reviewed by Jappelli and Pistaferri (2010), consistently finds estimates of the aggregate MPC generally ranging between 0.2 and 0.6 at an annual rate. Furthermore, Jappelli and Pistaferri argue that household with a high MPC usually tend to be liquidity constrained and to hold few household resources. In this context, liquidity constrained households are those households that don’t have access to credit, or are limited in the amount of borrowing. Such households use a relatively high fraction of any additional income they receive for consumption compared to households with a low MPC. In the case of a decrease in real interest rates, there should be a redistribution from net lending households with a low MPC to net borrowing households with a high MPC. This should then result in an increase in overall consumption.

As can be seen from figure 1, household final consumption expenditure has traditionally been relatively low for the Netherlands when compared to other Western countries. In a DNB working paper, Teppa (2014) finds a MPC of 0.21 for Dutch households. She also constructs

(14)

7

two fragility indicators which consist of a debt-to-asset ratio and a debt-to-income ratio. Hence, a higher amount of debt increases fragility, while a higher amount of assets and income decreases fragility. Teppa finds that less fragile Dutch households display a double MPC compared to those that are more fragile (0.44 vs 0.21, respectively).

Figure 1: Household final consumption expenditure (in % of GDP)

In conclusion, there exist different channels through which monetary policy can have an effect on inequality. I seek to contribute to this field of research by analysing the savings redistribution channel like Doepke and Schneider (2006) and Auclert (2017). Similar to Teppa (2014), I quantify marginal propensities to consume in the Netherlands. I then go further and analyse the exposure of households to changes in the real interest rate in the Netherlands and how this in turns affects aggregate consumption. As far as I am aware, there exists no other empirical research on the quantification of the savings redistribution channel in the Netherlands.

(15)

8

Chapter III: Methodology

This chapter covers my methodological approach to estimating the redistribution effects of monetary policy through changes in the real interest rate in the Netherlands. I first describe Auclert’s (2017) analysis of the redistribution channel. Then I provide a summary of my dataset and discuss how I construct my variables. Finally, I show how I arrive at the different estimates that I need and how I use these to measure the interest rate channel in the Netherlands.

3.1 Auclert’s (2017) Analysis of the Redistribution Channel

In his paper, Monetary Policy and the Redistribution Channel, Adrien Auclert evaluates the role of redistribution in the transmission mechanism of monetary policy to consumption. He identifies five transmission channels that affect aggregate consumption. Three of these channels are redistributive while two of the channels only affect the aggregate. Each of the channels depend on households’ marginal propensity to consume (MPC). The sum of these five channels constitutes the aggregation of consumer responses and is shown in figure 1:

𝑑𝐶 = 𝐸𝐼[ 𝑌_𝑖 𝑌𝑀𝑃𝐶̂ ] 𝑑𝑌𝑖 ⏟ 𝐴𝑔𝑔𝑟𝑒𝑔𝑎𝑡𝑒 𝑖𝑛𝑐𝑜𝑚𝑒 𝑐ℎ𝑎𝑛𝑛𝑒𝑙 + 𝐶𝑜𝑣𝐼(𝑀𝑃𝐶̂𝑖, 𝑑𝑌𝑖− 𝑌𝑖 𝑑𝑌 𝑌) ⏟ 𝐸𝑎𝑟𝑛𝑖𝑛𝑔𝑠 ℎ𝑒𝑡𝑒𝑟𝑜𝑔𝑒𝑖𝑛𝑒𝑡𝑦 𝑐ℎ𝑎𝑛𝑛𝑒𝑙 − 𝐶𝑜𝑣𝐼(𝑀𝑃𝐶̂ , 𝑁𝑁𝑃𝑖 𝑖) 𝑑𝑃 𝑃 ⏟ 𝐹𝑖𝑠ℎ𝑒𝑟 𝑐ℎ𝑎𝑛𝑛𝑒𝑙 + ( 𝐶𝑜𝑣_⏟_𝐼(𝑀𝑃𝐶̂ , 𝑈𝑅𝐸_𝑖 _𝑖) 𝐼𝑛𝑡𝑒𝑟𝑒𝑠𝑡 𝑟𝑎𝑡𝑒 𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒 𝑐ℎ𝑎𝑛𝑛𝑒𝑙 − 𝐸_⏟_𝐼[𝜎_𝑖(1 − 𝑀𝑃𝐶̂ )𝑐_𝑙 _𝑖] 𝑆𝑢𝑏𝑠𝑡𝑖𝑡𝑢𝑡𝑖𝑜𝑛 𝑐ℎ𝑎𝑛𝑛𝑒𝑙 )𝑑𝑅 𝑅

Figure 2: Auclert’s transmission channels of monetary policy and their effect on aggregate consumption.

Several comparisons can be drawn to the transmission channels of monetary policy classified by Coibion et al. (2012) which have been discussed in section 2.1. However, while the transmission channels classified by Coibion et al. (2012) capture the effect of monetary policy on inequality, Auclert defines channels that distinguish between aggregate and redistributive components. A description of Auclert’s channels and how they compare to those classified by Coibion et al. is given below.

