Distributional Effects of Monetary Policy: The Interest Rate Exposure of German Households

(1)

Distributional Effects of Monetary Policy:

The Interest Rate Exposure

of German Households

Master Thesis

Faculty of Economics and Business University of Amsterdam Supervisor: Dr. Christian A. Stoltenberg Submitted by: Elisa Vandy Student No. 11376724 M.Sc. Economics

(2)

(3)

ABSTRACT

This thesis examines how expansionary monetary policy affects aggregate consumption in Germany via wealth redistribution. I focus on changes in the real interest rate and measure how household balance sheets are affected. My results suggest that households that are hurt by lower real interest rates exhibit small consumption responses relative to households that benefit. This redistribution of wealth to agents with high marginal propensities to consume increases aggregate spending. In comparison to Italy and the U.S., my estimates point to a high exposure of households towards interest rate changes and suggest a large redistribution channel in Germany.

(4)

LIST OF ABBREVIATIONS

ARM Adjustable-Rate Mortgage

CEX Consumer Expenditure Survey of the United States EIS Elasticity of Intertemporal Substitution

FKP Financially Knowledgeable Person FRM Fixed-Rate Mortgage

HFCS European Household Finance and Consumption Survey MPC Marginal Propensity to Consume

NNP Net Nominal Positions

PHF German Panel on Household Finances

SHIW Italian Survey on Household Income and Wealth UMP Unconventional Monetary Policy

(6)

LIST OF TABLES

Table 1: Components of the Unhedged Real Interest Rate Exposure ... 13

Table 2: Summary statistics using the PHF data ... 17

Table 3: Correlations of URE with MPC and demographic variables ... 19

(7)

LIST OF FIGURES

Figure 1: Five transmission channels and their effect on aggregate consumption according to

Auclert (2016) ... 7

Figure 2: Balances in savings accounts in the PHF sample ... 18

Figure 3: Correlation between URE and MPC in the PHF population ... 20

Figure 4: The redistribution elasticity using the PHF sample ... 21

(8)

CHAPTER I: Introduction

The main goal of a central bank is to maintain price stability to stimulate economic growth and employment. It does so by determining interest rates, providing liquidity and surveilling the financial sector. Reducing inequality is no traditional target of central banks and the redistribution of income and wealth is no major concern for monetary policymakers. Recent studies such as Coibion, Gorodnichenko, Kueng and Silvia (2012) and Auclert (2016) show that monetary policy has an impact on inequality through conventional and unconventional measures. Thus, when deciding about monetary policy measures, central banks should consider the intentional or unintentional distributional implications of their policy choices.

This thesis analyses the complexity of effects that monetary policy may have on the distribution of wealth and income in Germany. Hereby, I focus on monetary policy that targets at changes in the real interest rate and examine how it affects aggregate spending of the household sector. By looking at the composition of household balance sheets, I measure the extent to which they are exposed to falling or rising policy rates, meaning how their wealth is affected. Since the European Central Bank has implemented a complex set of expansionary monetary policy measures since the Global Financial Crisis, I primarily concentrate on the effect of falling interest rates in Germany.

To capture the response of aggregate spending, I quantify each household’s marginal propensity to consume (MPC), in other words how their consumption is affected by a change in income. Eventually, my aim is to find a relation between households’ exposures towards changes in the interest rate and their consumption response. Overall consumption rises when those who gain from real interest rate falls exhibit higher MPC than households that are hurt by such changes. My contribution to research is to examine this specific transmission channel in Germany, which complements recent studies by Auclert (2016) on Italian and U.S. households. To my knowledge, this application to German household data is novel and sheds further light on the variety of distributional impacts of monetary policy.

Introducing more details about my analysis, I apply Auclert’s (2016) methodology in “Monetary Policy and the Redistribution Channel” to German household data. In his working paper Auclert

(9)

(2016) analyses the effects of real interest rate changes on aggregate consumption in Italy and the United States. His argument is that aggregate spending increases because those who benefit from real interest rate falls typically have higher MPC, such as net borrowers.

To quantify which parts of the balance sheet are affected by changing real interest rates, I define so-called unhedged real interest rate exposures (URE). This includes measuring the difference between all maturing assets and liabilities at a certain point in time. Auclert (2016) argues that households with a high amount of short-term assets, such as demand deposits, tend to have positive URE, whereas investors of long-term bonds and adjustable-rate mortgage holders exhibit negative URE. Using data from the German Panel on Household Finances (PHF), I measure the correlation between MPC and URE in the PHF population and analyse the redistributive component of monetary policy. In line with Auclert’s (2016) research, aggregate consumption responds to declines in the real interest rate via redistribution and intertemporal substitution, when it becomes cheaper to consume today instead of tomorrow. To assess the redistribution component, I quantify a redistribution elasticity and an elasticity of intertemporal substitution (EIS). This EIS is set up in a way that it captures the total response of aggregate consumption including the redistribution component.

My results suggest a small but significant negative correlation of URE with MPC. This supports Auclert’s (2016) findings that real interest rate falls distribute away from low MPC-agents with a positive exposure to households with high MPC and negative interest rate exposures. In comparison with Auclert’s (2016) estimates, I provide empirical evidence that the redistribution channel in Germany is very large relative to the United States and Italy. This indicates that German households are more exposed to changes in the real interest rate than other countries. Consequently, my empirical evidence supports the view that redistributional implications of monetary policy should be cautiously taken into account in Germany.

My thesis is structured as follows. In chapter 2, I provide a theoretical outline and an overview of recent empirical findings on the redistribution component of monetary policy. Then, I present my methodology and adjustment to Auclert’s (2016) empirical approach in chapter 3. Chapter 4 contains the results of my empirical analysis and in chapter 5 I evaluate limitations and potential measurement errors. Chapter 6 concludes.

