How do Treasury dealers manage their positions? *

Transcription

1 How do Treasury dealers manage their positions? * Michael J. Fleming and Joshua V. Rosenberg Federal Reserve Bank of New York October 31, 2006 Abstract We assess U.S. Treasury dealer positions from 1990 to 2006 and find that dealers absorb a large share of Treasury issuance inventory shocks. While dealers regularly hedge spot position changes with futures position changes, the degree of hedging is much smaller for spot position changes attributable to these large, predictable non-information-based inventory shocks than it is for other position changes. Dealers seem to be compensated for the inventory risk associated with their intertemporal intermediation of Treasury supply shocks in that accumulation of positions attributable to such shocks is associated with price appreciation in the subsequent week. Keywords: Treasury market, dealer, positions, hedging, issuance JEL Codes: G12, G20, G24 * PRELIMINARY COMMENTS WELCOME. We thank April Bang, Neel Krishnan, Michal Lementowski, Katherine Lewis, and Samuel Maurer for excellent research assistance. We are grateful to Bruno Biais for helpful comments. Views expressed are those of the authors and not necessarily those of the Federal Reserve Bank of New York or the Federal Reserve System. Both authors can be contacted at the Federal Reserve Bank of New York, 33 Liberty St., New York, NY (michael.fleming@ny.frb.org, joshua.rosenberg@ny.frb.org).

2 1. Introduction Theoretical models of dealer activity such as Kyle (1985) and Glosten and Milgrom (1985) assess the dealer s market-making function. In particular, these papers examine how dealers adjust their quotes to manage inventory risk in response to informed and uninformed customer trades. Models that incorporate interdealer trading include Ho and Stoll (1983) and Lyons (1996). Not surprisingly, the institutional setting in the government bond market is more complex than the market structure in these stylized models. Therefore, the actual position management behavior of dealers is of considerable interest. Cohen and Shin (2002), Fleming (2003), and Brandt and Kavajecz (2004) show that dealer order flow is informative for yields, which is consistent with dealers implementing speculative strategies based on private information. Vitale (1998) finds that U.K. gilt dealer trades are informative for yields, but customer trades are not. Massa and Simonov (2003) identify dealers in the Italian government bond market whose trades move market prices. There is also evidence of dealer hedging activity. Naik and Yadav (2003b) show that U.K. government bond dealers selectively take directional bets and hedge their spot exposure using futures positions rather than by adjusting their spot position. de Vassal (1998) finds an increase in Treasury note yields before monthly auctions, which he attributes to dealer hedging of anticipated new issuance in the when-issued market. In contrast, Vitale (1998) finds that dealer inventories do not affect gilt prices, which is consistent with an absence of hedging. 1

3 Our paper is the first to analyze the determinants of dealer positions in U.S. Treasury securities. We examine the response of dealer positions to several types of common inventory shocks (e.g., new issuance and maturing issues) to assess inventory management strategies. This analysis cannot be done using interdealer flows, because we are interested in whether these inventory shocks are absorbed by dealers or are transferred to customers. There is limited data on Treasury dealer positions and how these positions change over time. Our analysis is based on the weekly net positions of the primary government securities dealers compiled by the Federal Reserve Bank of New York and included in the FR 2004 reports for the July 1990 to June 2006 period. To our knowledge, no other paper has analyzed this positions data. 1 We find that dealers absorb a large share of Treasury issuance inventory shocks so that their positions increase substantively, even on a weekly basis, around auctions. Moreover, we find that dealers selectively hedge spot position changes with futures. In particular, dealers hedge a much smaller share of spot position changes attributable to these large, predictable Treasury issuance inventory shocks than they do of other position changes. Together, these results are consistent with an important role for dealers in intertemporal intermediation. The additional inventory risk taken on by dealers appears to be compensated, on average, by appreciation of these excess positions in the subsequent week. The paper is organized as follows. Section 2 discusses the role of Treasury dealers and the institutional features of the market. Section 3 discusses the data used in our empirical analysis. Section 4 presents our findings on the determinants of dealer positions and assesses the extent to which futures are used to hedge spot positions. Section 5 examines 1 Adrian and Fleming (2005) analyze changes in dealer leverage and the impact on dealer risk taking using FR 2004 financing data. FR 2004 trading volume and settlement fails data are analyzed in other studies. 2

