Bundling Cars and Car Loans: Interest Rate Markups and Car Prices Connan Snider UCLA This Version: September 2013 (THIS DOCUMENT IS A PRELIMINARY SKETCH PLEASE DO NO CITE OR DISTRIBUTE)

Abstract On the order of 25% of new car dealership pro…ts come from arranging …nancing for its customers. These pro…ts come from commissions, paid by the banks and …nance companies ultimately providing the loan, and from dealers charging car buyers an interest rate above the rate o¤ered by these banks and …nance companies-a practice known as dealer reserve. In this paper we emprically analyze the auto …nance market, focussing particularly on quantifying the importance of dealer reserve.

Department of Economics, UCLA, www.econ.ucla.edu/faculty/regular/Snider.html

[email protected],

1

ph:

(310)

804-7574.

1

Introduction

The majority of new car purchases in the U.S. are …nanced using dealer arranged or indirect …nancing. After negotiating the price of the car and the value of the trade in, if any, the dealership sales person passes the customer to a Finance and Insurance (F&I) manager. If the customer is …nancing the purchase through the dealership, the F&I manager …nds an outside bank or …nance company willing to …nance the purchase and o¤ers customers additional products such as insurance, service contracts, after market modi…cations, etc..

Dealerships are compensated for the loans

they originate either through …xed commissions or, more commonly, by being allowed to o¤er the customer an interest rate above the rate o¤ered by the lender, the "buy rate", and keeping the di¤erence, a practice known as dealer reserve.

Remuneration for arranging …nancing is an

important source of pro…ts for dealerships, accounting for on the order of 25% of total dealership pro…ts. Dealer reserve has been a controversial practice for many years and has led to numerous lawsuits, largely by plainti¤s alleging disclosure violations under the auspices of the Truth in Lending Act (See Hudson, Benoit, and Looney 1999 for a discussion of several cases …led in the 1990s). On the other hand, dealer …nance has largely been exempted from the new …nancial regulatory framework and, in particular, the Dodd-Frank regulatory regime.

In March 2013, however, the Consumer

Financial Protection Bureau (CFPB) indicated, in spite of its inability to regulate dealer …nance directly, that it would pursue banks lending through dealerships by making these banks liable, under Equal Credit Opportunity Act (ECOA), for di¤erential lending terms o¤ered to women, minorities, and other disadvantaged groups identi…ed by the ECOA. In its policy guidance the CFPB (CITE) suggests indirect lenders prohibit dealer reserve in order to avoid enforcement action and many commentators have suggested this is the ultimate goal of the new enforcement posture. In this paper we empirically analyze the indirect auto …nance market with the primary goal of understanding the role of dealer reserve and forecasting the impact of its elimination in favor of ‡at commissions.

To do this, our analysis focuses on the unique aspects of dealer arranged …nance.

Namely, we focus on the fact that …nance products sold by a dealer are always bundled with the dealer’s auto product and often with other products like service and insurance contracts. This fact di¤erentiates dealer arranged …nance from its cousins like mortgage brokerage, for which marking up buy rates is now prohibited (CITE). The bundling of the durable good, …nancing, and add ons like insurance and service contracts means that the pricing margins for each component are

2

interdependent. Regulating one margin will necessarily have implications for the other margins. Moreover, eliminating dealer reserve may have implications for the incentives faced by banks in setting buy rates and dealer compensation as well as incentives faced by F&I managers to search for good …nance terms for their customers. In part because of increasing legal and regulatory scrutiny, some individual lenders and dealerships have started moving towards ‡at compensation schemes.

Credit unions, in particular,

have tended to rely on ‡at commissions because of their mission of serving the interests of their, often narrow and cohesive, client base. Using an large data set of auto transactions, we exploit this variation in compensation practices to identify the impact of dealer reserve on the disparate margins of an auto transaction. In doing this we face several identi…cation challenges. First, in all markets, lenders o¤ering ‡at compensation schemes compete with lenders allowing dealer reserve. This means that dealers will tend to select lenders o¤ering ‡at compensation when incentives to do otherwise are weak, i.e. when the markup would be lowest. Second, whether or not a bank allows dealer reserve will depend on how much competition it faces and the nature of that competition, though, a priori, it is unclear how di¤erent levels of competition should a¤ect equilibrium dealer compensation. Finally, at the transaction level, consumers who negotiate poorly on one dimension of the transaction are likely to negotiate poorly on other dimensions as well so, for example, high interest rate markups will be correlated with high vehicle markups. In order to overcome these identi…cation challenges we combine data on auto transactions with instruments constructed from data on local bank market structure. The time sensitive nature of the F&I managers job in closing a transaction means dealer-lender relationships are an important determinant in selecting the a …nancier and these relationships are often forged locally over o¢ ce visits and business dinners. On the other hand, since for most banks auto lending, and especially auto lending in a speci…c market, represents only a small share of its business, it’s unlikely elements of bank market structure, such as deposit and branch shares, are correlated with unobservable determinants of auto transaction elements. Additionally we exploit the panel structure of the data to aid in identi…cation. We …nd dealer reserve, relative to …xed compensation, is utilized more heavily in less competitive banking markets as measured by branch or deposit concentration. Interestingly, without accounting for compensation form, buy rates are lower in more concentrated banking markets, however, the sign of this relationship reverses when we account for the fact that these less competitive markets are more likely to use dealer reserve compensation schemes. 3

This is consistent with a story of

indirect lending competition in which banks compete for loans either by o¤ering …xed payments to dealers and earning a larger margin on the di¤erence between o¤ered buy rates and their own cost of funding or by earning a smaller margin over funding costs and o¤ering no direct payment, allowing the dealer to earn compensation instead by marking up the buy rate. We also …nd that …nal car prices, after trade in allowance and customer cash rebates, are negatively related to interest rate markups. Reduced form, OLS speci…cations suggest the average e¤ect of eliminating dealer reserve is a 1-1.5% increase in …nal car prices. Instrumental variable speci…cations, using bank market structure variables to instrument for compensation terms, that address the aforementioned selection and endogeneity problems show stronger results, indicating the average e¤ect of eliminating dealer reserve would be a 5-9%. We delve deeper into one of the determinants of …nal vehicle price, trade in allowance, which is the actual cash value of a transaction, as booked by the dealer, less the credit the customer is given toward the purchase. OLS regressions suggest the average e¤ect of eliminating dealer reserve on trade in allowance is a $50-60 decrease in the trade in credit relative to the actual cash value of the trade in. Instrumental variables regressions tell a much di¤erent story, however, implying a $880-1200 average increase in trade-in credit relative to the actual cash value of the trade-in if dealer reserve were eliminated. We further …nd evidence that lower dealer reserves are associated with more e¤ort in selling service contract products.

