Inside Information and Market Making in the Secondary Mortgage Market

Steve DruckerColumbia Business School

& Chris Mayer

Columbia Business School & NBER

Introduction

Banks are increasingly moving loans off their balance sheets through securitization Securitization issuance has grown from $125

billion in 1985 to over $3.4 trillion by the end of 2006 (Federal Reserve)

Over $2 US trillion in private mortgage-backed securities (MBS)

Risk-based capital and regulators have helped drive securitization

New financing tools such as CDOs and the flood of capital looking for yield has further increased demand for securities

Securitization divides responsibilities among several parties; may exacerbate information asymmetries

Loan Originator/Servicer

Seller / Underwriter

Loans

Trust (Bankruptcy-Remote)

Loan Payments

Bond Payments

(via Trustee)

Investors/Traders

PrivateInformation

Underwriters sometimes act as traders or investors

What do underwriters do?

Underwriters play a crucial role in arranging IPOs of $US 100’s of billions of stock and bond offerings annually

Provide crucial information for buyer of securities when companies and debt issuances are first sold to the public

Open question: what are underwriters expected to do after the IPO Clients expect underwriters to support their issuances

if initial demand is low Market participants apparently expect underwriters

to support their securities in a post-IPO marketplace

Examine trading of prime MBS

Look at secondary market trading of securities backed by pools of prime mortgages Prime mortgages are usually for borrowers with good

credit and who obtain mortgages with balances above GSE limits (non-conforming loans)

Sample avoids the riskiest sub-prime mortgages that you read about in the WSJ—our results thus might underestimate magnitudes in the broader mortgage market

Mortgages are predominantly originated between 2001 and 2005

Non-conforming mortgages represent about one-half of securitized mortgages (US$2 trillion in 2006)

Empirical questions

Q1) How effectively do underwriters make a secondary market in their own securities? Do they bid more frequently on their own securities? Are they more likely to bid when their securities are less

liquid?

Q2) Do underwriters use non-public information when bidding on their own securities? If they bid more frequently, is it because they are better

informed than other possible bidders? Do they avoid bidding on securities that will perform the

worst in the future? Do they bid more aggressively on securities that will

perform better?

What we find

Underwriters bid more frequently for their own securities (83% vs. 66%), but they avoid the least liquid issues

When they bid, they win twice as frequently (20% vs. 11%)

When underwriters do not bid or submit very low bids, securities perform significantly worse (conditional on observables) Three-fold increase in the likelihood that a loan transitions from

30-day to 90 day delinquent/default (3.3% vs. 9.5%) 43% decrease in the 6-month ex-post prepayment rate (7.4% vs.

4.2%)

Some evidence that the underwriter’s winning margin predicts ex-post performance (but not for other winning bidders)

The securitization process

Mortgage Collateral

AAA

AAA

BBBBBB

Unrated

Senior Class

Mezzanine Classes

First-loss pieces

Accumulation of Losses

Institutional investors are major buyers of first-loss pieces First-loss pieces are the most informationally sensitive part

of the capital structure

Unique data from a large institutional buyer Anonymous institutional buyer of non-

investment grade first-loss pieces (unrated, B, or BB rated)

Does not participate in any other part of the securitization process

Sometimes own higher tranches, but only if they also own the first-loss position

How do first-loss buyers make money? Avoid delinquencies/defaults as long as

possible Losses flow from the bottom up Collect coupon payments in the meantime at a very

high yield Examine origination and servicing information to

ensure the correct process was followed

Get as many prepayments in the pool as possible May reduce the number of loans at risk of default Lowest rated tranche may be upgraded

Security description as of auction date

Lowest rated tranche represents 0.2% of total collateral, but holds the risk for the entire securitization

Defaults on even one or two loans with collateral of over 1,100 mortgages can wipe out the first-loss position

