The Federal Reserve Board eagle logo links to home page

Skip to: [Printable Version (PDF)] [Bibliography] [Footnotes]
Finance and Economics Discussion Series: 2014-60 Screen Reader version

The scarcity value of Treasury collateral: Repo market effects of security-specific supply and demand factors*

Stefania D'Amico*, Roger Fan*, and Yuriy Kitsul*
December 12, 2014

Abstract:

In the special collateral repo market, forward agreements are security-specific, which may magnify demand and supply effects. We quantify the scarcity value of Treasury collateral by estimating the impact of security-specific demand and supply factors on the repo rates of all outstanding U.S. Treasury securities. We find an economically and statistically significant scarcity premium. This scarcity effect is quite persistent, passes through to Treasury market prices, and explains a significant portion of the flow-effects of LSAP programs, providing additional evidence for the scarcity channel of QE. Through the same mechanism, the Fed's reverse repo operations could alleviate potential shortages of high-quality collateral.

Keywords: Treasury bonds, repo contracts, special repo rate, supply-demand factors, liquidity, large scale asset purchase programs, treasury auctions.

JEL Classification: G1, G12, G19, C23, E43.


A growing literature finds significant price responses to expected and unexpected changes in the net supply of various securities, including stocks (e.g., Kaul et al., 2000; Shleifer, 1986; Wurgler & Zhuravskaya, 2002; Greenwood, 2005) and bonds (e.g., D'Amico & King, 2013; Lou et al., 2013; Brandt & Kavajecz, 2004), suggesting the presence of a "scarcity premium." In very liquid cash markets, price impacts of anticipated and repeated supply shocks are typically shown to be temporary, as this premium is quickly arbitraged away.1 In these cases, however, the securities in question generally have a large pool of close substitutes. Consequently, arbitrage is relatively riskless, allowing quantity fluctuations in a particular security to be readily absorbed in a broader market. This both makes it harder to isolate supply effects empirically and, arguably, reduces their importance from an asset-pricing standpoint.

This paper examines supply effects in the context of a vast and liquid market where substitution across assets plays no role. In the special collateral repurchase agreement (SC repo) market, collateralized transactions are security-specific (i.e., the contract precludes the possibility to deliver substitutes); therefore, the scarcity of the underlying collateral should be the main determinant of the transaction's cost, that is, the repo rate. We provide evidence that, in the Treasury SC repo market, supply effects are significant and persistent: the repo rate on a specific security falls in response to a reduction in the amount of that security and remains lower for at least three months. This response measures a scarcity premium that we also show passes through to Treasury cash market prices, having potentially important implications for both the conduct of monetary policy through operations that change the available supply of Treasury collateral, and the Treasury's management of the auction cycle of its securities.

In particular, we quantify the scarcity value of Treasury collateral by estimating the impact of security-specific supply factors on the SC repo rates of all available U.S. Treasury securities.2 Exploiting the daily cross-sectional variation of these security-level data over a period of almost four years, we estimate panel regressions to carefully pin down quantity effects. Quantity variation in our sample comes mostly from purchased and sold amounts of Treasury securities under various Federal Reserve (Fed) programs.3 Since these programs were targeting yields in the Treasury cash market rather than the repo market, it is safe to assume that they were not directly responding to changes in SC repo rates and are therefore exogenous. By tracking cumulative price responses in the months following these quantity shocks, we can estimate impulse-responses and gain some understanding of whether the inability to substitute across securities exacerbates the supply effects' persistence. Finally, in our panel specification, time dummies sweep out any market-wide effects, including Fed and Treasury actions that affect the overall repo market. Therefore, our security-specific estimates can be considered a lower bound on the total supply effect; this bound is shown to be significant and quite persistent.

The estimated average elasticities of SC repo rates to collateral supply factors capture how the borrowing cost of a loan collateralized by a specific bond changes as that bond becomes more or less scarce. Therefore, these elasticities should measure the portion of the repo rate that is solely due to changes in the scarcity of the underlying collateral (i.e., its scarcity value) and not other idiosyncrasies of the specific security such as a change in liquidity and/or credit quality. This is also ensured by explicitly controlling for security-level measures of liquidity such as the bid-ask spread and "credit-quality" measures such as maturity, which determines the length of exposure to interest-rate risk and can be thought of as capturing Treasuries' riskiness. Finally, we estimate separate effects for on- and off-the-run securities.

Our results indicate that security-specific demand and supply factors are statistically significant and carry the expected signs. In particular, the coefficient on the amount purchased at the Fed's operations is negative and significant for both on- and off-the-run securities, implying an average effect of -0.8 and -0.3 basis points per billion dollars, respectively. This suggests that as the supply of a specific security available to private investors shrinks, the repo rate decreases (and the specialness spread increases) and borrowers of that security face an increased holding cost since they must lend money at relatively lower interest rates. In addition, these impacts are larger in shorter-term securities, with an average effect of -1.8 and -0.5 basis points per billion dollars, for on- and off-the-run securities, respectively. The estimated effects are quite persistent, staying significant for about three months. Conversely, the coefficient on the amount of off-the-run securities sold at the Fed's operations is positive and significant, implying an average effect of 0.2 basis points per billion dollars. This indicates that an increase in the available supply of Treasury securities pushes repo rates up (and specialness spreads down). The coefficient for the Treasury issued amount is also positive and significant, at 0.4 basis points per billion dollars.4 In addition, when we quantify the pass-through of these changes in the repo scarcity value to Treasury cash market prices, our estimates suggest that a one basis point increase in the predicted repo scarcity premium translates on average into a cash premium of about 0.4 basis points.

This pass-through of the SC repo scarcity premium to Treasury cash market prices, explains how perfectly anticipated changes in supply can still affect Treasury prices when they occur. As shown by Duffie (1996) and confirmed by Jordan & Jordan (1997) and Buraschi & Menini (2002), bonds that trade special in the repo market should trade at a premium in the cash market.5 Since we show that part of this repo scarcity premium originates from the Fed purchase operations and is priced in the Treasury cash premium, our results provide additional evidence in favor of the scarcity channel of quantitative easing (QE) (e.g., D'Amico et al., 2012; Krishnamurthy & Vissing-Jorgensen, 2011). Specifically, the estimated pass-through explains a significant portion of the Treasury price reactions to the Fed's actual purchase operations as estimated in D'Amico & King (2013), also known in the QE literature as flow- or pace-effects.

Our findings also have potentially important implications for both the future conduct of monetary policy through fixed-rate full-allotment (FRFA) reverse repos and the Treasury's management of the auction cycles of its securities.6 If the Fed decides to fully employ FRFA reverse repos, it could in theory become the largest (and most creditworthy) borrower in the repo market with the power to set a floor on repo rates (Martin et al., 2013). Our estimates suggest that, indeed, by changing the net supply of Treasury collateral, the Fed's reverse repos could potentially be successful in both controlling money market rates and alleviating shortages of high-quality collateral.7 Regarding Treasury auction cycle management, our results indicate that available options such as increasing the issuance at auction and/or reopening a security could reduce the scarcity premium by increasing the tradable supply.

Finally, our results can help quantify the potential impact on the repo market of new financial regulation that might affect the net supply of high-quality collateral such as Treasuries. For example, the new bank holding companies' supplementary leverage and liquidity coverage ratios might lead to a reduced willingness and ability to engage in repo transactions; and the mandatory central clearing of standardized over-the-counter derivatives (OTCD) will increase demand for high-quality assets by requiring initial margin on most OTCD transactions and limiting the re-hypothecation of pledged assets.8

The paper is organized as follows. Section 1 describes the data and the variables used in the empirical analysis, whose results are discussed in detail in Section 2. In Section 3 we estimate the pass-through of the repo scarcity premium to Treasury cash prices. In Section 4, we provide updated evidence on the impact of Treasury auction characteristics on SC repo rates. And Section 5 concludes.


1 Data Background and Description

1.1 Repo Market Background

A repo is a transaction involving the spot sale of a security coupled with a simultaneous forward agreement to buy back the same security, usually on the next day. Thus, it is similar to a collateralized overnight loan where the party providing the funds earns interest at the repo rate. In general collateral (GC) repos the acceptable collateral can be any of a variety of securities, while in specific or special collateral (SC) repos the contract is specific to the particular asset that serves as collateral.9 In this study, we limit our attention to Treasury collateral. The Treasury repo market is a vast market where the high quality of the collateral attracts many market participants and over the past decades has grown dramatically in size and popularity.10

In particular, GC repos are used by dealers and other levered accounts (such as hedge funds) as an inexpensive way to fund much of their activity. Money market mutual funds, corporate treasuries, and municipalities are among the most frequent cash providers in this market, as GC repos represent a relatively safe and liquid money-market instrument (Gorton & Metrick, 2012). SC repos are used by dealers and hedge funds to establish short positions (Duffie, 1996), that is, to borrow a specific security, which they then sell short in the secondary market in anticipation of a price reduction by the settlement date. This implies that anyone who sold that specific collateral short must deliver that bond and not some other bond and therefore will put extra value on that collateral. Mutual and pension funds, custodial agents, and other owners of Treasury securities can then borrow cash at an advantageous rate by lending specific securities, and eventually re-lend the money at a higher GC repo rate, capturing the spread between the two rates. These transactions are often open, that is, the agreement has an overnight tenor but continues until one of the counterparties decides to close it (Adrian et al., 2011).

Overall, the Treasury repo market, by facilitating market making, hedging, and speculative activities, has been fundamental in ensuring liquidity to the Treasury cash market. And in particular, by mitigating leverage constraints (e.g., Gromb & Vayanos, 2010), it has facilitated arbitrage trading, which is essential to Treasury market efficiency. On the other hand, the smooth functioning of the repo market and prevailing SC repo rates depend on the availability of the underlying Treasury collateral. The latter relation, which has been little investigated at the security level across all outstanding Treasury securities, is the main object of our study.