(10)

CHAPTER II: Literature Overview

This section examines recent research on the distributional effects of monetary policy. I present different channels through which monetary policy can affect inequality and compare various methodologies on the quantification of these channels. Eventually, I show that my research on the distributional effects of interest rate changes in Germany is novel and complements many recent empirical studies.

The effect of conventional and unconventional monetary policy is a priori ambiguous. This is the conclusion that many researchers draw from analysing how monetary policy may have an impact on economic inequality (Bundesbank (2016), Coibon et. al. (2012), Bank of England (2012), Bernoth et. al. (2016)). According to their studies this can be traced back to the variety of transmission channels through which the income and wealth distribution of the household sector is affected. In theory, five main transmission channels have been identified (Bundesbank (2016), Coibon et. al. (2012), DIW (2016)):

(a) Income Composition Channel: The variety of income sources such as business, financial and labour income shapes the extent to which households are affected by monetary policy. When asset prices increase and yields decline as a result of central banks’ asset purchase programmes, the income of households with a large share of financial assets falls. Business income on the other hand tends to rise in such a scenario. Households whose primary source of income is their labour earnings tend to be less concerned in the short-run. With a time lag however, these households may be affected as well via developments in the labour market such as changes in unemployment.

(b) Financial Segmentation Channel: The degree to which households participate in the financial market and the amount of shares they hold determines the extent to which they can benefit from expansionary monetary policy. These households tend to be situated on the upper part of the wealth and income distribution.

(c) Savings Redistribution Channel: In theoretical literature, a widely stated view is that changes in inflation or the interest rate affect net borrowers and savers heterogeneously. Whereas net borrowers benefit from higher inflation and lower policy rates because of

(11)

falling debt repayments, net savers are hurt because the real value of their savings declines. Thus, when central banks take measures to lower the long-term real interest rate, they redistribute away from savers to borrowers.

(d) Labour Earnings Heterogeneity Channel: When monetary policymakers help to restore the financial market after a recession, it likely affects the labour market via decreasing unemployment rates and rising labour demand in the medium- and long-run. Such a development tends to benefit low-income households more than high-income households as they are more affected by changes in unemployment. On top of that, heterogeneous effects on wages can be another outcome depending on the wage rigidity across sectors and employments.

(e) Portfolio Channel: Similar to the savings redistribution channel, households with a high fraction of assets that are not protected against inflation, such as demand deposits or cash holdings, are hurt by higher inflation rates. Since high-income households tend to have more diversified portfolios, higher inflation rates are more harmful to households with less accumulated wealth.

To shed more light on the relation between monetary policy and redistribution, empirical research has given much devotion to analyse the impact of rising asset prices. In this field of research, Saiki and Frost (2014) looked at the impact of unconventional monetary policy (UMP) in Japan and found evidence for higher inequality via the income composition (a) and the financial segmentation channel (b). By applying a vector auto regression model their results support the view that loose monetary policy mainly benefits wealthy households who actively participate in the financial market and invest their savings in equities. These households benefit from a rising business income that contributes to a larger income gap in Japan. Nonetheless, it has to be borne in mind that Saiki and Frost (2016) only focus on partial effects of UMP and disregard the savings redistribution and earnings heterogeneity channel that can have a diminishing effect on inequality in the medium- and long-run.

Similar to Saiki and Frost (2014), the Bank of England (2012) and Bernoth et. al. (2016) emphasise that the rise in asset prices is the main distributional effect caused by UMP in the

(12)

short-run. However, they point out that in the long-run UMP is meant to stimulate economic recovery, employment and higher inflation. Thus, the overall net effect of UMP cannot be specified ex ante as the above named channels have ambiguous effects on the households’ income distribution.

A different perspective is taken by Coibon et. al. (2012) who analyse the long-term effects of contractionary monetary policy in the United States since 1980. In their empirical analysis they conclude that contractionary monetary policy has contributed to consumption and income inequality through the labour earnings heterogeneity channel (d). Although their lack of detailed data on household balance sheets did not allow them to quantify the portfolio, savings redistribution and financial segmentation channel, their data suggests that these effects are comparatively small. The research of Domanski et. al. (2016) is in line with Coibon et. al. (2012) in so far that they agree on the negligibility of the distributional effects of low interest rates. However, with respect to the main driving factors of inequality their evaluations point to different transmission channels. In contrast to Coibon et. al. (2012), Domanski et. al. (2016) see rising equity prices as the main contribution of central banks to more inequality in advanced economies. Despite their empirical findings, they suggest that these households may have become more sensitive to changes in the real interest rate in recent years because their balance sheets have been growing at higher rates than the gross domestic product. Besides, they point out that the effect of continuingly low policy rates tends to have a stronger effect on wealth inequality in Germany because of the lowest quintile holding more than 50 per cent of their assets in fixed income deposits.

Different approaches to measuring the distributional effects of monetary policy are taken by Doepke and Schneider (2006) and Auclert (2016). Both analyse the savings redistribution channel (c) and find evidence for expansionary monetary policy contributing to less wealth inequality in the United States. While Doepke and Schneider (2006) evaluate the effect of higher inflation on savings redistribution, Auclert (2016) focuses on the impact of changes in the real interest rate. By quantitatively assessing the exposure of households’ balance sheets to changes in inflation and the real interest rate, both conclude that higher inflation and lower real interest rates redistribute wealth away from rich, net lending households to young, middle-class households with mortgage debt.

(13)

Concluding, there are various different channels through which monetary policy is transmitted to the real economy. Although much research has been done in this field since the Global Financial Crisis, only few analyse the interconnectedness of transmission channels. Thus, they only account for partial effects and give an incomplete picture of the relation between monetary policy and wealth inequality. Since most recent papers focus on the effects of higher asset prices induced by monetary policy, my analysis of the savings redistribution channel is together with Doepke and Schneider (2006) and Auclert (2016) a relatively new field of research. My contribution to research consists of analysing the exposure of households to changes in the real interest rate in Germany, thus applying Auclert’s (2016) empirical work on U.S. and Italian households to German households. To my knowledge, my investigation is new to the extent that there is no other empirical work on quantifying the savings redistribution channel in Germany.