Mean collateral size

Mean mtg. balance

Loss position: 1st 2nd 3rd

# Tranches in sample 106 99 36

Avg. tranche amount $1,102,052 $1,128,652 $1,297,220

$505,000,000

$548,371

Time line and trading

Trading information is typically unavailable, even to bidders Seller may choose to keep the security instead of selling it We have the bidder name and the bid amount on all

securities put up for sale, regardless of whether the offer was accepted

Mortgage Originations

Securitization

“Primary market”

Holding Period

Investor Requests Bids

3 days

Blind, First-Price Auction

“Secondary market”

Banks participate in various parts of the securitization and trading process

Trader only Integrated Underwriter onlyBarclays Bear Stearns ABN AmroCantor Fitzgerald Bank of America First HorizonCredit Suisse Chase Manhattan Wells FargoDescap Securities CitigroupGreenwich Capital CountrywideLehman Deutsche BankMorgan Stanley GoldmanMerrill Lynch UBS

Washington Mutual

Literature: Auctions with asymmetric information If all parties are equally informed based on public

information about securities All actions should be symmetric and traders earn zero

profits Some bidders may have a private signal, in which case

they can earn positive profits Alternative: Auctions with asymmetric information:

Informed party (I) competes against multiple uninformed bidders (U) who have access to the same public information (Engelbrecht-Wiggins, et. al 1983, Hendricks & Porter 1988)

Informed party (I) competes against multiple relatively uninformed bidders (U) who have a mix of public and private information, but whose private signal is less precise than the signal of the most informed party (Hausch 1987, Kagel & Levin 1999)

Bidding behavior of integrated banks

YesNo

Yes

Bid on Own Tranche?

Win Own Tranche?

40

40

Note: We threw out all bids with a yield more than 60% higher than the winning (lowest) yield

Securities issued by integrated banks

No

161

201

241

Integrated banks bid for and win their own securities at a much higher frequency #

OWN OWN Others OWN OthersIntegrated banks

Bank A 3 100.0% 67.2% 66.7% 15.6%Bank B 47 93.6% 79.9% 15.9% 5.8%Bank C 42 54.8% 51.8% 8.7% 7.8%Bank D 11 63.6% 30.0% 14.3% 0.0%Bank E 74 87.8% 59.3% 21.5% 14.1%Bank F 6 100.0% 70.6% 0.0% 8.4%Bank G 4 100.0% 85.2% 100.0% 18.3%Bank H 5 100.0% 80.5% 20.0% 6.8%Bank I 49 89.8% 66.7% 20.5% 13.3%Total 241 83.4% 66.0% 19.9% 10.8%TradersTotal

% Bid on Of Bid, % won

39.5% 9.3%

We also obtain data on underlying mortgages from LoanPerformance

Individual mortgage contracts that back the MBS tranches We use information from mortgage application, including

origination date, loan-to-value (LTV), FICO (consumer credit score), purpose of loan (cash-out, purchase, refi), housing type (single-family, condo), and mortgage type (fixed, floating)

Mostly 30-year mortgages Ex-post performance measures: 30 and 90-day

delinquencies, default/foreclosure, and prepayments

Deal-level data Deal id Underwriter

Integrated banks appear to bid on and win deals with similar ex-ante characteristics No Bid Bid, Lose Bid, Win

# Auctions 40 161 40

Origination Amount (a) $589,940 $569,319 $562,573Initial LTV 63.9% 65.3% 66.9%FICO Score 735.26 739.53 738.42

Mortgage Type

Single-Family Residence (a) 72.2% 76.4% 75.9%Condo 5.9% 5.0% 6.0%

Rate TypeFixed 79.0% 73.1% 62.1%

Mortgage PurposeHome Purchase 30.4% 29.8% 34.1%Refi 69.6% 70.2% 65.9%

(a) Denotes: No Bid ≠ Bid at 5% Level, (b) Denotes: Lose ≠ Win at 5% Level

No Bid Bid, Lose Bid, WinBid Information

Avg # bids (a) (b) 6.63 10.09 8.65Avg winning bid (% of par) 61.25 67.41 62.00Avg winning yield 19.09% 15.78% 18.71%