4 the relationship between position changes and contemporaneous and future yield changes. Section 6 concludes. 2. The role of Treasury dealers U.S. Treasury securities trade in a multiple-dealer over-the-counter market. The predominant marketmakers are the primary government securities dealers those dealers with whom the Federal Reserve Bank of New York interacts directly in the course of its open market operations. The dealers buy securities at auction, make markets for their customers, and buy and sell securities for hedging and speculative purposes. For additional details on the structure of the Treasury market, see Dupont and Sack (1999) and Fleming and Remolona (1999); dealer markets in general are discussed in Madhavan (2000). A. Participating in auctions There are certain features of the Treasury market that make it a good laboratory for understanding dealer behavior. In the Treasury market, the primary dealers have a special obligation to participate meaningfully in auctions of U.S. Treasury securities. 2 This underwriting function is somewhat analogous to the firm commitment underwriting of an IPO, e.g. Ritter (1987). Dealers can consolidate advance customer orders and act as a broker for customer orders at auction. However, Treasury dealers are also expected to place 2 Primary dealers are also expected to make reasonably good markets in their trading relationships with the Fed's trading desk and to provide the trading desk with market information and analysis that may be useful to the Federal Reserve in the formulation and implementation of monetary policy. Dealer responsibilities are listed in the Federal Reserve Bank of New York s January 22, 1992 memorandum, Administration of Relationships with Primary Dealers, posted at 3

5 competitive bids for their own account, and they are encouraged to ladder their bids to ensure that the entire issue is sold at a reasonable price. 3 Because of this underwriting role, dealer inventories are subject to significant shocks due to the auction cycle. Primary dealers regularly buy large shares of new issues at auction for their own accounts. Between May 2003 and December 2005, for example, primary dealers purchased an average 71% share of Treasury issues sold to the public (Fleming (2006)). However, dealers do not face the same level of adverse selection when taking on inventory from Treasury auctions as they do in market making. Dealers can hedge their auction purchases by selling Treasuries prior to issuance in the when-issued market, selling after issuance in the secondary market, or taking offsetting positions in derivatives markets. As discussed below, previous findings on the extent of dealer hedging of inventory risk are mixed, but previous studies focus on inventory risk due to market making, whereas we focus on risk due to common inventory shocks such as those from auctions. B. Making markets Another role of dealers is to make secondary markets, meeting the transaction needs of customers and other dealers by buying and selling securities for their own account. Until 1992, the primary dealers were required to maintain a 1% share of total customer activity reported by all primary dealers. Since then, making markets to customers has not been a criterion for being a primary dealer, but the dealers are nonetheless still the predominant market makers. In the third quarter of 2006, for example, primary dealers reported average 3 See U.S. Primary Dealers Say Fed Urges More Active Tsy Bids at Mtg, Market News International, June 15,

6 daily Treasury trading volume of $291 billion with customers and $220 billion with other dealers. 4 In their role as market makers, dealers acquire positions that deviate from desired levels, thus creating inventory risk. To mitigate this inventory risk, dealers can implement hedging strategies by offsetting their exposures with positions in other securities or by altering their quotes ( quote shading ) to reduce inventory exposure, e.g. Ho and Stoll (1981). For example, a dealer buying a less liquid off-the-run Treasury security from a customer, might immediately hedge its exposure by buying a more liquid on-the-run security of similar maturity. Alternatively, the dealer could reduce its offer price for this security to increase the likelihood that another customer will buy it. Dealers generally make markets in several types of debt instruments resulting in a more complicated inventory risk management problem. Perhaps due to the decentralized decision-making of individual traders within a dealing firm, dealers could manage inventory risk over asset classes separately, e.g. Naik and Yadav (2003a). However, portfolio theory suggests that risk management should occur at the portfolio level as discussed in Ho and Stoll (1983). C. Taking speculative positions In the course of buying and selling securities, dealers obtain exposure to interest rate risk. Dealers can choose to hedge, partially hedge, or leave unhedged this exposure depending on their perceived informational advantage over other market participants. For 4 From Federal Reserve Bank of New York s October 12, 2006 release on the Market Share of Primary Dealer Transaction Volume, posted at ftp://ftp.ny.frb.org/mshare/2006/oct/ms The customer figure is based on reported trading that did not occur through an interdealer broker and the dealer figure is based on trading that did occur through an interdealer broker. The latter figure includes significant double-counting as trades between primary dealers get reported by both parties. 5

7 example, a dealer that expects interest rates to fall in the near future might accumulate a long Treasury securities position. If interest rates do indeed fall, the dealer will be able to sell the securities at a higher price. Given that there is no asymmetric information about Treasury security cash flows, the ability of market participants to forecast future price changes is probably limited. Nonetheless, it is possible that some market participants are better able to forecast future yield changes because they have better information about discount rates. Such information might emanate from fundamentals, such as a superior ability to evaluate the state of the economy, or from technical considerations, such as knowledge of customer order flow and/or security ownership. As mentioned earlier, several studies have shown that dealer order flow is indeed informative. 3. Data Our data on dealer positions come from the Federal Reserve Bank of New York FR 2004A release. The Fed collects positions data from the primary dealers covering U.S. Treasury securities, agency debt securities, mortgage-backed securities (MBS) and corporate debt securities. The data are reported to the Fed on a weekly basis, as of the close of business each Wednesday. Summary data are then released by the Fed with a one-week lag. 5 The data are netted and aggregated across all dealers and are only available for broad categories of securities. We analyze dealer Treasury positions over the weeks ending July 4, 1990 to June 28, Spot positions data are available for the full sample, whereas futures and options data 5 The positions data were released with a four-week lag until January 15,

