The Effectiveness of M onetary Transm ission to the U S M arket for A utom obile C redit

The Effectiveness of M onetary

Transm ission to the U S M arket for

A utom obile Credit

Master Thesis

08-06-2018

Name: Bastiaan Mensink

Student number: s2763249

MSc: Economics

Supervisor: S. Pool

Faculty of Economics and Business

University of Groningen

A B S T R A C T

:

2 We are committed to using all the tools at our disposal to preserve the strength of our financial

institutions and stabilize our financial markets to minimize the spill-over into the rest of the economy.

Henry Paulson 1. I N T R O D U C T I O N

The above quote from the United States (US) Treasury Secretary Henry Paulson followed the announcement of the Federal Reserve (Fed) to inject $800 billion into the US economy on November 25, 2008 (“US Fed announces $800 billion stimulus,” 2008). The financial market turmoil after the collapse of the subprime mortgage market had led the Fed to decide that this stimulus package was a necessity to stabilize the financial system. The so-called large-scale asset purchases (LSAPs) included $500 billion of mortgage-backed securities (MBS) and $100 billion of government sponsored-enterprise (GSE) debt. An additional $200 billion was lent to investors in asset-backed securities (ABS). Since these LSAPs did not significantly achieve their aim of boosting consumer lending, extensions and new purchase programs came in a sequel to the $800 billion package. The third and last program of debt and security purchases ended in October 2014.1

These rounds of quantitative easing (QE) – a common term to describe the unusual increase of the monetary base – were not completely new phenomena. An earlier and prominent example are the expansionary policies by the Bank of Japan (BoJ) in 2001. At that time, the BoJ responded to a national trend of persistent deflation by means of aggressive purchases of government bonds (Bowman, Cai, Davies, & Kamin, 2011). This and similar ways of central bank’s balance sheet expansions are regularly considered to be unconventional policies (Williams, 2012). In the US, the implementation of QE programs were merely a consequence of ineffective cuts in the federal funds rate (FFR)2_{, the conventional tool of the Fed. After reaching the zero lower bound, the Fed had}

to look for other ways in which it could further stimulate aggregate demand at that time. By targeting different assets of commercial banks and investors, such as through the aforementioned purchases, the Fed provided liquidity to the system in exchange for credit market risk. The purchases should depress funding costs of commercial banks and finance companies and hence lead

1 _{Quoted from the Federal Reserve’s press release from October 29, 2014. This and further press releases are}

listed in T A B L E A .2 of the Appendix.

2 _{Note that the Fed does not always succeed in letting the FFR reach its target (see section 4.2). In the}

3 S O U R C E : F E D E R A L R E S E R V E E C O N O M I C D A T A

to lower interest rates on mortgages and other consumer loans, such as auto loans. John C. Williams, President of the Federal Reserve Bank of San Francisco, claimed that quantitative easing at least managed to push down automobile finance rates and thus to spur automobile sales via this way (Williams, 2012). F I G U R E 1 shows the remarkable increase in the number of outstanding

motor vehicle loans since quantitative easing took place, at least visible after the second and later purchase programs.

This paper examines whether quantitative easing programs of the Fed were indeed the reason for lower interest rates charged by automobile lenders and corresponding higher loan volumes. It mainly does so by investigating the impact of announcements – the so-called stock effects – and actual implementation – or flow effects – of the Fed’s purchase programs on different indicators of the US market for auto credit. The research framework is derived from Fratzscher, Duca, and Straub (2014) in which the effects of unconventional policies of the European Central Bank (ECB) on equity indices and government bond yields are studied. Logically, this model is altered to make it applicable for answering the leading research question. For instance, this implies that control variables are added and removed and that data are expanded to make them operable

F I G U R E 1 · The development of the level of outstanding motor vehicle loans owned and securitized (in

for the time period of interest. This period takes from the first quarter of 2003, in which the FFR fluctuated around levels comparable with the time of writing, to the last quarter of 2017. All announcements and implementations of the Fed’s LSAPs took place within this interval.

The analysis adds to existing literature on quantitative easing by placing a primary focus on the US automobile credit market. Only few authors attempted to investigate the monetary transmission to automobile loan rates or volumes, let alone the transmission of unconventional policies. In addition, this paper utilizes a notably different and new approach in the form of a thorough study on the impact of both flow and stock effects. These effects will be discussed in the next section, in which there is first made room for a debate on existing literature and quantitative easing programs. Other subsections inform of announcement anticipations, automobile interest rates components, monetary transmission channels, and applicable hypotheses. The sections thereafter examine the applied methodology, gathered data, and major findings. The empirical results show significant proof for conventional monetary spillovers, but no clear evidence for stock and flow effects. In the robustness checks section, other important control variables will be added to the model. It precedes the final conclusions, in which limitations are discussed and focus areas for further research are suggested.

2. T H E O R Y

2.1 L I T E R A T U R E R E V I E W

5 from other and earlier QE programs that did not target specific credit markets. The remainder of

this paper, however, will use the terms credit easing and quantitative easing interchangeably when describing the non-traditional expansion of the Fed’s balance sheet.

Williams (2011) summarizes existing studies that focus on the impact of LSAPs on a range of asset prices and interest rates, both in the US and in foreign countries. The Fed official shows a general agreement on the magnitude of LSAP effects on Treasury yields, whereas there is less agreement on the actual channels through which these effects work. Evidence slightly favors the portfolio balance channel over the signaling channel, but this remains far from conclusive. Krishnamurthy and Vissing-Jørgensen (2011) belong to the authors that investigate these and other channels in which intended interest rate cuts might have been rippled through to other rates. One example is the aforementioned signaling channel, which describes the credible commitment that the Fed gave through its QE announcements to keep the FFR low for a longer period. This promise is then, via the expectations hypothesis, expected to affect all other interest rates. Further channels at work are the duration risk channel, the liquidity channel, the safety premium channel, the prepayment risk premium channel, the default risk channel, and the inflation channel. Some of these channels might apply to the transmission of unconventional policies to interest rates on US automobile loans and will be discussed in more detail in section 2.6.

Fawley and Neely (2013) collect information on all the monetary easing programs performed in the aftermath of the financial crisis. These expansionary policies not only originated from the Fed, but also from the Bank of England (BoE), the BoJ, and the ECB. The authors schedule the dates on which the programs were announced, either through speeches or released statements. Additionally, they provide a detailed discussion of messages by the central banks in which build-offs or terminations of the easing programs were communicated. Details on these programs are of critical importance in the study to the effect of LSAP programs on automobile finance rates and the outstanding volume of automobile loans in the US.

existing automobile loans of different credit scores and consequently traded in the financial market. By means of this liquidity provision, the Fed may have influenced automobile loan interest rates and hence the amount of outstanding credit significantly. Besides, literature presents evidence that the credit easing programs also led to striking reductions in interest rates that were not directly targeted by the Fed. For instance, Gagnon, Raskin, Remache, and Sack (2011) find proof for widespread effects to markets for corporate bonds and interest rate swaps. These spillovers apparently took place around the time of QE announcements, which emphasizes the potency of monetary policies at the zero bound. The authors name lower risk premia, rather than lower expectation of future short-term interest rates, as the main reason for the reduction in longer-term interest rates.

