GSEs, Mortgage Rates, and Secondary Market Activities

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. GSEs, Mortgage Rates, and Secondary Market Activities Andreas Lehnert, Wayne Passmore, and Shane M. Sherlund 2006-30 NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

GSEs, Mortgage Rates, and ∗ Secondary Market Activities Andreas Lehnert WaynePassmore BoardofGovernorsofthe BoardofGovernorsofthe FederalReserveSystem FederalReserveSystem Washington, DC20551 Washington, DC20551 (202)452-3325 (202)452-6432 Andreas.Lehnert@frb.gov Wayne.Passmore@frb.gov Shane M.Sherlund BoardofGovernorsofthe FederalReserveSystem Washington, DC20551 (202)452-3589 Shane.M.Sherlund@frb.gov ThisVersion: September8,2006 Wewelcomeallcomments;pleasecontactusdirectlyforthelatestversion. ∗Revised version of FEDS 2005-7. Cathy Gessert provided excellent research assistance. We thank Ben Bernanke, Darrel Cohen, Karen Dynan, Kieran Fallon, Michael Fratantoni, Mike Gibson,DianaHancock,LarsPeterHansen,PaulKupiec,EllenMerry,SteveOliner,BobPribble,Tom Tallarini, andJonathan Wrightfor helpful commentsand suggestions onvariousdraftsofthispaper. Theopinions,analysis,andconclusionsofthispaperaresolelythoseoftheauthorsanddonot necessarilyreflectthoseoftheBoardofGovernorsoftheFederalReserveSystem.

GSEs, Mortgage Rates, and Secondary Market Activities Abstract Fannie Mae and Freddie Mac are government-sponsored enterprises (GSEs) that securitizemortgagesandissuemortgage-backed securities(MBS).Inaddition,the GSEs are active participants in the secondary mortgage market on behalf of their own investment portfolios. Because these portfolios have grown quite large, portfoliopurchases (inadditiontoMBSissuance)areoftenthoughttobeanimportant forceinthemortgagemarket. Usingmonthlydatafrom1993to2005weestimatea VARmodeloftherelationshipbetweenGSEsecondarymarketactivitiesandmortgageinterestratespreads. WefindthatGSEportfoliopurchaseshavenosignificant effects on either primary or secondary mortgage rate spreads. Further, we examine GSE activities and mortgage rate spreads in the wake of the 1998 debt crisis, and findthat GSEportfolio purchases did littletoaffect interest ratespaid bynew mortgage borrowers. This empirical finding is robust to alternative identification assumptions andtoalternative modelandvariablespecifications. Journal ofEconomicLiteratureclassification numbers: H81,G18,G21 Keywords: Mortgage finance, Government-Sponsored Enterprises, Financial stability

1 Introduction The housing-related government-sponsored enterprises (GSEs) Fannie Mae and Freddie Macsecuritize pools ofmortgages, thereby assuming their credit risk and allowing the resulting mortgage-backed securities (MBS) to trade as effectively AAA-rated securities. This process provides originators access to a liquid secondary market for their loans. Separately, the GSEsalso issue corporate bonds to finance large, highly leveraged, portfolios ofmortgages, often inthe form oftheir ownMBS. The GSEs, through their portfolios, are large investors in the U.S. mortgage market. Attheendof2004,GSE-issuedMBStotalednearly$2.7trillion,ornearly 35 percent of outstanding home mortgage debt. Atthe sametime, GSEportfolios totaled over $1.5 trillion, or more than 20 percent of total mortgage debt. In a typicalmonth,roughly40percentofnewlyoriginatedmortgagesaresecuritizedby the GSEs, and about 20 percent are bought by the GSEs’portfolios.1 Given their important role inmortgage markets, one mightexpect the quantities purchased by the GSEstoaffect the equilibrium prices inmortgage markets. Indeed, the GSEs’ effect onmortgage rates hasplayed akeyroleintherecent policy debates onhow toreformtheGSEs(Greenspan (2005b)). Earnings from mortgages held in the GSEs’ portfolios clearly benefit GSE shareholders. But these portfolios might also benefit mortgage originators and home buyers with conforming mortgages. Unusually heavy and sustained portfoliopurchasesmightbidupthepriceofnewmortgages,allowingoriginatorsgreater profitsortheopportunitytolowermortgagerates. However,theGSEsmustfinance 1Source:InsideMortgageFinance. 1

suchpurchases byissuingcorporate debt. Thustheextrademandformortgageassets created by portfolio purchases might be largely offset by the increase inGSE corporate debt. However, evenifGSEportfolio purchases donotaffect mortgage rates during normaltimes,thepurchasesmightactasastabilizationmechanismduringfinancial crises, with the GSEs acting as a buyer of last resort in the MBS market. The GSEsmightthenbuffermortgageoriginatorsfromfinancialmarketshocks,thereby limitingtheimpactofshocks onmortgageratesandmortgageborrowers. The ability of GSE portfolio purchases to affect MBS prices depends in part on whether investors view GSE-guaranteed MBSand GSEcorporate debt as substitutes. Roll (2003), among others, argues that foreign investors prefer holding GSE debt over GSE-guaranteed MBS because some GSE corporate bonds do not carry the prepayment risk inherent in MBS. In this view, GSE portfolio growth wouldstimulatelower-costforeigncapitaltoflowintoU.S.mortgagemarkets. By the same argument, however, this capital would flow out of corporate and Treasury markets. Moreover, other intermediaries can construct synthetic securities based on MBS that strip out prepayment risk. Given the size and diversity of the U.S.high-quality debtmarket(morethan$23trillionaccording totheFederalReserve’s FlowofFundsAccounts), theimportance offoreign investors inmortgage rate determination might be very small. Indeed, the large market for highly rated debt suggests that mortgage rates are set inworldwide capital markets and that GSEportfolios mighthavelittleinfluenceonmortgagerates. InvestorsdemandlowerreturnsonGSEcorporatedebtthanonthedebtofother comparablecorporations, partlybecauseinvestorsperceiveanimplicitgovernment guarantee on the debt. One might expect some of this implicit subsidy to flow to 2

mortgage borrowers. Previous literature has examined several channels by which the GSEs could affect mortgage rates. By law, the GSEs cannot buy mortgages largerthantheconformingloanlimit(suchlargemortgagesareknownasjumbos). Several papers have estimated the difference between mortgage rates on jumbo loans and those on conforming loans. Recent estimates of this spread range from 4 to 35 basis points, while older estimates are often times even higher.2 Other studies have examined the effect of the GSEs’ activities on conforming mortgage ratespreads.3 In thispaper weuse avector autoregression (VAR)approach and monthly data from March 1993 to December 2005 to estimate the effect of GSE secondary market activities—both grossportfolio purchases andMBSissuance—on primary and secondary mortgage rate spreads. OurmainfindingisthatGSEportfolio purchaseshaveessentiallynoshort-orlong-runeffectsoneitherprimaryorsecondary mortgage rate spreads. We also find some evidence that GSE portfolio purchases tend to rise following an increase in spreads; if spreads are mean-reverting, such behavior isconsistent withaprofit-maximizing portfolio strategy. Our results are subject to some obvious caveats. First, and most importantly from a policy perspective, our results are subject to the Lucas critique. We are notestimating thedeepparameters ofafullyspecified theoretical modelfeaturing optimizing forward-looking market participants. Thus, our estimated effects can 2SeeMcKenzie(2002)andAmbrose, LaCour-Little,andSanders(2004). Passmore, Sherlund, andBurgess(2005) estimatethat7basispoints, ofanestimated16basispoint jumbo-conforming spread, areattributabletotheGSEfunding advantage. Theother 9basispointsareattributableto differentcharacteristicsofthejumboandconformingmortgagemarkets,suchascompensationfor differingcreditandprepaymentrisks. 3Theendresultismixed:somestudiesconcludethattheGSEsdecreasemortgageratespreads— see Hendershott and Shilling (1989), Cotterman and Pearce (1996), Kolari, Fraser, and Anari (1998)—whileotherscanfindnosignificanteffects—seeRothberg,Nothaft,andGabriel(1989). 3

only describe the behavior of the endogenous variable under the policy regime of the sample period. However, our results can be used to evaluate the claims of an effect over the sample period. Further, given that secondary markets for nonconforming mortgagesaregrowinginsophistication andsize,theeffectiveness ofGSEactions (which primarily affect conforming mortgages) seemsmorelikely todiminishthantogrow. Second, we are limited by our data to studying the relationship among GSE actions and interest rates at a monthly frequency. If GSE actions and interacted spreads at a much higher frequency we might not find any relationships using monthly data. However, even at a monthly frequency, wedo find several interesting dynamic relationships; thus, we do not believe that our results are driven by time-aggregation bias. Moreover, we show that spreads and GSE actions are not particularly correlated within a month (although GSE actions are correlated with lagged shocks tospreads). Thus,thereissimplynotverymuchcausality toassign withinagivenmonth.4 Finally,otherstudiesoftherelationshipbetweenGSEportfolios and mortgage rates have used monthly data while none (to the best of our knowledge) havehadaccesstohigher-frequency data. Third,andrelated,theclassicstructuralVARmethodologyrequirestheeconometrician to make identifying assumptions about how the endogenous variables reacttooneanotherwithinthesameperiod. Thecommonlyusedtriangular ordering, for example, requires the econometrician to specify that some variables react to others only after a delay. In our model, this would require assuming that, for 4Asacounterexample,inflowstomutualfundsthatmainlybuyU.S.stocksareveryhighlycorrelatedwithU.S.stockpricemovements,evenwhendailydataareused.Inthatsetting,unlikeours, thereisalargeamount ofcausalitytoassignwithinaperiod. Theestimatedrelationshipbetween mutualfundflowsandstockpricesisstronglyaffectedbyassumingthatstockpricesdonotreactto fundflowswithinagivenperiod(orviceversa). 4

example,GSEportfoliopurchasescouldreacttochangesinspreadswithinagiven monthbutthatspreadscouldnotreacttoportfoliopurchases(orviceversa). However, in place of the standard structural VAR identifying assumptions, we use the weakeridentifyingassumptionssuggestedbyPesaranandShin(1998). Theseproduceimpulseresponsefunctionsthatarenotaffectedbytheorderingoftheshocks. In our robustness tests, we show that our results are essentially unchanged under severaldifferentidentifying assumptions. Our main results are also robust to a variety of alternative specifications. In particular, weestimated the effect of GSEactions on mortgage spreads in models that use (1) nonstationary techniques, (2) alternative scaling factors and variable definitions toproduce stationary series, and(3)avarietyoftimeseriesidentifying assumptions including thefullsetoftriangular shockorderings. Our paper is closest to the study of Naranjo and Toevs (2002), who estimate a long-run cointegrating relationship between GSE portfolio purchases and mortgage rate spreads. In contrast with our results, they conclude that GSE portfolio purchases lowerprimarymortgage ratespreads andthatportfolio purchases lower primary mortgage rates more than MBS issuance. Because they used proprietary datafromFannieMaetoconstructtheirdatasetwecannotattempttoreplicatetheir study. However,evenunderourclosestapproximationtotheirspecificationweare unabletoreproduce severalkeyfindingsfromtheirstudy.5 OurpaperisalsosimilartothestudyofGonzalez-Rivera(2001),whoestimates a long-run cointegrating relationship between secondary market spreads and portfolio purchases using monthly data from 1994 to 1999. She concludes that wider 5Foradiscussionofthedifferencebetweenreplicationandreproductionineconomics, aswell asananalysisoftheratesatwhicheconomicfindingscanbesuccessfullyreplicatedorreproduced, seeMcCullough,McGeary,andHarrison(2006). 5