(14)

CHAPTER III: METHODOLOGY

In this chapter I present my methodological approach to analysing the distributional effects of monetary policy via changes in the real interest rate in Germany. I begin with a presentation of Auclert’s (2016) empirical work, which I partly replicate and apply to German household data. Then, I give an outline of the dataset that I use for my analysis and introduce the different steps to measuring the exposure of households to changes in the interest rate.

3.1 Auclert’s (2016) Analysis of the Redistribution Channel

In his working paper “Monetary Policy and the Redistribution Channel” Auclert (2016) assesses the role of the transmission channels of monetary policy and their effect on aggregate consumption. He argues that households that benefit from monetary policy exhibit higher marginal propensities to consume (MPC) on average. Following this line of reasoning, aggregate consumption would increase, if monetary policy contributed to a redistribution of wealth away from low-MPC agents to high-MPC agents. Auclert (2016) summarises this by stating that redistribution is a channel in itself through which monetary policy has an effect on macroeconomic aggregates. !" = !! !_! !!"#! !" !""#$"%&$ !"#$%& !!!""#$ + !"#! !"#!, !!! − !! !" ! − !"#$%$&' !!"!#$%!&!'"( !!!""#$ !"#! !"#!, !!"! !" ! !"#!!" !!!""#$ + !"#_! !"#_!, !"#_! !"#$%$&# !"#$ !"#$%&'! !!!""#$ − !_! !_! 1 − !"#_! !_! !"#$%&%"%&'( !!!""#$ !" !

Figure 1: Five transmission channels and their effect on aggregate consumption according to Auclert (2016)

Figure 1 shows the channels that Auclert (2016) uses for defining the first-order response of aggregate consumption to macroeconomic shocks in form of a change in monetary policy. There are many parallels but also distinctive differences to the transmissions channels presented by the Bundesbank (2016) that I outlined in chapter 2. Most importantly, the Bundesbank (2016)

(15)

focuses on transmission channels that potentially affect inequality whereas Auclert (2016) distinguishes between aggregate and redistributive components. By quantifying the sum of these components, Auclert (2016) determines the effect of monetary policy on overall spending. In more detail, Auclert’s (2016) channels can be compared to the Bundesbank (2016) as follows:

(i) Aggregate income channel: In contrast to the Bundesbank’s (2016) channels outlined in section 2, an aggregate income effect plays a role in Auclert’s (2016) analysis. This effect is non-redistributive and does not depend on the individual source of income such as the income composition channel (a) defined by the Bundesbank (2016). Instead it displays the impact of monetary policy on aggregate spending through changes in total income across sectors.

(ii) Earnings heterogeneity channel: This channel describes how monetary policy influences aggregate consumption by affecting households’ individual incomes differently. It presents a redistributive component of monetary policy and can be regarded as a combination of the Bundesbank’s (2016) income composition (a) and the labour earnings heterogeneity channel (d).

(iii) Fisher channel: This redistributive transmission channel of monetary policy shows how individual households are affected by changes in inflation. Heterogeneous effects are measured by analysing their so-called net nominal positions (NNP), meaning the exposures of household balance sheets towards unexpected inflation changes. The NNP are calculated by subtracting the market value of nominal liabilities from the market value of all nominal assets. Recalling the Bundesbank’s (2016) outline, the Fisher channel is captured by the financial segmentation (b), the savings redistribution (c) and the portfolio channel (e). Hereby, the Bundesbank (2016) argues that inflation affects households through their participation in the financial market (a), whether they are net lenders or borrowers (c) and the extent to which their portfolios are diversified (e).

(iv) Interest rate exposure channel: Next to the Fisher channel, the interest rate exposure channel presents a redistributive component of monetary policy. It examines how household balance sheets are exposed to changes in the real interest rate and measures

(16)

heterogeneous effects across the household sector. Similar to unexpected inflation, a change in the real interest rate is captured by the financial segmentation (b), the savings redistribution (c) and the portfolio channel (e). Thus, these three channels can be regarded as a combination of Auclert’s (2016) Fisher and the interest rate exposure channel. (v) Substitution channel: The intertemporal substitution channel describes how households

shift their consumption across periods when the real interest rate changes. It presents the aggregate household response to changing opportunity costs of consumption and belongs to the non-redistributive effects of monetary policy. Consequently, this channel is not considered by the Bundesbank’s (2016) examination of distributive transmission channels.

As my methodological approach is based on Auclert’s (2016) analysis of the interest rate channel, I describe its mechanism in more detail. The interest rate exposure channel explains how aggregate consumption is affected by the households’ exposure to unexpected changes in the real interest rate. For this analysis, cross-sectional so-called unhedged real interest rate exposures (URE) are quantified by taking the difference between all maturing assets and liabilities at a certain point in time. Auclert (2016) argues that households who mainly invest their wealth in short-term assets such as savings deposits tend to have positive URE, whereas holders of adjustable-rate mortgages and agents with assets of longer maturities tend to have negative URE. His assumption is that the households that exhibit positive URE are likely to have lower marginal propensities to consume than households with negative URE that have their wealth stored in long-term investments. Consequently, the cross-sectional covariances between MPC and URE are assumed to be negative.

According to Japelli and Pistaferri (2010) high-MPC agents tend to be liquidity constrained with little household resources. They consume a high fraction of any additional income and increase their consumption more than low-MPC agents in case of income gains. Thus, overall consumption would rise, when interest rate falls redistributed away from households with low MPC and positive URE to high-MPC households with a negative exposure.