While there is plenty of studies focusing on the impacts of quantitative easing, only few authors write about the transmission to the US market for automobile credit. Mora (2014), for instance, notices a significant decrease in 48-month new car loan rates since monetary easing has started. Nevertheless, the spillovers of the Fed’s policies to these rates appear to be weakened compared to earlier periods. A fraction of this attenuation is assigned to existing pricing power in the market in times of low interest rates. Changes in lending risk and economic conditions are among the other reasons for a softer conventional transmission. Although Mora (2014) discriminates between monetary transmission in times of quantitative easing and spillovers in other periods, he does not investigate direct impacts of the Fed’s announcements on or implementation of unconventional policies. In an attempt to fill this gap in this and other literature, it is good to start a review on the Fed’s quantitative easing programs that took place between 2008 and 2014.

2.2 Q U A N T I T A T I V E E A S I N G

In fact, the period of active purchase programs by the Fed consisted of four distinct programs: QE1, QE2, Operation Twist, and QE3. At the time of writing, the central bank is shrinking its balance sheet through the sale of its security holdings. During the rounds of monetary easing, the Fed maintained the size of its balance sheet by reinvesting principal payments on maturing debt in both Treasuries and MBS. F I G U R E 2 shows the total amount of system open market account

7 S O U R C E : F E D E R A L R E S E R V E B A N K O F N E W Y O R K

an extension of Fawley and Neely (2013) and is updated with data from the Fed to make it suitable for the time period of interest. Details on the depicted expected effects will follow in section 3.2.

i. Q E 1

As mentioned in the introduction, quantitative easing in the US started with an $800 billion rescue package announced on November 25, 2008. This message arrived shortly after the first out of four quarters of national economic recession. The Fed initially designed the program to support the entire economy, but by spending the largest amount on housing GSE debt and MBS (i.e., more than 80% of the package) it showed its priority for the market of housing credit. The Federal Open Market Committee (FOMC) predominantly hoped to “reduce the cost and availability of credit for the purchase of houses, which in turn should support housing markets and foster improved conditions in financial markets more generally”. Six days after the release, Fed Chairman Ben Bernanke discussed the potential extension of the program to Treasuries for the first time. This extension became reality in March 2009, when the Fed bought $300 billion of these long-term securities. In addition, the bank purchased another $750 billion and $100 billion in MBS and GSE

S O U R C E : F A W L E Y & N E E L Y ( 2 0 1 3 ) , F E D E R A L R E S E R V E B A N K

T A B L E 1 · The major announcements of the Fed. The last column denotes the hypothesized effect of each announcement on average automobile loan rates, assuming no

market anticipations; ▼▼denotes a strong negative effect, ▼ a weak negative effect, ▬ a neutral effect, ▲ a weak positive effect, and ▲▲ a strong positive effect.

DATE PROGRAM BRIEF DESCRIPTION OF THE ANNOUNCEMENT MEDIUM EXPECTED EFFECT

11/25/2008 QE1 The Fed will purchase $500 billion in MBS and $100 billion in GSE debt; in addition, the Fed lends $200 billion to holders of debt backed by consumer loans, such as car loans.

FOMC statement ▼▼

12/1/2008 QE1 First suggestion of program extension through the purchase of Treasuries. Bernanke speech ▼

12/16/2008 QE1 First FOMC statement in which the program extension to Treasuries is suggested. FOMC statement ▼

1/28/2009 QE1 The Fed stands ready to expand and to buy Treasuries. FOMC statement ▼

3/18/2009 QE1 The Fed purchases $300 billion in long-term Treasuries; an additional $750 billion and $100 billion in MBS and GSE debt, respectively, will be purchased.

FOMC statement ▼

8/12/2009 QE1 Slowdown: all purchases will finish by the end of October instead of mid-September. FOMC statement ▬ 9/23/2009 QE1 Slowdown: finish of GSE debt and MBS purchases is further delayed to end March, 2010. FOMC statement ▬ 11/4/2009 QE1 QE1 is downsized: not $200 billion, but $175 billion of GSE debt will be bought at total. FOMC statement ▲ 8/10/2010 QE1 The Fed will maintain its balance sheet by reinvesting principal payments in Treasuries. FOMC statement ▼

8/27/2010 QE2 There is a role for additional easing, ‘should further action prove necessary’. Bernanke speech ▼

9/21/2010 QE2 Inflation is expected to be too low for some time, so inconsistent with the Fed’s mandate. FOMC statement ▼

10/12/2010 QE2 Members of the FOMC feel that additional accommodation might be appropriate. FOMC minutes released ▼

10/15/2010 QE2 Bernanke states that the Fed stands ready to further ease policy. Bernanke speech ▼

11/3/2010 QE2 Announcement of new purchases: the Fed is going to invest $600 billion in Treasuries. FOMC statement ▼▼

6/22/2011 QE2 QE2 announced to finish this month; principal payments will continue to be reinvested. FOMC statement ▲ 9/21/2011 Operation

Twist

Maturity extension program (‘Operation Twist’) announced: the Fed will trade $400 billion of Treasuries with 6-30 years remaining maturities for an equal amount of Treasuries with up to 3 years remaining maturities.

FOMC statement ▼▼

6/20/2012 Operation Twist

Maturity extension program extended: the Fed will continue with the purchase of long-term securities in exchange for short-long-term securities till the end of 2012. The program will continue with the pace at that time in force – that is, a monthly volume of $45 billion.

FOMC statement ▼

8/22/2012 QE3 Additional expansionary policies are likely to occur soon, according to FOMC members. FOMC minutes released ▼

9/13/2012 QE3 Third round of easing is kicked off: the Fed purchases MBS at a pace of $40 billion per month, as long as the economic outlook does not improve substantially.

FOMC statement ▼▼

12/12/2012 QE3 Expansion of the program: the Fed continues to purchase $45 billion of long-term Treasuries per month.

FOMC statement ▼

9 S O U R C E : F A W L E Y & N E E L Y ( 2 0 1 3 ) , F E D E R A L R E S E R V E B A N K

S O U R C E : F A W L E Y & N E E L Y ( 2 0 1 3 ) , F E D E R A L R E S E R V E B A N K

T A B L E 1 [CONTINUED] · The major announcements of the Fed. The last column denotes the hypothesized effect of each announcement on average automobile loan rates,

assuming no market anticipations; ▼▼denotes a strong negative effect, ▼ a weak negative effect, ▬ a neutral effect, ▲ a weak positive effect, and ▲▲ a strong positive effect.

12/18/2013 QE3 The pace of purchases is reduced by $10 billion per month. After each following meeting of the FOMC, the monthly pace is reduced by $10 billion.

FOMC minutes released ▲ 10/29/2014 QE3 The Fed announces to conclude its asset purchase program this month. The Committee

bases this decision on a substantial improvement in the outlook for the labor market.

