Liquidity in a Market for Unique Assets: Specified Pool and · PDF fileLiquidity in a Market for Unique Assets: Specified Pool and TBA Trading in the Mortgage Backed Securities Market

1  

Liquidity in a Market for Unique Assets: Specified Pool and TBA Trading in the Mortgage Backed Securities Market

Pengjie Gaoa, Paul Schultza, and Zhaogang Songb

Abstract Agency mortgage-backed securities are traded simultaneously in a market for specified pools and in a forward (TBA) market. Cheapest-to-deliver pricing in the TBA market increases liquidity because buyers do not bear costs of analyzing the mortgage pool. Pooling is not complete in the TBA market though, prices differ across dealers. TBA trading also increases liquidity by allowing dealers to hedge specified pool inventory. Dealers hedge almost all specified pool inventory for pools that are closely matched by TBA contracts. Specified pools that do not match TBA trading as closely are hedged less frequently. Specified pools that match TBA characteristics are cheapest to trade.

August, 2014

Preliminary, do not quote. Not for circulation.

aMendoza College of Business, University of Notre Dame. bBoard of Governors of the Federal Reserve.

2  

Banks and other mortgage originators typically sell their mortgages as part of a pool of mortgages

in a mortgage backed security (MBS). The pooling of mortgages for MBS occurs in two ways. Large

originators can assemble a pool of conforming mortgages and issue them through one of three agencies:

Fannie Mae, Freddie Mac, or Ginnie Mae. Small mortgage originators will typically sell loans to the

agencies through the cash window, and the agency will poll them with other mortgages into an MBS.

Large originators may also do this if they don’t want to keep loans on their balance sheet until a pool is

assembled. In pooling mortgages, the agencies attempt to pool similar loans together. Agency MBS are

pass-through securities. All MBS issued on an underlying mortgage pool represent a proportional claim

on the principal and interest payments of the underlying loans. After their initial sale to investors, MBS

trade freely in the over-the-counter market.

The secondary market for MBS is among the largest, most active, and most liquid of all securities

markets. At first glance, it is surprising that the market is so liquid because each MBS is unique,

composed of specific mortgages with their own prepayment characteristics. In this paper, we study the

institutional feature of this market that allows it to work so well – its structure of parallel trading in a to-

be-announced (TBA) forward market in MBS and a specified pool market in which specific MBS are

traded.

Agency MBS are default-free. Despite this, MBS with the same coupon and maturity can differ

significantly in value because of differences in the risk of prepayment. Mortgages give the borrower an

option to prepay the loan. This option may be exercised when interest rates fall and borrowers can switch

to new lower rate mortgages. Conversely, if mortgage rates rise, borrowers who would otherwise prepay

and move may be more likely to remain in a house. The value of the prepayment options that mortgage

lenders have written vary from MBS to MBS depending on the ability and willingness of the mortgage

holders to prepay. Homeowners with low credit scores, or who hold mortgages with high loan-to-value

ratios, or who hold mortgages that have been recently refinanced are regarded as less likely to exercise

their prepayment option. MBS composed of mortgages like these are considered to be particularly

valuable. On the other hand, borrowers who take out large mortgages are thought to be more financially

sophisticated and more likely to exercise prepayment options. MBS with a lot of large mortgages are less

valuable.

It is difficult to model prepayment. Stanton (1995) notes that some mortgages are not prepaid

even when their coupon rate is below current mortgage rates. He also observes that there is seasonality in

prepayments and that there is a “burnout” factor in prepayment. That is, if mortgages haven’t been repaid

after a decline in interest rates, it indicates that the remaining mortgage holders in a pool cannot easily

3  

refinance and prepayments will decline. To produce estimates of prepayments that approximate empirical

observations, Stanton relies on the assumption that different mortgage holders face different and probably

unobservable costs of prepayment.

The idiosyncratic prepayment risk of individual mortgages may be reduced through

diversification when the mortgages are assembled into MBS, but MBS differ significantly in the

characteristics of the constituent mortgages. The average size of the loans, refinancing history,

homeowner credit scores, house locations, loan-to-value ratios and other factors differ significantly across

MBS’s and are related to their prepayment risk. Each MBS has a unique combination of these

characteristics and each therefore carries its own unique prepayment risk. A potential buyer or seller of

these securities needs to use the characteristics of the mortgages in an MBS to evaluate the prepayment

risk and determine its value. Many potential buyers of MBS do not have the expertise to evaluate

prepayment risk, and face a winners curse in this market.

TBA trading provides a solution to the winner’s curse problem. Specific mortgage pools are not

traded in the TBA market. Instead, buyers and sellers agree on six parameters; the price, par value of the

securities, settlement date, maturity, coupon rate and issuer (Fannie Mae, Freddie Mac, or Ginnie Mae).

Sellers are expected to deliver the cheapest pool that meets the agreed-upon parameters. By purchasing in

the TBA market, uninformed investors pay the same price as sophisticated investors. Every buyer in the

TBA market, regardless of sophistication, has the same likelihood of receiving a more or less desirable

MBS.1 Because of this, the TBA market for MBS is among the most liquid securities markets even

though each MBS is a unique asset with unique prepayment risks.

The TBA market functions so well in part because the MBS that trade there are relatively

homogeneous. This is because the pools with the lowest prepayment risk can be sold for higher prices in

the specified pool market. The MBS traded in the TBA market have similar high risk of prepayment.

The TBA market contributes to the liquidity of the specified pool market by allowing dealers to

hedge their specified pool inventory risk. Trades in the MBS market are large, often in the tens of millions

of dollars. Without active efforts to manage inventory, dealers can acquire large long or short positions

that will leave them exposed to risks from changing interest rates or changing values of prepayment

options. The highly liquid TBA market allows dealers to hedge the risk of specified pool positions with

offsetting positions in the TBA market.

                                                            1 Pagano and Volpin  (2012) consider a market for asset‐backed securities in which some potential buyers cannot evaluate information needed for pricing the securities. They show that opaqueness can make the market more liquid. 

4  

In this paper, we find that, as expected, prices of specified pools are higher than TBA pools with

the same coupon and maturity. Also as expected, we find that trading costs are much lower in the TBA

market than in the specified pool market. Specified pools that are eligible for TBA trading carry lower

trading costs. Specified pools with the same maturity and coupon as actively traded TBA contracts also

have lower trading costs than other specified pools.

An unexpected finding is that the market recognizes that pools traded in the TBA market are not

perfect substitutes. The TBA market is an over the counter market without central clearing or settlement.

Investors pay different prices for contractually identical TBA trades with different dealers on the same

day. These price differences do not appear to be a result of counterparty risk. Less active dealers tend to

get higher prices when they sell in the TBA market than more active dealers. A possible explanation is

that less active dealers may be more likely to deliver a more valuable MBS, while the most active dealers

are more sophisticated and are more likely to sell an attractive MBS in the specified pool market.

Finally, we document that the TBA market appears to play a critical role in allowing dealers to

hedge specified pool inventory. The great majority of dealers match changes in their specified pool

inventory with offsetting changes in their TBA inventory. Dealers are less likely to hedge inventories of

specified pools with maturities or coupons that do not match TBA maturities and coupons. They are also

less likely to hedge specified pools that are not TBA eligible. Specified pools that are less likely to be

hedged have higher round-trip trading costs on average.

The rest of the paper is organized as follows. Section I discusses how the secondary market for

MBS operates. Section II documents that pooling in the TBA market is only partial, and that the identity

of counterparties affects prices. Section III compares liquidity in the TBA and specified pool markets. In

Section IV, we show that dealers use the TBA market to hedge specified pool positions. This improves

liquidity in the specified pool market. Section V draws conclusions.

I. How the Market for MBS Works

In the forward market, or TBA market for mortgage-backed securities, buyer and seller agree on a

price for agency mortgage backed securities to be delivered later. The actual securities to be delivered are

not determined. Instead, six parameters are specified – the issuer (Fannie Mae, Freddie Mac, or Ginnie

Mae), maturity, coupon, price, par amount, and settlement date. Contracts settle once a month. The TBA

market settlements are up to three months out, with contracts for the next two months particularly active.

5  

Sellers have an incentive to deliver the cheapest possible mortgage backed securities that meet the TBA

specifications - and buyers assume they will.

Equilibrium in the MBS market is a pooling equilibrium for MBS with a high probability of

prepayment, and a separating equilibrium for MBS with low prepayment risk. The pooling of MBS with

high prepayment risk occurs through trading in the TBA market. Low prepayment risk MBS will trade in

the TBA market if the price discount from being pooled with higher prepayment risk MBS is smaller than

the cost of convincing buyers of their low risk status. If a MBS has a low enough prepayment risk, it

becomes worthwhile to bear the costs of revealing its value to investors, and the MBS will trade in the

specified pool market. If the borrowers are more sophisticated, they are more likely to prepay and

refinance when interest rates fall. Holders of large mortgages, for example, are more likely to prepay and

hence pools of these mortgages are unlikely to trade as specified pools. Other mortgage backed securities

trade in the specified pool market because their mortgages have non-standard features (e.g. jumbo loans)

and are ineligible for TBA trading.

In this paper, we examine MBS trading using a dataset of all trades by all dealers who are FINRA

members over the May 16, 2011 through April, 2013. Data for each trade includes the maturity, coupon,

and issuer of the MBS, the price, par value, trade date, trade time, and settlement date for the trade, and

identifying numbers for dealers in the trade. Data includes both interdealer trades and trades between

dealers and customers, and both TBA and specified pool trades

Table 1 provides some summary statistics for MBS trading. Panel A reports the number of trades

of various types, and the volume from these trades. As is also noted by Vickery and Wright (2013), 90%

of mortgage backed security volume is in the TBA market. It is also interesting that that interdealer trades

account for almost half the volume in the TBA market, but a much smaller proportion of specified pool

volume. An explanation for greater interdealer trading in the TBA market could be that dealers lay off

inventory in the TBA market by trading with other dealers, and also hedge specified pool inventory with

interdealer TBA trades.

Panel A of Table 1 also provides information on the numbers of different types of TBA trades.

Over the May, 2011 through April, 2013, there are more than 3.3 million TBA trades. Outright trades

make up the majority of TBA trades. Dollar rolls are the second most common type of trade. Dollar rolls

are similar to repos. The seller of a dollar roll sells the front month TBA contract and simultaneously buys

a future month contract with the same characteristics at specified prices. Dollar rolls differ from repos in

that the securities that are purchased for delivery in the later month are “substantially similar” to the one

sold in the front month rather than the same securities. In addition, in a dollar roll, the buyer of the front

6  

month contract receives coupon and principal payments over the month. Dollar rolls tend to be larger size

trades, and are responsible for most of the buy, sell, and interdealer volume.

Stipulated trades are TBA trades in which the buyer requires the seller to deliver pools with

additional stipulated characteristics. The buyer could, for example, specify that no more than a certain

percentage of mortgages are on homes in California. Stipulated dollar rolls are dollar rolls that stipulate

additional characteristics of pools to be delivered. They are less common, and account for less than

30,000 trades.