(17)

In addition to the interest rate exposure channel, the substitution channel captures another response of households’ consumption to changes in the interest rate. Households increase their consumption today relative to the future when the opportunity cost of consuming today falls, because of lower returns on savings for example. This substitution effect depends on the households’ elasticity of intertemporal substitution (EIS). The EIS is referred to as !_! in figure 1. With negative covariances between MPC and URE, the interest rate exposure and the substitution channel both move in the same direction when the real interest rate changes. Consequently, Auclert’s (2016) model predicts that aggregate consumption increases when the real interest rate falls.

In my research, I will quantify the interest rate exposure and the substitution channel using German household data. My aim is to quantitatively compare both channels and evaluate the extent to which a fall in the real interest rate increases aggregate consumption via redistribution.

3.2 The German Panel on Household Finance (PHF)

This section gives an overview of the German Panel on Household Finance (PHF) that I use for quantifying the interest rate exposure of households in Germany. The PHF study is a panel survey on household finance and wealth in Germany that is conducted by the German Bundesbank and forms part of the European Household Finance and Consumption Survey (HFCS). The PHF data covers balance sheet characteristics, consumption patterns and information about income. It consists of a first wave that was conducted in 2010-2011 for a net sample of more than 3,500 randomly selected households, and a second wave in 2014 for a sample of 4,500 households.

Since wealthy households are deliberately oversampled, the PHF data gives an extensive insight into the wealth situation of German households and offers a detailed account on their asset and liability positions. Overall the interview is conducted with all household members who are aged above 16 years. Exceptions are sections about consumption, assets and liabilities that are only answered by the financially knowledgeable person (FKP) of the household. In my research I work with answers of the FKP of each household that participated in both waves. This leaves me with a smaller sample of 2,138 observations. Although the survey offers a detailed account on

(18)

income and employment characteristics on the individual level, I only work with aggregated household estimates obtained from the interview with the FKP. By only using the answers of the FKP, my aim is to maintain a high level of comparability of household consumption and income patterns.

With respect to the format, all data variables that ask for specific time periods have been converted to yearly values. Additionally, all nominal interest rates provided were converted into effective interest rates for higher comparability. Logical consistency checks and structural edits such as currency conversions were applied by the Bundesbank. The data quality is characterised by a low item non-response rate and a high unit non-response rate especially for questions directed at the individual (see Kalckreuth et. al. 2012). This high unit non-response underlines my necessity to work with aggregated household data reported by the FKP to maintain consistent estimates.

3.3 Marginal Propensities to Consume (MPC)

As drafted in section 3.1, Auclert (2016) analyses covariances of URE with MPC for evaluating the redistributive component of the interest rate channel. In this section, I offer a theoretical outline of the MPC and present my methodological approach to quantify it in the PHF data. In the following section 3.4, I give an account of Auclert’s (2016) measurement of the URE and show my adjustments to calculate the URE of the PHF population.

In theory the marginal propensity to consume measures the fraction of an unexpected rise in income that is used for consumption. Its purpose is to predict how well income transfers to the household sector cause aggregate spending to increase (see Johnson et. al., 2004). Expressed mathematically the MPC is defined as the change in consumption over the change in disposable income or as the derivative of aggregate consumption with respect to income:

!"#_! = !!! !!_!

Hereby !_! presents the consumption of each individual and !_! stands for disposable income. Since it is hard to obtain precise estimates of the MPC on the individual level, aggregates and

(19)

averages are often used in economic research (see Jappelli and Pistaferri, 2010). For my analysis however, it is crucial to capture the MPC at the cross-sectional level to account for the heterogeneity of consumption responses to income shocks. Instead of using aggregate consumption for calculating the MPC, I use the sum of the amount spent on food, meals and drinks inside and outside home as a proxy. Reason for this is the lack of data on aggregate consumption in the first wave of the PHF survey. As a measure of income I use an estimate of the monthly net disposable household income, which I convert into annual values. Having acquired these annual values for both survey waves, I measure the MPC as follows:

!"#_! =!"!

!"#$_{− !"} !!"#" !_!!"#$_{− !}

!!"#"

Hereby !"_! stands for the food consumption of each household i. Although the comparison of changes in food consumption and disposable income over both waves is merely an approximation of the MPC, I expect it to show the same trend as the MPC used in Auclert’s (2016) empirical work.

3.4 Quantifying the Unhedged Real Interest Rate Exposure (URE)

In Auclert’s (2016) analysis the unhedged real interest rate exposure is a method to quantify the present households’ net saving requirement from the point of view of yesterday. As outlined in section 3.1, the !"#_! measure the present difference between all maturing assets and liabilities, thus accounting for the composition of households’ balance sheets. Auclert (2016) measures the !"#! as follows:

!"#_! = !_! − !_! + !_! − !_!

Above !_! presents the total income and !_! includes consumption of durable goods and fixed-rate mortgage payments but does not capture house purchases. !! and !! present all assets and liabilities that mature over the year in which the !"!_! is measured. It is important to note that !_! captures all of the households’ savings deposits since they are typically withdrawable on demand and maturing over one year. Besides that, Auclert (2016) assumes an average duration of four years for financial assets in the U.S. and counts one-fourth of these assets to his annual

(20)

measurement of !!. Adjustable-rate mortgages (ARM) are treated as maturing over the year and counted towards !_!. Fixed-rate mortgages (FRM) on the other hand tend to have longer durations, so that they are not counted towards !_!. However, payments on FRM are captured in !! and form part of the consumption expenditures.

Table 1 illustrates how I apply Auclert’s (2016) construction of the URE for U.S. and Italian household data to the PHF dataset on German household finance. Although both waves of the PHF survey have a high comparability, the second wave has a larger question catalogue and offers a more detailed account on household consumption. Hence, I use the PHF data from 2014 for the construction of the URE.