8 are only available until June 27, Spot and futures positions are reported in terms of market value, whereas options positions are reported in terms of the delta-weighted value of the security underlying the option. In our analyses, we generally combine the options and futures positions and refer to the combined positions as futures positions for brevity. 7 Net dealer positions in Treasuries over our sample are shown in Figures 1 and 2. Figure 1 shows that spot bill positions tend to be positive over our sample period, whereas coupon positions tend to be negative and have dropped significantly since Figure 2 includes futures and combined spot and futures positions over the shorter futures sample. In both panels of Figure 2, there appears to be a negative correlation between spot and futures positions. We also see that bill and coupon futures positions are typically negative (i.e., dealers usually hold short positions in these contracts). In Table 1, we report summary statistics for the positions data. Coupon position levels fluctuate more than bill positions, with a standard deviation in levels for coupons of compared to for bills. However, including futures positions results in a somewhat smaller difference; the standard deviation of the combined position in levels for 6 Spot positions include securities transacted for immediate or forward delivery. For part of our sample, futures and options positions were reported as separate categories, and for another part of our sample futures positions were reported together with longer-term forward positions. For Treasuries and agencies, we exclude longer-term forward positions when reported together with futures positions. Longer-term forwards are a minimal part of the Treasury or agency debt markets, so our processing results in fairly consistent Treasury and agency series. For MBS, we include longer-term forward positions when reported together with futures positions. Longer-term forwards are important in the MBS market, but there are no MBS futures, so our processing results in a consistent MBS series. 7 The one exception is in Table 6, Panel C, where we look at dealer futures positions exclusively for consistency with other futures data with which we are comparing dealer positions. 8 Bills are reported as a distinct category for our entire sample. Coupons are reported in several buckets, but bucket definitions change (twice) during our sample. The current buckets are: due in 3 years or less, due in more than 3 but less than 6 years, due in more than 6 but less than 11 years, and due in more than 11 years. Inflation-protected securities are reported in their own bucket and are excluded from the paper (the data we retain reflects positions in inflation- protected securities from their introduction in January 1997 until the February 1998 reporting change separated out these positions). 7

9 coupons is reduced to while the bill standard deviation falls to However, bill positions exhibit higher volatility than coupon positions in terms of week-to-week changes. Bill positions are more strongly mean-reverting than coupon positions; one week spot autocorrelations are 0.77 and 0.98 respectively. Of more relevance in evaluating inventory control strategies is the mean reversion of the combined (hedged) position. In this case, we see that bill and coupon position autocorrelations drop to 0.74 and We can reject a random walk for hedged bill positions with an augmented Dickey-Fuller p-value less than 0.01, but the p-value for hedged coupon positions just barely fails to reject at In terms of inventory control, these findings suggest that dealers appear to target a fixed position. Our subsequent analysis of the effect of specific inventory shocks will provide a more refined test of dealer position management. While not the focus of our paper, we use data on dealer spot agency debt and MBS positions to better understand management of their Treasury positions. Corporate debt positions are not used because such data have only been collected by the Fed since July 4, Summary statistics on dealer agency debt and MBS positions, as well as other explanatory variables, are reported in Table 2. We also use data on Treasury security issuance. We collect data on the auction date, issuance size (in terms of par value), and time to maturity of every marketable Treasury security sold during our sample period. The issuance data are shown by week in Figure 3. The increase in bill issuance in the second half of 2001 can be attributed to the introduction of the 4-week bill series in July of that year, as well as Treasury s increased borrowing needs after September 11. 8

10 We also collect data on the redemption date, size, and time to maturity (at auction) of every Treasury security that matured or was called during our sample, as well as every security that was bought back in a debt buyback operation bonds with a total par value of $53 billion were called during our sample, all of which were originally issued between 1973 and 1981 (Treasury does not currently issue callable securities). Bonds with a market value of $87 billion were bought back in 45 operations between March 9, 2000 and April 25, The redemptions are plotted by week in Figure 4. Securities called and bought back are dwarfed by the maturing amounts of securities. Holdings of Treasury securities by the Federal Reserve and foreign central banks are also used in our analysis. These data are reported by the Federal Reserve on a weekly basis as of the close of business each Wednesday and thus match the timing of the dealer data. Note that while the Federal Reserve data are comprehensive, the foreign central bank data only include holdings held in custody at the Fed and not holdings held through other financial institutions. We also employ data on U.S. corporate bond issuance from Thomson Financial s SDC database. We exclude convertible issues, private placements, issues denominated in foreign currencies, and issues of financial institutions. The corporate issuance data are aggregated by week to correspond with our other data. Lastly, we use data on Treasury yields. We obtain constant maturity Treasury (CMT) rates from the U.S. Treasury Department for maturities of 3 month, 6 months, 2 years, 5 years, and 10 years. 9 Our sample also contains five flower bonds, which could be effectively redeemed before maturity by being used in lieu of cash to pay estate taxes. Maturing amounts of flower bonds are based on amounts outstanding at the end of the month preceding maturity (the last of which was in 1998). 10 The $87 billion figure excludes inflation-protected securities, which were bought back in one of the operations. As noted earlier, inflation-protected securities are excluded from the paper. 9