FOMC statement ▲

6/14/2017 Policy Normalization

The new President Yellen announces that the Fed is likely to shrink its balance sheet soon.

Yellen speech ▲

9/20/2017 Policy Normalization

The Fed starts to wind down its program by unloading $10 billion of purchased debt per month, at least from October to December 2017. The amount consists of $6 billion in Treasuries and $4 billion in GSE debt.

Yellen speech ▲▲

l

T A B L E 2 · Key announcements of the Fed with expected negative effects on average auto loan rates, assuming no market anticipations; symbols of the first column match

with the symbols shown in F I G U R E 3. Expected effects correspond to the effects of T A B L E 1; observed effects describe the actual change in the automobile loan rate

in the period after the announcement: ▼denotes a decrease in the average automobile loan rate, ▼▼denotes a decrease that is stronger than the change in the prior quarter.

SYMBOL PERIOD PROGRAM BRIEF DESCRIPTION OF THE ANNOUNCEMENT MEDIUM EXPECTED EFFECT OBSERVED EFFECT

Q4·2008 QE1 The Fed announces an $800 billion rescue program. Later in this quarter, there is a first suggestion of program extension to Treasuries.

FOMC statement, Bernanke speech

▼▼,

▼··

▼▼

Q1·2009 QE1 The Fed purchases $300 billion in Treasuries and an additional $850 billion in MBS and GSE debt.

FOMC statement ▼ ▼

Q3·2010 QE2 Bernanke keeps possibilities for additional easing open. Bernanke speech ▼ ▼▼

Q4·2010 QE2 The Fed will invest $600 billion in Treasuries. FOMC statement ▼▼ ▼

Q3·2011 Operation Twist

Maturity extension program announced: the Fed will purchase $400 billion of long-term securities and sell short-term securities.

FOMC statement ▼▼ ▼▼

Q2·2012 Operation Twist

Maturity extension program extended to the end of 2012, at the pace of approximately $45 billion per month.

FOMC statement ▼ ▼

Q3·2012 QE3 New program announced: the Fed will purchase $40 billon in MBS per month as long as the outlook does not improve substantially.

FOMC statement ▼▼ ▼▼

Q4·2012 QE3 Program extension: the Fed keeps investing $45 billion a month in long-term Treasuries, but does not sell short-term securities in turn.

S O U R C E : F A W L E Y & N E E L Y ( 2 0 1 3 ) , F E D E R A L R E S E R V E B A N K

T A B L E 3 · Key announcements of the Fed with expected negative effects on average automobile loan rates, assuming no market anticipations; symbols of the first column

match with the symbols shown in F I G U R E 4. Expected effects correspond to the effects of T A B L E 1; observed effects describe the actual change in the automobile

loan rate in the period after the announcement: ▲ denotes an increase in the average automobile loan rate, ▲▲ denotes an increase that is stronger than the adjustment in the prior quarter,▼▼denotes a decrease that is stronger than the adjustment in the prior quarter.

SYMBOL PERIOD PROGRAM BRIEF DESCRIPTION OF THE ANNOUNCEMENT MEDIUM EXPECTED EFFECT OBSERVED EFFECT

Q4·2009 QE1 QE1 downsized: the Fed purchases $25 million of GSE debt less. FOMC statement ▲ ▼▼

Q2·2011 QE2 The second program is announced to finish at the end of the quarter. FOMC statement ▲ ▲▲

Q2·2013 QE3 Bernanke claims that easing might finish at the end of 2013. Bernanke speech ▲▲ ▲▲

Q4·2013 QE3 The pace of the purchases is reduced by $10 billion per month. FOMC minutes released

▲ ▼▼

Q4·2014 QE3 The Fed concludes its asset purchase program at the end of October. FOMC statement ▲ ▲▲

Q2·2017 Policy Normalization

According to the President Yellen, the Fed is likely to shrink its balance sheet soon.

Yellen speech ▲ ▼▼

Q3·2017 Policy Normalization

The Fed starts to wind down its program by unloading $10 billion of purchased debt per month, at least from October to December 2017.

11 debt, respectively. Together with the November 2008 asset purchase program, the March 2009

extension became hereafter known as ‘QE1’. ii. Q E 2

Although the first round of monetary easing amounted to approximately 12% of the US economy, economic activity remained sluggish afterwards. The FOMC had to conclude that “progress toward its objectives has been disappointingly slow.” This realization persuaded the Fed to decide upon additional asset purchases in November 2010. The new program consisted of $600 billion investment in long-term Treasury securities and was meant to “promote a stronger pace of economic recovery”. In August 2010, the Fed had already communicated that it would reinvest principal payments of matured debt in new Treasuries. The second asset purchase program received criticism from both insiders and the public, for it would only drive up inflation and not achieve higher employment levels (Swanson, 2011). Bernanke and others, however, persisted, such that ‘QE2’ was born. Whereas QE1 was open-ended, QE2 officially expired in July 2011.

iii. O P E R A T I O N T W I S T

Shortly after QE2 was terminated, the Fed announced a maturity extension program similar to an earlier program in 1961. On February 2 that year, President John F. Kennedy declared that the Treasury and the Fed would cooperate to change the relative supplies of long-term and short-term securities for an amount of $8.8 billion (Swanson, 2011). Now, on September 21, 2011, the Fed announced to purchase $400 billion of Treasuries with 6-30 years remaining maturities in exchange for Treasuries with up to 3 years remaining maturities. Taken into account that the size of the US economy was much smaller in 1961, the 2011 program was relatively similar to its predecessor. Later, the maturity extension program therefore obtained the same name: ‘Operation Twist’. In July 2012, Operation Twist was extended with a final amount of $267 billion at a monthly pace of $45 billion. Two months after this extension, the Fed choose to implement a last and third round of easing.

iv. Q E 3

longer-term interest rates and to support mortgage markets and the broader financial climate. In December 2012, the program was extended to temporary monthly purchases of long-term Treasuries at a price of $45 billion. On July 19, 2013, President Bernanke stated that the first phase-out of purchases was expected to take place at the end of 2013. This declaration would be the trigger for the omega of a long period of monetary easing by the Fed. The gradual decrease in monthly pace of purchases ended on October 29, 2014, when the LSAPs were definitely concluded.

v. P O L I C Y N O R M A L I Z A T I O N

Almost three years after the last asset purchases, Fed Chairwoman Janet Yellen announced in a press conference that the Fed intended to shrink its balance sheet soon. SOMA holdings had increased massively and were ready to be reduced. The Fed indeed started to wind down its program in October 2017 through the monthly unloading of $10 billion of purchased debt until December 2017. After these three months, the pace of sale should gradually increase to $20 billion and eventually $30 billion per month. Yellen substantiated the balance sheet shrinkage with observations of the FOMC that “the economy is performing well” and that there is “confidence in the outlook for the real economy”. This still ongoing process of balance sheet build-offs will in the remaining part be referred to as ‘Policy Normalization’.

