Does Asset Supply Affect Asset Prices? Evidence from the Agency Bond Market (Job Market Paper)∗ Siddhartha Dastidar Ph.D. Student in Finance & Economics Graduate School of Business, Columbia University December 13, 2007

Abstract I examine the effect of a change in the relative supply of bonds issued by the Government Sponsored Enterprises (Fannie Mae and Freddie Mac) on their relative prices. I find that Fannie Mae spreads are tighter than Freddie Mac spreads after an exogenous negative supply shock to Fannie Mae bonds resulting from an undercapitalization announcement. Moreover, a decrease in issuance of Fannie Mae bonds after this event is associated with a decrease in Fannie Mae spreads relative to Freddie Mac spreads, but issuance volume and spreads are not positively related in the rest of the sample. As expected, this supply shock affected the short end of the term structure more than the long end. Collectively, these findings provide evidence that supply of financial assets affects prices, implying that the demand curve for financial assets is not perfectly elastic.

∗

I would especially like to thank my advisor Chris Mayer for copious guidance and support. I have also benefited from useful discussions with John Donaldson, Steve Drucker, Ray Fisman, Charles Jones, Joseph Stiglitz and Suresh Sundaresan. Doug Richardson from ICI was extremely helpful in providing mutual fund flow data. Corresponding email: [email protected]

1. Introduction This paper attempts to empirically examine the role of asset supply in determining asset prices. In the classical asset pricing setup, the supply of securities may not matter. This is true when the demand curve for the security is perfectly elastic, a commonly-held assumption in the asset pricing literature, which has traditionally focused on the role of demand-side variables. This will be true, for example, when the investors can replicate the payoffs for this security by investing in a combination of other securities and when the market is frictionless (see, for example, Black and Scholes, 1973). However, there are reasons to believe that the world of asset pricing is not well represented by this frictionless ideal. There may be certain constraints that restrict the smooth substitutability across states and time, or incentives or constraints faced by investors (e.g. compensation contracts, agreements with their investors) that restrict their participation in some securities. Further, incomplete markets may lead to a restricted asset span and hence constrain the replication of security payoffs. Under these circumstances, the supply of assets may plausibly be an important determinant of prices. Indeed, research reports by major investment banks often indicate that this is a strongly held view among practitioners.1 Issuers’ propensity to supply financial assets may vary over time (as in, for example, optimal dynamic capital structure models; see Goldstein, Ju and Leyland (2001)), and if such a world is not frictionless, these changes will lead to a relative glut (scarcity) in the market in some time periods, keeping prices lower (higher) than they would otherwise be. However, we face the standard identification problem in uncovering such effects. We observe ex-post issuances and prices, and it is difficult to decipher whether the issuance was in response to greater investor demand or higher issuer propensity to supply. Hence, to uncover the impact of asset supply on asset prices requires an unexpected change in an issuer’s propensity to supply assets, where we may trace out the differential effect on asset prices relative to a control. In this paper, I take advantage of an unanticipated shock to the supply of bonds to study the resulting impact on primary and secondary bond spreads in the Government Sponsored 1

For example, the Bank of America 2005 Mortgage Outlook Report stated, “On the supply side, we expect the universe of outstanding residential mortgage debt to grow by approximately $750 billion. However, on the securities side of the equation, we expect the outstanding volume of fixed-rate MBS to decline by $50 billion—$75 billion in 2005. Most of these declines will likely be concentrated in the agency MBS subsector and should provide a firm support for historically rich valuations.”

2

Enterprises (GSE) Agency market, where the main issuers are Fannie Mae and Freddie Mac. Using exogenous demand and supply shocks to Fannie Mae bonds resulting from accounting fraud and undercapitalization announcements respectively, I find that Fannie Mae spreads did indeed tighten relative to Freddie Mac (my control) after the negative supply shock but not after the negative demand shock. I further examine the relationship between bond spreads (adjusted for duration-matched treasury spreads) and current period issuance volume2 before and after the event in both the primary and secondary markets. I find that after the supply shock, relative to Freddie Mac, the relation between spreads and period volume of Fannie Mae switches to positive after the supply shock caused by the undercapitalization announcement and remains so for months, before eventually dying out. As expected, no such effect is observed for the demand shock. This indicates that, after the supply shock, lower volumes are associated with higher prices (i.e. lower spreads) for Fannie Mae bonds relative to Freddie Mac. Moreover, the short end of the term structure is affected more than the long end, consistent with the underlying reason (duration mismatch) behind the supply shock and expost data. The effect of this event on the credit default swap (CDS) spreads of Fannie Mae and Freddie Mac is also considered. This, collectively, offers compelling evidence that the supply shock has a significant effect on asset prices and suggests that supply factors need to be taken seriously in modeling asset prices3 . There is, of course, a prior literature that relaxes the assumptions of the frictionless asset pricing world, and there exists earlier work on the slope of the asset demand curve specifically. Most notably, Stiglitz (1972) and subsequent papers question the completeness of financial markets, and analyze the implications of market imperfections. On the empirical side, one strand of literature focuses on block trades to examine the issue of demand elasticity. Scholes (1972), Kraus and Stoll (1972) and Holthausen, Leftwich, and Mayers (1990) document price pressure in the reaction of equity prices to large block trades, while another set of authors (Shleifer (1986), Harris and Gurel (1986), Wurgler and Zhuravskaya (2002) and Chen, Noronha, and Singal (2004)) investigate the effects of changes in the composition of the S&P 500 Index. Finally, Hess and Frost (1982), Ofek and Richardson (2000) and Mitchell, Pul2

By current period volume, I refer to15-day issuance volume before and after the current date. This is a key dependent variable in my regressions. The results are robust to alternate definitions. 3 The equity prices and implied volatilities of Fannie Mae are fairly stable during this announcement, suggesting that the information content was minimal, in contrast to a demand shock a few months earlier (Figures 1-4). If demand was negatively affected because of this news, it would make my results harder to find.

3

vino, and Stafford (2004) examine the price impact of seasoned equity offerings, IPO lock-up expirations and mergers respectively. The current paper improves on this earlier literature in a number of ways. Most importantly, earlier work that attempted to measure the role of asset supply is plagued by difficulties in disentangling the effects of demand from the effect of supply, since many key observables are determined in equilibrium by the joint effects of demand and supply, and an observable supply proxy (that is orthogonal to demand) is difficult to find. Moreover, there is some endogeneity in the system (issuers responding to price movements) that is hard to break. This paper attempts to make some headway in these directions, by using an unanticipated event that can be identified as a supply shock, and looking at the consequent effect on bond spreads. Moreover, this paper uses joint movements in price and quantity to identify demand and supply shocks (similar to Cohen et. al., 2006), instead of looking only at price movements. Finally, I use a control group (Freddie Mac) that is arguably very similar to the treatment group (Fannie Mae), but unaffected by the exogenous shock. While most of the earlier inquiries look at the equity markets, there are some studies on the fixed income market too (such as Newman & Rierson, 2004). However, they find the supply effect to be mainly a short term phenomenon and attribute it to liquidity, whereas I find the effect to persist for months and attribute it to a fundamental supply crunch. Moreover, in their paper, it is difficult to argue that the shock they consider (issue of telecom bonds during the technology boom) is not driven by demand. There is also a fixed income credit pricing literature that looks at determinants of bond spreads (e.g. Collin-Dufresne, Goldstein & Martin (2001), Huang & Huang (2003)). Both structural models (which model the firm’s cash flow as a stochastic process e.g. Merton (1974), Fisher, Heinkel and Zechner (1989), Leland (1994), Anderson and Sundaresan (1996)) and reduced-form models (which model the firm’s default arrival e.g. Duffie and Singleton (1998)) have been used, but the effect of asset supply has not been explicitly incorporated. However, Collin-Dufresne, Goldstein & Martin (2001) and Duffie & Singleton (1997) conjecture that supply shocks may be affecting asset prices significantly. Boudoukh et. al. (1997) studies the pricing of Ginnie Mae backed mortgage pools and concludes that while the interest rate level and slope adequately proxy for the optionality, a missing common factor remains. Using credit default swap data, Longstaff, Mithal and Neis (2005) show that the non-default component of bonds varies strongly with market liquidity. These papers form the basis for including various

4

control variables in the regressions. Further, in the fixed income literature, there are very few (Jaffee (2003) is a notable exception) papers that study Agency bond markets, which is surprising considering the size of this segment (bigger than the Treasury and the corporate bond market), as listed in Table 1. This paper also represents an attempt to explore this market. The paper is organized as follows: Section 2 provides some background information on the Agencies. Section 3 describes the empirical setting. Section 4 details the data and the methodology. Section 5 discusses the results. Section 6 concludes.

2. Background4 The GSE Agency markets comprise Federal National Mortgage Association (Fannie Mae), Federal Home Loan Mortgage Corporation (Freddie Mac) and the Federal Home Loan Banks (FHLB). Government National Mortgage Association (Ginnie Mae) comes under the purview of FHLB. In this study, we restrict our attention to Fannie Mae (FNM) and Freddie Mac (FHLMC), since they have issued the bulk of the securities in this market, and are the most liquid. Also, for reasons explained later, Freddie Mac is a better control group for my analysis than FHLB. Fannie Mae and Freddie Mac (the agencies) were created in 1938 and 1970 respectively by a charter of the government to promote home ownership. One of the primary goals of creating Fannie Mae and Freddie Mac was "... to provide supplementary assistance to the secondary market for home mortgages by providing a degree of liquidity for mortgage investments, thereby improving the distribution of investment capital available for home mortgage financing." During the 1970s and early 1980s, Fannie Mae and Freddie Mac played a lead role in the development of the securitization market. However, with rising interest rates in the early 1980s, Fannie Mae’s cost of funds rose above the interest rate it was earning on its long-term, fixed-rate mortgages. This interest rate risk because of a duration mismatch was similar to that faced by the Savings and Loan industry. A crisis for Fannie Mae was averted combination of legislative tax relief, regulatory forbearance, and a decline in interest rates. In 1989, Congress restated Fannie Mae’s and Freddie Mac’s charters, directing the GSEs to "provide stability" and "ongoing assistance to 4

Parts of this section have been heavily borrowed from Gary Gensler’s congressional testimony in 2000 and Jaffee (2003).