There are about 1.66 million trades of specified pools. The great majority of these are TBA –

eligible, and could be sold in the TBA market if the seller so desired. The other pools may contain jumbo

loans or loans with credit scores that make them ineligible for TBA trading. Interdealer trades make up a

far smaller proportion of specified pool trades than TBA trades. As we show later, interdealer TBA trades

are used to manage specified pool inventory.

Panel B of Table 1 provides information on trade sizes. Trade sizes are far smaller for specified

pools than for TBA trades. Interdealer specified pool trades have an average size of only $3.32 million

dollars. Trade sizes are right-skewed, so the majority of trade sizes are smaller. Only 6.7% of specified

pool interdealer trades for at least $10 million par. It is interesting that specified pool trades with

customers tend to be larger than interdealer specified pool trades. The mean size trade with customers is

for $6.49 million par value, and 10.7% of the trades are for $10 million or more.

TBA trades are much larger than specified pool trades. Dollar rolls are especially large. The mean

trade size for interdealer trades of dollar rolls is $59.64 million while the mean size for trades with

customers is over $100 million.

Panel C of Table 1 reports the proportion of trades of different types for dealers with different

levels of activity. There are a large number of dealers, but trading is heavily concentrated in a few dealers.

Active dealers tend to do most of their trading in the TBA market, while inactive ones trade mainly in

specified pools. The table doesn’t show results for individual dealers, but the single most active dealer

accounts for 17.3% of all trades, but made almost no trades in the specified pool market. For the ten most

active dealers, the average proportion of volume from specified pools is 13.55%. For the twenty next

most active dealers, the proportion of volume from specified pools averages 26.16%. For dealers ranked

101 – 758 by number of trades, the proportion of volume from specified pools reaches 87.82%. Large

dealers trade mainly in the TBA market. Less active dealers trade mainly in the specified pool market. As

we have seen, TBA trades are usually much larger than specified pool trades. To compete effectively as a

7  

dealer in the TBA market requires more capital than it takes to trade specified pools – capital that the

inactive dealers may not have.

Panel C also reveals that the proportion of trades that are interdealer trades is higher for more

active dealers than for less active ones. Even for the least active dealers, however, the average proportion

of trades that are interdealer is over 44%.

Figure 1 shows the daily dollar of investor outright purchases from dealers, purchases of dollar

rolls, and purchases of specified pools for our sample period of May, 2011 through April, 2013. Several

facts about MBS trading stand out in the graph. First, the dollar volume of both outright and dollar roll

TBA trading dwarf that of specified pool trading. Specified pool trading is typically on the order of $5

billion to $10 billion per day. In contrast, there are several days when dollar rolls alone account for $100

billion dollars in trading. Total daily purchases in the TBA market typically exceed $50 billion and are

often greater than $100 billion. Second, there is a strong monthly seasonal component in the TBA

volume, particularly for dollar rolls. This reflects investors rolling over positions in the TBA market

before settlement dates. There appears to be seasonal component to specified pool trading, but it is much

weaker. Finally all of the time series appear stationary. There is no appreciable change in volume in the

TBA, dollar roll, or specified pool market over the two-year sample period.

II. Prices in the Specified Pool and TBA Markets

During the sample period, both TBA and specified pool prices increased. This can be attributed to

falling mortgage interest rates over this time period. Figure 2 shows weekly national average mortgage

rates, from Freddie Mac, for 15 and 30 year mortgages for the period from April, 2011 through April,

2013. For this period, 30-year rates were consistently higher than 15 year rates by about 75 basis points.

Rates decline by about 125 basis points between April, 2011 and October, 2012. The decline in rates

explains why prices of mortgage backed securities rose over the sample period. The decline in rates also

made prepayment an attractive option to many mortgage holders during this time.

To compare prices in the TBA and specified pool markets, we first calculate the average price of

interdealer trades in the TBA market for combinations of maturity and coupon for each sponsor, and each

settlement date each day. Likewise we calculate the average price for interdealer trades of specified pools

for maturity and coupon combinations each day. Prices of TBA and specified pool securities cannot be

directly compared because they have different settlement dates. To adjust for this, we calculate the “drop”

as the difference in price between the Fannie Mae TBA with the nearest settlement, and the Fannie Mae

TBA with the second nearest settlement. Fannie Mae TBAs are used because they are the most common.

The daily drop is the drop divided by the number of days between the two settlement dates. We then

8  

multiply the daily drop by the number of days between a specified pools settlement date and the nearest

TBA settlement date and add this to the specified pool price. This adjusted specified pool price can then

be compared with TBA prices. In practice though, adjusting for the drop does not have a big impact on

the difference between TBA and specified pool prices.

Figure3 shows average specified pool and average nearest settlement Fannie Mae TBA prices for

each day for MBS with different maturities and coupons. The first figure compares prices of 15 year, 3%

MBS traded in the TBA market with those traded as specified pools. Specified pool and TBA prices are

generally close, but there are fewer specified pool trades, and hence the series of average specified pool

prices appears more volatile. The second graph shows prices of 15 year, 4% TBA and specified pool

trades. Here, specified pool prices are slightly higher, especially toward the end of the period. The third

graph shows 30 year, 4% TBA and specified pool prices. At the start of the sample period, the prices are

very close. Starting in late 2011, the two price series diverge, with specified pool prices typically higher

by 1% to 2% of face value. The last graph shows TBA and specified pool prices for 30 year, 5% MBS. In

this case, specified pool prices are almost always higher.

III. Trading Costs in the Specified Pool and TBA Markets

To date, there has been little academic research on the microstructure of MBS markets.

Bessembinder et al (2013) examine trading of MBS and other structured credit products for the period

from May 16, 2011 through January 31, 2013. They estimate that 90% of TBA trades are over $1 million,

and two-third’s are interdealer trades. For other MBS trades, 39% are over $1 million and only 35% are

interdealer trades. They use a regression approach to look at trading costs as differences in price between

successive customer trades with dealers. A variable for change from sell to buy (+1) or buy to sell (-1) is

included in the regression along with variables for changes in bond and equity indices over the trade

period. One way trading costs are estimated to be 40 basis points for MBS, and just 1 basis point for

trading in the TBA market.

To start, we provide some summary statistics on round-trip trading costs (markups) for dollar

rolls, TBA outright trades, and specified pool trades for our sample period. For each CUSIP each day, we

calculate the weighted average purchase and sales prices using the dollar volume in a trade as the weight.

Only trades between customers and dealers are used. Each observation is one security on one day.

Observations are only included if there is at least one purchase and at least one sale on that day. Markups

are the differences between the average purchase price and average sale price. Table 2 presents average

markups for MBS with 30 years to maturity and coupon rates between 2.5% and 6%, and for MBS with

9  

15 years to maturity and coupon rates of 2.5% to 4%. These 12 maturity-coupon combinations account for

95% of the volume in the TBA market during our sample period. We report average markups separately

for TBA dollar rolls, TBA outright trades, and specified pool trades. Markups are reported as percentages,

so 1.000 is 1% of par value.

Markups for dollar rolls are particularly low, typically about one-half of a basis point. Markups on

TBA outright trades range from 0.29 basis points for 30-year, 4% TBA trades to 8.78 basis points for 15-

year, 4% TBAs. The more active the trading in a TBA contract, the lower the transactions costs. Specified

pool markups are much larger. The round trip costs on a 30-year, 3.5% specified pool averaged 10.56

basis points, as compared to 1.79 basis points for a 30-year, 3.5% TBA trade. Likewise, the markup on a

15-year, 3% specified pool averaged 31.47 basis points, while the markup for a similar TBA trade was

0.99 basis points.

Table 2 results suggest that specified pools are much more expensive to trade than TBAs.

Nevertheless, it is possible that we have underestimated trading costs for specified pools. In many cases,

dealers who purchase specified pools need to hold them several days before they can be sold. It is likely

that specified pools with purchases and sales on the same day, like those in Table 2, are more liquid and

have lower markups than other specified pools. This is less likely to be a problem for TBAs, which trade

much more frequently.

To estimate trading costs for MBS when purchases and sales occur days apart, we employ a

regression methodology like that in Bessembinder et al (2013). Each observation is consecutive trades in

an MBS with a specific CUSIP, but the regression includes observations from all CUSIPs with a

particular coupon and maturity. To estimate trading costs, we estimate the following regression:

Δ Δ Δ ∙ ln Δ ∙

Δ ∙ ∙ ln Σ , , 1

where ΔPt is the difference in prices between trade t and trade t-1, ΔQt, is +1 (-1) if trade t-1 is a sale to

(purchase from) a dealer and trade t is a purchase from (sale to) a dealer, lnSize is the natural log of the

par value of the securities in the trade in thousands of dollars, and TBA Eligible is a dummy variable that

takes a value of one if the specified pool is eligible to be traded as a TBA. Five return variables are also

included to capture changes in MBS values when consecutive trades take place on different days. They

are the percentage changes in 1) a U.S. Agency Fixed Rate MBS index, 2) a U.S. Treasury 7-10 year

Bond index, 3) a U.S. Investment Grade Corporate Bond Index, 4) a U.S. Corporate High-Yield Bond

Index, and 5) the S&P 500 index. If consecutive trades occur on the same day, all of these return values

are zero.

10  

Table 3 reports the regression results, with Panel A reporting results for specified pools and Panel B

reporting TBA results. The first row in Panel shows coefficients and robust t-statistics for specified pools

with 30 years to maturity and coupons of 2.5%. There are a total of 3,866 pairs of consecutive trades in

specified pools with the same CUSIP. The coefficient on ΔQ is 1.4110. If both of the consecutive trades

were for $1,000, and hence the log of the trade size was zero, and the TBA dummy was zero, round trip

trading costs would be 1.411%. The interactions between ΔQ and trade size and between ΔQ and TBA

eligible indicate that trading costs decrease with trade size and are lower for TBA eligible securities. In

most of the regressions the coefficient for the ΔQ x TBA eligible x trade size term is positive. This

suggests that trading costs do not decrease as fast with trade size for TBA eligible specified pools as they

do for specified pools that are not TBA eligible.

Panel B shows results for regressions using TBA trades. Here, of course, variables for TBA eligibility

are not included in the regressions. Coefficients on ΔQ are much lower for TBA trades than for specified

pool trades, indicating that transactions costs are lower for small TBA trades than for small specified pool

trades. The coefficients on the interaction between ΔQ and trade size are negative, indicating that TBA

trading costs decline with trade size. Regressions in Panel B have far more observations than Panel A

regressions.

Using the regression coefficients in Table 3, we estimate markups for TBA trades, TBA eligible

specified pools, and ineligible specified pools for trades of $100,000, $1,000,000, and $5,000,000.

Results are shown in Table 4.