OLS estimates suggest the average e¤ect eliminating dealer reserve

would be associated with a 5-6 percentage point decrease in service contract sales. Instrumental variable estimates again tell a dramatically di¤erent story. These IV models suggest eliminating dealer reserve would be associated with a 20-25% increase in service contract penetration. Given the prevalence of dealer reserve in all markets-an estimated 50% of all transactions are associated with interest rate markups- a total elimination of dealer reserve takes us quite far out of sample so we urge caution in interpreting these reduced form predictions. In order to capture some of the likely behavioral changes in consumers, dealers, and banks we introduce a structural model that allows dealers to adjust prices and substitute between banks, banks to adjust lending terms, and customers to adjust car and …nancing choices (ONLY A VERY SIMPLE "COUNTERFACTUAL" IN THIS SKETCH) in response to an elimination of dealer reserve. We …nd, in the new equilibrium, buy rates increase .57 percentage points on average as banks need to increase the margin between o¤ered buy rates and their funding costs to make up for the fact that, in the counterfactual, dealers are compensated from the banks pro…ts as opposed to the consumer’s surplus. The equilibrium 4

dispersion of buy rates also increases because, in the absence of dealer reserve, dealers have less incentive to choose a lender on the basis of its o¤ered buy rate.

The result is that, in the new

equilibrium without interest rate markups, customers on average only see a .25 percentage point decline in APR’s.

This is in contrast to a naive estimate of .85 percentage points, which is the

average interest rate markup. Moreover, roughly 40% of customers actually see higher APR’s. There is a very long literature in industrial organization focusing on various aspects of auto markets.

However, the literature on the vehicle …nance market and it’s relationship with the

product market is scant. (ADD CITES) The paper proceeds as follows: Section 2 describes the data. Section 3 discusses the new car …nancing market. Section 4 presents the empirical analysis. Section 5 presents our counterfactual analysis of a prohibition on dealer reserve.

2

Data

The dealer data are from a major automotive market research …rm and comprise a 20% sample of all new vehicle dealerships in the US. All new car transactions from a given dealer are recorded in the data. The data available for this study are disaggregated but not individual transaction level data. The data base allows users to create bins of transactions based on speci…ed categories and reports all bins for which there are 3 or more transactions. In trading o¤ this truncation/granularity problem, an observation in our primary …nancing data set is determined by the following identi…ers: Year Market-The database tracks 31 major dealer markets Vehicle Nameplate - E.g. Ford, Nissan, etc. Contract price by $2000 increments-The contract price of a transaction is price of the car negotiated prior to accounting for the trade in credit if applicable. In the absence of more detailed car characteristics, this identi…er is used occasionally to control, along with Nameplate, for product heterogeneity. Buy rate by 50bps increments- This is the "wholesale" rate o¤ered by the bank or …nance company to …nance the purchase.

In the absence of credit scores these bins are used to

control for individual level heterogeneity. 5

Lender - The name of bank/…nance company providing the loan Captive Indicator -Whether or not a lender is acting as a subsidiary of a particular auto manufacturer Type of Sale- (IN THIS SKETCH) We limit attention to purchases (as opposed to leases) that are …nanced through the dealer. For each observation of one of these bins the number of transactions as well as the average values of a large number of transaction characteristics are reported if the bin has more than 3 associated transactions.

Even at this level of aggregation many transactions from the full data base are

truncated leading to an oversampling of high volume lenders, in particular, captives that are more likely to have 3 or more transactions in a given bin. Table 1 summarizes the dealership data. Our bank data come primarily from publicly available sources and were compiled in the data base of SNL Financial, a market research …rm. Bank branch level data is derived from the FDIC Summary of Deposits (SOD), which contains information on branch locations and annual deposit totals for every bank and credit union branch that is part of the FDIC system. These data are used to construct measures of bank level competition such as Deposit Share, Branch Share, Deposit Herfindahl, Branch Herfindahl, and are further interacted with bank level …nancial data. Bank level data is derived from the Reports on Condition and Income (CALL reports) published by the Federal Reserve and FDIC. These data contain detailed information on assets and liabilities of all commercial banks, among many other variables. We use these data to construct measures of the importance of consumer and auto lending for a given bank and interact these measures with the geographically varying deposit data. A version of the call reports are also available for Credit Unions and these include even more relevant measures, for example, they include measures of the relative importance of indirect and direct lending for any given credit union. Finally, we use SNL’s proprietary interest rate data to construct measures of local interest rates (NOT USING THESE RIGHT NOW). This data is collected as part of a "secret shopper" program, run by SNL, in which SNL employees approach di¤erent branches and inquire about various bank interest rate products. These products include all standard deposit, mortgage, and, most importantly consumer loan products including the standard auto loan products. summarizes the variables we exploit in the bank data.

6

Table 2

3

New Car Financing

The auto lending market is large and represents a major share of new consumer credit. In 2012 there were around $750bn in new auto loans outstanding. To put this number in perspective, in 2012, auto loans represented over one quarter of total, non-mortgage consumer credit outstanding and was about 8% as large, in terms of outstanding debt, as the massive residential mortgage market. The indirect lending channel is also, by far, the most important in new auto loan originations, accounting for 70-80% of auto loan originations in 2012. The remainder of auto transactions are leases (about 20%), cash transactions, or are direct …nanced, for example when a customer comes with a pre-approved loan from their bank. The most important indirect lenders are captive …nance companies, comprising over half of the indirect lending market. Captives, often subsidiaries of a manufacturer, are dedicated to …nancing the products of a particular manufacturer.