5

the secondary mortgage market." Fannie Mae and Freddie Mac operate in two main lines of business — mortgage securitizations and management of mortgage portfolios. The first line comprises buying whole mortgages from originators based on certain criteria, pooling them into portfolios, and selling them to other investors, while guaranteeing the payments that the mortgage holders need to make. The GSEs charge a fee of about 19 basis points for this, and bear the credit risk of the homeowners. This has historically been more than adequate compensation for the realized loss levels. The other line of business, which has gained prominence over the last 15 years or so, is the investment in retained mortgage portfolios. Here, the agencies directly purchase many mortgage-related securities, including securities that they themselves have issued. Their retained portfolio has been rising rapidly compared to the mortgages they securitize and was close to 80% in 2001. While Freddie Mac had a much lower proportion in early on, the strategies of the two firms are now similar. The profit margin here depends on the difference between the interest earned on the mortgage assets, less the cost of the funding liability. So, this activity rises significantly when market interest rates are low, when the agencies borrow in the credit markets and use it to buy back mortgages. This differential return is compensation for bearing default risk, interest rate risk and liquidity risk, and stood at 104 basis points in 2001. This return is maximized by not completely hedging the source of these risks, the benefit of which reaches the shareholders of the firm. The agencies also have a relatively small proportion of their investments in non-housing financial assets. The agencies have to maximize their profits in this business line primarily because they enjoy special benefits, which provide them with significant advantage in the capital markets. Yet, while a portion of these benefits is passed on to homeowners in the form of lower mortgage rates, the rest of the cost reductions provide higher returns to GSEs’ shareholders. Studies conducted by Treasury, CBO, and GAO over the past ten years concluded that the GSEs retain a significant amount of their federal subsidy. Given the objectives for which Fannie and Freddie were established, this outcome is a contentious issue. Some of the benefits they enjoy are listed below: •

Their debt and mortgage-backed securities are exempt from registration with the

•

The GSEs are exempt from state and local corporate income taxes.

Securities and Exchange Commission.

6

•

The GSEs have a line of credit from the Treasury that authorizes Treasury to pur-

chase up to $2.25 billion of Fannie Mae’s and Freddie Mac’s obligations and up to $4 billion of the Federal Home Loan Bank System’s obligations. •

Banks are permitted to make unlimited investments in GSEs’ debt securities, whereas

•

GSE securities are eligible as collateral for public deposits and for loans from Federal

•

GSE securities are lawful investments for federal fiduciary and public funds.

there are limits placed on their investments in any other company’s debt securities. Reserve Banks and Federal Home Loan Banks. •

GSEs are authorized to use Federal Reserve Banks as their fiscal agents, including

issuing and transferring their securities through the book-entry system maintained by the Federal Reserve. These benefits translate to lower funding costs (much cheaper than other corporates with similar performance), lower leverage compared to other financial institutions of similar standing, and direct cost savings. These advantages are particularly significant to the GSEs because of the markets in which they operate. The U.S. capital markets are the most competitive and efficient in the world. Relatively small advantages, even those measured in single basis points, over time can allow firms to dominate their markets. 2.1. Capital Market Activities of the GSEs Ignoring the smaller components, the main assets of Fannie and Freddie are mortgage-related, which are funded by debt and equity. The assets comprise the whole mortgages that they buy to eventually securitize, as well as purchases of mortgage related securities, including repurchases of mortgages they had securitized earlier. This is by far the more substantial part of their portfolio. This is funded predominantly by debt (and some equity). As mentioned earlier, these firms use their GSE status to hold less equity per dollar of debt compared to other financial firms. Moreover, while mortgage assets are typically long term, these firms have chosen to fund them by primarily short term debt (>50% of total debt) thereby exacerbating the duration mismatch between assets and liabilities while maximizing the interest spread. Further, the debt issuance of these firms has grown at a very rapid rate, and currently is almost equal to the volume of debt that the US Treasury issues. The agencies hedge this interest rate risk by participating in the swap market. In fact, they are one of the largest players in that market, and do not post collateral, unlike all other 7

participants. So, since the agencies need to add duration to their liabilities, the specific trade they would do would be to pay fixed and receive floating in the swap. This strategy is effective since the term structure of swaps is usually flatter than the agency debt tem structure. The other significant risk that the agencies are exposed to is the prepayment risk of the mortgages. Most mortgage contracts in the US allow the homeowners to prepay the mortgages when they wish, without penalty. So, when rates go low, the homeowners tend to repay the loans, leading to a reinvestment risk for the agencies, since they are saddled with money exactly when rates are low. Another way to express this is the fact that the liabilities value may go up if they are fixed, but the asset value is capped because of negative convexity, leading to balance sheet erosion. The agencies hedge this risk by issuing callable bonds. However, this reduces their issuance proceeds since callable bonds are cheaper than non-callable bonds (as the lender is short an option). So, the agencies have only a relatively small proportion of their bonds as callable, whereas almost all the mortgages have prepayment options. The other popular mechanism to hedge this exposure would be to enter into swaptions to receive fixed and pay floating. Since all these hedges have costs associated with them, the agencies hedge against only the likely interest rate scenarios, leaving the relatively unlikely scenarios unhedged. While this certainly increases their profits, adverse interest rate movements can severely affect their balance sheet, as shown by OHFEO stress tests. So, to summarize, Fannie Mae and Freddie Mac are privately owned but government chartered firms that were set up to provide liquidity to the secondary market for mortgages. Over time, these firms have moved to becoming very large investors in mortgages. While they have been eminently profitable, they are carrying a lot of interest rate risk. One of the major drivers of these high profits has been the special benefits they receive as GSEs. 2.2. Debt Issuance and Trading in Fixed Income Markets As of December 31, 2004, according to the Bond Markets Association, the total fixed income markets outstanding debt was about $23.5 trillion, with Treasuries comprising 17%, corporates 20%, mortgages 22%, agencies 12%, money market 12%, municipals 9% and asset-backed 8%. However, from a secondary market liquidity perspective, the Agencies are very liquid, with a trading volume at 80 billion dollars about four times that of corporates. Daily trading volume in Treasuries is about 500 billion dollars. Further details on issuance and outstanding are listed in Table I. 8

[TABLE I — ABOUT HERE] Within the Agencies universe, on December 31, 2004, Fannie Mae had about 941 billion dollars of outstanding debt, Freddie Mac had 733 billion dollars and the Federal Home Loan bank System (which comprises of 12 banks) had about 870 billion dollars. More details of Fannie Mae and Freddie Mac debt financing activities are detailed in Table II. We observe that the long term debt issuance for both Fannie and Freddie went down (as did the issuance for most of the universe, since Fed was raising rates), but the reduction in Fannie was far larger than that of Freddie. In the short-term segment, Freddie Mac’s debt outstanding has been stable from 2003 to 2005, but Fannie Mae’s debt outstanding fell by almost 50% from 2004 to 2005. During my tests, I shall return to this and check if the patterns in the price data are consistent with this ex-post empirical observation. [TABLE II — ABOUT HERE]

3. The empirical setting — Event details Over the last few years, Fannie Mae and Freddie Mac have been in the midst of several irregularities. Shortly after the Enron debacle, in January 2003, Freddie Mac announced accounting errors. The regulator (OHFEO) found evidence of fraud, which prompted a senior management change and a financial clean-up. In July 2003, OHFEO announced an inquiry into Fannie Mae’s accounts, out of caution. On 21 September 2004, the regulator discovered accounting fraud and evidence of earning smoothing in Fannie. However, the capital available was found to be adequate. Over the next few months, there was a sequence of news events that suggested governance problems in the company. First, there was a disclosure by the company that the current top management compensation contracts were leniently worded, and the conditions for dismissal for cause were being strengthened. The external directors of the board took over all communication with the regulators (instead of the CEO). On September 29, it was announced that the Chief Financial Officer had sold company shares before the accounting scandal. The Justice Department and the SEC announced independent inquiries. On October 25, the SEC suggested that this inquiry could take months. Fannie failed to file quarterly reports with SEC. Fannie issued a $9billion loss warning on November 16, if the accounting inquiry had an adverse ruling. On December 16, SEC asked Fannie Mae to restate results. The earlier rumors of CEO departure were also confirmed. On 21 December 2004, the regulator announced that it had determined that Fannie Mae 9

was undercapitalized, and instructed Fannie Mae to scale down its business and submit a plan for recapitalization, which would be monitored on a weekly basis. Since this would necessitate a reduction in the balance sheet size, the amount of debt outstanding would need to go down. While the markets reacted to all these news events, I interpret that the event on September 21 primarily affected investor demand, whereas the news on December 21 mainly affected supply (negatively), since almost all of the negative demand related news effects would have been factored in over the past few months. Apriori, this interpretation is likely to make it harder to obtain the empirical evidence that I end up observing in the data. Also, the stock price did not react much to the undercapitalization announcement whereas it fell by about 15% following the fraud announcement on September 21, suggesting that the negative news of this latter (supply) shock had been priced in already, and the announcement would have more to do with expectations of future bond supply. Graphs of the stock price and the implied volatilities for Fannie Mae and Freddie Mac around these event dates are enclosed in Figures 1-4. Figure 5 shows a plot of the spreads over Treasury for the most liquid Fannie Mae and Freddie Mac bonds in the 10-year sector (data obtained from Lehman Brothers).