2.3 S T O C K V E R S U S F L O W E F F E C T S

13 This does not hold for flow effects. These assume that deviations between announcements

and actual purchases lead to yield differentials. Flow effects might either result from the obscurity of the announcement or from adjustments of the Fed’s policy after it is communicated (e.g., due to changes in the economic outlook). When flow effects dominate, the impact of quantitative easing on automobile loan rates may be observed at the time of the actual purchases rather than at the time of announcement. Elbourne et al. (2018) discuss existing shadow rates that merge direct and indirect impacts of unconventional monetary policies, but these mainly focus on unusual actions of the ECB. This paper will test for both stock and flow effects through the interchangeable inclusion of announcement dummies from T A B L E 1 and total SOMA holdings by the Fed in the

model specification.

2.4 A N T I C I P A T I O N V E R S U S A N N O U N C E M E N T E F F E C T S

Next to the difference between stock and flow effects, readers of literature monetary easing should be aware of the distinction between announcement and anticipation effects. When easing programs had caught US markets by surprise, it should have been reasonable to estimate stock or flow effects by simply adding up interest or loan stock movements in the window. Arguably, this assumption applies for the first announcements and implementations of the Fed’s unconventional policies. For later programs, however, markets began to anticipate on future QE expansions even before the Fed had announced such expansions (Gagnon, 2016). In these cases, any observed effect on auto loan rates at the time of the announcement should then be assigned to the deviation between market expectations and the actual announcement rather than by the announcement itself. Some event studies try to compensate for market expectations, for example by controlling for deviations between domestic and foreign bond yields (Churm, Joyce, Kapetanios, & Theodoridis, 2015). However, it typically remains hard to identify market expectations, especially when related to future programs of monetary easing. In addition, once expectations can be aggregated, they are difficult to quantify. It is for these reasons that the models that are used in this paper do not account for market expectations. Market anticipations are nonetheless not disregarded, but taken into consideration in the remaining sections.

schedules aggregated expectations that match with the later announcements of T A B L E 1. Once

these tables are compared, it can be observed that – in most cases – markets were already familiar with the type of announcement in advance of the statement or speech. Actual stock effects may therefore even be of opposite sign compared to the expected effects of T A B L E 1, which assume

no anticipation. That is, e.g., when primary dealers expected a larger amount of easing than announced, the effect of the QE announcement on auto lending rates might even be positive. Although anticipations thus do not show up in the methodology of this paper, it should be clear from this debate that announcements should not have the expected effects of T A B L E 1 per se.

2.5 A U T O M O B I L E F I N A N C E R A T E C O M P O N E N T S

There are multiple ways in which automobile purchases are financed in the US, but around two thirds of the automobiles are loan-financed (Larrimore, Dodini, & Thomas, 2016). In case of a loan contract, a distinction can be made between indirect financing – where the borrower arranges terms of his credit contract with an intermediate automobile dealer – and direct financing – where credit is arranged directly with the agency that services the contract (Shay, 1963). Although captive lenders, credit unions and automobile finance companies gain in popularity, commercial banks are still the major provider of automobile loans with a market share of approximately 35% (Zabritski, 2017). They are involved both in direct and indirect lending and offer loans for new and used vehicles. Interest rates on these loans are based on the funding and operating costs of the loan and a premium to compensate for existing credit risk (Shay, 1963). Each potential borrower obtains a credit score which measures default risk based on the client’s income, residential status, employment history, and other factors. Reasonably, borrowers with lower credit scores have to pay a higher interest rate on their car loan compared to principals with higher credit scores. Opposed to credit card debt rates, automobile loan interest rates are not directly tied to the FFR, but rather set by the free market. However, since funding costs and other commercial bank balance sheet items change as a consequence of FFR fluctuations, it is safe to assume that automobile finance rates also follow changes in the Fed’s conventional tool (Hsu, 2017). The methodology section accounts for this relation through controls for both the contemporaneous and lagged federal funds rate. Bernanke and Blinder (1992) further extend on market rigidities that might back the use of a lag of the FFR in the upcoming regression models.

15 a liquidity premium (Cox, Ingersoll, & Ross, 2005). The risk-free rate is mostly a theoretical

concept and can be compared with the so-called prime rate on auto loans, which is expected to be significantly related with the FFR. Other factors and premia can also be influenced by both conventional and unconventional policies, as will be hypothesized in the next section.

2.6 T R A N S M I S S I O N C H A N N E L S

Based on literature from Krishnamurthy and Vissing-Jørgensen (2011), Williams (2011), and other authors, conclusions can be drawn regarding the channels through which monetary transmission to the US automobile credit market may have occurred. Since there has been scarce earlier investigation on this conveyance, the following channels should be considered to be hypothetical:

✓ The confidence channel. By taking decisive actions as done in the different rounds of QE, the Fed might help restoring confidence in the financial system. There is trust at and between commercial banks that they will be supported when times get worse. As a consequence, the risk premium on automobile loan interest rates and uncertainty might decline and the number of supplied loans might increase.

✓ The portfolio balance channel. A change in the term premium on government bond yields – as affected by LSAPs – leads commercial banks to rebalance their portfolios, which might eventually have additional effects on automobile loan interest rates and thus on automobile loan volumes. For instance, as yields on Treasuries or on MBS decrease due to quantitative easing, investors might substitute Treasury and MBS holdings for automobile loan asset-backed securities. In this way, demand for auto loans might be pushed and interest rates depressed.

✓ The signaling channel. By implementing LSAPs, the Fed communicates a certain path it is going to walk in the longer term. As long-term interest rates are expected to get lower, commercial banks foresee cheaper funding costs for some time. Hence, they are likely to adjust their automobile loan rates downwards. It is difficult to distinguish between this channel and forward guidance (Campbell, Evans, Fisher, & Justiniano, 2012), which describes the anticipations of commercial banks on conventional policies of the Fed (i.e., changes in the federal funds rate). However, it is likely that this mechanism was at work. ✓ The liquidity channel. Liquidity in hands of commercial banks increased due to quantitative

lowered the liquidity premium on their automobile loan interest rates. Hence, automobile loan interest rates might be lower due to the LSAP programs.

Most of these channels are expected to set automobile loan rates in motion immediately after the announcements of QE programs take place. This expectation follows the assumption that commercial banks adjust their portfolios and interest rate directly after the message is brought to public, without any anticipation in advance. It is though not unlikely that anticipation or flow effects, observed at the time of actual asset purchases by the Fed, coexist. The methodology section shows translations of these theoretical arguments into different regression model applications.

2.7 H Y P O T H E S E S

All channels from the previous section argue in favor of negative effects of QE programs on automobile loan rates, either via flow or stock effects. Accordingly, lower interest rates are expected to result in a higher stock of outstanding loans. Furthermore, it is assumed that announcements of reductions and termination of asset purchases will lead to higher automobile loan rates. This and the preceding discussion can therefore be summarized by means of the following hypotheses: H Y P O T H E S I S 1: Announcements on the start of a quantitative easing program have, ceteris

paribus, led to lower automobile finance rates (a higher stock of outstanding automobile loans), whereas announcements on downsizing or termination have led to higher automobile finance rates (a lower stock of outstanding automobile loans).