Table 1: Components of the Unhedged Real Interest Rate Exposure, !"#! =

!_! −!_! +!_! −!_! Net disposable household income (annual) Total household expenditures excluding major purchases + Rental costs + Payments on FRM + Regular donations + Instalment payments for leased cars +

Repayments of student loan

Demand deposits + Savings accounts + Balances on all credit cards + (1/4)* Market value of all certificates + (1/4)* Market value of bonds + (1/4)* Value of publicly traded stocks Outstanding balance of all ARM + (1/4)* Outstanding credit line and credit card debt +

(1/5)* Uncollateralised loans +

(1/5)* Amount out-standing for unpaid bills

Note: Medians and means are only measured for the sub-sample of households that participated in both waves of the PHF survey.

When measuring the average duration of financial assets in the PHF data, I obtain mean values of four to five years that are in line with Auclert’s (2016) research. Counting one-fourth of financial assets in my measurement of !_! should give an approximation of all assets maturing over the year. For all maturing liabilities I count one-fourth or one-fifth of consumer credit

(21)

towards !! depending on my calculations of mean maturities in the PHF sample. In contrast to Auclert (2016), I construct my URE solely at an annual level because the majority of my data are measured in this time unit. It is important to note that neither non-financial assets, such as real estate, nor long-term investments are listed in !_! since they tend to have no or very long maturities. Student loans and instalment payments for leased cars are included in !_! because they typically consist of fixed-rate payments that are seen as regular expenses in my measures.

3.5 Measuring the Interest Rate Channel

Section 3.5 summarises Auclert’s (2016) measurement of the interest rate channel after having obtained precise estimates of MPC and URE. I present the extent to which I apply his approach to assess the distributional effects of interest rate changes in Germany.

To evaluate how aggregate consumption responds to changes in the real interest rate, Auclert (2016) makes several simplifications in his empirical study. Importantly, he focuses on a transitory change in the policy rate and assumes that all households have separable preferences. He finds that the partial elasticity of aggregate consumption to the real interest rate is given by

!"#! !"#!, !"#_! !_! !_! !! − ! !! 1 − !"#! !_! !_! !_! !

where !_! represents the redistribution elasticity, S is a scaling factor and ! is defined as the weighted average of the elasticity of intertemporal substitution (EIS):

! ∶= !_! !_! 1 − !"#! !! !_! 1 − !"#_! !_!

Since the EIS ! cannot be measured empirically at household level, Auclert (2016) provides a proxy !_! which assumes that the redistribution and substitution channel are equal in size:

!!∶= −!_!

! =

−!"#_! !"#_!, !"#_! !_! 1 − !"#_! !_!

(22)

Furthermore, !! is set up in a way to estimate the extent to which the elasticity of a representative agent ! differs from the heterogeneous elasticities measured empirically. In other words, !_! shows by how much ! from the representative agent economy has to be increased to account for the total response of aggregate spending including the redistributive component. Thus, a high value of the EIS !! suggests a large redistributive interest rate exposure channel. Although Auclert (2016) concedes that this approach is merely an approximation of the more complex redistribution channel, he is convinced of its practicability.

In my measurement of the redistribution channel, I quantify how aggregate consumption responds to a change in the real interest rate in line with Auclert (2016). It is important to note that the covariances between the marginal propensities to consume !"#! and the unhedged real interest rate exposures !"#_! are assumed to be negative when the redistribution channel is large. In this case, monetary policy that lowers the policy rate redistributes away from households with positive URE and low MPC to negative-URE agents with high MPC.

To scale the covariances, the !"#_! are divided by the consumption mean. Besides, I create dummy variables of several demographic characteristics for which I control when measuring the correlation between !"#_! and !"#_!. After having obtained these cross-sectional correlations I estimate the redistribution elasticity !_! and the equivalent elasticity of intertemporal substitution !_! using the PHF data. I then compare these values to the empirical findings of Auclert (2016) and evaluate the extent to which the redistribution channel plays a role in Germany when the real interest rate falls.

(23)

CHAPTER IV: EMPIRICAL FINDINGS

This chapter deals with the results of my empirical analysis using the PHF data. In section 4.1, I present my estimates for the marginal propensity to consume and the unhedged real interest rate exposure. I show my empirical evidence for the existence of a redistribution channel in the PHF population in section 4.2 and in section 4.3 I evaluate how monetary policy affects the wealth distribution in Germany via the interest rate channel.

4.1 Quantification of URE and MPC in the PHF population

Using the data obtained from the first and second wave of the PHF survey, I construct cross-sectional !"#_!. As described in more detail in section 3.3, I quantify the MPC by dividing the difference in food consumption by the difference in income over the period of both survey waves. In my results I find large MPC heterogeneity with negative outliers below -1 in the lowest percentiles. A possible reason for this empirical finding is the approximate measure of food expenditures in the PHF survey. The time span between the first and second wave is long and the panel households may have reported different values for their food expenditures even if there food consumption did not change in real terms. On top of that, food expenditures are only a fraction of the consumption measurement. Thus, a change in income may increase consumption through higher spending on durable goods but leaves food consumption unchanged. Besides, my construction of the MPC does not differentiate between transitory and permanent changes in income. Consequently, I cannot identify exogenous and unexpected changes in income, which according to Jappelli and Pistaferri (2010) increases the exactness of MPC measurements. When more micro data is collected in the upcoming waves of the PHF survey, changes in income and food consumption over time can be evaluated more precisely. For my measurement of the interest rate channel, I restrict the number of outliers in the !"#! to positive values. This leaves me with a sample of 1,384 households and a mean value of 0.49 for the MPC that is similar to Auclert’s (2016) MPC mean of 0.47. The latter mean has been measured on the basis of households’ self-reported consumption responses to a positive income transfer, thus capturing exogenous and transitory income changes.

(24)

Table 2: Summary statistics using the PHF data

Note: The normalised standard deviation is !"_! !!

!_!!_! for !!= !!, !!, !!, !!, !"#! and !"! !"#! for MPC.