4. Data and Methodology The sample period is from 1998 to 2005, though I mainly use daily bond-level data from mid-2004 onwards. Deal-level issuance data for Fannie and Freddie bonds has been obtained from SDC Platinum (now ThomsonOne Banker). All floating rate issuances have been excluded. Treasury rates and secondary market bond prices are obtained from Datastream. The methodology involves regressing spread levels (or spread differences, for robustness with secondary bond prices) on factors that could plausibly be affecting prices, on sub-samples before and after the demand and supply event dates. F-tests for structural breaks have been conducted. This paper primarily examines whether a supply effect exists, rather than focus on its magnitude. So, regression seems an appropriate technique to study this question, since it orthogonalizes the effects of various variables that may be contributing to bond spread movements. This becomes particularly important since prior literature has found it difficult to nail down what drives spreads in high grade bonds. So the methodology of choice has to be flexible and robust to model misspecification. Even if there were a widely accepted model for high-grade credit spreads, non-availability of balance sheet data and use of data at a daily 10

frequency would make it difficult to apply. It is likely that prices obtained from Datastream include prices other than actual transactions, although the panel data would probably not have been so heavily unbalanced (unrelated to issuance/ redemption) if the non-traded bonds were being mechanically priced. Warga and Welch (1993) have shown that using non-transactions data can lead to problematic inference, so this is a potential concern. I believe that, since I am dealing with the Agencies market, which is far more homogeneous and liquid than the corporate bond market (average trading volume is four times as high in Agencies compared to corporate bonds, across far fewer names), this problem is likely to be much less serious. Nevertheless, to be certain, I exclude all trades on dates that are more than 90 days after the issue date. This implies that I only look at bonds that are less than one quarter old, and there is no seasoning effect. While my results hold without this filter too, I report the cases only with these new bonds. Further, many of the agency issues are created based on reverse inquiry. These issues are smaller and unlikely to trade actively, so I exclude all issuances less than USD 250 million in my base cases. Again, my results hold without this filter too. These filters were suggested by Sarig and Warga (1989). I also include the deal issuance volume IssV ol as a control variable to control for the effect of size. Finally, I replicate my analysis using the primary market issuance price data, and these results hold. The demand factors are obtained from prior literature (e.g. Collin-Dufresne et. al. 2001). These include equity returns, S&P 500 returns, level (r2), slope (r10-r2) and curvature (r210), implied volatilities, probability of a negative jump. The equity and Treasury returns are obtained from Datastream. Data on implied volatility of options is obtained from OptionMetrics, as is the data on option prices and strikes required to estimate the jump probability using the methodology described in Collin-Dufresne et. al. 2001. Mutual fund monthly bond flows have also been included, as an additional explanatory variable, as documented by Fridson and Jonsson (1995). This mutual fund flow data is obtained from Investment Company Institute (ICI). I use the issuance volume of the particular security to control for variation in cross-sectional liquidity. No balance sheet data is used, since GSE balance sheet data is considered suspect for the sub-sample periods of interest around the event date. The lack of credible balance sheet data is also why Merton-type models cannot be used in this exercise. To consider the effect of changes in asset supply, I include variables related to the total bond issuance volume of Fannie Mae and Freddie Mac. These include the issuance on the current

11

day and the issuance over the past and next 2 weeks for both the firm and its competitor. Table III contains summary statistics for the 30-day window around the supply shock, based on which I report most results in the paper. A list of relevant variables is included in Table IV. [TABLES III AND IV — ABOUT HERE] I also obtain 5-year CDS spread data from Bloomberg for Fannie Mae and Freddie Mac senior debt. The data in Bloomberg goes back only till October 2004, so I can use it for some of my tests, but not all. I test my hypothesis regarding the effect of a change in asset supply on asset prices on two different samples, one containing secondary market spreads, and the other containing spreads from primary issuances. I compute duration and issuance yield for each bond, and use constant maturity Treasury yields to compute a spread over duration-matched Treasury yield. Since most Agency issuances have call features, I compute the bond duration both to first call and to maturity before computing the spread for the primary sample. All results are very similar and robust to both dependent variables. I report only results that use duration computed to maturity for all bonds, except for callable bonds where the coupon rate is higher than the yield, in which case the duration is computed to next call. This yield spread is the dependent variable in my regressions. Replicating the analysis excluding the callable bonds gives qualitatively similar results, which have not been reported. Spreads computed over the swap curve (instead of Treasuries) also gave similar results, but have not been reported since the Agencies are the largest players in the swap market, and the swap curve itself may be influenced by these announcements. The analysis for the secondary market has been replicated in differences instead of level, to alleviate concerns about non-stationarity. The results have not been reported, but hold good. While a formal test for stationarity was not done because of an unbalanced panel, plotting the autocorrelation function for some bonds did not indicate any significant concerns. The methodology involves regressing spreads (or spread changes/ differences, as the case may be) on all variables that prior literature has identified as important. All these relate to the demand side listed above. Most prior work has not manages to nail the determinants of high grade bond spreads, and this issue persists in my regressions too. After controlling for determinants identified in prior work, I introduce variables related to issuance volume as explanatory variables. The association between these issuance volume variables and spread may

12

either be because of residual demand factors that were not accounted for, or supply changes. If demand factors dominate the supply factors, one would expect a negative coefficient on 30-day own issuance volume volmain30 (i.e. spreads are higher, i.e. prices are lower, when quantity demanded is lower or vice versa). Endogeneity is also likely, with firms trying to issue more when prices are high, leading to the same effect on the coefficient. However, if the supply effect dominates the demand effect, one would see a positive coefficient. So, the results are biased downward, and not finding a significant positive coefficient does not rule out the importance of the supply effect. Cohen et. al. (2006) also use the joint movements in price and volume to distinguish a demand effect from a supply effect. Further, if the supply effect is strong, one may expect a negative coefficient on the firm dummy in some simple specifications with small event windows including only the basic demand controls. This is tested in the difference-in-difference specification below. 4.1. Specification details 4.1.1. Difference-in-difference (secondary sample) I begin by focusing on a narrow window around the event date (to isolate this event), and look at a difference-in-difference regression specification. Specifically, I consider ten days before and after the event, and regress spread (levels) on a Fannie Mae dummy variable, a post-event dummy, their interaction term and the most important demand-side variables (equity price, implied volatility and duration). The regression is detailed in equation 1 below: spread = α1 + β 1 F NMdum + β 2 Eventdum + β 3 F NMdum ∗ Eventdum

(1)

+β 5 f irmeqpr + β 6 impvol + β 7 duration + ε If a supply effect is present in the data, I would expect β 3 , the coefficient of the interaction term to be negative and significant, especially once controls are introduced. 4.1.2. Issuance volume — spread co-movement (primary and secondary samples) For a more elaborate test, I look at the sensitivity of the spreads to issuance volume, before and after the event. For a supply effect, the spread and volume for Fannie bonds should be positively correlated relative to Freddie after the event, but not before. I test this hypothesis using various specifications and window lengths, for both the primary and secondary bond spread samples. I regress spreads on the demand factors described above. To explain the 13

residual variation from this regression, I then include issuance volume for the day (volume), as well as the volume issued by this issuer and the competitor for the past 15 days and the next 15 days (the period volumes volmain30 and volcomp30), which prior work (e.g. see Newman & Rierson (2004)) indicates is in the information set of the investor at the time of issuance. The interaction between this 30-day issuance volume and the firm dummies (volmain30 ∗ F NMdum) is also included, the coefficient of which (δ 4 ) is of primary interest.

The change in this coefficient, as one moves from pre-event to post event, indicates whether there is any firm-specific change in the effect of volume on spreads centered around the event. I run this for various event windows but report only the 30-day window, before and after the event date. The results also hold for 60, 90 and 120 day windows, though the magnitudes get smaller (as one might expect) as one moves further out in time. The specification is spelt out below: 2 spread = α + β 1 f irmeqrtn + β 2 SP 500rtn + β 3 r2 + β 4 (r10 − r2 ) + β 5 r10 + β 6 f undf lows

+β 7 duration + β 8 impvol + β 9 corpsprd + β 10 Jump + β 11 IssV ol + ε

(2)

ε = γ + δ 1 volume + δ 2 volmain30 + δ 3 volcomp30 + δ 4 (volmain30 ∗ F NMdum)

During the estimation, I change the specification slightly to include the volume variables in the first stage regression itself without extracting residuals. With the secondary sample, I also include bond fixed effects to capture time-invariant bond specific unobserved variation. Thus, the identification in this specification is coming from within-bond variation. I also allow for arbitrary correlation across bonds at any point in time by clustering by day. Since the issuance data suggests a much larger decline in short-term debt, I look at the incremental supply effect for short-term debt by including a triple interaction (with a maturity dummy). I repeat this regression in differences for all event windows. 4.1.3. Matched Sample (secondary market data) I also create a matched sample of Fannie and Freddie bonds (and look at the relation between (Fannie spread — Freddie spread) and cross sectional differences in the explanatory variables. In this specification, the coefficient on the difference in 30-day volume for Fannie and Freddie is of interest. In the equations below, “diff” denotes the difference between a matched FannieFreddie pair. 14

2 spreaddiff = α + β 1 f irmeqprdif f + β 2 SP 500rtn + β 3 r2 + β 4 (r10 − r2 ) + β 5 r10 + β 6 f undf lows

+β 7 durationdiff + β 8 impvoldiff + β 9 corpsprddiff + β 10 Jumpdif f + ε ε = γ + δ 1 volumediff + δ 4 volmain30dif f Apart from being robust to different data samples (primary, secondary and matched), these results are also robust to the choice / construction of the explanatory issuance variables. They are also not sensitive to event window. These sensitivities have not been reported. The above regressions may be fine-tuned by considering issuance volumes net of redemptions. That is left for future research. I think of the September 21 2004 fraud announcement to be primarily a demand shock. December 21, 2004 primarily appears to result in a supply shock to Fannie Mae bonds, since the most of the negative news has already been announced. Violation of this will make my results harder to get. Also, this is empirically validated by examining the price-issuance co-movement detailed above. Since this analysis uses panel data, one may worry about the correlation in errors in the cross-section of bond prices driving these results. To confirm this, the analysis was repeated with bond fixed effects and clustering by time (i.e. day), and with bond clusters and time dummies. Moreover, Fama-Macbeth regressions (which take care of cross-sectional correlation) were also run. In all these cases, the effects did not die out. This provides additional comfort that the significance is not a reflection of the correlation in the error terms. 4.1.4. Credit Default Swap Spreads I then turn to the credit defaults swap market, and consider the 5-year spreads on Fannie Mae and Freddie Mac. The CDS is a synthetic contract created by a dealer, in contrast to bonds, which are physical securities issued by the firm. So, a supply shock is more likely to affect the bond spreads than the CDS spreads (“no-arbitrage” trades are difficult to implement here, as explained in the results section). The CDS spreads are more likely to respond to default probability changes. I test this hypothesis by replicating some of the above analysis (difference-in-difference) with CDS spreads data. Data constraints do not permit me to replicate the full study.