Markups decrease with trade size for trades of all types. For example, a TBA-eligible specified pool

with a 4% coupon and a 30 year maturity has a markup of 51 basis points for a trade of $100,000, 26

basis points for a trade of $1,000,000, and 8.6 basis points for a trade of $5,000,000. Table 4 also

demonstrates that markups are far lower for TBA trades than for TBA-eligible specified pools, which are,

in turn much lower than markups for specified pools that are not TBA-eligible. The markup for a round-

trip trade of $1,000,000 of 30-year 3.5% securities is 3.3 basis points for TBA trades, 18 basis points for

TBA-eligible specified pools, and 91.5 basis points for other specified pools. There are some maturity-

coupon combinations with few specified pool observations, such as 30-year maturities with coupons of

5.5% or 6%. In these cases there are few observations of specified pools that are not TBA-eligible, and

estimates are imprecise. Nevertheless, the pattern is clear: TBA trades are much cheaper than trades of

TBA-eligible specified pools, which in turn are much cheaper than trades of other specified pools.

IV. Counterparties and Pooling in the TBA Market

11  

TBA market trading can be thought of as a pooling equilibrium in which MBS that differ in

prepayment risk are nevertheless traded as if they were identical. The TBA market is a bilateral dealer

market and not a centrally cleared market though, so the pooling is incomplete. Counterparties matter in

bilateral dealer markets if they pose different default risks, but there is another reason why counterparties

may matter in the TBA market. In TBA trading, the seller of the MBS can deliver any pool that meets the

specified criteria. It is assumed by the buyer that the seller will deliver the least attractive (that is the most

prone to prepayment) pool. It is possible that different dealers will deliver pools of different quality.

We test this using all sales from dealers to customers of MBS in the TBA market. For maturity-

coupon combinations we estimate the following regression

, , . , , , , , 2

where Pricei,c,t is the price for the ith TBA sale of MBS with CUSIP c from dealers to customers on day t,

Avg.Pricec,t is the trade size weighted average price for TBA sales of CUSIP c on day t, Ln(Trade Sizei,c,t)

is the natural logarithm of the par value of the securities in the sale, and Dealern is a dummy variable that

takes a value of one for sales by dealer n.

Results are reported in Table 5 for the 12 most common maturity-coupon combinations. In each

case, the intercept of the regression is positive and significant, and the coefficient on the log trade size is

negative and significant. Dealers charge lower prices for large TBA purchases than for small TBA

purchases. Of more interest are the f-statistics that test whether fixed effects for dealers are significant.

For three of the less-frequently traded maturity-coupon combinations, f-tests indicate that sale prices do

not differ significantly across dealers. In the other cases, f-statistics indicate that the dealer fixed effects

are quite significant. Different dealers get different prices when selling in the TBA market. For example,

for the 30-year maturity, 3% coupon case, the f-statistic, for 100 dealers and 57,653 remaining degrees of

freedom is 30.531. The p-value for this is 0.000. Different dealers receive different prices for similar sales

in the TBA market.

A more active dealer may be better able to separate high quality pools for sale in the specified

pool market, leaving particularly poor quality pools to sell TBA. Inactive dealers, on the other hand, may

commit to sell in the TBA market, but not be able to separate pools into those that should be sold in the

specified pool market, and those that should be sold TBA. The quality of pools delivered with TBA sales

may be higher for inactive dealers.

12  

For each dealer each month, we calculate the total number of TBA purchases from, and sales to

customers over the previous three months. We define inactive dealers as those with 1,000 or fewer trades

in the previous three months. For each TBA CUSIP each day, we calculate the weighted average price

for sales from dealers to customers using the par value of trades as weights. We also calculate the average

trade size for the CUSIP that day, and the average natural log of the trade size. Then, for each sale in a

CUSIP with two or more sales during a day, we regress the sale price on a dummy variable for an inactive

dealer and the difference between the log of the trade size and the average log of trade size for that CUSIP

that day. That is,

3

Table 6 reports results. Panel A reports results when trades of all sizes are included. The first row

is the regression when TBA trades of MBS of all maturities and coupons are included together. The

coefficient on the difference in log sizes is -0.0110. When a dealer sells MBS in a larger trade than the

average for that day, he tends to receive a lower price. The coefficient on the inactive dealer dummy

variable is 0.0728. On average, the price that inactive dealers get when they sell in the TBA market is

7.28 basis points higher than active dealers receive. The robust t-statistic for the inactive dealer dummy is

37.01, reflecting both the significance of dealer identity, and the large number of observations (322,037)

in the regression.

The rest of the rows report results when only trades of particular coupons and maturities are

included in the regressions. In every case, the coefficient on the difference in log sizes is negative and

highly significant. In all cases save one, the coefficient on the inactive dealer dummy is positive and

highly significant. Inactive dealers get higher prices when they sell to customers in the TBA market. In

general, holding maturities constant, higher coupons are associated with larger coefficients on the inactive

dealer dummy. Higher coupons are associated with a higher rate of prepayment, so this suggests that

buyers recognize that they will get pools with lower prepayment rates in the TBA market if they buy from

inactive dealers.

A reader could question whether our adjustment for trade size was adequate and whether the

difference in prices received by active and inactive dealers reflects differences in trade size. To address

this, we repeat the analysis of Panel A, but only include trades that differ in size from the average trade

for that CUSIP on that day by less than $1,000,000. Results are shown in Panel B.

When the sample is restricted in this way, we lose about 95% of the observations. Despite this,

the coefficients on the inactive dealer dummy variables are generally positive and statistically significant.

13  

When trades from all maturity-coupon combinations are used, the coefficient on inactive dealer is 0.0902.

Inactive dealers receive prices that are 9.02 basis points more than active dealers. For 9 of the 12 specific

maturity-coupon combinations, coefficients on the inactive dealer dummy are positive and statistically

significant. They are of roughly the same magnitude as the coefficients from the regressions with all

trades.

We rerun the regressions in several alternate forms as robustness checks. We use the simple

average price for a CUSIP each day rather than the weighted average. We also use the difference in trade

sizes rather than the difference in log trade sizes. We also ran the regressions with fixed effects for the

date of the sale. None of these alternative methodologies produced different results – inactive dealers

received higher prices for sales in the TBA market.

We repeat the analysis of dealer sales to customers with a similar analysis of dealer purchases

from customers. For each CUSIP each day, we calculate the weighted price for purchases by dealers from

customers. We also calculate the average log of the trade size for these trades. We then regress the

difference between the price of a purchase and the average purchase price for each CUSIP each day on a

dummy for inactive dealers and the difference between the log trade size and the average log trade size.

Results are reported in Table 7.

Coefficients on the inactive dealer dummy are generally negative and statistically significant. For

example, when all maturity and coupon combinations are included in the regression, the coefficient on the

inactive dealer coefficient is -0.0172. Inactive dealers pay about 1.7 basis points less when purchasing

from customers than do active dealers. When specific maturity-coupon combinations are considered,

coefficients on inactive dealer dummies are generally negative and statistically significant. We repeat the

regressions using only trades with sizes within $1,000,000 of the mean size. Results are shown in Panel

B. Coefficients on the inactive dealer dummy are typically negative, but are not statistically significant in

most of the regressions using a specific maturity-coupon combination.

Results using dealer purchases from investors seem to mirror the results from dealer sales to

investors. In both cases, inactive dealers seem to get better prices. The results are much stronger,

however, for dealer sales to investors. When all trades are included in the same regression, dealers pay

under two basis points less to purchase, but sell for seven to eight basis points more. And, despite larger

sample sizes, the discount that inactive dealers receive when buying from customers is less likely to be

statistically significant than the premium they receive when selling to customers.

14  

TBA trading involves bilateral contracts. There is no central clearing, so the identity of one’s

counterparty can matter. There are two potential explanations for why active dealers receive lower prices

when they sell in the TBA market. One is that they are better informed. They are more likely to know

when MBS prices are likely to fall. The second potential explanation is that active dealers are likely to

deliver a less desirable mortgage pool when the TBA trade is settled. Less desirable in this context means

homeowners are more likely to prepay when interest rates fall, and less likely to prepay when interest

rates rise. Active dealers may sell less desirable pool in the TBA market if they are better than inactive

dealers at separating their most desirable pools and selling them in the specified pool market.

V. The TBA Market and Specified Pool Liquidity

Every specified pool is unique, and there are hundreds of thousands of them. That the market is as

liquid as it is owes a lot to parallel trading in the TBA market.

TBA trading contributes to specified pool liquidity in two ways. First, it provides benchmark

prices for specified pools. MBS traded in the TBA market are lower priced because they are more likely

to be prepaid. We would expect price discovery to take place in the TBA market both because it is more

liquid and also because the cheapest-to-deliver pools trade in the TBA market.

The TBA market also allows dealers to hedge positions in the specified pool market. Because

each specified pool is unique, it may take some time for dealers to sell them. The low trading costs in the

TBA market allow dealers to hedge the inventory they intend to sell in the specified pool market cheaply.

It is not clear whether dealers would hedge specified pool inventory from one issuer (say, Fannie

Mae) with offsetting TBA inventory by the same issuer, or whether they would hedge specified pool

inventory from one issuer with TBA trading in other issuers. To examine this, we calculate daily changes

in specified pool and TBA inventory for each dealer for each issuer, maturity, and coupon combination.

One combination, for example, would be 30-year, 3.5%, Freddie Mac MBS. Inventory changes for the

dealer are obtained by adding the par value of all purchases from customers and other dealers, and

subtracting all sales to customers or other dealers. We then estimate the following two regressions of

changes in TBA inventory on changes in specified pool inventory

∆ , ∆ . , , 4

∆ , ∆ . , , 5 .

The α2 coefficient in equation (4) indicates the proportion of specified pool inventory changes that are

offset by TBA trades in MBS issued by the same agency. If dealers completely hedged their specified

15  

pool inventory with TBA trades in MBS issued by the same agency, we would expect α2 to equal -1. The

β2 coefficient in equation (5) provides the proportion of specified pool inventory changes that are offset

by TBA trades in MBS issued by all agencies. These regressions are run separately for each dealer. Some

maturity/coupon combinations are not traded over the entire sample period. For example, 30 year 5%,

5.5%, and 6% MBS were only traded in the TBA market in the early part of the sample period. By 2013,

trading in these maturity/coupon combinations had virtually disappeared from the TBA market. Hence, in

estimating these regressions, we use only days when the dealer had a change in specified pool inventory.

Regression estimates are reported in Table 8. The weighted median α2 and β2 coefficients across

dealer regressions are reported. In calculating the coefficients, we weight by the number of observations

in each dealer regression. An examination of Table 8 reveals that the α2 coefficients are typically smaller

in absolute value and further from -1 than the β2 coefficients. For example, for Fannie Mae 30 year 3.5%

coupon MBS, the median α2 coefficient is -0.4756. This suggests that 47.56% of specified pool inventory

changes in 30 year 3.5% Fannie Mae specified pools are offset with Fannie Mae TBA trades. The median

β2 coefficient is -1.0670, suggesting that slightly over 100% of Fannie Mae specified pool inventory

changes are offset when TBA trades in MBS from all issuers are considered.