11 4. Explaining changes in dealer positions A. Determinants of dealer positions We examine the determinants of Treasury dealer positions by regressing weekly changes in dealer positions on our explanatory variables, including issuance, redemptions, changes in Federal Reserve holdings, changes in foreign central bank holdings, changes in dealer agency debt positions, changes in dealer MBS positions, corporate issuance, and the previous week s yield change. We assess spot positions (Panel A) as well as futures positions (Panel B) and consider bill and coupon positions both separately and together. 11 The results of our analysis are reported in Table 3. For spot positions, in Panel A, the most significant coefficient in all three models is issuance. The issuance coefficient can be interpreted as the change in dealer positions relative to auction amount. Dealer bill positions thus increase by an average of $302 million per $1 billion of bill issuance in a week, whereas dealer coupon positions increase by an average of $219 million per $1 billion of coupon issuance in a week. 12 The issuance coefficients indicate that dealers do not immediately sell, or hedge with other spot sales, large quantities of the Treasury securities they buy at auction. Fleming (2006) reports that dealers buy about 74% of bill issues at auction and 60% of coupon issues. A comparison of these figures with our issuance coefficients suggests that dealers sell, or 11 In the regressions of bill and coupon positions, the issuance, redemptions, and Federal Reserve holdings variables are defined accordingly (i.e., bills and coupons separately). Similarly, the representative Treasury yield used in the regressions differs, with a 6-month yield used for bills, a 5-year yield used for coupons, and a 2-year yield used for all positions. In contrast, foreign central bank holdings, agency debt positions, MBS positions, and corporate issuance are only defined in aggregate and thus do not vary across regressions. 12 Our issuance sizes include amounts purchased by the Federal Reserve (i.e., not offered to the public). For the coupon securities, we collected additional data excluding amounts offered to the Fed, re-estimated the regressions, and find that the issuance coefficient increases from to

12 hedge with other spot sales, roughly 60% of the Treasury securities they buy at auction in the same week. Surprisingly, the evidence in Table 3, Panel B for futures positions implies that dealers use futures only minimally to hedge this Treasury issuance inventory shock. For bill positions, the issuance coefficient is close to zero and statistically insignificant. For coupon positions, the coefficient is negative and statistically significant, but relatively small. On average, dealers reduce their coupon futures positions by $31 million for every $1 billion of coupon issuance in a week. Dealers thus seem to absorb, rather than hedge, most of the spot position changes attributable to Treasury issuance inventory shocks. Redemptions is another important determinant of spot position changes (no coefficients besides issuance are more than marginally significant for explaining futures position changes). Again, the coefficients can be interpreted as the change in dealer positions relative to redemption amount, so that dealer bill positions decline by an average of $282 million per $1 billion of bill redemptions in a week and dealer coupon positions decline by $47 million per $1 billion of coupon redemptions in a week. The much larger magnitude of the coefficient for bills compared to coupons implies that dealers hold a larger fraction of bills outstanding at maturity. Bills are often outstanding for only a few weeks, so the findings are suggestive of dealers holding larger amounts of these securities from auction through maturity. Changes in Federal Reserve holdings also help explain dealer position changes in coupon securities. The relationship is positive, consistent with dealers increasing positions when the Fed does, but the reason for this relationship is unclear. Changes in foreign central bank holdings are not significantly related to dealer position changes. 11

13 The only other coefficient that is significant in all three models is MBS positions. Hedging of other fixed-income securities with Treasuries (e.g., long MBS and short Treasuries) would lead to a negative relationship between these position changes and changes in Treasury positions. Alternatively, a change in expectations might cause dealers to increase or decrease positions across markets, leading to a positive relationship among position changes. We find that the MBS coefficients are all negative, consistent with a hedging explanation. For example, dealer coupon positions tend to decrease by $167 million per $1 billion increase in MBS positions. The fact that the coefficient is of similar size for bills is somewhat puzzling, however, because it seems unlikely that much MBS hedging is done with bills due to the maturity mismatch. Coefficients on changes in agency debt positions are insignificant or only marginally significant. The corporate issuance coefficient is significant for bills but not coupons. As noted earlier, we do not use dealer corporate debt positions in our analysis because this data has only been collected since Instead, we test whether corporate debt issuance helps explain dealer Treasury positions. The idea is that dealer underwriting of corporate issuance could lead to selling of Treasuries for hedging purposes. The negative coefficients are therefore expected, but it is puzzling that the coefficient is statistically significant only for bills, because one would expect hedging of corporate bonds to occur with Treasury coupon securities, which are more closely matched in terms of maturity. Lastly, coupon positions are negatively correlated with the previous week s yield change. So, during periods of rising yields, coupon positions tend to decline. Dealers, 12