secondarymarketspreadsincreaseportfoliopurchasesandfindsthatthe“errorcorrection term in the equation for portfolio purchases is not statistically significant” and therefore “it is mainly movements in the secondary market spread that will carryouttheadjustmenttowardequilibrium, intheveryshortterm”(p.33). Thus, Gonzalez-Rivera’s results also cast doubt on the ability of portfolio purchases to affectspreads.6 TheGSEs’large, highly leveraged portfolios pose risks tothe taxpayer andto thefinancialsystem morebroadly.7 Ourresults suggest thatcurbing thegrowthof theseportfoliosmightnotincreasemortgageratespaidbynewmortgageborrowers whilemitigatingtherisksposedtotaxpayers andthefinancialsystem.8 The remainder of the paper is organized as follows. In the next section, we introduce the VAR. The third section presents our data. Section 4 contains our results, ouranalysis ofGSEsecondary marketactivities onmortgageratespreads, andananalysis ofGSEactivities during thefinancial marketdistress oflate1998. Thenextsectionpresentsvariousrobustnesschecksandthefinalsectionconcludes. 2 VAR and Identification In this section wediscuss the economic environment in which our data are generated and our statistical approach. Broadly speaking, we estimate a vector autoregression(VAR)modelwithGSEactionsandmortgageratespreadsasendogenous 6The appendix contains our efforts to reproduce the results of Naranjo and Toevs (2002) and Gonzalez-Rivera(2001). 7Inthispaper,wefocusonlyonthepotentialbenefitsbroughtaboutbyGSEsecondarymarket activities. SeeLucasandMcDonald(2005)forestimatesoftheriskstotaxpayers, andGreenspan (2005a)andOfficeofManagementandBudget(2006)forassessmentsofthebroaderrisks. 8Incontrasttotheaccumulationofportfolioassets,theGSEs’securitizationofmortgagesgeneratesfewon-balance-sheetassetsandresultsinlittledebtissuance. Thus,issuesrelatedtomarket disciplineandsystemicriskdonotarisewithrespecttosecuritization. 6

variables. Our primary conceptual experiment is how one variable will evolve over time in reaction to a shock to a different variable, e.g., how mortgage rate spreads will change if GSE portfolio purchases suddenly increase. Following the literature, wecompute impulse response functions (IRFs)tomatchtheconceptual experiments. However, in our baseline specification, we do not use the standard (strong) identifying assumptions toconstruct our IRFs;instead wefollow Pesaran and Shin (1998) and use weaker identifying assumptions to construct generalized impulseresponse functions. 2.1 Overview Mortgage interest rate spreads are affected by investors’ expectations about mortgage risks (mainly credit and prepayment risks), financial market liquidity, investors’ expectations about the actions ofother participants (including the GSEs), andthecurrentlevelandexpectedtrajectoryofmortgageratespreads. Atthesame time,theGSEsarebuyingmortgagesfortheirowninvestmentportfoliosformany ofthesesamereasons. The theoretical connection among these variables could be quite complicated, inpartbecausetheequilibrium dependsonhowasmallnumberofentitiesexpects the others tobehave. In this paper, wedonot attempt toestimate thedeep parameters of such a theory-based structural model.9 Instead, in our reduced-form approachourgoalistocharacterizethestatisticalrelationshipamongtheendogenous variables, including thepotential stabilizing effects ofGSEactivities onmortgage ratespreads andtheGSEportfolio managers’reactions tomortgageratespreads. Our techniques allow us to examine the short- and long-run effects of GSE 9SeeBrunnermeierandPedersen(2005)foronesuchmodel. 7

portfoliopurchasesonmortgageratespreads. Notethatloweringmortgageratesin theshortrundoesnotrequirepermanentlyloweringthem,orvice-versa. TheGSEs might be able to dramatically affect mortgage rate spreads in the short run, but then seethese effects undone overtime, leaving mortgage ratespreads unchanged in the long run. Conversely, the GSEs might not be able to affect mortgage rate spreads much in the short run, but might be able to cumulate their effects over time,producing asignificantlong-run effect. Anobviousshortcomingofourdataisitsmonthlyfrequency. Financialmarket pricesandtradersroutinelyinteractatamuchhigherfrequency. Giventhatourdataaremonthly,itisdifficulttoascribecausation tocorrelated movementsbetween GSEactivities andmortgageratespreads. Thatis,ifGSEportfolio purchases rose and mortgage rate spreads fell within a month, we could not say to what degree GSE business managers reacted to larger-than-expected mortgage rate spreads by increasing portfolio purchases, andtowhatdegreelarger-than-expected GSEportfolio purchases pushed downmortgage rate spreads. Thestandard Cholesky-style identification scheme would require choosing a priori the direction of contemporaneous causality. Inthispaper,however,weuseamoregeneralidentificationstrategythateliminatestheneedtospecifyanaprioriorderingofvariableswithintheVAR.Pesaran and Shin (1998) (following Koop, Pesaran, and Potter (1996)) derive generalized impulse response functions that are invariant to the ordering of variables in the VAR. This procedure is a deviation from standard practice, so we will describe it insomedetailhere. 8

2.2 BasicModel Formally, we arrange the n endogenous variables (such as mortgage rate spreads and portfolio purchases) in each period t into the vector X and the m exogenous t variables(suchastherisk-freerateandtheslopeoftheyieldcurve)intothevector Z . We then write the structural relationship between the endogenous and exoget nousvariablesas: (1) Φ 0 X t = Φ 1 X t−1 +···+Φ p X t−p +Γ 0 Z t +Γ 1 Z t−1 +···+Γ k Z t−k +ε t . Here,Φ (n×n)andΓ (n×m)denotecoefficientmatrixesandε (n×1)denotes j j t the vector of fundamental shocks to the economic system. Because these shocks are taken to be independent we assume that their variance-covariance matrix is ′ diagonal andgivenby: E(ε ε ) = Λ . t t n In our primary specification, the vector of endogenous variables X includes t five variables: the secondary mortgage rate spread, the primary mortgage rate spread, implied volatility on ten-year Treasuries, gross GSE MBS issuance, and gross GSE portfolio purchases. The vector of exogenous variables Z includes t three variables: realized mortgage delinquencies, the ten-year Treasury rate, and the Treasury yield curve slope (1to 10 year). Eachof these variables is described inmoredetailinthenextsection. We cannot estimate the coefficients of the structural representation given by equation (1) directly. Instead, we estimate the coefficients of the reduced-form representation: (2) X t = A 1 X t−1 +···+A p X t−p +B 0 Z t +···+B k Z t−k +u t . 9

Hereu isthevectorofreduced-form errors. Inmovingfromequation(1)toequat tion (2) we left-multiplied both sides of equation (1) by Φ −1. Thus the reduced- 0 formerroru willbealinear combination ofthefundamental shocksu = Φ −1ε . t t 0 t Theerrorsu will,asaresult,becorrelatedacrossequationssothattheirvariancet covariance matrixgenerally willbenon-diagonal: E u u ′ = E Φ −1ε ε ′ Φ −1 ′ = Φ −1E ε ε ′ Φ −1 ′ = Φ −1Λ Φ −1 ′ ≡ S (cid:0) t t (cid:1) (cid:16) 0 t t 0 (cid:17) 0 (cid:0) t t (cid:1) 0 0 n 0 Our fundamental identification problem is to produce estimates of the structural parameters from the estimated reduced-form coefficients. The reduced-form parameterswillprovideuswithfewercoefficientsthanwerequiretopindownthe structural parameters, requiring us to make additional a priori assumptions about the structural parameters. More formally, the structural equation contains the followingfreeparameters: Φ , Φ ···Φ , Γ ···Γ , Λ → n2+pn2+kmnfreeparameters. 0 1 p 1 k n n | 2 { − z} n | p× {z n2 } | k× { n z m } |{ n z} ThematrixΦ hasonlyn2−nfreeparameters because thediagonal elementsare 0 assumed to be unity. The reduced-form estimates provide us with the following coefficientestimates: A ···A , B ···B , S → pn2+kmn+(n2+n)/2coefficients. 1 p 0 k | p× {z n2 } | k× { n z m } (n | 2+ {z n } )/2 Thus, the number of restrictions we have to impose in order to identify the structural parameters isthe difference between thetwo,or(n2 −n)/2. Inourbaseline 10

model,wheren = 5,werequire10restrictions. 2.3 Standard ImpulseResponse Functions In our results section we will focus on the impulse response functions implied by our estimated coefficients. The conceptual experiment is to compare the trajectoriesoftheendogenous variablesundertwoscenarios: inonescenario(thecontrol) weassumethat,attheendofperiodt−1nothingisknownaboutthefundamental shocksthatwillhittheeconomyinperiodt;intheotherscenario (theexperiment) we assume that at the end of period t−1 market participants become aware that nextperiod’sfundamentalshockwillbesuchthatu = δ. Wethencomputetheext pectedvaluesoftheendogenous variables inperiodst,t+1,t+2,andsoon;the impulse response function will be the difference between the expectations under theexperimentandunderthecontrol. Tosimplifynotation,assumethatthestructuralmodel(1)featuresonlyonelag oftheendogenous variables andnoexogenous variables (otherthantheshocks ε). This assumption is not as restrictive as it might seem because the estimated coefficients on the exogenous and extra lagged endogenous variables in the reducedform representation provide no net restrictions on the parameters of the structural representation.10 Thus,werewriteequation (2)as: ′ (2) X t = A 1 X t−1 +u t . 10Inpractice,weusetheAkaikeInformationCriterion(AIC)todeterminethenumberoflagsto includeinourempiricalspecification—seesection4.1. 11

′ Wecanuseequation (2)towriteX asthesumoftheu ’s: t t (3) X t = u t +A 1 u t−1 +A2 1 u t−2 +A3 1 u t−3 +··· . Ourconceptual experiment(theimpulseresponse function) canbewrittenas: E t−1 {X t+j |u t = δ}−E t−1 {X t+j }, j = 0,1,.... Fromequation (3)weseethatforeachperiodj thisdifference is: j E t−1 (cid:0) X t+j |u t = δ (cid:1) −E t−1 (cid:0) X t+j (cid:1) =A 1 δ. The shock to the reduced-form system δ is usually constructed to represent an innovation toone ofthefundamental shocks ε . Forexample, aunit innovation to t ′ thefirstequationinoursystemmightbewrittenasδ = (1, 0, 0, 0, 0). Because ε u = Φ −1ε ,δ wouldbewrittenas: δ = Φ −1δ . t 0 t 0 ε Underthisapproach, weneedanestimateofΦ inordertoconstruct δ. Aswe 0 discussed intheprevious section, weneedtoimpose(n2 −n)/2restrictions. The usual set of restrictions is that Φ be lower triangular, that is, all values above the 0 diagonalarezero. Withthisassumption,wecanformanestimateofΦ andΛ by 0 n constructing theCholeskydecomposition (or“matrixsquareroot”)ofS,thisisthe ′ unique lowertriangular matrix P such that P P = S. Then row iand column j 0 0 0 12

oftheestimateofΦ −1 canformed: 0 0 ifi < j,    (4) Φ − 0 1(i,j) =    1 ifi = j, d     P 0 (i,j)/Λ n (i,i) ifi > j.    c Andthevariance estimatesareformedasΛ (i,i) = P (i,i). n 0 c NoticethattheassumptionthatΦ islowertriangularisnotinnocuous;instead, 0 itisthesameasassumingthatsomevariablesdonotreacttoshockstootherequations within the same period.11 To fix ideas, consider the simplified state vector ′ x = (p , s ) where p are GSEportfolio purchases and s are secondary market t t t t t mortgageratespreads. Thenwewritethestructural equation (1)as: p  1 0  p t   ε t  = Φ 1 x t−1 +  φ21 1  s   εs   0  t   t  Because Φ is (by assumption) lower triangular, GSE purchases, p , do not react 0 t to spreads, s , within period t. Had we reversed the order of the state vector we t wouldhavemadetheopposite assumption: thatspreads donotreactwithinperiod ttoGSEportfolio purchases. 2.4 Generalized ImpulseResponse Functions The standard approach produces orthogonalized errors, that is, innovations to the structural shock process ε. Theapproach of Pesaran and Shin(1998) is instead to 11SeeSarnoandThornton(2004) foranexampleinwhichshockorderingaffectstheestimated IRFsinastandardtriangularidentificationscheme. 13

shock the reduced-form errors u . Because the variance-covariance matrix of u , t t S,isnotdiagonal,ashocktooneelementofuwouldnormallybeaccompaniedby “spillover” effectstotheotherelements. Ifweassumethattheshocksu aredistributed asmultivariate Normal,wecan t usethewell-knownformulafortheconditional expectation ofmultivariate normal random variables. Inparticular, ifu ∼ N(0,S) and u = δ (where u denotes t jt j jt thejthelement ofu ),thentheconditional expectation oftheotherelements ofu t t isgivenby: δ j E u |u = δ = ×S ×1 . t jt j j (cid:0) (cid:1) s jj Here 1 is the selector vector with zeros everywhere except for a value of 1 at j position j, while s is the jth-diagonal element of S. We estimate S from the jj variance-covariance matrixofthereduced-form errorsu . t Notice that we are making no extra assumptions about the relationship of the endogenousvariableswithinagivenperiod(theΦ matrix). Asaconsequence,this 0 technique will not deliver an estimate of Φ , and we will not be able to construct 0 orthogonal shocksδ . ε Nonetheless, thistechnique offerssomeimportantadvantages. AsPesaranand Shin (1998) show, the generalized and standard IRFs coincide when Φ is trian- 0 gular. In the moregeneral case, when Φ isnot triangular, the twoIRFswillonly 0 coincide for the variable in the state vector that is permitted to react to all other variables withinthesystem. However,this isprecisely thecasewhenthestandard IRFsaremostinfluenced bythe(strong) assumptionoftriangularity. 14