1.2 Repo Rate Data

Our proprietary data set is derived from the repo interdealer-broker market. It includes daily averages of SC repo rates quoted between 7:30 and 10 a.m. (Eastern time). This time window is chosen because trading in the repo market begins at about 7 a.m., remains active until about 10 a.m., and then becomes light until the market closes at 3 p.m. Repo transactions with specific collateral are bilateral and are executed on a delivery versus payment (DVP) basis (i.e., same-day settlement). In these transactions, a collateral security is delivered into a cash lender's account in exchange for funds. The exchange occurs via FedWire or a clearing bank. In contrast, GC repo transactions often occur via the tri-party repo market, in which securities and cash are placed on the balance sheet of a custodial agent.

The repo specialness spread is defined as the difference between the overnight GC repo rate and the corresponding SC repo rate. This spread measures how special a security is in the repo market. Figure 1 shows the specialness spread for the 10-year on-the-run Treasury security, which, as can be seen, displays a significant amount of variation over our sample. The largest spikes usually coincide with Treasury auction announcements.

Figure 1: Repo specialness spread for the on-the-run 10-year Treasury security

Figure 1: Repo specialness spread for the on-the-run 10-year Treasury security. In this figure, there a line plotting daily repo specialness spread for the on-the-run 10-year Treasury security over time, which is measured on y-axis with a range from about 0 basis points to about 350 basis points. The time period covered in the figure spans from May 1, 2009 through December 31, 2012. In general, the repo specialness spread oscillates around zero, but periodically exhibits run-ups and spikes, reaching the levels occasionally exceeding 300 basis points.  The most pronounced spikes occurred in June and July 2009, September and December 2010, and September 2011.


To compute this spread, we use two data sources for Treasury GC repo rates. The first source is the General Collateral Finance (GCF) Repo Index, which is a tri-party repo platform maintained by the Depository Trust & Clearing Corporation (DTCC).11 This market is characterized as being primarily inter-dealer, although some commercial banks and Fannie Mae also participate. It is a fairly active market although its size is still small compared to that of the overall tri-party repo market. The second source for the Treasury GC repo rate is the daily survey of primary dealers conducted by the New York Fed. Dealers are instructed to report overnight GC repo activity with non-affiliated entities such as money market funds (Bartolini et al., 2011). The survey does not specify the market segment in which dealers' repo transactions take place, thus the data capture tri-party, GCF and bilateral transactions.

Since results are very similar using both the GCF and GC repo rates, we only report those based on the GCF repo rate as the primary dealer survey data are restricted.12 Overall, in this study, the specialness spread is mainly used for graphical purposes and comparisons to previous studies, as time dummies in our specification control for market-wide effects such as variation in the GC repo rate.


1.3 Federal Reserve Operations

During our sample period, from March 2009 to December 2012, the Fed conducted two Large-Scale Asset Purchase (LSAP) programs, one Reinvestment program, and two Maturity Extension Programs (MEPs).13 These programs have significantly altered the available supply and maturity composition of collateral in the Treasury repo market. Thus, some of the most relevant explanatory variables used in this study are the security-level daily amounts purchased and sold by the Fed under these programs, obtained from the New York Fed.14 In our regressions, to better account for the relative scarcity of each CUSIP, we use the Fed's purchased/sold amount as a percentage of the privately-held amount outstanding.15

Summary statistics of the Fed operations are shown in Table 1. In our sample, the Fed has conducted 3162 purchases and 810 sales of securities across various operations, where most of the CUSIPs have been purchased or sold multiple times. The average purchase's size is $420 million or 1.68% of the security's privately-held amount outstanding; while, the average sale's size is about $710 million or 2.86% of the security's privately-held amount outstanding. The majority of operations were concentrated in off-the-run securities (about 96% of purchases and 98% of sales). However, the average size of on-the-run purchases is well above the average size of off-the-run purchases.


Table 1: Summary Statistics - Fed Operations

  Mean Std. Dev. N
Total, percent_bought 1.68 2.57 3162
Total, amt_bought 4.2e+08 7.4e+08  
Total, percent_sold 2.86 4.56 810
Total, amt_sold 7.1e+08 9.2e+08  
On-The-Run, percent_bought 7.91 6.45 127
On-The-Run, amt_bought 2.3e+09 1.9e+09  
On-The-Run, percent_sold 1.24 1.37 15
On-The-Run, amt_sold 4.2e+08 4.8e+08  
Off-The-Run, percent_bought 1.42 1.86 3035
Off-The-Run, amt_bought 3.4e+08 5.2e+08  
Off-The-Run, percent_sold 2.89 4.59 795
Off-The-Run, amt_sold 7.1e+08 9.3e+08  

Amounts bought and sold are measured in dollars

We expect the impact of a sale or purchase operation to differ between on-the-run and off-the-run securities. For example, demand for short positions, a significant driver of repo rates (Duffie, 1996), is usually concentrated in the most liquid securities, as short sellers value the ability to quickly buy back those securities to cover or unwind their positions (Vayanos & Weill, 2008; Duffie et al., 2007). Therefore, the repo rates of on-the-run securities should be more sensitive to quantity factors. For this reason, we separately estimate the effects of the Fed operations for on- and off-the-run securities, though the small number of Fed sales of on-the-run securities limits our statistical power. By reducing the collateral available to the repo market, Fed purchases should decrease the SC repo rate and increase the specialness spread of the CUSIP purchased. Fed sales should have the opposite effect.

It is, however, important to take into account that once the purchased securities entered in the SOMA portfolio, they then became available through the Fed's Securities Lending Program (SLP), under which at noon of each business day the Fed offers to lend up to 90% of the amount of each Treasury security owned by SOMA on an overnight basis. But the SLP has constraints on the amount of an individual issue a dealer can borrow (25% of the lendable holdings) and the daily amount a dealer can borrow in aggregate across all issues ($5 billion).16 The program works through an auction mechanism to make loan pricing a market-driven process. Primary dealers bid for a security's loan specifying the quantity and the loan fee. The minimum fee is imposed to provide an incentive to borrow securities whose SC repo rates are below the GC repo rate.

In our regressions, we control for security-level uncovered bids at the SLP auctions, as any dealer who was not able to obtain the desired amount at the SLP to cover its positions would then have to seek the securities in the repo market, potentially pushing up demand for certain securities.


1.4 Treasury Auction Cycle

There are three important periodic dates in the Treasury auction cycle: the auction announcement date, the auction date, and the issuance date. There is usually about one week from the announcement to the auction. During a typical auction cycle, the supply of Treasury collateral available to the repo market is at its highest level when the security is issued, therefore the repo specialness spread should be close to zero. As time passes, more and more of the security may be purchased by holders who are not very active in the repo market, consequently the security's availability may decline over time and the repo specialness spread may increase. When forward trading in the next security begins on the announcement date (when-issued trading opens), holders of short positions will usually roll out of the outstanding issue, implying that demand for that specific collateral should decrease and that the repo specialness spread will rapidly decline (see Fisher, 2002). Keane (1995) documents that repo specialness for on-the-run securities exhibits this repeated pattern, that is, it climbs with the time since the last auction until around the announcement of the next auction, after which it declines sharply.

Figures 2 and 3 show the auction cycle patterns in our sample for securities auctioned monthly (2-, 3-, 5-, and 7-year maturities) and quarterly (10-year maturities), respectively. In Figure 2, it is easy to note the same pattern documented by Keane (1995) for securities with a monthly auction cycle. In contrast, Figure 3 shows that the quarterly auction cycle of the 10-year note looks quite different, mainly because the Treasury has introduced two regular reopenings following each 10-year note auction. Therefore, it is possible to observe three separate auction sub-cycles: the most dramatic run-up in specialness spread takes place before the first reopening; a second run-up, similar in shape but smaller in magnitude, immediately follows and peaks just before the second reopening; and finally, during the third sub-cycle the specialness spread is practically flat. This would suggest that the increased availability of the on-the-run security after each reopening strongly diminishes the impact of the seasonal demand for short positions around these dates (Sundaresan, 1994).

Figure 2: Average daily repo specialness spread for Treasury securities with a 1-month auction cycle (2-, 3-, 5-, and 7-year maturities).

Figure 2: Average daily repo specialness spread for Treasury securities with a 1-month auction cycle (2-, 3-, 5-, and 7-year maturities). In this figure, there are a line and dots plotting average daily repo specialness spread for 2-, 3-, 5-, and 7-year Treasury securities, which is measured on y-axis with a range from 0 to 20 basis points. The x-axis measures the number of days that elapsed since a new Treasury security is issued. Grey dots represent the actual average specialness spread on each day since the issue date, with the line being a LOESS curve fitted through these dots. Also plotted is a vertical dashed line which marks the average time of the auction of the next security with the same maturity.  The fitted average repo specialness spread gradually increases from about zero on the day of issuance to about 15 basis points approximately 3 weeks after the issuance and then begins to gradually decline.
Grey dots are the average specialness spread on each day since the issue date, and the line is a fitted LOESS curve. The vertical dashed line marks the average time of the auction of the next security with the same maturity.


Figure 3: Average daily repo specialness spread for 10-year Treasury securities.