14 therefore, appear to adjust their targeted level of coupon holdings according to market conditions. In general, we find that the underwriting role of dealers is a key determinant of dealer positions. Largely due to the Treasury issuance and redemption variables, we can explain 49% of the variation in bill spot position changes and 27% of the variation in coupon spot position changes. In contrast, futures positions react little to dealer underwriting or any of our other explanatory variables. We are thus able to explain only 1.6% of the variation in coupon futures position changes, and none of the variation in bill futures position changes. B. Issuance and redemption effects by maturity In Table 4, we re-estimate the relation between spot position changes and our explanatory variables, focusing on variation across issue. Table 3 does not distinguish among particular bill and coupon issues, but the sectoral differences suggest that issuance and redemption effects might also vary across this dimension. Our results shown in Table 4 therefore refine the models of Table 3 by replacing the general bill and coupon issuance and redemption variables with issue-specific issuance and redemption variables. Coefficients for the other variables are qualitatively the same as in Table 3 and are omitted from the table to save space. 13 Consistent with the Table 3 results, issuance and redemptions have a larger effect on bill positions than coupon positions. Moreover, bill coefficients tend to decrease with maturity, while the coupon coefficients are more stable across maturities. For cashmanagement bills, which tend to have the shortest maturities, the issuance coefficient is Also omitted to save space are coefficients for securities not issued but redeemed during our sample (15-, 20-, 25-, and 40-year bonds), and coefficients for securities only issued a small number of times during our sample (4- and 7-year notes). 13

15 and the redemption coefficient is For 52-week bills, the longest term bills, the issuance coefficient is 0.26 and the redemption coefficient is Coefficients for 13-week and 26-week bills are imprecisely estimated, because weekly issue sizes of these bills are highly persistent. Shorter-term bills and especially cash-management bills might have a greater effect on dealer positions for various reasons. First, for the irregularly scheduled cash-management bills, dealers may not be able to effectively reduce inventory by selling to customers. Customers may avoid these securities because of their irregular issuance and maturity dates and short maturities. Second, dealers may be willing to hold large positions in cashmanagement bills because they are of such short maturity and hence have little interest rate risk. This is consistent with an inventory-management strategy focused on duration risk, which is also found by Naik and Yadav (2003b) for U.K. dealers. C. Issuance and redemption effects surrounding auctions and redemptions We also explore the effects of issuance and redemptions more closely by examining how these effects vary over time. The results in Table 3 suggest that issuance and redemptions have important effects in those weeks. One might also expect to see significant (and opposite) effects in adjacent weeks, especially for issuance. That is, given that dealer positions increase during auction weeks, one might expect to see positions decline in subsequent weeks as dealers sell off acquired positions to customers. Moreover, dealers might sell securities to customers in advance of auction in the when-issued market on the expectation of positions that will be acquired in the auction. We thus refine the models of Table 3 by adding issuance and redemption variables for the weeks immediately preceding and following auction weeks. As shown by the coefficients 14

16 reported in Table 5, spot positions do indeed tend to decrease in the weeks before and after auction. However, the effects are of small magnitude for both types of securities and only significant for bills for the week preceding auction. Several of the redemption coefficients for the adjacent weeks are significant and negative, but it is not obvious how to interpret them. Overall, the results show that dealers absorb large, predictable inventory shocks and that the effects of these shocks persist for more than one week. D. Selective hedging using futures We next explore the extent of dealer hedging using futures by regressing changes in futures positions on changes in spot positions, similar to Naik and Yadav (2003b). Futures and spot positions tend to move in opposite directions, as shown in Table 6, Panel A. The coefficient on bill spot positions is small and marginally significant and suggests that bill futures positions decrease (increase) by $14 million for every $1 billion increase (decrease) in bill spot positions. In contrast, the coefficient for coupon positions is both sizable in magnitude and highly significant and suggests that coupon futures positions decrease (increase) by $234 million for every $1 billion increase (decrease) in coupon spot positions. In both cases, the hedge ratios are significantly less than 1, indicating that dealers retain a significant amount of risk, despite their hedging activities. We are also interested in whether dealers selectively hedge their spot positions by adjusting the amount they hedge depending on the type of shock. To test for selective hedging, we regress changes in futures positions on fitted and residual values of spot position changes. The fitted and residual values are those generated from the regressions reported on in Table 3, Panel A, so that fitted position changes are inventory shocks that can be explained 15