H Y P O T H E S I S 2: An increase in SOMA holdings by the Fed goes, ceteris paribus, hand in hand with a decrease in automobile finance rates (an increase in the stock of outstanding auto loans), whereas a decrease in SOMA holdings corresponds to an increase automobile finance rates (an decrease in the stock of outstanding automobile loans).

17 3. M E T H O D O L O G Y

3.1 B E N C H M A R K M O D E L

In order to investigate the impact of the Fed’s unconventional policies on the US market for auto credit, a similar model to that of Fratzscher et al. (2014) will be utilized as a benchmark:

𝑦𝑡= 𝛼 + 𝛽1∙ FFR𝑡+ 𝛽2∙ FFR𝑡−1+ 𝜑 ∙ QE𝑡+ 𝛾 ∙ F𝑡 + 𝜀𝑡, (1) where 𝑦𝑡 denotes the dependent variable of interest at quarter t.3 These reflect US auto credit market indicators and include the average 48-month and 60-month new automobile loan interest rates at US commercial banks; the average automobile finance rate at US automobile finance companies (AFCs); and the outstanding stock of US motor vehicle loans owned and securitized. The coefficients 𝛽1 and 𝛽2 measure the extent to which the federal funds rate at quarter t (FFR𝑡) and the prior quarter t–1 (FFR𝑡−1) affects the dependent variables. In three of the four cases, 𝛽1 and 𝛽2 thus estimate the conventional transmission from the FFR to automobile finance rates. In the other case, the effect of FFR adjustments on the outstanding stock of automobile loans is measured. Coefficient 𝜑 estimates the impact of quantitative easing at quarter t (QE𝑡) in various ways. In the case of flow effects, QE𝑡 describes the average stock of SOMA holdings of the Federal Reserve at quarter t. In the case of stock effects, QE𝑡 describes a set of dummy variables that will be expounded in section 3.2. As will become clear, these dummy variables amount to 15 quarters of policy announcements from the Fed. All major announcements are listed in T A B L E 1, whereas

T A B L E 2 and T A B L E 3 plot the selected 15 crucial announcements with hypothesized strong

and weak positive and strong and weak negative effects, respectively. As noted in the table descriptions, it is assumed here that there are no market anticipations on these messages. An additional announcement comparison model will test whether classifications shown in T A B L E 2

and T A B L E 3 make sense. Finally, a program comparison model tests for the different easing

programs in order to measure the level of success of each program. The set F𝑡 captures contemporaneous control variables (see section 4.2). Lastly, a constant term 𝛼 and an error term 𝜀𝑡 are included. All applications of the benchmark model are estimated by means of Ordinary Least Squares (OLS).

3 _{Note that the set of monetary indicators MP}

𝑡 of Fratzscher et al. (2014) is replaced by the conventional

indicators for the FFR and the unconventional set QE𝑡. The lagged control variable set Z𝑡−1 is excluded due

As has been stated, the model is largely derived from a panel data regression model of Fratzscher et al. (2014), in which the authors analyzed the effect of policy announcements on the return on a country’s main equity index, the first difference of long-term government bond yields, and other dependent variables. Model (1) differs from its reference in at least three aspects. First, this model focuses purely on US unconventional policies rather than on monetary easing by the ECB. Second, equation (1) is logically concentrated on automobile credit determinants instead of the asset prices and bond yield differentials that are studied by Fratzscher et al. (2014). Third and finally, the set of control variables is different from Fratzscher et al. (2014) to make the regression suitable for the US market for automobile credit. Section 4.2 discusses which variables are incorporated and hence expected to affect auto loan rates or volumes. In the upcoming section, different specifications of model (1) will be laid out.

3.2 A P P L I C A T I O N S

i. S T O C K E F F E C T M O D E L

The first specification follows from H Y P O T H E S I S 1, which expects stock effects to play a role

in the monetary transmission to the market for automobile credit. That is, announcements of quantitative easing, ceteris paribus, directly lead to a reduction in automobile loan rates. This spillover should take place through the channels of section 2.6. The last column of T A B L E 1

19 in the benchmark model (1) by these dummy variables and their coefficients finally result in the

following estimated stock effect model:

𝑦𝑡= 𝛼 + 𝛽1∙ FFR𝑡+ 𝛽2∙ FFR𝑡−1+ 𝛿1∙ QE1+ 𝛿2∙ QE2+ 𝛿3∙ QE3+ 𝛿4∙ QE4+ 𝛿5∙ QE5+ 𝛿6∙ QE6+ 𝛿7∙ QE7+ 𝛿8∙ QE8+ 𝛿9∙ QE9+ 𝛿10∙ QE10+ 𝛿11∙ QE11+ 𝛿12∙ QE12+ 𝛿13∙ QE13+ 𝛿14∙ QE14+ 𝛿15∙ QE15+ 𝛾 ∙ F𝑡+ 𝜀𝑡, (2) where QE1 refers to announcement and equals one only in Q4·2008, QE2 to equaling one in Q1·2009, QE3 to ….equaling one in Q4·2009, and so on in a chronological order.

ii. F L O W E F F E C T M O D E L

The basis for the second extension is formed by H Y P O T H E S I S 2. This hypothesis followed

from the expectation that flow effects, thus the effects at the time of actual asset purchases from the Fed, play a dominant role in the market for automobile credit. As explained, the Federal Reserve holds assets that it has purchased in the open market in its SOMA holdings. These have been nearly all Treasury securities, such as notes and bonds, in the past (Gagnon et al., 2011). The increase in SOMA assets occurs through so-called outright open market operations (OMOs), which tended to be small in comparison with the total security market before the Great Recession. LSAPs, however, were designed to have a relatively large impact on targeted interest rates and asset prices. Hence, they led to striking increases in the Fed’s SOMA holdings, especially for MBS and agency debt accounts, as shown in F I G U R E 2. The relation between SOMA holdings and

the automobile credit market can be tested and summarized with the following flow effect model:

𝑦𝑡= 𝛼 + 𝛽1∙ FFR𝑡+ 𝛽2∙ FFR𝑡−1 + 𝜃 ∙ SOMA𝑡+ 𝛾 ∙ F𝑡+ 𝜀𝑡, (3) which is equal to model (1) except for the replacement of the term 𝜑 ∙ QE𝑡 by 𝜃 ∙ SOMA𝑡. The parameter 𝜃 estimates the influence of the continuous variable SOMA𝑡, which reflects the average SOMA holdings of the Fed at quarter t. Its sign is expected to be negative, as explained by the channels of section 2.6.

iii. A N N O U N C E M E N T C O M P A R I S O N M O D E L

Details on the statements of the Federal Reserve allow for a precise comparison between different types of announcement effects on the US market for automobile credit. This implies that the previously debated classifications, denoted in the last column of T A B L E 1 by the five different