Figure 3: Average daily repo specialness spread for 10-year Treasury securities. In this figure, there are three lines and dots plotting average daily repo specialness spread for a 10-year Treasury security, which is measured on y-axis with a range from 0 to 20 basis points. The x-axis measures the number of days that elapsed since a new 10-year Treasury security is issued. Also plotted are vertical dotted lines and a vertical dash line, which mark the average times of 10-year security reopening auctions (one and two months after the original issuance) and the average time of the new auction of the 10-year security (three months after the original issuance), respectively.  Grey dots represent the average specialness spread on each day since the issue date, and the three lines denote a LOESS curves fitted through three subsets of dots. The three dot subsets display repo spreads on the days prior to and a few days following the two reopenings and the next new auction, respectively. These lines represent three distinct sub-cycles, in each of which the repo spread first gradually increases and then declines.  The first line shows that the fitted average repo spread gradually increases from about 10 basis points to over about 25 business days (aside from a brief dip at the beginning) and reaches a level of roughly 85 basis points shortly before the first auction reopening (first dotted vertical line) and then begins to decline. The second line also begins at the level of about 10 basis points and, aside from a brief dip, shows a gradual increase to about 50 basis points in the third week after the first reopening.  Afterward, the line shows that the repo spread declines over the days around the second reopening (second dotted vertical line).    The third line which begins a few days after the second reopening (second dotted vertical line) and remains at the levels of 5-10 basis points.
Grey dots are the average specialness spread on each day since the issue date, and the line is a fitted LOESS curve. Vertical dotted lines mark the average times of reopening auctions, while the vertical dashed line denotes the average time of the auction of the next 10-year security.


In order to control for these auction-cycle effects, we construct a set of dummy variables that track the time elapsed since issuance for both the monthly and quarterly cycles.

1.5 Demand for Short Positions and Other Controls

In addition to quantifying changes in the available supply of collateral, we also aim to capture one of the most important demand factors in the repo market: demand for short positions. Duffie (1996), Duffie et al. (2007), and Vayanos & Weill (2008) all suggest that agents who create short positions prefer to trade securities that are expected to be liquid in the future, and often use reverse repo contracts to create these positions because they are less expensive than other options. Therefore, for a given supply of the security, the extent of specialness should be increasing in the demand for short positions.

To control for daily demand for short positions at the security level, on any given day and for each CUSIP, we compute the total amount of transactions initiated as reverse repos and subtract the total amount of transactions initiated as repos over the same period. This imbalance, which should capture the security's excess demand, can create price pressures in the specific security and might make it run special.


Table 2: Summary Statistics - Operation Days

On-The-Run, Mean On-The-Run, Std. Dev. Off-The-Run, Mean Off-The-Run, Std. Dev. Total, Mean Total, Std. Dev.
repo_avgrate 5.54 21.9 14.2 7.48 14 8.22
delta_repo -.188 6.83 .017 2.95 .0123 3.1
repo_spread 11.1 21.1 2.77 3.22 2.96 4.69
delta_repo_spread .136 6.56 -.0779 2.67 -.0729 2.82
repo_volume_sprd_std -.261 3.33 -.0263 .91 -.0318 1.03
bidaskspread 1.35 .56 3.15 2.42 3.1 2.41
delta_bidaskspread .00231 .573 -.00579 .921 -.0056 .914

SC repo rates and repo specialness spreads are measured in basis points.
Repo volume spreads are standardized and measured in standard deviations.
Bid-ask spreads are measured in cents.


Finally, since liquidity and specialness are often correlated (Duffie, 1996), especially for on-the-run securities, we explicitly control for securities' liquidity using individual bid-ask spreads measured in cents per hundred dollars.17 Securities with lower bid-ask spreads are more liquid, therefore we expect them to have lower repo rates and higher specialness spreads.


2 Empirical Results

We now turn to estimating the impact of the previously described security-specific demand and supply factors on SC repo rates (and repo specialness spreads) through a series of panel regressions. Various empirical specifications are estimated at a daily frequency where the dependent variable is the change in either the SC repo rate or the specialness spread for all outstanding nominal Treasury coupon securities. Unlike previous studies, we use changes rather than levels because these variables exhibit a high degree of serial correlation.

Another important advantage of using changes is that they mitigate any additional endogeneity concerns that might affect some of the controls and that are typical of exercises in which a price variable (the repo rate) is regressed on quantity factors. The rationale for this is based on the time at which repo rates are collected relative to when Fed operations, Treasury auctions, and SLP auctions are conducted. The SC repo rates are collected every morning from 7:30 to 10:00 a.m., while the regular Fed purchase and sale operations start at 10:15 a.m. and end at 11:00 a.m. In some cases, there can be a second operation between 1:15 and 2 p.m. of the same day. The SLP auctions start at 12 p.m. and end at 12:15 p.m.; and, the Treasury auction results for notes and bonds are normally announced at 1 p.m. This sequence of events implies that only the repo rate of the following morning will reflect information from these operations. At the same time, the change in the next day's repo rate cannot be factored into the Fed's and Treasury's operational decisions. Therefore, while the change in the repo rate from the morning of any given day to the next will reflect that day's operations, it will not affect the operations' implementation on the same day.

Our sample starts after the introduction of a repo fail charge by the Treasury Market Practices Group on May 1, 2009 to avoid a structural break in the series.18 However, due to data availability, specifications that include information on whether transactions were initiated as repos or reverse repos are estimated on a slightly shorter sample starting on June 23, 2009. We omit securities maturing in more than 15 years because the repo market in longer-term securities is very thin. As a result, our unbalanced panel consists of 347 CUSIPs.


2.1 Regression Specification

Our basic panel regression specification is:

\displaystyle \Delta SCR_{i, t} = \alpha + \beta_1 \Delta SF_{i, t} + \beta_2 \Delta DF_{i, t} + \beta_3 \Delta L_{i, t} + \beta_4 \tau_{i, t} + \beta_5 D_{i, t} + \gamma_t + \epsilon_{i, t} (1)

where for each security  i at time  t ,  \Delta SCR is the change in the SC repo rate in basis points;  \Delta SF represents changes in supply factors such as amount purchased and sold at each Fed operation rescaled by the security's privately-held amount outstanding;  \Delta DF represents changes in demand factors such as our proxy for short positions rescaled by the security's privately-held amount outstanding and the amount of uncovered bids at the SLP auctions;  \Delta L are controls for liquidity characteristics such as the change in the bid-ask spread;  \tau includes maturity and maturity squared;  D are dummies that control for the auction cycle discussed in Section 1.4; and  \gamma _{t} are daily time dummies that control for the evolution over time of common market-wide factors.

Indeed, the daily time dummies should completely absorb the variation in specialness spreads due to the variation in the Treasury GC repo rate, which summarizes the overall trading conditions in the Treasury repo market. This suggests that regressions with changes in SC repo rates or in specialness spreads are equivalent under this specification.

In addition, some variables are interacted with a dummy that divides the sample into two mutually exclusive subsamples: on-the-run vs. off-the-run securities. Finally, because Fed operations settle on the following day, we also use the two-day changes in the SC repo rate and specialness spread as dependent variables in our regressions. The rationale is that the impact of these operations might not be felt until the day in which the investors have to actually deliver or receive the security to or from the Fed.

The above specification is estimated using only days when Fed operations were conducted.19


2.2 Results

The results from the SC repo rate panel regression estimated starting on May 1, 2009 are reported in the first column of Table 3, while the second column shows the results for the two-day change in the same dependent variable.20,21 Both on- and off-the-run Fed purchases have negative and statistically significant effects on SC repo rates, although their size appears to be considerably larger for on-the-run purchases. The coefficient of -0.234 suggests that buying one percent of a security's outstanding par value would decrease the SC repo rate by 0.234 basis points, implying that on average a $1 billion purchase of on-the-run securities would decrease the SC repo rate by 0.81 basis points. In contrast, the coefficient for the off-the-run securities implies a decline of 0.33 basis points for a purchase of the same size.

This suggests the existence of a scarcity premium, as a reduction in the available supply of a specific security would push its repo rate down, indicating that on average investors must lend money at relatively lower rates to obtain that specific security, facing an additional cost. And owners of that security would obtain financing at a more attractive rate, enjoying an extra profit. The coefficients for the same variables in the second column are slightly larger, suggesting that on the settlement day the impact from these operations not only persists but increases.


Table 3: SC Repo Rate Regressions, from May

  (1) d_repo (2) d2_repo
percent_bought_offtherun -0.0792*** -0.103***
percent_bought_offtherun, t-stat (-6.50) (-6.64)
percent_sold_offtherun 0.0480*** 0.0542***
percent_sold_offtherun, t-stat (3.90) (5.32)
percent_bought_ontherun -0.234*** -0.281***
percent_bought_ontherun, t-stat (-5.09) (-4.23)
percent_sold_ontherun -0.153 -0.305
percent_sold_ontherun, t-stat (-0.38) (-0.46)
SLP_pct_uncovered_off -0.00302 0.00498
SLP_pct_uncovered_off, t-stat (-0.94) (1.34)
SLP_pct_uncovered_on -0.0115 0.0642
SLP_pct_uncovered_on, t-stat (-0.34) (1.63)
delta_bidaskspread 0.000166 0.000728
delta_bidaskspread, t-stat (0.03) (0.09)
maturity 0.0175*** 0.0190***
maturity, t-stat (3.86) (3.45)
maturity2 -0.00123*** -0.00132**
maturity2, t-stat (-3.63) (-3.28)
 N 89614 88821
 R^{2} 0.730 0.729
adj.  R^{2} 0.729 0.728

Heteroskedasticity-consistent t statistics in parentheses
*  p<0.05 , **  p<0.01 , ***  p<0.001


The impact of Fed sales is positive and significant only for the off-the-run securities, which is not surprising given the small number (15) of on-the-run sales in our sample. The coefficient of 0.048 suggests that selling one percent of a security's outstanding would increase the SC repo rate by 0.048 basis points, implying that a $1 billion sale would increase the SC repo rate by 0.2 basis points. The cumulative impact is again slightly bigger on the settlement day. None of the other variables shown, except for maturity, are statistically significant. One possible explanation for the lack of significance of the SLP coefficient is that, as explained in Section 1.3, each dealer's participation is capped, making this tool less effective in releasing demand pressure.