As a whole, the results in Table 8 strongly suggest that dealers do not differentiate between MBS

issued by different agencies when hedging specified pool inventory. So, henceforth, we study dealer

hedging after pooling together inventory changes across issuers for each maturity/coupon combination

(i.e. 30 years, 3.5%). For each dealer, we regress daily changes in TBA inventory on same-day changes

in specified pool inventory and on lagged changes in TBA inventory. Both TBA and specified pool

inventory changes are aggregated across all issuers for each dealer. That is,

∆ , ∆ . , ∆ , , 6

Panel A of Table 9 summarizes the results across individual dealer regressions for

maturity/coupon combinations. A number of dealers did very little trading in the TBA or in the specified

pool market, making it very difficult to estimate the regression for them. If we cannot estimate all

coefficients for a dealer, that dealer regression is dropped. In addition, in calculating mean and median

coefficients and t-statistics, we weight individual dealer regressions by the number of observations in

them. The fourth column reports the median coefficient on the same day change in specified pool

inventory. If dealers used the TBA market to hedge their specified pool inventory, we would expect these

coefficients to be negative. If the specified pool positions were completely offset, we would expect the

coefficient to equal -1.

16  

As an example, consider the results for regressions using 30 year, 4% MBS. These are shown in

the fourth row of the table. Regressions of changes in TBA inventory on changes in specified pool

inventory could be estimated for 52 dealers, and as the next column shows, there were 6,199 observations

across the 52 regressions. The median estimate of the coefficient on changes in specified pool inventory is

-1.0057, suggesting that median dealer offset specified pool inventory changes with TBA trades one-for-

one. The median t-statistic for the coefficient is -7.00, and the mean is -8.15. The evidence that dealers

offset specified pool inventory changes with TBA trades is quite strong.

The other rows of the table report results from regressions using MBS with different maturity-

coupon combinations. The results in these other rows are sometimes stronger and sometimes weaker than

the results for 30 year 4% MBS, but they also indicate that dealers use the TBA market to offset changes

in specified pool inventory.

The weakest results are for the regressions using 30 year 6% MBS. The median coefficient on

change in specified pool inventory is -0.3296. This maturity-coupon combination traded infrequently

during the sample period. The 16 regressions are estimated using only 150 total observations, so these

coefficients are likely to be estimated less accurately than others. In addition, the TBA market for 30 year

2.5% and 30 year 6% MBS is likely to have been less liquid over most of the sample period than the TBA

market for other MBS

The last two columns of the table show the median coefficients on the change on lagged TBA

inventory change, and the median t-statistics. In contrast to the strong evidence that dealers hedge their

specified pool inventory with TBA trades, evidence that lagged TBA inventory changes from the previous

day affect current TBA trading is weak. Coefficients on lagged TBA inventory changes, shown in the

second to the last column, are all negative, but close to zero, as are the median t-statistics. While there is

weak evidence that dealers offset previous-day changes in TBA inventory, it seems likely that dealers

seldom maintain unwanted TBA positions overnight.

Much of the trading in the specified pool market consists of pools with different maturities than

the 30 year and 15 year maturities traded in the TBA market. These specified pool trades cannot be

hedged as well in the TBA market. Nevertheless, it seems likely that some dealers will use the TBA

market to hedge changes in the inventory of specified pools with different maturities. To examine this, we

regress changes in TBA inventory on changes in inventory of specified pools with the same coupon and

maturity, and changes in inventory of specified pools with the same coupon but other maturities. That is

∆ , ∆ . 15 30 , ∆ . . , , . 7

17  

Other maturities are 10-14 and 16 -21 years for 15 year TBA inventory changes and 22-29 and 31–35

years for 30 year TBA inventory. Now, a day is included in the dealer regression if there is a change in

the dealer’s TBA inventory, specified pool with the same maturity inventory, or specified pool with a

different maturity inventory.

Summaries of the individual dealer regressions are reported in Panel B of Table 9. Coefficients

and t-statistics are again weighted by the number of observations in each dealer’s regression to produce

the means and medians reported in the table. As in Panel A, coefficients on changes in specified pool

inventories are generally negative and significant. Dealers offset changes in their specified pool inventory

positions by trading in the TBA market.

The last four columns describe the distribution of coefficients, and t-statistics of the coefficients,

for the change in inventory of specified pools with other maturities. The coefficients are negative, but

generally smaller in absolute value than the coefficients on changes in inventory for specified pools with

15 or 30 years maturity. For example, the median coefficient on changes in inventory of specified pools

with odd maturities is -0.4569 for the 30 year 4.5% TBA regressions, and -0.9094 for specified pools with

30 years to maturity and coupons of 4.5%. It appears that dealers hedge their inventory positions in

specified pools with odd maturities, but not as much as those with 15 or 30 years to maturity.

Dealers who acquire inventory positions in specified pools with maturities other than 15 or 30

years do seem to hedge part of their inventory changes in the TBA market. Coefficients are smaller

however, than for changes in inventory of specified pools with 15 or 30 years to maturity. This suggests

that dealers hedge a lower portion of inventory changes in odd maturity specified pools than changes in

specified pool inventories that match TBA maturities.

Specified pools may not be eligible for TBA trading if they contain jumbo loans, if the mortgage

holders do not meet criteria for creditworthiness, or for other reasons. These mortgage pools have

different characteristics than those that trade in the TBA market. Jumbo loans, for example, are thought to

be more likely to be prepaid. These differences mean that TBA positions will not hedge TBA-ineligible

inventory as effectively as they hedge TBA-eligible specific pools. Hence it is interesting to ask whether

dealers hedge positions in specified pools that are not TBA eligible by taking offsetting positions in the

TBA market. To examine this, we run the following regression

∆ , ∆ . , ∆ . , , . (8)

This regression is run separately for every dealer using only specified pools with 30 (or15) years

to maturity. Median coefficients and t-statistics are reported in Panel C of Table 9. In calculating means

and medians, each dealer regression is weighted by the number of observations in the regression. Days are

18  

only included in a regression if there was a change in either the TBA-eligible or TBA ineligible specified

pool inventory.

The median α2 coefficients, which can be interpreted as the proportion of TBA-eligible specified

pool that are hedged in the TBA market, range -0.7974 to -0.9863 across the different maturity-coupon

combinations with more than 2,000 total observations. The α3 coefficients provide estimates of the change

in TBA inventory when TBA-ineligible specified pools are traded rather than TBA-eligible specified

pools. In general, these coefficients are closer to zero. There appears to be less hedging of TBA- ineligible

specified pools than TBA-eligible specified pools.

VI. Hedging and Trading Costs in the Specified Pool Market

Dealers who hold an inventory of securities are exposed to the risk of price fluctuations. This is

the inventory holding cost of market making. There is a large literature in market microstructure that

indicates that greater inventory holding costs for dealers leads to greater trading costs. If, however,

dealers are able to hedge their inventory holdings, trading costs need not be affected by dealers’ need to

hold inventory. For example, Battalio and Schultz (2011) find that option market makers needed to hedge

using the underlying stock.

As we have seen, the specified pools that are most likely to be hedged in the TBA market are

TBA eligible with 15 or 30 years maturity. This suggests that specified pools with these characteristics

are likely to be the cheapest to trade. We have already documented that trading costs are lower for TBA

eligible specified pools.

The maturity of specified pools may also affect their liquidity. During our sample period, TBA

trading takes place almost entirely in 15 and 30 years securities. Prices in the TBA market are likely to be

most informative for pricing specified pools with 15 or 30 years to maturity rather than specified pools

with, say, 14 or 26 years to maturity. In addition, it is likely to be easier to hedge specified pool positions

in the TBA market if the specified pools have either 15 or 30 years to maturity. Hence we would expect

specified pools with exactly15 or 30 years to maturity to have lower trading costs than those with

somewhat different maturities.

To test this, we run the following regressions using all specified pool sales to and purchases from

customers:

Δ Δ Δ ∙ ln Δ ∙ Δ ∙

∙ ln ∆ 30 ∆ 30 ∙ ln

Σ , . 9

19  

Here, 30Year is a dummy variable that takes a value of one if the specified pool has exactly 30 or 15

years to maturity, and zero if the maturity is longer or shorter. All other variables are the same as in the

previous regressions. Before, we included all MBS with the same coupon and either 15 or 30 years to

maturity in the regressions used to calculate trading costs. Now we include all specified pools with 22-35

years to maturity in the 30-year MBS regressions, and all specified pools with 10-21 years to maturity in

the 15-year MBS regressions. We do not include TBA trades in the regressions.

Results are reported in Panel A of Table 10. For 10 of the 12 maturity-coupon combinations, the

coefficient on the interaction between the quote change and the dummy variable for exactly 30 or 15

years to maturity is negative. This indicates that, at least for small trades, trading costs are lower for

specified pools with exactly 15 or 30 years to maturity than for specified pools with other maturities. The

regression also contains an interaction between the quote-side change, the size of the trade, and the

dummy variable for an exact 30 year or 15 year maturity. The coefficient on this variable is always the

opposite sign of the coefficient on the interaction between quote change and exact 30 year or 15 year

maturity. Thus the only way to tell if having an exact 15 or 30 years decreases trading costs for a

particular size trade is to plug values for the trade size into the regressions.

Transaction cost estimates from the regressions are reported in Panel B of Table 10. Costs when

each trade in the round trip is for $100,000 par value are shown in columns 3-5. Trading costs when the

specified pool trades are TBA eligible and mature in 30 (or 15) years are shown first. The second cost

reported in the column is 0.5509. This indicates that the round-trip trading cost for TBA-eligible specified

pools with 30 years to maturity and a coupon of 3.0% is 0.5509, or 55 basis points on a $100 par value.

The next column reports trading costs when the specified pools are not TBA eligible but mature in 30 (or

15) years. For every maturity/coupon combination, trading costs are greater when the MBS is not TBA

eligible. So, for example, trading costs for specified pools with coupons of 3% and 30 years to maturity

rise from 55 basis points for TBA-eligible pools, to 156 basis points for those that are not TBA-eligible.

The following column provides round-trip trading costs for $100,000 par value specified pools that are

not TBA eligible, and that have maturities within the 22-35 (10-21) year range that are not 30 (15) years.

Trading cost estimates for these odd maturities exceed trading costs for other specified pools for 11 of the

12 maturity-coupon combinations.

Trading costs for round-trips of $1,000,000 par value and $5,000,000 par value are shown in the

following six columns of the table. Regardless of the characteristics of the specified pool, larger trade

sizes are associated with lower proportional trading costs. For a handful of maturity-coupon

combinations, point estimates of round-trip trading costs are negative. In these cases, regressions may

have been estimated with very few trades of $1,000,000 or $5,000,000.

20  

For large trades, as with small trades, specified pool trading costs are greater if they are not TBA-

eligible, and greater yet if their maturities are not 30 or 15 years. The greater the similarity between the

characteristics of a specified pool and the characteristics of MBS traded in the TBA market, the lower the

trading costs for the specified pool.

Trading costs are lower for specified pools with characteristics that are most similar to the MBS

that trade in the TBA market. These are also the specified pools for which dealers are most likely to hedge

inventory positions. It is tempting to conclude that dealer’s ability to hedge specified pool inventory

positions lowers trading costs for specified pools, and that may indeed be true. It is also possible though

that TBA trades provide clearer benchmark prices for the specified pools that most closely resemble them,

and the existence of good benchmark prices lowers trading costs.