3 Data We obtained consistent data on gross GSE portfolio purchases, gross GSE MBS issuance, and mortgage interest rate spreads at a monthly frequency for March 1993 to December 2005, for a total of 154 observations. In addition, our data set contains covariates designed tocontrolforcreditandprepayment risks. Our measure of GSE portfolio purchases is the sum of Fannie Mae and Freddie Mac’s gross retained portfolio purchases of mortgage assets, including whole loans, own MBS, and other MBS. Our measure of MBS issuance is the sum of FannieMaeandFreddieMac’sgrossissuanceofMBS.Thesedataareavailableon theGSEs’monthlysummaryreports. Wenormalize grossportfolio purchases and MBSissuance bythe amount ofmortgages originated. Wefollow other studies in usingthemonthlytotalvolumeofnewresidential mortgagesoriginated (bothpurchase and refinance) as our measure of total market size. We derive this measure from the time series of mortgage originations reported under the Home Mortgage Disclosure Act(HMDA). We use both primary and secondary market mortgage rates to compute our measures of mortgage rate spreads. The primary market mortgage rate is the monthly average interest rate on new 30-year fixed-rate mortgages, from FreddieMac’sPrimaryMortgageMarketSurvey. Thesecondary marketmortgagerate is the monthly average current-coupon yield on Fannie Mae and Freddie Mac 30year MBS, from Bloomberg. Mortgage rate spreads are taken with respect to a duration-matched Treasuryrate.12 12DurationsareforFannieMaeandFreddieMac30-yearMBS,fromBloomberg.However,these dataonlygobackto1997.Wethereforebackcastthedurationdataasafunctionofyieldcurveslopes (1-to10-yearand1-to5-year),mortgagerates,thecoupongap,theMBArefinancingindex,existinghouseprices,thefixed-rateadjustable-ratespread,andtheadjustable-rateshareoforiginations. 15

Theprimaryriskspricedintomortgagerates(butnotrisk-freerates)arecredit risk (the risk of default) and prepayment risk (the risk of early termination). As a proxy for credit risk, we use the realized serious delinquency rate on conforming mortgages owned by Fannie Mae.13 We proxy prepayment risk with a measure of forward-looking interest rates and implied volatilities. Inparticular, weuse the slope oftheTreasuryyield curve (1to10year)andthe10-year Treasury rate. We furtheraugmentthemodelbyincludingimpliedvolatilityon10-yearTreasuriesas an endogenous component in the VAR. The implied volatility is calculated from theoptions on10-year Treasuryfuturescontracts. In relating mortgage rate spreads to GSE secondary market activity data, several complications arise. First, relative to our mortgage rate spread data, MBS issuancedatacanbelagged. Thatis,sometimepassesbetweenwhenanewhomeownerlocksintoandclosesonamortgage,andmoretimepassesbetweenwhenthe mortgage closes andwhenitissecuritized and sold inMBS.Second, there canbe lagsbetweenwhenaGSEcommitstoamortgagepurchaseandwhenthepurchase is brought onto the GSEs’ books. Because the GSEs do not release enough data publicly to adjust for these lags, we control for these two issues by (1) including lagged terms in our VAR and (2) using Fannie Mae commitments as an alternative to portfolio purchases in our robustness checks.14 The lags in the VAR will thenmanifest thetimingdifferences asdelayed responses inourimpulseresponse Furtherdetailsareavailablefromtheauthorsuponrequest. TheupperpanelofFigure2showsthe duration-matchedTreasuryversusthe10-yearTreasuryyields. 13Werealizethatthisisinherentlyabackward-lookingmeasure. However,moreforward-looking measuressuchasOFHEO’srepeat-saleshousepriceindexes(whichwouldcapturechangesincollateralvaluethatcouldbeexpectedtotranslateintochangesindefaults)areavailableonlyatquarterly frequencies and are also quite smooth. As an alternative, we report results using corporate bond spreadsasaproxyforcreditriskinthenextsection. 14Wehavealsoexperimentedwithlaggingourmeasureoforiginationsinanefforttomatchthe timingofMBSissuanceandportfoliopurchases,withnoappreciabledifferenceonourresults. 16

function analysis. The expected extra return to holding mortgages once these risks have been pricedisknownastheoption-adjusted spread(OAS).Iftheoption-adjusted spread (OAS)onmortgagesismean-reverting,buyingmortgageswhiletheOASisunusuallyhighcouldbeaprofitablestrategy. RatherthanincludeanestimateoftheOAS directly in our primary specification, we simply include some of the components of an OASmodel. This strategy avoids any problems with including an estimated variable ontheright-hand sideofaregression.15 Descriptive statistics for the data are provided in table 1. Figures 1–3 plot the time series of Treasury yields and implied volatilities on 10-year Treasuries, primary and secondary mortgage rate spreads, GSEportfolio purchases and MBS issuance (relative to total originations), and the mortgage delinquency rate and the slope of the Treasury yield curve. Note that the debt crisis of late 1998 was associated with a sharp widening of spreads and volatility and increased portfolio purchases; in just two months, primary mortgage rate spreads rose about 95 basis pointsandportfoliopurchases(relativetooriginations)increasedabout10percent. 4 Estimation Results In this section we discuss our estimated generalized impulse response functions under our baseline specification. We find that unanticipated portfolio purchases have essentially no effect on mortgage rate spreads. We also show that GSE activities during the debt crisis of late 1998 were not extraordinary; further, had the GSEsnotreactedtothespreadwideningduringthisperiod,primaryandsecondary 15Note, however, that we report results using an OAS from 1997 through 2005 as part of our robustnesschecksintheappendix. Theseresultsaresimilartothoseofourprimaryspecification. 17

mortgageratespreadswouldhaveevolvedinaboutthesameway. 4.1 Baseline Specification Our baseline specification is a stationary vector autoregression in which the five endogenous variables are: (1)secondary marketmortgage ratespreads, (2)primarymarketmortgageratespreads,(3)interestratevolatility,(4)GSEMBSissuance, and (5) GSE portfolio purchases. Our three exogenous variables are the ten-year Treasury rate, the slope of the Treasury yield curve, and the serious delinquency rate on mortgages reported by Fannie Mae. The Akaike Information Criterion (AIC) suggests that the optimal specification features two lags for the endogenous variables andone lagforthe exogenous variables.16 Variable definitions and sources areexplained insection3. WeincludedTreasurymarketvolatilityasanendogenousvariablebecausePerli and Sack(2003) provide evidence thatmortgage hedging canamplify movements in Treasury rates. Thus the volatility of risk-free rates might itself be endogenous to secondary market prices and GSE actions. Note that volatility, taken together with the slope and level of the yield curve, contains significant information about thevalueoftheprepayment optionembeddedinmortgages. GrossMBSissuance,whichisincludedasanendogenousvariable,isnotcompletely under the GSEs’ control, especially from month to month. While it might not be a policy tool in the same way that portfolio purchases are, MBS issuance doesconveyinformation aboutthesizeoftheconforming mortgagemarket. Also, byincludingit,wecanmorecloselyaddresstheconclusionsofNaranjoandToevs. 16Strictexogeneitytestssuggestthatourcreditandprepaymentriskproxiesareexogenoustothe system. 18

As with many financial series, the raw endogenous variables of interest may not be stationary. Gross MBS issuance and portfolio purchases contain obvious trends;weuseanaturalscalingfactor(totalmortgageoriginations)toconvertthem into stationary variables. Ourscaled variables can be interpreted as the percent of originatedmortgagessecuritizedbytheGSEsandpurchasedbytheGSEsfortheir ownportfolios. Aswediscussedinsection3,thetimingofourdataonoriginations maynotmatchthetimingofourdataonGSEactions;weconsiderthisissueinour robustness testingbelow. Mortgage ratespreads, theyieldcurve, anddelinquency ratesmight,asshown in table 2, have unit roots in their levels. However, economic theory suggests otherwise. A unit root in spreads would suggest that any unexpected shock to spreadswouldbepermanent,but,apriori,weexpectspreadstobemeanreverting. However, such spreads may revert to their long-run means only slowly, rendering unitroottestslesspowerful. Wefolloweconomictheoryratherthanstrictstatistical results in our baseline specification; however, we also consider a nonstationary modelinourrobustness tests. Aswediscussedinsection2,standardidentificationschemesrequireassuming aparticular shockordering. However,assumptions about shockordering aremore likely to affect the estimated impulse response functions when the variables are stronglycorrelatedwithinperiods. Table3showsthecontemporaneouscorrelation between estimated residuals from the reduced-form VAR. This matrix is nearly block-diagonal amongmortgageratespreadsandGSEactivities,whichshowsthat thevariablesofinterestareonlyweaklycorrelatedwithinperiods. Thus,wewould expect (as we find) that choosing a particular order for the shocks does not significantly affect our results. That is, our results are robust to the choice of shock 19

orderingbecausethereisnotmuchcontemporaneous correlationbetweenGSEactivities and mortgage rate spreads. There is simply not very much causality to assignwithinagivenmonth. Moreover,inourbaselinespecification, wedonotusethetriangular decompositionthatrequiresanaprioriassumptionaboutshockorder. Instead,asdiscussed insection2.4,weusethegeneralizedimpulseresponsefunctions(PesaranandShin (1998)). Theseimpulseresponsefunctionsareinvarianttotheorderingofvariables intheVAR,andusethehistoricalcorrelationsamongthereduced-formresidualsto formulate the residual variance-covariance matrix, allowing for contemporaneous cross-correlations amongtheendogenous variables. 4.2 Impulse ResponseFunctions Figure 4 shows the estimated generalized impulse response functions under our baseline specification. For each variable (shown in the rows), we computed the effectofaone-standard deviationshocktoeachofthefiveequations(showninthe columns). Wesummarize theeffectofeach shock (reading downeachcolumn) in turn. Theprimaryresultsofinterestareshowninthetoprightgraphs: theresponse ofmortgageratespreads toGSEportfolio purchases. Effect of a Shock to the Secondary Mortgage Rate Spread. The first column offigure4givesthereactionofthefiveendogenousvariables(secondarymortgage ratespread, primarymortgageratespread, impliedvolatility, grossMBSissuance, and portfolio purchases) to aone standard deviation shock to the secondary mortgageratespread (7.0basis points). Asshowninthetoprow,ashocktosecondary mortgage rate spread tends to be fairly persistent, with ahalf life of around 3 to 4 20

months. The second graph shows how the primary mortgage rate spread reacts to thesamesecondary mortgageratespreadshock; asonewouldexpect, theprimary market spread reacts strongly, increasing 7.0basis points, andis highly correlated withthe secondary mortgage rate spread. Thethird row showsthe reaction ofimpliedvolatilityonten-yearTreasuriestotheshocktothesecondarymarketspread. Volatility increases about1/4basispoints, withahalflifeofabout3months. Thebottom twographs show theresponses ofGSEactivities tothe secondary mortgage rate shock. In effect, they show how GSE business decisions react to an unexpected widening of mortgage spreads. As shown in the fourth row, MBS issuance (measuredasashareoforiginations) isatfirstessentially unchanged, but buildsuptoanincreased 1.6percentbythefourthmonthfollowingtheshock,and thenslowlytrailsoff. Asshowninthebottomrow,ashocktothesecondarymarket mortgageratespreadincreasestheGSEportfoliopurchaseshareoforiginationsby 0.9percentalmostimmediately, withahalflifeofabout6months. EffectofaShocktothePrimaryMortgage RateSpread. Thesecondcolumn ofthefiguregivesthereaction oftheendogenous variablestoaonestandard deviationshocktotheprimarymortgagemarketspread (7.6basispoints). Thegeneral patterns closely mirror the reactions to a shock to the secondary mortgage rate spread. MBS issuance (fourth row) increases by 1.6 percent of originations by 4 monthsaftertheshock,withahalflifeabout5monthsafterthis(9monthsafterthe initial shocktotheprimarymortgageratespread). Theinitialimpactofthisshock istoincreasetheGSEportfoliopurchaseshareoforiginations by0.7percent(fifth row),withahalflifeofabout8months. 21