15

(3)

4.2. Additional Tests While I look for a positive significant value of δ 4 post-event (and the absence of a positive significant value pre-event) after the supply shock in the above regressions, the more precise test is actually whether δ 4 increased significantly from pre-event to post-event. Here, as described above, for every spread-issuance volume case, I run two separate regressions, before and after the event, to check price volume co-movements. However, this is equivalent to running one regression using both pre-event and post-event data and interacting all variables with the event dummy. This single regression allows me to conduct a F-test for structural breaks to test if the coefficient of interest changed significantly after the event. I do this for all my specifications, and in almost all cases, the F-test confirms that the coefficient changed at a 5% significance level. The only exceptions are situations where the coefficient pre-event is estimated with such a wide standard error that it is difficult to distinguish it from a coefficient with a very different point estimate. These tests have not been reported.

5. Results This section presents results on the existence of a supply effect related to two main specifications — a simple difference-in-difference test with very few controls and the main spreadissuance volume co-movement test. For robustness, I use two main datasets, from the secondary market and the primary market (from different vendors) respectively, and a matched sample derived from the secondary market data. I also use data from the CDS market as an additional robustness test. The difference-in-difference specification is applied to the main secondary market data, the matched sample and the CDS data, whereas the spread-issuance specification is applied to all the samples. The analysis on the secondary and primary markets (excluding the matched sample) has been performed on two events — a demand shock and a supply shock, with separate pre-event and post-event regressions for the spread-issuance co-movement. In the other cases, only the main event of interest (the supply shock) has been used. This analysis mainly uses daily bond-level data from May 2004 to June 2005. It is robust to various window lengths (30, 60, 90 and 120 days), but I present results only related to 30 days for the spread-issuance test, and 10 days for the difference-in-difference test. The secondary market data presented here includes callable bonds, since most of my sample contains callable 16

bonds, but the results hold without them too. I do not use option-adjusted spread data, since prior literature has documented the unreliability of these numbers (see, for example, Heidari and Wu (2004)), and my conversations with practitioners confirmed these problems. The results also hold without the bond age and issue size filters that I impose. These results are also robust to the way I define the issuance volume variable and regardless of whether spreads are computed over swaps or Treasuries. I also run both the specifications on the CDS dataset. Almost all the regressions contain fixed effects (either bond-level or date), so the firm dummy is not identified. In cases where some bond effects are dropped, the change in the firm dummy as one moves from pre-event to post-event is in the expected direction. I look for a negative coefficient of the interaction term in the difference-in-difference regressions, for reasons I explain in the relevant section. In the spread-issuance co-movement regression, the positive coefficient of the interaction term indicates a supply effect. This shows up only after the supply shock in all samples (excluding the CDS data, as it should be) / window lengths and not at any other time (including before the supply shock, before and after the demand shock, and also in a large-sample primary market regression from 1998 to 2004), providing comfort with the methodology. After demonstrating existence of the effect, I try to tease out the extent of the effect in different segments of the term structure using a variant of the spread-issuance specification and restricting myself to the secondary market data, motivating my hypothesis by the issuance data in Table II. As pointed out above, some of my analysis has been performed only on the secondary market sample; the prime reason for this is the larger sample size in that dataset. While I include many controls in the spread-issuance specification, prior literature has noted the economic soundness of these controls, but their weak empirical performance, especially for high grade corporate bonds. This is true in my sample too, and makes my main result harder to obtain, in principle. The secondary market analysis has been carried out in differences too and the results hold just as well, but only the analysis in levels has been reported here. Moreover, an additional, equivalent specification to the spread-issuance co-movement test has been run throughout, but not shown. This specification, discussed in section 4.2, provides an opportunity to test for structural breaks (by doing an F-test that the coefficient of the interaction term increased significantly after the supply shock but not otherwise). In almost all cases, this works out as one would expect. These tests have not been reported. 5.1. The Supply Effect in Secondary Bond Spreads 17

5.1.1. Difference-in-Difference As a preliminary analysis, I first regress the spread levels on event and firm dummies, their interaction and very few controls using a small 10-day pre and post-event windows (to isolate this event from others) around December 21, 2004 (the event date for the negative supply shock), using a difference-in-difference specification. I find that the coefficient of the interaction term is negative and significant (β 3 in equation 1), even after introducing daily fixed effects, implying that the spreads of Fannie Mae bonds are tighter relative to Freddie Mac after the (negative) supply shock but not before. I find no such result after the demand shock. These results are presented in Table V. [TABLE V — ABOUT HERE] The effect of the relative supply shock is significant, at a little over one basis point. This order of magnitude is a common feature of all the specifications that I run (with spread changes the numbers are obviously smaller) and is not necessarily evidence of economic insignificance of the effect. To put this in perspective, bid-ask spreads of one basis point are common in the very liquid securities. Research reports published by major investment banks routinely discuss trading opportunities worth a couple of basis points. Moreover, since dealers do not pay trading costs, they are usually open to trading in opportunities that offer one basis point. While I do not suggest that one should trade based on such a model (since this is not intended to be close to a true model for spreads; it merely othogonalizes the effects of the various variables that might potentially be affecting spreads), it is comforting that the numbers are not grossly different from what one would expect from the true data generating process. I include a further discussion on the magnitudes at the end of this section. 5.1.2. Price-issuance co-movement Table VI reports regression results for the movements in the security bond spread associated with various explanatory variables during 30-day pre and post-event windows around December 21, 2004, when the regulator announced that Fannie was inadequately capitalized. I interpret this event as an unanticipated supply shock, and surmise that the incremental negative effect of this news on investors’ assessment of Fannie’s risk was not substantive and had already been incorporated on earlier dates, after all the earlier bad news. If this was untrue, it makes it even more difficult to find supply-driven effects. Thus, not finding a significant positive value of δ 4 does not necessarily imply the absence of such supply effects, 18

but a possible omitted variable bias because of model mis-specification that is killing the result. The instability of the coefficients of the demand factors have been highlighted before by Collin-Dufrense et. al. (2001) and following papers, and I find the same issue here too. The regressions with only the demand factors are presented in columns 1 and 4. In the 30-day window, many of the demand variables at least have the economically intuitive sign, even if they are not significant. We find that spreads move inversely with equity returns (both for the firm and the market), increase with level and slope factors and decrease with curvature. One would also expect the spread to decrease as bond flow increases, but that is not observed post-event. The results for implied volatility switches and duration (term premium) appear to be going the wrong way. Corporate spreads are positively associated with security spreads, as expected. [TABLE VI — ABOUT HERE] In columns 2 (pre-event) and 5(post-event), I include variables related to issuance volume, specifically own issuance volume during the past and next 15 days (period volume), competitor issuance during the past and next 15 days, own issuance changes on the trade date and interaction between the Fannie Mae dummy and the own issuance 30-day volume to capture differential effects. This assumes that the next 15-day issuance is in the current information set, as documented by Newman and Rierson. In the 30-day window, the significant variables include the competitor’s 30-day volume (volcomp30) and the interaction between the Fannie Mae dummy and own-issuance 30-day volume, the coefficients of which are δ3 and δ 4 in the above specification in equation 2 respectively. δ 4 , which captures the effect of Fannie issuance on its spreads relative to Freddie, is of prime interest to us. It switches from negative (not significant) from column 2 to positive and significant (at the 1% level) in column 5, thereby indicating that Fannie Mae spreads were more strongly positively related to issuance volume relative to Freddie, pointing to a supply effect. Thus, after the shock, a decrease in Fannie’s issuance volume gets associated with a reduction in spreads relative to Freddie, possibly reflecting a scarcity value. The magnitude is around one basis point for every billion dollars of issuance. δ 3 is significant throughout, but changes signs from positive to negative. A negative coefficient of δ 3 suggests that increases in competitor issuance are associated with higher prices, possibly because of the competitor timing the market. A positive coefficient ( δ 2 ) of the period volume (volmain30 in the specification above) indicates that after controlling

19

for observable demand drivers, higher spreads (i.e. lower prices) are associated with higher volumes. This may indicate that (ignoring endogeneity for a moment), after parsing out the demand, a residual increase in spread may be associated with an increase in volume because a shortage of the two securities is pushing up prices or vice versa. Had the coefficient been negative, we may have argued that firms issue relatively less in times when prices are low (this argument implies that the system is endogenous), either because firms time the market to issue more securities when prices are favorable, or because their investment opportunities are correlated with the period when investors are willing to pay more. Focusing on exogenous event dates helps deal with the endogeneity issue to some extent. I shall discuss columns 3 and 6 later. Tables VII considers similar event windows around September 21, 2004, on which date the regulator announced that it had discovered accounting fraud in Fannie’s books (the regulator had been conducting an inquiry since July 17, 2003). However, capital levels were considered adequate at this stage. I interpret this as a demand shock, and hypothesize that the relation between spreads and volume for Fannie Mae relative to Freddie Mac (i.e. the interaction term) shall not be positive after the event. None of the volume-related variables are significant before September 21. Also, the coefficient of the interaction term (i.e. δ 4 ) remains insignificant after the event. The direct volume effect (i.e. δ 2 ), the own-volume effect (i.e. δ 2 ), and competitor volume effect (i.e. δ 3 ) all have negative signs, suggesting that issuers are responding to prices to time the market. So, there appears to be no evidence of a supply effect around the fraud announcement. [TABLE VII — ABOUT HERE] 5.1.3. Tests with a Matched Sample Tables VIII and IX look at the cross-sectional spread differences in the secondary market between matched Fannie and Freddie bonds. For the match, both bonds need to trade on the same day, and have a duration difference of less than one month. Using a simple difference-indifference approach as described above for secondary bond spreads, I find that the difference in spreads (i.e. Fannie Mae — Freddie Mac) tightened after the event. This is evident in Table VIII, by noting the negative and significant coefficient of the event dummy. [TABLE VIII — ABOUT HERE] Using a 30-day window, I also find, in a specification similar to the spread-issuance setup above, that the spread difference is positively associated with the difference in issuance volume 20