The results from the same specification but augmented with the proxy for short positions and therefore estimated starting on June 23, 2009 are reported in Table 4. The demand for short positions (  repo\_volume\_sprd ) has a negative and statistically significant impact on SC repo rates, although the coefficient's size is much smaller than that one of the Fed purchases. In this case, the split in on- and off-the-run securities (not shown) does not affect its magnitude. Since all the other coefficients remain practically unchanged when estimated in this slightly shorter sample period, for the remainder of this section we focus only on the results obtained using this sample period.


Table 4: SC Repo Rate Regressions

  (1) d_repo_spread (2)d2_repo_spread
percent_bought_offtherun -0.0847*** -0.107***
percent_bought_offtherun, t-stat (-6.56) (-6.43)
percent_sold_offtherun 0.0487*** 0.0549***
percent_sold_offtherun, t-stat (3.95) (5.41)
percent_bought_ontherun -0.227*** -0.270***
percent_bought_ontherun, t-stat (-4.51) (-3.68)
percent_sold_ontherun -0.167 -0.138
percent_sold_ontherun, t-stat (-0.40) (-0.24)
SLP_pct_uncovered_off -0.00310 0.00486
SLP_pct_uncovered_off, t-stat (-0.97) (1.31)
SLP_pct_uncovered_on -0.00933 0.0456
SLP_pct_uncovered_on, t-stat (-0.27) (1.19)
repo_volume_sprd_std -0.0369* -0.0267
repo_volume_sprd_std, t-stat (-2.20) (-1.21)
delta_bidaskspread 0.00328 0.00119
delta_bidaskspread, t-stat (0.53) (0.15)
maturity 0.0159*** 0.0179**
maturity, t-stat (3.46) (3.22)
maturity2 -0.00105** -0.00122**
maturity2, t-stat (-3.07) (-2.98)
 N 87337 86551
 R^{2} 0.735 0.737
adj.  R^{2} 0.733 0.736

Heteroskedasticity-consistent t statistics in parentheses
*  p<0.05 , **  p<0.01 , ***  p<0.001


Table 5 reports the same regression as Table 4, except using the change in the repo specialness spread instead of the SC repo rate as the dependent variable. As discussed earlier, since we include time dummies, it is only the variation in the specific repo rates that drive our estimates. So these results are extremely similar to the previous regression except for the flipped sign. This is because any factor that pushes the SC repo rate down will push the corresponding specialness spread up, and vice versa.


Table 5: Repo Specialness Spread Regressions

  (1) d_repo_spread (2) d2_repo_spread
percent_bought_offtherun 0.0851*** 0.107***
percent_bought_offtherun, t-stat (6.62) (6.42)
percent_sold_offtherun -0.0486*** -0.0552***
percent_sold_offtherun, t-stat (-3.94) (-5.46)
percent_bought_ontherun 0.227*** 0.240***
percent_bought_ontherun, t-stat (4.51) (3.50)
percent_sold_ontherun 0.272 0.142
percent_sold_ontherun, t-stat (0.66) (0.26)
SLP_pct_uncovered_off 0.00312 -0.00489
SLP_pct_uncovered_off, t-stat (0.97) (-1.32)
SLP_pct_uncovered_on -0.00362 -0.0390
SLP_pct_uncovered_on, t-stat (-0.10) (-1.03)
repo_volume_sprd_std 0.0363* 0.0321
repo_volume_sprd_std, t-stat (2.17) (1.34)
delta_bidaskspread -0.00427 0.00180
delta_bidaskspread, t-stat (-0.70) (0.24)
maturity -0.0143** -0.0175**
maturity, t-stat (-3.17) (-3.14)
maturity2 0.000930** 0.00119**
maturity2, t-stat (2.76) (2.92)
 N 87321 86546
 R^{2} 0.686 0.668
adj.  R^{2} 0.684 0.666

Heteroskedasticity-consistent t statistics in parentheses
*  p<0.05 , **  p<0.01 , ***  p<0.001


We next break our data into three subsamples based on the securities' maturity. In particular, we consider possible differences between securities with shorter maturities that were eligible for both sale and purchase operations conducted by the Fed (during the MEP the Fed sold only securities maturing in 3 years or less), those with medium-term maturities (3 to 7 years), and securities with longer maturities (7 to 15 years). Table 6 presents the results for these subsamples. The coefficients for on- and off-the-run Fed purchases are both significantly larger for shorter-term securities, implying an average effect of -1.78 and -0.51 basis points per billion dollars, respectively. Again, the strong economic and statistical significance of these results confirm the existence of scarcity premia.


Table 6: SC Repo Rate Regressions; 1-day Changes

  (1) 0-3 Years (2) 3-7 Years (3) 7-15 Years
percent_bought_offtherun -0.139** -0.0686*** -0.0787***
percent_bought_offtherun, t-stat (-3.27) (-3.39) (-3.58)
percent_sold_offtherun 0.0465***    
percent_sold_offtherun, t-stat (3.77)    
percent_bought_ontherun -0.553*** -0.100** -0.414
percent_bought_ontherun, t-stat (-3.98) (-3.19) (-0.94)
percent_sold_ontherun -0.306    
percent_sold_ontherun, t-stat (-0.63)    
SLP_pct_uncovered_off -0.00395* 0.0110 0.00557
SLP_pct_uncovered_off, t-stat (-2.05) (0.16) (0.41)
SLP_pct_uncovered_on, t-stat (-0.12) (-1.17) (0.12)
repo_volume_sprd_std -0.0781* -0.0273 0.0103
repo_volume_sprd_std, t-stat (-2.56) (-1.26) (0.30)
delta_bidaskspread 0.0119 0.00419 -0.0126
delta_bidaskspread, t-stat (1.21) (0.41) (-0.86)
maturity 0.0915** 0.00438 -0.00222
maturity, t-stat (2.87) (0.05) (-0.03)
maturity2 -0.0174 0.000577 0.0000327
maturity2, t-stat (-1.96) (0.06) (0.01)
 N 45886 30194 11257
 R^{2} 0.766 0.749 0.641
adj.  R^{2} 0.764 0.745 0.625

Heteroskedasticity-consistent t statistics in parentheses
*  p<0.05 , **  p<0.01 , ***  p<0.001



Table 7: SC Repo Rate Regressions; 2-day Changes

  (1) 0-3 Years (2) 3-7 Years (3) 7-15 Years
percent_bought_offtherun -0.215*** -0.0827*** -0.0884***
percent_bought_offtherun, t-stat (-3.58) (-3.45) (-3.58)
percent_sold_offtherun 0.0539***    
percent_sold_offtherun, t-stat (5.23)    
percent_bought_ontherun -0.625** -0.119* 0.112
percent_bought_ontherun, t-stat (-3.08) (-2.20) (0.25)
percent_sold_ontherun -0.350    
percent_sold_ontherun, t-stat (-0.51)    
SLP_pct_uncovered_off -0.000209 0.103 0.0142
SLP_pct_uncovered_off, t-stat (-0.07) (1.90) (0.72)
SLP_pct_uncovered_on 0.0626 0.0474 0.0305
SLP_pct_uncovered_on, t-stat (1.29) (0.52) (0.46)
repo_volume_sprd_std -0.0688 -0.0509* 0.0719
repo_volume_sprd_std, t-stat (-1.50) (-2.28) (1.56)
delta_bidaskspread_pct 0.00279 0.00309 0.00756
delta_bidaskspread_pct, t-stat (0.24) (0.24) (0.39)
maturity 0.138*** 0.0734 0.165
maturity, t-stat (3.58) (0.57) (1.84)
maturity2 -0.0263* -0.00625 -0.00718
maturity2 , t-stat (-2.36) (-0.49) (-1.81)
 N 45474 29963 11114
 R^{2} 0.775 0.745 0.648
adj.  R^{2} 0.773 0.741 0.632

Heteroskedasticity-consistent t statistics in parentheses
*  p<0.05 , **  p<0.01 , ***  p<0.001


Further, in the case of shorter-term securities, the coefficient on off-the-run uncovered bids at the SLP is negative and significant, suggesting that if investors were unable to obtain the desired quantity of a specific security at the SLP, then on average they would lend money in the repo market at a relatively lower rate in exchange of that particular security. Table 7 shows results from the same regressions but using the two-day change in the SC repo rate, confirming that on the settlement day the magnitude of all the significant coefficients is a bit bigger.

As before, in Table 8 we report the same subsample results as in Table 6 except using the repo specialness spread as the dependent variable. Again, we can see that the coefficients are almost identical except for the flipped sign.