EffectofaShocktoVolatility. Thethirdcolumnofthefiguregivesthereaction to a one standard deviation shock to implied volatility (0.5 basis points). Primary and secondary mortgage rate spreads (the top two rows) both increase following increasesinvolatility. Shockstovolatilityitself(thirdrow)arenotverypersistent, with a half life of 3 months. MBS issuance (fourth row) reacts slowly, increasing by1.1percentoforiginations bythefourthmonthaftertheshock(andahalflife3 monthsafterthis). Portfoliopurchases(bottomrow)increaseinthemonthsfollowing a shock to volatility, with purchases increasing by 1.6 percent of originations onemonthaftertheinitialshock(withahalflifeof1or2months). Effect of a Shock to MBS Issuance. The fourth column of the figure gives the reaction to a one standard deviation shock to gross MBS issuance (5.8 percent of originations). Primary and secondary mortgage rate spreads (the top two graphs) donotmovemorethanabasispointawayfromzero;further,thesemovementsare statistically insignificant. MBS issuance also has virtually no effect on volatility. MBS issuance itself (the fourth graph) shows little persistence, with a half life of only1or2months. Asshowninthebottomgraph,portfoliopurchasesincreaseby about1.1percent oforiginations withahalflifeof2or3months. Effect of a Shock to Portfolio Purchases. The final column of the figure gives the reactions toaone standard deviation shock toportfolio purchases (4.5 percent oforiginations). Bothsecondary andprimary marketspreads (thetoptwographs) increase between 1 and 2 basis points following an unexpected increase in portfolio purchases, although these increases are not statistically different from zero, and quickly trail off(half lives areabout 2months). Themiddle graph shows that 22

volatility increases by about 0.1 basis points due to the purchase shock. Interestingly, shocks to portfolio purchases push gross MBS issuance up (1.9 percent of originations) for several months (fourth graph). Portfolio purchases themselves showlittlepersistence, withahalflifeof1or2months. Effect of GSE Activities on Mortgage Rates Based on our impulse response analysis, we estimate that if the GSEs unexpectedly increase their portfolio purchases by $10 billion (about 3.7 percent of average monthly originations during 2004), the primary and secondary mortgage rate spreads would increase 1.4 and 1.3 basis points after one month, respectively. But if the GSEs instead unexpectedlyincreasedtheirsecuritization activity(thatis,theirgrossissuanceofMBS)by $10 billion, weestimate that primary and secondary mortgage rate spreads would decline 0.6 and 0.5 basis points, respectively. Note that none of these effects is statistically different fromzero. Theseresults suggest thatGSEportfolio purchases,inparticular, haveeconomicallyandstatistically negligibleeffectsonmortgage ratespreads. 4.3 Counterfactual: GSEs During Financial Crisis As we demonstrated, mortgage rate spreads do not react to an unexpected shock to portfolio purchases. However, we can also use our estimates to simulate how mortgage spreads would have evolved had the GSEs not followed their expected plans. These results are especially interesting during financial crises, when mortgagespreads widenabruptly. Inaddition, wecantesthowwellourmodelpredicts actual GSE behavior during financial crises; if the GSEs act to dampen crises by buying more mortgages than usual, our models should significantly underpredict 23

the volume of mortgage purchases based on wider spreads alone. If our models predict GSE behavior fairly well during the crisis period we know that the GSEs werenot(inthisepisodeatleast)“leaningagainstthewind”tostabilizemarketsin awaythatourstatistical workwouldn’t capturebecause ofitsrarity. We focus on the August–October 1998 Russian debt default/LTCM crisis. In figure 5, we plot our model’s predicted path for portfolio purchases and MBS issuance through this period as well as actual purchases and issuance. We also plot theresultsofcounterfactual simulations. Inthesesimulations, weuseourestimatedcoefficients topredict how mortgage spreads wouldhave evolved had portfolio purchases remainedflatthrough thecrisis. We find that our model predicts actual portfolio purchases and MBS issuance fairlywell,supportingtheviewthattheGSEs’behaviorthroughthisfinancialcrisis was not out of the ordinary. Further, our counterfactual simulations suggest that hadtheGSEskepttheirportfoliopurchasesflat,thepathofmortgageinterestrates throughthecrisiswouldhavefollowedessentiallythesamepathsaswhenportfolio purchases didreacttowidersecondary marketspreads. In the two months from the end of August 1998 to October 1998, secondary mortgage rate spreads widened about 85 basis points and primary mortgage rate spreadswidenednearly95basispoints. MBSissuancedeclinedabout7percentof originations andportfolio purchases increased byover9percentoforiginations. During this period spreads and purchases moved much more than during normal periods. As shown in the upper-left panel of figure 5, secondary mortgage rate spreads increased about 33basis points in September, another 52 basis points in October, before decreasing about 25 basis points in November, and remaining essentially unchanged inDecember. 24

We used our model estimates to conduct two related studies of this episode. First, we compared the actual trajectories of the endogenous variables to the predicted trajectories when we force secondary market spreads to follow their actual path over the episode, but allow all other endogenous variables to evolve as specified by the estimated model. These trajectories are given by the curves labeled “predicted” inthefigure. Asshownintheupper-right panel, themodeldoesanicejobinexplaining the evolution of the primary mortgage rate spread given the shocks to the secondary mortgage rate spread, though the actual primary market spread was a little wider thanourmodelpredicted. Themiddle-leftpanelshowsthatGSEportfoliopurchasesduringthisperiodcanbeexplainedalmostcompletelybytheirhistoricalpattern ofbuying mortgageswhenmortgage ratespreads arewide,whilethemiddle-right panelshowsthattheGSEMBSissuancewasalsofairlyordinary. Thelowerpanel shows that implied volatility was a bit higher than predicted by the model. In all, there wasnothing particularly special about the GSEactions during this period of financialmarketstress. Our second experiment estimates the evolution of the endogenous variables, especially mortgagespreads, hadGSEportfoliopurchasesbeenheldconstant. We set all variables to their August values, force portfolio purchases to be constant at their August levels, and take into account the series of shocks to secondary market spreads (from their August 1998 level). Otherwise, the endogenous variables evolve based on the estimated coefficients. These results are summarized by the curveslabeled“counterfactual” infigure5. In the counterfactual experiment, MBS issuance is somewhat below actual (middle-right),butprimaryandsecondarymortgageratespreadsandimpliedvolatil- 25

ity are essentially the same (and perhaps slightly lower) as the model originally predicted(uppertwoandlowerpanels). Thedifferencebetweenourcounterfactual experiment andourmodel’sprediction isourestimateoftheeffectofGSEportfoliopurchasesonmortgageratespreadsduringthisperiodofcrisis. Asshown,GSE portfolio purchases appear tohave had little effect on either primary or secondary mortgageratespreadsoronimpliedvolatility. 5 Robustness Tests Figure4gavetheestimatedimpulseresponsefunctionsunderourbaselinespecification. We now estimate and report a complete set of impulse response functions under avariety ofalternative specifications, data periods, and variable definitions. In particular: (1) we replace our duration-matched Treasury yields with constantmaturity yields; (2) we replace Treasury rates with swap rates; (3) we restrict the sampleto1993–1999; (4)werestrict thesampleto1993–2002; (5)wereplace the Freddie Macprimary mortgage rate with the rate on jumbo mortgages taken from MIRS; (6) we replace the Freddie Mac primary mortgage rate with the rate on conforming mortgages taken from MIRS; (7) we repeat our earlier counterfactual experiment using jumbo and conforming mortgage rate spreads; (8) we dispense withourscaling factor forGSEactivities andusefirstdifferences toforcestationarity; (9) we take differences of our normalized measures of GSE activities; (10) we use option-adjusted spreads; (11) we use Fannie Mae commitments instead of actualpurchases;and(12)weusecorporatebondspreadsasaproxyforcreditrisk. In all, our main result—that GSE portfolio purchases do not lower mortgage rate spreads—remains unchanged. 26

In addition to the alternative specifications reported here, we also estimated nonstationary specifications. Wereportresults fromthosespecifications alongside our attempts to reproduce the results of Gonzalez-Rivera (2001) and Naranjo and Toevs(2002). (SeesectionA.) 5.1 AlternativeIdentifying Assumptions OurfirstrobustnesstestistouseatriangularorCholesky-styleidentificationscheme inplaceofthePesaran-ShinIRFsweusedinourbaselinespecification. Triangular decompositions remain quite popular in the VAR literature, and have an obvious structuralinterpretation. However,withfiveendogenousvariablestherearesimply too many potential triangular shock orderings to be conveniently summarized. In this section we report results after stripping our system to two endogenous variables: portfoliopurchases andsecondary marketspreads.17 In the small system there are only two triangular decompositions: either purchasescannotreacttospreadswithinamonth,orspreadscannotreacttopurchases withinamonth. Neitherofthesealternativesiscompletelysatisfying,whichiswhy weprefer thePesaran-Shin identification scheme. Forcomparison, figure6shows the estimated Pesaran-Shin IRFs. Figure 7 shows the estimated IRFs under the assumption that spreads cannot react to purchases within a month; figure 8 shows theestimatedIRFsunder thealternative assumption thatpurchases cannot reactto spreads withinamonth. Our results are broadly unchanged: purchases do not affect spreads, while spreads leadtoincreased purchases, perhaps withalagofseveralmonths. 17Wemoveimpliedvolatilityintothesetofexogenousvariables. 27

5.2 Duration-Matched versus 10-YearTreasuries InourbaselinespecificationwecomputespreadsbetweenmortgageratesandTreasuries of the sameestimated duration. Figure 9 showsthe impulse-response functionsforaspecification inwhichtheprimaryandsecondary mortgageratespreads aretakenwithrespect totheconstant-maturity 10-year Treasuryrate,ratherthana duration-matchedTreasuryrate. Asshown,MBSissuanceincreasesby1.7percent (of originations) by 4 months after a 7.0 basis point shock to secondary market spreads, while portfolio purchases increase by 1.4 percent (of originations) by 2 months after the shock. Mortgage rate spreads, however, show statistically negligibleeffectsduetoshockstoGSEsecondary marketactivities. 5.3 Treasuries versus Swaps Asanotheralternativebenchmarkrisk-freeratewecomputespreadsbetweenmortgage rates and duration-matched swaps, rather than duration-matched Treasuries. Treasury rates are influenced by flights to quality and other factors that might not affect mortgage rates. Figure 10 shows the impulse-response functions for this specification. As shown, MBS issuance increases by 1.5 percent (of originations) by 5or 6months after a5.4basis point shock tosecondary market spreads, while portfoliopurchasesincreasebyabout0.7percent(oforiginations)by5or6months aftertheshock. MortgageratespreadsagainshowlittleeffectduetoshockstoGSE secondary marketactivities. 28

5.4 1993–1999SamplePeriod Ourbaseline specification uses data from March 1993 to December 2005. IfGSE actions were notably more effective in the earlier period, using the full sample mightmaskthisresult. Figure11showstheimpulse-response functions estimated over 1993 to 1999. As shown, MBS issuance increases by 1.0 percent (of originations) by 3 months after a 7.2 basis point shock to secondary market spreads, while portfolio purchases increase by 1.3 percent (of originations) immediately. Mortgageratespreads,however,showlittleeffectduetoshockstoGSEsecondary marketactivities. 5.5 1993–2002SamplePeriod Some commentators believe that GSE actions were restrained following revelationsaboutaccountingirregularitiesin2003. Iffinancialmarketsreacteddifferently during this period of restraint, the estimated IRFs using the full sample might be artificially dampened. Figure 12 shows the impulse-response functions for a specification estimated over1993 to2002. Asshown, MBSissuance increases by 1.1percent(oforiginations) by3monthsaftera7.2basispointshocktosecondary market spreads, while portfolio purchases increase by about 1.0 percent of (originations) immediately. Mortgageratespreads againshowlittleeffectduetoshocks toGSEsecondary marketactivities. 5.6 Jumbo Market Spread Inthenextspecification,weexaminetheeffectofGSEsecondarymarketactivities onthejumbomarketspread,measured asthespreadbetweentheaveragemonthly 29