(i) Aggregate income channel: This channel measures how monetary policy leads to a change in total income across different sectors and in turn affects aggregate spending. This is a non-redistributive effect and does not depend on an individual

(16)

9

source of income. As such, there is no clear comparison with any channel classified by Coibion et al. (2012).

(ii) Earnings heterogeneity channel: This channel describes how monetary expansions induce gains in aggregate earnings from labour and profits. Since some agents will benefit from such expansions while others will lose in relative terms, these gains are likely to be redistributive. This channel can be considered to be a combination of Coibion et al.’s income composition (1) and labour earnings heterogeneity (5) channels.

(iii) Fisher channel: This channel describes how unexpected inflation as a result of monetary policy leads to a revaluation of nominal balance sheets. Whether a household experiences gains or losses depends on its net nominal position (NNP) which measures the household’s exposure to such unexpected inflation. The NNP is calculated by taking the household’s market value of nominal assets and subtracting its market value of nominal liabilities. Therefore, nominal debtors will gain from unexpected inflation while nominal creditors will suffer losses. Such redistribution effects due to unexpected inflation are captured by the financial segmentation (2), portfolio (3) and savings redistribution (4) channels as classified by Coibion et al. (iv) Interest rate exposure channel: This channel examines how exposed households are to changes in the real interest rate based on their unhedged real interest rate exposures (URE). The URE is calculated by taking the difference between all maturing assets and liabilities at a point in time. A decrease in real interest rates will lead to a redistribution from positive URE households to negative URE households. Similar to the case of unexpected inflation, the change in the real interest rate is captured by the financial segmentation (2), portfolio (3) and savings redistribution (4) channels as classified by Coibion et al.

(v) Intertemporal substitution channel: This channel shows how a change in the real interest rate leads to a shift in households’ consumption across periods. Due to a change in opportunity costs of consumption, the aggregate consumption of households is adjusted. Like the aggregate income channel, this channel is non-redistributive. As such, it cannot be compared to any channel classified by Coibion et al.

(17)

10

Since my research is primarily focused on the interest rate exposure channel, I will discuss this channel’s mechanism in more detail. The interest rate exposure channel measures the way in which the aggregate consumption is affected by households’ exposures to unexpected changes in the real interest rate. Households’ unhedged interest rate exposures (UREs) are found by taking the difference in all maturing assets and liabilities at a certain point in time. Income is included in the assets while consumption is included in the liabilities. Therefore, households whose financial wealth is primarily invested in short-term certificates of deposits tend to have positive UREs, while households whose financial wealth is mainly invested in long-term bonds or adjustable-rate mortgage liabilities usually have negative UREs.

Jappelli and Pistaferri (2010) argue that household with a high MPC usually tend to be liquidity constrained and to hold few household resources. Such households use a relatively high fraction of any additional income they receive for consumption compared to households with a low MPC. Auclert (2017) follows this line of reasoning and argues that therefore in the case of a decrease in real interest rates, there should be a redistribution from households with a low MPC and a positive URE to households with a high MPC and negative URE. This should result in an increase in overall consumption.

Next to the interest rate exposure channel, I also focus on the substitution channel. This channel describes how households may change their consumption across periods based on changes in the real interest rate. In the case that a change in the interest rate results in lower opportunity costs of consumption today, households may increase their present consumption relative to the future accordingly. This substitution effect depends on the elasticity of intertemporal substitution (EIS) of the household. Following the same reasoning as above, if there is a negative relation between households’ MPCs and UREs, a decrease in the real interest rate should result in an increase in aggregate consumption.

3.2 LISS panel

This section provides a description of the Longitudinal Internet Studies for the Social Sciences (LISS) panel which I will use to measure the interest rate exposure channel in the Netherlands. The LISS panel is an online longitudinal survey that is representative of the Dutch-speaking population. CentERdata at Tilburg University has conducted and administered the survey since 2007. The panel consists of 4500 households, comprising 7000 individuals, and is based on a true probability sample of households which is drawn from the population register by Statistics

(18)

11

Netherlands. Households that are not able to participate otherwise are provided with a computer and an internet connection.

Online questionnaires are completed by panel members every month for about 15 to 30 minutes. Panel members are paid for each questionnaire that they complete. One member of the household provides the household data and updates this information at regular time intervals. This longitudinal study is repeated yearly and is designed to follow changes in the life course and living conditions of the panel members. In addition to the LISS Core Study, there is ample room to collect data for different research purposes.1

3.3 Construction of Dataset and Variables

This section describes how I construct my dataset and create the variables needed to calculate the MPC and URE which are needed to quantify the interest rate exposure channel. For the purpose of this thesis I focus on four modules available in the dataset. I use the modules “Background Variables” and “Time Use and Consumption” to gather information on income, consumption and background characteristics of households. For these modules I use the waves of 2010 and 2015. Furthermore, I use the “Assets” and “Housing” modules for data on assets and liabilities, using only the wave of 2010. I merge all these modules and construct the following variables:

Net annual household income: Computed by multiplying net monthly household income by twelve.