Each of the major manufacturers have a dedicated

captive subsidiary, though some captives are not entirely owned by the parent manufacturer. As discussed in Schultz-Mahlendorfer (2013), the role of captives is intimately connected with the bundling of product and product …nancing done at the dealership. They frequently o¤er special subsidized interest rates, e.g. 0% apr deals, to prime borrows and extend credit to subprime borrowers when other banks will not in an e¤ort to "move metal" and earn margins on the cars produced by their parent companies.

3.1

Banks and Indirect Lending

Commercial banks, especially large banks, and credit unions account for roughly one quarter of the overall indirect lending market, while independent …nance companies account for the remaining 20%. As shown in tables 3a-c there is substantial variation in the breakdown of …nancing sources across markets, nameplates, and customer credit worthiness. As shown in Table 3a, captives are responsible for almost all transactions at buy rates below 1% as these largely represent 0% apr promotions that only captives, with an interest in product sales, have incentive to participate in. Local banks, i.e. those with branch presence, command a substantial share of "prime" customers- those qualifying for interest rates between 1-8%.

Credit unions also achieve their highest penetration

rates at these levels, though clearly overall credit union penetration is low. It’s natural to think of the indirect lending market, from the perspective of dealers, as a large search market.

There are a large number of potential lenders for any given transaction but the

7

dealer is unsure which will give a given customer the best deal with the most compensation for himself.

Personal and local relationships then become important in determining which banks

are contacted and ultimately awarded the business originated by a dealer.

Under the heading

"Managing Bank Relations," one popular F&I Training manual says: "I suggest you invite all of your bank reps for one-on-one meetings to gather information and get acquainted. (Oh yes, the fun and free stu¤.)

The long friendly

discussions, the contests, the golf tournaments, the free lunches, the perks. As a Salesperson, I thought the Finance Manager only drank from free co¤ee cups and only wrote with free pens. I though [sic] he hadn’t paid for lunch in years. Man did he have it made!...Here’s the point. Bank relationships are very important." Local deposit shares and local branch presence is likely to be correlated with bank rep presence since loan o¢ cers often do double duty on the direct and indirect lending sides of the business. For any given level of local presence, the overall consumer lending presence of a bank is likely to matter as well. First, a large consumer lending operation is likely to mean more local bank reps all else equal. Second, for consumers with standard, good credit histories, these large consumer lenders may o¤er dealers speed and convenience in approving …nancing at consistently available terms. Moreover, the volume of business done by these banks allows these terms to be very competitive. Table 4 shows the results of a regression of a lender’s share of transactions in a year-market-buy rate bin in the JDP data on measures of local market and consumer lending presence in the banking market. The table con…rms that local branch and deposit market share is strongly associated with indirect lending share. Most of the e¤ect occurs whenever a bank has any deposit presence in the market (d_s) but this e¤ect also increases in the level of deposit presence. Similarly, measures of the importance of consumer and indirect lending at the bank level are positively correlated with indirect lending, though these parameters aren’t as stable across speci…cations. Local bank market is structure is also likely to a¤ect the buy terms and compensation o¤ered to dealers by banks and …nance companies. It’s not immediately clear the e¤ect indirect lender competition should have on equilibrium compensation mix.

Dealer reserve allows dealer to use

information not available to the buying bank to take it’s compensation from consumer surplus costing the bank less for a given amount of compensation, however, it introduces an ine¢ cient double marginalization problem. Table 5 foreshadows the …rst stage of our IV strategy and shows the results of a regression of interest rate markups on measures of bank market structure. 8

The

table excludes captives.

Less competitive markets, as measured by higher deposit market HHI,

are associated with higher interest rate markups suggesting that bank competition manifests itself in indirect loan competition by pushing banks to o¤er a larger fraction of compensation out of their own pro…ts.

After controlling for lender identities, the interaction of deposit share and

HHI is also positive, further suggesting market power increases the use of dealer reserve.

The

importance of consumer lending for a bank (cons_f rac) is negatively correlated with interest rate markups.

This would make sense if these banks specialize in quick approvals and standardized

terms for standard, prime credit pro…les, provided we don’t see these banks o¤ering higher levels of …xed compensation. Consistent with anecdotal evidence, credit unions are associated with smaller average markups. On average credit union interest rate markups are 35-50 basis points lower than non credit unions, which corresponds to a 40-70% decrease in the overall average markup. A bank’s o¤ered buy rates may also be a¤ected by competition among banks. Abstracting from risk, the pro…tability of a bought loan depends on the di¤erence between the buy rate and a bank’s cost of funds and the ‡at commission paid by the bank to the dealer. Several factors then interact in determining equilibrium buy rates. In less competitive markets we expect a bought loan to be more pro…table, however, this can be achieved in two ways. One way is through higher buy rates as banks earn a higher margins over their cost of funds.

A second way, suggested by the Table

5, is that, in less competitive markets, dealer reserve may be a relatively more important source of dealer compensation. Since a dealer’s compensation depends directly on the buy rate when it is compensated through dealer reserve but only indirectly, by making it more likely the consumer direct …nances or walks away entirely, when it receives ‡at compensation. Table 6 shows the e¤ect of competition on equilibrium buy rates. Column 1 of the table shows the result of the regression of buy rates on the basic bank competition variables. Notable here is the large and signi…cant negative e¤ect of deposit HHI on buy rates. Columns 2-5 include measures of average market-buy rate interest rate markups. Average markups are positively correlated with individual markups and are unlikely to su¤er from the same endogeneity problems associated with including measures of own interest rate markups, which are functionally related to buy rates. In these speci…cations the relationship between HHI and buy rates is positive suggesting the reason for negative association between HHI and buy rates in column 1 comes from the fact that dealer reserve is used more heavily in less competitive banking markets and buy rates are lower in markets where dealer reserve is used more heavily.