MIRSjumborateandtheaverage 10-year Treasury rateoverthelastfivebusiness days of the month, rather than the primary market spread.18 Figure 13 shows the impulse-response functions for this specification. As shown, MBS issuance increases by 1.2 percent (of originations) by 5 months after a 8.1 basis point shock to secondary market spreads (measured as the average over the last five business days of the month), while portfolio purchases increase by 1.4 percent (of originations)by2monthsaftertheshock. Littleeffectisapparentfroma(8.8basispoint) shock to the jumbo market spread. Secondary market and jumbo market spreads showlittleeffectduetoshockstoGSEportfoliopurchases,butthesecondarymarket spread exhibits a statistically significant 1.8 basis point decline due to a (5.9 percent of originations) shock to MBSissuance (this effect does not seem to flow through tonewmortgageborrowers). 5.7 Conforming Market Spread Inthenextspecification,weexaminetheeffectofGSEsecondarymarketactivities on the conforming market spread, rather than the jumbo market spread, measured asthespreadbetweentheaveragemonthlyMIRSconformingrateandtheaverage 10-yearTreasuryrateoverthelastfivebusinessdaysofthemonth. Figure14shows the impulse-response functions for this specification. As shown, MBS issuance increasesby1.0percent(oforiginations) by4monthsaftera8.1basispointshock tosecondary marketspreads, whileportfolio purchases increaseby1.4percent(of originations) by 2 months after the shock. Little effect is apparent from a (5.5 basispoint) shocktotheconforming marketspread. Secondary marketandjumbo 18Werepeatedtherobustnesstestsdescribedinthissectionandthenextsectionusingspreadsto duration-matchedTreasuryratesandobtainedthesameresults. 30

marketspreadsshowlittleeffectduetoshockstoGSEportfoliopurchases, butthe secondary market spread exhibits a statistically significant 1.7 basis point decline due to a (5.8 percent of originations) shock to MBS issuance (again, this effect doesnotseemtoflowthroughtonewmortgageborrowers). 5.8 Jumbo and ConformingMarket Spreads During Late 1998 Anobvious generalization of our counterfactual experiment from section 4.3 isto replacethesingleprimarymarketinterestratefromFreddieMacwiththeseparate jumboandconformingmortgageratesfromMIRS.Asshowninfigure15,ourestimatedmodelspredictsecondarymarketspreadswell,buttendtounderpredictboth jumboandconforming marketspreads. Also,GSEsecondary marketactivitiesare underestimated. When we hold GSE portfolio purchases constant over the crisis period, MBS issuance is somewhat smaller (middle-right), but primary and secondary mortgage rate spreads and volatility are essentially the same as the model originally predicted (upper two and lower panels). Once again, the difference betweenourcounterfactual experiment andourmodel’sprediction isourestimateof theeffectofGSEportfoliopurchasesonmortgageratespreadsduringthecrisisperiod. Asshown, GSEportfolio purchases appear to have had little effect oneither primaryorsecondary mortgageratespreads (orvolatility). 5.9 FirstDifferences: UnnormalizedGSE Activities Next, we consider a difference-stationary specification to address potential problemsstemmingfromunitrootsininterest ratespreads. GSEsecondary marketactivitiesaremeasuredinlogsandarenotnormalizedbyHMDAoriginations. Figure 16showsthecumulativeimpulse-responsefunctionsforthisfirst-differencesspec- 31

ification. As shown, MBS issuance eventually increases by 2.8 percent following a7.4basis point shock tosecondary market spreads, whileportfolio purchases increase by8.3percent. Mortgage ratespreads showlittle effect duetoa28percent shock to GSE portfolio purchases, but show a statistically significant 3.2 to 3.4 basispointdecline duetoa15percentshocktoMBSissuance. 5.10 FirstDifferences: NormalizedGSE Activities Wenext consider thesamedifference-stationary specification, but instead ofmeasuring GSE secondary market activities in unnormalized logs, we normalize by HMDAoriginations (as in our primary specification). Figure 17 shows the cumulativeimpulse-responsefunctionsforthisfirst-differencesspecification. Asshown, MBSissuanceeventuallydecreasesby1.1percent(oforiginations)followinga7.5 basis point shock tosecondary market spreads, while portfolio purchases increase by only 0.6percent (of originations). Mortgage rate spreads show little effect due toa5percent(oforiginations)shocktoGSEportfoliopurchases,butshowastatistically significant 3.4to3.8basis pointdecline duetoa6percent (oforiginations) shocktoMBSissuance. 5.11 Option-Adjusted Spreads Next, we examine the effects of option-adjusted spreads on our results. Optionadjusted spreads obviously require an estimate of the value of the prepayment option on a mortgage. Not only are these estimates model-dependent, in practice, market participants price different mortgages using different models. Thus, although they might carry the same coupon rates, the estimated OAS on a pool of high-balance unseasoned loans will differ from the estimated OAS on a pool 32

of low-balance seasoned loans. We used Bloomberg’s estimate of the value of the prepayment option on newly issued par GSE MBS as representing the value of the prepayment option to the average borrower. We obtained data for 1997– 2005 from Bloomberg and subtracted the value of the embedded option to prepay a mortgage from the unadjusted primary and secondary market spreads. Figure 18 shows the impulse-response functions for this specification. As shown, MBS issuance increases by 2.3 percent (of originations) by 3 or 4 months after a 10.4 basis point shock to the option-adjusted secondary market spread, while portfolio purchases increase by 2.2 percent (of originations) by 1 month after the shock. Option-adjusted spreads, however, show little effect due to shocks to GSE secondarymarketactivities. 5.12 FannieMaeCommitments Mortgageratesmightrespondtonewsaboutfutureportfolio purchasesratherthan to the purchases themselves. We used Fannie Mae’s commitments to purchase mortgages as proxy for news about future portfolio movements. Figure 19 shows the impulse-response functions for a specification in which Fannie Mae commitments are used in place of GSE portfolio purchases. As shown, MBS issuance increases by 1.3 percent (of originations) by 4 or 5 months after a 7.1 basis point shock to the secondary market spread, while portfolio purchases increase by 0.9 percent (oforiginations) immediately. Mortgageratespreads, however,showlittle effectduetoshockstoGSEsecondary marketactivities. 33

5.13 Corporate Bond Spreads as aProxy forCreditRisk Thecreditriskmeasureweincludeinourbaselinemodelisbackward-looking. As an alternative, we used corporate bond spreads as a more forward-looking proxy forcredit risk. Figure 20showstheimpulse-response functions foraspecification inwhichthespreadbetweenMoody’sBAA-andAAA-ratedindustrialbondyields isusedinsteadofFannieMae’sreportedseriousdelinquencyrate. Asshown,MBS issuanceincreasesby1.0percent(oforiginations)by4or5monthsaftera7.2basis point shock tothesecondary marketspread, whileportfolio purchases increase by 0.8percent(oforiginations) almostimmediately. Mortgageratespreads, however, showlittleeffectduetoshockstoGSEsecondary marketactivities. 6 Conclusion We examined the empirical connection between mortgage interest rates and GSE secondary market activities, especially GSEpurchases ofmortgages for their own portfolios. IfGSEportfoliopurchasesaffectedmortgagerates,theycouldstabilize mortgagemarkets. Thisbenefitwouldflowtoallmortgagemarketparticipants,not justGSEshareholders. Earlierstudieshaveconflictedwitheachother: NaranjoandToevs(2002)conclude that GSE activities significantly affect mortgage spreads, while Gonzalez- Rivera (2001) concludes that mortgage spreads drive portfolio purchases. Our findings are consistent with Gonzalez-Rivera’s study, and we were unable to reproduce NaranjoandToevs’findings. Wefound that portfolio purchases have economically andstatistically negligible effects on both primary and secondary mortgage rate spreads. Our results are 34

robust toalternative identifying assumptions andtoalternative modelandvariable specifications. Weexamined the debt crisis of late 1998 and found that GSEactivities generally followed the predictions of our model. Further, had the GSEs not reacted to mortgage ratespread widening through these episodes, weestimate thatmortgage spreads paid by new mortgage borrowers would have evolved in about the same way. 35

References Ambrose, B.,M.LaCour-Little, andA.Sanders(2004). Theeffectofconformingloanstatusonmortgageyieldspreads: Aloanlevelanalysis.RealEstate Economics32,541–69. Brunnermeier,M.K.andL.H.Pedersen(2005).Predatorytrading.TheJournal ofFinance60(4),1825–64. Cotterman, R. F. and J. E. Pearce (1996). The effects of the Federal National Mortgage Association and the Federal Home Loan Mortgage Corporation on conventional fixed-rate mortgage yields. In Studies on Privatizing Fannie Mae and Freddie Mac, pp. 97–168. Washington, DC: U.S. Department ofHousingandUrbanDevelopment,OfficeofPolicyDevelopmentandResearch. Gonzalez-Rivera, G.(2001). Linkagesbetweensecondary andprimarymarkets formortgages. JournalofFixedIncome11,29–36. Greenspan, A. (2005a, May 19). Remarks by Alan Greenspan, Chairman of the Board of Governors of the Federal Reserve System, via Satellite at the Conference on Housing, Mortgage Finance, and the Macroeconomy at the FederalReserveBankofAtlanta. Greenspan, A.(2005b,April6).StatementofAlanGreenspan,Chairmanofthe Board of Governors of the Federal Reserve System, before the Committee onBanking, Housing,andUrbanAffairsoftheUnitedStatesSenate. Hendershott, P. H. and J. D. Shilling (1989, June). The impact of the agencies onconventional fixed-rate mortgage yields. Journal ofRealEstate Finance andEconomics2(2),101–15. Kolari, J. W., D. R. Fraser, and A. Anari (1998, Winter). The effects of securitization on mortgage market yields: A cointegration analysis. Real Estate Economics26(4), 677–93. Koop,G.,M.H.Pesaran,andS.M.Potter(1996).Impulseresponseanalysisin nonlinear multivariatemodels.Journal ofEconometrics 74(1),199–147. Lucas, D. and R. McDonald (2005). An options-based approach to evaluating the risk of Fannie Mae and Freddie Mac. Manuscript, Department of Finance,KelloggSchool,Northwestern University, EvanstonIL. McCullough, B. D., K. A. McGeary, and T. D. Harrison (2006). Lessons from the JMCB archive. Journal of Money, Credit, and Banking 38(4), 1093– 1107. 36

McKenzie, J.(2002). Areconsideration ofthejumbo/non-jumbo mortgage rate differential. JournalofRealEstateFinanceandEconomics25,197–214. Naranjo,A.andA.Toevs(2002).Theeffectsofpurchasesofmortgagesandsecuritizationbygovernmentsponsoredenterprisesonmortgageyieldspreads andvolatility. JournalofRealEstateFinanceandEconomics25,173–96. Office of Management and Budget (2004). Analytical Perspectives, Budget of theUnitedStates Government, FiscalYear2005. Washington: Government PrintingOffice. Office of Management and Budget (2006). Analytical Perspectives, Budget of theUnitedStates Government, FiscalYear2007. Washington: Government PrintingOffice. Passmore, W., S. M. Sherlund, and G. Burgess (2005). The effect of housing government-sponsored enterprises on mortgage rates. Real Estate Economics33(3),427–463. Perli, R. and B. Sack (2003). Does mortgage hedging amplify movements in long-term interestrates? Journal ofFixedIncome13(3),7–17. Pesaran,H.andY.Shin(1998).Generalizedimpulseresponseanalysisinlinear multivariatemodels.EconomicsLetters58(1), 17–29. Roll, R. (2003). Benefits to homeowners from mortgage portfolios retained by FannieMaeandFreddieMac.JournalofFinancialServicesResearch23(1), 29–42. Rothberg, J. P., F. E. Nothaft, and S. A. Gabriel (1989). On the determinants of yield spreads between mortgage pass-through and Treasury securities. JournalofRealEstateFinanceandEconomics2(4),301–15. Sarno,L.andD.L.Thornton(2004).TheefficientmarkethypothesisandidentificationinstructuralVARs.FederalReserveBankofSt.LouisReview86(1), 49–60. 37