Mortgage PerformanceEx-Ante (6 Month)Delinq. rate (30 Day) 2.09% 2.06% 1.72%Delinq. rate (90 Day) 0.10% 0.08% 0.06%Prepayment rate 11.16% 13.09% 15.10%Ex-Post (6 Month)Delinq. rate (30 Day) 1.83% 1.74% 1.68%

Delinq. rate (90 Day) (b) 0.14% 0.11% 0.21%Prepayment rate (a) 8.77% 13.00% 15.76%

Integrated banks seem to bid on deals with more bidders and higher ex-post prepayments

(a) Denotes: No Bid ≠ Bid at 5% Level, (b) Denotes: Lose ≠ Win at 5% Level

What drives information asymmetries? Servicer is closer to the borrower and may have

additional information about potential losses Loan Performance does not list the property address or

certain borrower/loan characteristics like points paid Loans can be 30-day past due for benign reasons

(borrower was on vacation when payment was due) or for structural reasons (borrower was laid-off or had an illness)

We examine the transition from 30 to 90-days past due (or default) to examine whether integrated banks use their information in bidding

Servicers may also have additional observable information or insights that help predict the likelihood of prepayment

Informed bidders seem to avoid deals with the highest rate of progression to serious delinquencies

(a) Denotes: No Bid ≠ Bid at 5% Level(b) Denotes: Bid, Lose ≠ Bid, Win at 5% Level

No Bid Bid, Lose Bid, Win

% Loans 30 days delinquent (b) 0.70% 0.71% 0.50%

% 30 days delinquent that progress to 90+ days delinquent (a),(b) 11.5% 3.4% 7.5%

% Fixed rate mtg's 79.0% 73.1% 62.1%

Avg. LTV 63.9% 65.3% 66.9%

Avg. Fico 735.3 739.5 738.4

Integrated underwriters tend to avoid deals that progress to 30-days delinquent

Panel A

1Underwriter

BidsUnderwriter

WinsWinning Yield

Margin

Underwriter Winning Yield

Margin

1   0.002** -0.002** -0.004 0.174

  (2.08) (-2.49) (-0.63) (1.36)

Month vs. Trade Month: -4 0.002 -0.002 0.002 0.055 -0.005

(1.09) (-1.19) (1.23) (0.71) (-1.48)

Month vs. Trade Month: -3 -0.000 -0.000 0.001 0.002 -0.005

(-0.10) (-0.20) (1.14) (0.07) (-0.37)

Month vs. Trade Month: -2 0.000 -0.002 0.002 -0.005* 0.128

(0.19) (-0.56) (1.31) (-1.70) (1.15)

Month vs. Trade Month: 0 0.003** -0.001 -0.001* 0.093 -0.005**

(2.17) (-0.93) (-1.75) (1.38) (-2.28)

Month vs. Trade Month: 1 0.002** -0.002 0.000 0.016 -0.005

(2.00) (-1.45) (0.01) (0.50) (-0.86)

Month vs. Trade Month: 2 0.002 -0.002* 0.001 0.000 -0.005

(1.59) (-1.86) (0.96) (0.03) (-1.52)

Month vs. Trade Month: 3 0.003** -0.002*** 0.004** 0.003 0.062

(2.51) (-2.88) (2.01) (0.17) (0.44)

Month vs. Trade Month: 4 0.001 -0.001 0.003* 0.053 -0.005

(1.45) (-1.23) (1.67) (0.92) (-0.78)

Logit regression on all current loans in the 4 months before and after the auctionDependent variable: dummy=1 if loan becomes 30-day delinquent

Progression to serious delinquency varies by issuer, so we need to control for this in our regressions

Issuer

Informed No Bid

Informed Bid, Lose

Informed Bid, Win

Informed No Bid

Informed Bid, Lose

Informed Bid, Win

Bank B 9 155 37 0.0% 3.2% 2.7%Bank C 101 88 6 6.9% 3.4% 0.0%Bank D 18 23 4 22.2% 4.3% 0.0%Bank E 47 295 64 19.1% 5.4% 9.4%Bank G 0 0 44 . . 18.2%Bank H 0 30 6 . 0.0% 0.0%Bank I 33 432 39 12.1% 2.3% 0.0%Total 208 1023 200 11.5% 3.4% 7.5%