𝑦𝑡= 𝛼 + 𝛽1∙ FFR𝑡+ 𝛽2∙ FFR𝑡−1+ 𝛿1∙ STRONG_NEGATIVE𝑡+ 𝛿2∙ WEAK_NEGATIVE𝑡+ 𝛿3∙ WEAK_POSITIVE𝑡+ 𝛿4∙ STRONG_POSITIVE𝑡+ 𝛾 ∙ F𝑡+ 𝜀𝑡. (4) As can be noticed, neutral effects are neglected since they are expected to have no serious impact on the US automobile credit market. The other four classifications are represented by the variables

STRONG_NEGATIVE𝑡, WEAK_NEGATIVE𝑡, WEAK_POSITIVE𝑡, and STRONG_POSITIVE𝑡. Respectively, they represent the impact of the four program announcements (measured by 𝛿1); of voices for additional easing and messages of program expansions (measured by 𝛿2); of reports of program conclusions and pace reductions (measured by 𝛿3); and of two speeches about complete easing termination and debt unloading (measured by 𝛿4). STRONG_NEGATIVE𝑡 equals one in four quarters, WEAK_NEGATIVE𝑡 in five quarters, WEAK_POSITIVE𝑡 in four quarters, and STRONG_POSITIVE𝑡 in two quarters.

iv. P R O G R A M C O M P A R I S O N M O D E L

A last framework is specified in order to compare the four purchase programs of the Fed and the ongoing process of policy normalization. It divides all dummy variables into these five categories and tests for the impact of each program by means of the following model:

𝑦𝑡= 𝛼 + 𝛽1∙ FFR𝑡+ 𝛽2∙ FFR𝑡−1 + 𝛿1∙ 𝑄𝐸1𝑡+ 𝛿2∙ QE2𝑡+ 𝛿3∙ OPERATION_TWIST𝑡+ 𝛿4∙ QE3𝑡+ 𝛿5∙ POLICY_NORMALIZATION𝑡+ 𝛾 ∙ F𝑡+ 𝜀𝑡, (5) where 𝑄𝐸1𝑡represents three quarters of QE1 announcements, QE2𝑡equals one in three quarters of announcements on QE2, OPERATION_TWIST𝑡 denotes two quarters of announcements on Operation Twist, QE3𝑡 describes five quarters of QE3 announcements, and POLICY_NORMALIZATION𝑡 represents two quarters of announcements on Policy Normalization. See T A B L E 2 and T A B L E 3 for

details of these messages. Results and conspicuities of the program comparison model will be discussed extensively in section 5.5.

4. D A T A A N D F I N D I N G S 4.1 D E P E N D E N T V A R I A B L E S

21 expected to be mostly concerned with these two automobile lending rates. It is, however, also

interesting to test for potential monetary spillovers to lending rates at AFCs. Since these companies are major competitors of commercial banks, it might be observed that the estimated coefficients for average auto finance rates at AFCs have opposite signs compared to banks. For instance, a decrease in the FFR and thus in the funding costs of bank loans might result in higher lending rates at AFCs due to shifts in demand and, ultimately, market shares. Similar movements might occur in times of quantitative easing, which might have pulled away potential automobile borrowers from AFCs to banks. Because of this interest, the average automobile finance rates at AFCs are included as a third dependent variable. The fourth dependent variable is represented by the stock of automobile loans in order to see what monetary policies eventually did to car debt levels. In the following, data on the dependent variables are shortly described.

Average interest rate on 48-month new car loans at US commercial banks. The quarterly finance rate on 48-month new auto loans is the only commercial auto lending rate from which data are richly available for the whole time period of interest. Data stem from the Federal Reserve Board, are downloaded from the Federal Reserve Economic Data (FRED) database of the Federal Reserve Bank of St. Louis, and are collected through the Quarterly Report of Interest Rates on Selected Direct Consumer Installment Loans. In this report, 150 banks are asked to mention the rate at which the largest dollar volume of loans was made during the surveyed period. Data on these so-called “most common rates” are neither benchmarked nor seasonally adjusted. A main advantage of having specific data on 48-month new car loan interest rates is that there is no need to control for the term of the loan and the age of the vehicle.

Average interest rate on 60-month new car loans at commercial banks. The next dependent variable is the quarterly average new automobile finance rate on 60-month loans. Data are gathered and transformed in the same way as 48-month new car loan interest rates. Different from these rates, however, is that commercial banks did not have to report the average 60-month new car loan rate before July 2006. This implies that the time period of investigation of quantitative easing impacts on the most common 60-month loan rates is reduced to Q3·2006 till Q4·2017. Despite this sample curtailment, all announcements of T A B L E 1 still fall within the new time frame. This

release at the website of the Board of Governors of the Federal Reserve System. The rates encompass the majority of US captive and non-captive finance companies, ever-upcoming competitors of commercial banks (see section 2.5). Publication was ceased after January 2011, since foundation for the series had been deteriorated. Starting on May 7, 2015, series were resumed with a new source extending back to Q1·2008. These new averages are weighted for the average amount financed and are thus not similar to the earlier finance rates. This problem is solved with a structural break dummy variable (Dufour, 1980), equal to one in each quarter later than Q4·2010 – for which the new data are used – and zero for Q4·2010 and earlier.

Total outstanding US motor vehicle loans owned and securitized. Next to the potential spillovers of QE to auto loan interest rates, this study is interested in the impact of the Fed’s programs on the actual stock of outstanding automobile loans. The total of outstanding motor vehicle loans owned and securitized includes loans on passenger cars and other vehicles, such as minivans and pickup trucks. Data are gained from the FRED database of the Federal Reserve Bank of St. Louis and are averaged over each quarter of the time period of interest. The estimates are constructed as a sum of automobile loans outstanding at depository institutions, AFCs, credit unions, and nonfinancial business. Data are not seasonally adjusted, but are stationary.

4.2 I N D E P E N D E N T V A R I A B L E S O F I N T E R E S T

23 Announcements on large-scale asset purchases by the Federal Reserve. Data on the Fed’s

announcements regarding its quantitative easing programs are mainly collected from Fawley and Neely (2013). As explained before, the authors provide brief descriptions on major announcements of the central bank’s purchase programs and details on the medium through which these messages were brought to public. T A B L E 1 plots these and additional dates taken from releases and

statements placed at the website of the Fed. Selection of these three extra dates are based on received media attention and personal expectations of importance in the explanation of auto loan rates and stocks fluctuation. They include the conclusion of the Fed’s asset purchase programs, the first hint by Yellen about balance shrinkage and the announcement of actual debt unloading. URL links to all announcements can be found in T A B L E A .2 of the Appendix.

System open market account holdings of the Federal Reserve. Data on SOMA holdings of domestic securities stem from the website of the Federal Reserve Bank of New York. This bank has been selected by the FOMC as the one that executes transactions for the SOMA, for macroeconomic and other objectives. Data contain all the Fed’s dollar-denominated assets acquired through OMOs. Total holdings are averaged over each quarter as from the second quarter of 2003, when data became available for the first time. This implies that the number of observations for the flow effect model is 59 instead of 60.

are also downloaded from the Center for Microeconomic Data. They are summarized in another dataset of the New York Fed Consumer Credit Panel and are reported without the need for additional calculations. Opposed to the first measure of credit scores, these overall median credit scores also concern loans stemming from AFCs. The data are used for regressions on the average auto finance rates at AFCs and on the stock of outstanding motor vehicle loans. A disadvantage of these data is that they are only available from Q1·2004, meaning that the number of observations for the last two dependent variables is reduced to 56. Median credit scores are included in the control variable set F𝑡 of each model.