Table 8: Repo Specialness Spread Regressions; 1-day Changes

  (1) 0-3 Years (2) 3-7 Years (3) 7-15 Years
percent_bought_offtherun 0.138** 0.0701*** 0.0787***
percent_bought_offtherun, t-stat (3.25) (3.48) (3.58)
percent_sold_offtherun -0.0465***    
percent_sold_offtherun, t-stat (-3.77)    
percent_bought_ontherun 0.554*** 0.0955** 0.414
percent_bought_ontherun, t-stat (3.98) (3.04) (0.94)
percent_sold_ontherun 0.308    
percent_sold_ontherun, t-stat (0.63)    
SLP_pct_uncovered_off 0.00395* -0.0109 -0.00558
SLP_pct_uncovered_off, t-stat (2.05) (-0.16) (-0.41)
SLP_pct_uncovered_on 0.00544 -0.0329 -0.00705
SLP_pct_uncovered_on, t-stat (0.12) (-0.22) (-0.12)
repo_volume_sprd_std 0.0778* 0.0274 -0.0109
repo_volume_sprd_std, t-stat (2.55) (1.20) (-0.32)
delta_bidaskspread, t-stat (-1.53) (-0.40) (0.86)
maturity -0.0870** 0.0380 0.00245
maturity, t-stat (-2.79) (0.41) (0.04)
maturity2 0.0170 -0.00466 -0.0000485
maturity2 , t-stat (1.95) (-0.53) (-0.02)
 N 45878 30189 11254
 R^{2} 0.729 0.690 0.578
adj.  R^{2} 0.726 0.685 0.560

Heteroskedasticity-consistent t statistics in parentheses
*  p<0.05 , **  p<0.01 , ***  p<0.001


To account for possible correlations across the regression errors of collateral with comparable maturities, we also run the analysis with clustered standard errors. Table 9 shows the results of this robustness exercise. The first column shows estimates from the same model as the first column in Table 4, using heteroskedasticity-consistent standard errors. The second and third columns show the results from specifications where we allow for clustering within one- and three-year maturity buckets for each security. The results are robust to the type of standard error used, as the statistical significance of the estimated coefficients is practically unchanged. We perform the same exercise for the maturity subsample regressions presented in Tables 6 and 7 and obtain similar results (not shown). This is not surprising if, in the SC repo market, substitution across securities does not play any role. Therefore, quantity shocks would not be transmitted to similar securities, reducing cross-sectional correlations.


Table 9: SC Repo Rate Regressions; 1-day Changes

  (1) Robust (2) 1-yr Clust. (3) 3-yr Clust.
percent_bought_offtherun -0.0847*** -0.0847*** -0.0847***
percent_bought_offtherun, t-stat (-6.56) (-6.49) (-6.21)
percent_sold_offtherun 0.0487*** 0.0487*** 0.0487***
percent_sold_offtherun, t-stat (3.95) (3.69) (3.66)
percent_bought_ontherun -0.227*** -0.227*** -0.227***
percent_bought_ontherun, t-stat (-4.51) (-4.50) (-4.49)
percent_sold_ontherun -0.167 -0.167 -0.167
percent_sold_ontherun, t-stat (-0.40) (-0.40) (-0.40)
SLP_pct_uncovered_off -0.00310 -0.00310 -0.00310
SLP_pct_uncovered_off, t-stat (-0.97) (-0.95) (-0.96)
SLP_pct_uncovered_on -0.00933 -0.00933 -0.00933
SLP_pct_uncovered_on, t-stat (-0.27) (-0.27) (-0.27)
repo_volume_sprd_std -0.0369* -0.0369* -0.0369*
repo_volume_sprd_std, t-stat (-2.20) (-2.21) (-2.18)
delta_bidaskspread_pct 0.00328 0.00328 0.00328
delta_bidaskspread_pct, t-stat (0.53) (0.50) (0.48)
 N 87337 87337 87337
 R^{2} 0.735 0.735 0.735
adj.  R^{2} 0.733 0.733 0.733

Heteroskedasticity-consistent or clustered t statistics in
*  p<0.05 , **  p<0.01 , ***  p<0.001


2.3 Persistency

In addition to looking at the immediate impact of the security-specific demand and supply factors on SC repo rates, we also investigate their dynamic effects. Because the Fed's purchased (sold) amounts can be perceived by the market participants as a long lasting reduction (increase) in a security's available supply (conditional on their expectations about the time of the potential unwinding of the Fed balance sheet expansion), and because SC repo contracts rule out the possibility of delivering a close substitute security, we would expect these effects to be quite persistent.

To test this hypothesis, the top panel of Figure 4 shows, for the change in the SC repo rates, the cumulative response to the Fed off-the-run purchases in the  N -day period following the purchases (  N=1,\dotsc,100 ) and the corresponding 95% confidence interval.22 In the dynamic specification, in addition to the variables used in the baseline regressions (see Section 2.1), we also control for any future purchases that took place over the  N -day time period. It can be seen that the effect is quite persistent, as it converges toward zero very slowly and stays significant for at least three months (60 business days). Further, in the week following the purchase operation, on average, the estimated coefficient increases in magnitude to -0.12 (from -0.08), indicating that a $1 billion purchase would decrease the SC repo rate by 0.5 basis points, and only after about two months (40 business days) it stabilizes around the initial impact value. We repeat the same exercise for the coefficient on the amount sold at the Fed operations. As shown in the bottom panel of Figure 4, the effect is less persistent for sales, as it remains significant for about 15 business days.

Figure 4a. Fed Purchases

Figure 4a. Fed Purchases
Figure 4a Data

Figure 4b. Fed Sales

Figure 4b. Fed Sales
Figure 4b Data
Coefficients on the percentage bought or sold by the Fed from regressions using, as the dependent variable, cumulative changes in the SC repo rate over the  N -day period following each operation. Black points indicate the estimated coefficients for each period. Grey triangles indicate the 95% confidence interval for each of those coefficients. The lines are fitted LOESS curves.


Indeed, the estimated impulse response for the coefficient on the Fed's purchases confirms the existence of a significant scarcity premium for Treasury collateral that does not seem to dissipate quickly, at least in our sample. This is quite striking if we consider that our panel includes time dummies, thus this coefficient isolates the additional price impact of a change in supply on top of any common factor, measuring a lower bound of the supply effect; this bound is shown to be sizable and fairly persistent.

It certainly persists longer than the purchase effects of the Fed's first LSAP program in the Treasury cash market, which revert to zero after six days from the day of purchase (D'Amico & King, 2013). This can be due to the security-specific nature of SC repo contracts, which prevent the delivery of close substitutes. In other words, anyone who sold collateral short must deliver that specific bond and not some other bond, and therefore would put extra value on that specific collateral. The availability of similar bonds would not affect that value, at least until the position is closed.

The following is one possible mechanism behind the persistency of the supply effects. If there are a significant amount of open short positions established through reverse repos and the net supply of the underlying collateral decreases (in this particular case because of the Fed purchases), at impact the price of the Treasury collateral in the cash market would increase and the current and expected future repo rate would decrease (repo specialness spread would increase). Dealers would now have a few options: they may be forced to repurchase the bond at a significantly higher price and incur a substantial loss, which in aggregate would make the collateral's net supply decrease further; they can roll over in a new reverse repo offering cash at the lower SC repo rate to get that specific security and close the previous position; and, if the current contract is an open repo, they can roll over the same reverse repo contract (subject to changes in margin requirements) re-setting to the lower SC repo rate. All these possibilities, by either making the underlying collateral scarcer and/or by keeping the repo contract rolling, may cause SC repo rates to stay lower for longer, magnifying the persistency of the supply shock.


3 Pass-Through to Cash Market Prices

In light of the recent literature's findings that even perfectly anticipated changes in supply could have effects on Treasury cash prices (as shown by Lou et al., 2013, for Treasury auctions and D'Amico & King, 2013, for the Fed's Treasury LSAPs), and given the existence of well-documented links between a security's cash market price and repo market specialness (Jordan & Jordan, 1997; Duffie, 1996; Buraschi & Menini, 2002), it is natural to hypothesize that some of the LSAPs' price effects in the cash market might reflect changes in repo specialness spreads due to the Fed operations estimated in Section 2.2. In this section, we attempt to test this conjecture.

We begin by showing that, in our sample, a specific Treasury bond's cash price premium (relative to securities with the same coupon and maturity) indeed mostly reflects the magnitude of its repo specialness spread, and that this relation becomes stronger on the days of Fed purchase/sale operations. Since we already showed that the Fed's asset purchases are associated with higher repo specialness spreads (lower SC repo rates) and that these effects are quite persistent, the above relation to the cash price premium provides some support for our hypothesis. Namely, that one channel through which LSAPs affect Treasury prices (on the days of the actual operations) could be by impacting the scarcity value of Treasury collateral in the repo market. This can help explain why purchase/sale operations that were announced in advance, and whose total size and targeted securities were fairly predictable, might still trigger statistically significant responses in bond prices, known as pace- or flow-effects in the QE literature.

In particular, Table 10 shows results from a panel regression, motivated by the work of Jordan & Jordan (1997), in which levels of the securities' cash price premia are regressed on their repo specialness spreads. We also control for liquidity and risk differentials through the bid-ask spread, on-the-run dummy, and maturity squared. To measure each specific security's price premium in the cash market over an otherwise identical note (i.e., a note with the same coupon rate and time to maturity), we use the deviation of its observed yield from the Svensson (1994) zero-coupon yield curve.23 A higher spread implies that a security is more expensive than the curve would predict based on the security's fundamentals, and therefore is embodying a premium related to its specific characteristics, such as liquidity and repo financing advantages. As shown in the first column, running this regression in the full sample produces a positive and significant coefficient on the specialness spread.24 Further, this coefficient becomes larger if we restrict the sample to the days of the Fed operations, shown in the second column.


Table 10: Cash Market Premium; Levels

(1) All Days (2) Operations
Repo Spread 0.0487*** 0.0646***
Repo Spread, t-stat (10.99) (9.24)
Bid-Ask Spread -0.525*** -0.402***
Bid-Ask Spread, t-stat (-51.74) (-23.76)
On-the-run 1.019*** 0.683***
On-the-run, t-stat (14.46) (7.05)
Maturity-squared 0.233*** 0.248***
Maturity-squared, t-stat (82.80) (49.80)
 N 170203 92099
 R^{2} 0.283 0.258
adj.  R^{2} 0.277 0.251

t statistics in parentheses
*  p<0.05 , **  p<0.01 , ***  p<0.001


In addition, we quantify the pass-through of fluctuations in the repo scarcity value to cash market prices by regressing changes in securities' cash market premia on the changes in specialness spreads solely explained by the amounts purchased and sold by the Fed. The results of this exercise are presented in Table 11. Each column presents the estimated coefficients from a panel regression in which daily changes in cash price premia (within each of the four maturity groups) are regressed on the portion of daily changes in repo specialness spreads that our baseline regressions attribute to the Fed's sale and purchase operations (using the coefficients reported in Tables 5 and 8). Additionally, we control for changes in the security-level bid-ask spreads, include time fixed effects, and we divide the sample into off- and on-the-run securities.