VII. Conclusions

The secondary market for agency mortgage backed securities is among the largest and most liquid

securities markets in the world. In a way, this is surprising because each of the hundreds of thousands of

MBS is a claim on the cash flows of a different set of mortgages, and is therefore unique. An important

reason for the liquidity of this market is the existence of TBA trading, in which different MBS are traded

in a forward market on a cheapest-to-deliver basis. Round-trip trading costs in the TBA market are tiny,

typically only a few basis points.

TBA trading can be thought of as a pooling equilibrium in which MBS with a range of

prepayment risks are sold to buyers who do not know which MBS they will receive. The pooling,

however, is incomplete. Mortgage backed securities trade in a dealer market, and different dealers are

likely to deliver MBS of different quality. We find that less active dealers receive higher prices when they

sell in the TBA market than more active dealers. We hypothesize that less active dealers may be less able

to separate the best MBS for sale in the specified pool market, and that buyers recognize that they are

more likely to receive an MBS with low prepayment risk when they purchase from an inactive dealer.

Round-trip trading costs are much higher for specified pools. There is considerable variation in

trading costs across different types of specified pools. The pools that are TBA-eligible are cheaper to

trade than TBA-ineligible specified pools. Specified pools with 15 or 30 years to maturity are cheaper to

trade than otherwise similar specified pools with other maturities.

Differences in trading costs across different types of specified pools correspond closely to

differences in dealer hedging of specified pool positions. By examining daily net inventory changes for

specified pools and TBA contracts, we provide strong evidence that dealers offset changes in specified

21  

pool inventory by trading similar TBA contracts. This is particularly true for the most active dealers.

Dealers do far less trading in the TBA market to offset positions in TBA-ineligible pools than TBA-

eligible pools. Similarly, they offset changes in inventory of specified pools with 15 or 30 years to

maturity by TBA trading more than they offset changes in inventory of specified pools with other

maturities. Put another way, if a specified pool is very similar to MBS traded in the TBA market, a dealer

will hedge a position in the specified pool through TBA trading. If the specified pool has different

characteristics, it will not be hedged through TBA trading.

Positions in the specified pools that are most similar to the pools traded in the TBA market are

most likely to be hedged. These are also the specified pools that are cheapest to trade.

22  

References

Atanasov, Vladimir, and John Merrick, 2013, The effects of market frictions on asset prices: Evidence from agency MBS, working paper, College of William and Mary, available at http://ssrn.com/abstract=2023779.

Battalio, Robert, and Paul Schultz, 2011, Regulatory uncertainty and market liquidity: The 2008 short sale ban’s impact on equity option markets, Journal of Finance 66, 2013-2053.

Bessembinder, Hendrik, William Maxwell, and Kumar Venkataraman, 2006, Market transparency, liquidity externalities, and institutional trading costs in corporate bonds, Journal of Financial Economics 82, 251-288.

Bessembinder, Hendrik, William Maxwell, and Kumar Venkataraman, 2013, Trading activity and transaction costs in structured credit products, Financial Analysts Journal 69 (6), 55-68.

Citigroup Markets Quantitative Analysis, 2011, Specified Pools: Superior Prepayment Profiles Offer Added Value, Citigroup Global Markets.

Pagano, Marco, and Paolo Volpin, 2012, Securitization,Transparency, and Liquidity, Review of Financial Studies 25 (8), 2417-2453.

Song, Zhaogang and Haoxiang Zhu, 2014, Mortgage Dollar Roll, working paper, Board of Governors of the Federal Reserve System.

Stanton, Richard, 1995, Rational Prepayment and the Valuation of Mortgage-Backed Securities, Review of Financial Studies 8 (3), 677-708.

Vickery, James, and Joshua Wright, 2013, TBA Trading and Liquidity in the Agency MBS Market, FRBNY Economic Policy Review, May, 2013, 1-16.

23  

Table 1 Summary Statistics for MBS Trading in the TBA and Specified Pool Markets, May, 2011 – April, 2013.

Volume is in $1,000,000’s of face value. Panel A. Total trading by trade type. Number

Sells Volume

SellsNumber

BuysVolume

BuysNumber

Interdealer Volume

Interdealer TBA Trades Outright Trades 342,350 13,794,567 494,661 13,568,672 1,531,919 25,807,536

Dollar Rolls 139,134 16,415,923 147,964 17,020,740 544,153 32,525,031

Stipulated Trades 34,460 1,001,036 39,936 1,125,391 14,624 170,244

Stip. Dollar Rolls 8,456 429,988 17,665 1,026,649 1,711 28,310

Total TBA trading 533,664 32,238,176 691,017 32,144,801 2,092,407 58,531,121

Specified Pool Trades TBA Eligible 291,404 2,459,120 657,974 3,822,160 472,574 1,394,371

Non-Eligible 75,787 394,383 74,512 476,542 89,625 474,355

Total Specified Pool 367,191 2,853,503 732,486 4,298,702 562,199 1,868,726

Panel B. Trade sizes by trade type Interdealer Trades Trades with Customers

Number Avg. Trade Size

($millions) Percent >

$10 million

Number Avg. Trade Size

($millions) Percent >

$10 million Specified Pools 562,067 $3.32 6.7% 1,099,260 $6.49 10.7% TBA Outright 1,531,919 $16.71 37.1% 837,011 $32.64 37.2% TBA Dollar Roll 544,153 $59.64 60.8% 287,158 $116.43 68.1% TBA Stipulated 14,624 $11.64 12.0% 74,396 $28.58 36.4% TBA Stip. Rolls 1,711 $16.55 32.0% 26,121 $55.77 66.1%

Panel C. Dealer activity and trade type. Dealer Ranking by Number of

Trades

Percentage of Trades that are Specified Pools

Percentage of Volume from

Specified Pools

Percentage of Trades that are

Interdealer

Minimum Number Trades

Maximum Number Trades

1-10 23.83% 13.55% 56.37% 267,712 1,508,883 11-30 42.86% 26.16% 55.09% 62,397 237,655 31-50 56.44% 42.08% 53.73% 24,037 61,151 51-100 75.35% 63.29% 51.37% 4,758 22,690

101-758 91.36% 87.82% 44.73% 1 4,587

24  

Table 2. Markup estimates for MBS that had same-day purchases and sales.

Each observation is one security on one day. Average prices of purchases and sales are calculated, as well as weighted averages. Markups are the differences between the average purchase price and average sale price. 0.0726 is 7.26 basis points. Observations with markups of more than $10 or less than -$10 are excluded.

TBA Dollar Rolls TBA Outright Trades Specified Pools Maturity Coupon Markup Obs. Markup Obs. Markup Obs.

30 2.5% -0.0038 (-0.35)

158 0.0726 (6.86)

614 0.4549 (16.29)

237

30 3.0% 0.0037 (1.42)

1,412 0.0585 (9.11)

2,917 0.2162 (13.76)

1,551

30 3.5% 0.0036 (1.75)

2,445 0.0179 (4.89)

4,514 0.1056 (9.61)

3,483

30 4.0% 0.0036 (1.87)

2,376 0.0029 (0.78)

3,974 0.1284 (7.58)

2,191

30 4.5% 0.0001 (0.05)

2,061 0.0150 (3.46)

3,011 0.1399 (5.48)

1,608

30 5.0% 0.0026 (0.45)

1,774 0.0253 (8.77)

2,244 0.1390 (2.72)

364

30 5.5% 0.0047 (1.95)

1,240 0.0509 (5.86)

1,618 0.0798 (2.24)

33

30 6.0% 0.0044 (1.28)

908 0.0388 (5.83)

1,173 0.8959 (3.14)

11

15 2.5% 0.0055 (2.04)

753 0.0131 (3.28)

1,432 0.1567 (7.75)

897

15 3.0% 0.0066 (2.77)

903 0.0099 (2.41)

1,836 0.3147 (9.86)

1,082

15 3.5% 0.0034 (1.29)

665 0.0057 (1.00)

1,371 0.1217 (4.79)

505

15 4.0% 0.0017 (0.62)

567 0.0878 (6.02)

1,011 0.1263 (2.78)

351

25  

Table 3. Regressions of Price Changes on Quote Changes and Interactions Between Quote Changes and Other Variables

We regress changes in price between two consecutive trades of the same MBS on the change in trade type (ΔQ), on the interaction between ΔQ and the sum of the natural logs of the trade sizes of the two consecutive trades, on the interaction between ΔQ and a dummy variable that is one when the specified pool is TBA eligible, on the interaction between ΔQ, trade size and TBA eligibility, and on changes in the 1) a U.S. Agency Fixed Rate MBS index, 2) a U.S. Treasury 7-10 year Bond index, 3) a U.S. Investment Grade Corporate Bond Index, 4) a U.S. Corporate High-Yield Bond Index, and 5) the S&P 500 index. Consecutive trades are always of the same MBS, but trades from all MBS with the same coupon and maturity are included in the regressions. Robust t-statistics are reported in parentheses. Panel A. Specified pool trades only.

Maturity

Coupon

ΔQ

ΔQ x Trade Size

ΔQ x TBA Eligible

ΔQ x TBA Elg x Size

Return Variables

Obs.

R2

30 Years 2.5% 1.4110 (8.21)

-0.0563 (-5.00)

-0.2620 (-0.67)

-0.0052 (-0.23)

Yes 3,866 0.3533

30 Years 3.0% 2.7446 (16.46)

-0.1316 (-12.55)

-1.6959 (-8.85)

0.0788 (6.73)

Yes 18,789 0.1649

30 Years 3.5% 3.1419 (33.08)

-0.1612 (-29.92)

-2.6531 (-22.64)

0.1388 (20.84)

Yes 45,900 0.1861

30 Years 4.0% 2.0582 (11.49)

-0.1008 (-10.09)

-1.0493 (-5.59)

0.0466 (4.42)

Yes 46,982 0.2737

30 Years 4.5% 2.6057 (8.83)

-0.1345 (-7.62)

-1.5244 (-4.97)

0.0739 (4.02)

Yes 34,399 0.1888

30 Years 5.0% 4.6320 (9.67)

-0.2758 (-9.57)

-3.9847 (-7.52)

0.2363 (7.45)

Yes 5,296 0.2767

30 Years 5.5% 9.7998 (4.92)

-0.6160 (-4.81)

-9.9535 (-4.58)

0.6518 (4.68)

Yes 755 0.2549

30 Years 6.0% 1.5939 (0.66)

-0.0395 (-0.14)

-0.2586 (-0.11)

0.0012 (0.00)

Yes 225 0.2281

15 Years 2.5% 1.6578 (3.44)

-0.0821 (-2.91)

-0.1857 (-0.37)

0.0039 (0.13)

Yes 13,137 0.0952

15 Years 3.0% 3.3452 (7.81)

-0.1794 (-6.58)

-1.9187 (-4.33)

0.1070 (3.81)

Yes 19,499 0.0792

15 Years 3.5% 2.1303 (2.97)

-0.0949 (-2.01)

-1.0726 (-1.48)

0.0465 (0.98)

Yes 14,160 0.3055

15 Years 4.0% 5.2279 (3.77)

-0.3215 (-3.72)

-4.4545 (-3.20)

0.2872 (3.31)

Yes 7,964 0.3285

26  

Table 3, Panel B. TBA trades only.