TABLE 1: DescriptiveStatistics Variable Mean Median Std. Dev. Min Max GSEActivities(percentoforiginations) MBSIssuance 36.15 34.82 9.44 19.03 65.57 PortfolioPurchases 18.32 17.49 7.21 7.04 47.72 Mortgageratespreads(basispoints) PrimaryMarket 210.12 192.91 50.55 132.58 334.30 Secondary Market 170.34 158.41 38.05 109.78 269.33 Impliedvolatility(basispoints) Ten-yearTreasury 6.73 6.75 1.18 4.24 9.53 Explanatoryvariables YieldCurveSlopea 129.73 96.60 99.80 −38.02 313.67 Ten-yearTreasuryRateb 5.49 5.60 1.05 3.33 7.95 Delinquency Ratesc 54.29 55.00 5.94 45.00 79.00 NOTE. Statistics are for 154 monthly observations running from March 1993 through December2005. aTen-yearlessone-yearTreasuryrates,expressedinbasispoints. bExpressedinpercent. cFannieMae“seriousdelinquency”rateonmortgages,expressedinbasispoints. 38

TABLE 2: UnitRootTests ADF Phillips-Perron Intercept Intercept+ Intercept Intercept+ Variable Trend Trend GSEActivities(percentoforiginations) MBSIssuance −4.68⋆⋆ −4.68⋆⋆ −4.56⋆⋆ −4.57⋆⋆ PortfolioPurchases −4.74⋆⋆ −5.07⋆⋆ −4.61⋆⋆ −5.07⋆⋆ MortgageRateSpreads PrimaryMarket −1.53 −1.94 −1.83 −2.38 SecondaryMarket −1.64 −1.82 −2.09 −2.37 ImpliedVolatility Ten-yearTreasury −3.14⋆ −3.10 −2.99⋆ −2.97 ExplanatoryVariables YieldCurveSlope −1.72 −1.75 −1.71 −1.76 Ten-yearTreasuryRate −1.69 −3.56⋆ −1.52 −3.04 Delinquency Rates −0.58 −1.15 −0.16 −0.78 NOTE. AugmentedDickey-FullerandPhillips-Perron testsofnullhypotheses that indicatedserieshaveaunitroot. ⋆and⋆⋆denotestatisticalsignificanceatthe5-and 1-percent levels,respectively. 39

TABLE 3: Contemporaneous Correlation AmongReduced-Form Residuals MortgageRateSpreads GSEActivities Secondary Primary Implied MBS Portfolio Market Market Volatility Issuance Purchases Secondary Mkt.Spread 1.000 PrimaryMkt.Spread 0.921 1.000 ImpliedVolatility 0.520 0.506 1.000 MBSIssuance −0.071 −0.035 0.010 1.000 PortfolioPurchases 0.182 0.152 0.147 0.242 1.000 NOTE. Thewithin-period correlation betweenmortgage ratespreads andGSEactivities (shown in the lower left portion of the matrix) are a measure of the effect of different triangular identifying assumptions on the estimated impulse response functions. Becausethecorrelation islow,wedonotexpectourresults tobesensitivetodifferent identifying assumptions. 40

FIGURE 1: TreasuryYieldsandImpliedVolatility 8 7 6 5 4 3 2 1994 1996 1998 2000 2002 2004 Duration-matched Treasury 10-year Treasury tnecreP 10 9 8 7 6 5 4 1994 1996 1998 2000 2002 2004 Implied Volatility stnioP sisaB NOTE. Figure shows the time series of Treasury market data used in our analysis. The top panel gives durationmatched Treasury yields and 10-year Treasury yields; the bottom panel shows volatility on 10-year Treasuries impliedbyoptions prices. 41

FIGURE 2: MortgageRateSpreadsandGSESecondaryMarketActivities 350 300 250 200 150 100 1994 1996 1998 2000 2002 2004 Secondary Market Spread Primary Market Spread stnioP sisaB 70 60 50 40 30 20 10 0 1994 1996 1998 2000 2002 2004 MBS Issuance Portfolio Purchases snoitanigirO fo tnecreP NOTE. Figure shows the time series data on mortgage interest rate spreads and GSE actions used in our baseline specification. The top panel shows primary and secondary mortgage rate spreads (relative to duration-matched Treasury yields), and the bottom panel gross portfolio purchases and MBS issuance as a percentage of HMDA originations. 42

FIGURE 3: MortgageMarketCharacteristics 80 76 72 68 64 60 56 52 48 44 1994 1996 1998 2000 2002 2004 Serious Delinquency Rate stnioP sisaB 400 300 200 100 0 -100 1994 1996 1998 2000 2002 2004 Treasury Yield Curve Slope stnioP sisaB NOTE. Figure gives two of the exogenous variables from our baseline specification. The top panel shows Fannie Mae’s serious delinquency rate (defined as the percent of mortgages 90 days or more past due or in foreclosure). The spike at the end of our sample is related to the late-2005 hurricanes and almost completely reverses over the next six months (not in our sample). The bottom panel gives the slope of the Treasury yield curve (ten year minus oneyearTreasuryrates). 43

FIGURE 4: ImpulseResponseFunctions: BaselineSpecification Response of SECSPREAD to SECSPREAD Response of SECSPREAD to PRISPREAD Response of SECSPREAD to VOLATILITY Response of SECSPREAD to ISSUANCE Response of SECSPREAD to PURCHASES 10 10 10 10 10 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PRISPREAD to SECSPREAD Response of PRISPREAD to PRISPREAD Response of PRISPREAD to VOLATILITY Response of PRISPREAD to ISSUANCE Response of PRISPREAD to PURCHASES 10 10 10 10 10 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of VOLATILITY to SECSPREAD Response of VOLATILITY to PRISPREAD Response of VOLATILITY to VOLATILITY Response of VOLATILITY to ISSUANCE Response of VOLATILITY to PURCHASES .6 .6 .6 .6 .6 .5 .5 .5 .5 .5 .4 .4 .4 .4 .4 .3 .3 .3 .3 .3 .2 .2 .2 .2 .2 .1 .1 .1 .1 .1 .0 .0 .0 .0 .0 -.1 -.1 -.1 -.1 -.1 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of ISSUANCE to SECSPREAD Response of ISSUANCE to PRISPREAD Response of ISSUANCE to VOLATILITY Response of ISSUANCE to ISSUANCE Response of ISSUANCE to PURCHASES 7 7 7 7 7 6 6 6 6 6 5 5 5 5 5 4 4 4 4 4 3 3 3 3 3 2 2 2 2 2 1 1 1 1 1 0 0 0 0 0 -1 -1 -1 -1 -1 -2 -2 -2 -2 -2 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PURCHASES to SECSPREAD Response of PURCHASES to PRISPREAD Response of PURCHASES to VOLATILITY Response of PURCHASES to ISSUANCE Response of PURCHASES to PURCHASES 6 6 6 6 6 5 5 5 5 5 4 4 4 4 4 3 3 3 3 3 2 2 2 2 2 1 1 1 1 1 0 0 0 0 0 -1 -1 -1 -1 -1 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 NOTE. Each panel gives the effect in month j for a given variable of a shock to the indicated innovation. Thus, the upper right panel showsthe effect onsecondary mortgage ratespreads ofaonestandard deviation shock tothe portfolio purchase innovation. 44

FIGURE 5: ResponseDuringLiquidity Crisisof1998 220 200 180 160 140 120 AUG SEP OCT NOV DEC stnioP sisaB Secondary Mortgage Rate Spread 260 240 220 200 180 160 AUG SEP OCT NOV DEC stnioP sisaB Primary Mortgage Rate Spread 30 28 26 24 22 20 18 16 AUG SEP OCT NOV DEC snoitanigirO fo tnecreP Portfolio Purchases 48 44 40 36 32 28 AUG SEP OCT NOV DEC snoitanigirO fo tnecreP MBS Issuance 10 9 8 7 6 5 4 AUG SEP OCT NOV DEC Actual Predicted Counterfactual stnioP sisaB Implied Volatility NOTE. Each panel gives the month-by-month effects of the liquidity shock to secondary mortgage rate spreads. Thus, the middle leftgraph showsthat themodel does wellintracing out theeffects oftheliquidity shock onGSE portfolio purchases. Portfoliopurchases duringthisperiodhadlittleeffectonmortgageratespreads. 45

FIGURE 6: ImpulseResponseFunctions: SmallSystemResults(Pesaran-Shin) Response of SECSPREAD to SECSPREAD Response of SECSPREAD to PURCHASES 7 7 6 6 5 5 4 4 3 3 2 2 1 1 0 0 -1 -1 -2 -2 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 Response of PURCHASES to SECSPREAD Response of PURCHASES to PURCHASES 6 6 5 5 4 4 3 3 2 2 1 1 0 0 -1 -1 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 46

FIGURE 7: Impulse Response Functions: Small System Results (Cholesky/Purchases) Response of SECSPREAD to SECSPREAD Response of SECSPREAD to PURCHASES 7 7 6 6 5 5 4 4 3 3 2 2 1 1 0 0 -1 -1 -2 -2 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PURCHASES to SECSPREAD Response of PURCHASES to PURCHASES 6 6 5 5 4 4 3 3 2 2 1 1 0 0 -1 -1 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 47

FIGURE 8: Impulse Response Functions: Small System Results (Cholesky/Spreads) Response of PURCHASES to PURCHASES Response of PURCHASES to SECSPREAD 6 6 5 5 4 4 3 3 2 2 1 1 0 0 -1 -1 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of SECSPREAD to PURCHASES Response of SECSPREAD to SECSPREAD 7 7 6 6 5 5 4 4 3 3 2 2 1 1 0 0 -1 -1 -2 -2 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 48

FIGURE 9: ImpulseResponseFunctions: SpreadsRelativeto10-YearTreasuries Response of SECSPREAD to SECSPREAD Response of SECSPREAD to PRISPREAD Response of SECSPREAD to VOLATILITY Response of SECSPREAD to ISSUANCE Response of SECSPREAD to PURCHASES 10 10 10 10 10 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PRISPREAD to SECSPREAD Response of PRISPREAD to PRISPREAD Response of PRISPREAD to VOLATILITY Response of PRISPREAD to ISSUANCE Response of PRISPREAD to PURCHASES 10 10 10 10 10 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of VOLATILITY to SECSPREAD Response of VOLATILITY to PRISPREAD Response of VOLATILITY to VOLATILITY Response of VOLATILITY to ISSUANCE Response of VOLATILITY to PURCHASES .6 .6 .6 .6 .6 .5 .5 .5 .5 .5 .4 .4 .4 .4 .4 .3 .3 .3 .3 .3 .2 .2 .2 .2 .2 .1 .1 .1 .1 .1 .0 .0 .0 .0 .0 -.1 -.1 -.1 -.1 -.1 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of ISSUANCE to SECSPREAD Response of ISSUANCE to PRISPREAD Response of ISSUANCE to VOLATILITY Response of ISSUANCE to ISSUANCE Response of ISSUANCE to PURCHASES 7 7 7 7 7 6 6 6 6 6 5 5 5 5 5 4 4 4 4 4 3 3 3 3 3 2 2 2 2 2 1 1 1 1 1 0 0 0 0 0 -1 -1 -1 -1 -1 -2 -2 -2 -2 -2 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PURCHASES to SECSPREAD Response of PURCHASES to PRISPREAD Response of PURCHASES to VOLATILITY Response of PURCHASES to ISSUANCE Response of PURCHASES to PURCHASES 5 5 5 5 5 4 4 4 4 4 3 3 3 3 3 2 2 2 2 2 1 1 1 1 1 0 0 0 0 0 -1 -1 -1 -1 -1 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 49

FIGURE 10: ImpulseResponseFunctions: SpreadsRelativetoSwaps Response of SECSPREAD to SECSPREAD Response of SECSPREAD to PRISPREAD Response of SECSPREAD to VOLATILITY Response of SECSPREAD to ISSUANCE Response of SECSPREAD to PURCHASES 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PRISPREAD to SECSPREAD Response of PRISPREAD to PRISPREAD Response of PRISPREAD to VOLATILITY Response of PRISPREAD to ISSUANCE Response of PRISPREAD to PURCHASES 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of VOLATILITY to SECSPREAD Response of VOLATILITY to PRISPREAD Response of VOLATILITY to VOLATILITY Response of VOLATILITY to ISSUANCE Response of VOLATILITY to PURCHASES .5 .5 .5 .5 .5 .4 .4 .4 .4 .4 .3 .3 .3 .3 .3 .2 .2 .2 .2 .2 .1 .1 .1 .1 .1 .0 .0 .0 .0 .0 -.1 -.1 -.1 -.1 -.1 -.2 -.2 -.2 -.2 -.2 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of ISSUANCE to SECSPREAD Response of ISSUANCE to PRISPREAD Response of ISSUANCE to VOLATILITY Response of ISSUANCE to ISSUANCE Response of ISSUANCE to PURCHASES 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PURCHASES to SECSPREAD Response of PURCHASES to PRISPREAD Response of PURCHASES to VOLATILITY Response of PURCHASES to ISSUANCE Response of PURCHASES to PURCHASES 7 7 7 7 7 6 6 6 6 6 5 5 5 5 5 4 4 4 4 4 3 3 3 3 3 2 2 2 2 2 1 1 1 1 1 0 0 0 0 0 -1 -1 -1 -1 -1 -2 -2 -2 -2 -2 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 50