Total annual household spending: Computed as the sum of monthly household expenditures multiplied by twelve. The monthly household expenditures originate from the following questions.

The following questions are about the expenditure pattern of your household. Can you indicate for each type of expenditure how many euros your household spends on this on average, per month? Consider as a reference period the past 12 months.

1_{https://www.lissdata.nl/about-panel}

(19)

12

-mortgage: interest plus amortization (what matters is the gross amount, so before tax deduction);

-rent (NOT including costs of gas and electricity);

-general utilities (heating, electricity, water, telephone, Internet, etc; but NO insurances); -transport and means of transport (public transport; own car: gasoline/diesel and maintenance, but NOT insurances or the purchase of e.g. a car or [motor] bike);

-insurances (home insurance, car insurance, health insurance, etc.);

-childrens daycare (day care center, out-of-school supervision, guest parents, homework guidance, etc.);

-alimony and ﬁnancial support for children not (or no longer) living at home; -debts and loans (but NOT the mortgage);

-day trips and holidays with the whole family or part of the family (ﬂight tickets, hotel, restaurant bills for the family, etc.);

-expenditure on cleaning the house or maintaining the garden; -eating at home (food, drinks, candy, etc.);

-other.

Total household assets: Computed as the sum of the following variables:

-total balance of the household’s current accounts, savings accounts, term deposit accounts, savings bonds or savings certificates;

-total sum of the guaranteed minimum payout of your single-premium or life annuity insurances, or the total savings amount of your endowment insurance;

-total value of your investments (growth funds, share funds, bonds, debentures, stocks, options, warrants);

-value set by the most recent municipal property appraisal (Dutch: Wet Waardering Onroerende Zaken, WOZ).

(20)

13

3.4 Quantifying the Marginal Propensities to Consume

Both the MPC and URE need to be estimated in order to identify the redistributive component interest rate channel. This section provides a theoretical outline of the MPC and discusses different approaches to estimate the MPC, after which I discuss my own methodology to estimate the MPC out of the LISS Panel. The next section, 3.5, discusses Auclert’s measurement of the URE and the modifications I make to his methodology in order to estimate the URE in the LISS Panel.

Theoretically, the marginal propensity to consume measures the increase in consumption after an unexpected increase in income. In this way, it can predict how income transfers to the household sector affect the aggregate spending (Johnson et. al., 2004). Mathematically, the MPC is defined as the derivative of consumption over the derivative of disposable income or the change in consumption with respect to income:

𝑀𝑃𝐶_𝑖 =𝜕𝐶𝑖 𝜕𝑌_𝑖

Auclert (2017) utilizes different datasets to quantify the interest rate exposure channels in Italy and the U.S. Specifically, he uses data from the Italian Survey of Household Income and Wealth (SHIW), the U.S. Panel Study of Income Dynamics (PSID) and the U.S. Consumer Expenditure Survey (CE). For each of these surveys he uses a different approach to estimate the marginal propensity to consume out of expected and unexpected income shocks, which he bases on approaches described by Jappelli and Pistaferri (2010). The pros and cons of each of these approaches and their utility for my thesis will be briefly discussed below.

A first approach consists of asking households to self-report the part of any hypothetical windfall that they would immediately spend. While this approach has the advantage of taking into account expectations that households may have regarding changes in income, the drawback is that such self-reported answers to hypothetical situations may not reflect how households would actually behave in these situations. Furthermore, while Auclert uses this approach for the SHIW, such a question is not available in the LISS Panel.

A second approach is ‘semi-structural’ and computes MPCs out of transitory income shocks. An income process and a consumption functioned are postulated, after which the MPC is backed out from transitory shocks from the joint from the joint cross-sectional distribution of consumption changes and income changes. This approach can only calculate MPCs at the group level. Therefore, redistribution elasticities are found by first grouping households into

(21)

14

different bins and then estimating MPCs within bins and covariances across bins. However, this procedure leads to significantly larger error bands. Auclert uses this approach to quantify the interest rate exposure channel in the PSID.

A third approach identifies the MPC using exogenous income variation. The MPC is identified out of an increase in income that is expected in advance by exploiting the random variation in timing of the income increase across households. The drawback of this approach is that it does not estimate the MPC out of an unexpected increase in income, although this issue is mitigated if borrowing constraint are important or if households are surprised by the increase in income despite its announcement. This approach can also only calculate an MPC at the group level. Auclert uses this approach for the CE.

Since there is no approach available to quantify the MPC on a cross-sectional level that provides reliable estimates for my dataset, I use Jappelli and Pistaferri’s second approach to calculate MPCs at the group level. Such an approach is also utilized by Auclert (2017) and Kaplan and Violante (2014) using data from the PSID, and is also rather similar to the approach that Teppa (2014) uses to find an aggregate MPC in the LISS Panel.