Loans 30 days delinquent at auction date

# Loans 30 Days Delinquent @ Trade Date

% Progress to 90 Days

Loans in deals that the integrated bank avoids are 3X more likely to become seriously delinquent

1 2Integrated bank bids -0.072*** -0.062***

(-3.81) (-3.32)

Integrated bank wins 0.0303 0.006 (1.41) (0.30)

Loan characteristics YES YESSale year FE's YES YESUnderwriter FE's NO YESWeighted logit model, with weights proportional to # loans in a security

Dep. var: 90+ day delinquent =1 | 30 days delinquent

Coefficients are marginal effects, Residuals clustered by securityN = 1,431

Anatomy of a Serious Delinquency

OCTOBER NOVEMBER DEC.

A B D

Miss Oct. 1st Payment CapturedBy Loan Performance: 30 Days Del.

Miss Nov. 1st Payment60 Days Del.

Trade D

ate

: Nov. 15 th

A. (Oct. 1st) Borrower misses monthly payment. Servicer, by definition, is immediately aware of all missed payments.

B. (Nov. 1st) Borrower misses second monthly payment. As of the 2nd missed payment, loan is now marked 30 days passed due in loan performance for the month of November. (Note: Agents do not have access to loan performance data until one quarter later.)

C. (Nov. 15th) Auction date. The market will not find out about the Oct. and Nov. 1st missed payments until the remittance reports come out on Nov. 24th (D); however, if the servicer and trader are part of the same bank, the trader might have access to the servicer’s Nov. 1st information.

F. (Dec. 1st) Loan performance marks the loan 60 days late on its Oct. 1st payment.

E

Underwriter avoids bidding on MBS with worst defaults in the precise month when this information is non-public

1Underwriter

BidsUnderwriter

WinsWinning Yield

Margin

Underwriter Winning Yield

Margin

1   -0.000 0.028 -0.059 0.113

  (-0.08) (1.50) (-0.77) (0.18)

Month vs. Trade Month: -4 0.023 -0.001 -0.027 0.921 0.941

(0.47) (-0.05) (-1.41) (0.68) (0.83)

Month vs. Trade Month: -3 -0.000 0.023 -0.011 0.254 0.795

(-0.14) (0.66) (-0.36) (0.26) (0.40)

Month vs. Trade Month: -2 -0.000 0.012 -0.034* 0.107 -0.060

(-0.02) (0.24) (-1.71) (0.21) (-0.63)

Month vs. Trade Month: 0 -0.031 0.024 -0.007 -0.064 -0.055

(-0.94) (0.56) (-0.69) (-0.34) (-0.44)

Month vs. Trade Month: 1 0.201*** -0.065*** -0.023 0.941 0.941

(2.82) (-3.09) (-0.65) (1.56) (1.53)

Month vs. Trade Month: 2 0.034 -0.009 -0.022 0.016 0.945

(0.79) (-0.28) (-0.67) (0.02) (1.19)

Month vs. Trade Month: 3 0.055 -0.019 -0.031 0.643 0.921

(1.23) (-0.94) (-1.41) (0.76) (0.45)

Month vs. Trade Month: 4 0.050 -0.016 0.017 0.901 -0.060

(1.06) (-0.60) (0.45) (0.91) (-0.38)

Logit regression on all 30-day delinquent loans in the 4 months before and after the auctionDependent variable: dummy=1 if loan becomes 60-day delinquent