Column 6 of the table ignores the clear endogeneity

problems and includes own interest rate markups. 9

Since, in principle, banks can increase the amount of …xed compensation o¤ered to dealers in the absence of dealer reserve, it’s not clear how important dealer reserve is for ultimate dealer compensation. However, since compensation comes from appropriating consumer surplus under a dealer reserve scheme, our intuition suggests that equilibrium dealer compensation should be more lucrative. Table 7 shows the results of regressions of a measure of …nance reserve income on markups and competition and other controls. The dependent variable we use is the ratio of …nance reserve income to amount …nanced in order to avoid the composition e¤ects that arise because higher car prices and larger loans are associated with greater compensation. As is clear from the table, binaverage interest rate markups are strongly positively associated with overall compensation. A one percentage point increase in markup is associated with 1.6-1.7 cents additional dealer compensation for every dollar …nanced.1 The picture presented by the table is consistent with the story told by the previous two tables. Less Competition, as measured by HHI, is negatively related to compensation once we take account of compensation form but less competitive markets tend to rely more heavily on the more pro…table compensation scheme, dealer reserve.

3.2

The Relationship Between Finance, Auto, and Service/Insurance Products

The extent to which dealers trade o¤ the margins between elements of the bundle of car, …nance, and service contract/insurance products depends on the extent to which there are complementarities between elements of the bundle.

If complementarities are weak then pricing di¤erences due to

customer heterogeneity are likely to swamp di¤erences due to bundle pricing incentives and our estimated e¤ects will be weak. There are several aspects of a new car purchase that reinforce bundle pricing incentives. In many contexts consumers demonstrate strong willingness to pay for the convenience of making a single payment for several services (CITES; CHINTAGUNTA CABLE BUNDLES). There may also be cost complementarities for the dealer/manufacturer. For example, high service contract penetration means a steady volume of business for the dealership’s high …xed cost service department. As another example, there may be lower transaction costs for loans from dealer arranged banks because of well known policies for, for example, what happens in case of default. Some dealer sales techniques further institutionalize bundle pricing. F&I managers often practice so called "menu" and "package" selling. Menu and package selling involve the presentation 1

When dealer reserve is the sole source of compensation for a transaction this relationship is mechanical but if compensation for a particular transaction includes a dealer reserve and …xed component or some transactions in a bin include dealer reserve and others only …xed this is not so.

10

of a …xed set of …nance and service/insurance options with an associated monthly payment by the F&I manager. From a sales perspective, the idea is that by o¤ering a menu it’s more di¢ cult for the consumer to decline each of the options individually. The upshot, in terms of pricing, is that the F&I manager explicitly trades o¤ the di¤erent margins in determining a single markup for the menu item/package. A bank’s allowed "advance", or loan to value, is another important source of product interdependence. While banks di¤er in their policies regarding "front end" (car price) and "out the door" (car plus …nance products) advance, this means, in general, the higher the negotiated sale price the less room for selling …nance products.

Since loan to value is a driver of riskiness, buy rates are

also usually tied to the requested advance. So if the consumer is willing to put more down or the sales team is willing to give up gross by lowering the price, a higher margin can be charged on the loan.

4

Empirical Analysis

Though there is variation in the utilization of dealer reserves, the previous section makes it clear that in all markets, they are an important source of dealer pro…ts. A de facto ban on dealer reserve could be expected to have a large and cascading impact on many dealer, bank, and customer behavioral margins.

When dealer reserve is no longer available, the "one discriminatory rent"

hypothesis suggests increases on other margins, all else equal. F&I managers also may be inclined to work less hard to …nd good buy rates for their customers since the bene…t of an improved rate goes entirely to the customer.

Similarly, bank o¤ered buy rates may be forced to increase

to compensate for increased ‡at compensation which, as opposed to reserve based compensation, comes from the bank’s pro…ts instead of from consumer surplus. di¤erent loan terms may make di¤erent transaction choices.

Finally, customers, faced with

For example with lower interest

rates consumers may substitute away from direct …nancing sources or leases into dealer arranged …nancing and purchases. Our approach to estimating these e¤ects is to gradually peel back layers of these responses to trace out the likely and possible total e¤ects of the policy change. To do this we start by looking at the e¤ects of exogenously varying interest rate markups on vehicle prices, service contract income and trade in allowances. Our price measure of interest is the "out the door" price of the vehicle after customer rebates, dealer cash, trade in, etc.. For service and insurance contract sales we use 11

the sale penetration rate.

For trade ins, we use the trade in over/under allowance which is the

amount over or under the cash value booked by the dealer credited to the customer’s transaction. For each dependent variable we estimate the following linear speci…cation: Yjblmt =

0

+

+

K X

1 ir_markupjblmt k dk

+

1 Xjblmt

+

+

2 Ifmarkup 2 Dmt

= 0g +

3 Cred:U nion

+

4 Cred:U nion

Ifmarkup =(1) 0g

+ "jblmt

(2)

k=1

Where b indexes the buy rate bin, of the observation l is the lender, m the market, and t the year. We use the single index j to index a nameplate/contract price bin to indicate we think of these as being our representation of product heterogeneity. The coe¢ cients

1

4

are associated

with our endogenous regressors. We estimate separate e¤ects for credit unions because they appear to behave di¤erently and the set of circumstances that lead to a dealer choosing a credit union may be correlated with unobservables that also determine our dependent variables. The dk are a set of dummy variables controlling for product, market, lender, year, and buy rate bins. X is a set of control variables that include the dealer booked vehicle cost, days to turn (number of days on the lot), captive status, trade in cash value, and percent down. Dmt is a set of census demographic controls that include median income, percent blue collar workers, percent with a bachelor’s degree, percent black, and percent Hispanic. Table 8 shows the results of several speci…cations using …nal vehicle price, including customer cash rebates and trade in over/under allowance, as the dependent variable. The …rst 3 columns of table 8 are models estimated using OLS, while columns 4-6 are the IV counterparts. Each of the regressions include dummies for year, market, buy rate bin, nameplate, and lender identity for lenders with a su¢ cient number of total loans. Columns 2 and 5 include additional controls for the bin average trade in actual cash value as booked by the dealer, bin average percent of the transaction …nanced and the bin average percent down payment. Columns 3 and 6 additionally include the "contract price" of the car, which is the car price prior to adjusting for customer cash rebates and trade in allowance. We include this speci…cation to help further control for product level unobserved heterogeneity. Each of the speci…cations show the expected signs on the main variables of interest.