Like Auclert, I first rank households based on their Net annual household income. I then stratify these households into three different bins, with bin 1 containing the households with the lowest incomes and bin 3 containing the households with the highest incomes. Three bins are used since the regression coefficient needed to calculate the MPC would no longer be significant in at least one bin if more bins were used. Average URE should increase in income, and Auclert notes that in his surveys average URE increases more than one for one with income at the top of the distribution, owing an increase in maturing assets. Logically, my bin 1 should display the lowest average URE while my bin 3 should display the highest average URE. As such, the redistribution elasticity 𝜀_𝑟 (discussed in more detail later) can be easily determined by finding the covariance between each bin’s URE and MPC. I also carry out a robustness test using financial wealth as an alternative stratifying variable.

To measure the MPC, I use a linear speciﬁcation in which the percentage change in consumption is regressed over the percentage change in household income. I also control for several demographic characteristics of the head of the household. The corresponding equation reads as follows:

∆𝑙𝑛𝐶_𝑖𝑗 = 𝛼_𝑖𝑗+ 𝛽∆𝑙𝑛𝑌_𝑖𝑗 + 𝛾𝑋_𝑖𝑗 + 𝜖_𝑖𝑗

where 𝐶_𝑖𝑗 represents consumption, 𝛼_𝑖𝑗 is the household ﬁxed eﬀect capturing time-invariant unobserved heterogeneous characteristics such as household preferences, and 𝑌𝑖𝑗

(22)

15

represents household i’s income. 𝑋_𝑖𝑗 represents demographic characteristics of the head of the household, and the error term is denoted by 𝜖𝑖𝑗. The subscripts i and j denote the household’s

number and bin respectively. The parameter of interest is 𝛽, which represents the elasticity of consumption with respect to income. This parameter is then multiplied by the ratio of average consumption to average income of the bin in order to retrieve the bin’s MPC.

For my measure of consumption, I use Total annual household spending as defined in section 3.3. Income is measured by Net annual household income. Both are measured in 2010 and 2015. Data from 2010 is used because this is the only year for which data on income, consumption, assets and liabilities are all available, and all of these need to be measured in the same year in order to calculate the URE which will be discussed further below. I also use data on consumption and income from 2015 because few respondents report changes in income in the first few years after 2010, which would make it impossible to calculate the MPC. In waves later than 2015, many respondents who participated in 2010 have flowed out of the survey, which would significantly decrease the amount of observations.

3.5 Quantifying the Unhedged Real Interest Rate Exposures

As has been previously mentioned, the unhedged real interest rate exposure measures the household’s balance sheet exposure to changes in the real interest rate. As such, it measures the total resource flow that a household needs to invest over the first period of its consumption plan. It is calculated by taking the difference between all maturing assets (including income) and all maturing liabilities (including consumption) at a certain point in time, and is therefore represented by the following equation:

𝑈𝑅𝐸𝑖 = 𝑌𝑖 − 𝐶𝑖 + 𝐴𝑖− 𝐿𝑖

Like before, 𝑌𝑖 and 𝐶𝑖 represent respectively income and consumption, for which I will again

use Net annual household income and Total annual household spending respectively. 𝐴_𝑖 and 𝐿_𝑖 are respectively assets and liabilities maturing over the period over and above the amounts already included in 𝑌_𝑖 or 𝐶_𝑖. Unfortunately, assets are scarcely reported on by respondents in the LISS Panel. To retain a reasonable amount of observations, I only include the total balance of the household’s current accounts, savings accounts, term deposit accounts, savings bonds or savings certificates in my measurement of household assets. This should be the most substantive component of Total household assets as listed in section 3.3 for most households

(23)

16

(OECD, 2017). Next, very few respondent give non-missing values for the mortgage. Therefore, I do not include the mortgage, because doing so would leave me with only a few hundred observations at best. Thus, my estimate of the URE only includes a specific measure of household financial assets and no liabilities. Table 1 summarizes how I estimate the URE. For each variable I use data from 2010 as this is the only year for which data on all variables is available. For each bin I calculate the average URE normalised by the average consumption in the sample.

Table 1: Components of the Unhedged Real Interest Rate Exposure

𝑌_𝑖 −𝐶_𝑖 +𝐴_𝑖 −𝐿_𝑖

Net annual Total annual Total balance of the household’s -household income household spending current accounts, savings accounts,

term deposit accounts, savings bonds or savings certificates;

3.6 Measuring the Interest Rate Channel

This section provides a description of Aucler’s (2017) methodology of measuring the interest rate exposure channel and substitution channel after having obtained estimates of the MPC and URE. Furthermore, I also discuss how I use this approach to estimate the interest rate channel in the Netherlands.