Table 2 provides summary statistics of my URE and MPC measurements in the PHF population. Notably, the distribution of household income !_! and the sum of maturing liabilities !_! are similar to Auclert’s (2016) results from the SHIW survey on Italian household data with means of 36,114€ for !_! and 6,228€ for !_!. However, consumption !_! in the PHF population is nearly one third lower than the mean consumption in the SHIW sample. One potential reason is that the data on total expenditures in the PHF survey explicitly excludes major purchases of durable goods. On top of that, the interpretation of total spending may vary across households, which potentially leads to more heterogeneity in my consumption measures.

Outstandingly large are my estimates for the maturing assets !_! whose mean is more than three times as large as Auclert’s (2016) measure of !_!. This difference does not only raise the question of the inclusion of similar variables but also needs further investigation on the size of the !_! in the PHF sample. When analysing the maturing assets in the PHF data, I found the reason for the large !_! in the amount of savings deposits. This is in line with the Bundesbank (2016) arguing that current accounts and savings deposits were the most widespread financial assets in 2014 when the second wave of the PHF survey was conducted.

Variable Mean (€) Normalised SD

Estimated Household Income (!!, per year) 37,987 1.37

Consumption Including Mortgage Payments (!_!, per

year) 19,942 0.67

Deposits and Maturing Assets (!!) 43,485 5.95

ARM Mortgage Liabilities and Consumer Credit (!_!) 6,048 3.47 Unhedged Real Interest Rate Exposure (!"#!, per year) 55,482 5.68

Marginal Propensity to Consume 0.49 1.10

(25)

Note: Values are restricted to the sub-sample that is considered for the measurement of the MPC and URE.

Figure 2: Balances in savings accounts in the PHF sample

Figure 2 summarises how balances in savings accounts were distributed in the PHF sample in 2014. Since savings deposits account for more than 90 per cent of my mean !_!, I assume that they are the main determinant of all maturing assets in the PHF population. Besides, it needs to be borne in mind that wealthy households are oversampled in the PHF data, which potentially increases my mean estimate for !_!. To adjust for this oversampling, I provide another measurement of the !"#! in appendix A that excludes the top per cent of !! and !!. Eventually, both approaches present large estimates of the mean URE that are more than a triple of Auclert’s (2016) corresponding !"#_! estimates of 16,110€.

The attempt to lower my measures of !"#_! by excluding savings deposits from !_! leads to implausible correlation coefficients of !"#! with !"#!. On top of that, savings deposits present an important part of the financial assets that mature over the period of one year in Germany. Consequently, despite my large values of !_!, I do not exclude savings deposits from my measurement of the interest rate exposure of households in Germany.

300 1 000 3 600 13 000 40 000 100 000 180 000 20 000 40 000 60 000 80 000 100 000 120 000 140 000 160 000 180 000 200 000 5% 10% 25% 50% 75% 90% 95% B al an ce s in S av in gs A cc ou n ts (€ ) Percentiles of Savings Balances

(26)

4.2 Estimating the Interest Rate Exposure of the PHF population

In this section I present my estimates for the correlation of URE with MPC in the PHF population in Germany. I show that my correlation coefficients are small but significantly negative, suggesting that a fall in the real interest rate distributes away from households with positive URE and low MPC to households that exhibit high MPC and negative exposures. Table 3 summarises my cross-sectional correlation results of !"#_! with !"#_! and several demographic characteristics of the financially knowledgeable person (FKP) of each household. The partial correlation coefficients present the correlation between two specified variables after removing the effects of all the other variables remaining in the list. In this way, the partial correlation of URE with MPC controls for all the demographic characteristics listed in table 3. In line with Auclert (2016), my estimates point to a negative correlation of !"#_! with !"#_! that is significant at the five per cent level. Even the semipartial correlation estimates, which only control for demographic variables in the !"#_! but not in !"#_!, show a similarly negative estimate. Since demographic characteristics potentially affect both the MPC and the URE, I use the partial correlation estimates for the quantification of the interest rate channel in section 4.3.

Table 3: Correlations of URE with MPC and demographic variables

Variable Partial Corr. Semipartial Corr. Significance Value

MPC -0.0583** -0.0560 0.0303 Year of Birth -0.1666*** -0.1620 0.0000 Female -0.0561** -0.0539 0.0372 Married 0.0889*** 0.0856 0.0010 Born in Germany 0.0461* 0.0442 0.0875 High level of education 0.1570*** 0.1525 0.0000 Employed 0.0284 0.0272 0.2921

Note: High level of education is defined as having obtained a higher education entrance qualification. ***Significance at the 1 per cent level. **Significance at the 5 per cent level.

(27)

When analysing the demographic variables in table 3, my estimates suggest a highly significant correlation of URE with age, gender, marital status and the level of education. Although the correlation coefficients are insufficient to assume any causal relation, my results point to a systematic relation between demographics and the URE. Hereby, the URE tends to increase with age and the level of education, and it is higher for households whose financially knowledgeable person is male and married. This finding suggests that old, educated and married households tend to have large and positive interest rate exposures, whereas the URE is small for young households with a low level of education. Auclert (2016) and Doepke and Schneider (2006) point out that young, middle-class households with mortgage debt are the main net nominal borrowers in the economy, while elderly and rich households are net lenders on average. This is in line with my correlation coefficients because net borrowers tend to have negative URE and net lenders are likely to exhibit positive URE. Besides that, it is crucial to note that the large estimates of !_! are the main reason for the size of my estimates of !"#_!. Thus, a significant positive correlation of URE with age, marriage and education also points to a positive correlation with the size of savings deposits. An overestimation of the significance of my correlation estimates can thus not be excluded.

Note: The normalised URE is !"#!

!!!!.