Payoff rate varies by issuer, so we need to control for this in our regressions

Issuer

Informed No Bid

Informed Bid, Lose

Informed Bid, Win

Informed No Bid

Informed Bid, Lose

Informed Bid, Win

Bank A 0 689 1,378 . 5.7% 5.7%Bank B 1,924 31,872 6,842 10.9% 12.0% 13.4%Bank C 14,059 14,152 1,006 2.4% 5.3% 4.2%Bank D 2,200 3,952 759 2.6% 4.5% 3.2%Bank E 6,141 40,096 8,217 7.6% 8.1% 20.0%Bank F 0 1,638 0 . 5.7% .Bank G 0 0 11,180 . . 4.6%Bank H 0 5,890 511 . 6.7% 51.5%Bank I 5,503 46,065 9,987 5.1% 9.6% 10.8%Total 29,827 144,354 39,880 4.5% 9.0% 11.4%

Loans at auction date

# Loans @ Trade Date % Payoff in 6 Months

Integrated banks bid on pools whose loans subsequently prepay at a much faster rate

1 2Integrated bank bids 0.033*** 0.032***

(4.66) (3.71)

Integrated bank wins 0.007 0.009(0.86) (1.14)

Prepayment rate, prev 0.018*** 0.017***6 months (3.71) (3.67)

Loan characteristics YES YESSale year FE's YES YESUnderwriter FE's NO YES

Dep. Var.: Payoff in 6 Mo.

Weighted logit model, with weights proportional to # loans in a securityCoefficients are marginal effects, Residuals clustered by securityN = 214,061

Underwriters appear to predict payoffs and incorporate this into their bids more than traders do

1Underwriter

BidsUnderwriter

WinsWinning Yield

Margin

Underwriter Winning Yield

Margin

1   -0.001 -0.002 -0.002 -0.019**

  (-0.24) (-1.09) (-0.05) (-2.30)

Month vs. Trade Month: -4 0.003 0.002 0.003 0 0.96**

(0.65) (0.46) (0.95) (-0.01) (2.48)

Month vs. Trade Month: -3 -0.003 0.008** 0.003 -0.016 0.82*

(-1.11) (2.14) (1.28) (-1.03) (1.75)

Month vs. Trade Month: -2 -0.001 0.005* -0.002 0.022 0.023

(-0.39) (1.82) (-0.92) (0.71) (0.51)

Month vs. Trade Month: 0 0.001 0.000 0.003 0.199 0.022

(0.35) (-0.14) (0.93) (1.65) (0.29)

Month vs. Trade Month: 1 -0.003 0.003 0.004 0.037 0.78**

(-1.15) (1.08) (1.10) (0.60) (2.00)

Month vs. Trade Month: 2 -0.007*** 0.011*** 0.004 0.03 0.752*

(-3.08) (3.04) (1.14) (0.41) (1.72)

Month vs. Trade Month: 3 -0.009*** 0.013*** 0.003 -0.015 0.86**

(-3.13) (3.17) (1.19) (-0.59) (2.15)

Month vs. Trade Month: 4 -0.01*** 0.014** 0.005 -0.017 0.98***

(-3.02) (2.55) (0.99) (-1.03) (4.06)

Logit regression on all current loans in the 4 months before and after the auctionDependent variable: dummy=1 if loan pays off

Conclusions

Underwriters bid more frequently for their own securities (83% vs. 66%)

When they bid, they win twice as frequently (20% vs. 11%)

When underwriters do not bid or submit very low bids, securities perform significantly worse Three-fold increase in the likelihood that a loan transitions

from 30-day to 90 day delinquent/default (3.3% vs. 9.5%) 43% decrease in the 6-month ex-post payoff rate (7.4% vs.

4.2 %) Underwriter avoids bidding on MBS with worst

defaults in the precise month when this information is non-public

The winning margin of the underwriter, but no other winner, helps predict ex-post payoffs

Policy discussion

Underwriters exploit informational advantages May help explain the trend toward vertical integration

between originators, servicers, and underwriters/traders Suggests that underwriters do not support deals that, ex-

post, have the greatest problems (no implicit recourse)

May help explain part of the demise of CDOs is they were the marginal buyers of the worst performing MBS

Information asymmetries are important for MBS Hard to interpret secondary market prices as reflecting

the true underlying value of a security Regulators need to understand that this market may not

be so liquid in a downturn