We find that the pass-through is significant only for off-the-run securities, which is not surprising considering the smaller amount of observations for the on-the-run securities. In particular, a one basis point increase in the predicted repo specialness spread translates, on average, into a cash price premium of about 0.4 basis points (column 4). However, the magnitude and importance of the estimated pass-through differs across maturities. For the off-the-run securities with 3 to 7 years left to maturity, it has a similar magnitude of about 0.4 basis points and is statistically significant at a 1% level, while for the off-the-run securities with 7 to 15 years to maturity, it is positive but smaller in magnitude and is barely significant at a 5% level. The coefficient for off-the-run securities with maturities less than 3 years is positive but statistically insignificant, although results in this sector are harder to interpret because the corresponding predicted repo spreads reflect effects of both asset purchases and sales.


Table 11: Cash Market Premia: Pass-through from Fed Operations; 1-Day Changes

  (1) 0-3 Years (2) 3-7 Years (3) 7-15 Years (4) All Maturities
Predicted Repo Spread*Off-the-Run 0.106 0.386*** 0.118 0.378**
Predicted Repo Spread*Off-the-Run, t-stat (0.42) (5.55) (1.96) (3.13)
Predicted Repo Spread*On-the-Run -0.0146 0.00560 0.0589 0.0224
Predicted Repo Spread*On-the-Run, t-stat (-0.42) (0.16) (0.76) (0.74)
Observed Bid-Ask Spread -0.589*** -0.0964*** -0.0402*** -0.269***
Observed Bid-Ask Spread, t-stat (-4.90) (-36.37) (-13.76) (-6.09)
 N 45691 30187 11241 87119
 R^{2} 0.016 0.470 0.579 0.009
adj.  R^{2} 0.006 0.462 0.562 0.004

Heteroskedasticity-consistent t statistics in parentheses
*  p<0.05 , **  p<0.01 , ***  p<0.001


Overall, these findings suggest that, at least for longer-term off-the-run securities, the Fed asset purchase programs could affect Treasury security prices not only directly through the stock effect, but also indirectly by increasing the scarcity value of the Treasury collateral in the repo market, which translates into higher specialness spreads. These increases in the security's specialness are in turn reflected (and discounted) in the cash market, leading to higher price premia for relatively scarcer securities on the days of the operations.

This is part of the so-called flow-effect that D'Amico & King (2013) find to be about 3.5 basis points per $5 billion of Treasury securities purchased under the first LSAP. In Section 2.2, we show that a Fed purchase of the same size would increase the repo specialness spread by about 2.5 basis points, and here we show that this increase would cause the cash market premium to rise by about one basis point (  0.4 \times 2.5bp ), a significant portion of the 3.5 basis point effect.


4 Treasury Auction-Level Regressions

Similar to Jordan & Jordan (1997), we also run regressions at Treasury-auction frequency to examine the impact of issued amounts and demand at auctions on future specialness spreads. In particular, they show that both the auction tightness and the issue's distribution of ownership have significant effects on on-the-run specialness spreads and that these effects can last as long as four weeks after the auction.

Again, we omit the 30-year Treasury auctions because long-term treasuries are rarely traded on the repo market. This leaves us with 257 Treasury auctions (including reopenings) from May 1, 2009 to Dec 31, 2012. The data on Treasury auctions are from the Treasury Direct website.25 For each auction in our sample, we construct the average specialness spread in the  j th week (  j = 1, 2, 3, 4 ) after the security is issued and the average for the month after issuance.

To measure an unexpectedly large demand for the auctioned security, we use the spread between the 1 p.m. when-issued (WI) rate and the yield at the auction. The WI market is a forward market in which trading begins approximately two weeks before the security's auction and contracts are settled when the security is issued. As Duffie (1996) notes, short positions created in this market are often covered with securities obtained at the auction. An auction yield lower than the prevailing WI rate indicates an unexpectedly strong demand at the auction. Short-sellers unable to cover their positions at the auction will turn to the cash or repo markets, which can potentially push SC repo rates lower and specialness spreads higher. Previous studies have used the bid-to-cover ratio, which is defined as the total dollar amount of received bids divided by the total dollar amount of accepted bids. This measure does not, however, necessarily measure the unexpected auction tightness. Instead, since the 1 p.m. WI rate is a reliable measure of the rate expected to prevail at the auction, the spread between the WI rate and the auction yield better captures the unexpected component of the auction demand.

Duffie (1996) also notes that some institutional investors are unable or reluctant to lend their securities in the repo market (e.g., insurance companies, pension funds, and foreign central banks), while other investors like dealers and brokers are major participants in the repo market and routinely use it to finance their positions. Therefore, similarly to Jordan & Jordan (1997), we expect a high fraction of issuance awarded to dealers to be readily available as collateral in the repo market, though this supply may decrease over time as dealers sell their inventories to clients and no longer need repo financing.

In contrast, foreign official investors usually buy and hold their securities and are often legally prohibited from supplying them as collateral to the repo market. In recent years, many foreign governments and central banks, particularly those with elevated current accounts surpluses, have held substantial international reserves in the form of Treasuries. Foreign holdings of U.S. Treasury securities amount to about $5.5 trillion, roughly half of the total marketable Treasury debt. This suggests that a significantly smaller portion of Treasury collateral is available to more active investors in the repo market, such as dealers.

Using auction allotment data obtained from the Treasury, we control for the security's share awarded to dealers and foreign investors in our regressions.26 Securities with larger fractions awarded to dealers are expected to have lower specialness spreads as they are more readily available in the repo market, while the opposite is expected for securities with larger fractions awarded to foreign participants.

Similarly to Graveline & McBrady (2011), we also include the amount issued at each auction, as larger issues should be in greater supply in the repo market. Finally, we include dummies for auctions of each maturity in our sample (2-, 3-, 5-, 7-, and 10-year) and a dummy variable that tracks if the auction is a reopening. Summary statistics for our dependent variables are shown in Table 12.


Table 12: Summary Statistics - Treasury Auctions

  Mean Std. Dev.
foreign_pct 22 10.4
dealer_pct 53.2 9.57
amt_issued 3.0e+10 9.3e+09
whenissued_spread -4.63 1.94
 N 257  

Regression results for the average specialness spread are presented in Table 13. The first column shows the results for the first week after issuance. The coefficients on foreign allotment percentages are largely significant and have the expected signs, with higher dealer allotments resulting in lower specialness and larger foreign allotments resulting in higher specialness. Furthermore, in week four the foreign allotment's effect size quadruples, suggesting that when the demand for on-the-run securities is already high due to the auction cycle (discussed in Section 1.4), then this type of supply constraint magnifies the rise in repo specialness spreads. The effect of the amount issued is negative and significant, and persists throughout the inter-auction period, with much larger effect sizes in weeks three and four. In contrast, the coefficient on the spread between the WI rate and the auction yield is negative but not statistically significant. The coefficient on the reopening dummy shows that, on average, the specialness spread is lower after reopenings than initial auctions, particularly in the third and fourth weeks after the auction. This confirms the patterns shown in Figure 3. This result also has implications for the Treasury's management of auction cycles, as increasing the tradeable supply of highly-demanded securities through reopenings could help alleviate collateral scarcity.


Table 13: Treasury Auctions: Average Specialness Spread by Week

  (1) Week 1 (2) Week 2 (3) Week 3 (4) Week 4 (5) Total
dealer_pct -0.0949+ -0.0723 0.137 0.296 0.0840
dealer_pct, t-stat (-1.81) (-0.81) (0.56) (1.11) (0.58)
foreign_pct 0.139* 0.132 0.140 0.579* 0.252+
foreign_pct, t-stat (2.20) (1.10) (0.59) (2.09) (1.71)
amt_issued -3.85e-10** -3.48e-10+ -1.04e-09+ -1.76e-09* -8.52e-10*
amt_issued, t-stat (-2.95) (-1.65) (-1.78) (-2.51) (-2.42)
WI_auction_spread -0.255 -0.437 -2.071 -2.420 -1.302
WI_auction_spread, t-stat (-1.13) (-0.78) (-1.28) (-1.62) (-1.52)
dummy_reopening -2.393* -3.663 -16.85+ -39.87** -16.67**
dummy_reopening, t-stat (-2.19) (-1.58) (-1.74) (-3.12) (-2.79)
 N 257 257 257 257 257
 R^{2} 0.262 0.153 0.136 0.205 0.201
adj.  R^{2} 0.232 0.119 0.101 0.173 0.168

Heteroskedasticity-consistent t statistics in parentheses
+  p<.10 , *  p<.05 , **  p<.01


We largely confirm the conclusions of Jordan & Jordan (1997) regarding the effects of Treasury auction characteristics on repo specialness in a much larger sample. Our results also provide support for Duffie's conjecture that repo specialness should exhibit a non-linear behavior due to the determination of its equilibrium value (see Figures 2 and 3 in Duffie, 1996). When collateral supply exceeds its demand, the normal situation, small supply shifts should not affect specialness, which remains in a corner solution close to zero. However, during periods of excess demand, such as at the peak of the auction cycle (week four), we expect supply shifts to have large and significant impacts. Consistent with this theory, we find that the estimated effects of foreign allotments and amounts issued are much stronger during the third and fourth weeks after issuance, when securities are usually on special due to the Treasury auction cycle.