Higher

markups are associated with lower …nal prices with an additional increase in price associated with all of the transactions in a bin being …xed compensation transactions.

12

The OLS speci…cations

indicate the average e¤ect of eliminating dealer reserve (setting all interest rate markups to 0 and …xed to 1) would be a 1.1-1.7% increase in prices. The results also suggest car prices for vehicles …nanced by a credit union increase by 0-.5%. We use the previously discussed bank competition variables as instruments as well as interactions of these with deposit shares. As expected, the IV speci…cations show a signi…cantly larger e¤ect of interest rate markups on prices. These speci…cations suggest the average e¤ect of eliminating dealer reserve is a 5-9% increase for bank …nanced vehicles and 10-14% for credit union …nanced vehicles. Table 9 shows the results of several speci…cations taking trade in over/under allowance as the dependent variable.

Trade in allowance is the actual cash value of the trade in, as booked by

the dealer, less the amount credited to the customer so lower values of the dependent variable indicate a greater credit relative to actual cash value. Trade in allowance is one component of the …nal vehicle price so the table can be seen as digging deeper into the sources of the price change associated with interest rate markups. Again, the …rst 3 columns, show OLS speci…cations. The coe¢ cients on the variables of interest show results opposite to what we would expect.

Higher

interest rate markups are associate with smaller credits relative to actual values, while the dummy for all transactions in a bin having 0 markup is associated with greater credits relative to actual values. The implied average e¤ect of eliminating dealer reserve is a $50-$60 decrease in actual cash value minus credit. The instrumental variable speci…cations tell a very di¤erent story.

The signs of the e¤ects

reverse and the coe¢ cients increase signi…cantly. The implied average e¤ect of eliminating dealer reserve is a $880-1200 increase in the dependent variable. That is, assuming actual cash values are given, the average customer would expect to receive on the order of $1000 less for his/her trade in. Finally, table 10 shows the results of several speci…cations taking service contract penetration-the percent of transactions in a bin in which a customer purchases a service contract-as the dependent variable. The OLS speci…cations, again columns 1-3, show a minimal though signi…cant relationship between interest rate markups and service contract penetration.

The implied average e¤ect of

eliminating dealer reserve is a between 5-6 percentage point decrease in service contract penetration, inconsistent with our story of increased service contract sales e¤ort. The instrumental variables speci…cation again lead to a large change in estimated coe¢ cients. The coe¢ cients, however, have the same sign as in the OLS speci…cations including a positive coe¢ cient on the interest markup variable, which is not what we would expect. However, looking 13

at the implied average e¤ect of eliminating dealer reserve shows the larger coe¢ cient on the F ixed dummy outweighs the positive coe¢ cient on ir_markup and elimination is associated with a between 20-25 percentage point increase in penetration. As with all of these reduced form predicted average e¤ects, the magnitude of this e¤ect should be taken with a grain of salt since the imagined "treatment" is quite far out of sample.

4.1

Dealer-Finance Company Substitution Patterns

Changing dealer reserve policy is likely to cause dealers to do some substitution to other …nance sources, away from those lenders that are attractive because of permissive reserve policies.

For

example, it has been suggested that credit unions are "priced out of the market" because they are less likely to allow dealer reserve. We might then expect credit unions to be helped by a policy limiting dealer reserve. We use a latent pro…t discrete choice model demand model in the spirit of Berry (1994) to model dealer choice between lenders. Assume for each product-buy rate bin-market-period there is a set of available lenders, Ljbmt , for a dealer to choose from. For each such "market" there is also an outside option, called option 0, which we assume is a captive …nance company.

We assume the

captive is the outside option because our data set contains only customers that ultimately choose a dealer …nance option. Moreover, every customer that negotiates with and comes to an agreement with a sales person, even those paying with cash or other pre-arranged …nancing, is passed o¤ to the F&I manager who then can pitch an indirect …nance source. The dealer never chooses cash/direct …nance. Pro…ts of dealer i from choosing lender l 2 Ljbmt is given by: iljbmt

=

0

+

0

1 Fljbmt

+

0

2 Blmt

+

ljbmt

+ "L iljbmt

(3)

Where F is a set of variables describing the average terms o¤ered by lender l in market jbmt. We consider two sets of F variables.

One directly includes interest rate markup, buy rate, and

interactions and the other contains the normalized …nance reserve measure so that choice only indirectly in‡uenced by allowed markups.

B is a set of bank variables including measures of

branch presence, consumer, lending presence, market level competition, and interactions.

One

interpretation of these variables is that branch or consumer lending presence lowers the cost, to the dealer, of transacting with a particular …nance company. Another interpretation is that this model 14

is a reduced form of a search model where the probability of …nding a particular lender is a¤ected by it’s local and consumer focus characteristics. As in Berry (1994) the , represent unobserved lender characteristics. We assume these characteristics take the form:

ljbmt

=

K X

k dk

+

(4)

ljbmt

k=1

The dk are again dummy variables representing subsets of the characteristic bins and

ljbmt

are

iid, idiosyncratic "demand" shocks". The pro…ts associated with choosing the captive …nance option are assumed to take the form:

i0jbmt

=

0

01 F0jbmt

+ "L i0jbmt

(5)

Assuming the "’s are type 1 extreme value distributed the probability of dealer i choosing lender l is given by: Piljbmt =

exp( exp(

i0jbmt )

+

iljbmt )

X

(6)

exp(

il0 jbmt )

l0 2Ljbmt

Using the well known log odds inversion gives: ln(sljbmt )

ln(s0jbmt ) =

0

+

0

0

1 Fljbmt

01 F0jbmt

+

0

2 Blmt

+

K X

k dk

+

ljbmt

(7)

k=1

Table 11 reports the results of a regression of the di¤erence in log market shares on bin-average …nance terms and bank characteristics.

No dealer reserve in a bin (f ixed = 1) is strongly and

robustly negatively associated with dealer choice of a non captive bank.

Similarly when non-

captives compete with captives that allow no dealer reserve they are strongly advantaged. Columns 2 and 3 indicate average interest rate markups are positively correlated with bank choice. Column 1 shows the speci…cation most saturated with dummies.