Maturity

Coupon

ΔQ ΔQ x Trade

Size Return

Variables

Obs.

R2

30 Years 2.5% 0.7120 (7.45)

-0.0341 (-6.86)

Yes 6,770 0.0957

30 Years 3.0% 0.2422 (20.04)

-0.0112 (-17.87)

Yes 167,165 0.0924

30 Years 3.5% 0.1028 (14.56)

-0.0050 (-13.39)

Yes 212,745 0.0638

30 Years 4.0% 0.0554 (4.32)

-0.0041 (-6.15)

Yes 137,073 0.0522

30 Years 4.5% 0.1002 (5.31)

-0.0051 (-5.27)

Yes 67,942 0.0473

30 Years 5.0% 0.0909 (8.00)

-0.0040 (-7.17)

Yes 31,786 0.0479

30 Years 5.5% 0.1594 (10.57)

-0.0071 (-9.06)

Yes 18,295 0.0683

30 Years 6.0% 0.1493 (9.55)

-0.0067 (-8.07)

Yes 10,807 0.0428

15 Years 2.5% 0.0886 (15.09)

-0.0040 (-12.54)

Yes 51,189 0.1497

15 Years 3.0% 0.0518 (2.56)

-0.0031 (-2.90)

Yes 51,331 0.1028

15 Years 3.5% 0.0356 (2.39)

-0.0025 (-3.17)

Yes 28,478 0.1390

15 Years 4.0% 0.3511 (10.02)

-0.0163 (-8.90)

Yes 13,697 0.0575

27  

Table 4 Estimates of Round-Trip Trading Costs for TBA Trades, Specified Pool Trades of MBS that are TBA Eligible, and Other Specified Pool Trades. Estimates are based on regression coefficients in Table 3.

Trades = $100,000 Trades = $1,000,000 Trades = $5,000,000

Maturity

Coupon

TBA TBA

Eligible TBA Non-

Eligible

TBA TBA

Eligible TBA Non-

Eligible

TBA TBA

Eligible TBA Non-

Eligible 30 Years 2.5% 0.3978 0.5825 0.8921 0.2407 0.2992 0.6326 0.1309 0.1013 0.4512 30 Years 3.0% 0.1389 0.5626 1.5328 0.0872 0.3196 0.9268 0.0511 0.1497 0.5033 30 Years 3.5% 0.0565 0.2828 1.6573 0.0333 0.1798 0.9151 0.0171 0.1078 0.3962 30 Years 4.0% 0.0178 0.5100 1.1298 -0.0010 0.2605 0.6656 -0.0142 0.0861 0.3411 30 Years 4.5% 0.0532 0.5228 1.3677 0.0298 0.2435 0.7472 0.0133 0.0483 0.3142 30 Years 5.0% 0.0539 0.2815 2.0921 0.0354 0.0999 0.8222 0.0224 -0.0270 -0.0655 30 Years 5.5% 0.0942 0.1765 4.1265 0.0616 0.3416 1.2899 0.0388 0.4571 -0.6928 30 Years 6.0% 0.0875 0.9825 1.2296 0.0566 0.8061 1.0475 0.0350 0.6828 0.9202 15 Years 2.5% 0.0519 0.7519 0.9017 0.0335 0.3918 0.5236 0.0206 0.1401 0.2593 15 Years 3.0% 0.0234 0.7597 1.6928 0.0093 0.4263 0.8667 -0.0006 0.1933 0.2892 15 Years 3.5% 0.0129 0.6121 1.2564 0.0016 0.3892 0.8194 -0.0063 0.2334 0.5140 15 Years 4.0% 0.2009 0.4574 2.2669 0.1258 0.2994 0.7864 0.0733 0.1889 -0.2484

28  

Table 5

, , . , , ,

Maturity Coupon Intercept Log Size Dealer Fixed Effects Adj.R2 30 Yrs. 2.5% 5.4145

(9.54) -0.5628 (-9.11)

F(51, 2,674) = 0.842 p = 0.779 0.0210

30 Yrs. 3.0% 1.1250 (20.71)

-0.1174 (-19.55)

F(100, 57,653) = 30.531 p = 0.000 0.0541

30 Yrs. 3.5% 0.3986 (14.44)

-0.0402 (-13.12)

F(119, 79,894) = 16.480 p = 0.000 0.0255

30 Yrs. 4.0% 0.4088 (9.92)

-0.0385 (-8.45)

F(103, 58,034) = 79.493 p = 0.000 0.1255

30 Yrs. 4.5% 0.6185 (7.99)

-0.0546 (-6.38)

F(101, 32,493) = 97.577 p = 0.000 0.2358

30 Yrs. 5.0% 0.8875 (7.10)

-0.0840 (-6.17)

F(76, 15,785) = 13.403 p = 0.000 0.0621

30 Yrs. 5.5% 3.0974 (9.63)

-0.3020 (-8.19)

F(61, 8,935) = 8.308 p = 0.000 0.0654

30 Yrs. 6.0% 6.5417 (10.80)

-0.6469 (-9.15)

F(45, 5,221) = 7.412 p = 0.000 0.0897

15 Yrs. 2.5% 0.3266 (6.95)

-0.0345 (-6.58)

F(70, 20,613) = 0.434 p = 1.000 -0.0000

15Yrs. 3.0% 0.3877 (6.48)

-0.0407 (-5.91)

F(79, 22,318) = 64.224 p = 0.000 0.1841

15 Yrs. 3.5% 1.0997 (8.79)

-0.1172 (-8.04)

F(68, 13,910) = 17.556 p = 0.000 0.0796

15 Yrs. 4.0% 2.2229 (8.50)

-0.2370 (-7.63)

F(68, 7,183) = 0.364 p = 1.000 0.0037

29  

Table 6.

For every TBA CUSIP with two or more sales to customers on the same day, we calculate a size-weighted average price. For a sale i, Pricediffi is the difference between that sales price and the size-weighted average sales price for that CUSIP on the same day. The dealer who participates in sale i to a customer is defined as inactive if it completed fewer than 1,000 trades in the previous three months. For these dealers the dummy variable for inactive takes a value of one, for trades by other dealers, the value of the dummy variable is zero. Sizediffi is the difference between the log of the trade size of trade i, and the average log trade size of all sales in that CUSIP that day. We run the following regression using all TBA sales or all TBA sales in CUSIPs with a particular maturity-coupon combination.

T-statistics, reported in parentheses are robust. Panel A. All trades.

Maturity

Coupon

Inactive Dealer

Difference in Log Sizes

Constant

Obs.

R2

All

0.0728 (37.01)

-0.0110 (-45.24)

0.0142 (45.46)

322,037 0.0261

15 Years 2.5% 0.0340 (7.78)

-0.0040 (-13.27)

0.0055 (9.83)

20,321 0.0181

15 Years 3.0% 0.0253 (7.51)

-0.0034 (-9.03)

0.0054 (9.09)

21,786 0.0103

15 Years 3.5% 0.1614 (14.10)

-0.0036 (-6.92)

0.0029 (3.47)

13,188 0.0901

15 Years 4.0% 0.3321 (15.03)

-0.0090 (-6.34)

-0.0104 (-4.60)

6,389 0.1191

30 Years 2.5% -0.0140 (-0.78)

-0.0137 (-3.36)

0.0150 (2.74)

2,200 0.0073

30 Years 3.0% 0.0591 (15.06)

-0.0137 (-18.95)

0.0178 (20.75)

56,941 0.0251

30 Years 3.5% 0.0464 (14.53)

-0.0125 (-24.60)

0.0191 (28.83)

78,969 0.0226

30 Years 4.0% 0.0505 (11.11)

-0.0138 (-22.05)

0.0200 (23.89)

56,308 0.0243

30 Years 4.5% 0.0521 (10.38)

-0.0148 (-18.27)

0.0209 (18.64)

30,122 0.0312

30 Years 5.0% 0.0857 (6.26)

-0.0050 (-6.93)

0.0042 (3.80)

14,317 0.0198

30 Years 5.5% 0.0885 (7.74)

-0.0051 (-5.79)

0.0014 (1.59)

7,986 0.0522

30 Years 6.0% 0.0460 (6.26)

-0.0026 (-3.46)

-0.0011 (-0.80)

4,334 0.0221

30  

Table 6 (continued)

Panel B. Difference between trade size and average trade size is less than $1 million

Maturity

Coupon Inactive Dealer

Difference in Log Sizes

Constant

Obs.

R2

All

0.0902 (17.32)

-0.0050 (-1.65)

-0.0029 (-1.19)

16,705 0.0310

15 Years 2.5% 0.0177 (1.55)

0.0026 (0.43)

-0.0003 (-0.04)

654 0.0016

15 Years 3.0% 0.0744 (4.82)

0.0126 (1.60)

-0.0168 (-2.32)

959 0.0390

15 Years 3.5% 0.4826 (15.63)

0.0032 (0.32)

-0.0254 (-3.65)

1,030 0.3810

15 Years 4.0% 0.3263 (17.22)

0.0191 (1.65)

-0.0468 (-5.24)

1,999 0.1755

30 Years 2.5% -0.1068 (-2.08)

-0.2335 (-2.84)

0.0950 (2.53)

279 0.0777

30 Years 3.0% 0.0165 (1.11)

-0.0323 (-2.34)

0.0279 (2.26)

1,523 0.0166

30 Years 3.5% 0.0477 (4.58)

-0.0017 (-0.31)

0.0066 (1.36)

2,183 0.0101

30 Years 4.0% 0.0574 (2.51)

0.0008 (0.10)

0.0033 (0.47)

1,826 0.0098

30 Years 4.5% 0.0590 (3.02)

-0.0356 (-2.91)

0.0170 (1.96)

1,143 0.0361

30 Years 5.0% 0.2529 (4.10)

-0.0131 (-2.30)

0.0043 (1.54)

820 0.1524

30 Years 5.5% 0.0610 (3.81)

-0.0187 (-5.03)

-0.0002 (-0.05)

1,034 0.0576

30 Years 6.0% 0.0743 (6.97)

-0.0172 (-3.24)

0.0022 (0.55)

684 0.1401

31  

Table 7. For every TBA CUSIP with two or more purchases from customers on the same day, we calculate a size-weighted average price. For a purchase i, Pricediffi is the difference between that purchase price and the size-weighted average purchase price for that CUSIP on the same day. The dealer who participates in purchase i from a customer is defined as inactive if it completed fewer than 1,000 trades in the previous three months. For these dealers the dummy variable for inactive takes a value of one, for trades by other dealers, the value of the dummy variable is zero. Sizediffi is the difference between the log of the trade size of trade i, and the average log trade size of all purchases in that CUSIP that day. We run the following regression using all TBA purchases or all TBA purchases in CUSIPs with a particular maturity-coupon combination.