17 by issuance, redemptions, and other variables and residual changes are those position changes that cannot be explained by these variables. The results, shown in Table 6, Panel B, suggest that futures are indeed selectively used to hedge spot position changes. For bills, the coefficient on fitted position changes is close to zero (0.002) and statistically insignificant, whereas the coefficient on residual position changes is negative (-0.036) and statistically significant. For coupons, both coefficients are negative and statistically significant, but the coefficient on residual position changes (-0.280) is much larger in magnitude than the one on fitted changes (-0.088). For bills, the coefficients on fitted and residual position changes are significantly different from one another at the 5% level and for coupons the coefficients are significantly different at the 1% level. Dealers thus seem to hedge inventory shocks attributable to issuance and other variables to a much smaller degree than other position changes. This might reflect the generally non-information-based nature of inventory taken in after an auction compared to potentially information-based inventory resulting from market making. Naik and Yadav (2003b) find that the extent of dealer hedging depends on multiple factors that change over time such as the cost of hedging, capital requirements, and perceived informational asymmetry. While it may seem more natural to assume that dealers hedge spot positions with futures and this is the interpretation offered by Naik and Yadav (2003b) it is possible that dealers hedge futures positions with spot positions. Evidence offered in Table 6, Panel C, however, is consistent with the former interpretation. The panel shows that primary dealer futures position changes are highly correlated with the futures position changes of 16

18 commercial traders as reported by the Commodity Futures Trading Commission in the weekly Commitments of Traders Report. 14 Commercial traders are generally classified as those who use futures or options markets for hedging. 5. Compensation for intertemporal intermediation Our evidence suggests that dealers absorb large inventory shocks, especially around Treasury auctions. In light of this, it is useful to see whether dealers are compensated for the risk associated with these inventory changes by relating changes in positions to changes in yields. If dealers intertemporally intermediate supply shocks, one would expect that they would accumulate excess inventory at auction at a price below the long-run equilibrium market value and then resell their inventory in the secondary market in subsequent weeks at a higher price. This would lead to a positive relationship between position changes and contemporaneous yield changes and a negative relationship between position changes and future yield changes. In fact, press reports and some papers suggest that Treasury securities tend to cheapen going into an auction. 15 Intertemporal intermediation is not the only possible explanation for a relationship between position changes and yield changes. We would expect to see a negative relationship between position changes and future yield changes if dealers successfully speculate using private information, e.g., Naik and Yadav (2003a) and Massa and Simonov (2003). Since 14 We match the weekly Commitment of Traders positions data (reported as of the close each Tuesday) to the weekly FR 2004 positions data (reported as of the close each Wednesday). 15 Examples of press reports include, Treasurys End Modestly Lower On Refunding Pressures, Ramez Mikdashi, Dow Jones Newswires, November 8, 2004 and U.S. Treasuries Firm to Day's Highs as Stocks Sag, Ros Krasny, Reuters News, January 11, de Vassal (1998) finds mixed evidence on the behavior of yields around auctions. 17

19 data on dealer positions are only released to the public with a lag, any value positions data have at explaining future price changes would not necessarily be arbitraged away. It is useful to distinguish our analysis from the underpricing literature. There is evidence that participation in the primary market is profitable in that issues tend to be sold at auction at prices lower than those in the secondary market (see Cammack (1991), Spindt and Stolz (1992), and Simon (1994)). This literature focuses on the decision to buy at auction versus in the secondary market at the same time, and thus compares primary market prices to secondary market prices at auction time. In contrast, we are interested in dealers tendency to retain positions acquired at auction for an extended period and thus compare secondary market prices around auction time to secondary market prices in the future. Because we only examine secondary market prices, any apparent compensation to dealers for their intertemporal intermediation is distinct from any compensation to dealers for buying securities at auction versus in the secondary market. 16 In Panel A of Table 7, we regress weekly yield changes on contemporaneous and lagged values of weekly spot position changes. Bill yield changes are regressed on changes in bill positions and coupon yield changes on changes in coupon positions. The contemporaneous relationship between yields and positions is statistically insignificant for bills but negative and statistically significant for coupons. The 5-year yield coefficient, for example, indicates that a $1 billion increase in coupon positions tends to be accompanied by a 0.48 basis point decline in yield. This is consistent with the idea that dealer order flow is informative and an important determinant of dealer position changes. 16 In the absence of transaction costs, the intertemporal intermediation we observe need not involve dealers that acquired large positions at auction, but could instead involve dealers that acquired large positions in the secondary market around the time of auction. Because of transaction costs, it probably tends to be the same dealers who put capital at risk by bidding at auction and by holding positions acquired around auction time for an extended period. 18

20 When we examine the relation between the previous week s spot position changes and current yields, we find evidence that holding onto some excess inventory seems to be profitable, since a change in positions is generally followed by an appreciation in value. The coefficients on the previous week s position change are negative and statistically significant for all five yield changes, with the coefficient for the 5-year yield, for example, indicating that a $1 billion increase in coupon positions tends to be accompanied by a 0.19 basis point decline in yield the following week. Not surprisingly, the explanatory power of these models is relatively small (from 0.1% for the 6-month yield to 5.9% for the 5-year yield), suggesting that dealer spot position changes explain only a small share of the variation in yield changes. Our results comparing yield changes to contemporaneous and lagged futures position changes are presented in Panel B of Table 7. They show that futures position changes are, for the most part, positively correlated with yield changes. One interpretation of this positive relationship is that dealers trading futures are acting as liquidity providers, which means they are going long (short) Treasury futures as spot prices fall (rise). It is also not surprising that futures and spot positions have opposite relations to yields since futures and spot positions are negatively correlated with each other. In Panel C of Table 7 we present results comparing yield changes to changes in spot and futures positions combined. The results tend to be similar to those in Panel A for spot position changes alone, although the coefficients on contemporaneous position changes are smaller in magnitude and the explanatory power lower for the coupon securities. To further explore the relationship between position changes and yield changes, we separate our position change variables into fitted and residual components according to our Table 3 models. The fitted position changes are the position changes that can be explained 19