Here, the size of average interest rate

markup is not signi…cantly related to choice. There are two plausible explanations for this. First, with the exception of, generally non-binding, caps on markups banks exert no in‡uence on the size of markups so, while we expect whether or not a markup is allowed to be strongly correlated with choice, the actual size of markups is likely to be related to other factors.

The second is related

to the nature of our data. Small shares are associated with a smaller number of transactions and thus greater sampling variability of average markups. Since markups are bounded below by zero, this measurement error is likely to be negatively correlated with share. This second fact is also

15

a leading candidate explanation for the fact that higher average captive markups are positively correlated with non-captive choice. Conditional on a buy rate bin, buy rates are strongly negative associated with dealer choice and captive buy rates are positively correlated with dealer choice. We allow this buy rate sensitivity to vary according to whether or not the customer is a "prime" borrower (qualifying for a buy rate less than 5%) and …nd that the coe¢ cient is almost double for good credit risks. This makes sense as the competition for good credit risks is more intense and these people are more likely to arrive at the dealership with a "pre approved" loan that a dealer must beat to earn the lending business. Consistent with our story of competition for indirect loans, dealers are less sensitive to buy rates from lenders who rarely or never o¤er dealer reserve, which shows up as a positive coe¢ cient on the …xed-buy rate (f br) interaction variable and a negative coe¢ cient on the interest rate markup-buy rate (irb) variable. Table 12a-b shows the distribution of elasticities with respect to buy rate implied by the estimated model by buy rate bin, i.e. interest rate markups don’t change in response to a change in buy rates. Table 12a shows the elasticities assuming zero pass through of buy rates to interest rate markups and Table 12b shows the same elasticities assuming complete pass through of buy rates to markups, i.e. interest rate markups respond one for one to changes in buy rates.

4.2

"Counterfactual"

As eluded to above, there would likely be a strong response to the elimination of dealer reserve from banks. For example, forced to provide more dealer compensation from their own pro…ts, we might expect the markup of o¤ered buy rates over cost of funds to increase. A standard way to capture this response would be to model the game played by indirect lenders and solve the game with the added constraint.

We instead take a reduced form approach and use the results of column 2 of

Table 6 to predict the bank response. We then plug these reduced form counterfactual buy rates and interest rate markups (all of which are assumed set to 0) into our choice model to account for dealer substitution between the various …nance sources. Table 13 shows the counterfactual shares of the di¤erent …nance sources by buy rate bin, corresponding to the original shares presented in Table 3a.

16

Captives gain some share of "prime"

customer …nancing, at the expense of commercial banks with branch presence and other …nance companies, as captive buy rates increase less in the counterfactual, though from a higher base, and associated average interest rate markups are considerably lower.

For the higher buy rate bins

captives actually lose share, however. This is because the associated average interest rate markups are more similar to banks for these "subprime" consumers are dealers less sensitive to the larger change in buy rates by the banks and …nance companies for these customers. Credit unions do gain share at all buy rate bins, however, this gain is small.

This results

from the fact, demonstrated by the choice model, that credit unions are somewhat undesirable for reasons other than their, on average, less generous dealer reserve policies.

Banks with branch

presence gain share among the subprime buy rate bins. This is because dealer buy rate sensitivity is lower for these customers and, with the …nance terms component of utility decreasing for both banks and captives, the …xed lender …xed e¤ects and presence measures buoy their utility. Tables 14a and 14b show the actual and counterfactual distributions of buy rates for the various buy rate bins. The counterfactual distribution accounts for both counterfactual buy rates and the substitution among …nance sources that occurs as a result of these changes as well as changes in markups. The average buy rate in the original data is 4.47%, while the average in the counterfactual distribution is 5.03%, a .56 percentage point increase. Looking at di¤erent points in the original versus counterfactual distribution shows that much of this average e¤ect is driven by two factors. One is a right shift in the distribution, resulting from overall increasing buy rates in the counterfactual.

The second is an increase in the dispersion of the buy rate distribution result-

ing from the sensitivity to buy rates decreasing in a world with no dealer reserve.

Accordingly,

the table shows a smaller di¤erence in actual and counterfactual buy rates at the low end of the distribution within each buy rate bin and across buy rates bins. Of course, from the perspective of customers who care about the APR’s they face, these increased buy rates are o¤set by the fact that dealers can no longer charge a markup over those buy rates. Tables 15a-b presents the counterfactual from the perspective of changes in APRs faced by customers. Table 15a presents the naive counterfactual, which assumes that lender buy rates and dealer choices would not adjust to the policy change, i.e. the change in apr’s is just the negative of the distribution of interest rate markups.

Table 15b shows the distribution of APR changes

assuming lenders o¤er the counterfactual buy rates and dealers adjust their choices according to the choice model. Table 15b shows that the counterfactual average change in customer apr’s is .55 percentage points lower than the naive analysis would suggest, though on average still negative. 17

Moreover, table shows that roughly 40% of customers see their APR’s increase (WANT TO MODEL BANK PRICING COMPETITION, CONSUMER SUBSTITUTION BETWEEN CARS/DIRECT FINANCE-LEASING-CASH (E.G. NESTED LOGIT ALA DAS GUPTA, SIDDARTH, SILVA-RISSO) , CAR PRICING COMPETITION, AND TRACE THROUGH THE ENTIRE EQUILIBRIUM EFFECTS)

5

Appendix:Not Obvious Variable De…nitions

cons_frac- consumer loans/total loans cons_intensity - cons_frac*deposit_share ind_intensity- indirect loans/total loans (only for credit unions) herf - deposit HHI for banks in FDIC SOD data (not just those that appear in JDP data) d_s - dummy = 1 if a bank has any branch presence in the market cu_mkt_share - deposit share of credit unions in FDIC SOD data d_* - deposit_share interacted with * (for all * but s) mean_ir_markup - transaction weighted average interest rate markup for all bins of a given year, market, buy rate bin combination *_sw_mkt - market, share weighted average of * among all banks in FDIC SOD and call report data