after the event (at a 5% level), as shown in Table IX. So, this too provides some additional support to the hypothesis that relative issuance volume reductions are associated with some residual tightening of spreads. The equity price difference and the implied volatility difference turn out to be important too, with the expected signs. [TABLE IX — ABOUT HERE] 5.2. The Supply Effect in Primary Bond (Issuance) Spreads Table X repeats this analysis for the primary market, and finds very similar evidence. The coefficient of interest (i.e. of the interaction variable) is insignificant in all cases in the rights panel (as it should be) and significant at 5% post-event after the supply shock (in the left panel). This also validates the findings reported earlier that, subsequent to the announcement regarding Fannie Mae’s capital adequacy on Dec 21, 2004, there appears to be a strong positive relation between issuance volume and spreads of Fannie Mae, relative to Freddie Mac. Also, the point estimates are much larger here compared to the secondary markets, close to 17 basis points after the event. However, it is to be noted that, since we observe every bond here only once at issuance, the identification is not from within bond variation before and after the shock, as in the secondary market. To that extent, unobserved bond-level characteristics could possibly be driving these results, although firm fixed effects and event dummies have been included. Also, Boudoukh et. al. (1997) show that the optionality in mortgage-backed securities is adequately captured by the level and slope of the term structure. Since these variables have been included, along with other controls, it is not obvious what, aside from the supply effect, could be affecting these results. [TABLE X — ABOUT HERE] A similar regression (not reported) using primary issuance data from 1998 to 2004 was also run, to examine, whether this positive association shows up in longer time series. As expected there is no positive relation between the period issuance volume and spreads for Fannie relative to Freddie. A negative significant coefficient for δ1 indicates that there is probably some price impact associated with issuance. 5.3. Relative Supply Effect on Different Parts of the Term Structure So far, this paper has indicated that prices of Fannie Mae bonds rose relative to Freddie Mac bonds, following an announcement by the regulator asking Fannie Mae to increase its capital adequacy. Moreover, the joint movements of prices and issuance are exactly what one would 21

expect if it was a supply shock driving this change. Now, the main reason for the supply change, as detailed in the background, was the big duration mismatch in Fannie Mae’s books, where it was funding long-term mortgage investments by raising money in the short-term market. While this increased the returns, it also exposed the firm to huge interest rate risks. So, it is likely that the firm would need to reduce its short-term debt more than its long-term debt to satisfy its regulators. This was also explicit in the expert testimonies to Congress on this issue. The data in Table II shows that, from 2004 to 2005, while Freddie Mac kept its short term debt constant, Fannie Mae’s short term debt declined by about 50%. So, one would expect the supply effect to be much stronger in the shorter end than the long end. To test this empirically, I create a dummy variable (ShortDur) to flag all bonds with a duration of less than one year. I then interact this dummy variable with the earlier interaction term F NMDum ∗ volmain30 (i.e. a triple interaction) to check the differential price impact

on this short-term debt category. This is documented in columns 3 and 6 in Tables VI and VII, for demand and supply shock events. I find that the incremental impact is about as large as the main supply effect in Table VI, suggesting that spreads of these bonds tightened twice as much. However, it does not drive out the main supply effect, nor weaken it significantly. This ties in with the issuance data in Table II, where we observe that there was a decline in issuance of long —term bonds for Fannie Mae relative to Freddie Mac, but the relative decline in short-term outstanding was far larger. I find a supply effect in this segment in all time periods except before the fraud announcement, suggesting that the market anticipated a (separate) reduction in supply of the short-term securities because of the duration mismatch uproar, independent of the undercapitalization announcement. While the magnitudes of the incremental supply effect are relatively large in many periods (and tiny but significant in others), note that this is relative to the main supply effect (because of the triple interaction). This main supply effect is significantly negative, zero or positive depending on the event date and the window. I, however, do not find an incremental effect for the shorter duration bonds when I run a similar specification with the matched sample (see columns 3 and 6 in Table IX). Again, as a robustness check, I use alternative definitions of ShortDur. I get exactly the same patterns (not reported). 5.4. Evidence from the Credit Default Swap market I then look at the effect of this event on the credit default swap (CDS) market, and consider the 5-year spreads on Fannie Mae and Freddie Mac. I test this hypothesis by replicating some 22

of the above analysis with CDS spreads data (Data constraints do not permit me to replicate the full exercise). Apriori, it may appear ambiguous which way the CDS spreads are likely to move. The CDS is a synthetic contract created by a dealer, in contrast to bonds, which are physical securities issued by the firm. On one hand, CDS, being dealer-created synthetic securities, should be immune to a supply shock that affects Fannie bonds (the underlying in this case). Then again, it appears that the cash flows from being a protection seller and holding a long position in the risk free asset are the same as those from a long position in a Fannie bond. So, one may expect a portfolio that is long in a Fannie bond and short in a risk free bond to be priced at the same spread as the CDS. However, violation of this price does not necessarily permit an arbitrage. First, this trade involves borrowing at the risk free rate, which is not possible for ordinary market participants. Second, the CDS pays par upon the relevant credit event (which occurs at an uncertain time), whereas the other securities have clearly specified maturities. Finally, conversations with CDS traders indicated that the dealers typically hedge their positions in CDS using a CDS index and historical correlations of the index and the name (instead of a position in the underlying). All these factors weaken the link between CDS spreads and the spreads of the underlying bonds. However, the probability of default of the name is a key ingredient in pricing both the bond and the CDS, and so the CDS and the bond spreads generally move together. A supply shock (unrelated to changes in probability of default) is one instance when they may not. Table XI replicates the earlier difference-in-difference exercise with a 10-day window using CDS spreads instead of bond spreads. If there were a negative and significant coefficient of the interaction between firm and event dummy, I could have concluded the event had caused a supply effect (i.e. tightening in this case) on the CDS spreads. However, as expected, this is not the case. I also look at the bond issuance volume sensitivity to the CDS spreads (not reported). Again, none of the variables are significant. While this may have a lot to do with the relatively low power of this test for the CDS sample (there are fewer observations in the CDS dataset), I regard this as tentative evidence that no supply effect was seen in the synthetic dealer-created security, as conjectured. [TABLE XI — ABOUT HERE] 5.5. A comment about the magnitudes As mentioned earlier, the objective of this paper is to document the existence of a supply effect, rather than focus on its magnitude. This makes regression the ideal technique for studying this 23

question. While such regressions usually provide some indication of magnitude as well, there are several reasons why there is a downward bias in the data in this case. First, I obtain the demand-side variables from prior literature, which has not been very successful in explaining high grade bond spreads. This unexplained residual demand is likely to significantly dampen my findings. Second, this effect is measured relative to Freddie Mac. Since Freddie Mac is otherwise a similar company, a supply shock in Fannie Mae is likely to cause investors to push up demand for Freddie Mac bonds too, thereby reducing the relative difference between the two bonds (if they were perfect substitutes, there would be absolutely no relative effect). Next, the spread difference between Fannie Mae and Freddie Mac is not a large number, and is usually close to 0-1 basis point, according to market participants. Moreover, the other variables in the regressions are not explaining a particularly large proportion of the variation in the dependent variable. The small magnitude needs to be viewed in this context, and may not be representative of an economically negligible effect. Further, at first glance, the magnitudes of the coefficients δ 1 - δ 4 may be difficult to interpret. To aid this, one may note that the volume of issuance is denoted in millions in my dataset, and the average 30-day issuance around the supply shock is over five billions dollars (i.e. of the order 103 times the unit of measurement). So multiplying the coefficient by 1000 will give us the point estimate of the spread movement in basis points for one billion dollars of issuance. In the case of δ 4 , the relative spread effectively changes by over 1 basis point for 1 billion of issuance after the supply shock in the 30-day window in the secondary market, after incorporating this, which seems appropriate and in line with a moderate effect. The shortterm bonds increase by a further 0.8 basis point for a billion dollars of issuance. To put these numbers in context, from Table II, Fannie Mae issued about $155 billion of long-term debt in 2005, in addition to about 2.8 trillion dollars of short-term debt. The total outstanding debt of Fannie Mae on December 31, 2004 was 941 billion dollars, according to the Bond Market Association. These numbers, along with the fact that the average Fannie-Freddie spread was about 4 basis points during that time (of course, this depends on duration; this number is an unconditional mean) implies a high demand elasticity. This is striking, considering that most prior research ignores demand elasticity (and consequently the role of supply) while modeling asset prices. The movement in the primary market is larger, at about 17 basis points. The difference in magnitudes across the two markets is an interesting fact, and should be explored further for robustness in other asset classes.

24

6. Conclusion This paper documents that changes in issuer propensity to supply securities affect the prices of financial assets, using multiple data sets and a range of specifications. By using an unanticipated supply shock in the Agency markets and looking across asset maturities, I show that asset groups with larger supply shocks have bigger effects per unit volume. These effects are large considering the relative magnitudes in the Agency market. Moreover, in empirical specifications that do not manage to identify the other price influencers consistently (in line with prior research), the supply effect still stands out. This line of research may be extended in a number of directions. Most obviously, the same analyses can be applied to markets for other securities. For instance, Dastidar (2007) presents some evidence on the discontinuation of the 30-year Treasury bonds in 2001. Unlike this event, the magnitude of change after the Treasury discontinuation was much more pronounced, with yields on long-term bonds moving as much as 18 basis points in a single day. Also, using the same event dates, one may examine the effect of the supply glut in the mortgage-backed securities market following the Fannie Mae capital requirement in December 2004, which necessitated heavy selling of mortgages by Fannie to reduce the assets it held on its balance sheet. This is a potentially cleaner test, because the credit of Fannie Mae is much less important in the mortgage-backed securities market, since Fannie’s credit is relevant only if homeowners default. This has important policy implications regarding the drivers of mortgage rates. It may also make sense to look at the issuance volume variables net of redemption. This would require availability of high frequency redemption data. Also, since redemptions are perfectly predictable (assuming no unexpected calls), one might argue that that additional information may not be present in that data. While a supply effect has been documented here, this paper has not directly addressed whether supply is systematically priced. The supply effect I document varies both across time and in the cross section, since it has not been absorbed by time-invariant bonds fixed effects or day fixed-effects, and varies across the term structure depending on maturity. This provides some hope that this factor exists, since it goes beyond mere cross sectional (time invariant) differences or unobserved time series shocks. In developing a proxy for this factor, it will be useful to look for something that is orthogonal to current demand, yet has a good

25

predictive power for future issuance volume; this is an important topic for further work. The role of supply shocks also provides possible alternative explanations for some asset pricing results; for example, some instances of time-varying risk premia could instead have arisen from changes in issuer propensity to supply assets. Again, an increase in the supply of Treasury securities may be a reason why sustained government deficits lead to an increase in long-term rates (Dai and Philippon, 2004). However, the key in these situations will be once again to disentangle the demand and supply effects, since issuer propensity to supply is very likely to be correlated with investor demand. This too remains an avenue for further research.