21 by the independent variables in Table 3 (notably issuance and redemptions) and the residual position changes are the position changes that cannot be explained by the independent variables in Table 3 (for example, dealer speculative activity or inventory shocks due to market-making activity). In Table 8, Panel A we see that there is a positive relationship between fitted spot position changes and contemporaneous yields, although it is only statistically significant for 6-month and 2-year yields. For the 2-year yield, for example, the coefficient suggests that a $1 billion increase in dealer positions attributable to the explanatory variables is associated with a 0.28 basis point increase in yields. In Panel B, the results based on combined spot and futures positions in Panel B are generally stronger. The positive relationship is consistent with dealers accumulating positions during the week of an auction, with the excess supply also resulting in a decline in prices and an increase in yields. Looking at what happens in the subsequent week, we see that the excess inventory from the fitted component appreciates. Coefficients are negative and statistically significant for all securities. For example, 2-year yields decline by 0.25 basis points per $1 billion increase in coupon spot positions. Here, results based on combined spot and futures positions are generally weaker. The findings are consistent with the role of the dealer acting to intertemporally intermediate predictable supply shocks associated with auctions, with compensation for taking on the associated inventory risk. For coupon securities, residual position changes are significantly negatively correlated with contemporaneous positions. This is the same relationship that we saw for position changes in Table 7, Panel A, but the opposite of fitted position changes. An explanation for the negative residual position change coefficients is that dealers initiate 20

22 changes in their positions for reasons other than those explored in Table 3, perhaps speculative reasons, causing positions and yields to move inversely. We do not detect a relation between residual bill position changes and current or future yields, and results based on combined spot and futures positions are modestly weaker. Lastly, there is only weak evidence of a relationship between residual position changes and yield changes in the subsequent week. The coefficients are generally negative, and significantly so for the 2-year yield and spot positions, providing weak support for the hypothesis that dealers are successfully speculating. The results more generally suggest that while dealers tend to be compensated for position changes explainable by issuance and other factors, they tend not to be compensated for position changes that are not explained by these factors. The fitted results in Table 8 suggest that one or more of the explanatory variables in Table 3 causes positions and yields to move together in the same week and causes positions to move inversely to yields the following week. An obvious candidate variable is Treasury issuance, which explains most of the variation in fitted positions. In Table 9, we replace the fitted position change with the issuance variable, and explicitly identify auctions as the drivers of these results. Table 9 shows a positive relationship between contemporaneous issuance and yield changes (significant for three of the five yields), and a negative relationship between the issuance and yield changes the following week (significant for three of five yields). 21

23 6. Conclusion We use a unique dataset to assess the determinants of U.S. Treasury dealer positions and find that underwriting plays a key role. Dealers thus absorb a large share of Treasury issuance inventory shocks by retaining positions bought at auction instead of immediately selling these positions to other investors. Further results suggest that dealers still hold large shares of issues at maturity, especially issues with short durations. We also find that dealers engage in selective hedging, offsetting a much smaller share of spot position changes in the futures market when such changes are explained by issuance and other variables. Presumably there is less need to hedge such inventory shocks because they are not information-based. Such selective hedging is consistent with the behavior of U.K. government bond dealers documented by Naik and Yadav (2003b). Lastly, we find that dealers appear to be compensated for their intertemporal intermediation of Treasury supply shocks. Dealer position changes attributable to particular, identifiable factors (namely, issuance) tend to appreciate in the subsequent week. Moreover, issuance by itself helps explain Treasury yield changes, with yields rising (prices falling) in auction weeks and declining (prices rising) in subsequent weeks. 22