18

Table 1: Dealer Data Mean

St. Dev

5th Percentile

25th Percentile

50th Percentile

75th Percentile

95th Percentile

Contract Price 27460.78 8281.06 17021.00 21289.00 25381.00 31323.00 42978.00 Vehicle Price  25438.47 7894.69 15453.00 20054.00 24139.00 29203.00 40256.00 Customer Rebate 2393.46 1405.98 598.00 1238.00 2097.00 3377.00 4959.00 Rebate Penetration (%) 71.61 30.64 11.20 49.00 83.40 100.00 100.00 ‐298.79 925.06 ‐1830.00 ‐476.00 ‐79.00 108.00 540.00 Trade In Over/Under Allow Finance Reserve 702.21 326.24 266.00 481.00 650.00 865.00 1307.00 0.029 0.010 0.012 0.023 0.029 0.035 0.047 Finance Reserve/Amount Financed Flat Compensation* 0.110 0.313 0.000 0.000 0.000 0.000 1.000 Interest rate markup 0.800 0.537 0.000 0.360 0.810 1.190 1.690 Buy Rate 4.468 1.799 2.320 2.900 3.900 5.320 7.940 Sell Rate 5.267 1.965 2.900 3.830 4.960 6.270 9.160 Service contract penetration (%) 44.26 20.25 12.40 30.20 43.20 57.30 78.50 Service Contract Income 680.74 377.56 0.00 468.00 722.00 923.00 1247.00 Finance Product Income 858.96 500.87 129.00 498.00 816.00 1148.00 1755.00 Vehicle gross 805.08 804.38 ‐373.00 361.00 765.00 1197.00 2096.00 Vehicle profit(%) 2.79 2.67 ‐1.70 1.50 3.00 4.30 6.70 Source: JDPA  ; *‐Flat compensation defined as a bin where average apr is equal to average buy rate.  Some bins will have a positive average difference even though  some transactions had flat compensation

Table 2: Bank Data Summary Statistics Mean

St. Dev

5th Percentile

25th Percentile

50th Percentile

75th Percentile

95th Percentile

Credit union 0.024 0.152 0.000 0.000 0.000 0.000 0.000 Captive 0.646 0.478 0.000 0.000 1.000 1.000 1.000 Branches>0 0.255 0.436 0.000 0.000 0.000 1.000 1.000 Branches 888.546 1721.292 0.000 0.000 0.000 945.000 5119.000 Branch Share 0.015 0.033 0.000 0.000 0.000 0.001 0.094 Deposit Share 0.018 0.045 0.000 0.000 0.000 0.001 0.111 Consumer Loans/Total Loans 0.053 0.153 0.000 0.000 0.000 0.037 0.202 Auto loans/Total Loans 0.009 0.067 0.000 0.000 0.000 0.000 0.000 Indirect Loans/Total Loans (CU only)) 0.011 0.076 0.000 0.000 0.000 0.000 0.000 Deposit Herfindahl 0.101 0.036 0.048 0.072 0.106 0.126 0.156 Source: SNL Financial compiled from FDIC Summary of Deposits Reports and FDIC Reports on Condition and Income (CALL).  Summary of all indirect lenders in JDPA  data (including those lenders for which there is no bank data) weighted by number of dealer transactions

Table 3a:  Finance Source by Buy Rate  Buy Rate Bin 

Captive %

0‐1% 0.9935282 1‐2% 0.7717846 2‐3% 0.5670206 3‐4% 0.656401 4‐5% 0.595937 5‐6% 0.6225995 6‐7% 0.7568592 7‐8% 0.8609924 8‐9% 0.8993178 9‐10% 0.8800071 10‐11% 0.8234536 11‐12% 0.771259 >12% 0.4660453 Total 0.7233506 Source: JDPA and FDIC SOD 

. 

Credit Union % 

w/ Branch Presence % 

0.00000997 0.0183509 0.0226723 0.0202884 0.0274326 0.0329728 0.0169931 0.006474 0.0062783 0.0067523 0.0045653 0.0075109 0.0018305 0.0173765

0.0159584 0.1604932 0.3219652 0.2412119 0.2801431 0.2497805 0.1695759 0.0775635 0.0456142 0.0516156 0.0644015 0.086244 0.1095561 0.192529

Table 3b: Finance Source by Market Market

Captive %

Atlanta 0.7274778 Baltimore/Washin 0.7150043 Boston 0.6899363 California ‐ Nor 0.8658247 California ‐ Sou 0.8334367 Charlotte 0.6906208 Chicago 0.7068677 Cincinnati 0.7167852 Cleveland 0.6111111 Columbus 0.6883993 Dallas/Ft. Worth 0.7036218 Denver 0.7362637 Detroit 0.6616522 Houston 0.7208489 Indianapolis 0.6722509 Kansas City 0.7444968 Miami 0.7617799 Milwaukee 0.6253650 Minneapolis 0.6717400 Nevada 0.8213497 New York 0.7233171 Norfolk/Virginia 0.7894737 Oklahoma 0.6710249 Orlando 0.7923339 Philadelphia 0.6941200 Phoenix 0.6829024 Pittsburgh 0.6969471 San Antonio 0.4086679 Seattle/Portland 0.7300840 St. Louis 0.7160522 Tampa 0.7756386 Tennessee 0.8337332 Total 0.7233506 Source: JDPA and FDIC SOD 

Credit Union % 

w/ Branch Presence % 

0.0329781 0.0068721 0.0329936 0.0137557 0.0063012 0.0001692 0.0011264 0.0695881 0.0107254 0.0084358 0.0081138 0.0182139 0.0155677 0.0202264 0.0105807 0.0052058 0.0034794 0.0896399 0.0030218 0.0009793 0.0125898 0.0208652 0.0851265 0.0055586 0.0201555 0.0160210 0.0078570 0.0443173 0.0655932 0.0325810 0.0354038 0.0193211 0.0173765

0.1916748 0.2184082 0.1873500 0.1064242 0.1293591 0.2271734 0.2596721 0.2043602 0.3363748 0.2878447 0.1827967 0.0947100 0.2298864 0.1897913 0.2333147 0.0965938 0.1924032 0.2915059 0.1112451 0.2967565 0.2364884 0.1516631 0.2044464 0.1475612 0.2407201 0.2324633 0.1678785 0.0617449 0.1759287 0.1634990 0.1961954 0.1103368 0.1925290