26

References Amato, J., and E. Remolona, 2003, The Credit Spread Puzzle, BIS Quarterly Review, December 2003. Anderson, R.W., and S. M. Sundaresan, 1996, Design and Valuation of Debt Contracts, Review of Financial Studies 9(1), 37-68. Black, F., and M. Scholes (1973). "The Pricing of Options and Corporate Liabilities". Journal of Political Economy 81 (3): 637-654 Boudoukh, Jacob, Matthew Richardson, Richard Stanton, and Robert Whitelaw, 1997, Pricing mortgage-backed securities in a multifactor interest rate environment: A multivariate density estimation approach, Review of Financial Studies 10, 405—46. Chen, H., G. Noronha, and V. Singal, 2004, The Price Response to S&P 500 Index Additions and Deletions: Evidence of Asymmetry and a New Explanation, Forthcoming, Journal of Finance. Cohen, L., Deither and C. Malloy, 2006, Supply and Demand Shifts in the Shorting Market, Journal of Finance, Forthcoming Collin-Dufresne, P., R. S. Goldstein, and J. S. Martin, 2001, The Determinants of Credit Spread Changes, Journal of Finance 56, 2177 — 2207. Chaudhury, S and O. Ravi, 2005, Outlook for the Mortgage Market 2005, Bank of America Securities Mortgage-Backed Research Dastidar, S., 2007, Effect of a financial asset supply shock on prices: Evidence from Treasury Markets, Columbia University Working Paper. Dai, Q. and T. Philippon, 2004, Fiscal Policy and the Term Structure of Interest Rates, Working Paper Duffie, D., and K. J. Singleton, 1997, An Econometric Model of the Term Structure of InterestRate Swap Yields, Journal of Finance 52, 1287—1321. Duffie, D., and K. J. Singleton, 1998, Modeling term structures of Defaultable Bonds, Review of Financial Studies 12(4), 687—720. Duffie, D., and K. J. Singleton, 2003, Credit Risk: Pricing, Measurement, and Management, 27

Princeton University Press Fisher, E., R. Heinkel, and J. Zechner (1989). Dynamic Capital Structure Choice: Theory and Tests. Journal of Finance 44, 19—40. Fridson, M., and J. Jonsson (1995). Spread Versus Treasuries and the Riskiness of High-Yield Assets. Journal of Fixed Income, December 1995, 79—85. Goldstein, R, N. Ju and H. Leyland (2001). An EBIT-based model of Dynamic Capital Structure. Journal of Business, Vol. 74, No. 4, 483-512 Harris, L., and E. Gurel, 1986, Price and Volume Effects Associated with Changes in the S&P 500, Journal of Finance 41, 815—829. Heidari, Massoud and Wu, Liuren, 2004, "What Constitutes a Good Model? An Analysis of Models for Mortgage Backed Securities", Working paper Holthausen, R. W., R. W. Leftwich, and D. Mayers, 1990, Large-Block Transactions, the Speed of Response and Temporary and Permanent Stock-Price Effects, Journal of Financial Economics 26, 71—95. Huang, J. and M. Huang, 2003, How Much of the Corporate-Treasury Yield Spread is Due to Credit Risk? Working Paper, Stanford University Huang, J. and W. Kong, 2003, Explaining Credit Spread Changes: Some New Evidence from Option-Adjusted Spreads of Bond Indexes, Journal of Derivatives Jaffee, D., 2003, The Interest Rate Risk of Fannie Mae and Freddie Mac, Journal of Financial Services Research 24:1, 5-29. Leland, H. (1994). Corporate Debt Value, Bond Covenants, and Optimal Capital Structure, Journal of Finance 49, 1213—1252. Longstaff, Francis, S. Mithal and E. Neis, 2005, Corporate Yields Spreads: Default Risk or Liquidity? New Evidence from the Credit-Default Swap Market, Journal of Finance 60, 2213-2253. Merton, R.C. (1974), On the Pricing of Corporate Debt: The Risk Structure of Interest Rates, Journal of Finance 29, pp. 449-470. Newman, Yigal S., and Michael A. Rierson, 2004, Illiquidity Spillovers: Theory and Evidence 28

from European Telecom Bond Issuance, Working paper Sarig, Oded and A. Warga, Bond Price Data and Bond Market Liquidity, The Journal of Financial and Quantitative Analysis, Vol. 24, No. 3. (Sep., 1989), pp. 367-378. Scholes, M. S., 1972, The Market for Securities: Substitution vs. Price Pressure and the Effects of Information on Share Prices, Journal of Business 45, 179—211. Shleifer, A., 1986, Do Demand Curves for Stocks Slope Down?, Journal of Finance 41, 579— 590. Stiglitz, J. E, 1972, On the Optimality of the Stock Market Allocation of Investment, The Quarterly Journal of Economics, Vol. 86, No. 1. (Feb. 1972), pp. 25-60. Sundaresan, S., 2002, Fixed Income Markets and their Derivatives (Second Edition), South Western Thomson Learning Wurgler, J., and E. Zhuravskaya, 2002, Does Arbitrage Flatten Demand Curves for Stocks?, Journal of Business 75, 583—608.

29

30

0.15

0.1

10/25/2004

75

10/18/2004

10/11/2004

10/4/2004

9/27/2004

9/20/2004

9/13/2004

9/6/2004

8/30/2004

8/23/2004

8/16/2004

8/9/2004

8/2/2004

8/ 2/ 20 04 8/ 9/ 20 04 8/ 16 /2 00 4 8/ 23 /2 00 4 8/ 30 /2 00 4 9/ 6/ 20 04 9/ 13 /2 00 4 9/ 20 /2 00 4 9/ 27 /2 00 4 10 /4 /2 00 10 4 /1 1/ 20 04 10 /1 8/ 20 04 10 /2 5/ 20 04

Figure 1

Implied Vol of FNM and FRE around Sep 21 2004

0.45

0.4

0.35

0.3

0.25

0.2

FNM

FRE

0.05 Date

Figure 2

Equity Prices of FNM and FRE around Sep 21 2004

80

FNM

FRE

70

65

60

Date

11 /1 /2 00 4 11 /8 /2 00 4 11 /1 5/ 20 04 11 /2 2/ 20 04 11 /2 9/ 20 04 12 /6 /2 00 4 12 /1 3/ 20 04 12 /2 0/ 20 04 12 /2 7/ 20 04 1/ 3/ 20 05 1/ 10 /2 00 5

31 04

04

5

68

/2 00

20 05

/2 0

/2 0

04

4

04

04

20 0

/2 0

1/ 3/

/2 7

/2 0

/1 3

1/ 10

12

12

12

/2 0

/2 0

4

4

04

20 0

20 0

/2 0

/6 /

/2 9

12

11

/2 2

/1 5

/8 /

/1 /

$

11

11

11

11

Figure 3

FNM & FRE Equity Prices around Dec 21 2004

76

74

72

70

FNM

FRE

66

64

0.14

0.1 Date

Figure 4

FNM & FRE Imp Vol around Dec 21 2004

0.3

0.26

0.22

0.18

FNM

FRE

Table I: Fixed Income Market Size by Asset Class Asset Class

Outstanding

Long-term bond Issuance in

($ billion)

2004 ($ billion)

Treasuries

3943.6

853.3

Corporates

4704.5

711.0

Money market

2872.1

Asset-Backed

1827.8

896.6

Mortgage-related

5472.5

1760.0

Municipal

2018.6

360.3

Agencies

2745.1

896.7

Source: Bond Market Association, February 2005

Table II: Primary market Debt Activity of Fannie Mae and Freddie Mac 2003

2004

2005

Freddie Mac

277.1

199.2

172.7

Fannie Mae

347.8

252.2

155.4

Freddie Mac

189.0

181.1

183.4

Fannie Mae

346.0

320.3

172.5

Long-term Debt Issuance ($bn)

Short-term Debt Outstanding ($bn)

Source: Bond Market Association, February 2004, 2005

32

Table III: Summary Statistics (+/-) 30 days around Supply Shock Fannie Mae Mean Std. Dev 69.95 1.17 0.22 0.01 2.31 1.01 -0.06 0.04 394.82 721.20 6977.35 5540.32 820.31 1012.77 18.33 1.37 480

Variable Equity Price Implied Vol Duration P(Jump)** Daily Issuance Volume Issuer Period Vol Bond Issuance Volume CDS Spreads* Observations

Variable S&P Value r2 (r10 – r2) r102 Bond Flows Corp Spread * 12 observations for event study **Smirk of implied volatility curve (Collin-Dufrense et.al. 2001)

33

Market Mean Std. Dev 1768.94 18.16 305.64 9.99 118.50 10.27 179949.90 5988.54 1918.24 923.93 58.58 8.59

Freddie Mac Mean Std. Dev 70.00 1.82 0.19 0.01 3.86 2.41 -0.05 0.04 215.87 610.16 3616.22 1814.91 767.97 421.16 14.39 1.14 320

Table IV: List of Variables Variable Name

Description

Equity Price S&P Price r2 (r10 – r2) 2 r10 Bond Flows Duration Implied Vol Corp Spread