Figure 3: Correlation between URE and MPC in the PHF population

0 .1 .2 .3 .4 .5 Ma rg in a l Pro p e n si ty to C o n su me -1 0 1 2 3 4 Normalised URE

(28)

Figure 3 presents the partial correlation between URE with MPC for a sub-sample of the PHF population that excludes large outliers. Although the correlation coefficient remains negative in the reduced sample size, there is large heterogeneity in URE and MPC. This empirical correlation is less negative than Auclert’s (2016) estimate of -0.09 using Italian household data. Therefore, even if declines in the real interest rate distributed away from positive- to negative-URE agents, my small correlation estimates suggest that the effect on aggregate consumption would be lower than in Auclert’s (2016) household sample.

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

Germany

This section evaluates my results for the interest rate exposure channel and offers an interpretation of the role of monetary policy in Germany. For a better understanding of the redistribution channel and its effect on aggregate consumption, I compare my results to Auclert’s (2016) research on Italian and U.S. households.

My estimates for the components of the redistribution elasticity are shown in figure 4. As described in section 4.2, my correlation coefficient of about -0.06 is significant, but remains about one third lower than Auclert’s (2016) estimate. Both standard deviations of MPC and URE are large and suggest a lot of heterogeneity in the data.

!_! = !"##_! !"#_!,!"#! !_! !_! !!.!" !"_! !"#_! !.!" !"_! !"#! !_! !_! !.!" Figure 4: The redistribution elasticity using the PHF sample

Measuring the redistribution elasticity, !! is about -0.37 in the PHF sample. Despite the smaller magnitude of my correlation coefficient, Auclert (2016) does not measure a larger redistribution channel in his data. Instead his results show a redistribution elasticity of -0.06 for Italian households and -0.24 for the United States, which are both smaller in magnitude than my estimate of !_!. Also the exclusion of large outliers, presented in appendix A, suggests a large redistribution component. Looking at the decomposition of my !_!, the size can be explained by the large heterogeneity in MPC and URE. This increases the necessity to carefully assess my

(29)

findings and consider potential measurement errors in MPC and URE when evaluating the interest rate channel.

When the real long-term rate falls as a result of expansionary monetary policy, theory suggests that it lowers the return on savings so that households increase their consumption today relative to the future (Hall, 1988). This substitution channel forms an important part of the effect of interest rate changes on aggregate consumption. As outlined in section 3.5, Auclert (2016) assumes in his measurement that the substitution and the redistribution channel are equally large. He defines an elasticity of intertemporal substitution !_! that displays by how much to increase the EIS ! from the representative agent economy to capture the total response of aggregate consumption that includes the redistributive component.

Table 4 and figure 5 summarise and relate my results for the EIS !_! and the redistribution elasticity !_! to Auclert’s (2016) estimates. Whereas my estimate of the EIS !_! is 0.67 in the PHF population, Auclert (2016) measures an EIS of 0.30 for the U.S. and 0.12 for Italian household data. Hence, assuming that the EIS ! in the representative agent economy is similar in these nations, my findings suggest a large redistribution channel in Germany.

Table 4: Estimates of !_! and !_! using the PHF survey

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

Parameter Mean Estimate 95% CI

Redistribution Elasticity !_! -0.37 [-0.56, -0.21] Scaling Factor ! 0.55 [0.49, 0.60] Equivalent EIS !! = − !_! ! 0.67 [0.44, 0.92]

(30)

Note: Mean estimates are calculated on the basis of the PHF survey for Germany, 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).

Figure 5: The redistribution elasticity and the equivalent EIS in Germany, Italy and the U.S. Notably, Havranek (2014) argues that in cross-sectional data the EIS is 0.3-0.4 on average. He points out that in published literature such calibrations are often above this average, as they tend to suffer from selective reporting bias or misspecification. Therefore, potential measurement errors in my estimate of the EIS need to be borne in mind when evaluating the interest rate channel. Nonetheless, the value of my EIS is deeply below 1, which is consistent with findings in economic research (see Havranek, 2014).

Concluding, my results show that the interest rate exposure channel in the PHF sample is large relative to Italy and the United States. This leads to the assumption that the redistributive effect of a change in the real interest rate is higher in Germany than in other countries. My estimates suggest that the large size of savings deposits contributes to a positive unhedged real interest rate exposure of German households. This increases the magnitude of the interest rate channel as a transmission mechanism of monetary policy in my findings. Although my measurements support the existence of a large redistribution effect, potential measurement errors due to much heterogeneity in the PHF data should not be neglected.

-0.37 -0.06 -0.24 0.67 0.12 0.30 -0.60 -0.40 -0.20 0.00 0.20 0.40 0.60 0.80

Germany Italy United States

Redistribution Elasticity Equivalent EIS

(31)

CHAPTER V: LIMITATIONS

This chapter presents several limitations that need to be considered when evaluating my empirical results. Most importantly, it is necessary to interpret the real interest rate channel carefully because it only represents one of many transmission channels of monetary policy. Thus, when assessing the overall effects of monetary policy on aggregate consumption, my analysis only provides partial results. Eventually, the total effect can differ substantially from my empirical results, since other channels may work in the opposite direction to the interest rate channel. On top of that, there are medium- and long-term effects of monetary policy that are not captured in my analysis. For a precise measurement of these long-term channels, such as the labour earnings heterogeneity channel, longer panel studies are needed. Future research may thus disentangle these time-varying effects and provide a more complete picture of the interlinkage between monetary policy and inequality.

Other limitations include measurement errors in the marginal propensity to consume and overall consumption. First of all, I calculate the MPC on the basis of the difference in food consumption and net income between 2010 and 2014. Thus, any change in consumption that is caused by other factors than income contributes to impreciseness in my measurement. Furthermore, my simplified MPC does not differentiate between transitory and permanent changes in income. Japelli and Pistaferri (2010) argue that the MPC should be higher for permanent than for transitory changes in the net income. The analysis of longer panel studies can hereby identify external effects on consumption and contribute to a more precise estimate of the MPC.