FIGURE 11: ImpulseResponseFunctions: 1993–1999 SamplePeriod Response of SECSPREAD to SECSPREAD Response of SECSPREAD to PRISPREAD Response of SECSPREAD to VOLATILITY Response of SECSPREAD to ISSUANCE Response of SECSPREAD to PURCHASES 10 10 10 10 10 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PRISPREAD to SECSPREAD Response of PRISPREAD to PRISPREAD Response of PRISPREAD to VOLATILITY Response of PRISPREAD to ISSUANCE Response of PRISPREAD to PURCHASES 12 12 12 12 12 8 8 8 8 8 4 4 4 4 4 0 0 0 0 0 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of VOLATILITY to SECSPREAD Response of VOLATILITY to PRISPREAD Response of VOLATILITY to VOLATILITY Response of VOLATILITY to ISSUANCE Response of VOLATILITY to PURCHASES .8 .8 .8 .8 .8 .6 .6 .6 .6 .6 .4 .4 .4 .4 .4 .2 .2 .2 .2 .2 .0 .0 .0 .0 .0 -.2 -.2 -.2 -.2 -.2 -.4 -.4 -.4 -.4 -.4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of ISSUANCE to SECSPREAD Response of ISSUANCE to PRISPREAD Response of ISSUANCE to VOLATILITY Response of ISSUANCE to ISSUANCE Response of ISSUANCE to PURCHASES 6 6 6 6 6 5 5 5 5 5 4 4 4 4 4 3 3 3 3 3 2 2 2 2 2 1 1 1 1 1 0 0 0 0 0 -1 -1 -1 -1 -1 -2 -2 -2 -2 -2 -3 -3 -3 -3 -3 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PURCHASES to SECSPREAD Response of PURCHASES to PRISPREAD Response of PURCHASES to VOLATILITY Response of PURCHASES to ISSUANCE Response of PURCHASES to PURCHASES 5 5 5 5 5 4 4 4 4 4 3 3 3 3 3 2 2 2 2 2 1 1 1 1 1 0 0 0 0 0 -1 -1 -1 -1 -1 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 51

FIGURE 12: ImpulseResponseFunctions: 1993–2002 SamplePeriod Response of SECSPREAD to SECSPREAD Response of SECSPREAD to PRISPREAD Response of SECSPREAD to VOLATILITY Response of SECSPREAD to ISSUANCE Response of SECSPREAD to PURCHASES 10 10 10 10 10 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PRISPREAD to SECSPREAD Response of PRISPREAD to PRISPREAD Response of PRISPREAD to VOLATILITY Response of PRISPREAD to ISSUANCE Response of PRISPREAD to PURCHASES 12 12 12 12 12 8 8 8 8 8 4 4 4 4 4 0 0 0 0 0 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of VOLATILITY to SECSPREAD Response of VOLATILITY to PRISPREAD Response of VOLATILITY to VOLATILITY Response of VOLATILITY to ISSUANCE Response of VOLATILITY to PURCHASES .6 .6 .6 .6 .6 .5 .5 .5 .5 .5 .4 .4 .4 .4 .4 .3 .3 .3 .3 .3 .2 .2 .2 .2 .2 .1 .1 .1 .1 .1 .0 .0 .0 .0 .0 -.1 -.1 -.1 -.1 -.1 -.2 -.2 -.2 -.2 -.2 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of ISSUANCE to SECSPREAD Response of ISSUANCE to PRISPREAD Response of ISSUANCE to VOLATILITY Response of ISSUANCE to ISSUANCE Response of ISSUANCE to PURCHASES 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PURCHASES to SECSPREAD Response of PURCHASES to PRISPREAD Response of PURCHASES to VOLATILITY Response of PURCHASES to ISSUANCE Response of PURCHASES to PURCHASES 5 5 5 5 5 4 4 4 4 4 3 3 3 3 3 2 2 2 2 2 1 1 1 1 1 0 0 0 0 0 -1 -1 -1 -1 -1 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 52

FIGURE 13: ImpulseResponseFunctions: JumboMarketSpread Response of SECSPREAD to SECSPREAD Response of SECSPREAD to JSPREAD Response of SECSPREAD to VOLATILITY Response of SECSPREAD to ISSUANCE Response of SECSPREAD to PURCHASES 10 10 10 10 10 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of JSPREAD to SECSPREAD Response of JSPREAD to JSPREAD Response of JSPREAD to VOLATILITY Response of JSPREAD to ISSUANCE Response of JSPREAD to PURCHASES 10 10 10 10 10 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of VOLATILITY to SECSPREAD Response of VOLATILITY to JSPREAD Response of VOLATILITY to VOLATILITY Response of VOLATILITY to ISSUANCE Response of VOLATILITY to PURCHASES .6 .6 .6 .6 .6 .5 .5 .5 .5 .5 .4 .4 .4 .4 .4 .3 .3 .3 .3 .3 .2 .2 .2 .2 .2 .1 .1 .1 .1 .1 .0 .0 .0 .0 .0 -.1 -.1 -.1 -.1 -.1 -.2 -.2 -.2 -.2 -.2 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of ISSUANCE to SECSPREAD Response of ISSUANCE to JSPREAD Response of ISSUANCE to VOLATILITY Response of ISSUANCE to ISSUANCE Response of ISSUANCE to PURCHASES 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PURCHASES to SECSPREAD Response of PURCHASES to JSPREAD Response of PURCHASES to VOLATILITY Response of PURCHASES to ISSUANCE Response of PURCHASES to PURCHASES 5 5 5 5 5 4 4 4 4 4 3 3 3 3 3 2 2 2 2 2 1 1 1 1 1 0 0 0 0 0 -1 -1 -1 -1 -1 -2 -2 -2 -2 -2 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 53

FIGURE 14: ImpulseResponseFunctions: Conforming MarketSpread Response of SECSPREAD to SECSPREAD Response of SECSPREAD to CSPREAD Response of SECSPREAD to VOLATILITY Response of SECSPREAD to ISSUANCE Response of SECSPREAD to PURCHASES 10 10 10 10 10 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of CSPREAD to SECSPREAD Response of CSPREAD to CSPREAD Response of CSPREAD to VOLATILITY Response of CSPREAD to ISSUANCE Response of CSPREAD to PURCHASES 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of VOLATILITY to SECSPREAD Response of VOLATILITY to CSPREAD Response of VOLATILITY to VOLATILITY Response of VOLATILITY to ISSUANCE Response of VOLATILITY to PURCHASES .6 .6 .6 .6 .6 .5 .5 .5 .5 .5 .4 .4 .4 .4 .4 .3 .3 .3 .3 .3 .2 .2 .2 .2 .2 .1 .1 .1 .1 .1 .0 .0 .0 .0 .0 -.1 -.1 -.1 -.1 -.1 -.2 -.2 -.2 -.2 -.2 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of ISSUANCE to SECSPREAD Response of ISSUANCE to CSPREAD Response of ISSUANCE to VOLATILITY Response of ISSUANCE to ISSUANCE Response of ISSUANCE to PURCHASES 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PURCHASES to SECSPREAD Response of PURCHASES to CSPREAD Response of PURCHASES to VOLATILITY Response of PURCHASES to ISSUANCE Response of PURCHASES to PURCHASES 5 5 5 5 5 4 4 4 4 4 3 3 3 3 3 2 2 2 2 2 1 1 1 1 1 0 0 0 0 0 -1 -1 -1 -1 -1 -2 -2 -2 -2 -2 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 54

FIGURE 15: ResponseDuringLiquidity Crisisof1998 170 160 150 140 130 AUG SEP OCT NOV DEC stnioP sisaB Secondary Mortgage Rate Spread 260 250 240 230 220 210 200 190 AUG SEP OCT NOV DEC stnioP sisaB Primary Mortgage Rate Spreads 30 28 26 24 22 20 18 16 AUG SEP OCT NOV DEC snoitanigirO fo tnecreP Portfolio Purchases 46 44 42 40 38 36 34 32 30 AUG SEP OCT NOV DEC snoitanigirO fo tnecreP MBS Issuance 8.5 8.0 7.5 7.0 6.5 6.0 5.5 AUG SEP OCT NOV DEC Actual Jumbo Actual Conf. Predicted Jumbo Predicted Conf. Counterfactual Jumbo Counterfactual stnioP sisaB Implied Volatility 55

FIGURE 16: CumulativeImpulseResponseFunctions: FirstDifferences Response of SECSPREAD to SECSPREAD Response of SECSPREAD to PRISPREAD Response of SECSPREAD to VOLATILITY Response of SECSPREAD to ISSUANCE Response of SECSPREAD to PURCHASES 16 16 16 16 16 12 12 12 12 12 8 8 8 8 8 4 4 4 4 4 0 0 0 0 0 -4 -4 -4 -4 -4 -8 -8 -8 -8 -8 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PRISPREAD to SECSPREAD Response of PRISPREAD to PRISPREAD Response of PRISPREAD to VOLATILITY Response of PRISPREAD to ISSUANCE Response of PRISPREAD to PURCHASES 16 16 16 16 16 12 12 12 12 12 8 8 8 8 8 4 4 4 4 4 0 0 0 0 0 -4 -4 -4 -4 -4 -8 -8 -8 -8 -8 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of VOLATILITY to SECSPREAD Response of VOLATILITY to PRISPREAD Response of VOLATILITY to VOLATILITY Response of VOLATILITY to ISSUANCE Response of VOLATILITY to PURCHASES .8 .8 .8 .8 .8 .6 .6 .6 .6 .6 .4 .4 .4 .4 .4 .2 .2 .2 .2 .2 .0 .0 .0 .0 .0 -.2 -.2 -.2 -.2 -.2 -.4 -.4 -.4 -.4 -.4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of ISSUANCE to SECSPREAD Response of ISSUANCE to PRISPREAD Response of ISSUANCE to VOLATILITY Response of ISSUANCE to ISSUANCE Response of ISSUANCE to PURCHASES .20 .20 .20 .20 .20 .15 .15 .15 .15 .15 .10 .10 .10 .10 .10 .05 .05 .05 .05 .05 .00 .00 .00 .00 .00 -.05 -.05 -.05 -.05 -.05 -.10 -.10 -.10 -.10 -.10 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PURCHASES to SECSPREAD Response of PURCHASES to PRISPREAD Response of PURCHASES to VOLATILITY Response of PURCHASES to ISSUANCE Response of PURCHASES to PURCHASES .32 .32 .32 .32 .32 .28 .28 .28 .28 .28 .24 .24 .24 .24 .24 .20 .20 .20 .20 .20 .16 .16 .16 .16 .16 .12 .12 .12 .12 .12 .08 .08 .08 .08 .08 .04 .04 .04 .04 .04 .00 .00 .00 .00 .00 -.04 -.04 -.04 -.04 -.04 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 56