Auclert makes several simplifications in his paper in order to estimate the response of aggregate consumption to a change in the real interest rate. He looks at a transitory shock in the policy rate, and assumes that households have separable preferences. He defines the partial elasticity of aggregate consumption to the real interest rate as:

𝐶𝑜𝑣𝑙(𝑀𝑃𝐶𝑖, 𝑈𝑅𝐸_𝑖 𝐸𝑙[𝑐𝑖] ) ⏟ 𝜀𝑟 − 𝜎 [𝐸𝑙(1 − 𝑀𝑃𝐶𝑖) 𝑐_𝑖 𝐸𝑙[𝑐𝑖] ] ⏟ 𝑆

Here 𝜀_𝑟 represents the redistribution elasticity of the interest rate channel, 𝑆 is a scaling factor and 𝜎 is the weighted average of the elasticity of intertemporal substitution (EIS):

𝜎 ≔ 𝐸𝑙[𝜎𝑖

(1 − 𝑀𝑃𝐶_𝑖)𝑐_𝑖 𝐸𝑙[(1 − 𝑀𝑃𝐶𝑖)𝑐𝑖]

(24)

17

It is not possible to measure the EIS at the household level. Auclert (2017) therefore works with a proxy 𝜎_𝑟 for which he makes the assumption that the redistribution channel and the substitution channel are of equal size:

𝜎𝑟 ≔

−𝜀_𝑟

𝑆 =

−𝐶𝑜𝑣_𝑙(𝑀𝑃𝐶_𝑖, 𝑈𝑅𝐸_𝑖) 𝐸𝑙[(1 − 𝑀𝑃𝐶𝑖)𝑐𝑖]

Furthermore, 𝜎𝑟 is set up in such a way that it estimates the extent to which the elasticity of a

representative agent 𝜎 differs from the heterogeneous elasticities that are measured empirically. Hence, 𝜎_𝑟 shows by how much 𝜎 from the representative agent economy has to be increased to account for the total response of aggregate spending including the redistributive component. This means that a high value of 𝜎_𝑟 implies a large redistributive interest rate exposure channel, while a small value of 𝜎𝑟 suggests the opposite. While this approach is only an approximation

of the redistribution channel, Auclert (2017) is convinced of its practicability.

In order to measure the interest rate exposure channel in the Netherlands, I use a methodology similar to Auclert’s (2017). I first stratify households into different bins and calculate each bin’s MPC. I also find each bin’s URE normalized by the average consumption in the sample. After having obtained each bin’s MPC and normalized URE, I estimate the redistribution elasticity 𝜀𝑟 by finding the covariance of these MPCs and normalized UREs. This

covariance between the MPCs and normalized UREs is assumed to be negative when the redistribution channel is large. In the case of a decrease in real interest rates, there should then be a redistribution from households with a low MPC and a positive URE to households with a high MPC and negative URE. This should consequently result in an increase in overall consumption. Furthermore, I use the MPCs to estimate the scaling factor 𝑆 which is in turn needed to measure the elasticity of intertemporal substitution 𝜎_𝑟 in the LISS Panel.

(25)

18

Chapter IV: Empirical Findings

This chapter provides an overview and discussion of the results from my empirical analysis using data from the LISS Panel. Section 4.1 presents the summary statistics of my observations using data from the LISS Panel. Section 4.2 concerns empirical evidence regarding the redistribution channel in the LISS Panel population. Finally, section 4.3 evaluates the role of monetary policy through the interest rate channel in the Netherlands.

4.1 Summary Statistics in the LISS Panel

In chapter 3 I described how I construct my dataset and variables in order to quantify the MPC and URE in the LISS Panel. I am eventually left with a dataset of 951 households. Table 2 provides summary statistics regarding income, consumption, assets and the unhedged real interest rate exposure in my whole dataset. A mean of the marginal propensity to consume cannot be provided since no reliable estimate of the MPC can be quantified at the individual household level. However, the next section will provide an estimate of the aggregate MPC as well as estimates of the MPCs of the different bins based on the linear regression discussed in section 3.4. Appendix A provides summary statistics for each of the three bins.

Table 2: Summary Statistic using LISS Panel data

Variable Mean Normalized S.D.

Net annual household income Yi 33,922 2.10

Total annual household spending Ci 22,001 0.52

Total household financial assets Ai 32,877 6.00

Unhedged Real Interest Rate Exposure UREi 44,798 6.37

Numbers of observations 951

Note: Normalized standard deviations are calculated by taking 𝑆𝐷𝐼( 𝑋_𝑖

𝐸_𝑙[𝑐𝑖]) for Xi = Yi, Ci, Ai and UREi.

Income Yi is very close to Auclert’s (2017) results from the SHIW, but quite a bit lower

than his results from the PSID and CE. Consumption Ci is only a few thousand euros lower than

(26)

19

CE. A reason for this could be that the LISS Panel does not include any major purchases in its data on consumption. Next, assets Ai is higher than Auclert’s results in the SHIW and CE, but

lower than his results from the PSID. The URE is only a few thousands lower than Auclert’s results from the PSID, but far higher than his results from the SHIW and CE. However, it needs to be borne in mind that my estimate of the URE does not include any measure of liabilities such as the mortgage. It can reasonably be expected to be quite a bit lower if liabilities were to be included. Compared to Teppa’s (2014) summary statistics of the LISS Panel, my findings regarding income Yi and consumption Ci are rather similar. Assets Ai are somewhat higher than

Teppa’s findings, while Teppa’s measure of Assets includes more variables. Results differ because Teppa considers some different variables and studies different waves than I do. Overall, none of the summary statistics seem to be too much out of the ordinary.