The price of the issuer’s stock on that day, in dollars The price of the S&P500 on that day The two-year Treasury rate, in basis points The slope of the Treasury yield curve, in basis points The convexity of the yield curve, , in basis points squared The net bond mutual fund inflows in the previous month, in millions The duration of the bond, measured in years The Black Scholes implied volatility on the trade date (1% = 0.01) The spread of duration-matched corporate bonds over Treasury, expressed in basis points The slope of the smirk in the implied volatility curve as described in Collin-Dufrense, Goldstein and Martin (2001), used as a proxy for probability of large negative jumps in firm value Fannie Mae dummy A dummy variable that takes the value 1 after the event, 0 otherwise Interaction of the FNMDummy and the EventDummy The issuance volume for the current bond, in $million The issuance volume over the past and next 2 weeks for the competitor (i.e. non-issuer) in $million The issuance volume of the issuer for past and next two weeks The interaction of the FNMDummy and Issuer Period Volume Dummy Variable that takes the value 1 if duration < 1 year

P(Jump) FNMDummy PostEventDummy Event*FNMDummy Issuance Volume Comp Period Vol Issuer Period Vol FNM*Issuer Period Vol ShortDur Δ(Any of the above variables) FNM-FRE(any of the above variables)

Change in the relevant variable over one day Difference in the values of the relevant variables of Fannie Mae and Freddie Mac’s matched bonds

34

Table V: Difference-in-difference regressions around supply shock using bond spreads This table displays regression results for the secondary market trades of Fannie Mae and Freddie Mac fixed rate bonds around December 21, 2004 (when Fannie Mae undercapitalization was announced). The dependent variable is the spread of the bond over a duration-matched treasury in basis points on the trade date. Please refer to Table IV for the definitions of the explanatory variables. Robust standard errors clustered by bond are shown in parentheses below each coefficient estimate. 90%, 95% and 99% significance levels are denoted by one, two or three asterisks next to the relevant coefficients. 5.61

5.61

5.61

(9.102e-01)*** 1.03 (3.303e-01)*** 7.97 (4.393e+00)* -0.47

(9.118e-01)*** 1.22 (3.213e-01)*** 7.28 (4.426) -0.80

(9.221e-01)*** 0.30 (0.395) 7.33 (4.427) -1.15

(0.446) Equity Price

(3.635e-01)** -0.03

(3.795e-01)*** -0.29

Implied Vol

(0.104) 18.71

(1.678e-01)* 12.18

13.52

(7.551e+00)** 12.18

(7.629) 31.87

(5.474e+00)**

(10.140)

(1.487e+01)**

552 0.63 No

552 0.63 No

552 0.63 Yes

Duration PostEventDummy FNMDummy Event*FNMDummy

Constant Observations R-squared Day Fixed Effects

35

Table VI: Secondary Market Regression Results Around Supply Shock (30-day window) This table displays regression results for the secondary market trades of Fannie Mae and Freddie Mac fixed rate bonds during a 30-day window before and after December 21, 2004 (when Fannie Mae undercapitalization was announced). The dependent variable is the difference between the yield on the bond and the yield on a duration-matched treasury, expressed in basis points. Please refer to Table IV for the definitions of the explanatory variables. Robust standard errors are shown in parentheses below each coefficient estimate. Coefficients marked with one, two or three asterisks are significant to the 90%, 95% and 99% level respectively. 30-day window Pre-event Post-event (1) (2) (3) (4) (5) (6) -25.60 -27.48 -17.44 -8.83 -5.41 -26.55 Δ Equity Price (24.970) (20.310) (18.520) (32.860) (18.770) (26.830) 57.72 31.78 -12.13 -33.51 -29.65 29.09 Δ S&P 500 (60.390) (58.950) (50.150) (49.500) (42.200) (45.310) 4.77 8.57 8.01 28.25 37.25 30.13 r2 (2.881) (3.174e+00)** (3.169e+00)** (1.142e+01)** (8.292e+00)*** (1.220e+01)** 4.96 8.63 7.96 28.54 37.35 30.12 (r10 – r2) (2.873e+00)* (3.153e+00)** (3.141e+00)** (1.136e+01)** (8.279e+00)*** (1.218e+01)** -0.006 -0.010 -0.009 -0.033 -0.044 -0.035 r102 (0.0034) (3.739e-03)** (3.728e-03)** (1.331e-02)** (9.713e-03)*** (1.428e-02)** 0.0014 0.0009 0.0002 -0.0002 0.0001 -0.0002 Bond Flows (4.017e-04)*** (0.0006) (0.0006) (0.0011) (0.0008) (0.0015) 18.21 -28.43 -26.51 -17.91 -18.40 -22.06 Implied Vol (27.610) (36.570) (37.910) (15.900) (7.841e+00)** (1.062e+01)* -0.0043 -0.0041 -0.0023 0.0062 0.0062 0.0060 Issuance Volume (1.938e-04)*** (2.104e-04)*** (2.696e-04)*** (2.612e-04)*** (3.108e-04)*** (2.930e-04)*** 0.0255 0.0222 0.0114 0.1724 0.1736 0.1756 Corp Spread (0.055) (0.050) (0.058) (6.240e-02)** (8.021e-02)** (8.247e-02)** -22.08 -22.17 -21.06 -20.23 -20.30 -19.64 Duration (3.093e-01)*** (3.063e-01)*** (3.056e-01)*** (5.758e-01)*** (5.380e-01)*** (5.970e-01)*** 4.76 4.28 3.52 -9.08 -2.23 -4.53 P(Jump) (6.249) (5.324) (3.950) (7.442) (5.596) (8.055) 0.0018 0.0017 0.0002 0.0001 Daily Volume (8.467e-04)* (7.384e-04)** (0.00019) (0.00023) 0.0010 0.0013 -0.0013 -0.0016 Comp Period Volume (4.317e-04)** (4.530e-04)*** (3.385e-04)*** (3.949e-04)*** 0.0005 0.0003 0.0002 0.0001 Issuer Period Volume (0.00062) (0.00055) (8.201e-05)* (0.00012) -0.0005 -0.0008 0.0010 0.0012 FNM*Issuer Period Vol (0.00066) (0.00056) (2.981e-04)*** (3.290e-04)*** 0.0004 0.0008 FNM*Issuer Period Vol*ShortDur. (2.020e-04)*** (1.451e-04)*** -850.10 -1654.00 -1542.00 -5786.00 -7708.00 -6191.00 Constant (609.000) (6.711e+02)** (6.686e+02)** (2.444e+03)** (1.771e+03)*** (2.604e+03)** 466 466 466 334 334 334 Observations 0.37 0.37 0.55 0.35 0.36 0.39 R-squared Yes Yes Yes Yes Yes Yes Bond Fixed Effects

36

Table VII: Secondary Market Regression Results Around Demand Shock (30-day window) This table displays regression results for the secondary market trades of Fannie Mae and Freddie Mac fixed rate bonds during a 30-day window before and after September 21, 2004 (when Fannie Mae fraud was announced). The dependent variable is the difference between the yield on the bond and the yield on a duration-matched treasury, expressed in basis points. Please refer to Table IV for the definitions of the explanatory variables. Robust standard errors are shown in parentheses below each coefficient estimate. Coefficients marked with one, two or three asterisks are significant to the 90%, 95% and 99% level respectively. 30-day window Pre-event Post-event (1) (2) (3) (4) (5) (6) -2.82 -2.70 -2.37 -29.35 -32.57 -28.48 Δ Equity Price (26.410) (34.920) (35.240) (1.224e+01)** (8.949e+00)*** (9.766e+00)*** 74.99 70.29 69.72 76.16 71.04 59.89 Δ S&P 500 (52.490) (48.970) (49.210) (44.420) (3.195e+01)** (34.940) -1.00 -2.22 -2.23 -6.29 -3.26 -3.47 r2 (4.460) (4.601) (4.621) (4.429) (3.471) (3.396) -0.43 -1.64 -1.65 -5.94 -3.02 -3.24 (r10 – r2) (4.421) (4.565) (4.585) (4.417) (3.458) (3.383) 0.0014 0.0028 0.0029 0.0081 0.0043 0.0046 r102 (0.0053) (0.0055) (0.0055) (0.0053) (0.0042) (0.0041) -0.00047 -0.00050 -0.00050 0.00362 0.00013 0.00030 Bond Flows (1.530e-04)*** (1.628e-04)*** (1.632e-04)*** (9.281e-04)*** (0.0010) (0.0009) -13.75 -32.66 -32.70 1.13 8.85 9.80 Implied Vol (24.300) (25.780) (25.860) (17.610) (11.500) (11.000) -0.0131 -0.0131 -0.0136 0.0008 0.0010 0.0528 Issuance Volume (4.460e-04)*** (4.383e-04)*** (7.734e-04)*** (0.001) (4.723e-04)** (1.738e-03)*** -0.60 -0.60 -0.60 -0.50 -0.62 -0.55 Corp Spread (1.644e-01)*** (1.611e-01)*** (1.616e-01)*** (2.097e-01)** (1.347e-01)*** (1.545e-01)*** -32.13 -32.35 -32.35 -30.95 -31.08 -30.95 Duration (6.166e-01)*** (6.821e-01)*** (6.836e-01)*** (3.577e-01)*** (3.374e-01)*** (3.594e-01)*** -6.74 -3.62 -3.70 -8.42 -7.19 -6.34 P(Jump) (10.350) (10.210) (10.280) (7.089) (4.890) (4.651) -8.34E-05 -7.97E-05 -6.40E-04 -6.42E-04 Daily Volume (0.00034) (0.00034) (1.404e-04)*** (1.400e-04)*** -2.91E-05 -2.95E-05 -5.68E-04 -5.46E-04 Comp Period Volume (0.00030) (0.00030) (1.289e-04)*** (1.260e-04)*** 2.73E-04 2.74E-04 -5.80E-04 -5.65E-04 Issuer Period Volume (0.00029) (0.00029) (1.218e-04)*** (1.231e-04)*** -3.49E-04 -2.68E-04 1.20E-04 -1.59E-05 FNM*Issuer Period Vol (0.00030) (0.00029) (0.00023) (0.00022) -9.55E-05 1.84E-04 FNM*Issuer Period Vol*ShortDur. (0.00011) (1.051e-04)* 321.20 574.40 577.70 1318.00 745.10 769.50 Constant (926.000) (953.900) (958.200) (913.500) (711.400) (696.900) 305 305 305 323 323 323 Observations 0.78 0.78 0.78 0.67 0.68 0.68 R-squared Yes Yes Yes Yes Yes Yes Bond Fixed Effects