FIGURE 17: CumulativeImpulseResponseFunctions: FirstDifferences Response of SECSPREAD to SECSPREAD Response of SECSPREAD to PRISPREAD Response of SECSPREAD to VOLATILITY Response of SECSPREAD to ISSUANCE Response of SECSPREAD to PURCHASES 16 16 16 16 16 12 12 12 12 12 8 8 8 8 8 4 4 4 4 4 0 0 0 0 0 -4 -4 -4 -4 -4 -8 -8 -8 -8 -8 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PRISPREAD to SECSPREAD Response of PRISPREAD to PRISPREAD Response of PRISPREAD to VOLATILITY Response of PRISPREAD to ISSUANCE Response of PRISPREAD to PURCHASES 16 16 16 16 16 12 12 12 12 12 8 8 8 8 8 4 4 4 4 4 0 0 0 0 0 -4 -4 -4 -4 -4 -8 -8 -8 -8 -8 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of VOLATILITY to SECSPREAD Response of VOLATILITY to PRISPREAD Response of VOLATILITY to VOLATILITY Response of VOLATILITY to ISSUANCE Response of VOLATILITY to PURCHASES .8 .8 .8 .8 .8 .6 .6 .6 .6 .6 .4 .4 .4 .4 .4 .2 .2 .2 .2 .2 .0 .0 .0 .0 .0 -.2 -.2 -.2 -.2 -.2 -.4 -.4 -.4 -.4 -.4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of ISSUANCE to SECSPREAD Response of ISSUANCE to PRISPREAD Response of ISSUANCE to VOLATILITY Response of ISSUANCE to ISSUANCE Response of ISSUANCE to PURCHASES 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PURCHASES to SECSPREAD Response of PURCHASES to PRISPREAD Response of PURCHASES to VOLATILITY Response of PURCHASES to ISSUANCE Response of PURCHASES to PURCHASES 6 6 6 6 6 5 5 5 5 5 4 4 4 4 4 3 3 3 3 3 2 2 2 2 2 1 1 1 1 1 0 0 0 0 0 -1 -1 -1 -1 -1 -2 -2 -2 -2 -2 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 57

FIGURE 18: ImpulseResponseFunctions: Option-Adjusted Spreads Response of SECSPREAD to SECSPREAD Response of SECSPREAD to PRISPREAD Response of SECSPREAD to VOLATILITY Response of SECSPREAD to ISSUANCE Response of SECSPREAD to PURCHASES 15 15 15 15 15 10 10 10 10 10 5 5 5 5 5 0 0 0 0 0 -5 -5 -5 -5 -5 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PRISPREAD to SECSPREAD Response of PRISPREAD to PRISPREAD Response of PRISPREAD to VOLATILITY Response of PRISPREAD to ISSUANCE Response of PRISPREAD to PURCHASES 15 15 15 15 15 10 10 10 10 10 5 5 5 5 5 0 0 0 0 0 -5 -5 -5 -5 -5 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of VOLATILITY to SECSPREAD Response of VOLATILITY to PRISPREAD Response of VOLATILITY to VOLATILITY Response of VOLATILITY to ISSUANCE Response of VOLATILITY to PURCHASES .6 .6 .6 .6 .6 .5 .5 .5 .5 .5 .4 .4 .4 .4 .4 .3 .3 .3 .3 .3 .2 .2 .2 .2 .2 .1 .1 .1 .1 .1 .0 .0 .0 .0 .0 -.1 -.1 -.1 -.1 -.1 -.2 -.2 -.2 -.2 -.2 -.3 -.3 -.3 -.3 -.3 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of ISSUANCE to SECSPREAD Response of ISSUANCE to PRISPREAD Response of ISSUANCE to VOLATILITY Response of ISSUANCE to ISSUANCE Response of ISSUANCE to PURCHASES 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PURCHASES to SECSPREAD Response of PURCHASES to PRISPREAD Response of PURCHASES to VOLATILITY Response of PURCHASES to ISSUANCE Response of PURCHASES to PURCHASES 6 6 6 6 6 5 5 5 5 5 4 4 4 4 4 3 3 3 3 3 2 2 2 2 2 1 1 1 1 1 0 0 0 0 0 -1 -1 -1 -1 -1 -2 -2 -2 -2 -2 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 58

FIGURE 19: ImpulseResponseFunctions: FannieMaeCommitments Response of SECSPREAD to SECSPREAD Response of SECSPREAD to PRISPREAD Response of SECSPREAD to VOLATILITY Response of SECSPREAD to ISSUANCE Response of SECSPREAD to COMMITMENTS 10 10 10 10 10 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PRISPREAD to SECSPREAD Response of PRISPREAD to PRISPREAD Response of PRISPREAD to VOLATILITY Response of PRISPREAD to ISSUANCE Response of PRISPREAD to COMMITMENTS 10 10 10 10 10 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of VOLATILITY to SECSPREAD Response of VOLATILITY to PRISPREAD Response of VOLATILITY to VOLATILITY Response of VOLATILITY to ISSUANCE Response of VOLATILITY to COMMITMENTS .6 .6 .6 .6 .6 .5 .5 .5 .5 .5 .4 .4 .4 .4 .4 .3 .3 .3 .3 .3 .2 .2 .2 .2 .2 .1 .1 .1 .1 .1 .0 .0 .0 .0 .0 -.1 -.1 -.1 -.1 -.1 -.2 -.2 -.2 -.2 -.2 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of ISSUANCE to SECSPREAD Response of ISSUANCE to PRISPREAD Response of ISSUANCE to VOLATILITY Response of ISSUANCE to ISSUANCE Response of ISSUANCE to COMMITMENTS 4 4 4 4 4 3 3 3 3 3 2 2 2 2 2 1 1 1 1 1 0 0 0 0 0 -1 -1 -1 -1 -1 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of COMMITMENTS to SECSPREADResponse of COMMITMENTS to PRISPREADResponse of COMMITMENTS to VOLATILITY Response of COMMITMENTS to ISSUANCEResponse of COMMITMENTS to COMMITMENTS 4 4 4 4 4 3 3 3 3 3 2 2 2 2 2 1 1 1 1 1 0 0 0 0 0 -1 -1 -1 -1 -1 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 59

FIGURE 20: ImpulseResponse Functions: ProxyforCreditRisk Response of SECSPREAD to SECSPREAD Response of SECSPREAD to PRISPREAD Response of SECSPREAD to VOLATILITY Response of SECSPREAD to ISSUANCE Response of SECSPREAD to PURCHASES 10 10 10 10 10 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PRISPREAD to SECSPREAD Response of PRISPREAD to PRISPREAD Response of PRISPREAD to VOLATILITY Response of PRISPREAD to ISSUANCE Response of PRISPREAD to PURCHASES 10 10 10 10 10 8 8 8 8 8 6 6 6 6 6 4 4 4 4 4 2 2 2 2 2 0 0 0 0 0 -2 -2 -2 -2 -2 -4 -4 -4 -4 -4 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of VOLATILITY to SECSPREAD Response of VOLATILITY to PRISPREAD Response of VOLATILITY to VOLATILITY Response of VOLATILITY to ISSUANCE Response of VOLATILITY to PURCHASES .6 .6 .6 .6 .6 .5 .5 .5 .5 .5 .4 .4 .4 .4 .4 .3 .3 .3 .3 .3 .2 .2 .2 .2 .2 .1 .1 .1 .1 .1 .0 .0 .0 .0 .0 -.1 -.1 -.1 -.1 -.1 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of ISSUANCE to SECSPREAD Response of ISSUANCE to PRISPREAD Response of ISSUANCE to VOLATILITY Response of ISSUANCE to ISSUANCE Response of ISSUANCE to PURCHASES 7 7 7 7 7 6 6 6 6 6 5 5 5 5 5 4 4 4 4 4 3 3 3 3 3 2 2 2 2 2 1 1 1 1 1 0 0 0 0 0 -1 -1 -1 -1 -1 -2 -2 -2 -2 -2 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Response of PURCHASES to SECSPREAD Response of PURCHASES to PRISPREAD Response of PURCHASES to VOLATILITY Response of PURCHASES to ISSUANCE Response of PURCHASES to PURCHASES 6 6 6 6 6 5 5 5 5 5 4 4 4 4 4 3 3 3 3 3 2 2 2 2 2 1 1 1 1 1 0 0 0 0 0 -1 -1 -1 -1 -1 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 60

A Comparison with Other Studies A.1 Gonzalez-Rivera(2001) We were able to produce essentially the same results as Gonzalez-Rivera (2001) whenusingthesamemethodology, specification, anddataperiod(December1994 to December 1999). Gonzalez-Rivera reports the cointegrating relationship (⋆ denotesstatistical significance atthe95percentconfidence level): Purchases= −51.35+0.554⋆ Spread. Using our data over the same time period, we estimate the cointegrating relationshiptobe: Purchases= −51.47+0.551⋆ Spread. In both cases, these estimates suggest that increases in portfolio purchases are associated withincreases inmortgagemarketspreads. AswithGonzalez-Rivera, we also find that the error-correction term is statistically significant in the secondary market spread equation, but not statistically significant in the portfolio purchase equation. This suggests that secondary market spreads, not portfolio purchases, carry out any adjustment toward restoration of the long-run relationship. Interestingly,wecannotrejectthenullhypothesisofnocointegratingrelationshipbetween portfolio purchases andsecondary marketspreads. Estimatesofcointegratingrelationshipsideallyrequirelongtimesamples. While Gonzalez-Rivera was limited to essentially five years of data, our full sample is morethantwiceaslong. Whenweincludetheextrasixyearsofdatainoursample (2000–2005) weestimatethelong-run relationship tobe: Purchases= 30.13+0.015Spread. Here, we find no evidence of a long-run relationship between portfolio purchases and mortgage rate spreads, as we again cannot reject the null hypothesis of no cointegrating relationship. A.2 Naranjo andToevs(2002) In their paper, Naranjo and Toevs (2002) also posit a cointegrating relationship between mortgage market spreads and GSE activities.19 Naranjo and Toevs use 19NaranjoandToevsdonotincludeaconstantintheirlong-runcointegratingrelationship. Given the presence of prepayment and credit risk with mortgages, a long-run cointegrating relationship should, perhaps, include a (positive) constant. Despite this apparent omission, we continue as in NaranjoandToevs. 61

mortgage rate data from the FHFB’s Monthly Interest Rate Survey (MIRS) and cantherefore distinguish betweenjumboandconforming mortgagerates. Naranjo and Toevs use non-public data on Fannie Mae’s portfolio activity going back to 1986,sowecannotexactlyreplicatetheirresults. Weinsteadattempttoreproduce theirresults usingthesamespecification andcomparable data. NaranjoandToevs report the following cointegrating relationships based on their data from 1986 to 1998(⋆ denotes statistical significance atthe95percentconfidence level): Jumbo-ConformingSpread= 0.27⋆ Purchases Jumbo-ConformingSpread= 0.48⋆ Issuance ConformingSpread=−1.74⋆ Purchases ConformingSpread=−3.53⋆ Issuance JumboSpread=−0.84⋆ Purchases JumboSpread=−1.92⋆ Issuance. Theseresultssuggestthatincreasedpurchasesareassociatedwithdecreasedjumbo and conforming marketspreads andincreased jumbo-conforming spreads. NaranjoandToevsconclude thatincreased portfolio purchases decrease jumboandconformingmortgageratespreads, whileincreasing thejumbo-conforming spread. In addition to the data we use in the main part of the paper, we also collected data on jumboand conforming mortgage rates using theMIRSforMarch 1993 to December2005. UsingthedatafromMarch1993toDecember1998—ourclosest matchtoNaranjoandToevs’dataspan—weobtained thefollowingresults: Jumbo-ConformingSpread=−0.05Purchases Jumbo-ConformingSpread=−1.03⋆ Issuance ConformingSpread= 0.22⋆ Purchases ConformingSpread= 0.89⋆ Issuance JumboSpread= 0.32Purchases JumboSpread= 1.74⋆ Issuance. We find that securitization and portfolio purchases are both positively correlated withmortgagemarketspreads,butnegativelycorrelatedwiththejumbo-conforming spread,inthelongrun. Onlyinthefourthequationdowefindevidencesupporting asinglecointegrating relationship. 62

Usingourcompletedataset(March1993toDecember2005),weobtained: Jumbo-ConformingSpread= −0.01Purchases Jumbo-ConformingSpread= 0.01Securitization ConformingSpread= 0.43⋆ Purchases ConformingSpread= 1.62⋆ Securitization JumboSpread= 0.53⋆ Purchases JumboSpread= 1.84⋆ Securitization. Again we find that securitization and portfolio purchases are both positively correlated with mortgage market spreads, but negatively correlated with the jumboconforming spread, in the long run. Here, we find only find evidence of a single cointegrating relationship inthefourthandsixthequations. A.3 Discussion Notethat ourresults areperfectly consistent withthefindings ofGonzalez-Rivera (2001): mortgage rate spreads and GSE portfolio purchases are positively correlated. However,ourresultscontradict thosereportedinNaranjoandToevs(2002). Naranjo and Toevs find that mortgage rate spreads are negatively correlated with GSE secondary market activities, while we find a positive relationship; Naranjo and Toevs find a positive relationship between the jumbo-conforming spread and GSEsecondary market activities, whilewefind, ifanything atall, anegative relationship. 63