Net percentage of domestic banks reporting stronger demand for auto loans. Most of the independent variables merely play a role at the supply side of automobile loans. It is, however, not unlikely that auto loan interest rates at commercial banks also follow demand shocks. The models compensate for these shocks through the inclusion of a variable that captures the net percentage of domestic banks reporting stronger demand for auto loans for each quarter. Data on these percentages are downloaded from the FRED database and are rooted in the Senior Loan Officer Opinion Survey (SLOOS) on Bank Lending Practices. The Fed asks loan officers at each respondent bank to respond to the SLOOS up to six times a year. The net percentage is either positive, if the perceived demand of most commercial banks for auto loans has increased in the reported quarter, or negative, if perceived demand decreased. Shifts in demand are expected to be positively related with auto loan interest rates and negatively with outstanding car debt. Like median credit scores, the net percentage changes in demand are included in the control set F𝑡.

25 S O U R C E : F A W L E Y & N E E L Y (2 0 1 3 ) , F E D E R A L R E S E R V E B A N K O F N E W Y O R K

4.3 F I N D I N G S

F I G U R E 3 depicts the development of the average 48-month new automobile loan interest rates

of US commercial banks together with the effective FFR over the time period of interest. Each vertical line represents a quarter in which an announcement was made that was, ceteris paribus, expected to be followed by a decrease in the automobile loan rate. The lines are highlighted with symbols that correspond to the announcements of T A B L E 2. The last column of this table

represents the observed effects of each announcement in comparison with lending rate changes in the previous periods. It can be concluded from the figure that, while the effective FFR remained constant during the period of financial crisis, QE announcements are related with remarkable drops in 48-month new automobile loan rates. For five out of eight quarters, the rate reductions were even stronger than the decrease in the previous period (or increase, for the first and fifth announcement). F I G U R E 4 describes the same developments, but now combined with lines of

F I G U R E 3 · The development of the average 48-month new automobile loan interest rates of US

S O U R C E : F A W L E Y & N E E L Y (2 0 1 3 ) , F E D E R A L R E S E R V E B A N K O F N E W Y O R K

F I G U R E 4 · The development of the average 48-month new automobile loan interest rates of US

commercial banks and the effective FFR. Vertical lines represent moments of important announcements by the Federal Reserve, presented in T A B L E 3, with an expected positive impact on automobile loan rates.

F I G U R E 5 · The evolution of the net percentage of commercial banks reporting stronger demand for

automobile loans and the average 48-month new automobile loan interest rate over time.

27 announcements that were, ceteris paribus, expected to be followed by an increase in the automobile

loan rate. The symbols now correspond with the announcements of T A B L E 3. Quarters of

announcements have ambiguous outcomes, with only positive observed effects after declarations of purchase program terminations and after the speech in which debt unloading was communicated. For both F I G U R E 3 and F I G U R E 4, however, it should be noted that announcement

anticipation and lagged FFR effects are ignored. This implies, for example, that the drops observed

in F I G U R E 3 could merely represent a delayed consequence of the sharp FFR cuts at the end

of 2008 rather than a consequence of announcements on monetary easing. Or it could be that, in the case of F I G U R E 4, anticipations have caused stock effects to be of signs that are in contrast

to expectations. Empirical results should point out what the actual announcement effects are.

F I G U R E 5 plots the development of the percentage change in auto loan demand as

perceived by loan officers at commercial banks. It clearly shows higher automobile loan rates in quarters of low demand and lower charges in others, indicating a strong relation between the two. Reasonably, it should be noted that this figure only shows developments of the 48-month interest rate and auto loan demand and thus does not correct for third variables as the effective FFR.

4.4 D E S C R I P T I V E S T A T I S T I C S

T A B L E 4 shows summary statistics for all relevant variables that are used in the debated models.

5. E M P I R I C A L R E S U L T S 5.1 S T A T I O N A R I T Y

The fact that most of the data from the FRED database display periodic fluctuations or linear trends – combined with the regular non-stationary nature of nominal interest rates (Neely & Rapach, 2008) – potentially require the debated models to be first-differenced (Granger & Joyeux, 1980). Augmented Dickey-Fuller tests (Dickey & Fuller, 1979) are performed to measure the extent to which the dependent variables are non-stationary. T A B L E A .5 and T A B L E A .6 show the

results of these tests. The test statistic for a unit root in the interest rate on 48-month new auto loans (RATE_48) equals -0.72 with an approximate MacKinnon (1994) p-value of 0.84. Correcting

for seasonality reduces this p-value to 0.79, while compensation for both seasonality and trend lead

T A B L E 4 · Descriptive statistics on the variables of interest. Included are details on the Fed’s SOMA holdings

(SOMA) for the flow effect model (3); on the categorized dummy variables for the announcement comparison model

(4); and on the classified dummy variables for the program comparison model (5). Additional continuous variables are the remaining listed dependent and independent variables: the 48-month new automobile lending rate (RATE_48);

the 60-month new automobile lending rate (RATE_60); the combined average auto finance rate at AFCs (RATE_AFC); the stock of outstanding motor vehicle loans (LOANS); the effective FFR (FFR); the SOMA holdings of the Fed (SOMA);

the computed median credit score for auto loan applicants at commercial banks (MEDCS_BANKS); the overall median

credit score of auto borrowers (MEDCS_OVERALL); the net percentage change in automobile loan demand perceived by local officers of US commercial banks (DEMAND); and the CPI for all consumers and on all items (CPI).

variables observations mean standard deviation minimum maximum

29 to a p-value of 0.51. In all cases, the null hypothesis of a unit root cannot be rejected, so the data

are indeed likely to be non-stationary. Differencing the data solves the non-stationarity issue for

RATE_48, as is proved by a p-value close to zero for the differenced data (not shown in the tables).

The Dickey-Fuller test for the 60-month new car loan interest rate (RATE_60) has a test statistic

of -1.98 (p = 0.29) and of -2.60 (p = 0.09) after correction for seasonality. Inclusion of both the trend and seasonality in the test gives a positive test statistic (p = 1.00), so the null can again not be rejected at the 5% level for each of the cases. Meanwhile, series of the average auto finance rate at AFCs (RATE_AFC) have a Dickey-Fuller test statistic of -4.31 with an approximate p-value of

0.00, so the null hypothesis of a unit root can be rejected. However, once the test corrects for seasonality and trends, auto finance rates at AFCs are found to be non-stationary too. Finally, this also holds for the outstanding stock of loans (LOANS). Data on loans have a test statistic still

equal to 0.75 after compensation for seasonality and trends. Loan volumes are strongly correlated with their four-quarter lags (i.e., their relation is significant at the 1% level) and have a trend significant at the 5% level. All results imply that the models for each of the dependent variables have to be first-differenced. Tests on the modified dependent variables indicate that the new data do no longer suffer from non-stationarity problems.