5 Conclusion

In this study, we use security-level data to estimate the impact of changes in the demand and supply of Treasury collateral on the SC repo rates of all outstanding U.S. nominal Treasury securities. We find that demand and supply effects are economically and statistically significant in the SC repo market. Specifically, we estimate that a one-billion-dollar reduction in the available supply of Treasury collateral can increase the scarcity value of this collateral by 0.3 to 1.8 basis points depending on the security's characteristics, with the larger effects concentrated in on-the-run and shorter-term securities.

Further, we find that for longer-term off-the-run securities this scarcity value is reflected in the Treasury cash market prices. And in particular, we show that the portion of the increase in the repo scarcity premium due to the Fed purchase operations passes through to cash market premia, explaining a significant amount of the flow-effects of these operations. Therefore, our results provide further support for the scarcity channel of quantitative easing.

Our findings also suggest that, through the same mechanism, the Fed's FRFA reverse repo facility--one of the tools under consideration for the policy normalization process--can help tighten control over money market rates. For example, by increasing the availability of Treasury collateral to a wide range of market participants, it could reduce the scarcity premium embodied in these rates, especially when the appetite for high-quality assets increases. Figure 5 attempts to illustrate this point.

Figure 5a. Short Rates

Figure 5a. Short Rates.
Figure 5a Data

Figure 5b. Fed Reverse Repor Facility: Total Volume

Figure 5b. Fed Reverse Repor Facility: Total Volume.
Figure 5b Data

Figure 5c. Fed Reverse Repo Facility: Number of Bidders

Figure 5c. Fed Reverse Repo Facility: Number of Bidders.
Figure 5c Data
Figure 5. The top panel shows various money market rates as well as the rate offered at the Fed's reverse repo facility. The bottom two panels show some of this facility's statistics. The shaded area denotes the U.S. government shutdown of 2013.


The top panel of Figure 5 shows two of the most relevant overnight money market rates--the federal funds rate and the GCF Treasury repo rate--together with the repo rate set by the Fed for its reverse repo operational tests, which started at the end of September 2013. This panel shows that, although the operations' amounts are still capped, the Fed's reverse repo rate has generally been providing a floor for other money market rates during quarter- and year-end periods. These are periods when demand for Treasury securities increases, likely due to risk-shifting window dressing by intermediaries, who alter portfolios at disclosure dates to underrepresent their riskiness (e.g., Griffiths & Winters, 2005; Musto, 1997). Indeed, as shown in the bottom two panels, which plot the aggregate volume and the number of participants at each Fed reverse repo operations, demand for Treasury securities and participation at this facility have spiked at the end of each quarter.

Further, it is important to keep in mind that the impact of the Fed FRFA reverse repo facility is similar to, but less direct than the impact of an increase in the amount issued by the Treasury, a Treasury reopening, or a Fed LSAP sale operation.27

This discussion is not meant to provide a definite answer regarding the efficacy of this facility as a monetary policy tool and the sample is still too small for in-depth empirical analysis. However, we do think that this topic deserves further investigation, and that the type of analysis presented in this paper is well suited to evaluate some of the tools available to the Fed when implementing monetary policy with a very large balance sheet. This is crucial for understanding the issues surrounding the process of policy normalization and we leave it to future research.


Bibliography

T. Adrian, et al. (2011).
`Repo and securities lending'.
Staff Report 529, Federal Reserve Bank of New York.
Retrieved from http://www.nyfedeconomists.org/research/staff_reports/sr529.pdf.
A. Ajello, et al. (2012).
`No-Arbitrage Restrictions and the U.S. Treasury Market'.
Economic Perspectives 36(2):55-74.
Federal Reserve Bank of Chicago.
S. Banerjee & J. J. Graveline (2013).
`The Cost of Short-Selling Liquid Securities'.
Journal of Finance 68(2):637-664.
L. Bartolini, et al. (2011).
`Collateral values by asset class: Evidence from primary securities dealers'.
Review of Financial Studies 24(1):248-278.
M. W. Brandt & K. A. Kavajecz (2004).
`Price Discovery in the U.S. Treasury Market: The Impact of Orderflow and Liquidity on the Yield Curve'.
Journal of Finance 59(6):2623-2654.
A. Buraschi & D. Menini (2002).
`Liquidity risk and specialness'.
Journal of Financial Economics 64:243-284.
M. E. Cahill, et al. (2013).
`Duration risk versus local supply channel in Treasury yields: Evidence from the Federal Reserve's asset purchase announcements'.
Working Paper 35, Finance and Economics Discussion Series, Board of Governors of the Federal Reserve System.
J. A. Cherian, et al. (2004).
`A model of the convenience yields in on-the-run Treasuries'.
Review of Derivatives Research 7:79-97.
Committee on the Global Financial System (May 2013).
`Asset encumbrance, financial reform and the demand for collateral assets'.
CGFS Papers 49.
S. D'Amico, et al. (2012).
`The Federal Reserve's Large-Scale Asset Purchase Programmes: Rationale and Effects'.
The Economic Journal 122:415-446.
S. D'Amico & T. B. King (2013).
`Flow and stock effects of large-scale Treasury purchases: Evidence on the importance of local supply'.
Journal of Financial Economics 108:425-448.
D. Duffie (1996).
`Special repo rates'.
Journal of Finance 51:493-526.
D. Duffie, et al. (2002).
`Securities lending, shorting, and pricing'.
Journal of Financial Economics 66:307-339.
D. Duffie, et al. (2007).
`Valuation in over-the-counter markets'.
Review of Financial Studies 20(6):1865-1900.
R. Eldor, et al. (2006).
`A search-based theory of the on-the-run phenomenon'.
Journal of Business 79(4):2067-2097.
E. J. Elton & T. C. Green (1998).
`Tax and liquidity effects in pricing government bonds'.
Journal of Finance 53(5):1533-1562.
M. Fisher (2002).
`Special repo rates: An introduction'.
Economic Review 87(2):27-43.
Federal Reserve Bank of Atlanta.
M. J. Fleming (2002).
`Are larger Treasury issues more liquid? Evidence from bill reopenings'.
Journal of Money, Credit and Banking 34(3):707-735.
M. J. Fleming & K. D. Garbade (2003).
`The Repurchase Agreement Refined: GCF Repo ® '.
Current Issues in Economics and Finance 9(6).
Federal Reserve Bank of New York.
M. J. Fleming & K. D. Garbade (2004).
`Repurchase agreements with negative interest rates'.
Current Issues in Economics and Finance 10(5).
Federal Reserve Bank of New York.
M. J. Fleming & K. D. Garbade (2007).
`Dealer behavior in the specials market for US Treasury securities'.
Journal of Financial Intermediation 16:204-228.
M. J. Fleming, et al. (2010).
`Repo market effects of the Term Securities Lending Facility'.
Staff Report 426, Federal Reserve Bank of New York.
Retrieved from http://www.newyorkfed.org/research/staff_reports/sr426.pdf.
M. J. Fleming, et al. (2012).
`The Effects of Treasury Fails Charge on Market Functioning'.
Working Paper.
M. J. Fleming & J. V. Rosenberg (2007).
`How do Treasury dealers manage their positions?'.
Staff Report 299, Federal Reserve Bank of New York.
Retrieved from http://www.newyorkfed.org/research/staff_reports/sr299.pdf.
J.-S. Fontaine & R. Garcia (2012).
`Bond liquidity premia'.
Review of Financial Studies 25(4):1207-1254.
D. Goldreich, et al. (2005).
`The Price of Future Liquidity: Time-Varying Liquidity in the U.S. Treasury Market'.
Review of Finance 9(1):1-32.
G. Gorton & A. Metrick (2012).
`Securitized Banking and the Run on Repo'.
Journal of Financial Economics 104:425-451.
J. J. Graveline & M. R. McBrady (2011).
`Who makes on-the-run Treasuries special?'.
Journal of Financial Intermediation 20:620-632.
R. Greenwood (2005).
`Short- and long-term demand curves for stocks: Theory and evidence on the dynamics of arbitrage'.
Journal of Financial Economics 75(3):607-649.
R. Greenwood & D. Vayanos (2013).
`Bond Supply and Excess Bond Returns'.
Review of Financial Studies Forthcoming.
M. D. Griffiths & D. B. Winters (2005).
`The Turn of the Year in Money Markets: Tests of the Riskâ€Shifting Window Dressing and Preferred Habitat Hypotheses'.
Journal of Business 78(4):1337-1364.
D. Gromb & D. Vayanos (2010).
`Limits of Arbitrage: The State of the Theory'.
Working Paper 15821, National Bureau of Economic Research.
Retreived from http://www.nber.org/papers/w15821.
D. Heller & N. Vause (June 2011).
`Expansion of Central Clearing'.
BIS Quarterly Review .
B. D. Jordan & S. D. Jordan (1997).
`Special repo rates: An empirical analysis'.
Journal of Finance 52(5):2051-2072.
A. Kaul, et al. (2000).
`Demand Curves for Stocks Do Slope down: New Evidence from an Index Weights Adjustment'.
Journal of Finance 55(2):893-912.
F. Keane (1995).
`Expected repo specialness costs and the Treasury auction cycle'.
Research Paper 9504, Federal Reserve Bank of New York.
Retrieved from http://www.newyorkfed.org/research/staff_reports/research_papers/9504.pdf.
A. Krishnamurthy (2002).
`The bond/old-bond spread'.
Journal of Financial Economics 66:463-506.
A. Krishnamurthy & A. Vissing-Jorgensen (2011).
`The Effects of Quantitative Easing on Interest Rates: Channels and Implications for Policy'.
Working Paper 17555, National Bureau of Economic Research.
Retrieved from http://www.nber.org/papers/w17555.
F. A. Longstaff (2000).
`Arbitrage and the Expectations Hypothesis'.
Journal of Finance 55(2):989-994.
J. C. Lopez, et al. (2013).
`The Market for Collateral: The Potential Impact of Financial Regulation'.
Financial System Review (2):45-53.
Bank of Canada.
D. Lou, et al. (2013).
`Anticipated and Repeated Shocks in Liquid Markets'.
Review of Financial Studies 26(8):1891-1912.
A. Martin, et al. (2013).
`Federal Reserve Tools for Managing Rates and Reserves'.
Staff Report 642, Federal Reserve Bank of New York.
Retrieved from http://www.newyorkfed.org/research/staff_reports/sr642.pdf.
M. Mazzoleni (2013).
`Treasury yields and auction cycle management'.
Job Market Paper.
P. C. Moulton (2004).
`Relative repo specialness in U.S. Treasuries'.
Journal of Fixed Income 14(1):40-47.
D. K. Musto (1997).
`Portfolio Disclosures and Year-End Price Shifts'.
Journal of Finance 52(4):1563-1588.
D. K. Musto, et al. (2011).
`Notes on bonds: Liquidity at all costs in the Great Recession'.
Working paper.
Retrieved from http://www.fma.org/Napa/Papers/Notes_on_Bonds.pdf.
D. Nautz (1997).
`How auctions reveal information: A case study on German REPO rates'.
Journal of Money, Credit and Banking 29:17-25.
S. Potter (2013).
`Recent Developments in Monetary Policy Implementation'.
Speech delivered before the Money Marketeers of New York University, New York City, December 2. Available at http://www.newyorkfed.org/newsevents/speeches/2013/pot131202.html.
A. Shleifer (1986).
`Do Demand Curves for Stocks Slope Down?'.
Journal of Finance 41(3):579-590.
S. Sundaresan (1994).
`An Empirical Analysis of U.S. Treasury Auctions: Implications for Auction and Term Structure Theories'.
Journal of Fixed Income 4(2):35-50.
S. Sundaresan & Z. Wang (2009).
`Y2K Options and the Liquidity Premium in Treasury markets'.
Review of Financial Studies 22(3):1021-1056.
L. E. Svensson (1994).
`Estimating and Interpreting Forward Interest Rates: Sweden 1992-1994'.
Working Paper 4871, National Bureau of Economic Research.
Retreived from http://www.nber.org/papers/w4871.
D. Vayanos & P.-O. Weill (2008).
`A search-based theory of the on-the-run phenomenon'.
Journal of Finance 63(3):1361-1398.
J. Wurgler & E. Zhuravskaya (2002).
`Does Arbitrage Flatten Demand Curves for Stocks?'.
Journal of Business 75(4):583-608.