T-statistics, reported in parentheses are robust. Panel A. All purchases.

Maturity

Coupon

Inactive Dealer

Difference in Log Sizes

Constant

Obs.

R2

All

-0.0172 (-15.14)

0.0030 (21.54)

-0.0064 (-25.62)

461,574 0.0018

15 Years 2.5% -0.0155 (-7.16)

0.0051 (16.95)

-0.0086 (-16.43)

29,638 0.0107

15 Years 3.0% -0.0223 (-4.69)

0.0059 (13.82)

-0.0103 (-12.97)

27,269 0.0074

15 Years 3.5% -0.0163 (-4.14)

0.0033 (5.05)

-0.0036 (-2.71)

12,564 0.0035

15 Years 4.0% -0.0101 (-1.62)

0.0019 (2.31)

0.0004 (0.16)

5,482 0.0009

30 Years 2.5% -0.0132 (-0.94)

0.0119 (5.73)

-0.0117 (-4.15)

3,226 0.0093

30 Years 3.0% -0.0164 (-10.48)

0.0061 (25.43)

-0.0146 (-39.46)

107,695 0.0071

30 Years 3.5% -0.0216 (-10.96)

0.0029 (9.87)

-0.0068 (-15.02)

129,867 0.0022

30 Years 4.0% -0.0257 (-9.76)

0.0013 (3.02)

-0.0012 (-1.37)

74,164 0.0016

30 Years 4.5% -0.0265 (-6.13)

0.0008 (1.44)

0.0031 (2.72)

31,416 0.0023

30 Years 5.0% -0.0174 (-2.41)

0.0005 (0.86)

0.0030 (2.53)

14,404 0.0012

30 Years 5.5% 0.0158 (1.55)

0.0008 (1.64)

-0.0006 (-0.57)

8,246 0.0016

30 Years 6.0% 0.0198 (0.97)

0.0012 (1.25)

-0.0006 (-0.31)

4,741 0.0011

32  

Table 7. Panel B. Difference between trade size and average trade size is less than $1 million.

Maturity

Coupon Inactive Dealer

Difference in Log Sizes

Constant

Obs.

R2

All

-0.0182 (-3.89)

0.0068 (3.50)

-0.0067 (-4.38)

28,690 0.0016

15 Years 2.5% -0.0354 (-3.31)

0.0061 (1.50)

-0.0074 (-1.90)

1,131 0.0124

15 Years 3.0% -0.0081 (-0.40)

0.0333 (3.29)

-0.0353 (-3.76)

1,385 0.0130

15 Years 3.5% -0.0774 (-2.34)

0.0262 (2.37)

-0.0034 (-0.44)

795 0.0290

15 Years 4.0% -0.0222 (-0.71)

0.0182 (2.23)

0.0017 (0.21)

653 0.0067

30 Years 2.5% -0.0130 (-0.64)

0.0283 (1.23)

0.0001 (0.02)

671 0.0053

30 Years 3.0% -0.0146 (-2.02)

-0.0022 (0.28)

-0.0066 (-1.06)

5,052 0.0009

30 Years 3.5% -0.0101 (-1.26)

-0.0005 (-0.11)

-0.0053 (-1.79)

6,295 0.0003

30 Years 4.0% -0.0142 (-1.94)

-0.0008 (-0.16)

-0.0039 (-1.00)

3,948 0.0009

30 Years 4.5% -0.0056 (-0.24)

0.0013 (0.20)

-0.0030 (-0.59)

1,843 0.0001

30 Years 5.0% -0.0449 (-1.78)

0.0146 (2.55)

-0.0035 (-0.79)

1,013 0.0147

30 Years 5.5% -0.0337 (-1.44)

0.0076 (1.83)

-0.0027 (-0.77)

841 0.0117

30 Years 6.0% -0.0065 (-0.58)

0.0207 (2.05)

-0.0061 (-1.05)

477 0.0225

33  

Table 8.

∆ , ∆ . , ,

∆ , ∆ . , ,

Same Agency TBA All Agencies TBA

Maturity

Coupon Number Dealers

Sum of Obs.

Median α2 Coefficient

Median T-statistic

Median β2 Coefficient

Median T-statistic

Fannie Mae 30 Year 2.5% 22 431 -0.0035 -0.20 -1.2854 -6.88 30 Year 3.0% 50 2,818 -0.0348 -1.67 -1.0812 -7.66 30 Year 3.5% 61 4,866 -0.4756 -5.02 -1.0670 -6.32 30 Year 4.0% 67 4,826 -0.7775 -4.48 -0.9935 -5.14 30 Year 4.5% 54 2,914 -0.8973 -3.28 -1.1409 -4.07 30 Year 5.0% 37 1,166 -1.2914 -2.93 -1.3229 -2.67 15 Year 2.5% 53 2,748 0.0008 0.12 -0.9296 -8.58 15 Year 3.0% 68 3,857 -0.0003 -0.19 -0.8897 -3.26 15 Year 3.5% 48 2,519 -0.0430 -0.07 -1.1359 -2.04 15 Year 4.0% 38 1,170 -0.9631 -0.45 -0.8626 -0.46

Freddie Mac 30 Year 2.5% 47 1,173 -0.5200 -3.09 -0.5065 -2.99 30 Year 3.0% 70 3,414 -1.0446 -5.61 -1.0365 -5.34 30 Year 3.5% 84 5,527 -0.7788 -5.61 -0.8937 -6.63 30 Year 4.0% 71 4,568 -0.4202 -3.45 -1.1770 -5.14 30 Year 4.5% 68 2,800 -0.1651 -2.00 -0.9901 -3.71 30 Year 5.0% 48 1,458 -0.1423 -0.71 -0.9852 -2.28 15 Year 2.5% 64 2,339 -0.3575 -1.73 -0.3562 -1.73 15 Year 3.0% 61 3,198 -0.7464 -2.11 -0.7947 -2.40 15 Year 3.5% 60 2,084 -1.0168 -1.97 -0.8992 -2.22 15 Year 4.0% 46 1,262 -0.2046 -0.85 -0.6273 -0.80

Ginnie Mae 30 Year 2.5% 38 1,102 -0.0023 -0.02 -0.0287 -0.20 30 Year 3.0% 42 1,226 0.0000 0.08 0.1386 0.37 30 Year 3.5% 62 1,880 -0.6444 -7.24 -0.8448 -3.69 30 Year 4.0% 47 1,761 -0.7216 -5.19 -1.0001 -2.23 30 Year 4.5% 43 1,174 -0.8321 -5.86 -1.0261 -2.69 30 Year 5.0% 30 554 -1.0489 -2.69 -0.7755 -0.85 15 Year 2.5% 39 1,173 0.0059 0.28 -0.9046 -1.40 15 Year 3.0% 51 1,705 0.0239 0.37 -0.9293 -0.79 15 Year 3.5% 46 1,218 0.2634 0.35 0.0803 0.05 15 Year 4.0% 33 636 -0.4219 -1.00 -1.3078 -1.17

34  

Table 9. The impact of daily changes in specified pool inventory on same day changes in TBA inventory of MBS with the same coupon and maturity.

Panel A. Dealer observations weighted by number of days. A day is only included in a dealer regression if the change in specified pool inventory is different from zero. Standard errors of both coefficients must be greater than zero for a dealer regression to be included.

∆ , ∆ . , ∆ , ,

Regression Coefficients on Specified Pool Inventory Change

Coefficients on Lagged TBA Inventory Change

Maturity

Coupon

Number Dealers

Sum of Obs.

Median Coefficient

Mean Coefficient

Median T-statistic

Mean T-statistic

Median Coefficient

Mean Coefficient

Median T-statistic

Mean T-statistic

30 Year 2.5% 18 431 -0.8761 -0.7452 -6.70 -10.77 -0.1347 -0.1375 -0.30 -0.79 30 Year 3.0% 45 3,583 -0.9870 -0.9499 -9.93 -11.29 -0.0130 0.0131 -0.12 -0.07 30 Year 3.5% 59 6,708 -0.8983 -0.8801 -11.27 -11.42 -0.0570 -0.0329 -1.00 -1.06 30 Year 4.0% 52 6,199 -1.0057 -0.9312 -7.00 -8.15 -0.0793 -0.0564 -1.43 -0.88 30 Year 4.5% 49 3,886 -0.9842 -0.5265 -5.41 -6.93 -0.0050 -0.2092 -0.15 0.24 30 Year 5.0% 33 1,935 -0.7478 -0.8350 -3.46 -3.03 -0.0892 0.2010 -0.71 -0.48 30 Year 5.5% 21 454 -1.5847 0.0591 -2.26 -3.11 -2.2644 -3.1094 -0.20 0.51 30 Year 6.0% 16 150 -0.3296 -1.0166 -0.30 -0.19 0.3923 0.3617 0.42 0.81 15 Year 2.5% 41 3,821 -0.9118 -0.9620 -9.68 -11.25 -0.0665 -0.2007 -1.00 -1.05 15 Year 3.0% 50 5,513 -0.7645 -0.7005 -7.57 -7.51 -0.0635 -0.0827 -1.06 -1.11 15 Year 3.5% 48 3,545 -0.9087 -0.7793 -5.32 -5.15 0.0138 -0.0466 0.18 -0.50 15 Year 4.0% 38 2,289 -0.8122 -0.8035 -4.39 -4.81 -0.0177 -0.0699 -0.13 -0.14

35  

Table 9, Panel B. Dealer observations weighted by number of days. A day is only included in a dealer regression if the change in specified pool inventory, or the change in specified pool odd maturity inventory is different from zero. Standard errors of both coefficients must be greater than zero for a dealer regression to be included.

∆ , ∆ . 15 30 , ∆ . , ,

Sum of Obs.

Coefficients on Specified Pool Inventory Change

Coefficients on Odd Maturity Specified Pool Inventory Change

Maturity (Years)

Coupon

Number Dealers

Median Coefficient

Mean Coefficient

Median T-statistic

Mean T-statistic

Median Coefficient

Mean Coefficient

Median T-statistic

Mean T-statistic

22-35 2.5% 42 4,987 -0.0967 -0.3387 -1.55 -7.23 -0.0048 -0.0611 -0.24 -0.77 22-35 3.0% 63 6,844 -0.9433 -0.7687 -9.50 -11.11 -0.5333 -0.5576 -2.77 -3.27 22-35 3.5% 83 10,018 -0.7894 -0.7393 -9.18 -10.80 -0.7437 -0.6619 -3.18 -4.50 22-35 4.0% 82 10,929 -0.8428 -0.7423 -7.87 -8.41 -0.7442 -0.6497 -7.31 -9.85 22-35 4.5% 82 10,933 -0.9094 -0.7093 -6.11 -8.18 -0.4569 -0.4406 -6.92 -9.60 22-35 5.0% 75 11,699 -0.6736 -0.4182 -2.50 -3.55 -0.2486 -0.2259 -6.43 -7.50 22-35 5.5% 55 10,964 -0.1560 -0.5257 -0.92 -2.16 -0.1394 -0.1597 -5.90 -8.25 22-35 6.0% 56 10,821 -0.0528 -0.3255 -0.19 -0.35 -0.0626 -0.1097 -3.17 -4.79 10-21 2.5% 65 9,760 -0.7378 -0.6201 -8.29 -8.84 -0.0288 -0.1142 -0.83 -1.61 10-21 3.0% 70 11,157 -0.7567 -0.6704 -7.29 -8.25 -0.0735 -0.1373 -1.37 -2.16 10-21 3.5% 77 10,562 -0.5825 -0.5421 -3.07 -4.22 -0.1174 -0.1441 -1.76 -2.65 10-21 4.0% 74 10,981 -0.6390 -0.6687 -3.74 -5.02 -0.1763 -0.2001 -3.86 -4.68

36  

Table 9, Panel C. Dealer observations weighted by number of days. A day is only included in a dealer regression if the change in specified pool TBA-eligible inventory, or the change in specified pool TBA-ineligible inventory is different from zero. Standard errors of both coefficients must be greater than zero for a dealer regression to be included.