4.2 Estimating the Interest Rate Exposure in the LISS Panel

This section presents the estimates of the MPC and URE in the LISS Panel. I first present and analyse the estimates of the MPCs of the aggregate and of the different bins. Next, I present the estimates of the UREs of the different bins normalized by the average consumption.

Table 3: MPCs out of Income

Total Bin 1 Bin 2 Bin 3

Net annual household income (logs) .246*** .423*** .25*** .11**

Year -.004*** -.005** -.001 -.005**

Female .028 .004 .034 .024

Number of household members .013 .048 .021 .011

Partner -.090** -.245** .009 .66

Education -.010 -.044** .003 .26

Constant .318*** .467** -.037 .52

Implied MPC .16*** .39*** .18*** .06**

Number of observations 951 317 317 317 ***Significance at the 1 percent level. **Significance at the 5 percent level.

(27)

20

As described in section 3.4 I stratify my observations into three different bins of 317 observations each, and then use a linear equation in order to quantify the MPC for each bin. Furthermore, I described in section 3.5 how I quantify the URE of each bin and normalize them by the average consumption of the sample. Table 3 reports the results of my regression of the change in consumption (logs) on the change in income (logs) and several demographic characteristics of the household head. Results of the whole sample as well as of the different bins are presented.

The implied computed MPC of the whole sample is 0.16 which is highly significant. This means that a one euro increase in household income raises consumer spending by 16 cents. This is somewhat lower but still rather close to the aggregate MPC Teppa finds of 0.21. Once again, it needs to be borne in mind that Teppa uses a different linear equation in which she also controls for financial assets and fewer demographic characteristics, and uses different waves in her analysis. Jappelli and Pistaferri (2011) find that most estimates of the aggregate MPC out of income coming from survey data range between 0.2 and 0.6, hence my estimate of the aggregate MPC is slightly lower. The implied MPCs of the different bins are all significant at least at the 5 percent level. Bin 1 consists of households with the lowest net annual household income and has the highest implied MPC, while bin 2 consist of households with the highest net annual household income and has the lowest implied MPC. With regards to background characteristics of the household head, only Year and Partner turn out to be significant in the aggregate and in some of the bins. Both seem to have a negative relation with the MPC.

Table 4: Correlations of URE with demographic variables

Variable Partial corr. Significance value

Year .095*** .003

Female .033 .309

Numbers of household members -.023 .472

Partner .067** .040

Education .020 .530

(28)

21

Table 4 reports the correlations of the URE with demographic variables. The partial correlation coefficients represents the correlation between the two specified variables after the effects of all the other variables remaining in the list have been removed. Only the correlations of the URE with year and partner turn out to be significant. The URE tends to increase with age, and is higher for households whose household head lives together with a partner. Auclert (2017) and Doepke and Schneider (2006) show that older, rich households tend to be net lenders, while young, middle-class households with mortgage debt are generally net nominal borrowers. Furthermore, Auclert shows that the URE also increases with being married. While I do not measure whether the household head is married but rather whether they live together with a partner, I expect these effects to be similar. Since net lenders tend to exhibit a positive URE while net borrowers generally have a negative URE, my results seem to be in line with those of Auclert and Doepke and Schneider.

Figure 3 reports the distribution of the MPC by the URE in the LISS Panel. The figure reports the point estimate of the MPC together with confidence intervals within each bin. The dataset shows a negative correlation between the MPC and the URE. This implicates that indeed 𝜀𝑟 < 0: a fall in the interest rate increases consumption demand via the redistribution channel,

which is in line with Auclert (2017).

(29)

22

4.3 The Interest Rate Channel and Role of Monetary Policy in the

Netherlands

This section provides an evaluation of my results regarding the interest rate exposure channel and gives an interpretation of the role of monetary policy in the Netherlands. I also compare my results to Auclert’s (2017) results of the redistribution channel and its effect on aggregate consumption in Italy and the U.S.

I find the redistribution elasticity 𝜀̂𝑟 in the LISS panel by taking the covariance of the

MPCs and normalized UREs of the different bins. This results in a redistribution elasticity 𝜀̂𝑟 of

-0.16. Auclert (2017) finds a redistribution elasticity of -0.06 for Italian households and -0.24 for U.S. households. The redistribution elasticity for Dutch households therefore seems to take a moderate position when compared to Italian households and U.S. households. The scaling factor 𝑆̂ is found by taking the average of the MPCs of the bins and subtracting it from 1. This results in a scaling factor of 0.79. Auclert makes the assumption that the redistribution channel and the substitution channel are equally large. He therefore uses an elasticity of intertemporal substitution 𝜎_𝑟 that displays by how much to increase the EIS 𝜎_𝑟 from the representative agent economy to capture the total response of aggregate consumption that includes the redistributive

component.

Table 5: Estimates of 𝜀̂_𝑟 and 𝜎̂ using LISS Panel data _𝑟

Parameter Mean Estimate 95% CI

Redistribution Elasticity 𝜀̂_𝑟 -.159 [-.159, -.159]

Scaling Factor 𝑆̂ .790 [.740, .840]

EIS 𝜎𝑟=

−𝜀𝑟

𝑆 .201 [.189. .215]

Note: Confidence intervals are bootstrapped by resampling observations 100 times with replacement.