37

Table VIII: Matched Sample Difference-in-difference regressions around the Supply Shock This table displays regression results for the secondary market trades of Fannie Mae and Freddie Mac fixed rate bonds for +/- 10 days around December 21, 2004 (when Fannie Mae undercapitalization was announced). The dependent variable is the difference between the durationmatched spread over Treasury for a Fannie Mae bond and a matched Freddie Mac bond, expressed in basis points. Please refer to Table IV for the definitions of the explanatory variables. Robust standard errors are shown in parentheses below each coefficient estimate. 90%, 95% and 99% significance levels are denoted by one, two or three asterisks next to the relevant coefficients. -2.61

-0.38

Constant

(0.709)*** 117.20 (37.533)*** -45.55 (108.881) 6.15 (3.461) -21.51

(0.091)*** 59.18 (20.276)** 3.71 (11.973) -1.18 (0.269)*** -123.46

Observations

(4.048)*** 196

(1.622)*** 196

0.02 No

1.00 Yes

(FNM-FRE)Equity Price (FNM-FRE)Duration (FNM-FRE)Implied Vol PostEventDummy

R-squared Bond-Pair Fixed Effects

38

Table IX: Regressions around the Supply Shock with the Matched Sample data This table displays regression results for the secondary market trades of Fannie Mae and Freddie Mac fixed rate bonds during a 30-day window before and after December 21, 2004 (when Fannie Mae undercapitalization was announced). The dependent variable is the difference between the duration-matched spread over Treasury for a Fannie Mae bond and a matched Freddie Mac bond, expressed in basis points. Please refer to Table IV for the definitions of the explanatory variables. Robust standard errors are shown in parentheses below each coefficient estimate. Coefficients marked with one, two or three asterisks are significant to the 90%, 95% and 99% level respectively. 30-day window Pre-event Post-event (1) (2) (3) (4) (5) (6) -0.59 -0.45 -0.50 -0.37 -0.31 -0.31 (FNM-FRE)Equity Price (0.145)*** (0.162)** (0.170)*** (0.217) (0.111)** (0.112)** -0.013 -0.008 -0.010 0.014 0.013 0.014 S&P Value (0.008) (0.007) (0.007) (0.015) (0.017) (0.017) -1.81 -2.55 -2.48 -12.94 -13.41 -13.37 r2 (1.137) (1.259)* (1.453) (10.227) (6.567)* (6.616)* -1.85 -2.58 -2.51 -13.07 -13.49 -13.44 r10 – r2 (1.117) (1.256)* (1.450)* (10.214) (6.561)* (6.610)* 2 2.00E-03 3.00E-03 3.00E-03 1.50E-02 1.60E-02 1.60E-02 r10 (0.001)* (0.001)** (0.002)* (0.012) (0.008)* (0.008)* -9.98E-04 -9.46E-04 -1.02E-03 7.37E-04 9.94E-04 9.99E-04 Bond Flows (1.975e-04)*** (1.855e-04)*** (2.107e-04)*** (0.001) (0.001) (0.001) -0.30 -0.31 -0.35 0.54 0.55 0.56 Corp Spread (0.106)** (0.099)*** (0.108)*** (0.154)*** (0.160)*** (0.161)*** 5.45 7.55 10.91 103.22 51.02 84.35 (FNM-FRE)Duration (15.661) (15.068) (16.963) (132.496) (132.485) (108.600) (FNM-FRE)P(Jump) (FNM-FRE)Implied Vol

-4.98 (1.395)*** 12.61 (5.421)**

-5.05 (1.477)*** 5.68 (8.864) 4.72E-04 (2.070e-04)** 2.96E-04 (0.0002)

387.32 (242.179) 300 0.18 Yes

(FNM-FRE)Daily Volume (FNM-FRE)Period Volume

2.37 (1.657) -5.03 (9.218)

1.08 (2.120) -0.99 (6.322) 2.09E-04 (1.093e-04)* 1.28E-04 (5.908e-05)**

540.01 (272.802)* 300 0.18

-5.33 (1.614)*** 5.94 (9.927) 6.28E-04 (2.808e-04)** 4.89E-04 (0.0003) -5.78E-04 (0.0004) 527.00 (313.900) 300 0.21

2731.34 (2186.816) 297 0.18

2834.56 (1400.190)* 297 0.18

1.16 (2.123) -0.85 (6.311) 2.14E-04 (1.077e-04)* 1.34E-04 (6.189e-05)** -1.80E-05 (0.00003) 2825.00 (1.410e+03)* 297 0.18

Yes

Yes

Yes

Yes

Yes

(FNM-FRE)Period Volume*ShortDur Constant Observations R-squared Bond Pair Fixed Effects

39

Table X: Regressions around Supply and Demand Shocks with Primary Market Data This table displays regression results for the primary market issuances of Fannie Mae and Freddie Mac fixed rate bond issuances. The left panel shows regressions around the supply shock (undercapitalization announcement). The right panel shows regressions around the demand shock (fraud announcement). The dependent variable is the difference between the yield on the bond and the yield on a duration-matched treasury, expressed in basis points. Please refer to Table IV for the definitions of the explanatory variables. Robust standard errors are shown in parentheses below each coefficient estimate. 90%, 95% and 99% significance levels are denoted by one, two or three asterisks next to the relevant coefficients. Supply Shock (30-day window) Demand Shock(30-day window) Pre-event Post-event Pre-event Post-event 377.53 189.378 372.324 483.464 392.732 903.934 -176.973 -278.698 Equity Return (653.407) (937.381) (693.223) (822.369) (361.290) (484.764)* (429.441) (450.544) -432.157 -635.511 -483.164 -472.604 271.055 -15.848 810.763 655.996 S&P Return (547.630) (619.867) (835.589) (885.597) (319.645) (374.060) (625.821) (720.136) -160.09 -127.27 155.697 -4.333 -99.943 -53.413 118.232 120.029 r2 (103.542) (105.987) (694.567) (733.841) (90.381) (97.620) (68.326)* (69.177)* -161.192 -128.643 155.868 -4.482 -100.766 -55.312 116.303 118.04 r10 – r2 (103.762) (106.468) (693.875) (732.451) (89.476) (96.572) (67.452)* (68.291)* 0.188 0.148 -0.184 0.004 0.12 0.064 -0.144 -0.147 r102 (0.121) (0.125) (0.817) (0.863) (0.107) (0.116) (0.083)* (0.084)* 0.029 0.011 -0.003 -0.001 0.018 0.029 -0.004 0.008 Bond Flows (0.019) (0.023) (0.006) (0.005) (0.013) (1.617e-02)* (0.025) (0.030) -6.282 -5.402 0.189 0.053 4.067 3.831 5.085 4.291 Duration (4.113) (6.126) (3.294) (3.632) (3.279) (3.951) (3.624) (4.025) -444.949 -90.587 217.978 104.573 72.113 203.117 -99.31 -29.407 Implied Vol (674.308) (767.608) (513.460) (534.346) (397.313) (443.649) (280.416) (303.133) 0.439 0.403 0.254 0.273 0.11 0.11 0.076 0.103 Corp Spread (0.202)** (0.310) (0.216) (0.250) (0.133) (0.163) (0.148) (0.167) -122.899 -49.489 -77.846 -55.519 101.959 14.985 114.808 121.936 P(Jump) (123.503) (131.272) (137.807) (132.381) (151.706) (171.416) (124.529) (131.163) 13.097 -101.932 -2.583 30.347 -0.413 -25.665 0.665 -13.957 FNMDummy (17.364) (66.651) (17.757) (47.498) (10.618) (53.609) (15.425) (44.419) -0.002 -0.011 -0.012 -0.014 Volume (0.017) (6.306e-03)* (0.012) (0.011) 0.017 0.006 -0.005 -0.001 Comp Period Volume (8.121e-03)** (0.003) (0.005) (0.003) -0.007 -0.019 -0.004 -0.003 Issuer Period Volume (0.008) (8.771e-03)** (0.007) (0.003) 0.023 0.017 0.002 0.001 FNM*Issuer Period Vol -1.610E-02 (7.408e-03)** (0.005) (0.005) 34255.18 27488.174 -33040 1152.059 20977.154 11370.682 -23822.91 -24204.181 Constant (22064.566) (22566.136) (147600.000) (155880.141) (18855.826) (20350.746) (13971.008)* (14147.096)* 95 95 88 88 130 130 129 129 Observations 0.22 0.25 0.3 0.35 0.32 0.34 0.39 0.4 R-squared

40

Table XI: Difference-in-difference regressions for CDS Spreads around Supply Shock This table displays regression results for the credit default swaps (CDS) of Fannie Mae and Freddie Mac around December 21, 2004 (when Fannie Mae undercapitalization was announced). The dependent variable is the 5-year CDS spread in basis points on the trade date. Please refer to Table IV for the definitions of the explanatory variables. Robust standard errors are shown in parentheses below each coefficient estimate. 90%, 95% and 99% significance levels are denoted by one, two or three asterisks next to the relevant coefficients. 0.93 -8.33E-01 4.30 (1.011157)*** 1.32 (0.9776)

3.23 (0.615930)*** 3.30 (0.461642)*** 0.44 (0.4269)

Equity Price

0.16 (0.2530)

0.02 (0.1278)

Implied Vol

-21.02 (24.8447)

10.19 (11.9284)

13.64 (0.357007)***

5.89 (19.0919)

9.30 (9.5354)

24 0.88

24 0.89

24 0.99

No

No

Yes

PostEventDummy FNMDummy Event*FNMDummy

Constant Observations R-squared Day Fixed Effects

1.50 (0.504884)*** 3.65 (0.504884)*** 0.57 (0.7140)

41