24 References Adrian, Tobias, and Michael J. Fleming, 2005, What Financing Data Reveal about Dealer Leverage, Current Issues in Economics and Finance, 11, 1-7. Brandt, Michael W., and Kenneth 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, Cammack, Elizabeth B., 1991, Evidence on Bidding Strategies and the Information in Treasury Bill Auctions, Journal of Political Economy 99, Cohen, Benjamin H., and Hyun Song Shin, 2002, Positive Feedback Trading in the US Treasury Market, BIS Quarterly Review, de Vassal, Vladimir, 1998, Time and Seasonal Patterns in the Fixed-Income Markets, Journal of Fixed Income 7, Dupont, Dominique, and Brian Sack, 1999, The Treasury Securities Market: Overview and Recent Developments, Federal Reserve Bulletin 85, Fleming, Michael J., 2003, Measuring Treasury Market Liquidity, Federal Reserve Bank of New York Economic Policy Review 9, Fleming, Michael J., 2006, Who Buys U.S. Treasury Securities at Auction, Manuscript. Fleming, Michael J., and Eli M. Remolona, 1999, Price Formation and Liquidity in the U.S. Treasury Market: The Response to Public Information, Journal of Finance 54, Glosten, Lawrence R., and Paul R. Milgrom, 1985, Bid, Ask and Transaction Prices in a Specialist Market with Heterogeneously Informed Traders, Journal of Financial Economics 14, Ho, Thomas S. Y., and Hans R. Stoll, 1983, The Dynamics of Dealer Markets Under Competition, Journal of Finance 38, Ho, Thomas, and Hans R. Stoll, 1981, Optimal Dealer Pricing under Transactions and Return Uncertainty, Journal of Financial Economics 9, Kyle, Albert S., 1985, Continuous Auctions and Insider Trading, Econometrica 53, Lyons, Richard K., 1996, Optimal Transparency in a Dealer Market with an Application to Foreign Exchange, Journal of Financial Intermediation 5, Madhavan, Ananth, 2000, Market Microstructure: A Survey, Journal of Financial Markets 3,

25 Massa, Massimo, and Andrei Simonov, 2003, Reputation and Interdealer Trading: A Microstructure Analysis of the Treasury Bond Market, Journal of Financial Markets 6, Naik, Narayan Y., and Pradeep K. Yadav, 2003a, Do Dealer Firms Manage Inventory on a Stock-by-Stock or a Portfolio Basis? Journal of Financial Economics 69, Naik, Narayan Y., and Pradeep K. Yadav, 2003b, Risk Management with Derivatives by Dealers and Market Quality in Government Bond Markets., Journal of Finance 58, Ritter, Jay R., 1987, The Costs of Going Public, Journal of Financial Economics 19, Simon, David P., 1994, Markups, Quantity Risk, and Bidding Strategies at Treasury Coupon Auctions, Journal of Financial Economics 35, Spindt, Paul A., and Richard W. Stolz, 1992, Are US Treasury Bills Underpriced in the Primary Market? Journal of Banking and Finance 16, Vitale, Paolo, 1998, Two Months in the Life of Several Gilt-Edged Market Makers on the London Stock Exchange, Journal of International Financial Markets, Institutions and Money 8,

26 Table 1: Descriptive Statistics of Dealer Treasury Positions Panel A: Spot Positions Levels Weekly Changes Bill Positions Coupon Positions Bill Positions Coupon Positions Mean Standard deviation Minimum Maximum Number of observations Skewness Kurtosis Autocorrelation (1 week) ARCH p-value Dickey-Fuller p-value Panel B: Futures Positions Levels Weekly Changes Bill Positions Coupon Positions Bill Positions Coupon Positions Mean Standard deviation Minimum Maximum Number of observations Skewness Kurtosis Autocorrelation (1 week) ARCH p-value Dickey-Fuller p-value Panel C: Combined Spot and Futures Positions Levels Weekly Changes Bill Positions Coupon Positions Bill Positions Coupon Positions Mean Standard deviation Minimum Maximum Number of observations Skewness Kurtosis Autocorrelation (1 week) ARCH p-value Dickey-Fuller p-value Notes: The table reports descriptive statistics of primary dealer net spot (Panel A), futures (Panel B), and spot and futures combined (Panel C) positions in U.S. Treasury securities in billions of dollars for the weeks ending July 4, 1990 to June 28, 2006 (Panel A) or July 4, 1990 to June 27, 2001 (Panels B and C). ARCH p-values are from ARCH LM tests using four lagged squared values, while the Dickey- Fuller p-values are the from Augmented Dickey-Fuller tests with lag lengths chosen using the Schwarz information criterion.

27 Table 2: Descriptive Statistics of Explanatory Variables Variable Issuance Redemptions Federal Reserve holdings Foreign central bank holdings Agency debt positions MBS positions Corporate issuance All Bills Coupons (25.6) (20.7) (13.3) (24.0) (20.8) (13.3) (148.2) (42.9) (108.6) (249.5) 43.5 (28.5) 30.6 (14.9) 5.55 (4.69) Treasury yield (1.64) (1.72) (1.42) Notes: The table reports means (standard deviations) of weekly auction amounts of Treasury securities, weekly redemptions of Treasury securities (including buybacks, calls, and regular maturities), Federal Reserve holdings of Treasury securities, foreign central bank holdings of Treasury securities (held in custody at the Federal Reserve), primary dealer agency debt positions, primary dealer MBS positions, U.S. corporate bond issuance, and Treasury yields for the July 4, 1990 to June 28, 2006 period. Where possible, statistics are reported separately for bills, and coupon-bearing securities. For yields, the 2-year yield is used for combined, the 6-month yield for bills, and the 5-year yield for coupons. Yields are in percent and all other variables are in billions of dollars.