∆ , ∆ . , ∆ . , ,

Sum of Obs.

Coefficients on TBA-Eligible Specified Pool Inventory Change

Coefficients on TBA-Ineligible Specified Pool Inventory Change

Maturity

Coupon

Number Dealers

Median Coefficient

Mean Coefficient

Median T-statistic

Mean T-statistic

Median Coefficient

Mean Coefficient

Median T-statistic

Mean T-statistic

30 Year 2.5% 25 1,682 -0.5855 -0.6200 -3.17 -9.19 -0.0189 -0.0607 -0.14 -1.82 30 Year 3.0% 51 5,179 -0.9669 -0.8868 -9.07 -10.71 -0.4440 -0.1971 -1.29 -1.81 30 Year 3.5% 69 8,117 -0.9183 -0.8504 -9.48 -10.77 -0.6235 -0.5169 -1.90 -2.22 30 Year 4.0% 62 7,407 -0.9836 -0.9111 -6.65 -8.11 -0.9291 -0.7398 -1.62 -1.91 30 Year 4.5% 54 4,457 -0.9619 -0.5485 -5.37 -7.00 -0.8025 -0.6852 -0.93 -1.39 30 Year 5.0% 35 2,206 -0.7974 -0.8272 -3.71 -3.30 -0.5708 0.3305 -0.53 -0.66 30 Year 5.5% 17 410 -1.8508 -0.6517 -2.40 -3.91 0.0000 -0.5344 0.00 0.47 30 Year 6.0% 10 126 -0.3629 -0.2826 -0.39 -0.74 -2.0629 -1.7635 -0.17 -0.14 15 Year 2.5% 45 4,713 -0.8925 -0.8942 -8.59 -10.94 -0.4237 -0.5326 -0.71 -1.34 15 Year 3.0% 51 5,961 -0.7891 -0.7058 -7.72 -7.81 -0.1500 -0.1591 -0.36 -0.80 15 Year 3.5% 44 3,940 -0.8859 -0.7883 -5.36 -5.18 -0.1897 -0.1503 -0.23 -0.16 15 Year 4.0% 41 2,404 -0.8413 -0.8029 -4.48 -4.77 -0.3728 -5.1305 -0.33 -0.73

37  

Table 10. Panel A. Regressions of consecutive trade changes in the price of specified pools on changes in trade type and interactions between changes in trade type and trade sizes, TBA eligibility, and whether the specified pool maturity is exactly 15 or 30 years. The following regression is run:

Δ Δ Δ ∙ ln Δ ∙ Δ ∙ ∙ ln Δ∙ 30 Δ ∙ 30 ∙ ln Σ , .

ΔPt is the difference in price for two successive trades of a particular specified pool. ΔQt is the difference in trade type between two consecutive trades. It is +1 if trade t-1 was a sale by a customer to a dealer and trade t was a purchase by a customer from a dealer. It is -1 if trade t-1 was a purchase from a customer by a dealer and trade t was a sale to a customer from a dealer. TBA Eligible is a dummy variable that takes a value of one if the specified is eligible for TBA trading and zero otherwise. 30Year is a dummy variable that has a value of one if the specified pool has a maturity of 15 or 30 years, and zero if the maturity is different. Observations from different specified pools are included together in the regressions.

38  

Panel A. Regression results.

Maturity

Coupon

ΔQ

ΔQ x 30 (15)Year

ΔQ x TBA Eligible

ΔQ x Trade Size

ΔQ x 30 (15) Year x Size

ΔQ x TBA Elg x Size

Return Variables

Obs.

R2

22-35 Years 2.5% 2.3175 (15.04)

-0.8734 (-3.97)

-0.3596 (-1.33)

-0.0981 (-9.78)

.0376 (2.62)

0.0018 (0.11)

Yes 5,273 0.2725

22-35 Years 3.0% 1.8893 (10.18)

0.8809 (4.95)

-1.7384 (-11.67)

-0.0711 (-5.99)

-0.0606 (-5.40)

.0795 (8.36)

Yes 21,132 0.1577

22-35 Years 3.5% 4.0393 (34.12)

-1.2591 (-13.10)

-2.2529 (-20.78)

-0.2088 (-26.15)

0.0694 (10.41)

0.1143 (16.98)

Yes 53,731 0.2226

22-35 Years 4.0% 3.6452 (37.69)

-0.7740 (-13.87)

-1.8968 (-19.11)

-0.1875 (-28.15)

0.0433 (11.05)

0.0902 (13.66)

Yes 78,032 0.2458

22-35 Years 4.5% 3.7900 (29.73)

-0.7366 (-10.64)

-1.9792 (-15.02)

-0.1937 (-20.31)

0.0280 (5.84)

0.1013 (10.50)

Yes 78,637 0.1375

22-35 Years 5.0% 4.0192 (18.55)

-0.8149 (-3.47)

-2.3581 (-10.73)

-0.1961 (-11.37)

0.0033 (0.22)

0.1301 (7.44)

Yes 81,390 0.0540

22-35 Years 5.5% 3.8340 (15.99)

-1.0610 (-1.11)

-2.3237 (-9.54)

-0.1730 (-8.77)

0.0459 (0.71)

0.1125 (5.60)

Yes 89,313 0.0366

22-35 Years 6.0% 4.4117 (13.56)

-1.1289 (-0.97)

-2.0964 (-6.33)

-0.2045 (-7.45)

0.0635 (0.75)

0.1136 (4.05)

Yes 58,029 0.0605

10-21 Years 2.5% 1.9129 (5.20)

0.0662 (0.44)

-0.5584 (-1.49)

-0.0621 (-2.64)

-0.0130 (-1.38)

-0.0025 (-0.11)

Yes 18,580 0.0756

10-21 Years 3.0% 2.4315 (6.05)

-0.1051 (-0.95)

-0.8883 (-2.20)

-0.1146 (-4.38)

0.0016 (0.23)

0.0391 (1.49)

Yes 33,151 0.0936

10-21 Years 3.5% 2.8167 (8.52)

-0.0860 (-0.99)

-1.6705 (-5.05)

-0.1523 (-7.28)

0.0010 (0.17)

0.1029 (4.90)

Yes 32,511 0.2222

10-21 Years 4.0% 2.7111 (6.77)

-0.3957 (-3.87)

-1.5044 (-3.73)

-0.1706 (-6.37)

0.0142 (2.10)

0.1192 (4.42)

Yes 30,971 0.1891

39  

Table 10. Panel B. Trading cost estimates from the regressions.

Trades = $100,000 Trades = $1,000,000 Trades = $5,000,000

Maturity

Coup.

TBA Eligible, 30 Year

Not TBA Eligible, 30

Year

Not TBA Eligible, Not

30 Years

TBA Eligible, 30 Year

Not TBA Eligible, 30

Year

Not TBA Eligible, Not

30 Years

TBA Eligible, 30 Year

Not TBA Eligible, 30

Year

Not TBA Eligible, Not

30 Years 22-35 Yrs 2.5% 0.5450 0.8877 1.4144 0.2752 0.6095 0.9629 0.0867 0.4151 0.6473 22-35 Yrs 3.0% 0.5509 1.5571 1.2343 0.3105 0.9505 0.9069 0.1424 0.5266 0.6780 22-35 Yrs 3.5% 0.2968 1.4966 2.1163 0.1815 0.8549 1.1548 0.1010 0.4063 0.4827 22-35 Yrs 4.0% 0.4758 1.5426 1.9181 0.2265 0.8783 1.0546 0.0523 0.4140 0.4510 22-35 Yrs 4.5% 0.4807 1.5271 2.0058 0.1840 0.7639 1.1137 -0.0234 0.2305 0.4901 22-35 Yrs 5.0% 0.2679 1.4281 2.2130 -0.0213 0.5399 1.3099 -0.2234 -0.0809 0.6786 22-35 Yrs 5.5% 0.3152 1.6025 2.2408 0.2480 1.0172 1.4442 0.2011 0.6081 0.8873 22-35 Yrs 6.0% 0.9339 1.9842 2.5282 0.8075 1.3349 1.5864 0.7193 0.8811 0.9281 10-21 Yrs 2.5% 0.7062 1.2879 1.3410 0.3489 0.9423 1.0550 0.0991 0.7007 0.8552 10-21 Yrs 3.0% 0.7578 1.2861 1.3762 0.4177 0.7659 0.8485 0.1800 0.4024 0.4797 10-21 Yrs 3.5% 0.6141 1.3372 1.4144 0.3911 0.6405 0.7132 0.2352 0.1535 0.2331 10-21 Yrs 4.0% 0.4693 0.8755 1.1402 0.2984 0.1556 0.3548 0.1789 -0.3476 -0.1942

40  

Figure 1. TBA and specified pool volume (in thousands of dollars) by day. Includes only purchases by investors from dealers.

$0

$25,000,000

$50,000,000

$75,000,000

$100,000,000

$125,000,000

$150,000,000

5/16/2011 9/11/2011 5/8/2012 11/1/2012 4/30/2012

specified pool

TBA Outright

TBA Dollar Roll

41  

Figure 2.  Weekly 15 and 30 year mortgage rates. Source: Freddie Mac. 

0.00

1.00

2.00

3.00

4.00

5.00

4/28/2011 10/27/2011 4/26/2012 10/25/2012 4/25/2013

15 Year Rate

30 Year Rate

42  

Figure 3. Mean Daily Prices of MBS Traded in the TBA and Specified Pool Markets

15 year 3%

15 year 4%

98

99

100

101

102

103

104

105

106

107

108

5/19/2011 12/2/2011 4/30/2012 10/3/2012 4/18/2013

TBA

Specified Pool

100

102

104

106

108

110

5/16/2011 8/8/2011 11/8/2011 4/25/2012

TBA

Specified Pool

43  

30 year, 4.0%

30 Year, 5%

98

100

102

104

106

108

110

112

5/16/2011 10/12/2011 3/12/2012 8/8/2012 1/25/2013

TBA

Specified Pool

104

106

108

110

112

114

116

5/15/2011 8/25/2011 1/3/2012 7/18/2012

TBA

Specified Pool