Table 5 and figure 4 present my results of the redistribution elasticity 𝜀̂_𝑟 and the EIS 𝜎̂ in the _𝑟 LISS panel. I find an EIS 𝜎̂ of 0.20. Auclert measures and EIS of 0.12 for Italian households _𝑟 and 0.30 for U.S. households. Assuming that the EIS 𝜎 in the representative agent economy is similar in these nations, my findings show a redistribution channel in the Netherlands that is larger than in Italy and smaller than in the U.S.

(30)

23

observations 100 times with replacement. This bootstrapping takes place at the bin level, since bootstrapping at the household level can only yield confidence intervals for the MPCs. Bootstrapping at the bin level is not ideal, as it only takes into consideration aggregate MPCs and normalized UREs of the bins. As a result, this confidence interval differs very little from the actual estimate and should be interpreted with caution.

Figure 4: The redistribution elasticity and the equivalent EIS in the Netherlands, Italy and the United States.

Note: Mean estimates are calculated with data from the LISS Panel for the Netherlands, the SHIW survey for Italy and the CEX household dataset for the United States. The results for Italy and the U.S. are obtained from Auclert (2016).

The results of the robustness test using financial wealth as a stratifying variable instead of income are presented in Appendix B. Two bins are used here, since the regression coefficient needed to calculate the MPC would no longer be significant in at least one bin if more bins were used. The results of this test are quite similar, with the redistribution elasticity and elasticity of intertemporal substitution only deviating a few hundredths from my original results when income is used as a stratifying variable. Hence, my results suggest that the interest rate exposure channel in the Netherlands is larger than in Italy but smaller than in the U.S. The redistributive effect of a change in the real interest rate therefore seems to be moderate in the Netherlands in

(31)

24

comparison to other countries. However, measurement errors and limitations should be taken into account when interpreting these results.

(32)

25

Chapter V: Limitations

This chapter discusses several limitations that need to be considered when interpreting my empirical results. These limitations either arise from data availability or the construction of my methodology. Such limitations could be resolved in future studies if more elaborate and longitudinal data becomes available.

Due to data limitations, I am not able to take mortgages into consideration in my estimate of the URE. Very few respondents in the LISS panel report non-missing values for their mortgage, and dropping all observations with missing values for the mortgage leaves me with a dataset that is too small to provide any significant results. As a consequence, my URE contains no estimate of liabilities, which means that individual household UREs and the mean URE are generally higher than if liabilities were to be included. For similar reasons, my measure of assets is limited to current accounts, savings accounts, term deposit accounts, savings bonds and savings certificates. Other variables that could be included in assets such as investments, payouts of insurances and property values are sparsely reported on by respondents in the LISS panel and therefore not taken into consideration. While my measure of consumption is rather comprehensive by including all variables for consumption that are available in the LISS panel, it does not include any major purchases since this data is not available in the LISS panel. Consequently, my estimate of consumption may be somewhat too low compared to its real value.

Due to there being no reliable method available to measure the MPC at the household level, I stratify the households into different bins based on net annual household income, and then calculate the MPC of each bin. This process of binning households results in significantly larger error bands. Since I can only calculate the confidence interval for the redistribution elasticity on basis of the bins, this confidence interval ends up hardly differing from the actual estimate. Furthermore, Jappelli and Pistafferi (2010) argue that the MPC is higher for permanent changes in net income than for temporary changes. My measure of the MPC relies on the change in income between 2010 and 2015. Measuring the change in consumption over a longer period of time means that too many respondents flow out of the survey. If longer term data were available, it could potentially contribute to a more precise estimate of the MPC resulting from permanent changes in income.

Finally, it is important to consider that this thesis only examines one of the transmission channels of monetary policy. As such, the total effect of monetary policy may differ from the results presented in this thesis, as other channels may either strengthen the effect or work in the

(33)

26

opposite direction. In addition to this thesis, other studies that examine different transmission channels can provide a more complete view of the effect of monetary policy in the Netherlands.

(34)

27

Chapter VI: Conclusion

As my main result, I find evidence for the existence of a moderate redistributive interest rate channel of monetary policy in the Netherlands. After a decline in the policy rates, aggregate consumption of Dutch households increases through wealth redistribution and intertemporal substitution. I follow Auclert’s methodology to quantify households’ marginal propensities to consume and unhedged real interest rate exposures using the LISS panel. I find a negative correlation between the MPC and the URE which shows that households with a negative URE tend to have a higher MPC compared to households with a positive URE. Therefore, a fall in the real interest rate redistributes wealth away from low-MPC households with a positive URE to high-MPC households with a negative URE. Since the former group tends to fall on the higher part of the wealth distribution relative to the latter group, a fall in the policy rates increases aggregate consumption and decreases wealth inequality via the redistribution channel. In comparison to Auclert’s results, my estimates of the redistribution elasticity and the equivalent EIS are larger than in Italy but smaller than in the United States. Thus, there seems to exist a moderate redistribution channel in the Netherlands and the conduction of monetary policy should take redistributional effects into consideration.