5.2 S T O C K E F F E C T M O D E L

A result of the preceding discussion is that the stock effect model (2) is transformed into the model

∆𝑦𝑡= 𝛽1∙ ∆FFR𝑡+ 𝛽2∙ ∆FFR𝑡−1+ 𝛿1∙ ∆QE1+ 𝛿2∙ ∆QE2+ 𝛿3∙ ∆QE3+ 𝛿4∙ ∆QE4+ 𝛿5∙ ∆QE5+ 𝛿6∙ ∆QE6+ 𝛿7∙ ∆QE7+ 𝛿8∙ ∆QE8+ 𝛿9∙ ∆QE9+ 𝛿10∙ ∆QE10+ 𝛿11∙ ∆QE11+ 𝛿12∙ ∆QE12+ 𝛿13∙ ∆QE13+ 𝛿14∙ ∆QE14+ 𝛿15∙ ∆QE15+ 𝛾 ∙ ∆F𝑡+ ∆𝜀𝑡 (6) for each of the dependent variables RATE_48, RATE_60, RATE_AFC and LOANS. As can be noticed, the

constant 𝛼 is removed in the first-difference estimator. In the first instance, the set of control

variables F𝑡 only includes the variables of section 4.2: the median credit score (MEDCS_BANKS or

MEDCS_OVERALL); the net percentage increase in auto loan demand (DEMAND); and the consumer

price index for all consumers and on all items (CPI). This implies that 𝛾 ∙ ∆F𝑡= 𝛾1∙ ∆DEMAND𝑡 + 𝛾2∙ ∆MEDCS_BANKS𝑡+ 𝛾3∙ ∆CPI𝑡 for RATE_48 and RATE_60; 𝛾 ∙ ∆F𝑡= 𝛾2∙ ∆MEDCS_OVERALL𝑡+ 𝛾3∙ ∆CPI𝑡

for RATE_AFC; and, finally, 𝛾 ∙ ∆F𝑡= 𝛾1∙ ∆DEMAND𝑡 + 𝛾2∙ ∆MEDCS_OVERALL𝑡+ 𝛾3∙ ∆CPI𝑡 for LOANS.

There is no evidence of further autocorrelation for the first-difference models. The models for

RATE_48, RATE_AFC and LOANS nonetheless include White (1980) robust standard errors, but then

analysis by means of the Breusch-Pagan (1979) and Cook-Weisberg (1983) heteroscedasticity test. This test shows statistics that are significant enough to reject the null hypothesis of constant variance at the 5% level for RATE_48, RATE_AFC and LOANS. The regression of the stock effect model

of RATE_60 does not suffer from heteroscedasticity, at least not significant at the 5% level.

Therefore, only models for RATE_48, RATE_AFC, and LOANS include robust standard errors.

Results of the stock effect models are depicted in T A B L E 5. As expected, changes in the

effective federal funds rate significantly explains variation in average 48-month new automobile loan interest rates (i.e., at the 1% level). Additionally, the FFR is significantly related with 60-month new car loan rates at the 5% level. There is no proof for a relation between the quarterly change in average auto finance rate at AFCs or outstanding car debt accumulation and effective FFR adjustments. Perceived demand changes only significantly explain ∆RATE_48 and have a

negative rather than a positive sign, possibly due to reverse causation. The median credit score of borrowers is not significantly related with any of the dependent variables. Remarkably, the linkage with the CPI is found to be significant for all dependent variables except for the finance rates at AFCs, but with different from expected signs. Inflation is found to correspond with lower interest rates and with higher stocks of outstanding loans. The opposite signs should be assigned to other channels than the expected inflation premia discussed in section 4.2. A possible explanation is that the monetary expansions from the Fed simultaneously lowered interest rates and boosted prices, such that the eventual correlation between the two is negative. Another potential reason may be rooted in the presence of inflation uncertainty, which made it hard for banks and finance companies to anticipate on future inflation in the period of the Great Recession (Golob, 1994).

The stock effect model finds weak evidence in favor of the main variables of interest, which are the 15 quarters of announcements by the Fed. Nine of the included dummy variables are significantly related with the 48-month new car loan interest rates, either at the 1% or the 5% level. Strikingly, only two of them – that is, announcements on the introductions of Operation Twist and Policy Normalization – have the expected signs from T A B L E 2 and T A B L E 3. The

31

T A B L E 5 · Results of the stock effect models. Estimated coefficients match with model (6) for all dependent variables.

Symbols refer to the 15 announcements of the tables. Robust standard errors are between parentheses in column (1), (3), and (4), default standard errors in column (2). Significance is represented by * _{p < 0.10,} ** _{p < 0.05,} *** _{p < 0.01.}

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

COEFFICIENT SYMBOL ∆RATE_48 ∆RATE_60 ∆RATE_AFC ∆LOANS

to bad times in the automobile market than that it is a result of the announcement of the first round of easing. The other six quarters of announcements, such as messages on pace reductions, are insignificantly related with the market for automobile credit.

Apparently, the US market for automobile credit did not really react to messages of the central bank, at least not via the pure expected announcement channels as plotted in the tables. The adjusted R2 _{has become low for each model after differencing and is even negative for LOANS,}

pointing to potential model weaknesses. Significant statistics on the Ramsey (1969) Regression Equation Specification Error Test (RESET test) on each model further hint at specification issues, which are not solved by logarithm transformations or inclusions of quadratic terms. These imperfections should be kept in mind when reading the results. However, the conclusion that there is no evidence for the first hypothesis remains. Later sections will elaborate on possible alternative explanations.

5.3 F L O W E F F E C T M O D E L

The stationarity issues described in section 5.1 lead to adjustments in the flow effect model into ∆𝑦_𝑡= 𝛽₁∙ ∆FFR𝑡+ 𝛽2∙ ∆FFR𝑡−1+ 𝜃 ∙∆SOMA𝑡+ 𝛾 ∙∆F𝑡+∆𝜀𝑡 (7) for all the dependent variables. The constant 𝛼 is again removed. All models include controls for

median credit scores, demand developments, and inflation, although the ∆RATE_AFC regression

reasonably does not correct for commercial bank demand changes. The null hypothesis of constant variance cannot be rejected for RATE_48, RATE_60, and LOANS when using the Breusch- Pagan (1979)

and Cook-Weisberg (1983) test. The first-difference model for RATE_AFC, however, suffers from

heteroscedasticity and hence includes robust standard errors. Residuals are also found to be serially correlated, with an alternative Durbin (1970) test statistic of 0.02, which leads to a significant rejection of the null hypothesis of no serial correlation at the 5% level. Despite first-differencing, residuals of the new model for LOANS are also serially correlated. The null hypothesis of no

autocorrelation in the alternative Durbin (1970) test is significantly rejected at the 1% level. This implies that the model for ∆LOANS also includes robust standard errors. Results on all models are

presented in T A B L E 6, together with the goodness of fit measures. Whereas the impact of the