Cash/Direct Finance %

Table 3c: Finance Source By Nameplate Vehicles

Captive %

Acura 0.9348062 Audi 0.8164832 BMW 0.9179334 Buick 0.6788611 Cadillac 0.7831445 Chevrolet 0.5529450 Chrysler 0.7103590 Dodge 0.6250727 Fiat 0.6416521 Ford 0.6554378 GMC 0.6689959 Honda 0.8415560 Hummer 0.8035451 Hyundai 0.5391715 Infiniti 0.7723755 Isuzu 0.0000000 Jaguar 0.1505376 Jeep 0.5368224 Kia 0.3788184 Land Rover 0.0135325 Lexus 0.8819911 Lincoln 0.7657944 Mazda 0.8099106 Mercedes‐Benz 0.7890414 Mercury 0.7067957 Mini 0.8890441 Mitsubishi 0.7440000 Nissan 0.7014251 Pontiac 0.6198352 Porsche 0.7172727 RAM 0.6467848 Saab 0.3891213 Saturn 0.7467652 Scion 0.9404209 Subaru 0.8595775 Suzuki 0.5578594 Toyota 0.8825376 Volkswagen 0.8244453 Volvo 0.6370107 smart 0.7592157 Total 0.7233506 Source: JDPA and FDIC SOD  . 

Credit Union % 

w/ Branch Presence % 

0.0028677 0.0032666 0.0003848 0.0057494 0.0028496 0.0272633 0.0266764 0.0364834 0.0066899 0.0169608 0.0144347 0.0100282 0.0044313 0.0280210 0.0007035 0.0000000 0.0000000 0.0358567 0.0435572 0.0000000 0.0030302 0.0215899 0.0182148 0.0004170 0.0306513 0.0010867 0.0198519 0.0157225 0.0179389 0.0000000 0.0262200 0.0000000 0.0212569 0.0069302 0.0212687 0.0158561 0.0105867 0.0085842 0.0238253 0.0109804 0.0173765

0.0515000 0.1658119 0.0735039 0.2570686 0.1771746 0.3150439 0.1969095 0.2109588 0.2652705 0.2336422 0.2323033 0.1108017 0.1137371 0.3025469 0.1817400 0.6666667 0.1626344 0.3109793 0.3216522 0.1564148 0.0975288 0.1682420 0.2627154 0.1862455 0.2058883 0.0868941 0.1542222 0.1887422 0.2396703 0.2609091 0.1869610 0.5104603 0.1814541 0.0687939 0.0993043 0.1779235 0.0966271 0.1359342 0.5342301 0.1411765 0.1925290

Cash/Direct Finance %

(1)

Table 4: Lender Indirect Lending Share on Competition (2) (3) (4)

(5)

(6)

VARIABLES deposit_share

d_s

cu

cons_frac

cons_intensity

ind_intensity

c_s

ind_frac

captive_ind

Constant

0.043*** (0.008) 0.000 [0.027 - 0.059] 0.030*** (0.001) 0.000 [0.027 - 0.032] -0.015*** (0.002) 0.000 [-0.020 - -0.010]

0.041*** (0.008) 0.000 [0.025 - 0.057] 0.015*** (0.002) 0.000 [0.010 - 0.020] 0.008 (0.007) 0.236 [-0.005 - 0.022] -0.040*** (0.007) 0.000 [-0.053 - -0.027]

0.047*** 0.091*** 0.090*** 0.095*** (0.008) (0.009) (0.009) (0.009) 0.000 0.000 0.000 0.000 [0.031 - 0.063] [0.073 - 0.109] [0.071 - 0.108] [0.077 - 0.113] 0.016*** 0.025*** 0.003 0.010*** (0.002) (0.001) (0.003) (0.003) 0.000 0.000 0.185 0.000 [0.012 - 0.021] [0.022 - 0.028] [-0.002 - 0.008] [0.005 - 0.015] 0.023*** -0.010*** -0.005 0.022*** (0.007) (0.003) (0.008) (0.008) 0.001 0.000 0.564 0.007 [0.009 - 0.037] [-0.015 - -0.005] [-0.020 - 0.011] [0.006 - 0.038] 0.037*** -0.030*** 0.095*** (0.012) (0.008) (0.012) 0.001 0.000 0.000 [0.015 - 0.060] [-0.045 - -0.015] [0.070 - 0.119] -0.088*** -0.144*** (0.011) (0.011) 0.000 0.000 [-0.109 - -0.067] [-0.167 - -0.122] 0.015 0.037*** (0.009) (0.011) 0.102 0.001 [-0.003 - 0.034] [0.015 - 0.058] 0.024*** 0.015*** 0.031*** 0.011*** (0.003) (0.003) (0.003) (0.003) 0.000 0.000 0.000 0.001 [0.019 - 0.030] [0.009 - 0.020] [0.025 - 0.037] [0.004 - 0.017] 0.011 0.029*** (0.009) (0.011) 0.265 0.008 [-0.008 - 0.029] [0.008 - 0.050] 0.435*** 0.434*** 0.429*** (0.006) (0.006) (0.006) 0.000 0.000 0.000 [0.423 - 0.446] [0.422 - 0.445] [0.417 - 0.441] 1.097 1.087 1.082 0.326 0.321 0.324 (483.962) (483.871) (483.793) (429.092) (429.018) (428.924) 0.998 0.998 0.998 0.999 0.999 0.999 947.457 - 949.65[-947.288 - 949.462]947.139 - 949.304-840.681 - 841.333-840.541 - 841.184-840.354 - 841.002

Observations 213,257 213,257 213,257 371,771 371,771 371,771 R-squared 0.345 0.345 0.346 0.679 0.679 0.680 R2 0.345 0.345 0.346 0.679 0.679 0.680 Include Captives no no no yes yes yes dummies y,v,p,b,m y,v,p,b,m y,v,p,b,m y,v,p,b,m y,v,p,b,m y,v,p,b,m Notes: Dependent variable: Share of year, market, nameplate, contract price bin and buyrate bin transactions. Y-year, v-nameplate, p-contract price bin, b-buyrate bin, m-market dummies. Standard errors in brackets *** p