Taking a look at the macroeconomic environment, real long-term interest rates are rather low in the Netherlands. Interest rates of government bonds maturing in ten years in the Netherlands were 0.670 in May 2018 according to the OECD (2017). This is quite low compared to interest rates of 2.180 in Italy and 2.980 in the United States. Moreover, during this time the three month money-market rate was -0.33 while the inflation rate measured by the consumer price index (CPI) was 1.74 percent. (OECD, 2017) There is not much room left for a decrease in policy rates in such a macroeconomic environment. While this may lead to the expectation that the interest rate channel is small, my results suggest that there exists a moderate redistribution channel in the Netherlands. Hence, a rise in the real interest rate can still affect the inequality of income and wealth of Dutch households, which should be taken into account when monetary policy is conducted.

(35)

28

Bibliography

Auclert, A. (2016): “Monetary Policy and the Redistribution Channel,” Working Paper, Stanford University.

CentERdata (n.d.): “About the Panel”. Available at https://www.lissdata.nl/about-panel. Coibion, O., Gorodnichenko, Y., Kueng, L., Silvia, J. (2012): “Innocent Bystanders? Monetary Policy and Inequality in the U.S.,” Working Paper 18170, National Bureau of Economic Research.

Doepke, M., Schneider, M. (2006): “Inflation and the Redistribution of Nominal Wealth,” Journal of Political Economy, 114 (6), 1069-1097.

Domanski, D., Scatigna, M., Zabai, A. (2016): “Wealth Inequality and Monetary Policy,” BIS Quarterly Review, March 2016, 45-64.

Inui, M., Sudo, N., Yamada, T. (2017): “Effects of Monetary Policy Shocks on Inequality in Japan,” Working Paper 17-E-3, Bank of Japan.

Jappelli, T., Pistaferri, L. (2010): “The Consumption Response to Income Changes,” Annual Review of Economics, 2 (1), 479-506.

Johnson, D. S., Parker, J. A., Souleles, N. S. (2004): “Household Expenditure and the Income Tax Rebates of 2001,” Working Paper 10784, National Bureau of Economic Research.

Kaplan, G., Violante, G. L. (2014): A Model of the Consumption Response to Fiscal Stimulus Payments,” Econometrica, 82 (4), 1199-1239.

OECD (2017a): “Household financial assets,” doi: 10.1787/7519b9dc-en (Accessed on 20 May 2018)

OECD (2017b): “Inflation (CPI) (indicator),” doi: 10.1787/eee82e6e-en (Accessed on 02 June 2018).

OECD (2017c): “Long-term interest rates (indicator),” doi: 10.1787/662d712c-en (Accessed on 02 June 2018).

OECD (2017d): “Short-term interest rates (indicator),” doi: 10.1787/2cc37d77-en (Accessed on 02 June 2018).

(36)

29

Saiki, A., Frost, J. (2014): “How Does Unconventional Monetary Policy Affect Inequality? Evidence from Japan,” Working Paper No. 423, De Nederlandsche Bank.

Teppa, F. (2014): “Consumption Behaviour and Financial Crisis in the Netherlands,” Working Paper 453, De Nederlandsche Bank.

(37)

30

Appendix A: Summary Statistics of Bins

Table A1: Summary Statistics of bins using LISS Panel data

Bin 1 Bin 2 Bin 3 Mean S.D Mean S.D Mean S.D Net annual household income Yi 17,329 .21 30,377 .17 54,059 3.44

Total annual household spending Ci 16,079 .52 21,450 .31 28,473 .55

Total household financial assets Ai 20,511 1.87 40,784 9.91 37,337 2.55

Unhedged real interest rate exposure UREi 21,761 1.89 49,711 9.90 62,923 4.39

Numbers of observations 317 317 317

Note: Normalized standard deviations are calculated by taking 𝑆𝐷𝐼( 𝑋_𝑖

(38)

31

Appendix B: Robustness Test

Table B1: MPCs out of Income (using Financial Wealth as Stratifying Variable)

Total Bin 1 Bin2

Net annual household income (logs) .246*** .298*** .153**

Year -.004*** -.005*** -.003**

Female .028 .057 -.006

Number of household members .013 .016 .024

Partner -.090** -.141** -.057

Education -.010 -.020 -.003

Constant .318*** .371*** .213

Implied MPC .16*** .20*** .10**

Number of observations 951 476 475

***Significance at the 1 percent level. **Significance at the 5 percent level. Figure B1: Marginal Propensity to Consume and the Redistribution Channel

(39)

32

Table B2: Estimates of 𝜀̂_𝑟 and 𝜎̂ using LISS Panel data _𝑟

Parameter Mean Estimate 95% CI

Redistribution Elasticity 𝜀̂_𝑟 -.140 [-.140, -.140]

Scaling Factor 𝑆̂ .853 [.781, .925]

EIS 𝜎_𝑟= −𝜀𝑟

𝑆 .164 [.151, .179]

Redistributional effects of monetary policy in the Netherlands : real interest rate exposures of Dutch households