Footnotes

* We are grateful to Gadi Barlevy, Francois Gourio, Frank Keane, Thomas King, Eric LeSueur, Dina Marchioni, Sean Savage, John Sporn, seminar participants, and the MOMA group at the Federal Reserve Bank of Chicago for useful discussions and comments. We also thank Dominic Anene, Long Bui, Scott Konzem, and Tanya Perkins for their help with data preparation. All errors and omissions are our sole responsibility. The views expressed in this paper are those of the authors alone and do not necessarily reflect the views of the Federal Reserve Bank of Chicago, the Board of Governors of the Federal Reserve System, the Federal Reserve System, or their staff. Return to Text
* Federal Reserve Bank of Chicago. Email: sdamico@frbchi.org. Phone: +1 (312) 322-5873 Return to Text
* Federal Reserve Bank of Chicago. Email: rfan@frbchi.org. Phone: +1 (312) 322-4019 Return to Text
* Federal Reserve Board. Email: yuriy.kitsul@frb.gov. Phone: +1 (202) 452-2967 Return to Text
1. See Lou et al. (2013) for price responses around Treasury auctions, and D'Amico & King (2013) for price reactions to the Federal Reserve's Treasury purchase operations. Both studies indicate that these supply effects reverse after few days. Return to Text
2. Except for Jordan & Jordan (1997), which uses Treasury auction results on 39 distinct notes from September 1991 to December 1992, most other studies focus on the specialness spreads of a few on-the-run Treasury securities and use mainly aggregate demand variables (e.g., interest-rate-risk hedging demand, buy-and-hold investors' demand, and mortgage-convexity hedging demand); see Moulton (2004) and Graveline & McBrady (2011). Return to Text
3. From March 2009 to December 2012, the Fed conducted two Large-Scale Asset Purchase programs by removing $900 billion of Treasury securities from the market, and two Maturity Extension Programs by purchasing a total of $667 billion of Treasury securities with maturity between 6 and 30 years and selling an equal amount of securities with remaining maturity of 3 years or less. Return to Text
4. This result is consistent with Mazzoleni (2013) estimates of a reduction of 0.55 basis points in the yield premium of a two- or a five-year note for an additional issuance of one billion dollars. Return to Text
5. Other important studies that examine the relationship between price differentials in the Treasury cash market and funding conditions in the repo market in various contexts include Krishnamurthy (2002), Goldreich et al. (2005), Musto et al. (2011), Fontaine & Garcia (2012), and Banerjee & Graveline (2013). Return to Text
6. The September 2013 FOMC meeting authorized the New York Fed to start operational tests of fixed-rate overnight reverse repos. The FRFA reverse repo facility allows a wide range of market participants to deposit unlimited amounts of cash at a fixed rate in exchange for Treasury securities held in the SOMA portfolio. See http://www.newyorkfed.org/markets/rrp_faq.html for more information on these operations. Return to Text
7. See Potter (2013) for a more detailed discussion on the overnight FRFA reverse repo facility and its objectives. Return to Text
8. For more details, see the May 2013 report of the Committee on the Global Financial System for discussions on various factors that could potentially affect availability of collateral assets. Return to Text
9. For more details on the special collateral repo market see Fisher (2002). Return to Text
10. For example, as of November 14, 2013, the total amount of U.S. Treasury overnight repos and reverse repos entered into by primary dealers was about $1.6 trillion (FR-2004 data); for comparison, the average daily traded volume in the Treasury cash market over the week ending on November 6 was about $500 billion. Return to Text
11. DTCC GCF rate data are publicly available at http://www.dtcc.com/charts/dtcc-gcf-repo-index.aspx. Return to Text
12. For more detail about the differences between GC repo rate and the GCF Repo see Fleming & Garbade (2003). Return to Text
13. For more details on these programs, see Cahill et al. (2013). Return to Text
14. SOMA operation and holding data by CUSIP are publicly available on the New York Fed's website: http://www.ny.frb.org/markets/pomo/display/index.cfm. Return to Text
15. "Privately held" Treasury securities are defined here as any security not held by the Federal Reserve and is calculated by subtracting the par value held in the SOMA portfolio from the total outstanding par value, which are obtained from CRSP. Source: CRSP®, Center for Research in Security Prices, Booth School of Business, The University of Chicago. Used with permission. All rights reserved. http://www.crsp.uchicago.edu. Return to Text
16. See Fleming & Garbade (2007) for more details on the SLP. Data are publically available at the New York Fed's website: http://www.newyorkfed.org/markets/securitieslending.html. Return to Text
17. Composites of bid and ask price quotations for individual Treasury securities are obtained by the New York Fed. Return to Text
18. See http://www.newyorkfed.org/tmpg/tmpg_faq_033109.pdf for details of the fails charge implementation. Fleming et al. (2012) show that this triggered striking changes in the willingness to receive negative interest rates on cash pledged to secure borrowing of certain securities. Return to Text
19. We obtain very similar results if we use every day in the sample. Return to Text
20. For brevity, we do not show the coefficients for the time and auction cycle dummies. Return to Text
21. In our regressions, we discard observations for which the 1-day change in the SC repo rate exceeds 40 basis points or the 2-day change exceeds 60 basis points. These threshold choices seems reasonable, since in our full sample over 99.9% of observations are within each threshold. Return to Text
22. The small sample size for the on-the-run securities limits our ability to test for dynamic effects. Return to Text
23. The yield curve is estimated excluding on-the-run and first off-the-run Treasury securities. The deviation is computed as the predicted minus actual yield and is maintained by the staff of the Board of Governors of the Federal Reserve System. Return to Text
24. In our regressions, we include security and time fixed-effects and discard observations for which the cash price premium exceeds 50 basis points in absolute value. This threshold choice seems reasonable, since in our full sample the 1% and 99% percentiles of price premium measures are about -16 bps and 22 bps, respectively, while their 0.1% and 99.99% percentiles are -116 and 44 basis points, respectively. Return to Text
25. See http://www.treasurydirect.gov/RI/OFGateway. When-issued rate data are obtained from Tradeweb. Return to Text
26. Investor class auction allotment data are available at http://www.treasury.gov/resource-center/data-chart-center/Pages/investor_class_auction.aspx. "Dealers and brokers" includes primary dealers, dealer departments at commercial banks, and other non-bank dealers and brokers. "Foreign investors" includes private foreign investors, foreign central banks, and other non-private foreign entities. Return to Text
27. Under a reverse repo with the Fed, the securities sold by the Fed to the counterparty may be held on the counterparty's balance sheet but are in the tri-party system, making them unavailable for the counterparty to satisfy margin requirements (Potter, 2013). Return to Text

This version is optimized for use by screen readers. Descriptions for all mathematical expressions are provided in LaTex format. A printable pdf version is available. Return to Text