Besides, I exclude households that show negative MPC from my analysis of the correlation of URE with MPC. My attempt to exclude outliers may however decrease the representativeness of my sample. If, for example, the negative-MPC agents were high-income households on average, my sample restriction would affect more wealthy than poor households. Although my summary statistics do not suggest an above-average exclusion of wealthy households, the MPC needs cautious interpretation. Japelli and Pistaferri (2010) also point out that households that take part in a panel study have little incentives to provide truthful answers to the interview questions. In relation to this, Coibon et. al. (2012) argue that households tend to underreport consumption

(32)

expenditures. Also my summary statistics show very low consumption expenditures in comparison to Auclert’s (2016) estimates and suggest that consumption is underreported. Another contribution to my low values on spending is the specific exclusion of major purchases such as cars, household appliances and furniture in the consumption measure of the PHF survey. Consequently, impreciseness in my quantification of total household consumption increases potential measurement errors in the URE.

(33)

CHAPTER VI: CONCLUSION

Overall, my results provide evidence for the existence of a redistributive interest rate channel of monetary policy in Germany. When real long-term rates fall, aggregate consumption increases due to intertemporal substitution and redistribution of wealth. Following Auclert’s (2016) methodology, I quantify a cross-sectional measure of the unhedged real interest rate exposure in the PHF sample and find a negative correlation with the marginal propensity to consume. My correlation coefficients are small but significant. They show that households with a positive URE tend to have a lower MPC relative to households with a negative URE. In this transmission channel, monetary policy is redistributing away from positive- to negative-URE households and thus from low- to high-MPC agents. Inequality falls because households with a high MPC tend to be on the lower part of the wealth distribution relative to low-MPC agents. My estimates show that the redistribution elasticity and the equivalent EIS are very large in Germany compared to the U.S. and Italy. Thus, monetary policy in Germany should be pursued cautiously because the redistributional effects of changes in the interest rate can be potentially large.

Furthermore, my results suggest that the URE in the PHF study is highly positive on average. Households tend to have large savings deposits that mature over one year. Hence, German households are generally very exposed to changes in the real interest rate compared to Italian and U.S. households in Auclert’s (2016) analysis. Eventually, real long-term interest rates are very low in Germany. According to the OECD (2017) interest rates on government bonds maturing in ten years were at 0.220 in April 2017. Furthermore, the three-month money market rate was negative at -0.33 in April 2017. The inflation rate measured by the consumer price index (CPI) amounted to 1.96 per cent in April 2017 (see OECD, 2017). In this macroeconomic environment, there is little scope for monetary policy further decreasing the real interest rate. This proximity to the zero lower bound feeds the view that the interest rate transmission channel is small and insignificant. However, my research suggests a large redistribution channel and shows that households’ balance sheets are highly exposed to any change in the real interest rate. A rise in the real interest rate can thus significantly affect the income and wealth distribution in Germany. Future monetary policy should thus not neglect the interest rate transmission channel and its high redistributional potential when setting the policy rates.

(34)

REFERENCES

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

Bank of England (2012): “The Distributional Effects of Asset Purchases,” Quarterly Bulletin Q3/2012, 254-266.

Bernoth, K., König, P., Beckers, B. (2016): “ECB Asset Purchases May Affect Wealth Distribution,” DIW Economic Bulletin No. 7/2016, 75-81.

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.

Deutsche Bundesbank (2016a): “Distributional Effects of Monetary Policy,” Monthly Report, 68 (9), 13-36.

Deutsche Bundesbank (2016b): “Household Wealth and Finances in Germany: Results of the 2014 Survey,” Monthly Report, March 2016, 57-82.

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.

Hall, R. (1988): “Intertemporal Substitution in Consumption,” Journal of Political Economy, 96 (2), 339-357.

Havranek, T. (2014): “Measuring Intertemporal Substitution: Importance of Method Choices and Selective Reporting,” Working Paper, Czech National Bank and Charles University, Prague. 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.

(35)

Kalckreuth, U., Eisele, M., Le Blanc, J., Schmidt, T., Zhu, J. (2012): “The PHF: A Comprehensive Panel Survey on Household Finances and Wealth in Germany,” Discussion Paper No. 13/2012, Deutsche Bundesbank.

OECD (2017a): “Inflation (CPI) (indicator),” doi: 10.1787/eee82e6e-en (Accessed on 05 June 2017).

OECD (2017b): “Long-term interest rates (indicator),” doi: 10.1787/662d712c-en (Accessed on 05 June 2017).

OECD (2017c): “Short-term interest rates (indicator),” doi: 10.1787/2cc37d77-en (Accessed on 05 June 2017).

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

(36)

Appendix A

Table A1: Summary statistics adjusted for outliers using the PHF data

Note: The normalised standard deviation is !!_! !!

!_!!_! for !!= !!, !!, !!, !!, !"#! and !"! !"#! for MPC. The

highest per cent of each !! and !! have been omitted to correct for large outliers.

Table A2: Estimates of !! and !! adjusted for outliers using the PHF survey

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

Variable Mean (€) Normalized SD

Estimated Household Income (!_!, per year) 37,585 3.09

Consumption Including Mortgage Payments (!_!) 8,540 0.93

Deposits and Maturing Assets (!!) 38,026 7.46

ARM Mortgage Liabilities and Consumer Credit (!!) 1,515 1.08 Unhedged Real Interest Rate Exposure (!"#_!, per year) 65,556 9.07

Marginal Propensity to Consume 0.49 1.10

Count 1,365

Parametre Mean Estimate 95% CI

Redistribution Elasticity !! -0.61 [-0.80, -0.44] Scaling Factor ! 0.59 [0.53, 0.65] Equivalent EIS !_! = −!! ! 1.04 [0.84, 1.23]

(37)

Figure A1: Correlation between MPC and URE adjusted for outliers in the PHF population 0 .1 .2 .3 .4 .5 Ma rg in a l Pro p e n si ty to C o n su me 0 5 10 Normalised URE

Distributional Effects of Monetary Policy: The Interest Rate Exposure of German Households